program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "5.33.5"}, {"coremlc-version", "1877.0.48"}, {"coremltools-component-torch", "2.0.1"}, {"coremltools-version", "7.0b1"}})]
{
    func main<ios16>(tensor<fp16, [2, 320, 128, 128]> additional_residual_0, tensor<fp16, [2, 320, 128, 128]> additional_residual_1, tensor<fp16, [2, 320, 128, 128]> additional_residual_2, tensor<fp16, [2, 320, 64, 64]> additional_residual_3, tensor<fp16, [2, 640, 64, 64]> additional_residual_4, tensor<fp16, [2, 640, 64, 64]> additional_residual_5, tensor<fp16, [2, 640, 32, 32]> additional_residual_6, tensor<fp16, [2, 1280, 32, 32]> additional_residual_7, tensor<fp16, [2, 1280, 32, 32]> additional_residual_8, tensor<fp16, [2, 1280, 32, 32]> additional_residual_9, tensor<fp16, [2, 2048, 1, 77]> encoder_hidden_states, tensor<fp16, [2, 4, 128, 128]> sample, tensor<fp16, [2, 1280]> text_embeds, tensor<fp16, [12]> time_ids, tensor<fp16, [2]> timestep) {
            tensor<int32, []> var_34 = const()[name = tensor<string, []>("op_34"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_51_axes_0 = const()[name = tensor<string, []>("op_51_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1]> var_51_cast = expand_dims(axes = var_51_axes_0, x = timestep)[name = tensor<string, []>("op_51_cast")];
            tensor<fp16, [1, 160]> var_53_to_fp16 = const()[name = tensor<string, []>("op_53_to_fp16"), val = tensor<fp16, [1, 160]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
            tensor<fp16, [2, 160]> emb_3_cast = mul(x = var_51_cast, y = var_53_to_fp16)[name = tensor<string, []>("emb_3_cast")];
            tensor<fp16, [2, 160]> var_58_cast = sin(x = emb_3_cast)[name = tensor<string, []>("op_58_cast")];
            tensor<fp16, [2, 160]> var_59_cast = cos(x = emb_3_cast)[name = tensor<string, []>("op_59_cast")];
            tensor<bool, []> emb_7_interleave_0 = const()[name = tensor<string, []>("emb_7_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 320]> emb_7_cast = concat(axis = var_34, interleave = emb_7_interleave_0, values = (var_58_cast, var_59_cast))[name = tensor<string, []>("emb_7_cast")];
            tensor<int32, [2]> var_63_begin_0 = const()[name = tensor<string, []>("op_63_begin_0"), val = tensor<int32, [2]>([0, 160])];
            tensor<int32, [2]> var_63_end_0 = const()[name = tensor<string, []>("op_63_end_0"), val = tensor<int32, [2]>([2, 320])];
            tensor<bool, [2]> var_63_end_mask_0 = const()[name = tensor<string, []>("op_63_end_mask_0"), val = tensor<bool, [2]>([true, true])];
            tensor<fp16, [2, 160]> var_63_cast = slice_by_index(begin = var_63_begin_0, end = var_63_end_0, end_mask = var_63_end_mask_0, x = emb_7_cast)[name = tensor<string, []>("op_63_cast")];
            tensor<int32, [2]> var_65_begin_0 = const()[name = tensor<string, []>("op_65_begin_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [2]> var_65_end_0 = const()[name = tensor<string, []>("op_65_end_0"), val = tensor<int32, [2]>([2, 160])];
            tensor<bool, [2]> var_65_end_mask_0 = const()[name = tensor<string, []>("op_65_end_mask_0"), val = tensor<bool, [2]>([true, false])];
            tensor<fp16, [2, 160]> var_65_cast = slice_by_index(begin = var_65_begin_0, end = var_65_end_0, end_mask = var_65_end_mask_0, x = emb_7_cast)[name = tensor<string, []>("op_65_cast")];
            tensor<bool, []> sample_1_interleave_0 = const()[name = tensor<string, []>("sample_1_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 320]> sample_1_cast = concat(axis = var_34, interleave = sample_1_interleave_0, values = (var_63_cast, var_65_cast))[name = tensor<string, []>("sample_1_cast")];
            tensor<int32, []> var_68 = const()[name = tensor<string, []>("op_68"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> var_75_axes_0 = const()[name = tensor<string, []>("op_75_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2, 320, 1]> var_75_cast = expand_dims(axes = var_75_axes_0, x = sample_1_cast)[name = tensor<string, []>("op_75_cast")];
            tensor<int32, [1]> input_1_axes_0 = const()[name = tensor<string, []>("input_1_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2, 320, 1, 1]> input_1_cast = expand_dims(axes = input_1_axes_0, x = var_75_cast)[name = tensor<string, []>("input_1_cast")];
            tensor<int32, [2]> var_79 = const()[name = tensor<string, []>("op_79"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_81 = const()[name = tensor<string, []>("op_81"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> input_3_pad_type_0 = const()[name = tensor<string, []>("input_3_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> input_3_pad_0 = const()[name = tensor<string, []>("input_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 320, 1, 1]> time_embedding_linear_1_weight_to_fp16 = const()[name = tensor<string, []>("time_embedding_linear_1_weight_to_fp16"), val = tensor<fp16, [1280, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(448)))];
            tensor<fp16, [1280]> time_embedding_linear_1_bias_to_fp16 = const()[name = tensor<string, []>("time_embedding_linear_1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(819712)))];
            tensor<fp16, [2, 1280, 1, 1]> input_3_cast = conv(bias = time_embedding_linear_1_bias_to_fp16, dilations = var_81, groups = var_68, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = var_79, weight = time_embedding_linear_1_weight_to_fp16, x = input_1_cast)[name = tensor<string, []>("input_3_cast")];
            tensor<fp16, [2, 1280, 1, 1]> input_5_cast = silu(x = input_3_cast)[name = tensor<string, []>("input_5_cast")];
            tensor<int32, [2]> var_87 = const()[name = tensor<string, []>("op_87"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_89 = const()[name = tensor<string, []>("op_89"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> emb_pad_type_0 = const()[name = tensor<string, []>("emb_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> emb_pad_0 = const()[name = tensor<string, []>("emb_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> time_embedding_linear_2_weight_to_fp16 = const()[name = tensor<string, []>("time_embedding_linear_2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(822336)))];
            tensor<fp16, [1280]> time_embedding_linear_2_bias_to_fp16 = const()[name = tensor<string, []>("time_embedding_linear_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4099200)))];
            tensor<fp16, [2, 1280, 1, 1]> emb_cast = conv(bias = time_embedding_linear_2_bias_to_fp16, dilations = var_89, groups = var_68, pad = emb_pad_0, pad_type = emb_pad_type_0, strides = var_87, weight = time_embedding_linear_2_weight_to_fp16, x = input_5_cast)[name = tensor<string, []>("emb_cast")];
            tensor<int32, []> var_95 = const()[name = tensor<string, []>("op_95"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_112_axes_0 = const()[name = tensor<string, []>("op_112_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [12, 1]> var_112_cast = expand_dims(axes = var_112_axes_0, x = time_ids)[name = tensor<string, []>("op_112_cast")];
            tensor<fp16, [1, 128]> var_114_to_fp16 = const()[name = tensor<string, []>("op_114_to_fp16"), val = tensor<fp16, [1, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4101824)))];
            tensor<fp16, [12, 128]> emb_11_cast = mul(x = var_112_cast, y = var_114_to_fp16)[name = tensor<string, []>("emb_11_cast")];
            tensor<fp16, [12, 128]> var_119_cast = sin(x = emb_11_cast)[name = tensor<string, []>("op_119_cast")];
            tensor<fp16, [12, 128]> var_120_cast = cos(x = emb_11_cast)[name = tensor<string, []>("op_120_cast")];
            tensor<bool, []> emb_15_interleave_0 = const()[name = tensor<string, []>("emb_15_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [12, 256]> emb_15_cast = concat(axis = var_95, interleave = emb_15_interleave_0, values = (var_119_cast, var_120_cast))[name = tensor<string, []>("emb_15_cast")];
            tensor<int32, [2]> var_124_begin_0 = const()[name = tensor<string, []>("op_124_begin_0"), val = tensor<int32, [2]>([0, 128])];
            tensor<int32, [2]> var_124_end_0 = const()[name = tensor<string, []>("op_124_end_0"), val = tensor<int32, [2]>([12, 256])];
            tensor<bool, [2]> var_124_end_mask_0 = const()[name = tensor<string, []>("op_124_end_mask_0"), val = tensor<bool, [2]>([true, true])];
            tensor<fp16, [12, 128]> var_124_cast = slice_by_index(begin = var_124_begin_0, end = var_124_end_0, end_mask = var_124_end_mask_0, x = emb_15_cast)[name = tensor<string, []>("op_124_cast")];
            tensor<int32, [2]> var_126_begin_0 = const()[name = tensor<string, []>("op_126_begin_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [2]> var_126_end_0 = const()[name = tensor<string, []>("op_126_end_0"), val = tensor<int32, [2]>([12, 128])];
            tensor<bool, [2]> var_126_end_mask_0 = const()[name = tensor<string, []>("op_126_end_mask_0"), val = tensor<bool, [2]>([true, false])];
            tensor<fp16, [12, 128]> var_126_cast = slice_by_index(begin = var_126_begin_0, end = var_126_end_0, end_mask = var_126_end_mask_0, x = emb_15_cast)[name = tensor<string, []>("op_126_cast")];
            tensor<bool, []> time_embeds_1_interleave_0 = const()[name = tensor<string, []>("time_embeds_1_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [12, 256]> time_embeds_1_cast = concat(axis = var_95, interleave = time_embeds_1_interleave_0, values = (var_124_cast, var_126_cast))[name = tensor<string, []>("time_embeds_1_cast")];
            tensor<int32, [2]> var_134 = const()[name = tensor<string, []>("op_134"), val = tensor<int32, [2]>([2, -1])];
            tensor<fp16, [2, 1536]> time_embeds_cast = reshape(shape = var_134, x = time_embeds_1_cast)[name = tensor<string, []>("time_embeds_cast")];
            tensor<int32, []> var_137 = const()[name = tensor<string, []>("op_137"), val = tensor<int32, []>(-1)];
            tensor<bool, []> sample_3_interleave_0 = const()[name = tensor<string, []>("sample_3_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 2816]> sample_3_cast = concat(axis = var_137, interleave = sample_3_interleave_0, values = (text_embeds, time_embeds_cast))[name = tensor<string, []>("sample_3_cast")];
            tensor<int32, []> var_139 = const()[name = tensor<string, []>("op_139"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> var_146_axes_0 = const()[name = tensor<string, []>("op_146_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2, 2816, 1]> var_146_cast = expand_dims(axes = var_146_axes_0, x = sample_3_cast)[name = tensor<string, []>("op_146_cast")];
            tensor<int32, [1]> input_7_axes_0 = const()[name = tensor<string, []>("input_7_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2, 2816, 1, 1]> input_7_cast = expand_dims(axes = input_7_axes_0, x = var_146_cast)[name = tensor<string, []>("input_7_cast")];
            tensor<int32, [2]> var_150 = const()[name = tensor<string, []>("op_150"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_152 = const()[name = tensor<string, []>("op_152"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> input_9_pad_type_0 = const()[name = tensor<string, []>("input_9_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> input_9_pad_0 = const()[name = tensor<string, []>("input_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2816, 1, 1]> add_embedding_linear_1_weight_to_fp16 = const()[name = tensor<string, []>("add_embedding_linear_1_weight_to_fp16"), val = tensor<fp16, [1280, 2816, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4102144)))];
            tensor<fp16, [1280]> add_embedding_linear_1_bias_to_fp16 = const()[name = tensor<string, []>("add_embedding_linear_1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11311168)))];
            tensor<fp16, [2, 1280, 1, 1]> input_9_cast = conv(bias = add_embedding_linear_1_bias_to_fp16, dilations = var_152, groups = var_139, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = var_150, weight = add_embedding_linear_1_weight_to_fp16, x = input_7_cast)[name = tensor<string, []>("input_9_cast")];
            tensor<fp16, [2, 1280, 1, 1]> input_11_cast = silu(x = input_9_cast)[name = tensor<string, []>("input_11_cast")];
            tensor<int32, [2]> var_158 = const()[name = tensor<string, []>("op_158"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_160 = const()[name = tensor<string, []>("op_160"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> aug_emb_pad_type_0 = const()[name = tensor<string, []>("aug_emb_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> aug_emb_pad_0 = const()[name = tensor<string, []>("aug_emb_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> add_embedding_linear_2_weight_to_fp16 = const()[name = tensor<string, []>("add_embedding_linear_2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11313792)))];
            tensor<fp16, [1280]> add_embedding_linear_2_bias_to_fp16 = const()[name = tensor<string, []>("add_embedding_linear_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14590656)))];
            tensor<fp16, [2, 1280, 1, 1]> aug_emb_cast = conv(bias = add_embedding_linear_2_bias_to_fp16, dilations = var_160, groups = var_139, pad = aug_emb_pad_0, pad_type = aug_emb_pad_type_0, strides = var_158, weight = add_embedding_linear_2_weight_to_fp16, x = input_11_cast)[name = tensor<string, []>("aug_emb_cast")];
            tensor<fp16, [2, 1280, 1, 1]> input_21_cast = add(x = emb_cast, y = aug_emb_cast)[name = tensor<string, []>("input_21_cast")];
            tensor<int32, []> var_168 = const()[name = tensor<string, []>("op_168"), val = tensor<int32, []>(1)];
            tensor<int32, [2]> var_171 = const()[name = tensor<string, []>("op_171"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_173 = const()[name = tensor<string, []>("op_173"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> input_15_pad_type_0 = const()[name = tensor<string, []>("input_15_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> input_15_pad_0 = const()[name = tensor<string, []>("input_15_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 4, 3, 3]> conv_in_weight_to_fp16 = const()[name = tensor<string, []>("conv_in_weight_to_fp16"), val = tensor<fp16, [320, 4, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14593280)))];
            tensor<fp16, [320]> conv_in_bias_to_fp16 = const()[name = tensor<string, []>("conv_in_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14616384)))];
            tensor<fp16, [2, 320, 128, 128]> input_15_cast = conv(bias = conv_in_bias_to_fp16, dilations = var_173, groups = var_168, pad = input_15_pad_0, pad_type = input_15_pad_type_0, strides = var_171, weight = conv_in_weight_to_fp16, x = sample)[name = tensor<string, []>("input_15_cast")];
            tensor<int32, []> var_182 = const()[name = tensor<string, []>("op_182"), val = tensor<int32, []>(1)];
            tensor<int32, [5]> reshape_0_shape_0 = const()[name = tensor<string, []>("reshape_0_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 128, 128])];
            tensor<fp16, [2, 32, 10, 128, 128]> reshape_0_cast = reshape(shape = reshape_0_shape_0, x = input_15_cast)[name = tensor<string, []>("reshape_0_cast")];
            tensor<int32, [3]> reduce_mean_0_axes_0 = const()[name = tensor<string, []>("reduce_mean_0_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_0_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_0_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_0_cast = reduce_mean(axes = reduce_mean_0_axes_0, keep_dims = reduce_mean_0_keep_dims_0, x = reshape_0_cast)[name = tensor<string, []>("reduce_mean_0_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> sub_0_cast = sub(x = reshape_0_cast, y = reduce_mean_0_cast)[name = tensor<string, []>("sub_0_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> square_0_cast = square(x = sub_0_cast)[name = tensor<string, []>("square_0_cast")];
            tensor<int32, [3]> reduce_mean_2_axes_0 = const()[name = tensor<string, []>("reduce_mean_2_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_2_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_2_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_2_cast = reduce_mean(axes = reduce_mean_2_axes_0, keep_dims = reduce_mean_2_keep_dims_0, x = square_0_cast)[name = tensor<string, []>("reduce_mean_2_cast")];
            tensor<fp16, []> add_0_y_0_to_fp16 = const()[name = tensor<string, []>("add_0_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_0_cast = add(x = reduce_mean_2_cast, y = add_0_y_0_to_fp16)[name = tensor<string, []>("add_0_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_0_cast = sqrt(x = add_0_cast)[name = tensor<string, []>("sqrt_0_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> real_div_0_cast = real_div(x = sub_0_cast, y = sqrt_0_cast)[name = tensor<string, []>("real_div_0_cast")];
            tensor<int32, [4]> reshape_1_shape_0 = const()[name = tensor<string, []>("reshape_1_shape_0"), val = tensor<int32, [4]>([2, 320, 128, 128])];
            tensor<fp16, [2, 320, 128, 128]> reshape_1_cast = reshape(shape = reshape_1_shape_0, x = real_div_0_cast)[name = tensor<string, []>("reshape_1_cast")];
            tensor<fp16, [320]> add_1_mean_0_to_fp16 = const()[name = tensor<string, []>("add_1_mean_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14617088)))];
            tensor<fp16, [320]> add_1_variance_0_to_fp16 = const()[name = tensor<string, []>("add_1_variance_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14617792)))];
            tensor<fp16, [320]> add_1_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_1_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14618496)))];
            tensor<fp16, [320]> add_1_beta_0_to_fp16 = const()[name = tensor<string, []>("add_1_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14619200)))];
            tensor<fp16, []> add_1_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_1_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 128, 128]> add_1_cast = batch_norm(beta = add_1_beta_0_to_fp16, epsilon = add_1_epsilon_0_to_fp16, gamma = add_1_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_1_cast)[name = tensor<string, []>("add_1_cast")];
            tensor<fp16, [2, 320, 128, 128]> input_19_cast = silu(x = add_1_cast)[name = tensor<string, []>("input_19_cast")];
            tensor<int32, [2]> var_200 = const()[name = tensor<string, []>("op_200"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_202 = const()[name = tensor<string, []>("op_202"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_1_pad_type_0 = const()[name = tensor<string, []>("hidden_states_1_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_1_pad_0 = const()[name = tensor<string, []>("hidden_states_1_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 320, 3, 3]> down_blocks_0_resnets_0_conv1_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [320, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14619904)))];
            tensor<fp16, [320]> down_blocks_0_resnets_0_conv1_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16463168)))];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_1_cast = conv(bias = down_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = var_202, groups = var_182, pad = hidden_states_1_pad_0, pad_type = hidden_states_1_pad_type_0, strides = var_200, weight = down_blocks_0_resnets_0_conv1_weight_to_fp16, x = input_19_cast)[name = tensor<string, []>("hidden_states_1_cast")];
            tensor<fp16, [2, 1280, 1, 1]> input_23_cast = silu(x = input_21_cast)[name = tensor<string, []>("input_23_cast")];
            tensor<int32, [2]> var_208 = const()[name = tensor<string, []>("op_208"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_210 = const()[name = tensor<string, []>("op_210"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_1_pad_type_0 = const()[name = tensor<string, []>("temb_1_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_1_pad_0 = const()[name = tensor<string, []>("temb_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1280, 1, 1]> down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [320, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16463872)))];
            tensor<fp16, [320]> down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17283136)))];
            tensor<fp16, [2, 320, 1, 1]> temb_1_cast = conv(bias = down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_210, groups = var_182, pad = temb_1_pad_0, pad_type = temb_1_pad_type_0, strides = var_208, weight = down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16, x = input_23_cast)[name = tensor<string, []>("temb_1_cast")];
            tensor<fp16, [2, 320, 128, 128]> input_25_cast = add(x = hidden_states_1_cast, y = temb_1_cast)[name = tensor<string, []>("input_25_cast")];
            tensor<int32, [5]> reshape_4_shape_0 = const()[name = tensor<string, []>("reshape_4_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 128, 128])];
            tensor<fp16, [2, 32, 10, 128, 128]> reshape_4_cast = reshape(shape = reshape_4_shape_0, x = input_25_cast)[name = tensor<string, []>("reshape_4_cast")];
            tensor<int32, [3]> reduce_mean_3_axes_0 = const()[name = tensor<string, []>("reduce_mean_3_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_3_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_3_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_3_cast = reduce_mean(axes = reduce_mean_3_axes_0, keep_dims = reduce_mean_3_keep_dims_0, x = reshape_4_cast)[name = tensor<string, []>("reduce_mean_3_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> sub_2_cast = sub(x = reshape_4_cast, y = reduce_mean_3_cast)[name = tensor<string, []>("sub_2_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> square_1_cast = square(x = sub_2_cast)[name = tensor<string, []>("square_1_cast")];
            tensor<int32, [3]> reduce_mean_5_axes_0 = const()[name = tensor<string, []>("reduce_mean_5_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_5_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_5_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_5_cast = reduce_mean(axes = reduce_mean_5_axes_0, keep_dims = reduce_mean_5_keep_dims_0, x = square_1_cast)[name = tensor<string, []>("reduce_mean_5_cast")];
            tensor<fp16, []> add_2_y_0_to_fp16 = const()[name = tensor<string, []>("add_2_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_2_cast = add(x = reduce_mean_5_cast, y = add_2_y_0_to_fp16)[name = tensor<string, []>("add_2_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_1_cast = sqrt(x = add_2_cast)[name = tensor<string, []>("sqrt_1_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> real_div_1_cast = real_div(x = sub_2_cast, y = sqrt_1_cast)[name = tensor<string, []>("real_div_1_cast")];
            tensor<int32, [4]> reshape_5_shape_0 = const()[name = tensor<string, []>("reshape_5_shape_0"), val = tensor<int32, [4]>([2, 320, 128, 128])];
            tensor<fp16, [2, 320, 128, 128]> reshape_5_cast = reshape(shape = reshape_5_shape_0, x = real_div_1_cast)[name = tensor<string, []>("reshape_5_cast")];
            tensor<fp16, [320]> add_3_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_3_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17283840)))];
            tensor<fp16, [320]> add_3_beta_0_to_fp16 = const()[name = tensor<string, []>("add_3_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17284544)))];
            tensor<fp16, []> add_3_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_3_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 128, 128]> add_3_cast = batch_norm(beta = add_3_beta_0_to_fp16, epsilon = add_3_epsilon_0_to_fp16, gamma = add_3_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_5_cast)[name = tensor<string, []>("add_3_cast")];
            tensor<fp16, [2, 320, 128, 128]> input_29_cast = silu(x = add_3_cast)[name = tensor<string, []>("input_29_cast")];
            tensor<int32, [2]> var_220 = const()[name = tensor<string, []>("op_220"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_222 = const()[name = tensor<string, []>("op_222"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_3_pad_type_0 = const()[name = tensor<string, []>("hidden_states_3_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_3_pad_0 = const()[name = tensor<string, []>("hidden_states_3_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 320, 3, 3]> down_blocks_0_resnets_0_conv2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [320, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17285248)))];
            tensor<fp16, [320]> down_blocks_0_resnets_0_conv2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19128512)))];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_3_cast = conv(bias = down_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = var_222, groups = var_182, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = var_220, weight = down_blocks_0_resnets_0_conv2_weight_to_fp16, x = input_29_cast)[name = tensor<string, []>("hidden_states_3_cast")];
            tensor<fp16, [2, 320, 128, 128]> input_31_cast = add(x = input_15_cast, y = hidden_states_3_cast)[name = tensor<string, []>("input_31_cast")];
            tensor<int32, [5]> reshape_8_shape_0 = const()[name = tensor<string, []>("reshape_8_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 128, 128])];
            tensor<fp16, [2, 32, 10, 128, 128]> reshape_8_cast = reshape(shape = reshape_8_shape_0, x = input_31_cast)[name = tensor<string, []>("reshape_8_cast")];
            tensor<int32, [3]> reduce_mean_6_axes_0 = const()[name = tensor<string, []>("reduce_mean_6_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_6_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_6_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_6_cast = reduce_mean(axes = reduce_mean_6_axes_0, keep_dims = reduce_mean_6_keep_dims_0, x = reshape_8_cast)[name = tensor<string, []>("reduce_mean_6_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> sub_4_cast = sub(x = reshape_8_cast, y = reduce_mean_6_cast)[name = tensor<string, []>("sub_4_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> square_2_cast = square(x = sub_4_cast)[name = tensor<string, []>("square_2_cast")];
            tensor<int32, [3]> reduce_mean_8_axes_0 = const()[name = tensor<string, []>("reduce_mean_8_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_8_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_8_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_8_cast = reduce_mean(axes = reduce_mean_8_axes_0, keep_dims = reduce_mean_8_keep_dims_0, x = square_2_cast)[name = tensor<string, []>("reduce_mean_8_cast")];
            tensor<fp16, []> add_4_y_0_to_fp16 = const()[name = tensor<string, []>("add_4_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_4_cast = add(x = reduce_mean_8_cast, y = add_4_y_0_to_fp16)[name = tensor<string, []>("add_4_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_2_cast = sqrt(x = add_4_cast)[name = tensor<string, []>("sqrt_2_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> real_div_2_cast = real_div(x = sub_4_cast, y = sqrt_2_cast)[name = tensor<string, []>("real_div_2_cast")];
            tensor<int32, [4]> reshape_9_shape_0 = const()[name = tensor<string, []>("reshape_9_shape_0"), val = tensor<int32, [4]>([2, 320, 128, 128])];
            tensor<fp16, [2, 320, 128, 128]> reshape_9_cast = reshape(shape = reshape_9_shape_0, x = real_div_2_cast)[name = tensor<string, []>("reshape_9_cast")];
            tensor<fp16, [320]> add_5_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_5_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19129216)))];
            tensor<fp16, [320]> add_5_beta_0_to_fp16 = const()[name = tensor<string, []>("add_5_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19129920)))];
            tensor<fp16, []> add_5_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_5_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 128, 128]> add_5_cast = batch_norm(beta = add_5_beta_0_to_fp16, epsilon = add_5_epsilon_0_to_fp16, gamma = add_5_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_9_cast)[name = tensor<string, []>("add_5_cast")];
            tensor<fp16, [2, 320, 128, 128]> input_35_cast = silu(x = add_5_cast)[name = tensor<string, []>("input_35_cast")];
            tensor<int32, [2]> var_237 = const()[name = tensor<string, []>("op_237"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_239 = const()[name = tensor<string, []>("op_239"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_5_pad_type_0 = const()[name = tensor<string, []>("hidden_states_5_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_5_pad_0 = const()[name = tensor<string, []>("hidden_states_5_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 320, 3, 3]> down_blocks_0_resnets_1_conv1_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_1_conv1_weight_to_fp16"), val = tensor<fp16, [320, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19130624)))];
            tensor<fp16, [320]> down_blocks_0_resnets_1_conv1_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_1_conv1_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20973888)))];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_5_cast = conv(bias = down_blocks_0_resnets_1_conv1_bias_to_fp16, dilations = var_239, groups = var_182, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = var_237, weight = down_blocks_0_resnets_1_conv1_weight_to_fp16, x = input_35_cast)[name = tensor<string, []>("hidden_states_5_cast")];
            tensor<int32, [2]> var_245 = const()[name = tensor<string, []>("op_245"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_247 = const()[name = tensor<string, []>("op_247"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_3_pad_type_0 = const()[name = tensor<string, []>("temb_3_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_3_pad_0 = const()[name = tensor<string, []>("temb_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1280, 1, 1]> down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [320, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20974592)))];
            tensor<fp16, [320]> down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21793856)))];
            tensor<fp16, [2, 320, 1, 1]> temb_3_cast = conv(bias = down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_247, groups = var_182, pad = temb_3_pad_0, pad_type = temb_3_pad_type_0, strides = var_245, weight = down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16, x = input_23_cast)[name = tensor<string, []>("temb_3_cast")];
            tensor<fp16, [2, 320, 128, 128]> input_39_cast = add(x = hidden_states_5_cast, y = temb_3_cast)[name = tensor<string, []>("input_39_cast")];
            tensor<int32, [5]> reshape_12_shape_0 = const()[name = tensor<string, []>("reshape_12_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 128, 128])];
            tensor<fp16, [2, 32, 10, 128, 128]> reshape_12_cast = reshape(shape = reshape_12_shape_0, x = input_39_cast)[name = tensor<string, []>("reshape_12_cast")];
            tensor<int32, [3]> reduce_mean_9_axes_0 = const()[name = tensor<string, []>("reduce_mean_9_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_9_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_9_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_9_cast = reduce_mean(axes = reduce_mean_9_axes_0, keep_dims = reduce_mean_9_keep_dims_0, x = reshape_12_cast)[name = tensor<string, []>("reduce_mean_9_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> sub_6_cast = sub(x = reshape_12_cast, y = reduce_mean_9_cast)[name = tensor<string, []>("sub_6_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> square_3_cast = square(x = sub_6_cast)[name = tensor<string, []>("square_3_cast")];
            tensor<int32, [3]> reduce_mean_11_axes_0 = const()[name = tensor<string, []>("reduce_mean_11_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_11_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_11_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_11_cast = reduce_mean(axes = reduce_mean_11_axes_0, keep_dims = reduce_mean_11_keep_dims_0, x = square_3_cast)[name = tensor<string, []>("reduce_mean_11_cast")];
            tensor<fp16, []> add_6_y_0_to_fp16 = const()[name = tensor<string, []>("add_6_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_6_cast = add(x = reduce_mean_11_cast, y = add_6_y_0_to_fp16)[name = tensor<string, []>("add_6_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_3_cast = sqrt(x = add_6_cast)[name = tensor<string, []>("sqrt_3_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> real_div_3_cast = real_div(x = sub_6_cast, y = sqrt_3_cast)[name = tensor<string, []>("real_div_3_cast")];
            tensor<int32, [4]> reshape_13_shape_0 = const()[name = tensor<string, []>("reshape_13_shape_0"), val = tensor<int32, [4]>([2, 320, 128, 128])];
            tensor<fp16, [2, 320, 128, 128]> reshape_13_cast = reshape(shape = reshape_13_shape_0, x = real_div_3_cast)[name = tensor<string, []>("reshape_13_cast")];
            tensor<fp16, [320]> add_7_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_7_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21794560)))];
            tensor<fp16, [320]> add_7_beta_0_to_fp16 = const()[name = tensor<string, []>("add_7_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21795264)))];
            tensor<fp16, []> add_7_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_7_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 128, 128]> add_7_cast = batch_norm(beta = add_7_beta_0_to_fp16, epsilon = add_7_epsilon_0_to_fp16, gamma = add_7_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_13_cast)[name = tensor<string, []>("add_7_cast")];
            tensor<fp16, [2, 320, 128, 128]> input_43_cast = silu(x = add_7_cast)[name = tensor<string, []>("input_43_cast")];
            tensor<int32, [2]> var_257 = const()[name = tensor<string, []>("op_257"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_259 = const()[name = tensor<string, []>("op_259"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_7_pad_type_0 = const()[name = tensor<string, []>("hidden_states_7_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_7_pad_0 = const()[name = tensor<string, []>("hidden_states_7_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 320, 3, 3]> down_blocks_0_resnets_1_conv2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_1_conv2_weight_to_fp16"), val = tensor<fp16, [320, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21795968)))];
            tensor<fp16, [320]> down_blocks_0_resnets_1_conv2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_1_conv2_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23639232)))];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_7_cast = conv(bias = down_blocks_0_resnets_1_conv2_bias_to_fp16, dilations = var_259, groups = var_182, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = var_257, weight = down_blocks_0_resnets_1_conv2_weight_to_fp16, x = input_43_cast)[name = tensor<string, []>("hidden_states_7_cast")];
            tensor<fp16, [2, 320, 128, 128]> input_45_cast = add(x = input_31_cast, y = hidden_states_7_cast)[name = tensor<string, []>("input_45_cast")];
            tensor<int32, [2]> var_266 = const()[name = tensor<string, []>("op_266"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> var_268 = const()[name = tensor<string, []>("op_268"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> input_47_pad_type_0 = const()[name = tensor<string, []>("input_47_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> input_47_pad_0 = const()[name = tensor<string, []>("input_47_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 320, 3, 3]> down_blocks_0_downsamplers_0_conv_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_downsamplers_0_conv_weight_to_fp16"), val = tensor<fp16, [320, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23639936)))];
            tensor<fp16, [320]> down_blocks_0_downsamplers_0_conv_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_downsamplers_0_conv_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25483200)))];
            tensor<fp16, [2, 320, 64, 64]> input_47_cast = conv(bias = down_blocks_0_downsamplers_0_conv_bias_to_fp16, dilations = var_268, groups = var_182, pad = input_47_pad_0, pad_type = input_47_pad_type_0, strides = var_266, weight = down_blocks_0_downsamplers_0_conv_weight_to_fp16, x = input_45_cast)[name = tensor<string, []>("input_47_cast")];
            tensor<bool, []> var_293 = const()[name = tensor<string, []>("op_293"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_298 = const()[name = tensor<string, []>("op_298"), val = tensor<int32, []>(1)];
            tensor<int32, [5]> reshape_16_shape_0 = const()[name = tensor<string, []>("reshape_16_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 64, 64])];
            tensor<fp16, [2, 32, 10, 64, 64]> reshape_16_cast = reshape(shape = reshape_16_shape_0, x = input_47_cast)[name = tensor<string, []>("reshape_16_cast")];
            tensor<int32, [3]> reduce_mean_12_axes_0 = const()[name = tensor<string, []>("reduce_mean_12_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_12_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_12_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_12_cast = reduce_mean(axes = reduce_mean_12_axes_0, keep_dims = reduce_mean_12_keep_dims_0, x = reshape_16_cast)[name = tensor<string, []>("reduce_mean_12_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> sub_8_cast = sub(x = reshape_16_cast, y = reduce_mean_12_cast)[name = tensor<string, []>("sub_8_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> square_4_cast = square(x = sub_8_cast)[name = tensor<string, []>("square_4_cast")];
            tensor<int32, [3]> reduce_mean_14_axes_0 = const()[name = tensor<string, []>("reduce_mean_14_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_14_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_14_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_14_cast = reduce_mean(axes = reduce_mean_14_axes_0, keep_dims = reduce_mean_14_keep_dims_0, x = square_4_cast)[name = tensor<string, []>("reduce_mean_14_cast")];
            tensor<fp16, []> add_8_y_0_to_fp16 = const()[name = tensor<string, []>("add_8_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_8_cast = add(x = reduce_mean_14_cast, y = add_8_y_0_to_fp16)[name = tensor<string, []>("add_8_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_4_cast = sqrt(x = add_8_cast)[name = tensor<string, []>("sqrt_4_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_4_cast = real_div(x = sub_8_cast, y = sqrt_4_cast)[name = tensor<string, []>("real_div_4_cast")];
            tensor<int32, [4]> reshape_17_shape_0 = const()[name = tensor<string, []>("reshape_17_shape_0"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> reshape_17_cast = reshape(shape = reshape_17_shape_0, x = real_div_4_cast)[name = tensor<string, []>("reshape_17_cast")];
            tensor<fp16, [320]> add_9_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_9_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25483904)))];
            tensor<fp16, [320]> add_9_beta_0_to_fp16 = const()[name = tensor<string, []>("add_9_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25484608)))];
            tensor<fp16, []> add_9_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_9_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 64, 64]> add_9_cast = batch_norm(beta = add_9_beta_0_to_fp16, epsilon = add_9_epsilon_0_to_fp16, gamma = add_9_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_17_cast)[name = tensor<string, []>("add_9_cast")];
            tensor<fp16, [2, 320, 64, 64]> input_51_cast = silu(x = add_9_cast)[name = tensor<string, []>("input_51_cast")];
            tensor<int32, [2]> var_321 = const()[name = tensor<string, []>("op_321"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_323 = const()[name = tensor<string, []>("op_323"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_9_pad_type_0 = const()[name = tensor<string, []>("hidden_states_9_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_9_pad_0 = const()[name = tensor<string, []>("hidden_states_9_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 320, 3, 3]> down_blocks_1_resnets_0_conv1_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [640, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25485312)))];
            tensor<fp16, [640]> down_blocks_1_resnets_0_conv1_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29171776)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_9_cast = conv(bias = down_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = var_323, groups = var_298, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = var_321, weight = down_blocks_1_resnets_0_conv1_weight_to_fp16, x = input_51_cast)[name = tensor<string, []>("hidden_states_9_cast")];
            tensor<int32, [2]> var_329 = const()[name = tensor<string, []>("op_329"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_331 = const()[name = tensor<string, []>("op_331"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_5_pad_type_0 = const()[name = tensor<string, []>("temb_5_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_5_pad_0 = const()[name = tensor<string, []>("temb_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1280, 1, 1]> down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [640, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29173120)))];
            tensor<fp16, [640]> down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30811584)))];
            tensor<fp16, [2, 640, 1, 1]> temb_5_cast = conv(bias = down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_331, groups = var_298, pad = temb_5_pad_0, pad_type = temb_5_pad_type_0, strides = var_329, weight = down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16, x = input_23_cast)[name = tensor<string, []>("temb_5_cast")];
            tensor<fp16, [2, 640, 64, 64]> input_55_cast = add(x = hidden_states_9_cast, y = temb_5_cast)[name = tensor<string, []>("input_55_cast")];
            tensor<int32, [5]> reshape_20_shape_0 = const()[name = tensor<string, []>("reshape_20_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 64, 64])];
            tensor<fp16, [2, 32, 20, 64, 64]> reshape_20_cast = reshape(shape = reshape_20_shape_0, x = input_55_cast)[name = tensor<string, []>("reshape_20_cast")];
            tensor<int32, [3]> reduce_mean_15_axes_0 = const()[name = tensor<string, []>("reduce_mean_15_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_15_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_15_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_15_cast = reduce_mean(axes = reduce_mean_15_axes_0, keep_dims = reduce_mean_15_keep_dims_0, x = reshape_20_cast)[name = tensor<string, []>("reduce_mean_15_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> sub_10_cast = sub(x = reshape_20_cast, y = reduce_mean_15_cast)[name = tensor<string, []>("sub_10_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> square_5_cast = square(x = sub_10_cast)[name = tensor<string, []>("square_5_cast")];
            tensor<int32, [3]> reduce_mean_17_axes_0 = const()[name = tensor<string, []>("reduce_mean_17_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_17_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_17_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_17_cast = reduce_mean(axes = reduce_mean_17_axes_0, keep_dims = reduce_mean_17_keep_dims_0, x = square_5_cast)[name = tensor<string, []>("reduce_mean_17_cast")];
            tensor<fp16, []> add_10_y_0_to_fp16 = const()[name = tensor<string, []>("add_10_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_10_cast = add(x = reduce_mean_17_cast, y = add_10_y_0_to_fp16)[name = tensor<string, []>("add_10_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_5_cast = sqrt(x = add_10_cast)[name = tensor<string, []>("sqrt_5_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_5_cast = real_div(x = sub_10_cast, y = sqrt_5_cast)[name = tensor<string, []>("real_div_5_cast")];
            tensor<int32, [4]> reshape_21_shape_0 = const()[name = tensor<string, []>("reshape_21_shape_0"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> reshape_21_cast = reshape(shape = reshape_21_shape_0, x = real_div_5_cast)[name = tensor<string, []>("reshape_21_cast")];
            tensor<fp16, [640]> add_11_mean_0_to_fp16 = const()[name = tensor<string, []>("add_11_mean_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30812928)))];
            tensor<fp16, [640]> add_11_variance_0_to_fp16 = const()[name = tensor<string, []>("add_11_variance_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30814272)))];
            tensor<fp16, [640]> add_11_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_11_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30815616)))];
            tensor<fp16, [640]> add_11_beta_0_to_fp16 = const()[name = tensor<string, []>("add_11_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30816960)))];
            tensor<fp16, []> add_11_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_11_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 64, 64]> add_11_cast = batch_norm(beta = add_11_beta_0_to_fp16, epsilon = add_11_epsilon_0_to_fp16, gamma = add_11_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_21_cast)[name = tensor<string, []>("add_11_cast")];
            tensor<fp16, [2, 640, 64, 64]> input_59_cast = silu(x = add_11_cast)[name = tensor<string, []>("input_59_cast")];
            tensor<int32, [2]> var_341 = const()[name = tensor<string, []>("op_341"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_343 = const()[name = tensor<string, []>("op_343"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_11_pad_type_0 = const()[name = tensor<string, []>("hidden_states_11_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_11_pad_0 = const()[name = tensor<string, []>("hidden_states_11_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 640, 3, 3]> down_blocks_1_resnets_0_conv2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [640, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30818304)))];
            tensor<fp16, [640]> down_blocks_1_resnets_0_conv2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38191168)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_11_cast = conv(bias = down_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = var_343, groups = var_298, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = var_341, weight = down_blocks_1_resnets_0_conv2_weight_to_fp16, x = input_59_cast)[name = tensor<string, []>("hidden_states_11_cast")];
            tensor<int32, [2]> var_348 = const()[name = tensor<string, []>("op_348"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_350 = const()[name = tensor<string, []>("op_350"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_1_pad_type_0 = const()[name = tensor<string, []>("x_1_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_1_pad_0 = const()[name = tensor<string, []>("x_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 320, 1, 1]> down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [640, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38192512)))];
            tensor<fp16, [640]> down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38602176)))];
            tensor<fp16, [2, 640, 64, 64]> x_1_cast = conv(bias = down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_350, groups = var_298, pad = x_1_pad_0, pad_type = x_1_pad_type_0, strides = var_348, weight = down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16, x = input_47_cast)[name = tensor<string, []>("x_1_cast")];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_13_cast = add(x = x_1_cast, y = hidden_states_11_cast)[name = tensor<string, []>("hidden_states_13_cast")];
            tensor<int32, [5]> reshape_24_shape_0 = const()[name = tensor<string, []>("reshape_24_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 64, 64])];
            tensor<fp16, [2, 32, 20, 64, 64]> reshape_24_cast = reshape(shape = reshape_24_shape_0, x = hidden_states_13_cast)[name = tensor<string, []>("reshape_24_cast")];
            tensor<int32, [3]> reduce_mean_18_axes_0 = const()[name = tensor<string, []>("reduce_mean_18_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_18_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_18_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_18_cast = reduce_mean(axes = reduce_mean_18_axes_0, keep_dims = reduce_mean_18_keep_dims_0, x = reshape_24_cast)[name = tensor<string, []>("reduce_mean_18_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> sub_12_cast = sub(x = reshape_24_cast, y = reduce_mean_18_cast)[name = tensor<string, []>("sub_12_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> square_6_cast = square(x = sub_12_cast)[name = tensor<string, []>("square_6_cast")];
            tensor<int32, [3]> reduce_mean_20_axes_0 = const()[name = tensor<string, []>("reduce_mean_20_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_20_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_20_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_20_cast = reduce_mean(axes = reduce_mean_20_axes_0, keep_dims = reduce_mean_20_keep_dims_0, x = square_6_cast)[name = tensor<string, []>("reduce_mean_20_cast")];
            tensor<fp16, []> add_12_y_0_to_fp16 = const()[name = tensor<string, []>("add_12_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_12_cast = add(x = reduce_mean_20_cast, y = add_12_y_0_to_fp16)[name = tensor<string, []>("add_12_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_6_cast = sqrt(x = add_12_cast)[name = tensor<string, []>("sqrt_6_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_6_cast = real_div(x = sub_12_cast, y = sqrt_6_cast)[name = tensor<string, []>("real_div_6_cast")];
            tensor<int32, [4]> reshape_25_shape_0 = const()[name = tensor<string, []>("reshape_25_shape_0"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> reshape_25_cast = reshape(shape = reshape_25_shape_0, x = real_div_6_cast)[name = tensor<string, []>("reshape_25_cast")];
            tensor<fp16, [640]> add_13_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_13_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38603520)))];
            tensor<fp16, [640]> add_13_beta_0_to_fp16 = const()[name = tensor<string, []>("add_13_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38604864)))];
            tensor<fp16, []> add_13_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_13_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 64, 64]> add_13_cast = batch_norm(beta = add_13_beta_0_to_fp16, epsilon = add_13_epsilon_0_to_fp16, gamma = add_13_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_25_cast)[name = tensor<string, []>("add_13_cast")];
            tensor<int32, [2]> var_372 = const()[name = tensor<string, []>("op_372"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_374 = const()[name = tensor<string, []>("op_374"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_15_pad_type_0 = const()[name = tensor<string, []>("hidden_states_15_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_15_pad_0 = const()[name = tensor<string, []>("hidden_states_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_proj_in_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38606208)))];
            tensor<fp16, [640]> down_blocks_1_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_proj_in_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39425472)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_15_cast = conv(bias = down_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = var_374, groups = var_298, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = var_372, weight = down_blocks_1_attentions_0_proj_in_weight_to_fp16, x = add_13_cast)[name = tensor<string, []>("hidden_states_15_cast")];
            tensor<int32, [4]> var_379 = const()[name = tensor<string, []>("op_379"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<fp16, [2, 640, 1, 4096]> inputs_1_cast = reshape(shape = var_379, x = hidden_states_15_cast)[name = tensor<string, []>("inputs_1_cast")];
            tensor<int32, [1]> var_389 = const()[name = tensor<string, []>("op_389"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_1_cast = reduce_mean(axes = var_389, keep_dims = var_293, x = inputs_1_cast)[name = tensor<string, []>("channels_mean_1_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_1_cast = sub(x = inputs_1_cast, y = channels_mean_1_cast)[name = tensor<string, []>("zero_mean_1_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_1_cast = mul(x = zero_mean_1_cast, y = zero_mean_1_cast)[name = tensor<string, []>("zero_mean_sq_1_cast")];
            tensor<int32, [1]> var_393 = const()[name = tensor<string, []>("op_393"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_394_cast = reduce_mean(axes = var_393, keep_dims = var_293, x = zero_mean_sq_1_cast)[name = tensor<string, []>("op_394_cast")];
            tensor<fp16, []> var_395_to_fp16 = const()[name = tensor<string, []>("op_395_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_396_cast = add(x = var_394_cast, y = var_395_to_fp16)[name = tensor<string, []>("op_396_cast")];
            tensor<fp16, []> denom_1_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_1_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_1_cast = rsqrt(epsilon = denom_1_epsilon_0_to_fp16, x = var_396_cast)[name = tensor<string, []>("denom_1_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_1_cast = mul(x = zero_mean_1_cast, y = denom_1_cast)[name = tensor<string, []>("out_1_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_400_to_fp16 = const()[name = tensor<string, []>("op_400_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39426816)))];
            tensor<fp16, [2, 640, 1, 4096]> var_401_cast = add(x = out_1_cast, y = var_400_to_fp16)[name = tensor<string, []>("op_401_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_403_to_fp16 = const()[name = tensor<string, []>("op_403_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39428160)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_17_cast = mul(x = var_401_cast, y = var_403_to_fp16)[name = tensor<string, []>("hidden_states_17_cast")];
            tensor<int32, [2]> var_410 = const()[name = tensor<string, []>("op_410"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_412 = const()[name = tensor<string, []>("op_412"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_1_pad_type_0 = const()[name = tensor<string, []>("q_1_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_1_pad_0 = const()[name = tensor<string, []>("q_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39429504)))];
            tensor<fp16, [2, 640, 1, 4096]> q_1_cast = conv(dilations = var_412, groups = var_298, pad = q_1_pad_0, pad_type = q_1_pad_type_0, strides = var_410, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_17_cast)[name = tensor<string, []>("q_1_cast")];
            tensor<int32, [2]> var_416 = const()[name = tensor<string, []>("op_416"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_418 = const()[name = tensor<string, []>("op_418"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_1_pad_type_0 = const()[name = tensor<string, []>("k_1_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_1_pad_0 = const()[name = tensor<string, []>("k_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40248768)))];
            tensor<fp16, [2, 640, 1, 4096]> k_1_cast = conv(dilations = var_418, groups = var_298, pad = k_1_pad_0, pad_type = k_1_pad_type_0, strides = var_416, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_17_cast)[name = tensor<string, []>("k_1_cast")];
            tensor<int32, [2]> var_422 = const()[name = tensor<string, []>("op_422"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_424 = const()[name = tensor<string, []>("op_424"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_1_pad_type_0 = const()[name = tensor<string, []>("v_1_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_1_pad_0 = const()[name = tensor<string, []>("v_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41068032)))];
            tensor<fp16, [2, 640, 1, 4096]> v_1_cast = conv(dilations = var_424, groups = var_298, pad = v_1_pad_0, pad_type = v_1_pad_type_0, strides = var_422, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_17_cast)[name = tensor<string, []>("v_1_cast")];
            tensor<int32, [4]> var_428_begin_0 = const()[name = tensor<string, []>("op_428_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_428_end_0 = const()[name = tensor<string, []>("op_428_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_428_end_mask_0 = const()[name = tensor<string, []>("op_428_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_428_cast = slice_by_index(begin = var_428_begin_0, end = var_428_end_0, end_mask = var_428_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_428_cast")];
            tensor<int32, [4]> var_432_begin_0 = const()[name = tensor<string, []>("op_432_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_432_end_0 = const()[name = tensor<string, []>("op_432_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_432_end_mask_0 = const()[name = tensor<string, []>("op_432_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_432_cast = slice_by_index(begin = var_432_begin_0, end = var_432_end_0, end_mask = var_432_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_432_cast")];
            tensor<int32, [4]> var_436_begin_0 = const()[name = tensor<string, []>("op_436_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_436_end_0 = const()[name = tensor<string, []>("op_436_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_436_end_mask_0 = const()[name = tensor<string, []>("op_436_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_436_cast = slice_by_index(begin = var_436_begin_0, end = var_436_end_0, end_mask = var_436_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_436_cast")];
            tensor<int32, [4]> var_440_begin_0 = const()[name = tensor<string, []>("op_440_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_440_end_0 = const()[name = tensor<string, []>("op_440_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_440_end_mask_0 = const()[name = tensor<string, []>("op_440_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_440_cast = slice_by_index(begin = var_440_begin_0, end = var_440_end_0, end_mask = var_440_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_440_cast")];
            tensor<int32, [4]> var_444_begin_0 = const()[name = tensor<string, []>("op_444_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_444_end_0 = const()[name = tensor<string, []>("op_444_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_444_end_mask_0 = const()[name = tensor<string, []>("op_444_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_444_cast = slice_by_index(begin = var_444_begin_0, end = var_444_end_0, end_mask = var_444_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_444_cast")];
            tensor<int32, [4]> var_448_begin_0 = const()[name = tensor<string, []>("op_448_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_448_end_0 = const()[name = tensor<string, []>("op_448_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_448_end_mask_0 = const()[name = tensor<string, []>("op_448_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_448_cast = slice_by_index(begin = var_448_begin_0, end = var_448_end_0, end_mask = var_448_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_448_cast")];
            tensor<int32, [4]> var_452_begin_0 = const()[name = tensor<string, []>("op_452_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_452_end_0 = const()[name = tensor<string, []>("op_452_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_452_end_mask_0 = const()[name = tensor<string, []>("op_452_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_452_cast = slice_by_index(begin = var_452_begin_0, end = var_452_end_0, end_mask = var_452_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_452_cast")];
            tensor<int32, [4]> var_456_begin_0 = const()[name = tensor<string, []>("op_456_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_456_end_0 = const()[name = tensor<string, []>("op_456_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_456_end_mask_0 = const()[name = tensor<string, []>("op_456_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_456_cast = slice_by_index(begin = var_456_begin_0, end = var_456_end_0, end_mask = var_456_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_456_cast")];
            tensor<int32, [4]> var_460_begin_0 = const()[name = tensor<string, []>("op_460_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_460_end_0 = const()[name = tensor<string, []>("op_460_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_460_end_mask_0 = const()[name = tensor<string, []>("op_460_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_460_cast = slice_by_index(begin = var_460_begin_0, end = var_460_end_0, end_mask = var_460_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_460_cast")];
            tensor<int32, [4]> var_464_begin_0 = const()[name = tensor<string, []>("op_464_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_464_end_0 = const()[name = tensor<string, []>("op_464_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_464_end_mask_0 = const()[name = tensor<string, []>("op_464_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_464_cast = slice_by_index(begin = var_464_begin_0, end = var_464_end_0, end_mask = var_464_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_464_cast")];
            tensor<int32, [4]> k_3_perm_0 = const()[name = tensor<string, []>("k_3_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_471_begin_0 = const()[name = tensor<string, []>("op_471_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_471_end_0 = const()[name = tensor<string, []>("op_471_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_471_end_mask_0 = const()[name = tensor<string, []>("op_471_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 640]> transpose_139 = transpose(perm = k_3_perm_0, x = k_1_cast)[name = tensor<string, []>("transpose_139")];
            tensor<fp16, [2, 4096, 1, 64]> var_471_cast = slice_by_index(begin = var_471_begin_0, end = var_471_end_0, end_mask = var_471_end_mask_0, x = transpose_139)[name = tensor<string, []>("op_471_cast")];
            tensor<int32, [4]> var_475_begin_0 = const()[name = tensor<string, []>("op_475_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_475_end_0 = const()[name = tensor<string, []>("op_475_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_475_end_mask_0 = const()[name = tensor<string, []>("op_475_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_475_cast = slice_by_index(begin = var_475_begin_0, end = var_475_end_0, end_mask = var_475_end_mask_0, x = transpose_139)[name = tensor<string, []>("op_475_cast")];
            tensor<int32, [4]> var_479_begin_0 = const()[name = tensor<string, []>("op_479_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_479_end_0 = const()[name = tensor<string, []>("op_479_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_479_end_mask_0 = const()[name = tensor<string, []>("op_479_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_479_cast = slice_by_index(begin = var_479_begin_0, end = var_479_end_0, end_mask = var_479_end_mask_0, x = transpose_139)[name = tensor<string, []>("op_479_cast")];
            tensor<int32, [4]> var_483_begin_0 = const()[name = tensor<string, []>("op_483_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_483_end_0 = const()[name = tensor<string, []>("op_483_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_483_end_mask_0 = const()[name = tensor<string, []>("op_483_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_483_cast = slice_by_index(begin = var_483_begin_0, end = var_483_end_0, end_mask = var_483_end_mask_0, x = transpose_139)[name = tensor<string, []>("op_483_cast")];
            tensor<int32, [4]> var_487_begin_0 = const()[name = tensor<string, []>("op_487_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_487_end_0 = const()[name = tensor<string, []>("op_487_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_487_end_mask_0 = const()[name = tensor<string, []>("op_487_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_487_cast = slice_by_index(begin = var_487_begin_0, end = var_487_end_0, end_mask = var_487_end_mask_0, x = transpose_139)[name = tensor<string, []>("op_487_cast")];
            tensor<int32, [4]> var_491_begin_0 = const()[name = tensor<string, []>("op_491_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_491_end_0 = const()[name = tensor<string, []>("op_491_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 384])];
            tensor<bool, [4]> var_491_end_mask_0 = const()[name = tensor<string, []>("op_491_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_491_cast = slice_by_index(begin = var_491_begin_0, end = var_491_end_0, end_mask = var_491_end_mask_0, x = transpose_139)[name = tensor<string, []>("op_491_cast")];
            tensor<int32, [4]> var_495_begin_0 = const()[name = tensor<string, []>("op_495_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_495_end_0 = const()[name = tensor<string, []>("op_495_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 448])];
            tensor<bool, [4]> var_495_end_mask_0 = const()[name = tensor<string, []>("op_495_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_495_cast = slice_by_index(begin = var_495_begin_0, end = var_495_end_0, end_mask = var_495_end_mask_0, x = transpose_139)[name = tensor<string, []>("op_495_cast")];
            tensor<int32, [4]> var_499_begin_0 = const()[name = tensor<string, []>("op_499_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_499_end_0 = const()[name = tensor<string, []>("op_499_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 512])];
            tensor<bool, [4]> var_499_end_mask_0 = const()[name = tensor<string, []>("op_499_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_499_cast = slice_by_index(begin = var_499_begin_0, end = var_499_end_0, end_mask = var_499_end_mask_0, x = transpose_139)[name = tensor<string, []>("op_499_cast")];
            tensor<int32, [4]> var_503_begin_0 = const()[name = tensor<string, []>("op_503_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_503_end_0 = const()[name = tensor<string, []>("op_503_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 576])];
            tensor<bool, [4]> var_503_end_mask_0 = const()[name = tensor<string, []>("op_503_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_503_cast = slice_by_index(begin = var_503_begin_0, end = var_503_end_0, end_mask = var_503_end_mask_0, x = transpose_139)[name = tensor<string, []>("op_503_cast")];
            tensor<int32, [4]> var_507_begin_0 = const()[name = tensor<string, []>("op_507_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_507_end_0 = const()[name = tensor<string, []>("op_507_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 640])];
            tensor<bool, [4]> var_507_end_mask_0 = const()[name = tensor<string, []>("op_507_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_507_cast = slice_by_index(begin = var_507_begin_0, end = var_507_end_0, end_mask = var_507_end_mask_0, x = transpose_139)[name = tensor<string, []>("op_507_cast")];
            tensor<int32, [4]> var_509_begin_0 = const()[name = tensor<string, []>("op_509_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_509_end_0 = const()[name = tensor<string, []>("op_509_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_509_end_mask_0 = const()[name = tensor<string, []>("op_509_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_509_cast = slice_by_index(begin = var_509_begin_0, end = var_509_end_0, end_mask = var_509_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_509_cast")];
            tensor<int32, [4]> var_513_begin_0 = const()[name = tensor<string, []>("op_513_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_513_end_0 = const()[name = tensor<string, []>("op_513_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_513_end_mask_0 = const()[name = tensor<string, []>("op_513_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_513_cast = slice_by_index(begin = var_513_begin_0, end = var_513_end_0, end_mask = var_513_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_513_cast")];
            tensor<int32, [4]> var_517_begin_0 = const()[name = tensor<string, []>("op_517_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_517_end_0 = const()[name = tensor<string, []>("op_517_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_517_end_mask_0 = const()[name = tensor<string, []>("op_517_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_517_cast = slice_by_index(begin = var_517_begin_0, end = var_517_end_0, end_mask = var_517_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_517_cast")];
            tensor<int32, [4]> var_521_begin_0 = const()[name = tensor<string, []>("op_521_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_521_end_0 = const()[name = tensor<string, []>("op_521_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_521_end_mask_0 = const()[name = tensor<string, []>("op_521_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_521_cast = slice_by_index(begin = var_521_begin_0, end = var_521_end_0, end_mask = var_521_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_521_cast")];
            tensor<int32, [4]> var_525_begin_0 = const()[name = tensor<string, []>("op_525_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_525_end_0 = const()[name = tensor<string, []>("op_525_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_525_end_mask_0 = const()[name = tensor<string, []>("op_525_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_525_cast = slice_by_index(begin = var_525_begin_0, end = var_525_end_0, end_mask = var_525_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_525_cast")];
            tensor<int32, [4]> var_529_begin_0 = const()[name = tensor<string, []>("op_529_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_529_end_0 = const()[name = tensor<string, []>("op_529_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_529_end_mask_0 = const()[name = tensor<string, []>("op_529_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_529_cast = slice_by_index(begin = var_529_begin_0, end = var_529_end_0, end_mask = var_529_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_529_cast")];
            tensor<int32, [4]> var_533_begin_0 = const()[name = tensor<string, []>("op_533_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_533_end_0 = const()[name = tensor<string, []>("op_533_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_533_end_mask_0 = const()[name = tensor<string, []>("op_533_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_533_cast = slice_by_index(begin = var_533_begin_0, end = var_533_end_0, end_mask = var_533_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_533_cast")];
            tensor<int32, [4]> var_537_begin_0 = const()[name = tensor<string, []>("op_537_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_537_end_0 = const()[name = tensor<string, []>("op_537_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_537_end_mask_0 = const()[name = tensor<string, []>("op_537_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_537_cast = slice_by_index(begin = var_537_begin_0, end = var_537_end_0, end_mask = var_537_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_537_cast")];
            tensor<int32, [4]> var_541_begin_0 = const()[name = tensor<string, []>("op_541_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_541_end_0 = const()[name = tensor<string, []>("op_541_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_541_end_mask_0 = const()[name = tensor<string, []>("op_541_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_541_cast = slice_by_index(begin = var_541_begin_0, end = var_541_end_0, end_mask = var_541_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_541_cast")];
            tensor<int32, [4]> var_545_begin_0 = const()[name = tensor<string, []>("op_545_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_545_end_0 = const()[name = tensor<string, []>("op_545_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_545_end_mask_0 = const()[name = tensor<string, []>("op_545_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_545_cast = slice_by_index(begin = var_545_begin_0, end = var_545_end_0, end_mask = var_545_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_545_cast")];
            tensor<string, []> var_549_equation_0 = const()[name = tensor<string, []>("op_549_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_549_cast = einsum(equation = var_549_equation_0, values = (var_471_cast, var_428_cast))[name = tensor<string, []>("op_549_cast")];
            tensor<fp16, []> var_550_to_fp16 = const()[name = tensor<string, []>("op_550_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_1_cast = mul(x = var_549_cast, y = var_550_to_fp16)[name = tensor<string, []>("aw_1_cast")];
            tensor<string, []> var_553_equation_0 = const()[name = tensor<string, []>("op_553_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_553_cast = einsum(equation = var_553_equation_0, values = (var_475_cast, var_432_cast))[name = tensor<string, []>("op_553_cast")];
            tensor<fp16, []> var_554_to_fp16 = const()[name = tensor<string, []>("op_554_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_3_cast = mul(x = var_553_cast, y = var_554_to_fp16)[name = tensor<string, []>("aw_3_cast")];
            tensor<string, []> var_557_equation_0 = const()[name = tensor<string, []>("op_557_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_557_cast = einsum(equation = var_557_equation_0, values = (var_479_cast, var_436_cast))[name = tensor<string, []>("op_557_cast")];
            tensor<fp16, []> var_558_to_fp16 = const()[name = tensor<string, []>("op_558_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5_cast = mul(x = var_557_cast, y = var_558_to_fp16)[name = tensor<string, []>("aw_5_cast")];
            tensor<string, []> var_561_equation_0 = const()[name = tensor<string, []>("op_561_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_561_cast = einsum(equation = var_561_equation_0, values = (var_483_cast, var_440_cast))[name = tensor<string, []>("op_561_cast")];
            tensor<fp16, []> var_562_to_fp16 = const()[name = tensor<string, []>("op_562_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_7_cast = mul(x = var_561_cast, y = var_562_to_fp16)[name = tensor<string, []>("aw_7_cast")];
            tensor<string, []> var_565_equation_0 = const()[name = tensor<string, []>("op_565_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_565_cast = einsum(equation = var_565_equation_0, values = (var_487_cast, var_444_cast))[name = tensor<string, []>("op_565_cast")];
            tensor<fp16, []> var_566_to_fp16 = const()[name = tensor<string, []>("op_566_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_9_cast = mul(x = var_565_cast, y = var_566_to_fp16)[name = tensor<string, []>("aw_9_cast")];
            tensor<string, []> var_569_equation_0 = const()[name = tensor<string, []>("op_569_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_569_cast = einsum(equation = var_569_equation_0, values = (var_491_cast, var_448_cast))[name = tensor<string, []>("op_569_cast")];
            tensor<fp16, []> var_570_to_fp16 = const()[name = tensor<string, []>("op_570_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_11_cast = mul(x = var_569_cast, y = var_570_to_fp16)[name = tensor<string, []>("aw_11_cast")];
            tensor<string, []> var_573_equation_0 = const()[name = tensor<string, []>("op_573_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_573_cast = einsum(equation = var_573_equation_0, values = (var_495_cast, var_452_cast))[name = tensor<string, []>("op_573_cast")];
            tensor<fp16, []> var_574_to_fp16 = const()[name = tensor<string, []>("op_574_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_13_cast = mul(x = var_573_cast, y = var_574_to_fp16)[name = tensor<string, []>("aw_13_cast")];
            tensor<string, []> var_577_equation_0 = const()[name = tensor<string, []>("op_577_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_577_cast = einsum(equation = var_577_equation_0, values = (var_499_cast, var_456_cast))[name = tensor<string, []>("op_577_cast")];
            tensor<fp16, []> var_578_to_fp16 = const()[name = tensor<string, []>("op_578_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_15_cast = mul(x = var_577_cast, y = var_578_to_fp16)[name = tensor<string, []>("aw_15_cast")];
            tensor<string, []> var_581_equation_0 = const()[name = tensor<string, []>("op_581_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_581_cast = einsum(equation = var_581_equation_0, values = (var_503_cast, var_460_cast))[name = tensor<string, []>("op_581_cast")];
            tensor<fp16, []> var_582_to_fp16 = const()[name = tensor<string, []>("op_582_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_17_cast = mul(x = var_581_cast, y = var_582_to_fp16)[name = tensor<string, []>("aw_17_cast")];
            tensor<string, []> var_585_equation_0 = const()[name = tensor<string, []>("op_585_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_585_cast = einsum(equation = var_585_equation_0, values = (var_507_cast, var_464_cast))[name = tensor<string, []>("op_585_cast")];
            tensor<fp16, []> var_586_to_fp16 = const()[name = tensor<string, []>("op_586_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_19_cast = mul(x = var_585_cast, y = var_586_to_fp16)[name = tensor<string, []>("aw_19_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_588_cast = softmax(axis = var_298, x = aw_1_cast)[name = tensor<string, []>("op_588_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_589_cast = softmax(axis = var_298, x = aw_3_cast)[name = tensor<string, []>("op_589_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_590_cast = softmax(axis = var_298, x = aw_5_cast)[name = tensor<string, []>("op_590_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_591_cast = softmax(axis = var_298, x = aw_7_cast)[name = tensor<string, []>("op_591_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_592_cast = softmax(axis = var_298, x = aw_9_cast)[name = tensor<string, []>("op_592_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_593_cast = softmax(axis = var_298, x = aw_11_cast)[name = tensor<string, []>("op_593_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_594_cast = softmax(axis = var_298, x = aw_13_cast)[name = tensor<string, []>("op_594_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_595_cast = softmax(axis = var_298, x = aw_15_cast)[name = tensor<string, []>("op_595_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_596_cast = softmax(axis = var_298, x = aw_17_cast)[name = tensor<string, []>("op_596_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_597_cast = softmax(axis = var_298, x = aw_19_cast)[name = tensor<string, []>("op_597_cast")];
            tensor<string, []> var_599_equation_0 = const()[name = tensor<string, []>("op_599_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_599_cast = einsum(equation = var_599_equation_0, values = (var_509_cast, var_588_cast))[name = tensor<string, []>("op_599_cast")];
            tensor<string, []> var_601_equation_0 = const()[name = tensor<string, []>("op_601_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_601_cast = einsum(equation = var_601_equation_0, values = (var_513_cast, var_589_cast))[name = tensor<string, []>("op_601_cast")];
            tensor<string, []> var_603_equation_0 = const()[name = tensor<string, []>("op_603_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_603_cast = einsum(equation = var_603_equation_0, values = (var_517_cast, var_590_cast))[name = tensor<string, []>("op_603_cast")];
            tensor<string, []> var_605_equation_0 = const()[name = tensor<string, []>("op_605_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_605_cast = einsum(equation = var_605_equation_0, values = (var_521_cast, var_591_cast))[name = tensor<string, []>("op_605_cast")];
            tensor<string, []> var_607_equation_0 = const()[name = tensor<string, []>("op_607_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_607_cast = einsum(equation = var_607_equation_0, values = (var_525_cast, var_592_cast))[name = tensor<string, []>("op_607_cast")];
            tensor<string, []> var_609_equation_0 = const()[name = tensor<string, []>("op_609_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_609_cast = einsum(equation = var_609_equation_0, values = (var_529_cast, var_593_cast))[name = tensor<string, []>("op_609_cast")];
            tensor<string, []> var_611_equation_0 = const()[name = tensor<string, []>("op_611_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_611_cast = einsum(equation = var_611_equation_0, values = (var_533_cast, var_594_cast))[name = tensor<string, []>("op_611_cast")];
            tensor<string, []> var_613_equation_0 = const()[name = tensor<string, []>("op_613_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_613_cast = einsum(equation = var_613_equation_0, values = (var_537_cast, var_595_cast))[name = tensor<string, []>("op_613_cast")];
            tensor<string, []> var_615_equation_0 = const()[name = tensor<string, []>("op_615_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_615_cast = einsum(equation = var_615_equation_0, values = (var_541_cast, var_596_cast))[name = tensor<string, []>("op_615_cast")];
            tensor<string, []> var_617_equation_0 = const()[name = tensor<string, []>("op_617_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_617_cast = einsum(equation = var_617_equation_0, values = (var_545_cast, var_597_cast))[name = tensor<string, []>("op_617_cast")];
            tensor<bool, []> input_63_interleave_0 = const()[name = tensor<string, []>("input_63_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_63_cast = concat(axis = var_298, interleave = input_63_interleave_0, values = (var_599_cast, var_601_cast, var_603_cast, var_605_cast, var_607_cast, var_609_cast, var_611_cast, var_613_cast, var_615_cast, var_617_cast))[name = tensor<string, []>("input_63_cast")];
            tensor<int32, [2]> var_623 = const()[name = tensor<string, []>("op_623"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_625 = const()[name = tensor<string, []>("op_625"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_627_pad_type_0 = const()[name = tensor<string, []>("op_627_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_627_pad_0 = const()[name = tensor<string, []>("op_627_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41887296)))];
            tensor<fp16, [640]> down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42706560)))];
            tensor<fp16, [2, 640, 1, 4096]> var_627_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_625, groups = var_298, pad = var_627_pad_0, pad_type = var_627_pad_type_0, strides = var_623, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_63_cast)[name = tensor<string, []>("op_627_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_3_cast = add(x = var_627_cast, y = inputs_1_cast)[name = tensor<string, []>("inputs_3_cast")];
            tensor<int32, [1]> var_631 = const()[name = tensor<string, []>("op_631"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_3_cast = reduce_mean(axes = var_631, keep_dims = var_293, x = inputs_3_cast)[name = tensor<string, []>("channels_mean_3_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_3_cast = sub(x = inputs_3_cast, y = channels_mean_3_cast)[name = tensor<string, []>("zero_mean_3_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_3_cast = mul(x = zero_mean_3_cast, y = zero_mean_3_cast)[name = tensor<string, []>("zero_mean_sq_3_cast")];
            tensor<int32, [1]> var_635 = const()[name = tensor<string, []>("op_635"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_636_cast = reduce_mean(axes = var_635, keep_dims = var_293, x = zero_mean_sq_3_cast)[name = tensor<string, []>("op_636_cast")];
            tensor<fp16, []> var_637_to_fp16 = const()[name = tensor<string, []>("op_637_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_638_cast = add(x = var_636_cast, y = var_637_to_fp16)[name = tensor<string, []>("op_638_cast")];
            tensor<fp16, []> denom_3_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_3_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_3_cast = rsqrt(epsilon = denom_3_epsilon_0_to_fp16, x = var_638_cast)[name = tensor<string, []>("denom_3_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_3_cast = mul(x = zero_mean_3_cast, y = denom_3_cast)[name = tensor<string, []>("out_3_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_642_to_fp16 = const()[name = tensor<string, []>("op_642_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42707904)))];
            tensor<fp16, [2, 640, 1, 4096]> var_643_cast = add(x = out_3_cast, y = var_642_to_fp16)[name = tensor<string, []>("op_643_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_645_to_fp16 = const()[name = tensor<string, []>("op_645_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42709248)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_19_cast = mul(x = var_643_cast, y = var_645_to_fp16)[name = tensor<string, []>("hidden_states_19_cast")];
            tensor<int32, [2]> var_652 = const()[name = tensor<string, []>("op_652"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_654 = const()[name = tensor<string, []>("op_654"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_3_pad_type_0 = const()[name = tensor<string, []>("q_3_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_3_pad_0 = const()[name = tensor<string, []>("q_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42710592)))];
            tensor<fp16, [2, 640, 1, 4096]> q_3_cast = conv(dilations = var_654, groups = var_298, pad = q_3_pad_0, pad_type = q_3_pad_type_0, strides = var_652, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_19_cast)[name = tensor<string, []>("q_3_cast")];
            tensor<int32, [2]> var_658 = const()[name = tensor<string, []>("op_658"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_660 = const()[name = tensor<string, []>("op_660"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_5_pad_type_0 = const()[name = tensor<string, []>("k_5_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_5_pad_0 = const()[name = tensor<string, []>("k_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43529856)))];
            tensor<fp16, [2, 640, 1, 77]> k_5_cast = conv(dilations = var_660, groups = var_298, pad = k_5_pad_0, pad_type = k_5_pad_type_0, strides = var_658, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_5_cast")];
            tensor<int32, [2]> var_664 = const()[name = tensor<string, []>("op_664"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_666 = const()[name = tensor<string, []>("op_666"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_3_pad_type_0 = const()[name = tensor<string, []>("v_3_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_3_pad_0 = const()[name = tensor<string, []>("v_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46151360)))];
            tensor<fp16, [2, 640, 1, 77]> v_3_cast = conv(dilations = var_666, groups = var_298, pad = v_3_pad_0, pad_type = v_3_pad_type_0, strides = var_664, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_3_cast")];
            tensor<int32, [4]> var_670_begin_0 = const()[name = tensor<string, []>("op_670_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_670_end_0 = const()[name = tensor<string, []>("op_670_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_670_end_mask_0 = const()[name = tensor<string, []>("op_670_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_670_cast = slice_by_index(begin = var_670_begin_0, end = var_670_end_0, end_mask = var_670_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_670_cast")];
            tensor<int32, [4]> var_674_begin_0 = const()[name = tensor<string, []>("op_674_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_674_end_0 = const()[name = tensor<string, []>("op_674_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_674_end_mask_0 = const()[name = tensor<string, []>("op_674_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_674_cast = slice_by_index(begin = var_674_begin_0, end = var_674_end_0, end_mask = var_674_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_674_cast")];
            tensor<int32, [4]> var_678_begin_0 = const()[name = tensor<string, []>("op_678_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_678_end_0 = const()[name = tensor<string, []>("op_678_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_678_end_mask_0 = const()[name = tensor<string, []>("op_678_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_678_cast = slice_by_index(begin = var_678_begin_0, end = var_678_end_0, end_mask = var_678_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_678_cast")];
            tensor<int32, [4]> var_682_begin_0 = const()[name = tensor<string, []>("op_682_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_682_end_0 = const()[name = tensor<string, []>("op_682_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_682_end_mask_0 = const()[name = tensor<string, []>("op_682_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_682_cast = slice_by_index(begin = var_682_begin_0, end = var_682_end_0, end_mask = var_682_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_682_cast")];
            tensor<int32, [4]> var_686_begin_0 = const()[name = tensor<string, []>("op_686_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_686_end_0 = const()[name = tensor<string, []>("op_686_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_686_end_mask_0 = const()[name = tensor<string, []>("op_686_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_686_cast = slice_by_index(begin = var_686_begin_0, end = var_686_end_0, end_mask = var_686_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_686_cast")];
            tensor<int32, [4]> var_690_begin_0 = const()[name = tensor<string, []>("op_690_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_690_end_0 = const()[name = tensor<string, []>("op_690_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_690_end_mask_0 = const()[name = tensor<string, []>("op_690_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_690_cast = slice_by_index(begin = var_690_begin_0, end = var_690_end_0, end_mask = var_690_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_690_cast")];
            tensor<int32, [4]> var_694_begin_0 = const()[name = tensor<string, []>("op_694_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_694_end_0 = const()[name = tensor<string, []>("op_694_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_694_end_mask_0 = const()[name = tensor<string, []>("op_694_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_694_cast = slice_by_index(begin = var_694_begin_0, end = var_694_end_0, end_mask = var_694_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_694_cast")];
            tensor<int32, [4]> var_698_begin_0 = const()[name = tensor<string, []>("op_698_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_698_end_0 = const()[name = tensor<string, []>("op_698_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_698_end_mask_0 = const()[name = tensor<string, []>("op_698_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_698_cast = slice_by_index(begin = var_698_begin_0, end = var_698_end_0, end_mask = var_698_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_698_cast")];
            tensor<int32, [4]> var_702_begin_0 = const()[name = tensor<string, []>("op_702_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_702_end_0 = const()[name = tensor<string, []>("op_702_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_702_end_mask_0 = const()[name = tensor<string, []>("op_702_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_702_cast = slice_by_index(begin = var_702_begin_0, end = var_702_end_0, end_mask = var_702_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_702_cast")];
            tensor<int32, [4]> var_706_begin_0 = const()[name = tensor<string, []>("op_706_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_706_end_0 = const()[name = tensor<string, []>("op_706_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_706_end_mask_0 = const()[name = tensor<string, []>("op_706_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_706_cast = slice_by_index(begin = var_706_begin_0, end = var_706_end_0, end_mask = var_706_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_706_cast")];
            tensor<int32, [4]> k_7_perm_0 = const()[name = tensor<string, []>("k_7_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_713_begin_0 = const()[name = tensor<string, []>("op_713_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_713_end_0 = const()[name = tensor<string, []>("op_713_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_713_end_mask_0 = const()[name = tensor<string, []>("op_713_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> transpose_138 = transpose(perm = k_7_perm_0, x = k_5_cast)[name = tensor<string, []>("transpose_138")];
            tensor<fp16, [2, 77, 1, 64]> var_713_cast = slice_by_index(begin = var_713_begin_0, end = var_713_end_0, end_mask = var_713_end_mask_0, x = transpose_138)[name = tensor<string, []>("op_713_cast")];
            tensor<int32, [4]> var_717_begin_0 = const()[name = tensor<string, []>("op_717_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_717_end_0 = const()[name = tensor<string, []>("op_717_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_717_end_mask_0 = const()[name = tensor<string, []>("op_717_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_717_cast = slice_by_index(begin = var_717_begin_0, end = var_717_end_0, end_mask = var_717_end_mask_0, x = transpose_138)[name = tensor<string, []>("op_717_cast")];
            tensor<int32, [4]> var_721_begin_0 = const()[name = tensor<string, []>("op_721_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_721_end_0 = const()[name = tensor<string, []>("op_721_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_721_end_mask_0 = const()[name = tensor<string, []>("op_721_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_721_cast = slice_by_index(begin = var_721_begin_0, end = var_721_end_0, end_mask = var_721_end_mask_0, x = transpose_138)[name = tensor<string, []>("op_721_cast")];
            tensor<int32, [4]> var_725_begin_0 = const()[name = tensor<string, []>("op_725_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_725_end_0 = const()[name = tensor<string, []>("op_725_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_725_end_mask_0 = const()[name = tensor<string, []>("op_725_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_725_cast = slice_by_index(begin = var_725_begin_0, end = var_725_end_0, end_mask = var_725_end_mask_0, x = transpose_138)[name = tensor<string, []>("op_725_cast")];
            tensor<int32, [4]> var_729_begin_0 = const()[name = tensor<string, []>("op_729_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_729_end_0 = const()[name = tensor<string, []>("op_729_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_729_end_mask_0 = const()[name = tensor<string, []>("op_729_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_729_cast = slice_by_index(begin = var_729_begin_0, end = var_729_end_0, end_mask = var_729_end_mask_0, x = transpose_138)[name = tensor<string, []>("op_729_cast")];
            tensor<int32, [4]> var_733_begin_0 = const()[name = tensor<string, []>("op_733_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_733_end_0 = const()[name = tensor<string, []>("op_733_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_733_end_mask_0 = const()[name = tensor<string, []>("op_733_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_733_cast = slice_by_index(begin = var_733_begin_0, end = var_733_end_0, end_mask = var_733_end_mask_0, x = transpose_138)[name = tensor<string, []>("op_733_cast")];
            tensor<int32, [4]> var_737_begin_0 = const()[name = tensor<string, []>("op_737_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_737_end_0 = const()[name = tensor<string, []>("op_737_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_737_end_mask_0 = const()[name = tensor<string, []>("op_737_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_737_cast = slice_by_index(begin = var_737_begin_0, end = var_737_end_0, end_mask = var_737_end_mask_0, x = transpose_138)[name = tensor<string, []>("op_737_cast")];
            tensor<int32, [4]> var_741_begin_0 = const()[name = tensor<string, []>("op_741_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_741_end_0 = const()[name = tensor<string, []>("op_741_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_741_end_mask_0 = const()[name = tensor<string, []>("op_741_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_741_cast = slice_by_index(begin = var_741_begin_0, end = var_741_end_0, end_mask = var_741_end_mask_0, x = transpose_138)[name = tensor<string, []>("op_741_cast")];
            tensor<int32, [4]> var_745_begin_0 = const()[name = tensor<string, []>("op_745_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_745_end_0 = const()[name = tensor<string, []>("op_745_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_745_end_mask_0 = const()[name = tensor<string, []>("op_745_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_745_cast = slice_by_index(begin = var_745_begin_0, end = var_745_end_0, end_mask = var_745_end_mask_0, x = transpose_138)[name = tensor<string, []>("op_745_cast")];
            tensor<int32, [4]> var_749_begin_0 = const()[name = tensor<string, []>("op_749_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_749_end_0 = const()[name = tensor<string, []>("op_749_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_749_end_mask_0 = const()[name = tensor<string, []>("op_749_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_749_cast = slice_by_index(begin = var_749_begin_0, end = var_749_end_0, end_mask = var_749_end_mask_0, x = transpose_138)[name = tensor<string, []>("op_749_cast")];
            tensor<int32, [4]> var_751_begin_0 = const()[name = tensor<string, []>("op_751_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_751_end_0 = const()[name = tensor<string, []>("op_751_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_751_end_mask_0 = const()[name = tensor<string, []>("op_751_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_751_cast = slice_by_index(begin = var_751_begin_0, end = var_751_end_0, end_mask = var_751_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_751_cast")];
            tensor<int32, [4]> var_755_begin_0 = const()[name = tensor<string, []>("op_755_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_755_end_0 = const()[name = tensor<string, []>("op_755_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_755_end_mask_0 = const()[name = tensor<string, []>("op_755_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_755_cast = slice_by_index(begin = var_755_begin_0, end = var_755_end_0, end_mask = var_755_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_755_cast")];
            tensor<int32, [4]> var_759_begin_0 = const()[name = tensor<string, []>("op_759_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_759_end_0 = const()[name = tensor<string, []>("op_759_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_759_end_mask_0 = const()[name = tensor<string, []>("op_759_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_759_cast = slice_by_index(begin = var_759_begin_0, end = var_759_end_0, end_mask = var_759_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_759_cast")];
            tensor<int32, [4]> var_763_begin_0 = const()[name = tensor<string, []>("op_763_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_763_end_0 = const()[name = tensor<string, []>("op_763_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_763_end_mask_0 = const()[name = tensor<string, []>("op_763_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_763_cast = slice_by_index(begin = var_763_begin_0, end = var_763_end_0, end_mask = var_763_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_763_cast")];
            tensor<int32, [4]> var_767_begin_0 = const()[name = tensor<string, []>("op_767_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_767_end_0 = const()[name = tensor<string, []>("op_767_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_767_end_mask_0 = const()[name = tensor<string, []>("op_767_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_767_cast = slice_by_index(begin = var_767_begin_0, end = var_767_end_0, end_mask = var_767_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_767_cast")];
            tensor<int32, [4]> var_771_begin_0 = const()[name = tensor<string, []>("op_771_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_771_end_0 = const()[name = tensor<string, []>("op_771_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_771_end_mask_0 = const()[name = tensor<string, []>("op_771_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_771_cast = slice_by_index(begin = var_771_begin_0, end = var_771_end_0, end_mask = var_771_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_771_cast")];
            tensor<int32, [4]> var_775_begin_0 = const()[name = tensor<string, []>("op_775_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_775_end_0 = const()[name = tensor<string, []>("op_775_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_775_end_mask_0 = const()[name = tensor<string, []>("op_775_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_775_cast = slice_by_index(begin = var_775_begin_0, end = var_775_end_0, end_mask = var_775_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_775_cast")];
            tensor<int32, [4]> var_779_begin_0 = const()[name = tensor<string, []>("op_779_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_779_end_0 = const()[name = tensor<string, []>("op_779_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_779_end_mask_0 = const()[name = tensor<string, []>("op_779_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_779_cast = slice_by_index(begin = var_779_begin_0, end = var_779_end_0, end_mask = var_779_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_779_cast")];
            tensor<int32, [4]> var_783_begin_0 = const()[name = tensor<string, []>("op_783_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_783_end_0 = const()[name = tensor<string, []>("op_783_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_783_end_mask_0 = const()[name = tensor<string, []>("op_783_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_783_cast = slice_by_index(begin = var_783_begin_0, end = var_783_end_0, end_mask = var_783_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_783_cast")];
            tensor<int32, [4]> var_787_begin_0 = const()[name = tensor<string, []>("op_787_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_787_end_0 = const()[name = tensor<string, []>("op_787_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_787_end_mask_0 = const()[name = tensor<string, []>("op_787_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_787_cast = slice_by_index(begin = var_787_begin_0, end = var_787_end_0, end_mask = var_787_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_787_cast")];
            tensor<string, []> var_791_equation_0 = const()[name = tensor<string, []>("op_791_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_791_cast = einsum(equation = var_791_equation_0, values = (var_713_cast, var_670_cast))[name = tensor<string, []>("op_791_cast")];
            tensor<fp16, []> var_792_to_fp16 = const()[name = tensor<string, []>("op_792_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_21_cast = mul(x = var_791_cast, y = var_792_to_fp16)[name = tensor<string, []>("aw_21_cast")];
            tensor<string, []> var_795_equation_0 = const()[name = tensor<string, []>("op_795_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_795_cast = einsum(equation = var_795_equation_0, values = (var_717_cast, var_674_cast))[name = tensor<string, []>("op_795_cast")];
            tensor<fp16, []> var_796_to_fp16 = const()[name = tensor<string, []>("op_796_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_23_cast = mul(x = var_795_cast, y = var_796_to_fp16)[name = tensor<string, []>("aw_23_cast")];
            tensor<string, []> var_799_equation_0 = const()[name = tensor<string, []>("op_799_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_799_cast = einsum(equation = var_799_equation_0, values = (var_721_cast, var_678_cast))[name = tensor<string, []>("op_799_cast")];
            tensor<fp16, []> var_800_to_fp16 = const()[name = tensor<string, []>("op_800_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_25_cast = mul(x = var_799_cast, y = var_800_to_fp16)[name = tensor<string, []>("aw_25_cast")];
            tensor<string, []> var_803_equation_0 = const()[name = tensor<string, []>("op_803_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_803_cast = einsum(equation = var_803_equation_0, values = (var_725_cast, var_682_cast))[name = tensor<string, []>("op_803_cast")];
            tensor<fp16, []> var_804_to_fp16 = const()[name = tensor<string, []>("op_804_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_27_cast = mul(x = var_803_cast, y = var_804_to_fp16)[name = tensor<string, []>("aw_27_cast")];
            tensor<string, []> var_807_equation_0 = const()[name = tensor<string, []>("op_807_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_807_cast = einsum(equation = var_807_equation_0, values = (var_729_cast, var_686_cast))[name = tensor<string, []>("op_807_cast")];
            tensor<fp16, []> var_808_to_fp16 = const()[name = tensor<string, []>("op_808_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_29_cast = mul(x = var_807_cast, y = var_808_to_fp16)[name = tensor<string, []>("aw_29_cast")];
            tensor<string, []> var_811_equation_0 = const()[name = tensor<string, []>("op_811_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_811_cast = einsum(equation = var_811_equation_0, values = (var_733_cast, var_690_cast))[name = tensor<string, []>("op_811_cast")];
            tensor<fp16, []> var_812_to_fp16 = const()[name = tensor<string, []>("op_812_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_31_cast = mul(x = var_811_cast, y = var_812_to_fp16)[name = tensor<string, []>("aw_31_cast")];
            tensor<string, []> var_815_equation_0 = const()[name = tensor<string, []>("op_815_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_815_cast = einsum(equation = var_815_equation_0, values = (var_737_cast, var_694_cast))[name = tensor<string, []>("op_815_cast")];
            tensor<fp16, []> var_816_to_fp16 = const()[name = tensor<string, []>("op_816_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_33_cast = mul(x = var_815_cast, y = var_816_to_fp16)[name = tensor<string, []>("aw_33_cast")];
            tensor<string, []> var_819_equation_0 = const()[name = tensor<string, []>("op_819_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_819_cast = einsum(equation = var_819_equation_0, values = (var_741_cast, var_698_cast))[name = tensor<string, []>("op_819_cast")];
            tensor<fp16, []> var_820_to_fp16 = const()[name = tensor<string, []>("op_820_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_35_cast = mul(x = var_819_cast, y = var_820_to_fp16)[name = tensor<string, []>("aw_35_cast")];
            tensor<string, []> var_823_equation_0 = const()[name = tensor<string, []>("op_823_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_823_cast = einsum(equation = var_823_equation_0, values = (var_745_cast, var_702_cast))[name = tensor<string, []>("op_823_cast")];
            tensor<fp16, []> var_824_to_fp16 = const()[name = tensor<string, []>("op_824_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_37_cast = mul(x = var_823_cast, y = var_824_to_fp16)[name = tensor<string, []>("aw_37_cast")];
            tensor<string, []> var_827_equation_0 = const()[name = tensor<string, []>("op_827_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_827_cast = einsum(equation = var_827_equation_0, values = (var_749_cast, var_706_cast))[name = tensor<string, []>("op_827_cast")];
            tensor<fp16, []> var_828_to_fp16 = const()[name = tensor<string, []>("op_828_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_39_cast = mul(x = var_827_cast, y = var_828_to_fp16)[name = tensor<string, []>("aw_39_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_830_cast = softmax(axis = var_298, x = aw_21_cast)[name = tensor<string, []>("op_830_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_831_cast = softmax(axis = var_298, x = aw_23_cast)[name = tensor<string, []>("op_831_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_832_cast = softmax(axis = var_298, x = aw_25_cast)[name = tensor<string, []>("op_832_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_833_cast = softmax(axis = var_298, x = aw_27_cast)[name = tensor<string, []>("op_833_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_834_cast = softmax(axis = var_298, x = aw_29_cast)[name = tensor<string, []>("op_834_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_835_cast = softmax(axis = var_298, x = aw_31_cast)[name = tensor<string, []>("op_835_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_836_cast = softmax(axis = var_298, x = aw_33_cast)[name = tensor<string, []>("op_836_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_837_cast = softmax(axis = var_298, x = aw_35_cast)[name = tensor<string, []>("op_837_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_838_cast = softmax(axis = var_298, x = aw_37_cast)[name = tensor<string, []>("op_838_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_839_cast = softmax(axis = var_298, x = aw_39_cast)[name = tensor<string, []>("op_839_cast")];
            tensor<string, []> var_841_equation_0 = const()[name = tensor<string, []>("op_841_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_841_cast = einsum(equation = var_841_equation_0, values = (var_751_cast, var_830_cast))[name = tensor<string, []>("op_841_cast")];
            tensor<string, []> var_843_equation_0 = const()[name = tensor<string, []>("op_843_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_843_cast = einsum(equation = var_843_equation_0, values = (var_755_cast, var_831_cast))[name = tensor<string, []>("op_843_cast")];
            tensor<string, []> var_845_equation_0 = const()[name = tensor<string, []>("op_845_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_845_cast = einsum(equation = var_845_equation_0, values = (var_759_cast, var_832_cast))[name = tensor<string, []>("op_845_cast")];
            tensor<string, []> var_847_equation_0 = const()[name = tensor<string, []>("op_847_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_847_cast = einsum(equation = var_847_equation_0, values = (var_763_cast, var_833_cast))[name = tensor<string, []>("op_847_cast")];
            tensor<string, []> var_849_equation_0 = const()[name = tensor<string, []>("op_849_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_849_cast = einsum(equation = var_849_equation_0, values = (var_767_cast, var_834_cast))[name = tensor<string, []>("op_849_cast")];
            tensor<string, []> var_851_equation_0 = const()[name = tensor<string, []>("op_851_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_851_cast = einsum(equation = var_851_equation_0, values = (var_771_cast, var_835_cast))[name = tensor<string, []>("op_851_cast")];
            tensor<string, []> var_853_equation_0 = const()[name = tensor<string, []>("op_853_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_853_cast = einsum(equation = var_853_equation_0, values = (var_775_cast, var_836_cast))[name = tensor<string, []>("op_853_cast")];
            tensor<string, []> var_855_equation_0 = const()[name = tensor<string, []>("op_855_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_855_cast = einsum(equation = var_855_equation_0, values = (var_779_cast, var_837_cast))[name = tensor<string, []>("op_855_cast")];
            tensor<string, []> var_857_equation_0 = const()[name = tensor<string, []>("op_857_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_857_cast = einsum(equation = var_857_equation_0, values = (var_783_cast, var_838_cast))[name = tensor<string, []>("op_857_cast")];
            tensor<string, []> var_859_equation_0 = const()[name = tensor<string, []>("op_859_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_859_cast = einsum(equation = var_859_equation_0, values = (var_787_cast, var_839_cast))[name = tensor<string, []>("op_859_cast")];
            tensor<bool, []> input_67_interleave_0 = const()[name = tensor<string, []>("input_67_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_67_cast = concat(axis = var_298, interleave = input_67_interleave_0, values = (var_841_cast, var_843_cast, var_845_cast, var_847_cast, var_849_cast, var_851_cast, var_853_cast, var_855_cast, var_857_cast, var_859_cast))[name = tensor<string, []>("input_67_cast")];
            tensor<int32, [2]> var_865 = const()[name = tensor<string, []>("op_865"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_867 = const()[name = tensor<string, []>("op_867"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_869_pad_type_0 = const()[name = tensor<string, []>("op_869_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_869_pad_0 = const()[name = tensor<string, []>("op_869_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48772864)))];
            tensor<fp16, [640]> down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49592128)))];
            tensor<fp16, [2, 640, 1, 4096]> var_869_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_867, groups = var_298, pad = var_869_pad_0, pad_type = var_869_pad_type_0, strides = var_865, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_67_cast)[name = tensor<string, []>("op_869_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_5_cast = add(x = var_869_cast, y = inputs_3_cast)[name = tensor<string, []>("inputs_5_cast")];
            tensor<int32, [1]> var_873 = const()[name = tensor<string, []>("op_873"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_5_cast = reduce_mean(axes = var_873, keep_dims = var_293, x = inputs_5_cast)[name = tensor<string, []>("channels_mean_5_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_5_cast = sub(x = inputs_5_cast, y = channels_mean_5_cast)[name = tensor<string, []>("zero_mean_5_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_5_cast = mul(x = zero_mean_5_cast, y = zero_mean_5_cast)[name = tensor<string, []>("zero_mean_sq_5_cast")];
            tensor<int32, [1]> var_877 = const()[name = tensor<string, []>("op_877"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_878_cast = reduce_mean(axes = var_877, keep_dims = var_293, x = zero_mean_sq_5_cast)[name = tensor<string, []>("op_878_cast")];
            tensor<fp16, []> var_879_to_fp16 = const()[name = tensor<string, []>("op_879_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_880_cast = add(x = var_878_cast, y = var_879_to_fp16)[name = tensor<string, []>("op_880_cast")];
            tensor<fp16, []> denom_5_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_5_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_5_cast = rsqrt(epsilon = denom_5_epsilon_0_to_fp16, x = var_880_cast)[name = tensor<string, []>("denom_5_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_5_cast = mul(x = zero_mean_5_cast, y = denom_5_cast)[name = tensor<string, []>("out_5_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_884_to_fp16 = const()[name = tensor<string, []>("op_884_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49593472)))];
            tensor<fp16, [2, 640, 1, 4096]> var_885_cast = add(x = out_5_cast, y = var_884_to_fp16)[name = tensor<string, []>("op_885_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_887_to_fp16 = const()[name = tensor<string, []>("op_887_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49594816)))];
            tensor<fp16, [2, 640, 1, 4096]> input_69_cast = mul(x = var_885_cast, y = var_887_to_fp16)[name = tensor<string, []>("input_69_cast")];
            tensor<int32, [2]> var_895 = const()[name = tensor<string, []>("op_895"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_897 = const()[name = tensor<string, []>("op_897"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_899_pad_type_0 = const()[name = tensor<string, []>("op_899_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_899_pad_0 = const()[name = tensor<string, []>("op_899_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [5120, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [5120, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49596160)))];
            tensor<fp16, [5120]> down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56149824)))];
            tensor<fp16, [2, 5120, 1, 4096]> var_899_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_897, groups = var_298, pad = var_899_pad_0, pad_type = var_899_pad_type_0, strides = var_895, weight = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_69_cast)[name = tensor<string, []>("op_899_cast")];
            tensor<int32, [2]> var_900_split_sizes_0 = const()[name = tensor<string, []>("op_900_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_900_axis_0 = const()[name = tensor<string, []>("op_900_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 4096]> var_900_cast_0, tensor<fp16, [2, 2560, 1, 4096]> var_900_cast_1 = split(axis = var_900_axis_0, split_sizes = var_900_split_sizes_0, x = var_899_cast)[name = tensor<string, []>("op_900_cast")];
            tensor<string, []> var_902_mode_0 = const()[name = tensor<string, []>("op_902_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 4096]> var_902_cast = gelu(mode = var_902_mode_0, x = var_900_cast_1)[name = tensor<string, []>("op_902_cast")];
            tensor<fp16, [2, 2560, 1, 4096]> input_71_cast = mul(x = var_900_cast_0, y = var_902_cast)[name = tensor<string, []>("input_71_cast")];
            tensor<int32, [2]> var_906 = const()[name = tensor<string, []>("op_906"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_908 = const()[name = tensor<string, []>("op_908"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_910_pad_type_0 = const()[name = tensor<string, []>("op_910_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_910_pad_0 = const()[name = tensor<string, []>("op_910_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2560, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [640, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56160128)))];
            tensor<fp16, [640]> down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59436992)))];
            tensor<fp16, [2, 640, 1, 4096]> var_910_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_908, groups = var_298, pad = var_910_pad_0, pad_type = var_910_pad_type_0, strides = var_906, weight = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_71_cast)[name = tensor<string, []>("op_910_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_7_cast = add(x = var_910_cast, y = inputs_5_cast)[name = tensor<string, []>("inputs_7_cast")];
            tensor<int32, [1]> var_920 = const()[name = tensor<string, []>("op_920"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_7_cast = reduce_mean(axes = var_920, keep_dims = var_293, x = inputs_7_cast)[name = tensor<string, []>("channels_mean_7_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_7_cast = sub(x = inputs_7_cast, y = channels_mean_7_cast)[name = tensor<string, []>("zero_mean_7_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_7_cast = mul(x = zero_mean_7_cast, y = zero_mean_7_cast)[name = tensor<string, []>("zero_mean_sq_7_cast")];
            tensor<int32, [1]> var_924 = const()[name = tensor<string, []>("op_924"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_925_cast = reduce_mean(axes = var_924, keep_dims = var_293, x = zero_mean_sq_7_cast)[name = tensor<string, []>("op_925_cast")];
            tensor<fp16, []> var_926_to_fp16 = const()[name = tensor<string, []>("op_926_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_927_cast = add(x = var_925_cast, y = var_926_to_fp16)[name = tensor<string, []>("op_927_cast")];
            tensor<fp16, []> denom_7_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_7_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_7_cast = rsqrt(epsilon = denom_7_epsilon_0_to_fp16, x = var_927_cast)[name = tensor<string, []>("denom_7_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_7_cast = mul(x = zero_mean_7_cast, y = denom_7_cast)[name = tensor<string, []>("out_7_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_931_to_fp16 = const()[name = tensor<string, []>("op_931_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59438336)))];
            tensor<fp16, [2, 640, 1, 4096]> var_932_cast = add(x = out_7_cast, y = var_931_to_fp16)[name = tensor<string, []>("op_932_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_934_to_fp16 = const()[name = tensor<string, []>("op_934_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59439680)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_23_cast = mul(x = var_932_cast, y = var_934_to_fp16)[name = tensor<string, []>("hidden_states_23_cast")];
            tensor<int32, [2]> var_941 = const()[name = tensor<string, []>("op_941"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_943 = const()[name = tensor<string, []>("op_943"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_5_pad_type_0 = const()[name = tensor<string, []>("q_5_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_5_pad_0 = const()[name = tensor<string, []>("q_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59441024)))];
            tensor<fp16, [2, 640, 1, 4096]> q_5_cast = conv(dilations = var_943, groups = var_298, pad = q_5_pad_0, pad_type = q_5_pad_type_0, strides = var_941, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_23_cast)[name = tensor<string, []>("q_5_cast")];
            tensor<int32, [2]> var_947 = const()[name = tensor<string, []>("op_947"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_949 = const()[name = tensor<string, []>("op_949"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_9_pad_type_0 = const()[name = tensor<string, []>("k_9_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_9_pad_0 = const()[name = tensor<string, []>("k_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60260288)))];
            tensor<fp16, [2, 640, 1, 4096]> k_9_cast = conv(dilations = var_949, groups = var_298, pad = k_9_pad_0, pad_type = k_9_pad_type_0, strides = var_947, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_23_cast)[name = tensor<string, []>("k_9_cast")];
            tensor<int32, [2]> var_953 = const()[name = tensor<string, []>("op_953"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_955 = const()[name = tensor<string, []>("op_955"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_5_pad_type_0 = const()[name = tensor<string, []>("v_5_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_5_pad_0 = const()[name = tensor<string, []>("v_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61079552)))];
            tensor<fp16, [2, 640, 1, 4096]> v_5_cast = conv(dilations = var_955, groups = var_298, pad = v_5_pad_0, pad_type = v_5_pad_type_0, strides = var_953, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_23_cast)[name = tensor<string, []>("v_5_cast")];
            tensor<int32, [4]> var_959_begin_0 = const()[name = tensor<string, []>("op_959_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_959_end_0 = const()[name = tensor<string, []>("op_959_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_959_end_mask_0 = const()[name = tensor<string, []>("op_959_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_959_cast = slice_by_index(begin = var_959_begin_0, end = var_959_end_0, end_mask = var_959_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_959_cast")];
            tensor<int32, [4]> var_963_begin_0 = const()[name = tensor<string, []>("op_963_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_963_end_0 = const()[name = tensor<string, []>("op_963_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_963_end_mask_0 = const()[name = tensor<string, []>("op_963_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_963_cast = slice_by_index(begin = var_963_begin_0, end = var_963_end_0, end_mask = var_963_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_963_cast")];
            tensor<int32, [4]> var_967_begin_0 = const()[name = tensor<string, []>("op_967_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_967_end_0 = const()[name = tensor<string, []>("op_967_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_967_end_mask_0 = const()[name = tensor<string, []>("op_967_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_967_cast = slice_by_index(begin = var_967_begin_0, end = var_967_end_0, end_mask = var_967_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_967_cast")];
            tensor<int32, [4]> var_971_begin_0 = const()[name = tensor<string, []>("op_971_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_971_end_0 = const()[name = tensor<string, []>("op_971_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_971_end_mask_0 = const()[name = tensor<string, []>("op_971_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_971_cast = slice_by_index(begin = var_971_begin_0, end = var_971_end_0, end_mask = var_971_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_971_cast")];
            tensor<int32, [4]> var_975_begin_0 = const()[name = tensor<string, []>("op_975_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_975_end_0 = const()[name = tensor<string, []>("op_975_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_975_end_mask_0 = const()[name = tensor<string, []>("op_975_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_975_cast = slice_by_index(begin = var_975_begin_0, end = var_975_end_0, end_mask = var_975_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_975_cast")];
            tensor<int32, [4]> var_979_begin_0 = const()[name = tensor<string, []>("op_979_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_979_end_0 = const()[name = tensor<string, []>("op_979_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_979_end_mask_0 = const()[name = tensor<string, []>("op_979_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_979_cast = slice_by_index(begin = var_979_begin_0, end = var_979_end_0, end_mask = var_979_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_979_cast")];
            tensor<int32, [4]> var_983_begin_0 = const()[name = tensor<string, []>("op_983_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_983_end_0 = const()[name = tensor<string, []>("op_983_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_983_end_mask_0 = const()[name = tensor<string, []>("op_983_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_983_cast = slice_by_index(begin = var_983_begin_0, end = var_983_end_0, end_mask = var_983_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_983_cast")];
            tensor<int32, [4]> var_987_begin_0 = const()[name = tensor<string, []>("op_987_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_987_end_0 = const()[name = tensor<string, []>("op_987_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_987_end_mask_0 = const()[name = tensor<string, []>("op_987_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_987_cast = slice_by_index(begin = var_987_begin_0, end = var_987_end_0, end_mask = var_987_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_987_cast")];
            tensor<int32, [4]> var_991_begin_0 = const()[name = tensor<string, []>("op_991_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_991_end_0 = const()[name = tensor<string, []>("op_991_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_991_end_mask_0 = const()[name = tensor<string, []>("op_991_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_991_cast = slice_by_index(begin = var_991_begin_0, end = var_991_end_0, end_mask = var_991_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_991_cast")];
            tensor<int32, [4]> var_995_begin_0 = const()[name = tensor<string, []>("op_995_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_995_end_0 = const()[name = tensor<string, []>("op_995_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_995_end_mask_0 = const()[name = tensor<string, []>("op_995_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_995_cast = slice_by_index(begin = var_995_begin_0, end = var_995_end_0, end_mask = var_995_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_995_cast")];
            tensor<int32, [4]> k_11_perm_0 = const()[name = tensor<string, []>("k_11_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_1002_begin_0 = const()[name = tensor<string, []>("op_1002_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1002_end_0 = const()[name = tensor<string, []>("op_1002_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_1002_end_mask_0 = const()[name = tensor<string, []>("op_1002_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 640]> transpose_137 = transpose(perm = k_11_perm_0, x = k_9_cast)[name = tensor<string, []>("transpose_137")];
            tensor<fp16, [2, 4096, 1, 64]> var_1002_cast = slice_by_index(begin = var_1002_begin_0, end = var_1002_end_0, end_mask = var_1002_end_mask_0, x = transpose_137)[name = tensor<string, []>("op_1002_cast")];
            tensor<int32, [4]> var_1006_begin_0 = const()[name = tensor<string, []>("op_1006_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_1006_end_0 = const()[name = tensor<string, []>("op_1006_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_1006_end_mask_0 = const()[name = tensor<string, []>("op_1006_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1006_cast = slice_by_index(begin = var_1006_begin_0, end = var_1006_end_0, end_mask = var_1006_end_mask_0, x = transpose_137)[name = tensor<string, []>("op_1006_cast")];
            tensor<int32, [4]> var_1010_begin_0 = const()[name = tensor<string, []>("op_1010_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_1010_end_0 = const()[name = tensor<string, []>("op_1010_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_1010_end_mask_0 = const()[name = tensor<string, []>("op_1010_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1010_cast = slice_by_index(begin = var_1010_begin_0, end = var_1010_end_0, end_mask = var_1010_end_mask_0, x = transpose_137)[name = tensor<string, []>("op_1010_cast")];
            tensor<int32, [4]> var_1014_begin_0 = const()[name = tensor<string, []>("op_1014_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_1014_end_0 = const()[name = tensor<string, []>("op_1014_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_1014_end_mask_0 = const()[name = tensor<string, []>("op_1014_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1014_cast = slice_by_index(begin = var_1014_begin_0, end = var_1014_end_0, end_mask = var_1014_end_mask_0, x = transpose_137)[name = tensor<string, []>("op_1014_cast")];
            tensor<int32, [4]> var_1018_begin_0 = const()[name = tensor<string, []>("op_1018_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_1018_end_0 = const()[name = tensor<string, []>("op_1018_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_1018_end_mask_0 = const()[name = tensor<string, []>("op_1018_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1018_cast = slice_by_index(begin = var_1018_begin_0, end = var_1018_end_0, end_mask = var_1018_end_mask_0, x = transpose_137)[name = tensor<string, []>("op_1018_cast")];
            tensor<int32, [4]> var_1022_begin_0 = const()[name = tensor<string, []>("op_1022_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_1022_end_0 = const()[name = tensor<string, []>("op_1022_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 384])];
            tensor<bool, [4]> var_1022_end_mask_0 = const()[name = tensor<string, []>("op_1022_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1022_cast = slice_by_index(begin = var_1022_begin_0, end = var_1022_end_0, end_mask = var_1022_end_mask_0, x = transpose_137)[name = tensor<string, []>("op_1022_cast")];
            tensor<int32, [4]> var_1026_begin_0 = const()[name = tensor<string, []>("op_1026_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_1026_end_0 = const()[name = tensor<string, []>("op_1026_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 448])];
            tensor<bool, [4]> var_1026_end_mask_0 = const()[name = tensor<string, []>("op_1026_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1026_cast = slice_by_index(begin = var_1026_begin_0, end = var_1026_end_0, end_mask = var_1026_end_mask_0, x = transpose_137)[name = tensor<string, []>("op_1026_cast")];
            tensor<int32, [4]> var_1030_begin_0 = const()[name = tensor<string, []>("op_1030_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_1030_end_0 = const()[name = tensor<string, []>("op_1030_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 512])];
            tensor<bool, [4]> var_1030_end_mask_0 = const()[name = tensor<string, []>("op_1030_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1030_cast = slice_by_index(begin = var_1030_begin_0, end = var_1030_end_0, end_mask = var_1030_end_mask_0, x = transpose_137)[name = tensor<string, []>("op_1030_cast")];
            tensor<int32, [4]> var_1034_begin_0 = const()[name = tensor<string, []>("op_1034_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1034_end_0 = const()[name = tensor<string, []>("op_1034_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 576])];
            tensor<bool, [4]> var_1034_end_mask_0 = const()[name = tensor<string, []>("op_1034_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1034_cast = slice_by_index(begin = var_1034_begin_0, end = var_1034_end_0, end_mask = var_1034_end_mask_0, x = transpose_137)[name = tensor<string, []>("op_1034_cast")];
            tensor<int32, [4]> var_1038_begin_0 = const()[name = tensor<string, []>("op_1038_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_1038_end_0 = const()[name = tensor<string, []>("op_1038_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 640])];
            tensor<bool, [4]> var_1038_end_mask_0 = const()[name = tensor<string, []>("op_1038_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1038_cast = slice_by_index(begin = var_1038_begin_0, end = var_1038_end_0, end_mask = var_1038_end_mask_0, x = transpose_137)[name = tensor<string, []>("op_1038_cast")];
            tensor<int32, [4]> var_1040_begin_0 = const()[name = tensor<string, []>("op_1040_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1040_end_0 = const()[name = tensor<string, []>("op_1040_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_1040_end_mask_0 = const()[name = tensor<string, []>("op_1040_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1040_cast = slice_by_index(begin = var_1040_begin_0, end = var_1040_end_0, end_mask = var_1040_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1040_cast")];
            tensor<int32, [4]> var_1044_begin_0 = const()[name = tensor<string, []>("op_1044_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1044_end_0 = const()[name = tensor<string, []>("op_1044_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_1044_end_mask_0 = const()[name = tensor<string, []>("op_1044_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1044_cast = slice_by_index(begin = var_1044_begin_0, end = var_1044_end_0, end_mask = var_1044_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1044_cast")];
            tensor<int32, [4]> var_1048_begin_0 = const()[name = tensor<string, []>("op_1048_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1048_end_0 = const()[name = tensor<string, []>("op_1048_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_1048_end_mask_0 = const()[name = tensor<string, []>("op_1048_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1048_cast = slice_by_index(begin = var_1048_begin_0, end = var_1048_end_0, end_mask = var_1048_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1048_cast")];
            tensor<int32, [4]> var_1052_begin_0 = const()[name = tensor<string, []>("op_1052_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1052_end_0 = const()[name = tensor<string, []>("op_1052_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_1052_end_mask_0 = const()[name = tensor<string, []>("op_1052_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1052_cast = slice_by_index(begin = var_1052_begin_0, end = var_1052_end_0, end_mask = var_1052_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1052_cast")];
            tensor<int32, [4]> var_1056_begin_0 = const()[name = tensor<string, []>("op_1056_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1056_end_0 = const()[name = tensor<string, []>("op_1056_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_1056_end_mask_0 = const()[name = tensor<string, []>("op_1056_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1056_cast = slice_by_index(begin = var_1056_begin_0, end = var_1056_end_0, end_mask = var_1056_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1056_cast")];
            tensor<int32, [4]> var_1060_begin_0 = const()[name = tensor<string, []>("op_1060_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1060_end_0 = const()[name = tensor<string, []>("op_1060_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_1060_end_mask_0 = const()[name = tensor<string, []>("op_1060_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1060_cast = slice_by_index(begin = var_1060_begin_0, end = var_1060_end_0, end_mask = var_1060_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1060_cast")];
            tensor<int32, [4]> var_1064_begin_0 = const()[name = tensor<string, []>("op_1064_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_1064_end_0 = const()[name = tensor<string, []>("op_1064_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_1064_end_mask_0 = const()[name = tensor<string, []>("op_1064_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1064_cast = slice_by_index(begin = var_1064_begin_0, end = var_1064_end_0, end_mask = var_1064_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1064_cast")];
            tensor<int32, [4]> var_1068_begin_0 = const()[name = tensor<string, []>("op_1068_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_1068_end_0 = const()[name = tensor<string, []>("op_1068_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_1068_end_mask_0 = const()[name = tensor<string, []>("op_1068_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1068_cast = slice_by_index(begin = var_1068_begin_0, end = var_1068_end_0, end_mask = var_1068_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1068_cast")];
            tensor<int32, [4]> var_1072_begin_0 = const()[name = tensor<string, []>("op_1072_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_1072_end_0 = const()[name = tensor<string, []>("op_1072_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_1072_end_mask_0 = const()[name = tensor<string, []>("op_1072_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1072_cast = slice_by_index(begin = var_1072_begin_0, end = var_1072_end_0, end_mask = var_1072_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1072_cast")];
            tensor<int32, [4]> var_1076_begin_0 = const()[name = tensor<string, []>("op_1076_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_1076_end_0 = const()[name = tensor<string, []>("op_1076_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_1076_end_mask_0 = const()[name = tensor<string, []>("op_1076_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1076_cast = slice_by_index(begin = var_1076_begin_0, end = var_1076_end_0, end_mask = var_1076_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1076_cast")];
            tensor<string, []> var_1080_equation_0 = const()[name = tensor<string, []>("op_1080_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1080_cast = einsum(equation = var_1080_equation_0, values = (var_1002_cast, var_959_cast))[name = tensor<string, []>("op_1080_cast")];
            tensor<fp16, []> var_1081_to_fp16 = const()[name = tensor<string, []>("op_1081_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_41_cast = mul(x = var_1080_cast, y = var_1081_to_fp16)[name = tensor<string, []>("aw_41_cast")];
            tensor<string, []> var_1084_equation_0 = const()[name = tensor<string, []>("op_1084_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1084_cast = einsum(equation = var_1084_equation_0, values = (var_1006_cast, var_963_cast))[name = tensor<string, []>("op_1084_cast")];
            tensor<fp16, []> var_1085_to_fp16 = const()[name = tensor<string, []>("op_1085_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_43_cast = mul(x = var_1084_cast, y = var_1085_to_fp16)[name = tensor<string, []>("aw_43_cast")];
            tensor<string, []> var_1088_equation_0 = const()[name = tensor<string, []>("op_1088_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1088_cast = einsum(equation = var_1088_equation_0, values = (var_1010_cast, var_967_cast))[name = tensor<string, []>("op_1088_cast")];
            tensor<fp16, []> var_1089_to_fp16 = const()[name = tensor<string, []>("op_1089_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_45_cast = mul(x = var_1088_cast, y = var_1089_to_fp16)[name = tensor<string, []>("aw_45_cast")];
            tensor<string, []> var_1092_equation_0 = const()[name = tensor<string, []>("op_1092_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1092_cast = einsum(equation = var_1092_equation_0, values = (var_1014_cast, var_971_cast))[name = tensor<string, []>("op_1092_cast")];
            tensor<fp16, []> var_1093_to_fp16 = const()[name = tensor<string, []>("op_1093_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_47_cast = mul(x = var_1092_cast, y = var_1093_to_fp16)[name = tensor<string, []>("aw_47_cast")];
            tensor<string, []> var_1096_equation_0 = const()[name = tensor<string, []>("op_1096_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1096_cast = einsum(equation = var_1096_equation_0, values = (var_1018_cast, var_975_cast))[name = tensor<string, []>("op_1096_cast")];
            tensor<fp16, []> var_1097_to_fp16 = const()[name = tensor<string, []>("op_1097_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_49_cast = mul(x = var_1096_cast, y = var_1097_to_fp16)[name = tensor<string, []>("aw_49_cast")];
            tensor<string, []> var_1100_equation_0 = const()[name = tensor<string, []>("op_1100_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1100_cast = einsum(equation = var_1100_equation_0, values = (var_1022_cast, var_979_cast))[name = tensor<string, []>("op_1100_cast")];
            tensor<fp16, []> var_1101_to_fp16 = const()[name = tensor<string, []>("op_1101_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_51_cast = mul(x = var_1100_cast, y = var_1101_to_fp16)[name = tensor<string, []>("aw_51_cast")];
            tensor<string, []> var_1104_equation_0 = const()[name = tensor<string, []>("op_1104_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1104_cast = einsum(equation = var_1104_equation_0, values = (var_1026_cast, var_983_cast))[name = tensor<string, []>("op_1104_cast")];
            tensor<fp16, []> var_1105_to_fp16 = const()[name = tensor<string, []>("op_1105_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_53_cast = mul(x = var_1104_cast, y = var_1105_to_fp16)[name = tensor<string, []>("aw_53_cast")];
            tensor<string, []> var_1108_equation_0 = const()[name = tensor<string, []>("op_1108_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1108_cast = einsum(equation = var_1108_equation_0, values = (var_1030_cast, var_987_cast))[name = tensor<string, []>("op_1108_cast")];
            tensor<fp16, []> var_1109_to_fp16 = const()[name = tensor<string, []>("op_1109_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_55_cast = mul(x = var_1108_cast, y = var_1109_to_fp16)[name = tensor<string, []>("aw_55_cast")];
            tensor<string, []> var_1112_equation_0 = const()[name = tensor<string, []>("op_1112_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1112_cast = einsum(equation = var_1112_equation_0, values = (var_1034_cast, var_991_cast))[name = tensor<string, []>("op_1112_cast")];
            tensor<fp16, []> var_1113_to_fp16 = const()[name = tensor<string, []>("op_1113_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_57_cast = mul(x = var_1112_cast, y = var_1113_to_fp16)[name = tensor<string, []>("aw_57_cast")];
            tensor<string, []> var_1116_equation_0 = const()[name = tensor<string, []>("op_1116_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1116_cast = einsum(equation = var_1116_equation_0, values = (var_1038_cast, var_995_cast))[name = tensor<string, []>("op_1116_cast")];
            tensor<fp16, []> var_1117_to_fp16 = const()[name = tensor<string, []>("op_1117_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_59_cast = mul(x = var_1116_cast, y = var_1117_to_fp16)[name = tensor<string, []>("aw_59_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1119_cast = softmax(axis = var_298, x = aw_41_cast)[name = tensor<string, []>("op_1119_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1120_cast = softmax(axis = var_298, x = aw_43_cast)[name = tensor<string, []>("op_1120_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1121_cast = softmax(axis = var_298, x = aw_45_cast)[name = tensor<string, []>("op_1121_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1122_cast = softmax(axis = var_298, x = aw_47_cast)[name = tensor<string, []>("op_1122_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1123_cast = softmax(axis = var_298, x = aw_49_cast)[name = tensor<string, []>("op_1123_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1124_cast = softmax(axis = var_298, x = aw_51_cast)[name = tensor<string, []>("op_1124_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1125_cast = softmax(axis = var_298, x = aw_53_cast)[name = tensor<string, []>("op_1125_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1126_cast = softmax(axis = var_298, x = aw_55_cast)[name = tensor<string, []>("op_1126_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1127_cast = softmax(axis = var_298, x = aw_57_cast)[name = tensor<string, []>("op_1127_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1128_cast = softmax(axis = var_298, x = aw_59_cast)[name = tensor<string, []>("op_1128_cast")];
            tensor<string, []> var_1130_equation_0 = const()[name = tensor<string, []>("op_1130_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1130_cast = einsum(equation = var_1130_equation_0, values = (var_1040_cast, var_1119_cast))[name = tensor<string, []>("op_1130_cast")];
            tensor<string, []> var_1132_equation_0 = const()[name = tensor<string, []>("op_1132_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1132_cast = einsum(equation = var_1132_equation_0, values = (var_1044_cast, var_1120_cast))[name = tensor<string, []>("op_1132_cast")];
            tensor<string, []> var_1134_equation_0 = const()[name = tensor<string, []>("op_1134_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1134_cast = einsum(equation = var_1134_equation_0, values = (var_1048_cast, var_1121_cast))[name = tensor<string, []>("op_1134_cast")];
            tensor<string, []> var_1136_equation_0 = const()[name = tensor<string, []>("op_1136_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1136_cast = einsum(equation = var_1136_equation_0, values = (var_1052_cast, var_1122_cast))[name = tensor<string, []>("op_1136_cast")];
            tensor<string, []> var_1138_equation_0 = const()[name = tensor<string, []>("op_1138_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1138_cast = einsum(equation = var_1138_equation_0, values = (var_1056_cast, var_1123_cast))[name = tensor<string, []>("op_1138_cast")];
            tensor<string, []> var_1140_equation_0 = const()[name = tensor<string, []>("op_1140_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1140_cast = einsum(equation = var_1140_equation_0, values = (var_1060_cast, var_1124_cast))[name = tensor<string, []>("op_1140_cast")];
            tensor<string, []> var_1142_equation_0 = const()[name = tensor<string, []>("op_1142_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1142_cast = einsum(equation = var_1142_equation_0, values = (var_1064_cast, var_1125_cast))[name = tensor<string, []>("op_1142_cast")];
            tensor<string, []> var_1144_equation_0 = const()[name = tensor<string, []>("op_1144_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1144_cast = einsum(equation = var_1144_equation_0, values = (var_1068_cast, var_1126_cast))[name = tensor<string, []>("op_1144_cast")];
            tensor<string, []> var_1146_equation_0 = const()[name = tensor<string, []>("op_1146_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1146_cast = einsum(equation = var_1146_equation_0, values = (var_1072_cast, var_1127_cast))[name = tensor<string, []>("op_1146_cast")];
            tensor<string, []> var_1148_equation_0 = const()[name = tensor<string, []>("op_1148_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1148_cast = einsum(equation = var_1148_equation_0, values = (var_1076_cast, var_1128_cast))[name = tensor<string, []>("op_1148_cast")];
            tensor<bool, []> input_73_interleave_0 = const()[name = tensor<string, []>("input_73_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_73_cast = concat(axis = var_298, interleave = input_73_interleave_0, values = (var_1130_cast, var_1132_cast, var_1134_cast, var_1136_cast, var_1138_cast, var_1140_cast, var_1142_cast, var_1144_cast, var_1146_cast, var_1148_cast))[name = tensor<string, []>("input_73_cast")];
            tensor<int32, [2]> var_1154 = const()[name = tensor<string, []>("op_1154"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1156 = const()[name = tensor<string, []>("op_1156"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_1158_pad_type_0 = const()[name = tensor<string, []>("op_1158_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_1158_pad_0 = const()[name = tensor<string, []>("op_1158_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61898816)))];
            tensor<fp16, [640]> down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62718080)))];
            tensor<fp16, [2, 640, 1, 4096]> var_1158_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_1156, groups = var_298, pad = var_1158_pad_0, pad_type = var_1158_pad_type_0, strides = var_1154, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_73_cast)[name = tensor<string, []>("op_1158_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_9_cast = add(x = var_1158_cast, y = inputs_7_cast)[name = tensor<string, []>("inputs_9_cast")];
            tensor<int32, [1]> var_1162 = const()[name = tensor<string, []>("op_1162"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_9_cast = reduce_mean(axes = var_1162, keep_dims = var_293, x = inputs_9_cast)[name = tensor<string, []>("channels_mean_9_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_9_cast = sub(x = inputs_9_cast, y = channels_mean_9_cast)[name = tensor<string, []>("zero_mean_9_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_9_cast = mul(x = zero_mean_9_cast, y = zero_mean_9_cast)[name = tensor<string, []>("zero_mean_sq_9_cast")];
            tensor<int32, [1]> var_1166 = const()[name = tensor<string, []>("op_1166"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_1167_cast = reduce_mean(axes = var_1166, keep_dims = var_293, x = zero_mean_sq_9_cast)[name = tensor<string, []>("op_1167_cast")];
            tensor<fp16, []> var_1168_to_fp16 = const()[name = tensor<string, []>("op_1168_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_1169_cast = add(x = var_1167_cast, y = var_1168_to_fp16)[name = tensor<string, []>("op_1169_cast")];
            tensor<fp16, []> denom_9_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_9_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_9_cast = rsqrt(epsilon = denom_9_epsilon_0_to_fp16, x = var_1169_cast)[name = tensor<string, []>("denom_9_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_9_cast = mul(x = zero_mean_9_cast, y = denom_9_cast)[name = tensor<string, []>("out_9_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_1173_to_fp16 = const()[name = tensor<string, []>("op_1173_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62719424)))];
            tensor<fp16, [2, 640, 1, 4096]> var_1174_cast = add(x = out_9_cast, y = var_1173_to_fp16)[name = tensor<string, []>("op_1174_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_1176_to_fp16 = const()[name = tensor<string, []>("op_1176_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62720768)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_25_cast = mul(x = var_1174_cast, y = var_1176_to_fp16)[name = tensor<string, []>("hidden_states_25_cast")];
            tensor<int32, [2]> var_1183 = const()[name = tensor<string, []>("op_1183"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1185 = const()[name = tensor<string, []>("op_1185"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_7_pad_type_0 = const()[name = tensor<string, []>("q_7_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_7_pad_0 = const()[name = tensor<string, []>("q_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62722112)))];
            tensor<fp16, [2, 640, 1, 4096]> q_7_cast = conv(dilations = var_1185, groups = var_298, pad = q_7_pad_0, pad_type = q_7_pad_type_0, strides = var_1183, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_25_cast)[name = tensor<string, []>("q_7_cast")];
            tensor<int32, [2]> var_1189 = const()[name = tensor<string, []>("op_1189"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1191 = const()[name = tensor<string, []>("op_1191"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_13_pad_type_0 = const()[name = tensor<string, []>("k_13_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_13_pad_0 = const()[name = tensor<string, []>("k_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63541376)))];
            tensor<fp16, [2, 640, 1, 77]> k_13_cast = conv(dilations = var_1191, groups = var_298, pad = k_13_pad_0, pad_type = k_13_pad_type_0, strides = var_1189, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_13_cast")];
            tensor<int32, [2]> var_1195 = const()[name = tensor<string, []>("op_1195"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1197 = const()[name = tensor<string, []>("op_1197"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_7_pad_type_0 = const()[name = tensor<string, []>("v_7_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_7_pad_0 = const()[name = tensor<string, []>("v_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66162880)))];
            tensor<fp16, [2, 640, 1, 77]> v_7_cast = conv(dilations = var_1197, groups = var_298, pad = v_7_pad_0, pad_type = v_7_pad_type_0, strides = var_1195, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_7_cast")];
            tensor<int32, [4]> var_1201_begin_0 = const()[name = tensor<string, []>("op_1201_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1201_end_0 = const()[name = tensor<string, []>("op_1201_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_1201_end_mask_0 = const()[name = tensor<string, []>("op_1201_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1201_cast = slice_by_index(begin = var_1201_begin_0, end = var_1201_end_0, end_mask = var_1201_end_mask_0, x = q_7_cast)[name = tensor<string, []>("op_1201_cast")];
            tensor<int32, [4]> var_1205_begin_0 = const()[name = tensor<string, []>("op_1205_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1205_end_0 = const()[name = tensor<string, []>("op_1205_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_1205_end_mask_0 = const()[name = tensor<string, []>("op_1205_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1205_cast = slice_by_index(begin = var_1205_begin_0, end = var_1205_end_0, end_mask = var_1205_end_mask_0, x = q_7_cast)[name = tensor<string, []>("op_1205_cast")];
            tensor<int32, [4]> var_1209_begin_0 = const()[name = tensor<string, []>("op_1209_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1209_end_0 = const()[name = tensor<string, []>("op_1209_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_1209_end_mask_0 = const()[name = tensor<string, []>("op_1209_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1209_cast = slice_by_index(begin = var_1209_begin_0, end = var_1209_end_0, end_mask = var_1209_end_mask_0, x = q_7_cast)[name = tensor<string, []>("op_1209_cast")];
            tensor<int32, [4]> var_1213_begin_0 = const()[name = tensor<string, []>("op_1213_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1213_end_0 = const()[name = tensor<string, []>("op_1213_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_1213_end_mask_0 = const()[name = tensor<string, []>("op_1213_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1213_cast = slice_by_index(begin = var_1213_begin_0, end = var_1213_end_0, end_mask = var_1213_end_mask_0, x = q_7_cast)[name = tensor<string, []>("op_1213_cast")];
            tensor<int32, [4]> var_1217_begin_0 = const()[name = tensor<string, []>("op_1217_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1217_end_0 = const()[name = tensor<string, []>("op_1217_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_1217_end_mask_0 = const()[name = tensor<string, []>("op_1217_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1217_cast = slice_by_index(begin = var_1217_begin_0, end = var_1217_end_0, end_mask = var_1217_end_mask_0, x = q_7_cast)[name = tensor<string, []>("op_1217_cast")];
            tensor<int32, [4]> var_1221_begin_0 = const()[name = tensor<string, []>("op_1221_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1221_end_0 = const()[name = tensor<string, []>("op_1221_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_1221_end_mask_0 = const()[name = tensor<string, []>("op_1221_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1221_cast = slice_by_index(begin = var_1221_begin_0, end = var_1221_end_0, end_mask = var_1221_end_mask_0, x = q_7_cast)[name = tensor<string, []>("op_1221_cast")];
            tensor<int32, [4]> var_1225_begin_0 = const()[name = tensor<string, []>("op_1225_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_1225_end_0 = const()[name = tensor<string, []>("op_1225_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_1225_end_mask_0 = const()[name = tensor<string, []>("op_1225_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1225_cast = slice_by_index(begin = var_1225_begin_0, end = var_1225_end_0, end_mask = var_1225_end_mask_0, x = q_7_cast)[name = tensor<string, []>("op_1225_cast")];
            tensor<int32, [4]> var_1229_begin_0 = const()[name = tensor<string, []>("op_1229_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_1229_end_0 = const()[name = tensor<string, []>("op_1229_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_1229_end_mask_0 = const()[name = tensor<string, []>("op_1229_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1229_cast = slice_by_index(begin = var_1229_begin_0, end = var_1229_end_0, end_mask = var_1229_end_mask_0, x = q_7_cast)[name = tensor<string, []>("op_1229_cast")];
            tensor<int32, [4]> var_1233_begin_0 = const()[name = tensor<string, []>("op_1233_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_1233_end_0 = const()[name = tensor<string, []>("op_1233_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_1233_end_mask_0 = const()[name = tensor<string, []>("op_1233_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1233_cast = slice_by_index(begin = var_1233_begin_0, end = var_1233_end_0, end_mask = var_1233_end_mask_0, x = q_7_cast)[name = tensor<string, []>("op_1233_cast")];
            tensor<int32, [4]> var_1237_begin_0 = const()[name = tensor<string, []>("op_1237_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_1237_end_0 = const()[name = tensor<string, []>("op_1237_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_1237_end_mask_0 = const()[name = tensor<string, []>("op_1237_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1237_cast = slice_by_index(begin = var_1237_begin_0, end = var_1237_end_0, end_mask = var_1237_end_mask_0, x = q_7_cast)[name = tensor<string, []>("op_1237_cast")];
            tensor<int32, [4]> k_15_perm_0 = const()[name = tensor<string, []>("k_15_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_1244_begin_0 = const()[name = tensor<string, []>("op_1244_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1244_end_0 = const()[name = tensor<string, []>("op_1244_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_1244_end_mask_0 = const()[name = tensor<string, []>("op_1244_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> transpose_136 = transpose(perm = k_15_perm_0, x = k_13_cast)[name = tensor<string, []>("transpose_136")];
            tensor<fp16, [2, 77, 1, 64]> var_1244_cast = slice_by_index(begin = var_1244_begin_0, end = var_1244_end_0, end_mask = var_1244_end_mask_0, x = transpose_136)[name = tensor<string, []>("op_1244_cast")];
            tensor<int32, [4]> var_1248_begin_0 = const()[name = tensor<string, []>("op_1248_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_1248_end_0 = const()[name = tensor<string, []>("op_1248_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_1248_end_mask_0 = const()[name = tensor<string, []>("op_1248_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1248_cast = slice_by_index(begin = var_1248_begin_0, end = var_1248_end_0, end_mask = var_1248_end_mask_0, x = transpose_136)[name = tensor<string, []>("op_1248_cast")];
            tensor<int32, [4]> var_1252_begin_0 = const()[name = tensor<string, []>("op_1252_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_1252_end_0 = const()[name = tensor<string, []>("op_1252_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_1252_end_mask_0 = const()[name = tensor<string, []>("op_1252_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1252_cast = slice_by_index(begin = var_1252_begin_0, end = var_1252_end_0, end_mask = var_1252_end_mask_0, x = transpose_136)[name = tensor<string, []>("op_1252_cast")];
            tensor<int32, [4]> var_1256_begin_0 = const()[name = tensor<string, []>("op_1256_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_1256_end_0 = const()[name = tensor<string, []>("op_1256_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_1256_end_mask_0 = const()[name = tensor<string, []>("op_1256_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1256_cast = slice_by_index(begin = var_1256_begin_0, end = var_1256_end_0, end_mask = var_1256_end_mask_0, x = transpose_136)[name = tensor<string, []>("op_1256_cast")];
            tensor<int32, [4]> var_1260_begin_0 = const()[name = tensor<string, []>("op_1260_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_1260_end_0 = const()[name = tensor<string, []>("op_1260_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_1260_end_mask_0 = const()[name = tensor<string, []>("op_1260_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1260_cast = slice_by_index(begin = var_1260_begin_0, end = var_1260_end_0, end_mask = var_1260_end_mask_0, x = transpose_136)[name = tensor<string, []>("op_1260_cast")];
            tensor<int32, [4]> var_1264_begin_0 = const()[name = tensor<string, []>("op_1264_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_1264_end_0 = const()[name = tensor<string, []>("op_1264_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_1264_end_mask_0 = const()[name = tensor<string, []>("op_1264_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1264_cast = slice_by_index(begin = var_1264_begin_0, end = var_1264_end_0, end_mask = var_1264_end_mask_0, x = transpose_136)[name = tensor<string, []>("op_1264_cast")];
            tensor<int32, [4]> var_1268_begin_0 = const()[name = tensor<string, []>("op_1268_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_1268_end_0 = const()[name = tensor<string, []>("op_1268_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_1268_end_mask_0 = const()[name = tensor<string, []>("op_1268_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1268_cast = slice_by_index(begin = var_1268_begin_0, end = var_1268_end_0, end_mask = var_1268_end_mask_0, x = transpose_136)[name = tensor<string, []>("op_1268_cast")];
            tensor<int32, [4]> var_1272_begin_0 = const()[name = tensor<string, []>("op_1272_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_1272_end_0 = const()[name = tensor<string, []>("op_1272_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_1272_end_mask_0 = const()[name = tensor<string, []>("op_1272_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1272_cast = slice_by_index(begin = var_1272_begin_0, end = var_1272_end_0, end_mask = var_1272_end_mask_0, x = transpose_136)[name = tensor<string, []>("op_1272_cast")];
            tensor<int32, [4]> var_1276_begin_0 = const()[name = tensor<string, []>("op_1276_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1276_end_0 = const()[name = tensor<string, []>("op_1276_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_1276_end_mask_0 = const()[name = tensor<string, []>("op_1276_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1276_cast = slice_by_index(begin = var_1276_begin_0, end = var_1276_end_0, end_mask = var_1276_end_mask_0, x = transpose_136)[name = tensor<string, []>("op_1276_cast")];
            tensor<int32, [4]> var_1280_begin_0 = const()[name = tensor<string, []>("op_1280_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_1280_end_0 = const()[name = tensor<string, []>("op_1280_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_1280_end_mask_0 = const()[name = tensor<string, []>("op_1280_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1280_cast = slice_by_index(begin = var_1280_begin_0, end = var_1280_end_0, end_mask = var_1280_end_mask_0, x = transpose_136)[name = tensor<string, []>("op_1280_cast")];
            tensor<int32, [4]> var_1282_begin_0 = const()[name = tensor<string, []>("op_1282_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1282_end_0 = const()[name = tensor<string, []>("op_1282_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_1282_end_mask_0 = const()[name = tensor<string, []>("op_1282_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1282_cast = slice_by_index(begin = var_1282_begin_0, end = var_1282_end_0, end_mask = var_1282_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_1282_cast")];
            tensor<int32, [4]> var_1286_begin_0 = const()[name = tensor<string, []>("op_1286_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1286_end_0 = const()[name = tensor<string, []>("op_1286_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_1286_end_mask_0 = const()[name = tensor<string, []>("op_1286_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1286_cast = slice_by_index(begin = var_1286_begin_0, end = var_1286_end_0, end_mask = var_1286_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_1286_cast")];
            tensor<int32, [4]> var_1290_begin_0 = const()[name = tensor<string, []>("op_1290_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1290_end_0 = const()[name = tensor<string, []>("op_1290_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_1290_end_mask_0 = const()[name = tensor<string, []>("op_1290_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1290_cast = slice_by_index(begin = var_1290_begin_0, end = var_1290_end_0, end_mask = var_1290_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_1290_cast")];
            tensor<int32, [4]> var_1294_begin_0 = const()[name = tensor<string, []>("op_1294_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1294_end_0 = const()[name = tensor<string, []>("op_1294_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_1294_end_mask_0 = const()[name = tensor<string, []>("op_1294_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1294_cast = slice_by_index(begin = var_1294_begin_0, end = var_1294_end_0, end_mask = var_1294_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_1294_cast")];
            tensor<int32, [4]> var_1298_begin_0 = const()[name = tensor<string, []>("op_1298_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1298_end_0 = const()[name = tensor<string, []>("op_1298_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_1298_end_mask_0 = const()[name = tensor<string, []>("op_1298_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1298_cast = slice_by_index(begin = var_1298_begin_0, end = var_1298_end_0, end_mask = var_1298_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_1298_cast")];
            tensor<int32, [4]> var_1302_begin_0 = const()[name = tensor<string, []>("op_1302_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1302_end_0 = const()[name = tensor<string, []>("op_1302_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_1302_end_mask_0 = const()[name = tensor<string, []>("op_1302_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1302_cast = slice_by_index(begin = var_1302_begin_0, end = var_1302_end_0, end_mask = var_1302_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_1302_cast")];
            tensor<int32, [4]> var_1306_begin_0 = const()[name = tensor<string, []>("op_1306_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_1306_end_0 = const()[name = tensor<string, []>("op_1306_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_1306_end_mask_0 = const()[name = tensor<string, []>("op_1306_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1306_cast = slice_by_index(begin = var_1306_begin_0, end = var_1306_end_0, end_mask = var_1306_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_1306_cast")];
            tensor<int32, [4]> var_1310_begin_0 = const()[name = tensor<string, []>("op_1310_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_1310_end_0 = const()[name = tensor<string, []>("op_1310_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_1310_end_mask_0 = const()[name = tensor<string, []>("op_1310_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1310_cast = slice_by_index(begin = var_1310_begin_0, end = var_1310_end_0, end_mask = var_1310_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_1310_cast")];
            tensor<int32, [4]> var_1314_begin_0 = const()[name = tensor<string, []>("op_1314_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_1314_end_0 = const()[name = tensor<string, []>("op_1314_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_1314_end_mask_0 = const()[name = tensor<string, []>("op_1314_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1314_cast = slice_by_index(begin = var_1314_begin_0, end = var_1314_end_0, end_mask = var_1314_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_1314_cast")];
            tensor<int32, [4]> var_1318_begin_0 = const()[name = tensor<string, []>("op_1318_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_1318_end_0 = const()[name = tensor<string, []>("op_1318_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_1318_end_mask_0 = const()[name = tensor<string, []>("op_1318_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1318_cast = slice_by_index(begin = var_1318_begin_0, end = var_1318_end_0, end_mask = var_1318_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_1318_cast")];
            tensor<string, []> var_1322_equation_0 = const()[name = tensor<string, []>("op_1322_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1322_cast = einsum(equation = var_1322_equation_0, values = (var_1244_cast, var_1201_cast))[name = tensor<string, []>("op_1322_cast")];
            tensor<fp16, []> var_1323_to_fp16 = const()[name = tensor<string, []>("op_1323_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_61_cast = mul(x = var_1322_cast, y = var_1323_to_fp16)[name = tensor<string, []>("aw_61_cast")];
            tensor<string, []> var_1326_equation_0 = const()[name = tensor<string, []>("op_1326_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1326_cast = einsum(equation = var_1326_equation_0, values = (var_1248_cast, var_1205_cast))[name = tensor<string, []>("op_1326_cast")];
            tensor<fp16, []> var_1327_to_fp16 = const()[name = tensor<string, []>("op_1327_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_63_cast = mul(x = var_1326_cast, y = var_1327_to_fp16)[name = tensor<string, []>("aw_63_cast")];
            tensor<string, []> var_1330_equation_0 = const()[name = tensor<string, []>("op_1330_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1330_cast = einsum(equation = var_1330_equation_0, values = (var_1252_cast, var_1209_cast))[name = tensor<string, []>("op_1330_cast")];
            tensor<fp16, []> var_1331_to_fp16 = const()[name = tensor<string, []>("op_1331_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_65_cast = mul(x = var_1330_cast, y = var_1331_to_fp16)[name = tensor<string, []>("aw_65_cast")];
            tensor<string, []> var_1334_equation_0 = const()[name = tensor<string, []>("op_1334_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1334_cast = einsum(equation = var_1334_equation_0, values = (var_1256_cast, var_1213_cast))[name = tensor<string, []>("op_1334_cast")];
            tensor<fp16, []> var_1335_to_fp16 = const()[name = tensor<string, []>("op_1335_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_67_cast = mul(x = var_1334_cast, y = var_1335_to_fp16)[name = tensor<string, []>("aw_67_cast")];
            tensor<string, []> var_1338_equation_0 = const()[name = tensor<string, []>("op_1338_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1338_cast = einsum(equation = var_1338_equation_0, values = (var_1260_cast, var_1217_cast))[name = tensor<string, []>("op_1338_cast")];
            tensor<fp16, []> var_1339_to_fp16 = const()[name = tensor<string, []>("op_1339_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_69_cast = mul(x = var_1338_cast, y = var_1339_to_fp16)[name = tensor<string, []>("aw_69_cast")];
            tensor<string, []> var_1342_equation_0 = const()[name = tensor<string, []>("op_1342_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1342_cast = einsum(equation = var_1342_equation_0, values = (var_1264_cast, var_1221_cast))[name = tensor<string, []>("op_1342_cast")];
            tensor<fp16, []> var_1343_to_fp16 = const()[name = tensor<string, []>("op_1343_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_71_cast = mul(x = var_1342_cast, y = var_1343_to_fp16)[name = tensor<string, []>("aw_71_cast")];
            tensor<string, []> var_1346_equation_0 = const()[name = tensor<string, []>("op_1346_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1346_cast = einsum(equation = var_1346_equation_0, values = (var_1268_cast, var_1225_cast))[name = tensor<string, []>("op_1346_cast")];
            tensor<fp16, []> var_1347_to_fp16 = const()[name = tensor<string, []>("op_1347_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_73_cast = mul(x = var_1346_cast, y = var_1347_to_fp16)[name = tensor<string, []>("aw_73_cast")];
            tensor<string, []> var_1350_equation_0 = const()[name = tensor<string, []>("op_1350_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1350_cast = einsum(equation = var_1350_equation_0, values = (var_1272_cast, var_1229_cast))[name = tensor<string, []>("op_1350_cast")];
            tensor<fp16, []> var_1351_to_fp16 = const()[name = tensor<string, []>("op_1351_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_75_cast = mul(x = var_1350_cast, y = var_1351_to_fp16)[name = tensor<string, []>("aw_75_cast")];
            tensor<string, []> var_1354_equation_0 = const()[name = tensor<string, []>("op_1354_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1354_cast = einsum(equation = var_1354_equation_0, values = (var_1276_cast, var_1233_cast))[name = tensor<string, []>("op_1354_cast")];
            tensor<fp16, []> var_1355_to_fp16 = const()[name = tensor<string, []>("op_1355_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_77_cast = mul(x = var_1354_cast, y = var_1355_to_fp16)[name = tensor<string, []>("aw_77_cast")];
            tensor<string, []> var_1358_equation_0 = const()[name = tensor<string, []>("op_1358_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1358_cast = einsum(equation = var_1358_equation_0, values = (var_1280_cast, var_1237_cast))[name = tensor<string, []>("op_1358_cast")];
            tensor<fp16, []> var_1359_to_fp16 = const()[name = tensor<string, []>("op_1359_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_79_cast = mul(x = var_1358_cast, y = var_1359_to_fp16)[name = tensor<string, []>("aw_79_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1361_cast = softmax(axis = var_298, x = aw_61_cast)[name = tensor<string, []>("op_1361_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1362_cast = softmax(axis = var_298, x = aw_63_cast)[name = tensor<string, []>("op_1362_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1363_cast = softmax(axis = var_298, x = aw_65_cast)[name = tensor<string, []>("op_1363_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1364_cast = softmax(axis = var_298, x = aw_67_cast)[name = tensor<string, []>("op_1364_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1365_cast = softmax(axis = var_298, x = aw_69_cast)[name = tensor<string, []>("op_1365_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1366_cast = softmax(axis = var_298, x = aw_71_cast)[name = tensor<string, []>("op_1366_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1367_cast = softmax(axis = var_298, x = aw_73_cast)[name = tensor<string, []>("op_1367_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1368_cast = softmax(axis = var_298, x = aw_75_cast)[name = tensor<string, []>("op_1368_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1369_cast = softmax(axis = var_298, x = aw_77_cast)[name = tensor<string, []>("op_1369_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1370_cast = softmax(axis = var_298, x = aw_79_cast)[name = tensor<string, []>("op_1370_cast")];
            tensor<string, []> var_1372_equation_0 = const()[name = tensor<string, []>("op_1372_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1372_cast = einsum(equation = var_1372_equation_0, values = (var_1282_cast, var_1361_cast))[name = tensor<string, []>("op_1372_cast")];
            tensor<string, []> var_1374_equation_0 = const()[name = tensor<string, []>("op_1374_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1374_cast = einsum(equation = var_1374_equation_0, values = (var_1286_cast, var_1362_cast))[name = tensor<string, []>("op_1374_cast")];
            tensor<string, []> var_1376_equation_0 = const()[name = tensor<string, []>("op_1376_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1376_cast = einsum(equation = var_1376_equation_0, values = (var_1290_cast, var_1363_cast))[name = tensor<string, []>("op_1376_cast")];
            tensor<string, []> var_1378_equation_0 = const()[name = tensor<string, []>("op_1378_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1378_cast = einsum(equation = var_1378_equation_0, values = (var_1294_cast, var_1364_cast))[name = tensor<string, []>("op_1378_cast")];
            tensor<string, []> var_1380_equation_0 = const()[name = tensor<string, []>("op_1380_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1380_cast = einsum(equation = var_1380_equation_0, values = (var_1298_cast, var_1365_cast))[name = tensor<string, []>("op_1380_cast")];
            tensor<string, []> var_1382_equation_0 = const()[name = tensor<string, []>("op_1382_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1382_cast = einsum(equation = var_1382_equation_0, values = (var_1302_cast, var_1366_cast))[name = tensor<string, []>("op_1382_cast")];
            tensor<string, []> var_1384_equation_0 = const()[name = tensor<string, []>("op_1384_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1384_cast = einsum(equation = var_1384_equation_0, values = (var_1306_cast, var_1367_cast))[name = tensor<string, []>("op_1384_cast")];
            tensor<string, []> var_1386_equation_0 = const()[name = tensor<string, []>("op_1386_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1386_cast = einsum(equation = var_1386_equation_0, values = (var_1310_cast, var_1368_cast))[name = tensor<string, []>("op_1386_cast")];
            tensor<string, []> var_1388_equation_0 = const()[name = tensor<string, []>("op_1388_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1388_cast = einsum(equation = var_1388_equation_0, values = (var_1314_cast, var_1369_cast))[name = tensor<string, []>("op_1388_cast")];
            tensor<string, []> var_1390_equation_0 = const()[name = tensor<string, []>("op_1390_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1390_cast = einsum(equation = var_1390_equation_0, values = (var_1318_cast, var_1370_cast))[name = tensor<string, []>("op_1390_cast")];
            tensor<bool, []> input_75_interleave_0 = const()[name = tensor<string, []>("input_75_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_75_cast = concat(axis = var_298, interleave = input_75_interleave_0, values = (var_1372_cast, var_1374_cast, var_1376_cast, var_1378_cast, var_1380_cast, var_1382_cast, var_1384_cast, var_1386_cast, var_1388_cast, var_1390_cast))[name = tensor<string, []>("input_75_cast")];
            tensor<int32, [2]> var_1396 = const()[name = tensor<string, []>("op_1396"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1398 = const()[name = tensor<string, []>("op_1398"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_1400_pad_type_0 = const()[name = tensor<string, []>("op_1400_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_1400_pad_0 = const()[name = tensor<string, []>("op_1400_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68784384)))];
            tensor<fp16, [640]> down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(69603648)))];
            tensor<fp16, [2, 640, 1, 4096]> var_1400_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_1398, groups = var_298, pad = var_1400_pad_0, pad_type = var_1400_pad_type_0, strides = var_1396, weight = down_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_75_cast)[name = tensor<string, []>("op_1400_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_11_cast = add(x = var_1400_cast, y = inputs_9_cast)[name = tensor<string, []>("inputs_11_cast")];
            tensor<int32, [1]> var_1404 = const()[name = tensor<string, []>("op_1404"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_11_cast = reduce_mean(axes = var_1404, keep_dims = var_293, x = inputs_11_cast)[name = tensor<string, []>("channels_mean_11_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_11_cast = sub(x = inputs_11_cast, y = channels_mean_11_cast)[name = tensor<string, []>("zero_mean_11_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_11_cast = mul(x = zero_mean_11_cast, y = zero_mean_11_cast)[name = tensor<string, []>("zero_mean_sq_11_cast")];
            tensor<int32, [1]> var_1408 = const()[name = tensor<string, []>("op_1408"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_1409_cast = reduce_mean(axes = var_1408, keep_dims = var_293, x = zero_mean_sq_11_cast)[name = tensor<string, []>("op_1409_cast")];
            tensor<fp16, []> var_1410_to_fp16 = const()[name = tensor<string, []>("op_1410_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_1411_cast = add(x = var_1409_cast, y = var_1410_to_fp16)[name = tensor<string, []>("op_1411_cast")];
            tensor<fp16, []> denom_11_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_11_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_11_cast = rsqrt(epsilon = denom_11_epsilon_0_to_fp16, x = var_1411_cast)[name = tensor<string, []>("denom_11_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_11_cast = mul(x = zero_mean_11_cast, y = denom_11_cast)[name = tensor<string, []>("out_11_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_1415_to_fp16 = const()[name = tensor<string, []>("op_1415_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(69604992)))];
            tensor<fp16, [2, 640, 1, 4096]> var_1416_cast = add(x = out_11_cast, y = var_1415_to_fp16)[name = tensor<string, []>("op_1416_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_1418_to_fp16 = const()[name = tensor<string, []>("op_1418_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(69606336)))];
            tensor<fp16, [2, 640, 1, 4096]> input_77_cast = mul(x = var_1416_cast, y = var_1418_to_fp16)[name = tensor<string, []>("input_77_cast")];
            tensor<int32, [2]> var_1426 = const()[name = tensor<string, []>("op_1426"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1428 = const()[name = tensor<string, []>("op_1428"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_1430_pad_type_0 = const()[name = tensor<string, []>("op_1430_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_1430_pad_0 = const()[name = tensor<string, []>("op_1430_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [5120, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [5120, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(69607680)))];
            tensor<fp16, [5120]> down_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76161344)))];
            tensor<fp16, [2, 5120, 1, 4096]> var_1430_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_1428, groups = var_298, pad = var_1430_pad_0, pad_type = var_1430_pad_type_0, strides = var_1426, weight = down_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_77_cast)[name = tensor<string, []>("op_1430_cast")];
            tensor<int32, [2]> var_1431_split_sizes_0 = const()[name = tensor<string, []>("op_1431_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_1431_axis_0 = const()[name = tensor<string, []>("op_1431_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 4096]> var_1431_cast_0, tensor<fp16, [2, 2560, 1, 4096]> var_1431_cast_1 = split(axis = var_1431_axis_0, split_sizes = var_1431_split_sizes_0, x = var_1430_cast)[name = tensor<string, []>("op_1431_cast")];
            tensor<string, []> var_1433_mode_0 = const()[name = tensor<string, []>("op_1433_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 4096]> var_1433_cast = gelu(mode = var_1433_mode_0, x = var_1431_cast_1)[name = tensor<string, []>("op_1433_cast")];
            tensor<fp16, [2, 2560, 1, 4096]> input_79_cast = mul(x = var_1431_cast_0, y = var_1433_cast)[name = tensor<string, []>("input_79_cast")];
            tensor<int32, [2]> var_1437 = const()[name = tensor<string, []>("op_1437"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1439 = const()[name = tensor<string, []>("op_1439"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_1441_pad_type_0 = const()[name = tensor<string, []>("op_1441_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_1441_pad_0 = const()[name = tensor<string, []>("op_1441_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2560, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor<fp16, [640, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76171648)))];
            tensor<fp16, [640]> down_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79448512)))];
            tensor<fp16, [2, 640, 1, 4096]> var_1441_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_1439, groups = var_298, pad = var_1441_pad_0, pad_type = var_1441_pad_type_0, strides = var_1437, weight = down_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_79_cast)[name = tensor<string, []>("op_1441_cast")];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_29_cast = add(x = var_1441_cast, y = inputs_11_cast)[name = tensor<string, []>("hidden_states_29_cast")];
            tensor<int32, [4]> var_1443 = const()[name = tensor<string, []>("op_1443"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> input_81_cast = reshape(shape = var_1443, x = hidden_states_29_cast)[name = tensor<string, []>("input_81_cast")];
            tensor<int32, [2]> var_1447 = const()[name = tensor<string, []>("op_1447"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1449 = const()[name = tensor<string, []>("op_1449"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_31_pad_type_0 = const()[name = tensor<string, []>("hidden_states_31_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_31_pad_0 = const()[name = tensor<string, []>("hidden_states_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_proj_out_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79449856)))];
            tensor<fp16, [640]> down_blocks_1_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_proj_out_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80269120)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_31_cast = conv(bias = down_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = var_1449, groups = var_298, pad = hidden_states_31_pad_0, pad_type = hidden_states_31_pad_type_0, strides = var_1447, weight = down_blocks_1_attentions_0_proj_out_weight_to_fp16, x = input_81_cast)[name = tensor<string, []>("hidden_states_31_cast")];
            tensor<fp16, [2, 640, 64, 64]> input_83_cast = add(x = hidden_states_31_cast, y = hidden_states_13_cast)[name = tensor<string, []>("input_83_cast")];
            tensor<int32, [5]> reshape_28_shape_0 = const()[name = tensor<string, []>("reshape_28_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 64, 64])];
            tensor<fp16, [2, 32, 20, 64, 64]> reshape_28_cast = reshape(shape = reshape_28_shape_0, x = input_83_cast)[name = tensor<string, []>("reshape_28_cast")];
            tensor<int32, [3]> reduce_mean_21_axes_0 = const()[name = tensor<string, []>("reduce_mean_21_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_21_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_21_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_21_cast = reduce_mean(axes = reduce_mean_21_axes_0, keep_dims = reduce_mean_21_keep_dims_0, x = reshape_28_cast)[name = tensor<string, []>("reduce_mean_21_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> sub_14_cast = sub(x = reshape_28_cast, y = reduce_mean_21_cast)[name = tensor<string, []>("sub_14_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> square_7_cast = square(x = sub_14_cast)[name = tensor<string, []>("square_7_cast")];
            tensor<int32, [3]> reduce_mean_23_axes_0 = const()[name = tensor<string, []>("reduce_mean_23_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_23_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_23_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_23_cast = reduce_mean(axes = reduce_mean_23_axes_0, keep_dims = reduce_mean_23_keep_dims_0, x = square_7_cast)[name = tensor<string, []>("reduce_mean_23_cast")];
            tensor<fp16, []> add_14_y_0_to_fp16 = const()[name = tensor<string, []>("add_14_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_14_cast = add(x = reduce_mean_23_cast, y = add_14_y_0_to_fp16)[name = tensor<string, []>("add_14_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_7_cast = sqrt(x = add_14_cast)[name = tensor<string, []>("sqrt_7_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_7_cast = real_div(x = sub_14_cast, y = sqrt_7_cast)[name = tensor<string, []>("real_div_7_cast")];
            tensor<int32, [4]> reshape_29_shape_0 = const()[name = tensor<string, []>("reshape_29_shape_0"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> reshape_29_cast = reshape(shape = reshape_29_shape_0, x = real_div_7_cast)[name = tensor<string, []>("reshape_29_cast")];
            tensor<fp16, [640]> add_15_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_15_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80270464)))];
            tensor<fp16, [640]> add_15_beta_0_to_fp16 = const()[name = tensor<string, []>("add_15_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80271808)))];
            tensor<fp16, []> add_15_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_15_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 64, 64]> add_15_cast = batch_norm(beta = add_15_beta_0_to_fp16, epsilon = add_15_epsilon_0_to_fp16, gamma = add_15_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_29_cast)[name = tensor<string, []>("add_15_cast")];
            tensor<fp16, [2, 640, 64, 64]> input_87_cast = silu(x = add_15_cast)[name = tensor<string, []>("input_87_cast")];
            tensor<int32, [2]> var_1464 = const()[name = tensor<string, []>("op_1464"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1466 = const()[name = tensor<string, []>("op_1466"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_33_pad_type_0 = const()[name = tensor<string, []>("hidden_states_33_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_33_pad_0 = const()[name = tensor<string, []>("hidden_states_33_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 640, 3, 3]> down_blocks_1_resnets_1_conv1_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_1_conv1_weight_to_fp16"), val = tensor<fp16, [640, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80273152)))];
            tensor<fp16, [640]> down_blocks_1_resnets_1_conv1_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_1_conv1_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87646016)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_33_cast = conv(bias = down_blocks_1_resnets_1_conv1_bias_to_fp16, dilations = var_1466, groups = var_298, pad = hidden_states_33_pad_0, pad_type = hidden_states_33_pad_type_0, strides = var_1464, weight = down_blocks_1_resnets_1_conv1_weight_to_fp16, x = input_87_cast)[name = tensor<string, []>("hidden_states_33_cast")];
            tensor<int32, [2]> var_1472 = const()[name = tensor<string, []>("op_1472"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1474 = const()[name = tensor<string, []>("op_1474"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_7_pad_type_0 = const()[name = tensor<string, []>("temb_7_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_7_pad_0 = const()[name = tensor<string, []>("temb_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1280, 1, 1]> down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [640, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87647360)))];
            tensor<fp16, [640]> down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89285824)))];
            tensor<fp16, [2, 640, 1, 1]> temb_7_cast = conv(bias = down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_1474, groups = var_298, pad = temb_7_pad_0, pad_type = temb_7_pad_type_0, strides = var_1472, weight = down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16, x = input_23_cast)[name = tensor<string, []>("temb_7_cast")];
            tensor<fp16, [2, 640, 64, 64]> input_91_cast = add(x = hidden_states_33_cast, y = temb_7_cast)[name = tensor<string, []>("input_91_cast")];
            tensor<int32, [5]> reshape_32_shape_0 = const()[name = tensor<string, []>("reshape_32_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 64, 64])];
            tensor<fp16, [2, 32, 20, 64, 64]> reshape_32_cast = reshape(shape = reshape_32_shape_0, x = input_91_cast)[name = tensor<string, []>("reshape_32_cast")];
            tensor<int32, [3]> reduce_mean_24_axes_0 = const()[name = tensor<string, []>("reduce_mean_24_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_24_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_24_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_24_cast = reduce_mean(axes = reduce_mean_24_axes_0, keep_dims = reduce_mean_24_keep_dims_0, x = reshape_32_cast)[name = tensor<string, []>("reduce_mean_24_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> sub_16_cast = sub(x = reshape_32_cast, y = reduce_mean_24_cast)[name = tensor<string, []>("sub_16_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> square_8_cast = square(x = sub_16_cast)[name = tensor<string, []>("square_8_cast")];
            tensor<int32, [3]> reduce_mean_26_axes_0 = const()[name = tensor<string, []>("reduce_mean_26_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_26_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_26_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_26_cast = reduce_mean(axes = reduce_mean_26_axes_0, keep_dims = reduce_mean_26_keep_dims_0, x = square_8_cast)[name = tensor<string, []>("reduce_mean_26_cast")];
            tensor<fp16, []> add_16_y_0_to_fp16 = const()[name = tensor<string, []>("add_16_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_16_cast = add(x = reduce_mean_26_cast, y = add_16_y_0_to_fp16)[name = tensor<string, []>("add_16_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_8_cast = sqrt(x = add_16_cast)[name = tensor<string, []>("sqrt_8_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_8_cast = real_div(x = sub_16_cast, y = sqrt_8_cast)[name = tensor<string, []>("real_div_8_cast")];
            tensor<int32, [4]> reshape_33_shape_0 = const()[name = tensor<string, []>("reshape_33_shape_0"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> reshape_33_cast = reshape(shape = reshape_33_shape_0, x = real_div_8_cast)[name = tensor<string, []>("reshape_33_cast")];
            tensor<fp16, [640]> add_17_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_17_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89287168)))];
            tensor<fp16, [640]> add_17_beta_0_to_fp16 = const()[name = tensor<string, []>("add_17_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89288512)))];
            tensor<fp16, []> add_17_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_17_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 64, 64]> add_17_cast = batch_norm(beta = add_17_beta_0_to_fp16, epsilon = add_17_epsilon_0_to_fp16, gamma = add_17_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_33_cast)[name = tensor<string, []>("add_17_cast")];
            tensor<fp16, [2, 640, 64, 64]> input_95_cast = silu(x = add_17_cast)[name = tensor<string, []>("input_95_cast")];
            tensor<int32, [2]> var_1484 = const()[name = tensor<string, []>("op_1484"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1486 = const()[name = tensor<string, []>("op_1486"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_35_pad_type_0 = const()[name = tensor<string, []>("hidden_states_35_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_35_pad_0 = const()[name = tensor<string, []>("hidden_states_35_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 640, 3, 3]> down_blocks_1_resnets_1_conv2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_1_conv2_weight_to_fp16"), val = tensor<fp16, [640, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89289856)))];
            tensor<fp16, [640]> down_blocks_1_resnets_1_conv2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_1_conv2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96662720)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_35_cast = conv(bias = down_blocks_1_resnets_1_conv2_bias_to_fp16, dilations = var_1486, groups = var_298, pad = hidden_states_35_pad_0, pad_type = hidden_states_35_pad_type_0, strides = var_1484, weight = down_blocks_1_resnets_1_conv2_weight_to_fp16, x = input_95_cast)[name = tensor<string, []>("hidden_states_35_cast")];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_37_cast = add(x = input_83_cast, y = hidden_states_35_cast)[name = tensor<string, []>("hidden_states_37_cast")];
            tensor<int32, [5]> reshape_36_shape_0 = const()[name = tensor<string, []>("reshape_36_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 64, 64])];
            tensor<fp16, [2, 32, 20, 64, 64]> reshape_36_cast = reshape(shape = reshape_36_shape_0, x = hidden_states_37_cast)[name = tensor<string, []>("reshape_36_cast")];
            tensor<int32, [3]> reduce_mean_27_axes_0 = const()[name = tensor<string, []>("reduce_mean_27_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_27_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_27_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_27_cast = reduce_mean(axes = reduce_mean_27_axes_0, keep_dims = reduce_mean_27_keep_dims_0, x = reshape_36_cast)[name = tensor<string, []>("reduce_mean_27_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> sub_18_cast = sub(x = reshape_36_cast, y = reduce_mean_27_cast)[name = tensor<string, []>("sub_18_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> square_9_cast = square(x = sub_18_cast)[name = tensor<string, []>("square_9_cast")];
            tensor<int32, [3]> reduce_mean_29_axes_0 = const()[name = tensor<string, []>("reduce_mean_29_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_29_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_29_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_29_cast = reduce_mean(axes = reduce_mean_29_axes_0, keep_dims = reduce_mean_29_keep_dims_0, x = square_9_cast)[name = tensor<string, []>("reduce_mean_29_cast")];
            tensor<fp16, []> add_18_y_0_to_fp16 = const()[name = tensor<string, []>("add_18_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_18_cast = add(x = reduce_mean_29_cast, y = add_18_y_0_to_fp16)[name = tensor<string, []>("add_18_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_9_cast = sqrt(x = add_18_cast)[name = tensor<string, []>("sqrt_9_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_9_cast = real_div(x = sub_18_cast, y = sqrt_9_cast)[name = tensor<string, []>("real_div_9_cast")];
            tensor<int32, [4]> reshape_37_shape_0 = const()[name = tensor<string, []>("reshape_37_shape_0"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> reshape_37_cast = reshape(shape = reshape_37_shape_0, x = real_div_9_cast)[name = tensor<string, []>("reshape_37_cast")];
            tensor<fp16, [640]> add_19_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_19_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96664064)))];
            tensor<fp16, [640]> add_19_beta_0_to_fp16 = const()[name = tensor<string, []>("add_19_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96665408)))];
            tensor<fp16, []> add_19_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_19_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 64, 64]> add_19_cast = batch_norm(beta = add_19_beta_0_to_fp16, epsilon = add_19_epsilon_0_to_fp16, gamma = add_19_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_37_cast)[name = tensor<string, []>("add_19_cast")];
            tensor<int32, [2]> var_1508 = const()[name = tensor<string, []>("op_1508"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1510 = const()[name = tensor<string, []>("op_1510"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_39_pad_type_0 = const()[name = tensor<string, []>("hidden_states_39_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_39_pad_0 = const()[name = tensor<string, []>("hidden_states_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_proj_in_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96666752)))];
            tensor<fp16, [640]> down_blocks_1_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_proj_in_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97486016)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_39_cast = conv(bias = down_blocks_1_attentions_1_proj_in_bias_to_fp16, dilations = var_1510, groups = var_298, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = var_1508, weight = down_blocks_1_attentions_1_proj_in_weight_to_fp16, x = add_19_cast)[name = tensor<string, []>("hidden_states_39_cast")];
            tensor<int32, [4]> var_1515 = const()[name = tensor<string, []>("op_1515"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<fp16, [2, 640, 1, 4096]> inputs_13_cast = reshape(shape = var_1515, x = hidden_states_39_cast)[name = tensor<string, []>("inputs_13_cast")];
            tensor<int32, [1]> var_1525 = const()[name = tensor<string, []>("op_1525"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_13_cast = reduce_mean(axes = var_1525, keep_dims = var_293, x = inputs_13_cast)[name = tensor<string, []>("channels_mean_13_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_13_cast = sub(x = inputs_13_cast, y = channels_mean_13_cast)[name = tensor<string, []>("zero_mean_13_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_13_cast = mul(x = zero_mean_13_cast, y = zero_mean_13_cast)[name = tensor<string, []>("zero_mean_sq_13_cast")];
            tensor<int32, [1]> var_1529 = const()[name = tensor<string, []>("op_1529"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_1530_cast = reduce_mean(axes = var_1529, keep_dims = var_293, x = zero_mean_sq_13_cast)[name = tensor<string, []>("op_1530_cast")];
            tensor<fp16, []> var_1531_to_fp16 = const()[name = tensor<string, []>("op_1531_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_1532_cast = add(x = var_1530_cast, y = var_1531_to_fp16)[name = tensor<string, []>("op_1532_cast")];
            tensor<fp16, []> denom_13_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_13_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_13_cast = rsqrt(epsilon = denom_13_epsilon_0_to_fp16, x = var_1532_cast)[name = tensor<string, []>("denom_13_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_13_cast = mul(x = zero_mean_13_cast, y = denom_13_cast)[name = tensor<string, []>("out_13_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_1536_to_fp16 = const()[name = tensor<string, []>("op_1536_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97487360)))];
            tensor<fp16, [2, 640, 1, 4096]> var_1537_cast = add(x = out_13_cast, y = var_1536_to_fp16)[name = tensor<string, []>("op_1537_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_1539_to_fp16 = const()[name = tensor<string, []>("op_1539_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97488704)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_41_cast = mul(x = var_1537_cast, y = var_1539_to_fp16)[name = tensor<string, []>("hidden_states_41_cast")];
            tensor<int32, [2]> var_1546 = const()[name = tensor<string, []>("op_1546"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1548 = const()[name = tensor<string, []>("op_1548"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_9_pad_type_0 = const()[name = tensor<string, []>("q_9_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_9_pad_0 = const()[name = tensor<string, []>("q_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97490048)))];
            tensor<fp16, [2, 640, 1, 4096]> q_9_cast = conv(dilations = var_1548, groups = var_298, pad = q_9_pad_0, pad_type = q_9_pad_type_0, strides = var_1546, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_41_cast)[name = tensor<string, []>("q_9_cast")];
            tensor<int32, [2]> var_1552 = const()[name = tensor<string, []>("op_1552"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1554 = const()[name = tensor<string, []>("op_1554"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_17_pad_type_0 = const()[name = tensor<string, []>("k_17_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_17_pad_0 = const()[name = tensor<string, []>("k_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98309312)))];
            tensor<fp16, [2, 640, 1, 4096]> k_17_cast = conv(dilations = var_1554, groups = var_298, pad = k_17_pad_0, pad_type = k_17_pad_type_0, strides = var_1552, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_41_cast)[name = tensor<string, []>("k_17_cast")];
            tensor<int32, [2]> var_1558 = const()[name = tensor<string, []>("op_1558"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1560 = const()[name = tensor<string, []>("op_1560"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_9_pad_type_0 = const()[name = tensor<string, []>("v_9_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_9_pad_0 = const()[name = tensor<string, []>("v_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99128576)))];
            tensor<fp16, [2, 640, 1, 4096]> v_9_cast = conv(dilations = var_1560, groups = var_298, pad = v_9_pad_0, pad_type = v_9_pad_type_0, strides = var_1558, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_41_cast)[name = tensor<string, []>("v_9_cast")];
            tensor<int32, [4]> var_1564_begin_0 = const()[name = tensor<string, []>("op_1564_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1564_end_0 = const()[name = tensor<string, []>("op_1564_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_1564_end_mask_0 = const()[name = tensor<string, []>("op_1564_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1564_cast = slice_by_index(begin = var_1564_begin_0, end = var_1564_end_0, end_mask = var_1564_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_1564_cast")];
            tensor<int32, [4]> var_1568_begin_0 = const()[name = tensor<string, []>("op_1568_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1568_end_0 = const()[name = tensor<string, []>("op_1568_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_1568_end_mask_0 = const()[name = tensor<string, []>("op_1568_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1568_cast = slice_by_index(begin = var_1568_begin_0, end = var_1568_end_0, end_mask = var_1568_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_1568_cast")];
            tensor<int32, [4]> var_1572_begin_0 = const()[name = tensor<string, []>("op_1572_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1572_end_0 = const()[name = tensor<string, []>("op_1572_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_1572_end_mask_0 = const()[name = tensor<string, []>("op_1572_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1572_cast = slice_by_index(begin = var_1572_begin_0, end = var_1572_end_0, end_mask = var_1572_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_1572_cast")];
            tensor<int32, [4]> var_1576_begin_0 = const()[name = tensor<string, []>("op_1576_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1576_end_0 = const()[name = tensor<string, []>("op_1576_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_1576_end_mask_0 = const()[name = tensor<string, []>("op_1576_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1576_cast = slice_by_index(begin = var_1576_begin_0, end = var_1576_end_0, end_mask = var_1576_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_1576_cast")];
            tensor<int32, [4]> var_1580_begin_0 = const()[name = tensor<string, []>("op_1580_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1580_end_0 = const()[name = tensor<string, []>("op_1580_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_1580_end_mask_0 = const()[name = tensor<string, []>("op_1580_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1580_cast = slice_by_index(begin = var_1580_begin_0, end = var_1580_end_0, end_mask = var_1580_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_1580_cast")];
            tensor<int32, [4]> var_1584_begin_0 = const()[name = tensor<string, []>("op_1584_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1584_end_0 = const()[name = tensor<string, []>("op_1584_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_1584_end_mask_0 = const()[name = tensor<string, []>("op_1584_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1584_cast = slice_by_index(begin = var_1584_begin_0, end = var_1584_end_0, end_mask = var_1584_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_1584_cast")];
            tensor<int32, [4]> var_1588_begin_0 = const()[name = tensor<string, []>("op_1588_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_1588_end_0 = const()[name = tensor<string, []>("op_1588_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_1588_end_mask_0 = const()[name = tensor<string, []>("op_1588_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1588_cast = slice_by_index(begin = var_1588_begin_0, end = var_1588_end_0, end_mask = var_1588_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_1588_cast")];
            tensor<int32, [4]> var_1592_begin_0 = const()[name = tensor<string, []>("op_1592_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_1592_end_0 = const()[name = tensor<string, []>("op_1592_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_1592_end_mask_0 = const()[name = tensor<string, []>("op_1592_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1592_cast = slice_by_index(begin = var_1592_begin_0, end = var_1592_end_0, end_mask = var_1592_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_1592_cast")];
            tensor<int32, [4]> var_1596_begin_0 = const()[name = tensor<string, []>("op_1596_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_1596_end_0 = const()[name = tensor<string, []>("op_1596_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_1596_end_mask_0 = const()[name = tensor<string, []>("op_1596_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1596_cast = slice_by_index(begin = var_1596_begin_0, end = var_1596_end_0, end_mask = var_1596_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_1596_cast")];
            tensor<int32, [4]> var_1600_begin_0 = const()[name = tensor<string, []>("op_1600_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_1600_end_0 = const()[name = tensor<string, []>("op_1600_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_1600_end_mask_0 = const()[name = tensor<string, []>("op_1600_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1600_cast = slice_by_index(begin = var_1600_begin_0, end = var_1600_end_0, end_mask = var_1600_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_1600_cast")];
            tensor<int32, [4]> k_19_perm_0 = const()[name = tensor<string, []>("k_19_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_1607_begin_0 = const()[name = tensor<string, []>("op_1607_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1607_end_0 = const()[name = tensor<string, []>("op_1607_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_1607_end_mask_0 = const()[name = tensor<string, []>("op_1607_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 640]> transpose_135 = transpose(perm = k_19_perm_0, x = k_17_cast)[name = tensor<string, []>("transpose_135")];
            tensor<fp16, [2, 4096, 1, 64]> var_1607_cast = slice_by_index(begin = var_1607_begin_0, end = var_1607_end_0, end_mask = var_1607_end_mask_0, x = transpose_135)[name = tensor<string, []>("op_1607_cast")];
            tensor<int32, [4]> var_1611_begin_0 = const()[name = tensor<string, []>("op_1611_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_1611_end_0 = const()[name = tensor<string, []>("op_1611_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_1611_end_mask_0 = const()[name = tensor<string, []>("op_1611_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1611_cast = slice_by_index(begin = var_1611_begin_0, end = var_1611_end_0, end_mask = var_1611_end_mask_0, x = transpose_135)[name = tensor<string, []>("op_1611_cast")];
            tensor<int32, [4]> var_1615_begin_0 = const()[name = tensor<string, []>("op_1615_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_1615_end_0 = const()[name = tensor<string, []>("op_1615_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_1615_end_mask_0 = const()[name = tensor<string, []>("op_1615_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1615_cast = slice_by_index(begin = var_1615_begin_0, end = var_1615_end_0, end_mask = var_1615_end_mask_0, x = transpose_135)[name = tensor<string, []>("op_1615_cast")];
            tensor<int32, [4]> var_1619_begin_0 = const()[name = tensor<string, []>("op_1619_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_1619_end_0 = const()[name = tensor<string, []>("op_1619_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_1619_end_mask_0 = const()[name = tensor<string, []>("op_1619_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1619_cast = slice_by_index(begin = var_1619_begin_0, end = var_1619_end_0, end_mask = var_1619_end_mask_0, x = transpose_135)[name = tensor<string, []>("op_1619_cast")];
            tensor<int32, [4]> var_1623_begin_0 = const()[name = tensor<string, []>("op_1623_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_1623_end_0 = const()[name = tensor<string, []>("op_1623_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_1623_end_mask_0 = const()[name = tensor<string, []>("op_1623_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1623_cast = slice_by_index(begin = var_1623_begin_0, end = var_1623_end_0, end_mask = var_1623_end_mask_0, x = transpose_135)[name = tensor<string, []>("op_1623_cast")];
            tensor<int32, [4]> var_1627_begin_0 = const()[name = tensor<string, []>("op_1627_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_1627_end_0 = const()[name = tensor<string, []>("op_1627_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 384])];
            tensor<bool, [4]> var_1627_end_mask_0 = const()[name = tensor<string, []>("op_1627_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1627_cast = slice_by_index(begin = var_1627_begin_0, end = var_1627_end_0, end_mask = var_1627_end_mask_0, x = transpose_135)[name = tensor<string, []>("op_1627_cast")];
            tensor<int32, [4]> var_1631_begin_0 = const()[name = tensor<string, []>("op_1631_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_1631_end_0 = const()[name = tensor<string, []>("op_1631_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 448])];
            tensor<bool, [4]> var_1631_end_mask_0 = const()[name = tensor<string, []>("op_1631_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1631_cast = slice_by_index(begin = var_1631_begin_0, end = var_1631_end_0, end_mask = var_1631_end_mask_0, x = transpose_135)[name = tensor<string, []>("op_1631_cast")];
            tensor<int32, [4]> var_1635_begin_0 = const()[name = tensor<string, []>("op_1635_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_1635_end_0 = const()[name = tensor<string, []>("op_1635_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 512])];
            tensor<bool, [4]> var_1635_end_mask_0 = const()[name = tensor<string, []>("op_1635_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1635_cast = slice_by_index(begin = var_1635_begin_0, end = var_1635_end_0, end_mask = var_1635_end_mask_0, x = transpose_135)[name = tensor<string, []>("op_1635_cast")];
            tensor<int32, [4]> var_1639_begin_0 = const()[name = tensor<string, []>("op_1639_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1639_end_0 = const()[name = tensor<string, []>("op_1639_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 576])];
            tensor<bool, [4]> var_1639_end_mask_0 = const()[name = tensor<string, []>("op_1639_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1639_cast = slice_by_index(begin = var_1639_begin_0, end = var_1639_end_0, end_mask = var_1639_end_mask_0, x = transpose_135)[name = tensor<string, []>("op_1639_cast")];
            tensor<int32, [4]> var_1643_begin_0 = const()[name = tensor<string, []>("op_1643_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_1643_end_0 = const()[name = tensor<string, []>("op_1643_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 640])];
            tensor<bool, [4]> var_1643_end_mask_0 = const()[name = tensor<string, []>("op_1643_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1643_cast = slice_by_index(begin = var_1643_begin_0, end = var_1643_end_0, end_mask = var_1643_end_mask_0, x = transpose_135)[name = tensor<string, []>("op_1643_cast")];
            tensor<int32, [4]> var_1645_begin_0 = const()[name = tensor<string, []>("op_1645_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1645_end_0 = const()[name = tensor<string, []>("op_1645_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_1645_end_mask_0 = const()[name = tensor<string, []>("op_1645_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1645_cast = slice_by_index(begin = var_1645_begin_0, end = var_1645_end_0, end_mask = var_1645_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_1645_cast")];
            tensor<int32, [4]> var_1649_begin_0 = const()[name = tensor<string, []>("op_1649_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1649_end_0 = const()[name = tensor<string, []>("op_1649_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_1649_end_mask_0 = const()[name = tensor<string, []>("op_1649_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1649_cast = slice_by_index(begin = var_1649_begin_0, end = var_1649_end_0, end_mask = var_1649_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_1649_cast")];
            tensor<int32, [4]> var_1653_begin_0 = const()[name = tensor<string, []>("op_1653_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1653_end_0 = const()[name = tensor<string, []>("op_1653_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_1653_end_mask_0 = const()[name = tensor<string, []>("op_1653_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1653_cast = slice_by_index(begin = var_1653_begin_0, end = var_1653_end_0, end_mask = var_1653_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_1653_cast")];
            tensor<int32, [4]> var_1657_begin_0 = const()[name = tensor<string, []>("op_1657_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1657_end_0 = const()[name = tensor<string, []>("op_1657_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_1657_end_mask_0 = const()[name = tensor<string, []>("op_1657_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1657_cast = slice_by_index(begin = var_1657_begin_0, end = var_1657_end_0, end_mask = var_1657_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_1657_cast")];
            tensor<int32, [4]> var_1661_begin_0 = const()[name = tensor<string, []>("op_1661_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1661_end_0 = const()[name = tensor<string, []>("op_1661_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_1661_end_mask_0 = const()[name = tensor<string, []>("op_1661_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1661_cast = slice_by_index(begin = var_1661_begin_0, end = var_1661_end_0, end_mask = var_1661_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_1661_cast")];
            tensor<int32, [4]> var_1665_begin_0 = const()[name = tensor<string, []>("op_1665_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1665_end_0 = const()[name = tensor<string, []>("op_1665_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_1665_end_mask_0 = const()[name = tensor<string, []>("op_1665_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1665_cast = slice_by_index(begin = var_1665_begin_0, end = var_1665_end_0, end_mask = var_1665_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_1665_cast")];
            tensor<int32, [4]> var_1669_begin_0 = const()[name = tensor<string, []>("op_1669_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_1669_end_0 = const()[name = tensor<string, []>("op_1669_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_1669_end_mask_0 = const()[name = tensor<string, []>("op_1669_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1669_cast = slice_by_index(begin = var_1669_begin_0, end = var_1669_end_0, end_mask = var_1669_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_1669_cast")];
            tensor<int32, [4]> var_1673_begin_0 = const()[name = tensor<string, []>("op_1673_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_1673_end_0 = const()[name = tensor<string, []>("op_1673_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_1673_end_mask_0 = const()[name = tensor<string, []>("op_1673_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1673_cast = slice_by_index(begin = var_1673_begin_0, end = var_1673_end_0, end_mask = var_1673_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_1673_cast")];
            tensor<int32, [4]> var_1677_begin_0 = const()[name = tensor<string, []>("op_1677_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_1677_end_0 = const()[name = tensor<string, []>("op_1677_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_1677_end_mask_0 = const()[name = tensor<string, []>("op_1677_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1677_cast = slice_by_index(begin = var_1677_begin_0, end = var_1677_end_0, end_mask = var_1677_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_1677_cast")];
            tensor<int32, [4]> var_1681_begin_0 = const()[name = tensor<string, []>("op_1681_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_1681_end_0 = const()[name = tensor<string, []>("op_1681_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_1681_end_mask_0 = const()[name = tensor<string, []>("op_1681_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1681_cast = slice_by_index(begin = var_1681_begin_0, end = var_1681_end_0, end_mask = var_1681_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_1681_cast")];
            tensor<string, []> var_1685_equation_0 = const()[name = tensor<string, []>("op_1685_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1685_cast = einsum(equation = var_1685_equation_0, values = (var_1607_cast, var_1564_cast))[name = tensor<string, []>("op_1685_cast")];
            tensor<fp16, []> var_1686_to_fp16 = const()[name = tensor<string, []>("op_1686_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_81_cast = mul(x = var_1685_cast, y = var_1686_to_fp16)[name = tensor<string, []>("aw_81_cast")];
            tensor<string, []> var_1689_equation_0 = const()[name = tensor<string, []>("op_1689_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1689_cast = einsum(equation = var_1689_equation_0, values = (var_1611_cast, var_1568_cast))[name = tensor<string, []>("op_1689_cast")];
            tensor<fp16, []> var_1690_to_fp16 = const()[name = tensor<string, []>("op_1690_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_83_cast = mul(x = var_1689_cast, y = var_1690_to_fp16)[name = tensor<string, []>("aw_83_cast")];
            tensor<string, []> var_1693_equation_0 = const()[name = tensor<string, []>("op_1693_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1693_cast = einsum(equation = var_1693_equation_0, values = (var_1615_cast, var_1572_cast))[name = tensor<string, []>("op_1693_cast")];
            tensor<fp16, []> var_1694_to_fp16 = const()[name = tensor<string, []>("op_1694_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_85_cast = mul(x = var_1693_cast, y = var_1694_to_fp16)[name = tensor<string, []>("aw_85_cast")];
            tensor<string, []> var_1697_equation_0 = const()[name = tensor<string, []>("op_1697_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1697_cast = einsum(equation = var_1697_equation_0, values = (var_1619_cast, var_1576_cast))[name = tensor<string, []>("op_1697_cast")];
            tensor<fp16, []> var_1698_to_fp16 = const()[name = tensor<string, []>("op_1698_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_87_cast = mul(x = var_1697_cast, y = var_1698_to_fp16)[name = tensor<string, []>("aw_87_cast")];
            tensor<string, []> var_1701_equation_0 = const()[name = tensor<string, []>("op_1701_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1701_cast = einsum(equation = var_1701_equation_0, values = (var_1623_cast, var_1580_cast))[name = tensor<string, []>("op_1701_cast")];
            tensor<fp16, []> var_1702_to_fp16 = const()[name = tensor<string, []>("op_1702_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_89_cast = mul(x = var_1701_cast, y = var_1702_to_fp16)[name = tensor<string, []>("aw_89_cast")];
            tensor<string, []> var_1705_equation_0 = const()[name = tensor<string, []>("op_1705_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1705_cast = einsum(equation = var_1705_equation_0, values = (var_1627_cast, var_1584_cast))[name = tensor<string, []>("op_1705_cast")];
            tensor<fp16, []> var_1706_to_fp16 = const()[name = tensor<string, []>("op_1706_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_91_cast = mul(x = var_1705_cast, y = var_1706_to_fp16)[name = tensor<string, []>("aw_91_cast")];
            tensor<string, []> var_1709_equation_0 = const()[name = tensor<string, []>("op_1709_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1709_cast = einsum(equation = var_1709_equation_0, values = (var_1631_cast, var_1588_cast))[name = tensor<string, []>("op_1709_cast")];
            tensor<fp16, []> var_1710_to_fp16 = const()[name = tensor<string, []>("op_1710_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_93_cast = mul(x = var_1709_cast, y = var_1710_to_fp16)[name = tensor<string, []>("aw_93_cast")];
            tensor<string, []> var_1713_equation_0 = const()[name = tensor<string, []>("op_1713_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1713_cast = einsum(equation = var_1713_equation_0, values = (var_1635_cast, var_1592_cast))[name = tensor<string, []>("op_1713_cast")];
            tensor<fp16, []> var_1714_to_fp16 = const()[name = tensor<string, []>("op_1714_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_95_cast = mul(x = var_1713_cast, y = var_1714_to_fp16)[name = tensor<string, []>("aw_95_cast")];
            tensor<string, []> var_1717_equation_0 = const()[name = tensor<string, []>("op_1717_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1717_cast = einsum(equation = var_1717_equation_0, values = (var_1639_cast, var_1596_cast))[name = tensor<string, []>("op_1717_cast")];
            tensor<fp16, []> var_1718_to_fp16 = const()[name = tensor<string, []>("op_1718_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_97_cast = mul(x = var_1717_cast, y = var_1718_to_fp16)[name = tensor<string, []>("aw_97_cast")];
            tensor<string, []> var_1721_equation_0 = const()[name = tensor<string, []>("op_1721_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1721_cast = einsum(equation = var_1721_equation_0, values = (var_1643_cast, var_1600_cast))[name = tensor<string, []>("op_1721_cast")];
            tensor<fp16, []> var_1722_to_fp16 = const()[name = tensor<string, []>("op_1722_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_99_cast = mul(x = var_1721_cast, y = var_1722_to_fp16)[name = tensor<string, []>("aw_99_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1724_cast = softmax(axis = var_298, x = aw_81_cast)[name = tensor<string, []>("op_1724_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1725_cast = softmax(axis = var_298, x = aw_83_cast)[name = tensor<string, []>("op_1725_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1726_cast = softmax(axis = var_298, x = aw_85_cast)[name = tensor<string, []>("op_1726_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1727_cast = softmax(axis = var_298, x = aw_87_cast)[name = tensor<string, []>("op_1727_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1728_cast = softmax(axis = var_298, x = aw_89_cast)[name = tensor<string, []>("op_1728_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1729_cast = softmax(axis = var_298, x = aw_91_cast)[name = tensor<string, []>("op_1729_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1730_cast = softmax(axis = var_298, x = aw_93_cast)[name = tensor<string, []>("op_1730_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1731_cast = softmax(axis = var_298, x = aw_95_cast)[name = tensor<string, []>("op_1731_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1732_cast = softmax(axis = var_298, x = aw_97_cast)[name = tensor<string, []>("op_1732_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_1733_cast = softmax(axis = var_298, x = aw_99_cast)[name = tensor<string, []>("op_1733_cast")];
            tensor<string, []> var_1735_equation_0 = const()[name = tensor<string, []>("op_1735_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1735_cast = einsum(equation = var_1735_equation_0, values = (var_1645_cast, var_1724_cast))[name = tensor<string, []>("op_1735_cast")];
            tensor<string, []> var_1737_equation_0 = const()[name = tensor<string, []>("op_1737_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1737_cast = einsum(equation = var_1737_equation_0, values = (var_1649_cast, var_1725_cast))[name = tensor<string, []>("op_1737_cast")];
            tensor<string, []> var_1739_equation_0 = const()[name = tensor<string, []>("op_1739_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1739_cast = einsum(equation = var_1739_equation_0, values = (var_1653_cast, var_1726_cast))[name = tensor<string, []>("op_1739_cast")];
            tensor<string, []> var_1741_equation_0 = const()[name = tensor<string, []>("op_1741_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1741_cast = einsum(equation = var_1741_equation_0, values = (var_1657_cast, var_1727_cast))[name = tensor<string, []>("op_1741_cast")];
            tensor<string, []> var_1743_equation_0 = const()[name = tensor<string, []>("op_1743_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1743_cast = einsum(equation = var_1743_equation_0, values = (var_1661_cast, var_1728_cast))[name = tensor<string, []>("op_1743_cast")];
            tensor<string, []> var_1745_equation_0 = const()[name = tensor<string, []>("op_1745_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1745_cast = einsum(equation = var_1745_equation_0, values = (var_1665_cast, var_1729_cast))[name = tensor<string, []>("op_1745_cast")];
            tensor<string, []> var_1747_equation_0 = const()[name = tensor<string, []>("op_1747_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1747_cast = einsum(equation = var_1747_equation_0, values = (var_1669_cast, var_1730_cast))[name = tensor<string, []>("op_1747_cast")];
            tensor<string, []> var_1749_equation_0 = const()[name = tensor<string, []>("op_1749_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1749_cast = einsum(equation = var_1749_equation_0, values = (var_1673_cast, var_1731_cast))[name = tensor<string, []>("op_1749_cast")];
            tensor<string, []> var_1751_equation_0 = const()[name = tensor<string, []>("op_1751_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1751_cast = einsum(equation = var_1751_equation_0, values = (var_1677_cast, var_1732_cast))[name = tensor<string, []>("op_1751_cast")];
            tensor<string, []> var_1753_equation_0 = const()[name = tensor<string, []>("op_1753_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1753_cast = einsum(equation = var_1753_equation_0, values = (var_1681_cast, var_1733_cast))[name = tensor<string, []>("op_1753_cast")];
            tensor<bool, []> input_99_interleave_0 = const()[name = tensor<string, []>("input_99_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_99_cast = concat(axis = var_298, interleave = input_99_interleave_0, values = (var_1735_cast, var_1737_cast, var_1739_cast, var_1741_cast, var_1743_cast, var_1745_cast, var_1747_cast, var_1749_cast, var_1751_cast, var_1753_cast))[name = tensor<string, []>("input_99_cast")];
            tensor<int32, [2]> var_1759 = const()[name = tensor<string, []>("op_1759"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1761 = const()[name = tensor<string, []>("op_1761"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_1763_pad_type_0 = const()[name = tensor<string, []>("op_1763_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_1763_pad_0 = const()[name = tensor<string, []>("op_1763_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99947840)))];
            tensor<fp16, [640]> down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100767104)))];
            tensor<fp16, [2, 640, 1, 4096]> var_1763_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_1761, groups = var_298, pad = var_1763_pad_0, pad_type = var_1763_pad_type_0, strides = var_1759, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_99_cast)[name = tensor<string, []>("op_1763_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_15_cast = add(x = var_1763_cast, y = inputs_13_cast)[name = tensor<string, []>("inputs_15_cast")];
            tensor<int32, [1]> var_1767 = const()[name = tensor<string, []>("op_1767"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_15_cast = reduce_mean(axes = var_1767, keep_dims = var_293, x = inputs_15_cast)[name = tensor<string, []>("channels_mean_15_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_15_cast = sub(x = inputs_15_cast, y = channels_mean_15_cast)[name = tensor<string, []>("zero_mean_15_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_15_cast = mul(x = zero_mean_15_cast, y = zero_mean_15_cast)[name = tensor<string, []>("zero_mean_sq_15_cast")];
            tensor<int32, [1]> var_1771 = const()[name = tensor<string, []>("op_1771"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_1772_cast = reduce_mean(axes = var_1771, keep_dims = var_293, x = zero_mean_sq_15_cast)[name = tensor<string, []>("op_1772_cast")];
            tensor<fp16, []> var_1773_to_fp16 = const()[name = tensor<string, []>("op_1773_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_1774_cast = add(x = var_1772_cast, y = var_1773_to_fp16)[name = tensor<string, []>("op_1774_cast")];
            tensor<fp16, []> denom_15_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_15_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_15_cast = rsqrt(epsilon = denom_15_epsilon_0_to_fp16, x = var_1774_cast)[name = tensor<string, []>("denom_15_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_15_cast = mul(x = zero_mean_15_cast, y = denom_15_cast)[name = tensor<string, []>("out_15_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_1778_to_fp16 = const()[name = tensor<string, []>("op_1778_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100768448)))];
            tensor<fp16, [2, 640, 1, 4096]> var_1779_cast = add(x = out_15_cast, y = var_1778_to_fp16)[name = tensor<string, []>("op_1779_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_1781_to_fp16 = const()[name = tensor<string, []>("op_1781_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100769792)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_43_cast = mul(x = var_1779_cast, y = var_1781_to_fp16)[name = tensor<string, []>("hidden_states_43_cast")];
            tensor<int32, [2]> var_1788 = const()[name = tensor<string, []>("op_1788"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1790 = const()[name = tensor<string, []>("op_1790"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_11_pad_type_0 = const()[name = tensor<string, []>("q_11_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_11_pad_0 = const()[name = tensor<string, []>("q_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100771136)))];
            tensor<fp16, [2, 640, 1, 4096]> q_11_cast = conv(dilations = var_1790, groups = var_298, pad = q_11_pad_0, pad_type = q_11_pad_type_0, strides = var_1788, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_43_cast)[name = tensor<string, []>("q_11_cast")];
            tensor<int32, [2]> var_1794 = const()[name = tensor<string, []>("op_1794"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1796 = const()[name = tensor<string, []>("op_1796"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_21_pad_type_0 = const()[name = tensor<string, []>("k_21_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_21_pad_0 = const()[name = tensor<string, []>("k_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101590400)))];
            tensor<fp16, [2, 640, 1, 77]> k_21_cast = conv(dilations = var_1796, groups = var_298, pad = k_21_pad_0, pad_type = k_21_pad_type_0, strides = var_1794, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_21_cast")];
            tensor<int32, [2]> var_1800 = const()[name = tensor<string, []>("op_1800"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1802 = const()[name = tensor<string, []>("op_1802"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_11_pad_type_0 = const()[name = tensor<string, []>("v_11_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_11_pad_0 = const()[name = tensor<string, []>("v_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104211904)))];
            tensor<fp16, [2, 640, 1, 77]> v_11_cast = conv(dilations = var_1802, groups = var_298, pad = v_11_pad_0, pad_type = v_11_pad_type_0, strides = var_1800, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_11_cast")];
            tensor<int32, [4]> var_1806_begin_0 = const()[name = tensor<string, []>("op_1806_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1806_end_0 = const()[name = tensor<string, []>("op_1806_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_1806_end_mask_0 = const()[name = tensor<string, []>("op_1806_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1806_cast = slice_by_index(begin = var_1806_begin_0, end = var_1806_end_0, end_mask = var_1806_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_1806_cast")];
            tensor<int32, [4]> var_1810_begin_0 = const()[name = tensor<string, []>("op_1810_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1810_end_0 = const()[name = tensor<string, []>("op_1810_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_1810_end_mask_0 = const()[name = tensor<string, []>("op_1810_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1810_cast = slice_by_index(begin = var_1810_begin_0, end = var_1810_end_0, end_mask = var_1810_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_1810_cast")];
            tensor<int32, [4]> var_1814_begin_0 = const()[name = tensor<string, []>("op_1814_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1814_end_0 = const()[name = tensor<string, []>("op_1814_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_1814_end_mask_0 = const()[name = tensor<string, []>("op_1814_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1814_cast = slice_by_index(begin = var_1814_begin_0, end = var_1814_end_0, end_mask = var_1814_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_1814_cast")];
            tensor<int32, [4]> var_1818_begin_0 = const()[name = tensor<string, []>("op_1818_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1818_end_0 = const()[name = tensor<string, []>("op_1818_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_1818_end_mask_0 = const()[name = tensor<string, []>("op_1818_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1818_cast = slice_by_index(begin = var_1818_begin_0, end = var_1818_end_0, end_mask = var_1818_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_1818_cast")];
            tensor<int32, [4]> var_1822_begin_0 = const()[name = tensor<string, []>("op_1822_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1822_end_0 = const()[name = tensor<string, []>("op_1822_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_1822_end_mask_0 = const()[name = tensor<string, []>("op_1822_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1822_cast = slice_by_index(begin = var_1822_begin_0, end = var_1822_end_0, end_mask = var_1822_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_1822_cast")];
            tensor<int32, [4]> var_1826_begin_0 = const()[name = tensor<string, []>("op_1826_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1826_end_0 = const()[name = tensor<string, []>("op_1826_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_1826_end_mask_0 = const()[name = tensor<string, []>("op_1826_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1826_cast = slice_by_index(begin = var_1826_begin_0, end = var_1826_end_0, end_mask = var_1826_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_1826_cast")];
            tensor<int32, [4]> var_1830_begin_0 = const()[name = tensor<string, []>("op_1830_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_1830_end_0 = const()[name = tensor<string, []>("op_1830_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_1830_end_mask_0 = const()[name = tensor<string, []>("op_1830_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1830_cast = slice_by_index(begin = var_1830_begin_0, end = var_1830_end_0, end_mask = var_1830_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_1830_cast")];
            tensor<int32, [4]> var_1834_begin_0 = const()[name = tensor<string, []>("op_1834_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_1834_end_0 = const()[name = tensor<string, []>("op_1834_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_1834_end_mask_0 = const()[name = tensor<string, []>("op_1834_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1834_cast = slice_by_index(begin = var_1834_begin_0, end = var_1834_end_0, end_mask = var_1834_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_1834_cast")];
            tensor<int32, [4]> var_1838_begin_0 = const()[name = tensor<string, []>("op_1838_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_1838_end_0 = const()[name = tensor<string, []>("op_1838_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_1838_end_mask_0 = const()[name = tensor<string, []>("op_1838_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1838_cast = slice_by_index(begin = var_1838_begin_0, end = var_1838_end_0, end_mask = var_1838_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_1838_cast")];
            tensor<int32, [4]> var_1842_begin_0 = const()[name = tensor<string, []>("op_1842_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_1842_end_0 = const()[name = tensor<string, []>("op_1842_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_1842_end_mask_0 = const()[name = tensor<string, []>("op_1842_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1842_cast = slice_by_index(begin = var_1842_begin_0, end = var_1842_end_0, end_mask = var_1842_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_1842_cast")];
            tensor<int32, [4]> k_23_perm_0 = const()[name = tensor<string, []>("k_23_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_1849_begin_0 = const()[name = tensor<string, []>("op_1849_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1849_end_0 = const()[name = tensor<string, []>("op_1849_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_1849_end_mask_0 = const()[name = tensor<string, []>("op_1849_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> transpose_134 = transpose(perm = k_23_perm_0, x = k_21_cast)[name = tensor<string, []>("transpose_134")];
            tensor<fp16, [2, 77, 1, 64]> var_1849_cast = slice_by_index(begin = var_1849_begin_0, end = var_1849_end_0, end_mask = var_1849_end_mask_0, x = transpose_134)[name = tensor<string, []>("op_1849_cast")];
            tensor<int32, [4]> var_1853_begin_0 = const()[name = tensor<string, []>("op_1853_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_1853_end_0 = const()[name = tensor<string, []>("op_1853_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_1853_end_mask_0 = const()[name = tensor<string, []>("op_1853_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1853_cast = slice_by_index(begin = var_1853_begin_0, end = var_1853_end_0, end_mask = var_1853_end_mask_0, x = transpose_134)[name = tensor<string, []>("op_1853_cast")];
            tensor<int32, [4]> var_1857_begin_0 = const()[name = tensor<string, []>("op_1857_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_1857_end_0 = const()[name = tensor<string, []>("op_1857_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_1857_end_mask_0 = const()[name = tensor<string, []>("op_1857_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1857_cast = slice_by_index(begin = var_1857_begin_0, end = var_1857_end_0, end_mask = var_1857_end_mask_0, x = transpose_134)[name = tensor<string, []>("op_1857_cast")];
            tensor<int32, [4]> var_1861_begin_0 = const()[name = tensor<string, []>("op_1861_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_1861_end_0 = const()[name = tensor<string, []>("op_1861_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_1861_end_mask_0 = const()[name = tensor<string, []>("op_1861_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1861_cast = slice_by_index(begin = var_1861_begin_0, end = var_1861_end_0, end_mask = var_1861_end_mask_0, x = transpose_134)[name = tensor<string, []>("op_1861_cast")];
            tensor<int32, [4]> var_1865_begin_0 = const()[name = tensor<string, []>("op_1865_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_1865_end_0 = const()[name = tensor<string, []>("op_1865_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_1865_end_mask_0 = const()[name = tensor<string, []>("op_1865_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1865_cast = slice_by_index(begin = var_1865_begin_0, end = var_1865_end_0, end_mask = var_1865_end_mask_0, x = transpose_134)[name = tensor<string, []>("op_1865_cast")];
            tensor<int32, [4]> var_1869_begin_0 = const()[name = tensor<string, []>("op_1869_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_1869_end_0 = const()[name = tensor<string, []>("op_1869_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_1869_end_mask_0 = const()[name = tensor<string, []>("op_1869_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1869_cast = slice_by_index(begin = var_1869_begin_0, end = var_1869_end_0, end_mask = var_1869_end_mask_0, x = transpose_134)[name = tensor<string, []>("op_1869_cast")];
            tensor<int32, [4]> var_1873_begin_0 = const()[name = tensor<string, []>("op_1873_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_1873_end_0 = const()[name = tensor<string, []>("op_1873_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_1873_end_mask_0 = const()[name = tensor<string, []>("op_1873_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1873_cast = slice_by_index(begin = var_1873_begin_0, end = var_1873_end_0, end_mask = var_1873_end_mask_0, x = transpose_134)[name = tensor<string, []>("op_1873_cast")];
            tensor<int32, [4]> var_1877_begin_0 = const()[name = tensor<string, []>("op_1877_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_1877_end_0 = const()[name = tensor<string, []>("op_1877_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_1877_end_mask_0 = const()[name = tensor<string, []>("op_1877_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1877_cast = slice_by_index(begin = var_1877_begin_0, end = var_1877_end_0, end_mask = var_1877_end_mask_0, x = transpose_134)[name = tensor<string, []>("op_1877_cast")];
            tensor<int32, [4]> var_1881_begin_0 = const()[name = tensor<string, []>("op_1881_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1881_end_0 = const()[name = tensor<string, []>("op_1881_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_1881_end_mask_0 = const()[name = tensor<string, []>("op_1881_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1881_cast = slice_by_index(begin = var_1881_begin_0, end = var_1881_end_0, end_mask = var_1881_end_mask_0, x = transpose_134)[name = tensor<string, []>("op_1881_cast")];
            tensor<int32, [4]> var_1885_begin_0 = const()[name = tensor<string, []>("op_1885_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_1885_end_0 = const()[name = tensor<string, []>("op_1885_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_1885_end_mask_0 = const()[name = tensor<string, []>("op_1885_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1885_cast = slice_by_index(begin = var_1885_begin_0, end = var_1885_end_0, end_mask = var_1885_end_mask_0, x = transpose_134)[name = tensor<string, []>("op_1885_cast")];
            tensor<int32, [4]> var_1887_begin_0 = const()[name = tensor<string, []>("op_1887_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1887_end_0 = const()[name = tensor<string, []>("op_1887_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_1887_end_mask_0 = const()[name = tensor<string, []>("op_1887_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1887_cast = slice_by_index(begin = var_1887_begin_0, end = var_1887_end_0, end_mask = var_1887_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_1887_cast")];
            tensor<int32, [4]> var_1891_begin_0 = const()[name = tensor<string, []>("op_1891_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1891_end_0 = const()[name = tensor<string, []>("op_1891_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_1891_end_mask_0 = const()[name = tensor<string, []>("op_1891_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1891_cast = slice_by_index(begin = var_1891_begin_0, end = var_1891_end_0, end_mask = var_1891_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_1891_cast")];
            tensor<int32, [4]> var_1895_begin_0 = const()[name = tensor<string, []>("op_1895_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1895_end_0 = const()[name = tensor<string, []>("op_1895_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_1895_end_mask_0 = const()[name = tensor<string, []>("op_1895_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1895_cast = slice_by_index(begin = var_1895_begin_0, end = var_1895_end_0, end_mask = var_1895_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_1895_cast")];
            tensor<int32, [4]> var_1899_begin_0 = const()[name = tensor<string, []>("op_1899_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1899_end_0 = const()[name = tensor<string, []>("op_1899_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_1899_end_mask_0 = const()[name = tensor<string, []>("op_1899_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1899_cast = slice_by_index(begin = var_1899_begin_0, end = var_1899_end_0, end_mask = var_1899_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_1899_cast")];
            tensor<int32, [4]> var_1903_begin_0 = const()[name = tensor<string, []>("op_1903_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1903_end_0 = const()[name = tensor<string, []>("op_1903_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_1903_end_mask_0 = const()[name = tensor<string, []>("op_1903_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1903_cast = slice_by_index(begin = var_1903_begin_0, end = var_1903_end_0, end_mask = var_1903_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_1903_cast")];
            tensor<int32, [4]> var_1907_begin_0 = const()[name = tensor<string, []>("op_1907_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1907_end_0 = const()[name = tensor<string, []>("op_1907_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_1907_end_mask_0 = const()[name = tensor<string, []>("op_1907_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1907_cast = slice_by_index(begin = var_1907_begin_0, end = var_1907_end_0, end_mask = var_1907_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_1907_cast")];
            tensor<int32, [4]> var_1911_begin_0 = const()[name = tensor<string, []>("op_1911_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_1911_end_0 = const()[name = tensor<string, []>("op_1911_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_1911_end_mask_0 = const()[name = tensor<string, []>("op_1911_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1911_cast = slice_by_index(begin = var_1911_begin_0, end = var_1911_end_0, end_mask = var_1911_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_1911_cast")];
            tensor<int32, [4]> var_1915_begin_0 = const()[name = tensor<string, []>("op_1915_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_1915_end_0 = const()[name = tensor<string, []>("op_1915_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_1915_end_mask_0 = const()[name = tensor<string, []>("op_1915_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1915_cast = slice_by_index(begin = var_1915_begin_0, end = var_1915_end_0, end_mask = var_1915_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_1915_cast")];
            tensor<int32, [4]> var_1919_begin_0 = const()[name = tensor<string, []>("op_1919_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_1919_end_0 = const()[name = tensor<string, []>("op_1919_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_1919_end_mask_0 = const()[name = tensor<string, []>("op_1919_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1919_cast = slice_by_index(begin = var_1919_begin_0, end = var_1919_end_0, end_mask = var_1919_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_1919_cast")];
            tensor<int32, [4]> var_1923_begin_0 = const()[name = tensor<string, []>("op_1923_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_1923_end_0 = const()[name = tensor<string, []>("op_1923_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_1923_end_mask_0 = const()[name = tensor<string, []>("op_1923_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1923_cast = slice_by_index(begin = var_1923_begin_0, end = var_1923_end_0, end_mask = var_1923_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_1923_cast")];
            tensor<string, []> var_1927_equation_0 = const()[name = tensor<string, []>("op_1927_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1927_cast = einsum(equation = var_1927_equation_0, values = (var_1849_cast, var_1806_cast))[name = tensor<string, []>("op_1927_cast")];
            tensor<fp16, []> var_1928_to_fp16 = const()[name = tensor<string, []>("op_1928_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_101_cast = mul(x = var_1927_cast, y = var_1928_to_fp16)[name = tensor<string, []>("aw_101_cast")];
            tensor<string, []> var_1931_equation_0 = const()[name = tensor<string, []>("op_1931_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1931_cast = einsum(equation = var_1931_equation_0, values = (var_1853_cast, var_1810_cast))[name = tensor<string, []>("op_1931_cast")];
            tensor<fp16, []> var_1932_to_fp16 = const()[name = tensor<string, []>("op_1932_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_103_cast = mul(x = var_1931_cast, y = var_1932_to_fp16)[name = tensor<string, []>("aw_103_cast")];
            tensor<string, []> var_1935_equation_0 = const()[name = tensor<string, []>("op_1935_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1935_cast = einsum(equation = var_1935_equation_0, values = (var_1857_cast, var_1814_cast))[name = tensor<string, []>("op_1935_cast")];
            tensor<fp16, []> var_1936_to_fp16 = const()[name = tensor<string, []>("op_1936_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_105_cast = mul(x = var_1935_cast, y = var_1936_to_fp16)[name = tensor<string, []>("aw_105_cast")];
            tensor<string, []> var_1939_equation_0 = const()[name = tensor<string, []>("op_1939_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1939_cast = einsum(equation = var_1939_equation_0, values = (var_1861_cast, var_1818_cast))[name = tensor<string, []>("op_1939_cast")];
            tensor<fp16, []> var_1940_to_fp16 = const()[name = tensor<string, []>("op_1940_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_107_cast = mul(x = var_1939_cast, y = var_1940_to_fp16)[name = tensor<string, []>("aw_107_cast")];
            tensor<string, []> var_1943_equation_0 = const()[name = tensor<string, []>("op_1943_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1943_cast = einsum(equation = var_1943_equation_0, values = (var_1865_cast, var_1822_cast))[name = tensor<string, []>("op_1943_cast")];
            tensor<fp16, []> var_1944_to_fp16 = const()[name = tensor<string, []>("op_1944_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_109_cast = mul(x = var_1943_cast, y = var_1944_to_fp16)[name = tensor<string, []>("aw_109_cast")];
            tensor<string, []> var_1947_equation_0 = const()[name = tensor<string, []>("op_1947_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1947_cast = einsum(equation = var_1947_equation_0, values = (var_1869_cast, var_1826_cast))[name = tensor<string, []>("op_1947_cast")];
            tensor<fp16, []> var_1948_to_fp16 = const()[name = tensor<string, []>("op_1948_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_111_cast = mul(x = var_1947_cast, y = var_1948_to_fp16)[name = tensor<string, []>("aw_111_cast")];
            tensor<string, []> var_1951_equation_0 = const()[name = tensor<string, []>("op_1951_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1951_cast = einsum(equation = var_1951_equation_0, values = (var_1873_cast, var_1830_cast))[name = tensor<string, []>("op_1951_cast")];
            tensor<fp16, []> var_1952_to_fp16 = const()[name = tensor<string, []>("op_1952_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_113_cast = mul(x = var_1951_cast, y = var_1952_to_fp16)[name = tensor<string, []>("aw_113_cast")];
            tensor<string, []> var_1955_equation_0 = const()[name = tensor<string, []>("op_1955_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1955_cast = einsum(equation = var_1955_equation_0, values = (var_1877_cast, var_1834_cast))[name = tensor<string, []>("op_1955_cast")];
            tensor<fp16, []> var_1956_to_fp16 = const()[name = tensor<string, []>("op_1956_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_115_cast = mul(x = var_1955_cast, y = var_1956_to_fp16)[name = tensor<string, []>("aw_115_cast")];
            tensor<string, []> var_1959_equation_0 = const()[name = tensor<string, []>("op_1959_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1959_cast = einsum(equation = var_1959_equation_0, values = (var_1881_cast, var_1838_cast))[name = tensor<string, []>("op_1959_cast")];
            tensor<fp16, []> var_1960_to_fp16 = const()[name = tensor<string, []>("op_1960_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_117_cast = mul(x = var_1959_cast, y = var_1960_to_fp16)[name = tensor<string, []>("aw_117_cast")];
            tensor<string, []> var_1963_equation_0 = const()[name = tensor<string, []>("op_1963_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_1963_cast = einsum(equation = var_1963_equation_0, values = (var_1885_cast, var_1842_cast))[name = tensor<string, []>("op_1963_cast")];
            tensor<fp16, []> var_1964_to_fp16 = const()[name = tensor<string, []>("op_1964_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_119_cast = mul(x = var_1963_cast, y = var_1964_to_fp16)[name = tensor<string, []>("aw_119_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1966_cast = softmax(axis = var_298, x = aw_101_cast)[name = tensor<string, []>("op_1966_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1967_cast = softmax(axis = var_298, x = aw_103_cast)[name = tensor<string, []>("op_1967_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1968_cast = softmax(axis = var_298, x = aw_105_cast)[name = tensor<string, []>("op_1968_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1969_cast = softmax(axis = var_298, x = aw_107_cast)[name = tensor<string, []>("op_1969_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1970_cast = softmax(axis = var_298, x = aw_109_cast)[name = tensor<string, []>("op_1970_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1971_cast = softmax(axis = var_298, x = aw_111_cast)[name = tensor<string, []>("op_1971_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1972_cast = softmax(axis = var_298, x = aw_113_cast)[name = tensor<string, []>("op_1972_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1973_cast = softmax(axis = var_298, x = aw_115_cast)[name = tensor<string, []>("op_1973_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1974_cast = softmax(axis = var_298, x = aw_117_cast)[name = tensor<string, []>("op_1974_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_1975_cast = softmax(axis = var_298, x = aw_119_cast)[name = tensor<string, []>("op_1975_cast")];
            tensor<string, []> var_1977_equation_0 = const()[name = tensor<string, []>("op_1977_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1977_cast = einsum(equation = var_1977_equation_0, values = (var_1887_cast, var_1966_cast))[name = tensor<string, []>("op_1977_cast")];
            tensor<string, []> var_1979_equation_0 = const()[name = tensor<string, []>("op_1979_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1979_cast = einsum(equation = var_1979_equation_0, values = (var_1891_cast, var_1967_cast))[name = tensor<string, []>("op_1979_cast")];
            tensor<string, []> var_1981_equation_0 = const()[name = tensor<string, []>("op_1981_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1981_cast = einsum(equation = var_1981_equation_0, values = (var_1895_cast, var_1968_cast))[name = tensor<string, []>("op_1981_cast")];
            tensor<string, []> var_1983_equation_0 = const()[name = tensor<string, []>("op_1983_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1983_cast = einsum(equation = var_1983_equation_0, values = (var_1899_cast, var_1969_cast))[name = tensor<string, []>("op_1983_cast")];
            tensor<string, []> var_1985_equation_0 = const()[name = tensor<string, []>("op_1985_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1985_cast = einsum(equation = var_1985_equation_0, values = (var_1903_cast, var_1970_cast))[name = tensor<string, []>("op_1985_cast")];
            tensor<string, []> var_1987_equation_0 = const()[name = tensor<string, []>("op_1987_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1987_cast = einsum(equation = var_1987_equation_0, values = (var_1907_cast, var_1971_cast))[name = tensor<string, []>("op_1987_cast")];
            tensor<string, []> var_1989_equation_0 = const()[name = tensor<string, []>("op_1989_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1989_cast = einsum(equation = var_1989_equation_0, values = (var_1911_cast, var_1972_cast))[name = tensor<string, []>("op_1989_cast")];
            tensor<string, []> var_1991_equation_0 = const()[name = tensor<string, []>("op_1991_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1991_cast = einsum(equation = var_1991_equation_0, values = (var_1915_cast, var_1973_cast))[name = tensor<string, []>("op_1991_cast")];
            tensor<string, []> var_1993_equation_0 = const()[name = tensor<string, []>("op_1993_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1993_cast = einsum(equation = var_1993_equation_0, values = (var_1919_cast, var_1974_cast))[name = tensor<string, []>("op_1993_cast")];
            tensor<string, []> var_1995_equation_0 = const()[name = tensor<string, []>("op_1995_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_1995_cast = einsum(equation = var_1995_equation_0, values = (var_1923_cast, var_1975_cast))[name = tensor<string, []>("op_1995_cast")];
            tensor<bool, []> input_101_interleave_0 = const()[name = tensor<string, []>("input_101_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_101_cast = concat(axis = var_298, interleave = input_101_interleave_0, values = (var_1977_cast, var_1979_cast, var_1981_cast, var_1983_cast, var_1985_cast, var_1987_cast, var_1989_cast, var_1991_cast, var_1993_cast, var_1995_cast))[name = tensor<string, []>("input_101_cast")];
            tensor<int32, [2]> var_2001 = const()[name = tensor<string, []>("op_2001"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2003 = const()[name = tensor<string, []>("op_2003"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_2005_pad_type_0 = const()[name = tensor<string, []>("op_2005_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_2005_pad_0 = const()[name = tensor<string, []>("op_2005_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106833408)))];
            tensor<fp16, [640]> down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107652672)))];
            tensor<fp16, [2, 640, 1, 4096]> var_2005_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_2003, groups = var_298, pad = var_2005_pad_0, pad_type = var_2005_pad_type_0, strides = var_2001, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_101_cast)[name = tensor<string, []>("op_2005_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_17_cast = add(x = var_2005_cast, y = inputs_15_cast)[name = tensor<string, []>("inputs_17_cast")];
            tensor<int32, [1]> var_2009 = const()[name = tensor<string, []>("op_2009"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_17_cast = reduce_mean(axes = var_2009, keep_dims = var_293, x = inputs_17_cast)[name = tensor<string, []>("channels_mean_17_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_17_cast = sub(x = inputs_17_cast, y = channels_mean_17_cast)[name = tensor<string, []>("zero_mean_17_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_17_cast = mul(x = zero_mean_17_cast, y = zero_mean_17_cast)[name = tensor<string, []>("zero_mean_sq_17_cast")];
            tensor<int32, [1]> var_2013 = const()[name = tensor<string, []>("op_2013"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_2014_cast = reduce_mean(axes = var_2013, keep_dims = var_293, x = zero_mean_sq_17_cast)[name = tensor<string, []>("op_2014_cast")];
            tensor<fp16, []> var_2015_to_fp16 = const()[name = tensor<string, []>("op_2015_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_2016_cast = add(x = var_2014_cast, y = var_2015_to_fp16)[name = tensor<string, []>("op_2016_cast")];
            tensor<fp16, []> denom_17_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_17_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_17_cast = rsqrt(epsilon = denom_17_epsilon_0_to_fp16, x = var_2016_cast)[name = tensor<string, []>("denom_17_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_17_cast = mul(x = zero_mean_17_cast, y = denom_17_cast)[name = tensor<string, []>("out_17_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_2020_to_fp16 = const()[name = tensor<string, []>("op_2020_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107654016)))];
            tensor<fp16, [2, 640, 1, 4096]> var_2021_cast = add(x = out_17_cast, y = var_2020_to_fp16)[name = tensor<string, []>("op_2021_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_2023_to_fp16 = const()[name = tensor<string, []>("op_2023_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107655360)))];
            tensor<fp16, [2, 640, 1, 4096]> input_103_cast = mul(x = var_2021_cast, y = var_2023_to_fp16)[name = tensor<string, []>("input_103_cast")];
            tensor<int32, [2]> var_2031 = const()[name = tensor<string, []>("op_2031"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2033 = const()[name = tensor<string, []>("op_2033"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_2035_pad_type_0 = const()[name = tensor<string, []>("op_2035_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_2035_pad_0 = const()[name = tensor<string, []>("op_2035_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [5120, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [5120, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107656704)))];
            tensor<fp16, [5120]> down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114210368)))];
            tensor<fp16, [2, 5120, 1, 4096]> var_2035_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_2033, groups = var_298, pad = var_2035_pad_0, pad_type = var_2035_pad_type_0, strides = var_2031, weight = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_103_cast)[name = tensor<string, []>("op_2035_cast")];
            tensor<int32, [2]> var_2036_split_sizes_0 = const()[name = tensor<string, []>("op_2036_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_2036_axis_0 = const()[name = tensor<string, []>("op_2036_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 4096]> var_2036_cast_0, tensor<fp16, [2, 2560, 1, 4096]> var_2036_cast_1 = split(axis = var_2036_axis_0, split_sizes = var_2036_split_sizes_0, x = var_2035_cast)[name = tensor<string, []>("op_2036_cast")];
            tensor<string, []> var_2038_mode_0 = const()[name = tensor<string, []>("op_2038_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 4096]> var_2038_cast = gelu(mode = var_2038_mode_0, x = var_2036_cast_1)[name = tensor<string, []>("op_2038_cast")];
            tensor<fp16, [2, 2560, 1, 4096]> input_105_cast = mul(x = var_2036_cast_0, y = var_2038_cast)[name = tensor<string, []>("input_105_cast")];
            tensor<int32, [2]> var_2042 = const()[name = tensor<string, []>("op_2042"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2044 = const()[name = tensor<string, []>("op_2044"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_2046_pad_type_0 = const()[name = tensor<string, []>("op_2046_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_2046_pad_0 = const()[name = tensor<string, []>("op_2046_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2560, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [640, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114220672)))];
            tensor<fp16, [640]> down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117497536)))];
            tensor<fp16, [2, 640, 1, 4096]> var_2046_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_2044, groups = var_298, pad = var_2046_pad_0, pad_type = var_2046_pad_type_0, strides = var_2042, weight = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_105_cast)[name = tensor<string, []>("op_2046_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_19_cast = add(x = var_2046_cast, y = inputs_17_cast)[name = tensor<string, []>("inputs_19_cast")];
            tensor<int32, [1]> var_2056 = const()[name = tensor<string, []>("op_2056"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_19_cast = reduce_mean(axes = var_2056, keep_dims = var_293, x = inputs_19_cast)[name = tensor<string, []>("channels_mean_19_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_19_cast = sub(x = inputs_19_cast, y = channels_mean_19_cast)[name = tensor<string, []>("zero_mean_19_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_19_cast = mul(x = zero_mean_19_cast, y = zero_mean_19_cast)[name = tensor<string, []>("zero_mean_sq_19_cast")];
            tensor<int32, [1]> var_2060 = const()[name = tensor<string, []>("op_2060"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_2061_cast = reduce_mean(axes = var_2060, keep_dims = var_293, x = zero_mean_sq_19_cast)[name = tensor<string, []>("op_2061_cast")];
            tensor<fp16, []> var_2062_to_fp16 = const()[name = tensor<string, []>("op_2062_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_2063_cast = add(x = var_2061_cast, y = var_2062_to_fp16)[name = tensor<string, []>("op_2063_cast")];
            tensor<fp16, []> denom_19_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_19_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_19_cast = rsqrt(epsilon = denom_19_epsilon_0_to_fp16, x = var_2063_cast)[name = tensor<string, []>("denom_19_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_19_cast = mul(x = zero_mean_19_cast, y = denom_19_cast)[name = tensor<string, []>("out_19_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_2067_to_fp16 = const()[name = tensor<string, []>("op_2067_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117498880)))];
            tensor<fp16, [2, 640, 1, 4096]> var_2068_cast = add(x = out_19_cast, y = var_2067_to_fp16)[name = tensor<string, []>("op_2068_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_2070_to_fp16 = const()[name = tensor<string, []>("op_2070_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117500224)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_47_cast = mul(x = var_2068_cast, y = var_2070_to_fp16)[name = tensor<string, []>("hidden_states_47_cast")];
            tensor<int32, [2]> var_2077 = const()[name = tensor<string, []>("op_2077"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2079 = const()[name = tensor<string, []>("op_2079"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_13_pad_type_0 = const()[name = tensor<string, []>("q_13_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_13_pad_0 = const()[name = tensor<string, []>("q_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117501568)))];
            tensor<fp16, [2, 640, 1, 4096]> q_13_cast = conv(dilations = var_2079, groups = var_298, pad = q_13_pad_0, pad_type = q_13_pad_type_0, strides = var_2077, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_47_cast)[name = tensor<string, []>("q_13_cast")];
            tensor<int32, [2]> var_2083 = const()[name = tensor<string, []>("op_2083"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2085 = const()[name = tensor<string, []>("op_2085"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_25_pad_type_0 = const()[name = tensor<string, []>("k_25_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_25_pad_0 = const()[name = tensor<string, []>("k_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118320832)))];
            tensor<fp16, [2, 640, 1, 4096]> k_25_cast = conv(dilations = var_2085, groups = var_298, pad = k_25_pad_0, pad_type = k_25_pad_type_0, strides = var_2083, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_47_cast)[name = tensor<string, []>("k_25_cast")];
            tensor<int32, [2]> var_2089 = const()[name = tensor<string, []>("op_2089"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2091 = const()[name = tensor<string, []>("op_2091"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_13_pad_type_0 = const()[name = tensor<string, []>("v_13_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_13_pad_0 = const()[name = tensor<string, []>("v_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119140096)))];
            tensor<fp16, [2, 640, 1, 4096]> v_13_cast = conv(dilations = var_2091, groups = var_298, pad = v_13_pad_0, pad_type = v_13_pad_type_0, strides = var_2089, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_47_cast)[name = tensor<string, []>("v_13_cast")];
            tensor<int32, [4]> var_2095_begin_0 = const()[name = tensor<string, []>("op_2095_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2095_end_0 = const()[name = tensor<string, []>("op_2095_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_2095_end_mask_0 = const()[name = tensor<string, []>("op_2095_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2095_cast = slice_by_index(begin = var_2095_begin_0, end = var_2095_end_0, end_mask = var_2095_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_2095_cast")];
            tensor<int32, [4]> var_2099_begin_0 = const()[name = tensor<string, []>("op_2099_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_2099_end_0 = const()[name = tensor<string, []>("op_2099_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_2099_end_mask_0 = const()[name = tensor<string, []>("op_2099_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2099_cast = slice_by_index(begin = var_2099_begin_0, end = var_2099_end_0, end_mask = var_2099_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_2099_cast")];
            tensor<int32, [4]> var_2103_begin_0 = const()[name = tensor<string, []>("op_2103_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_2103_end_0 = const()[name = tensor<string, []>("op_2103_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_2103_end_mask_0 = const()[name = tensor<string, []>("op_2103_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2103_cast = slice_by_index(begin = var_2103_begin_0, end = var_2103_end_0, end_mask = var_2103_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_2103_cast")];
            tensor<int32, [4]> var_2107_begin_0 = const()[name = tensor<string, []>("op_2107_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_2107_end_0 = const()[name = tensor<string, []>("op_2107_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_2107_end_mask_0 = const()[name = tensor<string, []>("op_2107_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2107_cast = slice_by_index(begin = var_2107_begin_0, end = var_2107_end_0, end_mask = var_2107_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_2107_cast")];
            tensor<int32, [4]> var_2111_begin_0 = const()[name = tensor<string, []>("op_2111_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_2111_end_0 = const()[name = tensor<string, []>("op_2111_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_2111_end_mask_0 = const()[name = tensor<string, []>("op_2111_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2111_cast = slice_by_index(begin = var_2111_begin_0, end = var_2111_end_0, end_mask = var_2111_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_2111_cast")];
            tensor<int32, [4]> var_2115_begin_0 = const()[name = tensor<string, []>("op_2115_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_2115_end_0 = const()[name = tensor<string, []>("op_2115_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_2115_end_mask_0 = const()[name = tensor<string, []>("op_2115_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2115_cast = slice_by_index(begin = var_2115_begin_0, end = var_2115_end_0, end_mask = var_2115_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_2115_cast")];
            tensor<int32, [4]> var_2119_begin_0 = const()[name = tensor<string, []>("op_2119_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_2119_end_0 = const()[name = tensor<string, []>("op_2119_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_2119_end_mask_0 = const()[name = tensor<string, []>("op_2119_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2119_cast = slice_by_index(begin = var_2119_begin_0, end = var_2119_end_0, end_mask = var_2119_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_2119_cast")];
            tensor<int32, [4]> var_2123_begin_0 = const()[name = tensor<string, []>("op_2123_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_2123_end_0 = const()[name = tensor<string, []>("op_2123_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_2123_end_mask_0 = const()[name = tensor<string, []>("op_2123_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2123_cast = slice_by_index(begin = var_2123_begin_0, end = var_2123_end_0, end_mask = var_2123_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_2123_cast")];
            tensor<int32, [4]> var_2127_begin_0 = const()[name = tensor<string, []>("op_2127_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_2127_end_0 = const()[name = tensor<string, []>("op_2127_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_2127_end_mask_0 = const()[name = tensor<string, []>("op_2127_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2127_cast = slice_by_index(begin = var_2127_begin_0, end = var_2127_end_0, end_mask = var_2127_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_2127_cast")];
            tensor<int32, [4]> var_2131_begin_0 = const()[name = tensor<string, []>("op_2131_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_2131_end_0 = const()[name = tensor<string, []>("op_2131_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_2131_end_mask_0 = const()[name = tensor<string, []>("op_2131_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2131_cast = slice_by_index(begin = var_2131_begin_0, end = var_2131_end_0, end_mask = var_2131_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_2131_cast")];
            tensor<int32, [4]> k_27_perm_0 = const()[name = tensor<string, []>("k_27_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_2138_begin_0 = const()[name = tensor<string, []>("op_2138_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2138_end_0 = const()[name = tensor<string, []>("op_2138_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_2138_end_mask_0 = const()[name = tensor<string, []>("op_2138_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 640]> transpose_133 = transpose(perm = k_27_perm_0, x = k_25_cast)[name = tensor<string, []>("transpose_133")];
            tensor<fp16, [2, 4096, 1, 64]> var_2138_cast = slice_by_index(begin = var_2138_begin_0, end = var_2138_end_0, end_mask = var_2138_end_mask_0, x = transpose_133)[name = tensor<string, []>("op_2138_cast")];
            tensor<int32, [4]> var_2142_begin_0 = const()[name = tensor<string, []>("op_2142_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_2142_end_0 = const()[name = tensor<string, []>("op_2142_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_2142_end_mask_0 = const()[name = tensor<string, []>("op_2142_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_2142_cast = slice_by_index(begin = var_2142_begin_0, end = var_2142_end_0, end_mask = var_2142_end_mask_0, x = transpose_133)[name = tensor<string, []>("op_2142_cast")];
            tensor<int32, [4]> var_2146_begin_0 = const()[name = tensor<string, []>("op_2146_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_2146_end_0 = const()[name = tensor<string, []>("op_2146_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_2146_end_mask_0 = const()[name = tensor<string, []>("op_2146_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_2146_cast = slice_by_index(begin = var_2146_begin_0, end = var_2146_end_0, end_mask = var_2146_end_mask_0, x = transpose_133)[name = tensor<string, []>("op_2146_cast")];
            tensor<int32, [4]> var_2150_begin_0 = const()[name = tensor<string, []>("op_2150_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_2150_end_0 = const()[name = tensor<string, []>("op_2150_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_2150_end_mask_0 = const()[name = tensor<string, []>("op_2150_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_2150_cast = slice_by_index(begin = var_2150_begin_0, end = var_2150_end_0, end_mask = var_2150_end_mask_0, x = transpose_133)[name = tensor<string, []>("op_2150_cast")];
            tensor<int32, [4]> var_2154_begin_0 = const()[name = tensor<string, []>("op_2154_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_2154_end_0 = const()[name = tensor<string, []>("op_2154_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_2154_end_mask_0 = const()[name = tensor<string, []>("op_2154_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_2154_cast = slice_by_index(begin = var_2154_begin_0, end = var_2154_end_0, end_mask = var_2154_end_mask_0, x = transpose_133)[name = tensor<string, []>("op_2154_cast")];
            tensor<int32, [4]> var_2158_begin_0 = const()[name = tensor<string, []>("op_2158_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_2158_end_0 = const()[name = tensor<string, []>("op_2158_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 384])];
            tensor<bool, [4]> var_2158_end_mask_0 = const()[name = tensor<string, []>("op_2158_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_2158_cast = slice_by_index(begin = var_2158_begin_0, end = var_2158_end_0, end_mask = var_2158_end_mask_0, x = transpose_133)[name = tensor<string, []>("op_2158_cast")];
            tensor<int32, [4]> var_2162_begin_0 = const()[name = tensor<string, []>("op_2162_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_2162_end_0 = const()[name = tensor<string, []>("op_2162_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 448])];
            tensor<bool, [4]> var_2162_end_mask_0 = const()[name = tensor<string, []>("op_2162_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_2162_cast = slice_by_index(begin = var_2162_begin_0, end = var_2162_end_0, end_mask = var_2162_end_mask_0, x = transpose_133)[name = tensor<string, []>("op_2162_cast")];
            tensor<int32, [4]> var_2166_begin_0 = const()[name = tensor<string, []>("op_2166_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_2166_end_0 = const()[name = tensor<string, []>("op_2166_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 512])];
            tensor<bool, [4]> var_2166_end_mask_0 = const()[name = tensor<string, []>("op_2166_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_2166_cast = slice_by_index(begin = var_2166_begin_0, end = var_2166_end_0, end_mask = var_2166_end_mask_0, x = transpose_133)[name = tensor<string, []>("op_2166_cast")];
            tensor<int32, [4]> var_2170_begin_0 = const()[name = tensor<string, []>("op_2170_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2170_end_0 = const()[name = tensor<string, []>("op_2170_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 576])];
            tensor<bool, [4]> var_2170_end_mask_0 = const()[name = tensor<string, []>("op_2170_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_2170_cast = slice_by_index(begin = var_2170_begin_0, end = var_2170_end_0, end_mask = var_2170_end_mask_0, x = transpose_133)[name = tensor<string, []>("op_2170_cast")];
            tensor<int32, [4]> var_2174_begin_0 = const()[name = tensor<string, []>("op_2174_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_2174_end_0 = const()[name = tensor<string, []>("op_2174_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 640])];
            tensor<bool, [4]> var_2174_end_mask_0 = const()[name = tensor<string, []>("op_2174_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_2174_cast = slice_by_index(begin = var_2174_begin_0, end = var_2174_end_0, end_mask = var_2174_end_mask_0, x = transpose_133)[name = tensor<string, []>("op_2174_cast")];
            tensor<int32, [4]> var_2176_begin_0 = const()[name = tensor<string, []>("op_2176_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2176_end_0 = const()[name = tensor<string, []>("op_2176_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_2176_end_mask_0 = const()[name = tensor<string, []>("op_2176_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2176_cast = slice_by_index(begin = var_2176_begin_0, end = var_2176_end_0, end_mask = var_2176_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_2176_cast")];
            tensor<int32, [4]> var_2180_begin_0 = const()[name = tensor<string, []>("op_2180_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_2180_end_0 = const()[name = tensor<string, []>("op_2180_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_2180_end_mask_0 = const()[name = tensor<string, []>("op_2180_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2180_cast = slice_by_index(begin = var_2180_begin_0, end = var_2180_end_0, end_mask = var_2180_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_2180_cast")];
            tensor<int32, [4]> var_2184_begin_0 = const()[name = tensor<string, []>("op_2184_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_2184_end_0 = const()[name = tensor<string, []>("op_2184_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_2184_end_mask_0 = const()[name = tensor<string, []>("op_2184_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2184_cast = slice_by_index(begin = var_2184_begin_0, end = var_2184_end_0, end_mask = var_2184_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_2184_cast")];
            tensor<int32, [4]> var_2188_begin_0 = const()[name = tensor<string, []>("op_2188_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_2188_end_0 = const()[name = tensor<string, []>("op_2188_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_2188_end_mask_0 = const()[name = tensor<string, []>("op_2188_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2188_cast = slice_by_index(begin = var_2188_begin_0, end = var_2188_end_0, end_mask = var_2188_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_2188_cast")];
            tensor<int32, [4]> var_2192_begin_0 = const()[name = tensor<string, []>("op_2192_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_2192_end_0 = const()[name = tensor<string, []>("op_2192_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_2192_end_mask_0 = const()[name = tensor<string, []>("op_2192_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2192_cast = slice_by_index(begin = var_2192_begin_0, end = var_2192_end_0, end_mask = var_2192_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_2192_cast")];
            tensor<int32, [4]> var_2196_begin_0 = const()[name = tensor<string, []>("op_2196_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_2196_end_0 = const()[name = tensor<string, []>("op_2196_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_2196_end_mask_0 = const()[name = tensor<string, []>("op_2196_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2196_cast = slice_by_index(begin = var_2196_begin_0, end = var_2196_end_0, end_mask = var_2196_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_2196_cast")];
            tensor<int32, [4]> var_2200_begin_0 = const()[name = tensor<string, []>("op_2200_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_2200_end_0 = const()[name = tensor<string, []>("op_2200_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_2200_end_mask_0 = const()[name = tensor<string, []>("op_2200_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2200_cast = slice_by_index(begin = var_2200_begin_0, end = var_2200_end_0, end_mask = var_2200_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_2200_cast")];
            tensor<int32, [4]> var_2204_begin_0 = const()[name = tensor<string, []>("op_2204_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_2204_end_0 = const()[name = tensor<string, []>("op_2204_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_2204_end_mask_0 = const()[name = tensor<string, []>("op_2204_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2204_cast = slice_by_index(begin = var_2204_begin_0, end = var_2204_end_0, end_mask = var_2204_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_2204_cast")];
            tensor<int32, [4]> var_2208_begin_0 = const()[name = tensor<string, []>("op_2208_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_2208_end_0 = const()[name = tensor<string, []>("op_2208_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_2208_end_mask_0 = const()[name = tensor<string, []>("op_2208_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2208_cast = slice_by_index(begin = var_2208_begin_0, end = var_2208_end_0, end_mask = var_2208_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_2208_cast")];
            tensor<int32, [4]> var_2212_begin_0 = const()[name = tensor<string, []>("op_2212_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_2212_end_0 = const()[name = tensor<string, []>("op_2212_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_2212_end_mask_0 = const()[name = tensor<string, []>("op_2212_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2212_cast = slice_by_index(begin = var_2212_begin_0, end = var_2212_end_0, end_mask = var_2212_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_2212_cast")];
            tensor<string, []> var_2216_equation_0 = const()[name = tensor<string, []>("op_2216_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2216_cast = einsum(equation = var_2216_equation_0, values = (var_2138_cast, var_2095_cast))[name = tensor<string, []>("op_2216_cast")];
            tensor<fp16, []> var_2217_to_fp16 = const()[name = tensor<string, []>("op_2217_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_121_cast = mul(x = var_2216_cast, y = var_2217_to_fp16)[name = tensor<string, []>("aw_121_cast")];
            tensor<string, []> var_2220_equation_0 = const()[name = tensor<string, []>("op_2220_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2220_cast = einsum(equation = var_2220_equation_0, values = (var_2142_cast, var_2099_cast))[name = tensor<string, []>("op_2220_cast")];
            tensor<fp16, []> var_2221_to_fp16 = const()[name = tensor<string, []>("op_2221_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_123_cast = mul(x = var_2220_cast, y = var_2221_to_fp16)[name = tensor<string, []>("aw_123_cast")];
            tensor<string, []> var_2224_equation_0 = const()[name = tensor<string, []>("op_2224_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2224_cast = einsum(equation = var_2224_equation_0, values = (var_2146_cast, var_2103_cast))[name = tensor<string, []>("op_2224_cast")];
            tensor<fp16, []> var_2225_to_fp16 = const()[name = tensor<string, []>("op_2225_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_125_cast = mul(x = var_2224_cast, y = var_2225_to_fp16)[name = tensor<string, []>("aw_125_cast")];
            tensor<string, []> var_2228_equation_0 = const()[name = tensor<string, []>("op_2228_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2228_cast = einsum(equation = var_2228_equation_0, values = (var_2150_cast, var_2107_cast))[name = tensor<string, []>("op_2228_cast")];
            tensor<fp16, []> var_2229_to_fp16 = const()[name = tensor<string, []>("op_2229_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_127_cast = mul(x = var_2228_cast, y = var_2229_to_fp16)[name = tensor<string, []>("aw_127_cast")];
            tensor<string, []> var_2232_equation_0 = const()[name = tensor<string, []>("op_2232_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2232_cast = einsum(equation = var_2232_equation_0, values = (var_2154_cast, var_2111_cast))[name = tensor<string, []>("op_2232_cast")];
            tensor<fp16, []> var_2233_to_fp16 = const()[name = tensor<string, []>("op_2233_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_129_cast = mul(x = var_2232_cast, y = var_2233_to_fp16)[name = tensor<string, []>("aw_129_cast")];
            tensor<string, []> var_2236_equation_0 = const()[name = tensor<string, []>("op_2236_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2236_cast = einsum(equation = var_2236_equation_0, values = (var_2158_cast, var_2115_cast))[name = tensor<string, []>("op_2236_cast")];
            tensor<fp16, []> var_2237_to_fp16 = const()[name = tensor<string, []>("op_2237_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_131_cast = mul(x = var_2236_cast, y = var_2237_to_fp16)[name = tensor<string, []>("aw_131_cast")];
            tensor<string, []> var_2240_equation_0 = const()[name = tensor<string, []>("op_2240_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2240_cast = einsum(equation = var_2240_equation_0, values = (var_2162_cast, var_2119_cast))[name = tensor<string, []>("op_2240_cast")];
            tensor<fp16, []> var_2241_to_fp16 = const()[name = tensor<string, []>("op_2241_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_133_cast = mul(x = var_2240_cast, y = var_2241_to_fp16)[name = tensor<string, []>("aw_133_cast")];
            tensor<string, []> var_2244_equation_0 = const()[name = tensor<string, []>("op_2244_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2244_cast = einsum(equation = var_2244_equation_0, values = (var_2166_cast, var_2123_cast))[name = tensor<string, []>("op_2244_cast")];
            tensor<fp16, []> var_2245_to_fp16 = const()[name = tensor<string, []>("op_2245_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_135_cast = mul(x = var_2244_cast, y = var_2245_to_fp16)[name = tensor<string, []>("aw_135_cast")];
            tensor<string, []> var_2248_equation_0 = const()[name = tensor<string, []>("op_2248_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2248_cast = einsum(equation = var_2248_equation_0, values = (var_2170_cast, var_2127_cast))[name = tensor<string, []>("op_2248_cast")];
            tensor<fp16, []> var_2249_to_fp16 = const()[name = tensor<string, []>("op_2249_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_137_cast = mul(x = var_2248_cast, y = var_2249_to_fp16)[name = tensor<string, []>("aw_137_cast")];
            tensor<string, []> var_2252_equation_0 = const()[name = tensor<string, []>("op_2252_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2252_cast = einsum(equation = var_2252_equation_0, values = (var_2174_cast, var_2131_cast))[name = tensor<string, []>("op_2252_cast")];
            tensor<fp16, []> var_2253_to_fp16 = const()[name = tensor<string, []>("op_2253_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_139_cast = mul(x = var_2252_cast, y = var_2253_to_fp16)[name = tensor<string, []>("aw_139_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2255_cast = softmax(axis = var_298, x = aw_121_cast)[name = tensor<string, []>("op_2255_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2256_cast = softmax(axis = var_298, x = aw_123_cast)[name = tensor<string, []>("op_2256_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2257_cast = softmax(axis = var_298, x = aw_125_cast)[name = tensor<string, []>("op_2257_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2258_cast = softmax(axis = var_298, x = aw_127_cast)[name = tensor<string, []>("op_2258_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2259_cast = softmax(axis = var_298, x = aw_129_cast)[name = tensor<string, []>("op_2259_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2260_cast = softmax(axis = var_298, x = aw_131_cast)[name = tensor<string, []>("op_2260_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2261_cast = softmax(axis = var_298, x = aw_133_cast)[name = tensor<string, []>("op_2261_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2262_cast = softmax(axis = var_298, x = aw_135_cast)[name = tensor<string, []>("op_2262_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2263_cast = softmax(axis = var_298, x = aw_137_cast)[name = tensor<string, []>("op_2263_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_2264_cast = softmax(axis = var_298, x = aw_139_cast)[name = tensor<string, []>("op_2264_cast")];
            tensor<string, []> var_2266_equation_0 = const()[name = tensor<string, []>("op_2266_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2266_cast = einsum(equation = var_2266_equation_0, values = (var_2176_cast, var_2255_cast))[name = tensor<string, []>("op_2266_cast")];
            tensor<string, []> var_2268_equation_0 = const()[name = tensor<string, []>("op_2268_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2268_cast = einsum(equation = var_2268_equation_0, values = (var_2180_cast, var_2256_cast))[name = tensor<string, []>("op_2268_cast")];
            tensor<string, []> var_2270_equation_0 = const()[name = tensor<string, []>("op_2270_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2270_cast = einsum(equation = var_2270_equation_0, values = (var_2184_cast, var_2257_cast))[name = tensor<string, []>("op_2270_cast")];
            tensor<string, []> var_2272_equation_0 = const()[name = tensor<string, []>("op_2272_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2272_cast = einsum(equation = var_2272_equation_0, values = (var_2188_cast, var_2258_cast))[name = tensor<string, []>("op_2272_cast")];
            tensor<string, []> var_2274_equation_0 = const()[name = tensor<string, []>("op_2274_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2274_cast = einsum(equation = var_2274_equation_0, values = (var_2192_cast, var_2259_cast))[name = tensor<string, []>("op_2274_cast")];
            tensor<string, []> var_2276_equation_0 = const()[name = tensor<string, []>("op_2276_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2276_cast = einsum(equation = var_2276_equation_0, values = (var_2196_cast, var_2260_cast))[name = tensor<string, []>("op_2276_cast")];
            tensor<string, []> var_2278_equation_0 = const()[name = tensor<string, []>("op_2278_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2278_cast = einsum(equation = var_2278_equation_0, values = (var_2200_cast, var_2261_cast))[name = tensor<string, []>("op_2278_cast")];
            tensor<string, []> var_2280_equation_0 = const()[name = tensor<string, []>("op_2280_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2280_cast = einsum(equation = var_2280_equation_0, values = (var_2204_cast, var_2262_cast))[name = tensor<string, []>("op_2280_cast")];
            tensor<string, []> var_2282_equation_0 = const()[name = tensor<string, []>("op_2282_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2282_cast = einsum(equation = var_2282_equation_0, values = (var_2208_cast, var_2263_cast))[name = tensor<string, []>("op_2282_cast")];
            tensor<string, []> var_2284_equation_0 = const()[name = tensor<string, []>("op_2284_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2284_cast = einsum(equation = var_2284_equation_0, values = (var_2212_cast, var_2264_cast))[name = tensor<string, []>("op_2284_cast")];
            tensor<bool, []> input_107_interleave_0 = const()[name = tensor<string, []>("input_107_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_107_cast = concat(axis = var_298, interleave = input_107_interleave_0, values = (var_2266_cast, var_2268_cast, var_2270_cast, var_2272_cast, var_2274_cast, var_2276_cast, var_2278_cast, var_2280_cast, var_2282_cast, var_2284_cast))[name = tensor<string, []>("input_107_cast")];
            tensor<int32, [2]> var_2290 = const()[name = tensor<string, []>("op_2290"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2292 = const()[name = tensor<string, []>("op_2292"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_2294_pad_type_0 = const()[name = tensor<string, []>("op_2294_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_2294_pad_0 = const()[name = tensor<string, []>("op_2294_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119959360)))];
            tensor<fp16, [640]> down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120778624)))];
            tensor<fp16, [2, 640, 1, 4096]> var_2294_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_2292, groups = var_298, pad = var_2294_pad_0, pad_type = var_2294_pad_type_0, strides = var_2290, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_107_cast)[name = tensor<string, []>("op_2294_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_21_cast = add(x = var_2294_cast, y = inputs_19_cast)[name = tensor<string, []>("inputs_21_cast")];
            tensor<int32, [1]> var_2298 = const()[name = tensor<string, []>("op_2298"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_21_cast = reduce_mean(axes = var_2298, keep_dims = var_293, x = inputs_21_cast)[name = tensor<string, []>("channels_mean_21_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_21_cast = sub(x = inputs_21_cast, y = channels_mean_21_cast)[name = tensor<string, []>("zero_mean_21_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_21_cast = mul(x = zero_mean_21_cast, y = zero_mean_21_cast)[name = tensor<string, []>("zero_mean_sq_21_cast")];
            tensor<int32, [1]> var_2302 = const()[name = tensor<string, []>("op_2302"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_2303_cast = reduce_mean(axes = var_2302, keep_dims = var_293, x = zero_mean_sq_21_cast)[name = tensor<string, []>("op_2303_cast")];
            tensor<fp16, []> var_2304_to_fp16 = const()[name = tensor<string, []>("op_2304_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_2305_cast = add(x = var_2303_cast, y = var_2304_to_fp16)[name = tensor<string, []>("op_2305_cast")];
            tensor<fp16, []> denom_21_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_21_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_21_cast = rsqrt(epsilon = denom_21_epsilon_0_to_fp16, x = var_2305_cast)[name = tensor<string, []>("denom_21_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_21_cast = mul(x = zero_mean_21_cast, y = denom_21_cast)[name = tensor<string, []>("out_21_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_2309_to_fp16 = const()[name = tensor<string, []>("op_2309_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120779968)))];
            tensor<fp16, [2, 640, 1, 4096]> var_2310_cast = add(x = out_21_cast, y = var_2309_to_fp16)[name = tensor<string, []>("op_2310_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_2312_to_fp16 = const()[name = tensor<string, []>("op_2312_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120781312)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_49_cast = mul(x = var_2310_cast, y = var_2312_to_fp16)[name = tensor<string, []>("hidden_states_49_cast")];
            tensor<int32, [2]> var_2319 = const()[name = tensor<string, []>("op_2319"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2321 = const()[name = tensor<string, []>("op_2321"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_15_pad_type_0 = const()[name = tensor<string, []>("q_15_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_15_pad_0 = const()[name = tensor<string, []>("q_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120782656)))];
            tensor<fp16, [2, 640, 1, 4096]> q_15_cast = conv(dilations = var_2321, groups = var_298, pad = q_15_pad_0, pad_type = q_15_pad_type_0, strides = var_2319, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_49_cast)[name = tensor<string, []>("q_15_cast")];
            tensor<int32, [2]> var_2325 = const()[name = tensor<string, []>("op_2325"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2327 = const()[name = tensor<string, []>("op_2327"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_29_pad_type_0 = const()[name = tensor<string, []>("k_29_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_29_pad_0 = const()[name = tensor<string, []>("k_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121601920)))];
            tensor<fp16, [2, 640, 1, 77]> k_29_cast = conv(dilations = var_2327, groups = var_298, pad = k_29_pad_0, pad_type = k_29_pad_type_0, strides = var_2325, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_29_cast")];
            tensor<int32, [2]> var_2331 = const()[name = tensor<string, []>("op_2331"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2333 = const()[name = tensor<string, []>("op_2333"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_15_pad_type_0 = const()[name = tensor<string, []>("v_15_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_15_pad_0 = const()[name = tensor<string, []>("v_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124223424)))];
            tensor<fp16, [2, 640, 1, 77]> v_15_cast = conv(dilations = var_2333, groups = var_298, pad = v_15_pad_0, pad_type = v_15_pad_type_0, strides = var_2331, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_15_cast")];
            tensor<int32, [4]> var_2337_begin_0 = const()[name = tensor<string, []>("op_2337_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2337_end_0 = const()[name = tensor<string, []>("op_2337_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_2337_end_mask_0 = const()[name = tensor<string, []>("op_2337_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2337_cast = slice_by_index(begin = var_2337_begin_0, end = var_2337_end_0, end_mask = var_2337_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_2337_cast")];
            tensor<int32, [4]> var_2341_begin_0 = const()[name = tensor<string, []>("op_2341_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_2341_end_0 = const()[name = tensor<string, []>("op_2341_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_2341_end_mask_0 = const()[name = tensor<string, []>("op_2341_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2341_cast = slice_by_index(begin = var_2341_begin_0, end = var_2341_end_0, end_mask = var_2341_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_2341_cast")];
            tensor<int32, [4]> var_2345_begin_0 = const()[name = tensor<string, []>("op_2345_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_2345_end_0 = const()[name = tensor<string, []>("op_2345_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_2345_end_mask_0 = const()[name = tensor<string, []>("op_2345_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2345_cast = slice_by_index(begin = var_2345_begin_0, end = var_2345_end_0, end_mask = var_2345_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_2345_cast")];
            tensor<int32, [4]> var_2349_begin_0 = const()[name = tensor<string, []>("op_2349_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_2349_end_0 = const()[name = tensor<string, []>("op_2349_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_2349_end_mask_0 = const()[name = tensor<string, []>("op_2349_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2349_cast = slice_by_index(begin = var_2349_begin_0, end = var_2349_end_0, end_mask = var_2349_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_2349_cast")];
            tensor<int32, [4]> var_2353_begin_0 = const()[name = tensor<string, []>("op_2353_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_2353_end_0 = const()[name = tensor<string, []>("op_2353_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_2353_end_mask_0 = const()[name = tensor<string, []>("op_2353_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2353_cast = slice_by_index(begin = var_2353_begin_0, end = var_2353_end_0, end_mask = var_2353_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_2353_cast")];
            tensor<int32, [4]> var_2357_begin_0 = const()[name = tensor<string, []>("op_2357_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_2357_end_0 = const()[name = tensor<string, []>("op_2357_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_2357_end_mask_0 = const()[name = tensor<string, []>("op_2357_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2357_cast = slice_by_index(begin = var_2357_begin_0, end = var_2357_end_0, end_mask = var_2357_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_2357_cast")];
            tensor<int32, [4]> var_2361_begin_0 = const()[name = tensor<string, []>("op_2361_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_2361_end_0 = const()[name = tensor<string, []>("op_2361_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_2361_end_mask_0 = const()[name = tensor<string, []>("op_2361_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2361_cast = slice_by_index(begin = var_2361_begin_0, end = var_2361_end_0, end_mask = var_2361_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_2361_cast")];
            tensor<int32, [4]> var_2365_begin_0 = const()[name = tensor<string, []>("op_2365_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_2365_end_0 = const()[name = tensor<string, []>("op_2365_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_2365_end_mask_0 = const()[name = tensor<string, []>("op_2365_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2365_cast = slice_by_index(begin = var_2365_begin_0, end = var_2365_end_0, end_mask = var_2365_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_2365_cast")];
            tensor<int32, [4]> var_2369_begin_0 = const()[name = tensor<string, []>("op_2369_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_2369_end_0 = const()[name = tensor<string, []>("op_2369_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_2369_end_mask_0 = const()[name = tensor<string, []>("op_2369_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2369_cast = slice_by_index(begin = var_2369_begin_0, end = var_2369_end_0, end_mask = var_2369_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_2369_cast")];
            tensor<int32, [4]> var_2373_begin_0 = const()[name = tensor<string, []>("op_2373_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_2373_end_0 = const()[name = tensor<string, []>("op_2373_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_2373_end_mask_0 = const()[name = tensor<string, []>("op_2373_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_2373_cast = slice_by_index(begin = var_2373_begin_0, end = var_2373_end_0, end_mask = var_2373_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_2373_cast")];
            tensor<int32, [4]> k_31_perm_0 = const()[name = tensor<string, []>("k_31_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_2380_begin_0 = const()[name = tensor<string, []>("op_2380_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2380_end_0 = const()[name = tensor<string, []>("op_2380_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_2380_end_mask_0 = const()[name = tensor<string, []>("op_2380_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> transpose_132 = transpose(perm = k_31_perm_0, x = k_29_cast)[name = tensor<string, []>("transpose_132")];
            tensor<fp16, [2, 77, 1, 64]> var_2380_cast = slice_by_index(begin = var_2380_begin_0, end = var_2380_end_0, end_mask = var_2380_end_mask_0, x = transpose_132)[name = tensor<string, []>("op_2380_cast")];
            tensor<int32, [4]> var_2384_begin_0 = const()[name = tensor<string, []>("op_2384_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_2384_end_0 = const()[name = tensor<string, []>("op_2384_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_2384_end_mask_0 = const()[name = tensor<string, []>("op_2384_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2384_cast = slice_by_index(begin = var_2384_begin_0, end = var_2384_end_0, end_mask = var_2384_end_mask_0, x = transpose_132)[name = tensor<string, []>("op_2384_cast")];
            tensor<int32, [4]> var_2388_begin_0 = const()[name = tensor<string, []>("op_2388_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_2388_end_0 = const()[name = tensor<string, []>("op_2388_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_2388_end_mask_0 = const()[name = tensor<string, []>("op_2388_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2388_cast = slice_by_index(begin = var_2388_begin_0, end = var_2388_end_0, end_mask = var_2388_end_mask_0, x = transpose_132)[name = tensor<string, []>("op_2388_cast")];
            tensor<int32, [4]> var_2392_begin_0 = const()[name = tensor<string, []>("op_2392_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_2392_end_0 = const()[name = tensor<string, []>("op_2392_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_2392_end_mask_0 = const()[name = tensor<string, []>("op_2392_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2392_cast = slice_by_index(begin = var_2392_begin_0, end = var_2392_end_0, end_mask = var_2392_end_mask_0, x = transpose_132)[name = tensor<string, []>("op_2392_cast")];
            tensor<int32, [4]> var_2396_begin_0 = const()[name = tensor<string, []>("op_2396_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_2396_end_0 = const()[name = tensor<string, []>("op_2396_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_2396_end_mask_0 = const()[name = tensor<string, []>("op_2396_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2396_cast = slice_by_index(begin = var_2396_begin_0, end = var_2396_end_0, end_mask = var_2396_end_mask_0, x = transpose_132)[name = tensor<string, []>("op_2396_cast")];
            tensor<int32, [4]> var_2400_begin_0 = const()[name = tensor<string, []>("op_2400_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_2400_end_0 = const()[name = tensor<string, []>("op_2400_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_2400_end_mask_0 = const()[name = tensor<string, []>("op_2400_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2400_cast = slice_by_index(begin = var_2400_begin_0, end = var_2400_end_0, end_mask = var_2400_end_mask_0, x = transpose_132)[name = tensor<string, []>("op_2400_cast")];
            tensor<int32, [4]> var_2404_begin_0 = const()[name = tensor<string, []>("op_2404_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_2404_end_0 = const()[name = tensor<string, []>("op_2404_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_2404_end_mask_0 = const()[name = tensor<string, []>("op_2404_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2404_cast = slice_by_index(begin = var_2404_begin_0, end = var_2404_end_0, end_mask = var_2404_end_mask_0, x = transpose_132)[name = tensor<string, []>("op_2404_cast")];
            tensor<int32, [4]> var_2408_begin_0 = const()[name = tensor<string, []>("op_2408_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_2408_end_0 = const()[name = tensor<string, []>("op_2408_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_2408_end_mask_0 = const()[name = tensor<string, []>("op_2408_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2408_cast = slice_by_index(begin = var_2408_begin_0, end = var_2408_end_0, end_mask = var_2408_end_mask_0, x = transpose_132)[name = tensor<string, []>("op_2408_cast")];
            tensor<int32, [4]> var_2412_begin_0 = const()[name = tensor<string, []>("op_2412_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2412_end_0 = const()[name = tensor<string, []>("op_2412_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_2412_end_mask_0 = const()[name = tensor<string, []>("op_2412_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2412_cast = slice_by_index(begin = var_2412_begin_0, end = var_2412_end_0, end_mask = var_2412_end_mask_0, x = transpose_132)[name = tensor<string, []>("op_2412_cast")];
            tensor<int32, [4]> var_2416_begin_0 = const()[name = tensor<string, []>("op_2416_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_2416_end_0 = const()[name = tensor<string, []>("op_2416_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_2416_end_mask_0 = const()[name = tensor<string, []>("op_2416_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2416_cast = slice_by_index(begin = var_2416_begin_0, end = var_2416_end_0, end_mask = var_2416_end_mask_0, x = transpose_132)[name = tensor<string, []>("op_2416_cast")];
            tensor<int32, [4]> var_2418_begin_0 = const()[name = tensor<string, []>("op_2418_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2418_end_0 = const()[name = tensor<string, []>("op_2418_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_2418_end_mask_0 = const()[name = tensor<string, []>("op_2418_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2418_cast = slice_by_index(begin = var_2418_begin_0, end = var_2418_end_0, end_mask = var_2418_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_2418_cast")];
            tensor<int32, [4]> var_2422_begin_0 = const()[name = tensor<string, []>("op_2422_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_2422_end_0 = const()[name = tensor<string, []>("op_2422_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_2422_end_mask_0 = const()[name = tensor<string, []>("op_2422_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2422_cast = slice_by_index(begin = var_2422_begin_0, end = var_2422_end_0, end_mask = var_2422_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_2422_cast")];
            tensor<int32, [4]> var_2426_begin_0 = const()[name = tensor<string, []>("op_2426_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_2426_end_0 = const()[name = tensor<string, []>("op_2426_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_2426_end_mask_0 = const()[name = tensor<string, []>("op_2426_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2426_cast = slice_by_index(begin = var_2426_begin_0, end = var_2426_end_0, end_mask = var_2426_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_2426_cast")];
            tensor<int32, [4]> var_2430_begin_0 = const()[name = tensor<string, []>("op_2430_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_2430_end_0 = const()[name = tensor<string, []>("op_2430_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_2430_end_mask_0 = const()[name = tensor<string, []>("op_2430_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2430_cast = slice_by_index(begin = var_2430_begin_0, end = var_2430_end_0, end_mask = var_2430_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_2430_cast")];
            tensor<int32, [4]> var_2434_begin_0 = const()[name = tensor<string, []>("op_2434_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_2434_end_0 = const()[name = tensor<string, []>("op_2434_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_2434_end_mask_0 = const()[name = tensor<string, []>("op_2434_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2434_cast = slice_by_index(begin = var_2434_begin_0, end = var_2434_end_0, end_mask = var_2434_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_2434_cast")];
            tensor<int32, [4]> var_2438_begin_0 = const()[name = tensor<string, []>("op_2438_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_2438_end_0 = const()[name = tensor<string, []>("op_2438_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_2438_end_mask_0 = const()[name = tensor<string, []>("op_2438_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2438_cast = slice_by_index(begin = var_2438_begin_0, end = var_2438_end_0, end_mask = var_2438_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_2438_cast")];
            tensor<int32, [4]> var_2442_begin_0 = const()[name = tensor<string, []>("op_2442_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_2442_end_0 = const()[name = tensor<string, []>("op_2442_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_2442_end_mask_0 = const()[name = tensor<string, []>("op_2442_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2442_cast = slice_by_index(begin = var_2442_begin_0, end = var_2442_end_0, end_mask = var_2442_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_2442_cast")];
            tensor<int32, [4]> var_2446_begin_0 = const()[name = tensor<string, []>("op_2446_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_2446_end_0 = const()[name = tensor<string, []>("op_2446_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_2446_end_mask_0 = const()[name = tensor<string, []>("op_2446_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2446_cast = slice_by_index(begin = var_2446_begin_0, end = var_2446_end_0, end_mask = var_2446_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_2446_cast")];
            tensor<int32, [4]> var_2450_begin_0 = const()[name = tensor<string, []>("op_2450_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_2450_end_0 = const()[name = tensor<string, []>("op_2450_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_2450_end_mask_0 = const()[name = tensor<string, []>("op_2450_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2450_cast = slice_by_index(begin = var_2450_begin_0, end = var_2450_end_0, end_mask = var_2450_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_2450_cast")];
            tensor<int32, [4]> var_2454_begin_0 = const()[name = tensor<string, []>("op_2454_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_2454_end_0 = const()[name = tensor<string, []>("op_2454_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_2454_end_mask_0 = const()[name = tensor<string, []>("op_2454_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2454_cast = slice_by_index(begin = var_2454_begin_0, end = var_2454_end_0, end_mask = var_2454_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_2454_cast")];
            tensor<string, []> var_2458_equation_0 = const()[name = tensor<string, []>("op_2458_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_2458_cast = einsum(equation = var_2458_equation_0, values = (var_2380_cast, var_2337_cast))[name = tensor<string, []>("op_2458_cast")];
            tensor<fp16, []> var_2459_to_fp16 = const()[name = tensor<string, []>("op_2459_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_141_cast = mul(x = var_2458_cast, y = var_2459_to_fp16)[name = tensor<string, []>("aw_141_cast")];
            tensor<string, []> var_2462_equation_0 = const()[name = tensor<string, []>("op_2462_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_2462_cast = einsum(equation = var_2462_equation_0, values = (var_2384_cast, var_2341_cast))[name = tensor<string, []>("op_2462_cast")];
            tensor<fp16, []> var_2463_to_fp16 = const()[name = tensor<string, []>("op_2463_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_143_cast = mul(x = var_2462_cast, y = var_2463_to_fp16)[name = tensor<string, []>("aw_143_cast")];
            tensor<string, []> var_2466_equation_0 = const()[name = tensor<string, []>("op_2466_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_2466_cast = einsum(equation = var_2466_equation_0, values = (var_2388_cast, var_2345_cast))[name = tensor<string, []>("op_2466_cast")];
            tensor<fp16, []> var_2467_to_fp16 = const()[name = tensor<string, []>("op_2467_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_145_cast = mul(x = var_2466_cast, y = var_2467_to_fp16)[name = tensor<string, []>("aw_145_cast")];
            tensor<string, []> var_2470_equation_0 = const()[name = tensor<string, []>("op_2470_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_2470_cast = einsum(equation = var_2470_equation_0, values = (var_2392_cast, var_2349_cast))[name = tensor<string, []>("op_2470_cast")];
            tensor<fp16, []> var_2471_to_fp16 = const()[name = tensor<string, []>("op_2471_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_147_cast = mul(x = var_2470_cast, y = var_2471_to_fp16)[name = tensor<string, []>("aw_147_cast")];
            tensor<string, []> var_2474_equation_0 = const()[name = tensor<string, []>("op_2474_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_2474_cast = einsum(equation = var_2474_equation_0, values = (var_2396_cast, var_2353_cast))[name = tensor<string, []>("op_2474_cast")];
            tensor<fp16, []> var_2475_to_fp16 = const()[name = tensor<string, []>("op_2475_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_149_cast = mul(x = var_2474_cast, y = var_2475_to_fp16)[name = tensor<string, []>("aw_149_cast")];
            tensor<string, []> var_2478_equation_0 = const()[name = tensor<string, []>("op_2478_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_2478_cast = einsum(equation = var_2478_equation_0, values = (var_2400_cast, var_2357_cast))[name = tensor<string, []>("op_2478_cast")];
            tensor<fp16, []> var_2479_to_fp16 = const()[name = tensor<string, []>("op_2479_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_151_cast = mul(x = var_2478_cast, y = var_2479_to_fp16)[name = tensor<string, []>("aw_151_cast")];
            tensor<string, []> var_2482_equation_0 = const()[name = tensor<string, []>("op_2482_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_2482_cast = einsum(equation = var_2482_equation_0, values = (var_2404_cast, var_2361_cast))[name = tensor<string, []>("op_2482_cast")];
            tensor<fp16, []> var_2483_to_fp16 = const()[name = tensor<string, []>("op_2483_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_153_cast = mul(x = var_2482_cast, y = var_2483_to_fp16)[name = tensor<string, []>("aw_153_cast")];
            tensor<string, []> var_2486_equation_0 = const()[name = tensor<string, []>("op_2486_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_2486_cast = einsum(equation = var_2486_equation_0, values = (var_2408_cast, var_2365_cast))[name = tensor<string, []>("op_2486_cast")];
            tensor<fp16, []> var_2487_to_fp16 = const()[name = tensor<string, []>("op_2487_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_155_cast = mul(x = var_2486_cast, y = var_2487_to_fp16)[name = tensor<string, []>("aw_155_cast")];
            tensor<string, []> var_2490_equation_0 = const()[name = tensor<string, []>("op_2490_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_2490_cast = einsum(equation = var_2490_equation_0, values = (var_2412_cast, var_2369_cast))[name = tensor<string, []>("op_2490_cast")];
            tensor<fp16, []> var_2491_to_fp16 = const()[name = tensor<string, []>("op_2491_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_157_cast = mul(x = var_2490_cast, y = var_2491_to_fp16)[name = tensor<string, []>("aw_157_cast")];
            tensor<string, []> var_2494_equation_0 = const()[name = tensor<string, []>("op_2494_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_2494_cast = einsum(equation = var_2494_equation_0, values = (var_2416_cast, var_2373_cast))[name = tensor<string, []>("op_2494_cast")];
            tensor<fp16, []> var_2495_to_fp16 = const()[name = tensor<string, []>("op_2495_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_159_cast = mul(x = var_2494_cast, y = var_2495_to_fp16)[name = tensor<string, []>("aw_159_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_2497_cast = softmax(axis = var_298, x = aw_141_cast)[name = tensor<string, []>("op_2497_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_2498_cast = softmax(axis = var_298, x = aw_143_cast)[name = tensor<string, []>("op_2498_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_2499_cast = softmax(axis = var_298, x = aw_145_cast)[name = tensor<string, []>("op_2499_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_2500_cast = softmax(axis = var_298, x = aw_147_cast)[name = tensor<string, []>("op_2500_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_2501_cast = softmax(axis = var_298, x = aw_149_cast)[name = tensor<string, []>("op_2501_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_2502_cast = softmax(axis = var_298, x = aw_151_cast)[name = tensor<string, []>("op_2502_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_2503_cast = softmax(axis = var_298, x = aw_153_cast)[name = tensor<string, []>("op_2503_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_2504_cast = softmax(axis = var_298, x = aw_155_cast)[name = tensor<string, []>("op_2504_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_2505_cast = softmax(axis = var_298, x = aw_157_cast)[name = tensor<string, []>("op_2505_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_2506_cast = softmax(axis = var_298, x = aw_159_cast)[name = tensor<string, []>("op_2506_cast")];
            tensor<string, []> var_2508_equation_0 = const()[name = tensor<string, []>("op_2508_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2508_cast = einsum(equation = var_2508_equation_0, values = (var_2418_cast, var_2497_cast))[name = tensor<string, []>("op_2508_cast")];
            tensor<string, []> var_2510_equation_0 = const()[name = tensor<string, []>("op_2510_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2510_cast = einsum(equation = var_2510_equation_0, values = (var_2422_cast, var_2498_cast))[name = tensor<string, []>("op_2510_cast")];
            tensor<string, []> var_2512_equation_0 = const()[name = tensor<string, []>("op_2512_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2512_cast = einsum(equation = var_2512_equation_0, values = (var_2426_cast, var_2499_cast))[name = tensor<string, []>("op_2512_cast")];
            tensor<string, []> var_2514_equation_0 = const()[name = tensor<string, []>("op_2514_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2514_cast = einsum(equation = var_2514_equation_0, values = (var_2430_cast, var_2500_cast))[name = tensor<string, []>("op_2514_cast")];
            tensor<string, []> var_2516_equation_0 = const()[name = tensor<string, []>("op_2516_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2516_cast = einsum(equation = var_2516_equation_0, values = (var_2434_cast, var_2501_cast))[name = tensor<string, []>("op_2516_cast")];
            tensor<string, []> var_2518_equation_0 = const()[name = tensor<string, []>("op_2518_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2518_cast = einsum(equation = var_2518_equation_0, values = (var_2438_cast, var_2502_cast))[name = tensor<string, []>("op_2518_cast")];
            tensor<string, []> var_2520_equation_0 = const()[name = tensor<string, []>("op_2520_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2520_cast = einsum(equation = var_2520_equation_0, values = (var_2442_cast, var_2503_cast))[name = tensor<string, []>("op_2520_cast")];
            tensor<string, []> var_2522_equation_0 = const()[name = tensor<string, []>("op_2522_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2522_cast = einsum(equation = var_2522_equation_0, values = (var_2446_cast, var_2504_cast))[name = tensor<string, []>("op_2522_cast")];
            tensor<string, []> var_2524_equation_0 = const()[name = tensor<string, []>("op_2524_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2524_cast = einsum(equation = var_2524_equation_0, values = (var_2450_cast, var_2505_cast))[name = tensor<string, []>("op_2524_cast")];
            tensor<string, []> var_2526_equation_0 = const()[name = tensor<string, []>("op_2526_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_2526_cast = einsum(equation = var_2526_equation_0, values = (var_2454_cast, var_2506_cast))[name = tensor<string, []>("op_2526_cast")];
            tensor<bool, []> input_109_interleave_0 = const()[name = tensor<string, []>("input_109_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_109_cast = concat(axis = var_298, interleave = input_109_interleave_0, values = (var_2508_cast, var_2510_cast, var_2512_cast, var_2514_cast, var_2516_cast, var_2518_cast, var_2520_cast, var_2522_cast, var_2524_cast, var_2526_cast))[name = tensor<string, []>("input_109_cast")];
            tensor<int32, [2]> var_2532 = const()[name = tensor<string, []>("op_2532"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2534 = const()[name = tensor<string, []>("op_2534"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_2536_pad_type_0 = const()[name = tensor<string, []>("op_2536_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_2536_pad_0 = const()[name = tensor<string, []>("op_2536_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126844928)))];
            tensor<fp16, [640]> down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127664192)))];
            tensor<fp16, [2, 640, 1, 4096]> var_2536_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_2534, groups = var_298, pad = var_2536_pad_0, pad_type = var_2536_pad_type_0, strides = var_2532, weight = down_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_109_cast)[name = tensor<string, []>("op_2536_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_23_cast = add(x = var_2536_cast, y = inputs_21_cast)[name = tensor<string, []>("inputs_23_cast")];
            tensor<int32, [1]> var_2540 = const()[name = tensor<string, []>("op_2540"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_23_cast = reduce_mean(axes = var_2540, keep_dims = var_293, x = inputs_23_cast)[name = tensor<string, []>("channels_mean_23_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_23_cast = sub(x = inputs_23_cast, y = channels_mean_23_cast)[name = tensor<string, []>("zero_mean_23_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_23_cast = mul(x = zero_mean_23_cast, y = zero_mean_23_cast)[name = tensor<string, []>("zero_mean_sq_23_cast")];
            tensor<int32, [1]> var_2544 = const()[name = tensor<string, []>("op_2544"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_2545_cast = reduce_mean(axes = var_2544, keep_dims = var_293, x = zero_mean_sq_23_cast)[name = tensor<string, []>("op_2545_cast")];
            tensor<fp16, []> var_2546_to_fp16 = const()[name = tensor<string, []>("op_2546_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_2547_cast = add(x = var_2545_cast, y = var_2546_to_fp16)[name = tensor<string, []>("op_2547_cast")];
            tensor<fp16, []> denom_23_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_23_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_23_cast = rsqrt(epsilon = denom_23_epsilon_0_to_fp16, x = var_2547_cast)[name = tensor<string, []>("denom_23_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_23_cast = mul(x = zero_mean_23_cast, y = denom_23_cast)[name = tensor<string, []>("out_23_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_2551_to_fp16 = const()[name = tensor<string, []>("op_2551_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127665536)))];
            tensor<fp16, [2, 640, 1, 4096]> var_2552_cast = add(x = out_23_cast, y = var_2551_to_fp16)[name = tensor<string, []>("op_2552_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_2554_to_fp16 = const()[name = tensor<string, []>("op_2554_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127666880)))];
            tensor<fp16, [2, 640, 1, 4096]> input_111_cast = mul(x = var_2552_cast, y = var_2554_to_fp16)[name = tensor<string, []>("input_111_cast")];
            tensor<int32, [2]> var_2562 = const()[name = tensor<string, []>("op_2562"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2564 = const()[name = tensor<string, []>("op_2564"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_2566_pad_type_0 = const()[name = tensor<string, []>("op_2566_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_2566_pad_0 = const()[name = tensor<string, []>("op_2566_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [5120, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [5120, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127668224)))];
            tensor<fp16, [5120]> down_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134221888)))];
            tensor<fp16, [2, 5120, 1, 4096]> var_2566_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_2564, groups = var_298, pad = var_2566_pad_0, pad_type = var_2566_pad_type_0, strides = var_2562, weight = down_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_111_cast)[name = tensor<string, []>("op_2566_cast")];
            tensor<int32, [2]> var_2567_split_sizes_0 = const()[name = tensor<string, []>("op_2567_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_2567_axis_0 = const()[name = tensor<string, []>("op_2567_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 4096]> var_2567_cast_0, tensor<fp16, [2, 2560, 1, 4096]> var_2567_cast_1 = split(axis = var_2567_axis_0, split_sizes = var_2567_split_sizes_0, x = var_2566_cast)[name = tensor<string, []>("op_2567_cast")];
            tensor<string, []> var_2569_mode_0 = const()[name = tensor<string, []>("op_2569_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 4096]> var_2569_cast = gelu(mode = var_2569_mode_0, x = var_2567_cast_1)[name = tensor<string, []>("op_2569_cast")];
            tensor<fp16, [2, 2560, 1, 4096]> input_113_cast = mul(x = var_2567_cast_0, y = var_2569_cast)[name = tensor<string, []>("input_113_cast")];
            tensor<int32, [2]> var_2573 = const()[name = tensor<string, []>("op_2573"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2575 = const()[name = tensor<string, []>("op_2575"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_2577_pad_type_0 = const()[name = tensor<string, []>("op_2577_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_2577_pad_0 = const()[name = tensor<string, []>("op_2577_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2560, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor<fp16, [640, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134232192)))];
            tensor<fp16, [640]> down_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137509056)))];
            tensor<fp16, [2, 640, 1, 4096]> var_2577_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_2575, groups = var_298, pad = var_2577_pad_0, pad_type = var_2577_pad_type_0, strides = var_2573, weight = down_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_113_cast)[name = tensor<string, []>("op_2577_cast")];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_53_cast = add(x = var_2577_cast, y = inputs_23_cast)[name = tensor<string, []>("hidden_states_53_cast")];
            tensor<int32, [4]> var_2579 = const()[name = tensor<string, []>("op_2579"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> input_115_cast = reshape(shape = var_2579, x = hidden_states_53_cast)[name = tensor<string, []>("input_115_cast")];
            tensor<int32, [2]> var_2583 = const()[name = tensor<string, []>("op_2583"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2585 = const()[name = tensor<string, []>("op_2585"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_55_pad_type_0 = const()[name = tensor<string, []>("hidden_states_55_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_55_pad_0 = const()[name = tensor<string, []>("hidden_states_55_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_proj_out_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137510400)))];
            tensor<fp16, [640]> down_blocks_1_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_proj_out_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138329664)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_55_cast = conv(bias = down_blocks_1_attentions_1_proj_out_bias_to_fp16, dilations = var_2585, groups = var_298, pad = hidden_states_55_pad_0, pad_type = hidden_states_55_pad_type_0, strides = var_2583, weight = down_blocks_1_attentions_1_proj_out_weight_to_fp16, x = input_115_cast)[name = tensor<string, []>("hidden_states_55_cast")];
            tensor<fp16, [2, 640, 64, 64]> input_117_cast = add(x = hidden_states_55_cast, y = hidden_states_37_cast)[name = tensor<string, []>("input_117_cast")];
            tensor<int32, [2]> var_2592 = const()[name = tensor<string, []>("op_2592"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> var_2594 = const()[name = tensor<string, []>("op_2594"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> input_119_pad_type_0 = const()[name = tensor<string, []>("input_119_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> input_119_pad_0 = const()[name = tensor<string, []>("input_119_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 640, 3, 3]> down_blocks_1_downsamplers_0_conv_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_downsamplers_0_conv_weight_to_fp16"), val = tensor<fp16, [640, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138331008)))];
            tensor<fp16, [640]> down_blocks_1_downsamplers_0_conv_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_downsamplers_0_conv_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145703872)))];
            tensor<fp16, [2, 640, 32, 32]> input_119_cast = conv(bias = down_blocks_1_downsamplers_0_conv_bias_to_fp16, dilations = var_2594, groups = var_298, pad = input_119_pad_0, pad_type = input_119_pad_type_0, strides = var_2592, weight = down_blocks_1_downsamplers_0_conv_weight_to_fp16, x = input_117_cast)[name = tensor<string, []>("input_119_cast")];
            tensor<bool, []> var_2629 = const()[name = tensor<string, []>("op_2629"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_2634 = const()[name = tensor<string, []>("op_2634"), val = tensor<int32, []>(1)];
            tensor<int32, [5]> reshape_40_shape_0 = const()[name = tensor<string, []>("reshape_40_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 32, 32])];
            tensor<fp16, [2, 32, 20, 32, 32]> reshape_40_cast = reshape(shape = reshape_40_shape_0, x = input_119_cast)[name = tensor<string, []>("reshape_40_cast")];
            tensor<int32, [3]> reduce_mean_30_axes_0 = const()[name = tensor<string, []>("reduce_mean_30_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_30_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_30_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_30_cast = reduce_mean(axes = reduce_mean_30_axes_0, keep_dims = reduce_mean_30_keep_dims_0, x = reshape_40_cast)[name = tensor<string, []>("reduce_mean_30_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> sub_20_cast = sub(x = reshape_40_cast, y = reduce_mean_30_cast)[name = tensor<string, []>("sub_20_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> square_10_cast = square(x = sub_20_cast)[name = tensor<string, []>("square_10_cast")];
            tensor<int32, [3]> reduce_mean_32_axes_0 = const()[name = tensor<string, []>("reduce_mean_32_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_32_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_32_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_32_cast = reduce_mean(axes = reduce_mean_32_axes_0, keep_dims = reduce_mean_32_keep_dims_0, x = square_10_cast)[name = tensor<string, []>("reduce_mean_32_cast")];
            tensor<fp16, []> add_20_y_0_to_fp16 = const()[name = tensor<string, []>("add_20_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_20_cast = add(x = reduce_mean_32_cast, y = add_20_y_0_to_fp16)[name = tensor<string, []>("add_20_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_10_cast = sqrt(x = add_20_cast)[name = tensor<string, []>("sqrt_10_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_10_cast = real_div(x = sub_20_cast, y = sqrt_10_cast)[name = tensor<string, []>("real_div_10_cast")];
            tensor<int32, [4]> reshape_41_shape_0 = const()[name = tensor<string, []>("reshape_41_shape_0"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> reshape_41_cast = reshape(shape = reshape_41_shape_0, x = real_div_10_cast)[name = tensor<string, []>("reshape_41_cast")];
            tensor<fp16, [640]> add_21_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_21_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145705216)))];
            tensor<fp16, [640]> add_21_beta_0_to_fp16 = const()[name = tensor<string, []>("add_21_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145706560)))];
            tensor<fp16, []> add_21_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_21_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 32, 32]> add_21_cast = batch_norm(beta = add_21_beta_0_to_fp16, epsilon = add_21_epsilon_0_to_fp16, gamma = add_21_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_41_cast)[name = tensor<string, []>("add_21_cast")];
            tensor<fp16, [2, 640, 32, 32]> input_123_cast = silu(x = add_21_cast)[name = tensor<string, []>("input_123_cast")];
            tensor<int32, [2]> var_2655 = const()[name = tensor<string, []>("op_2655"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2657 = const()[name = tensor<string, []>("op_2657"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_57_pad_type_0 = const()[name = tensor<string, []>("hidden_states_57_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_57_pad_0 = const()[name = tensor<string, []>("hidden_states_57_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 640, 3, 3]> down_blocks_2_resnets_0_conv1_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145707904)))];
            tensor<fp16, [1280]> down_blocks_2_resnets_0_conv1_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160453568)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_57_cast = conv(bias = down_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = var_2657, groups = var_2634, pad = hidden_states_57_pad_0, pad_type = hidden_states_57_pad_type_0, strides = var_2655, weight = down_blocks_2_resnets_0_conv1_weight_to_fp16, x = input_123_cast)[name = tensor<string, []>("hidden_states_57_cast")];
            tensor<int32, [2]> var_2663 = const()[name = tensor<string, []>("op_2663"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2665 = const()[name = tensor<string, []>("op_2665"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_9_pad_type_0 = const()[name = tensor<string, []>("temb_9_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_9_pad_0 = const()[name = tensor<string, []>("temb_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160456192)))];
            tensor<fp16, [1280]> down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163733056)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_9_cast = conv(bias = down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_2665, groups = var_2634, pad = temb_9_pad_0, pad_type = temb_9_pad_type_0, strides = var_2663, weight = down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16, x = input_23_cast)[name = tensor<string, []>("temb_9_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_127_cast = add(x = hidden_states_57_cast, y = temb_9_cast)[name = tensor<string, []>("input_127_cast")];
            tensor<int32, [5]> reshape_44_shape_0 = const()[name = tensor<string, []>("reshape_44_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_44_cast = reshape(shape = reshape_44_shape_0, x = input_127_cast)[name = tensor<string, []>("reshape_44_cast")];
            tensor<int32, [3]> reduce_mean_33_axes_0 = const()[name = tensor<string, []>("reduce_mean_33_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_33_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_33_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_33_cast = reduce_mean(axes = reduce_mean_33_axes_0, keep_dims = reduce_mean_33_keep_dims_0, x = reshape_44_cast)[name = tensor<string, []>("reduce_mean_33_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_22_cast = sub(x = reshape_44_cast, y = reduce_mean_33_cast)[name = tensor<string, []>("sub_22_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_11_cast = square(x = sub_22_cast)[name = tensor<string, []>("square_11_cast")];
            tensor<int32, [3]> reduce_mean_35_axes_0 = const()[name = tensor<string, []>("reduce_mean_35_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_35_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_35_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_35_cast = reduce_mean(axes = reduce_mean_35_axes_0, keep_dims = reduce_mean_35_keep_dims_0, x = square_11_cast)[name = tensor<string, []>("reduce_mean_35_cast")];
            tensor<fp16, []> add_22_y_0_to_fp16 = const()[name = tensor<string, []>("add_22_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_22_cast = add(x = reduce_mean_35_cast, y = add_22_y_0_to_fp16)[name = tensor<string, []>("add_22_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_11_cast = sqrt(x = add_22_cast)[name = tensor<string, []>("sqrt_11_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_11_cast = real_div(x = sub_22_cast, y = sqrt_11_cast)[name = tensor<string, []>("real_div_11_cast")];
            tensor<int32, [4]> reshape_45_shape_0 = const()[name = tensor<string, []>("reshape_45_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_45_cast = reshape(shape = reshape_45_shape_0, x = real_div_11_cast)[name = tensor<string, []>("reshape_45_cast")];
            tensor<fp16, [1280]> add_23_mean_0_to_fp16 = const()[name = tensor<string, []>("add_23_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163735680)))];
            tensor<fp16, [1280]> add_23_variance_0_to_fp16 = const()[name = tensor<string, []>("add_23_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163738304)))];
            tensor<fp16, [1280]> add_23_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_23_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163740928)))];
            tensor<fp16, [1280]> add_23_beta_0_to_fp16 = const()[name = tensor<string, []>("add_23_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163743552)))];
            tensor<fp16, []> add_23_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_23_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 32, 32]> add_23_cast = batch_norm(beta = add_23_beta_0_to_fp16, epsilon = add_23_epsilon_0_to_fp16, gamma = add_23_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_45_cast)[name = tensor<string, []>("add_23_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_131_cast = silu(x = add_23_cast)[name = tensor<string, []>("input_131_cast")];
            tensor<int32, [2]> var_2675 = const()[name = tensor<string, []>("op_2675"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2677 = const()[name = tensor<string, []>("op_2677"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_59_pad_type_0 = const()[name = tensor<string, []>("hidden_states_59_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_59_pad_0 = const()[name = tensor<string, []>("hidden_states_59_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> down_blocks_2_resnets_0_conv2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163746176)))];
            tensor<fp16, [1280]> down_blocks_2_resnets_0_conv2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(193237440)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_59_cast = conv(bias = down_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = var_2677, groups = var_2634, pad = hidden_states_59_pad_0, pad_type = hidden_states_59_pad_type_0, strides = var_2675, weight = down_blocks_2_resnets_0_conv2_weight_to_fp16, x = input_131_cast)[name = tensor<string, []>("hidden_states_59_cast")];
            tensor<int32, [2]> var_2682 = const()[name = tensor<string, []>("op_2682"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2684 = const()[name = tensor<string, []>("op_2684"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_3_pad_type_0 = const()[name = tensor<string, []>("x_3_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_3_pad_0 = const()[name = tensor<string, []>("x_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 640, 1, 1]> down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [1280, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(193240064)))];
            tensor<fp16, [1280]> down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194878528)))];
            tensor<fp16, [2, 1280, 32, 32]> x_3_cast = conv(bias = down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_2684, groups = var_2634, pad = x_3_pad_0, pad_type = x_3_pad_type_0, strides = var_2682, weight = down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16, x = input_119_cast)[name = tensor<string, []>("x_3_cast")];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_61_cast = add(x = x_3_cast, y = hidden_states_59_cast)[name = tensor<string, []>("hidden_states_61_cast")];
            tensor<int32, [5]> reshape_48_shape_0 = const()[name = tensor<string, []>("reshape_48_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_48_cast = reshape(shape = reshape_48_shape_0, x = hidden_states_61_cast)[name = tensor<string, []>("reshape_48_cast")];
            tensor<int32, [3]> reduce_mean_36_axes_0 = const()[name = tensor<string, []>("reduce_mean_36_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_36_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_36_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_36_cast = reduce_mean(axes = reduce_mean_36_axes_0, keep_dims = reduce_mean_36_keep_dims_0, x = reshape_48_cast)[name = tensor<string, []>("reduce_mean_36_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_24_cast = sub(x = reshape_48_cast, y = reduce_mean_36_cast)[name = tensor<string, []>("sub_24_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_12_cast = square(x = sub_24_cast)[name = tensor<string, []>("square_12_cast")];
            tensor<int32, [3]> reduce_mean_38_axes_0 = const()[name = tensor<string, []>("reduce_mean_38_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_38_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_38_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_38_cast = reduce_mean(axes = reduce_mean_38_axes_0, keep_dims = reduce_mean_38_keep_dims_0, x = square_12_cast)[name = tensor<string, []>("reduce_mean_38_cast")];
            tensor<fp16, []> add_24_y_0_to_fp16 = const()[name = tensor<string, []>("add_24_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_24_cast = add(x = reduce_mean_38_cast, y = add_24_y_0_to_fp16)[name = tensor<string, []>("add_24_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_12_cast = sqrt(x = add_24_cast)[name = tensor<string, []>("sqrt_12_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_12_cast = real_div(x = sub_24_cast, y = sqrt_12_cast)[name = tensor<string, []>("real_div_12_cast")];
            tensor<int32, [4]> reshape_49_shape_0 = const()[name = tensor<string, []>("reshape_49_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_49_cast = reshape(shape = reshape_49_shape_0, x = real_div_12_cast)[name = tensor<string, []>("reshape_49_cast")];
            tensor<fp16, [1280]> add_25_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_25_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194881152)))];
            tensor<fp16, [1280]> add_25_beta_0_to_fp16 = const()[name = tensor<string, []>("add_25_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194883776)))];
            tensor<fp16, []> add_25_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_25_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 32, 32]> add_25_cast = batch_norm(beta = add_25_beta_0_to_fp16, epsilon = add_25_epsilon_0_to_fp16, gamma = add_25_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_49_cast)[name = tensor<string, []>("add_25_cast")];
            tensor<int32, [2]> var_2722 = const()[name = tensor<string, []>("op_2722"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2724 = const()[name = tensor<string, []>("op_2724"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_63_pad_type_0 = const()[name = tensor<string, []>("hidden_states_63_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_63_pad_0 = const()[name = tensor<string, []>("hidden_states_63_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_proj_in_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194886400)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_proj_in_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198163264)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_63_cast = conv(bias = down_blocks_2_attentions_0_proj_in_bias_to_fp16, dilations = var_2724, groups = var_2634, pad = hidden_states_63_pad_0, pad_type = hidden_states_63_pad_type_0, strides = var_2722, weight = down_blocks_2_attentions_0_proj_in_weight_to_fp16, x = add_25_cast)[name = tensor<string, []>("hidden_states_63_cast")];
            tensor<int32, [4]> var_2729 = const()[name = tensor<string, []>("op_2729"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_25_cast = reshape(shape = var_2729, x = hidden_states_63_cast)[name = tensor<string, []>("inputs_25_cast")];
            tensor<int32, [1]> var_2739 = const()[name = tensor<string, []>("op_2739"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_25_cast = reduce_mean(axes = var_2739, keep_dims = var_2629, x = inputs_25_cast)[name = tensor<string, []>("channels_mean_25_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_25_cast = sub(x = inputs_25_cast, y = channels_mean_25_cast)[name = tensor<string, []>("zero_mean_25_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_25_cast = mul(x = zero_mean_25_cast, y = zero_mean_25_cast)[name = tensor<string, []>("zero_mean_sq_25_cast")];
            tensor<int32, [1]> var_2743 = const()[name = tensor<string, []>("op_2743"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_2744_cast = reduce_mean(axes = var_2743, keep_dims = var_2629, x = zero_mean_sq_25_cast)[name = tensor<string, []>("op_2744_cast")];
            tensor<fp16, []> var_2745_to_fp16 = const()[name = tensor<string, []>("op_2745_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_2746_cast = add(x = var_2744_cast, y = var_2745_to_fp16)[name = tensor<string, []>("op_2746_cast")];
            tensor<fp16, []> denom_25_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_25_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_25_cast = rsqrt(epsilon = denom_25_epsilon_0_to_fp16, x = var_2746_cast)[name = tensor<string, []>("denom_25_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_25_cast = mul(x = zero_mean_25_cast, y = denom_25_cast)[name = tensor<string, []>("out_25_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_2750_to_fp16 = const()[name = tensor<string, []>("op_2750_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198165888)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_2751_cast = add(x = out_25_cast, y = var_2750_to_fp16)[name = tensor<string, []>("op_2751_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_2753_to_fp16 = const()[name = tensor<string, []>("op_2753_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198168512)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_65_cast = mul(x = var_2751_cast, y = var_2753_to_fp16)[name = tensor<string, []>("hidden_states_65_cast")];
            tensor<int32, [2]> var_2760 = const()[name = tensor<string, []>("op_2760"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2762 = const()[name = tensor<string, []>("op_2762"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_17_pad_type_0 = const()[name = tensor<string, []>("q_17_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_17_pad_0 = const()[name = tensor<string, []>("q_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198171136)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_17_cast = conv(dilations = var_2762, groups = var_2634, pad = q_17_pad_0, pad_type = q_17_pad_type_0, strides = var_2760, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_65_cast)[name = tensor<string, []>("q_17_cast")];
            tensor<int32, [2]> var_2766 = const()[name = tensor<string, []>("op_2766"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2768 = const()[name = tensor<string, []>("op_2768"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_33_pad_type_0 = const()[name = tensor<string, []>("k_33_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_33_pad_0 = const()[name = tensor<string, []>("k_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(201448000)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_33_cast = conv(dilations = var_2768, groups = var_2634, pad = k_33_pad_0, pad_type = k_33_pad_type_0, strides = var_2766, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_65_cast)[name = tensor<string, []>("k_33_cast")];
            tensor<int32, [2]> var_2772 = const()[name = tensor<string, []>("op_2772"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2774 = const()[name = tensor<string, []>("op_2774"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_17_pad_type_0 = const()[name = tensor<string, []>("v_17_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_17_pad_0 = const()[name = tensor<string, []>("v_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(204724864)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_17_cast = conv(dilations = var_2774, groups = var_2634, pad = v_17_pad_0, pad_type = v_17_pad_type_0, strides = var_2772, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_65_cast)[name = tensor<string, []>("v_17_cast")];
            tensor<int32, [4]> var_2778_begin_0 = const()[name = tensor<string, []>("op_2778_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2778_end_0 = const()[name = tensor<string, []>("op_2778_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2778_end_mask_0 = const()[name = tensor<string, []>("op_2778_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2778_cast = slice_by_index(begin = var_2778_begin_0, end = var_2778_end_0, end_mask = var_2778_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2778_cast")];
            tensor<int32, [4]> var_2782_begin_0 = const()[name = tensor<string, []>("op_2782_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_2782_end_0 = const()[name = tensor<string, []>("op_2782_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_2782_end_mask_0 = const()[name = tensor<string, []>("op_2782_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2782_cast = slice_by_index(begin = var_2782_begin_0, end = var_2782_end_0, end_mask = var_2782_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2782_cast")];
            tensor<int32, [4]> var_2786_begin_0 = const()[name = tensor<string, []>("op_2786_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_2786_end_0 = const()[name = tensor<string, []>("op_2786_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_2786_end_mask_0 = const()[name = tensor<string, []>("op_2786_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2786_cast = slice_by_index(begin = var_2786_begin_0, end = var_2786_end_0, end_mask = var_2786_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2786_cast")];
            tensor<int32, [4]> var_2790_begin_0 = const()[name = tensor<string, []>("op_2790_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_2790_end_0 = const()[name = tensor<string, []>("op_2790_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_2790_end_mask_0 = const()[name = tensor<string, []>("op_2790_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2790_cast = slice_by_index(begin = var_2790_begin_0, end = var_2790_end_0, end_mask = var_2790_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2790_cast")];
            tensor<int32, [4]> var_2794_begin_0 = const()[name = tensor<string, []>("op_2794_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_2794_end_0 = const()[name = tensor<string, []>("op_2794_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_2794_end_mask_0 = const()[name = tensor<string, []>("op_2794_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2794_cast = slice_by_index(begin = var_2794_begin_0, end = var_2794_end_0, end_mask = var_2794_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2794_cast")];
            tensor<int32, [4]> var_2798_begin_0 = const()[name = tensor<string, []>("op_2798_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_2798_end_0 = const()[name = tensor<string, []>("op_2798_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_2798_end_mask_0 = const()[name = tensor<string, []>("op_2798_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2798_cast = slice_by_index(begin = var_2798_begin_0, end = var_2798_end_0, end_mask = var_2798_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2798_cast")];
            tensor<int32, [4]> var_2802_begin_0 = const()[name = tensor<string, []>("op_2802_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_2802_end_0 = const()[name = tensor<string, []>("op_2802_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_2802_end_mask_0 = const()[name = tensor<string, []>("op_2802_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2802_cast = slice_by_index(begin = var_2802_begin_0, end = var_2802_end_0, end_mask = var_2802_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2802_cast")];
            tensor<int32, [4]> var_2806_begin_0 = const()[name = tensor<string, []>("op_2806_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_2806_end_0 = const()[name = tensor<string, []>("op_2806_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_2806_end_mask_0 = const()[name = tensor<string, []>("op_2806_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2806_cast = slice_by_index(begin = var_2806_begin_0, end = var_2806_end_0, end_mask = var_2806_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2806_cast")];
            tensor<int32, [4]> var_2810_begin_0 = const()[name = tensor<string, []>("op_2810_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_2810_end_0 = const()[name = tensor<string, []>("op_2810_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_2810_end_mask_0 = const()[name = tensor<string, []>("op_2810_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2810_cast = slice_by_index(begin = var_2810_begin_0, end = var_2810_end_0, end_mask = var_2810_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2810_cast")];
            tensor<int32, [4]> var_2814_begin_0 = const()[name = tensor<string, []>("op_2814_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_2814_end_0 = const()[name = tensor<string, []>("op_2814_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_2814_end_mask_0 = const()[name = tensor<string, []>("op_2814_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2814_cast = slice_by_index(begin = var_2814_begin_0, end = var_2814_end_0, end_mask = var_2814_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2814_cast")];
            tensor<int32, [4]> var_2818_begin_0 = const()[name = tensor<string, []>("op_2818_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_2818_end_0 = const()[name = tensor<string, []>("op_2818_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_2818_end_mask_0 = const()[name = tensor<string, []>("op_2818_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2818_cast = slice_by_index(begin = var_2818_begin_0, end = var_2818_end_0, end_mask = var_2818_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2818_cast")];
            tensor<int32, [4]> var_2822_begin_0 = const()[name = tensor<string, []>("op_2822_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_2822_end_0 = const()[name = tensor<string, []>("op_2822_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_2822_end_mask_0 = const()[name = tensor<string, []>("op_2822_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2822_cast = slice_by_index(begin = var_2822_begin_0, end = var_2822_end_0, end_mask = var_2822_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2822_cast")];
            tensor<int32, [4]> var_2826_begin_0 = const()[name = tensor<string, []>("op_2826_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_2826_end_0 = const()[name = tensor<string, []>("op_2826_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_2826_end_mask_0 = const()[name = tensor<string, []>("op_2826_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2826_cast = slice_by_index(begin = var_2826_begin_0, end = var_2826_end_0, end_mask = var_2826_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2826_cast")];
            tensor<int32, [4]> var_2830_begin_0 = const()[name = tensor<string, []>("op_2830_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_2830_end_0 = const()[name = tensor<string, []>("op_2830_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_2830_end_mask_0 = const()[name = tensor<string, []>("op_2830_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2830_cast = slice_by_index(begin = var_2830_begin_0, end = var_2830_end_0, end_mask = var_2830_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2830_cast")];
            tensor<int32, [4]> var_2834_begin_0 = const()[name = tensor<string, []>("op_2834_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_2834_end_0 = const()[name = tensor<string, []>("op_2834_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_2834_end_mask_0 = const()[name = tensor<string, []>("op_2834_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2834_cast = slice_by_index(begin = var_2834_begin_0, end = var_2834_end_0, end_mask = var_2834_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2834_cast")];
            tensor<int32, [4]> var_2838_begin_0 = const()[name = tensor<string, []>("op_2838_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_2838_end_0 = const()[name = tensor<string, []>("op_2838_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_2838_end_mask_0 = const()[name = tensor<string, []>("op_2838_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2838_cast = slice_by_index(begin = var_2838_begin_0, end = var_2838_end_0, end_mask = var_2838_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2838_cast")];
            tensor<int32, [4]> var_2842_begin_0 = const()[name = tensor<string, []>("op_2842_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_2842_end_0 = const()[name = tensor<string, []>("op_2842_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_2842_end_mask_0 = const()[name = tensor<string, []>("op_2842_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2842_cast = slice_by_index(begin = var_2842_begin_0, end = var_2842_end_0, end_mask = var_2842_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2842_cast")];
            tensor<int32, [4]> var_2846_begin_0 = const()[name = tensor<string, []>("op_2846_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_2846_end_0 = const()[name = tensor<string, []>("op_2846_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_2846_end_mask_0 = const()[name = tensor<string, []>("op_2846_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2846_cast = slice_by_index(begin = var_2846_begin_0, end = var_2846_end_0, end_mask = var_2846_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2846_cast")];
            tensor<int32, [4]> var_2850_begin_0 = const()[name = tensor<string, []>("op_2850_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_2850_end_0 = const()[name = tensor<string, []>("op_2850_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_2850_end_mask_0 = const()[name = tensor<string, []>("op_2850_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2850_cast = slice_by_index(begin = var_2850_begin_0, end = var_2850_end_0, end_mask = var_2850_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2850_cast")];
            tensor<int32, [4]> var_2854_begin_0 = const()[name = tensor<string, []>("op_2854_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_2854_end_0 = const()[name = tensor<string, []>("op_2854_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_2854_end_mask_0 = const()[name = tensor<string, []>("op_2854_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2854_cast = slice_by_index(begin = var_2854_begin_0, end = var_2854_end_0, end_mask = var_2854_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_2854_cast")];
            tensor<int32, [4]> k_35_perm_0 = const()[name = tensor<string, []>("k_35_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_2861_begin_0 = const()[name = tensor<string, []>("op_2861_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2861_end_0 = const()[name = tensor<string, []>("op_2861_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_2861_end_mask_0 = const()[name = tensor<string, []>("op_2861_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_131 = transpose(perm = k_35_perm_0, x = k_33_cast)[name = tensor<string, []>("transpose_131")];
            tensor<fp16, [2, 1024, 1, 64]> var_2861_cast = slice_by_index(begin = var_2861_begin_0, end = var_2861_end_0, end_mask = var_2861_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2861_cast")];
            tensor<int32, [4]> var_2865_begin_0 = const()[name = tensor<string, []>("op_2865_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_2865_end_0 = const()[name = tensor<string, []>("op_2865_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_2865_end_mask_0 = const()[name = tensor<string, []>("op_2865_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2865_cast = slice_by_index(begin = var_2865_begin_0, end = var_2865_end_0, end_mask = var_2865_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2865_cast")];
            tensor<int32, [4]> var_2869_begin_0 = const()[name = tensor<string, []>("op_2869_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_2869_end_0 = const()[name = tensor<string, []>("op_2869_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_2869_end_mask_0 = const()[name = tensor<string, []>("op_2869_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2869_cast = slice_by_index(begin = var_2869_begin_0, end = var_2869_end_0, end_mask = var_2869_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2869_cast")];
            tensor<int32, [4]> var_2873_begin_0 = const()[name = tensor<string, []>("op_2873_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_2873_end_0 = const()[name = tensor<string, []>("op_2873_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_2873_end_mask_0 = const()[name = tensor<string, []>("op_2873_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2873_cast = slice_by_index(begin = var_2873_begin_0, end = var_2873_end_0, end_mask = var_2873_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2873_cast")];
            tensor<int32, [4]> var_2877_begin_0 = const()[name = tensor<string, []>("op_2877_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_2877_end_0 = const()[name = tensor<string, []>("op_2877_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_2877_end_mask_0 = const()[name = tensor<string, []>("op_2877_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2877_cast = slice_by_index(begin = var_2877_begin_0, end = var_2877_end_0, end_mask = var_2877_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2877_cast")];
            tensor<int32, [4]> var_2881_begin_0 = const()[name = tensor<string, []>("op_2881_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_2881_end_0 = const()[name = tensor<string, []>("op_2881_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_2881_end_mask_0 = const()[name = tensor<string, []>("op_2881_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2881_cast = slice_by_index(begin = var_2881_begin_0, end = var_2881_end_0, end_mask = var_2881_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2881_cast")];
            tensor<int32, [4]> var_2885_begin_0 = const()[name = tensor<string, []>("op_2885_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_2885_end_0 = const()[name = tensor<string, []>("op_2885_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_2885_end_mask_0 = const()[name = tensor<string, []>("op_2885_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2885_cast = slice_by_index(begin = var_2885_begin_0, end = var_2885_end_0, end_mask = var_2885_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2885_cast")];
            tensor<int32, [4]> var_2889_begin_0 = const()[name = tensor<string, []>("op_2889_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_2889_end_0 = const()[name = tensor<string, []>("op_2889_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_2889_end_mask_0 = const()[name = tensor<string, []>("op_2889_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2889_cast = slice_by_index(begin = var_2889_begin_0, end = var_2889_end_0, end_mask = var_2889_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2889_cast")];
            tensor<int32, [4]> var_2893_begin_0 = const()[name = tensor<string, []>("op_2893_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2893_end_0 = const()[name = tensor<string, []>("op_2893_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_2893_end_mask_0 = const()[name = tensor<string, []>("op_2893_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2893_cast = slice_by_index(begin = var_2893_begin_0, end = var_2893_end_0, end_mask = var_2893_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2893_cast")];
            tensor<int32, [4]> var_2897_begin_0 = const()[name = tensor<string, []>("op_2897_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_2897_end_0 = const()[name = tensor<string, []>("op_2897_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_2897_end_mask_0 = const()[name = tensor<string, []>("op_2897_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2897_cast = slice_by_index(begin = var_2897_begin_0, end = var_2897_end_0, end_mask = var_2897_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2897_cast")];
            tensor<int32, [4]> var_2901_begin_0 = const()[name = tensor<string, []>("op_2901_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_2901_end_0 = const()[name = tensor<string, []>("op_2901_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_2901_end_mask_0 = const()[name = tensor<string, []>("op_2901_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2901_cast = slice_by_index(begin = var_2901_begin_0, end = var_2901_end_0, end_mask = var_2901_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2901_cast")];
            tensor<int32, [4]> var_2905_begin_0 = const()[name = tensor<string, []>("op_2905_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_2905_end_0 = const()[name = tensor<string, []>("op_2905_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_2905_end_mask_0 = const()[name = tensor<string, []>("op_2905_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2905_cast = slice_by_index(begin = var_2905_begin_0, end = var_2905_end_0, end_mask = var_2905_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2905_cast")];
            tensor<int32, [4]> var_2909_begin_0 = const()[name = tensor<string, []>("op_2909_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_2909_end_0 = const()[name = tensor<string, []>("op_2909_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_2909_end_mask_0 = const()[name = tensor<string, []>("op_2909_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2909_cast = slice_by_index(begin = var_2909_begin_0, end = var_2909_end_0, end_mask = var_2909_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2909_cast")];
            tensor<int32, [4]> var_2913_begin_0 = const()[name = tensor<string, []>("op_2913_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_2913_end_0 = const()[name = tensor<string, []>("op_2913_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_2913_end_mask_0 = const()[name = tensor<string, []>("op_2913_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2913_cast = slice_by_index(begin = var_2913_begin_0, end = var_2913_end_0, end_mask = var_2913_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2913_cast")];
            tensor<int32, [4]> var_2917_begin_0 = const()[name = tensor<string, []>("op_2917_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_2917_end_0 = const()[name = tensor<string, []>("op_2917_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_2917_end_mask_0 = const()[name = tensor<string, []>("op_2917_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2917_cast = slice_by_index(begin = var_2917_begin_0, end = var_2917_end_0, end_mask = var_2917_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2917_cast")];
            tensor<int32, [4]> var_2921_begin_0 = const()[name = tensor<string, []>("op_2921_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_2921_end_0 = const()[name = tensor<string, []>("op_2921_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_2921_end_mask_0 = const()[name = tensor<string, []>("op_2921_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2921_cast = slice_by_index(begin = var_2921_begin_0, end = var_2921_end_0, end_mask = var_2921_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2921_cast")];
            tensor<int32, [4]> var_2925_begin_0 = const()[name = tensor<string, []>("op_2925_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_2925_end_0 = const()[name = tensor<string, []>("op_2925_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_2925_end_mask_0 = const()[name = tensor<string, []>("op_2925_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2925_cast = slice_by_index(begin = var_2925_begin_0, end = var_2925_end_0, end_mask = var_2925_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2925_cast")];
            tensor<int32, [4]> var_2929_begin_0 = const()[name = tensor<string, []>("op_2929_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_2929_end_0 = const()[name = tensor<string, []>("op_2929_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_2929_end_mask_0 = const()[name = tensor<string, []>("op_2929_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2929_cast = slice_by_index(begin = var_2929_begin_0, end = var_2929_end_0, end_mask = var_2929_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2929_cast")];
            tensor<int32, [4]> var_2933_begin_0 = const()[name = tensor<string, []>("op_2933_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_2933_end_0 = const()[name = tensor<string, []>("op_2933_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_2933_end_mask_0 = const()[name = tensor<string, []>("op_2933_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2933_cast = slice_by_index(begin = var_2933_begin_0, end = var_2933_end_0, end_mask = var_2933_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2933_cast")];
            tensor<int32, [4]> var_2937_begin_0 = const()[name = tensor<string, []>("op_2937_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_2937_end_0 = const()[name = tensor<string, []>("op_2937_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_2937_end_mask_0 = const()[name = tensor<string, []>("op_2937_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2937_cast = slice_by_index(begin = var_2937_begin_0, end = var_2937_end_0, end_mask = var_2937_end_mask_0, x = transpose_131)[name = tensor<string, []>("op_2937_cast")];
            tensor<int32, [4]> var_2939_begin_0 = const()[name = tensor<string, []>("op_2939_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2939_end_0 = const()[name = tensor<string, []>("op_2939_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2939_end_mask_0 = const()[name = tensor<string, []>("op_2939_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2939_cast = slice_by_index(begin = var_2939_begin_0, end = var_2939_end_0, end_mask = var_2939_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_2939_cast")];
            tensor<int32, [4]> var_2943_begin_0 = const()[name = tensor<string, []>("op_2943_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_2943_end_0 = const()[name = tensor<string, []>("op_2943_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_2943_end_mask_0 = const()[name = tensor<string, []>("op_2943_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2943_cast = slice_by_index(begin = var_2943_begin_0, end = var_2943_end_0, end_mask = var_2943_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_2943_cast")];
            tensor<int32, [4]> var_2947_begin_0 = const()[name = tensor<string, []>("op_2947_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_2947_end_0 = const()[name = tensor<string, []>("op_2947_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_2947_end_mask_0 = const()[name = tensor<string, []>("op_2947_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2947_cast = slice_by_index(begin = var_2947_begin_0, end = var_2947_end_0, end_mask = var_2947_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_2947_cast")];
            tensor<int32, [4]> var_2951_begin_0 = const()[name = tensor<string, []>("op_2951_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_2951_end_0 = const()[name = tensor<string, []>("op_2951_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_2951_end_mask_0 = const()[name = tensor<string, []>("op_2951_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2951_cast = slice_by_index(begin = var_2951_begin_0, end = var_2951_end_0, end_mask = var_2951_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_2951_cast")];
            tensor<int32, [4]> var_2955_begin_0 = const()[name = tensor<string, []>("op_2955_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_2955_end_0 = const()[name = tensor<string, []>("op_2955_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_2955_end_mask_0 = const()[name = tensor<string, []>("op_2955_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2955_cast = slice_by_index(begin = var_2955_begin_0, end = var_2955_end_0, end_mask = var_2955_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_2955_cast")];
            tensor<int32, [4]> var_2959_begin_0 = const()[name = tensor<string, []>("op_2959_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_2959_end_0 = const()[name = tensor<string, []>("op_2959_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_2959_end_mask_0 = const()[name = tensor<string, []>("op_2959_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2959_cast = slice_by_index(begin = var_2959_begin_0, end = var_2959_end_0, end_mask = var_2959_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_2959_cast")];
            tensor<int32, [4]> var_2963_begin_0 = const()[name = tensor<string, []>("op_2963_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_2963_end_0 = const()[name = tensor<string, []>("op_2963_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_2963_end_mask_0 = const()[name = tensor<string, []>("op_2963_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2963_cast = slice_by_index(begin = var_2963_begin_0, end = var_2963_end_0, end_mask = var_2963_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_2963_cast")];
            tensor<int32, [4]> var_2967_begin_0 = const()[name = tensor<string, []>("op_2967_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_2967_end_0 = const()[name = tensor<string, []>("op_2967_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_2967_end_mask_0 = const()[name = tensor<string, []>("op_2967_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2967_cast = slice_by_index(begin = var_2967_begin_0, end = var_2967_end_0, end_mask = var_2967_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_2967_cast")];
            tensor<int32, [4]> var_2971_begin_0 = const()[name = tensor<string, []>("op_2971_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_2971_end_0 = const()[name = tensor<string, []>("op_2971_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_2971_end_mask_0 = const()[name = tensor<string, []>("op_2971_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2971_cast = slice_by_index(begin = var_2971_begin_0, end = var_2971_end_0, end_mask = var_2971_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_2971_cast")];
            tensor<int32, [4]> var_2975_begin_0 = const()[name = tensor<string, []>("op_2975_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_2975_end_0 = const()[name = tensor<string, []>("op_2975_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_2975_end_mask_0 = const()[name = tensor<string, []>("op_2975_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2975_cast = slice_by_index(begin = var_2975_begin_0, end = var_2975_end_0, end_mask = var_2975_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_2975_cast")];
            tensor<int32, [4]> var_2979_begin_0 = const()[name = tensor<string, []>("op_2979_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_2979_end_0 = const()[name = tensor<string, []>("op_2979_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_2979_end_mask_0 = const()[name = tensor<string, []>("op_2979_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2979_cast = slice_by_index(begin = var_2979_begin_0, end = var_2979_end_0, end_mask = var_2979_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_2979_cast")];
            tensor<int32, [4]> var_2983_begin_0 = const()[name = tensor<string, []>("op_2983_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_2983_end_0 = const()[name = tensor<string, []>("op_2983_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_2983_end_mask_0 = const()[name = tensor<string, []>("op_2983_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2983_cast = slice_by_index(begin = var_2983_begin_0, end = var_2983_end_0, end_mask = var_2983_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_2983_cast")];
            tensor<int32, [4]> var_2987_begin_0 = const()[name = tensor<string, []>("op_2987_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_2987_end_0 = const()[name = tensor<string, []>("op_2987_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_2987_end_mask_0 = const()[name = tensor<string, []>("op_2987_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2987_cast = slice_by_index(begin = var_2987_begin_0, end = var_2987_end_0, end_mask = var_2987_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_2987_cast")];
            tensor<int32, [4]> var_2991_begin_0 = const()[name = tensor<string, []>("op_2991_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_2991_end_0 = const()[name = tensor<string, []>("op_2991_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_2991_end_mask_0 = const()[name = tensor<string, []>("op_2991_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2991_cast = slice_by_index(begin = var_2991_begin_0, end = var_2991_end_0, end_mask = var_2991_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_2991_cast")];
            tensor<int32, [4]> var_2995_begin_0 = const()[name = tensor<string, []>("op_2995_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_2995_end_0 = const()[name = tensor<string, []>("op_2995_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_2995_end_mask_0 = const()[name = tensor<string, []>("op_2995_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2995_cast = slice_by_index(begin = var_2995_begin_0, end = var_2995_end_0, end_mask = var_2995_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_2995_cast")];
            tensor<int32, [4]> var_2999_begin_0 = const()[name = tensor<string, []>("op_2999_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_2999_end_0 = const()[name = tensor<string, []>("op_2999_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_2999_end_mask_0 = const()[name = tensor<string, []>("op_2999_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2999_cast = slice_by_index(begin = var_2999_begin_0, end = var_2999_end_0, end_mask = var_2999_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_2999_cast")];
            tensor<int32, [4]> var_3003_begin_0 = const()[name = tensor<string, []>("op_3003_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_3003_end_0 = const()[name = tensor<string, []>("op_3003_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_3003_end_mask_0 = const()[name = tensor<string, []>("op_3003_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3003_cast = slice_by_index(begin = var_3003_begin_0, end = var_3003_end_0, end_mask = var_3003_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_3003_cast")];
            tensor<int32, [4]> var_3007_begin_0 = const()[name = tensor<string, []>("op_3007_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_3007_end_0 = const()[name = tensor<string, []>("op_3007_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_3007_end_mask_0 = const()[name = tensor<string, []>("op_3007_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3007_cast = slice_by_index(begin = var_3007_begin_0, end = var_3007_end_0, end_mask = var_3007_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_3007_cast")];
            tensor<int32, [4]> var_3011_begin_0 = const()[name = tensor<string, []>("op_3011_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_3011_end_0 = const()[name = tensor<string, []>("op_3011_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_3011_end_mask_0 = const()[name = tensor<string, []>("op_3011_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3011_cast = slice_by_index(begin = var_3011_begin_0, end = var_3011_end_0, end_mask = var_3011_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_3011_cast")];
            tensor<int32, [4]> var_3015_begin_0 = const()[name = tensor<string, []>("op_3015_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_3015_end_0 = const()[name = tensor<string, []>("op_3015_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_3015_end_mask_0 = const()[name = tensor<string, []>("op_3015_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3015_cast = slice_by_index(begin = var_3015_begin_0, end = var_3015_end_0, end_mask = var_3015_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_3015_cast")];
            tensor<string, []> var_3019_equation_0 = const()[name = tensor<string, []>("op_3019_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3019_cast = einsum(equation = var_3019_equation_0, values = (var_2861_cast, var_2778_cast))[name = tensor<string, []>("op_3019_cast")];
            tensor<fp16, []> var_3020_to_fp16 = const()[name = tensor<string, []>("op_3020_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_161_cast = mul(x = var_3019_cast, y = var_3020_to_fp16)[name = tensor<string, []>("aw_161_cast")];
            tensor<string, []> var_3023_equation_0 = const()[name = tensor<string, []>("op_3023_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3023_cast = einsum(equation = var_3023_equation_0, values = (var_2865_cast, var_2782_cast))[name = tensor<string, []>("op_3023_cast")];
            tensor<fp16, []> var_3024_to_fp16 = const()[name = tensor<string, []>("op_3024_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_163_cast = mul(x = var_3023_cast, y = var_3024_to_fp16)[name = tensor<string, []>("aw_163_cast")];
            tensor<string, []> var_3027_equation_0 = const()[name = tensor<string, []>("op_3027_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3027_cast = einsum(equation = var_3027_equation_0, values = (var_2869_cast, var_2786_cast))[name = tensor<string, []>("op_3027_cast")];
            tensor<fp16, []> var_3028_to_fp16 = const()[name = tensor<string, []>("op_3028_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_165_cast = mul(x = var_3027_cast, y = var_3028_to_fp16)[name = tensor<string, []>("aw_165_cast")];
            tensor<string, []> var_3031_equation_0 = const()[name = tensor<string, []>("op_3031_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3031_cast = einsum(equation = var_3031_equation_0, values = (var_2873_cast, var_2790_cast))[name = tensor<string, []>("op_3031_cast")];
            tensor<fp16, []> var_3032_to_fp16 = const()[name = tensor<string, []>("op_3032_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_167_cast = mul(x = var_3031_cast, y = var_3032_to_fp16)[name = tensor<string, []>("aw_167_cast")];
            tensor<string, []> var_3035_equation_0 = const()[name = tensor<string, []>("op_3035_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3035_cast = einsum(equation = var_3035_equation_0, values = (var_2877_cast, var_2794_cast))[name = tensor<string, []>("op_3035_cast")];
            tensor<fp16, []> var_3036_to_fp16 = const()[name = tensor<string, []>("op_3036_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_169_cast = mul(x = var_3035_cast, y = var_3036_to_fp16)[name = tensor<string, []>("aw_169_cast")];
            tensor<string, []> var_3039_equation_0 = const()[name = tensor<string, []>("op_3039_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3039_cast = einsum(equation = var_3039_equation_0, values = (var_2881_cast, var_2798_cast))[name = tensor<string, []>("op_3039_cast")];
            tensor<fp16, []> var_3040_to_fp16 = const()[name = tensor<string, []>("op_3040_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_171_cast = mul(x = var_3039_cast, y = var_3040_to_fp16)[name = tensor<string, []>("aw_171_cast")];
            tensor<string, []> var_3043_equation_0 = const()[name = tensor<string, []>("op_3043_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3043_cast = einsum(equation = var_3043_equation_0, values = (var_2885_cast, var_2802_cast))[name = tensor<string, []>("op_3043_cast")];
            tensor<fp16, []> var_3044_to_fp16 = const()[name = tensor<string, []>("op_3044_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_173_cast = mul(x = var_3043_cast, y = var_3044_to_fp16)[name = tensor<string, []>("aw_173_cast")];
            tensor<string, []> var_3047_equation_0 = const()[name = tensor<string, []>("op_3047_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3047_cast = einsum(equation = var_3047_equation_0, values = (var_2889_cast, var_2806_cast))[name = tensor<string, []>("op_3047_cast")];
            tensor<fp16, []> var_3048_to_fp16 = const()[name = tensor<string, []>("op_3048_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_175_cast = mul(x = var_3047_cast, y = var_3048_to_fp16)[name = tensor<string, []>("aw_175_cast")];
            tensor<string, []> var_3051_equation_0 = const()[name = tensor<string, []>("op_3051_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3051_cast = einsum(equation = var_3051_equation_0, values = (var_2893_cast, var_2810_cast))[name = tensor<string, []>("op_3051_cast")];
            tensor<fp16, []> var_3052_to_fp16 = const()[name = tensor<string, []>("op_3052_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_177_cast = mul(x = var_3051_cast, y = var_3052_to_fp16)[name = tensor<string, []>("aw_177_cast")];
            tensor<string, []> var_3055_equation_0 = const()[name = tensor<string, []>("op_3055_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3055_cast = einsum(equation = var_3055_equation_0, values = (var_2897_cast, var_2814_cast))[name = tensor<string, []>("op_3055_cast")];
            tensor<fp16, []> var_3056_to_fp16 = const()[name = tensor<string, []>("op_3056_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_179_cast = mul(x = var_3055_cast, y = var_3056_to_fp16)[name = tensor<string, []>("aw_179_cast")];
            tensor<string, []> var_3059_equation_0 = const()[name = tensor<string, []>("op_3059_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3059_cast = einsum(equation = var_3059_equation_0, values = (var_2901_cast, var_2818_cast))[name = tensor<string, []>("op_3059_cast")];
            tensor<fp16, []> var_3060_to_fp16 = const()[name = tensor<string, []>("op_3060_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_181_cast = mul(x = var_3059_cast, y = var_3060_to_fp16)[name = tensor<string, []>("aw_181_cast")];
            tensor<string, []> var_3063_equation_0 = const()[name = tensor<string, []>("op_3063_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3063_cast = einsum(equation = var_3063_equation_0, values = (var_2905_cast, var_2822_cast))[name = tensor<string, []>("op_3063_cast")];
            tensor<fp16, []> var_3064_to_fp16 = const()[name = tensor<string, []>("op_3064_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_183_cast = mul(x = var_3063_cast, y = var_3064_to_fp16)[name = tensor<string, []>("aw_183_cast")];
            tensor<string, []> var_3067_equation_0 = const()[name = tensor<string, []>("op_3067_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3067_cast = einsum(equation = var_3067_equation_0, values = (var_2909_cast, var_2826_cast))[name = tensor<string, []>("op_3067_cast")];
            tensor<fp16, []> var_3068_to_fp16 = const()[name = tensor<string, []>("op_3068_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_185_cast = mul(x = var_3067_cast, y = var_3068_to_fp16)[name = tensor<string, []>("aw_185_cast")];
            tensor<string, []> var_3071_equation_0 = const()[name = tensor<string, []>("op_3071_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3071_cast = einsum(equation = var_3071_equation_0, values = (var_2913_cast, var_2830_cast))[name = tensor<string, []>("op_3071_cast")];
            tensor<fp16, []> var_3072_to_fp16 = const()[name = tensor<string, []>("op_3072_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_187_cast = mul(x = var_3071_cast, y = var_3072_to_fp16)[name = tensor<string, []>("aw_187_cast")];
            tensor<string, []> var_3075_equation_0 = const()[name = tensor<string, []>("op_3075_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3075_cast = einsum(equation = var_3075_equation_0, values = (var_2917_cast, var_2834_cast))[name = tensor<string, []>("op_3075_cast")];
            tensor<fp16, []> var_3076_to_fp16 = const()[name = tensor<string, []>("op_3076_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_189_cast = mul(x = var_3075_cast, y = var_3076_to_fp16)[name = tensor<string, []>("aw_189_cast")];
            tensor<string, []> var_3079_equation_0 = const()[name = tensor<string, []>("op_3079_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3079_cast = einsum(equation = var_3079_equation_0, values = (var_2921_cast, var_2838_cast))[name = tensor<string, []>("op_3079_cast")];
            tensor<fp16, []> var_3080_to_fp16 = const()[name = tensor<string, []>("op_3080_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_191_cast = mul(x = var_3079_cast, y = var_3080_to_fp16)[name = tensor<string, []>("aw_191_cast")];
            tensor<string, []> var_3083_equation_0 = const()[name = tensor<string, []>("op_3083_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3083_cast = einsum(equation = var_3083_equation_0, values = (var_2925_cast, var_2842_cast))[name = tensor<string, []>("op_3083_cast")];
            tensor<fp16, []> var_3084_to_fp16 = const()[name = tensor<string, []>("op_3084_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_193_cast = mul(x = var_3083_cast, y = var_3084_to_fp16)[name = tensor<string, []>("aw_193_cast")];
            tensor<string, []> var_3087_equation_0 = const()[name = tensor<string, []>("op_3087_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3087_cast = einsum(equation = var_3087_equation_0, values = (var_2929_cast, var_2846_cast))[name = tensor<string, []>("op_3087_cast")];
            tensor<fp16, []> var_3088_to_fp16 = const()[name = tensor<string, []>("op_3088_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_195_cast = mul(x = var_3087_cast, y = var_3088_to_fp16)[name = tensor<string, []>("aw_195_cast")];
            tensor<string, []> var_3091_equation_0 = const()[name = tensor<string, []>("op_3091_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3091_cast = einsum(equation = var_3091_equation_0, values = (var_2933_cast, var_2850_cast))[name = tensor<string, []>("op_3091_cast")];
            tensor<fp16, []> var_3092_to_fp16 = const()[name = tensor<string, []>("op_3092_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_197_cast = mul(x = var_3091_cast, y = var_3092_to_fp16)[name = tensor<string, []>("aw_197_cast")];
            tensor<string, []> var_3095_equation_0 = const()[name = tensor<string, []>("op_3095_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3095_cast = einsum(equation = var_3095_equation_0, values = (var_2937_cast, var_2854_cast))[name = tensor<string, []>("op_3095_cast")];
            tensor<fp16, []> var_3096_to_fp16 = const()[name = tensor<string, []>("op_3096_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_199_cast = mul(x = var_3095_cast, y = var_3096_to_fp16)[name = tensor<string, []>("aw_199_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3098_cast = softmax(axis = var_2634, x = aw_161_cast)[name = tensor<string, []>("op_3098_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3099_cast = softmax(axis = var_2634, x = aw_163_cast)[name = tensor<string, []>("op_3099_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3100_cast = softmax(axis = var_2634, x = aw_165_cast)[name = tensor<string, []>("op_3100_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3101_cast = softmax(axis = var_2634, x = aw_167_cast)[name = tensor<string, []>("op_3101_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3102_cast = softmax(axis = var_2634, x = aw_169_cast)[name = tensor<string, []>("op_3102_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3103_cast = softmax(axis = var_2634, x = aw_171_cast)[name = tensor<string, []>("op_3103_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3104_cast = softmax(axis = var_2634, x = aw_173_cast)[name = tensor<string, []>("op_3104_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3105_cast = softmax(axis = var_2634, x = aw_175_cast)[name = tensor<string, []>("op_3105_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3106_cast = softmax(axis = var_2634, x = aw_177_cast)[name = tensor<string, []>("op_3106_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3107_cast = softmax(axis = var_2634, x = aw_179_cast)[name = tensor<string, []>("op_3107_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3108_cast = softmax(axis = var_2634, x = aw_181_cast)[name = tensor<string, []>("op_3108_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3109_cast = softmax(axis = var_2634, x = aw_183_cast)[name = tensor<string, []>("op_3109_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3110_cast = softmax(axis = var_2634, x = aw_185_cast)[name = tensor<string, []>("op_3110_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3111_cast = softmax(axis = var_2634, x = aw_187_cast)[name = tensor<string, []>("op_3111_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3112_cast = softmax(axis = var_2634, x = aw_189_cast)[name = tensor<string, []>("op_3112_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3113_cast = softmax(axis = var_2634, x = aw_191_cast)[name = tensor<string, []>("op_3113_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3114_cast = softmax(axis = var_2634, x = aw_193_cast)[name = tensor<string, []>("op_3114_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3115_cast = softmax(axis = var_2634, x = aw_195_cast)[name = tensor<string, []>("op_3115_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3116_cast = softmax(axis = var_2634, x = aw_197_cast)[name = tensor<string, []>("op_3116_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3117_cast = softmax(axis = var_2634, x = aw_199_cast)[name = tensor<string, []>("op_3117_cast")];
            tensor<string, []> var_3119_equation_0 = const()[name = tensor<string, []>("op_3119_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3119_cast = einsum(equation = var_3119_equation_0, values = (var_2939_cast, var_3098_cast))[name = tensor<string, []>("op_3119_cast")];
            tensor<string, []> var_3121_equation_0 = const()[name = tensor<string, []>("op_3121_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3121_cast = einsum(equation = var_3121_equation_0, values = (var_2943_cast, var_3099_cast))[name = tensor<string, []>("op_3121_cast")];
            tensor<string, []> var_3123_equation_0 = const()[name = tensor<string, []>("op_3123_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3123_cast = einsum(equation = var_3123_equation_0, values = (var_2947_cast, var_3100_cast))[name = tensor<string, []>("op_3123_cast")];
            tensor<string, []> var_3125_equation_0 = const()[name = tensor<string, []>("op_3125_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3125_cast = einsum(equation = var_3125_equation_0, values = (var_2951_cast, var_3101_cast))[name = tensor<string, []>("op_3125_cast")];
            tensor<string, []> var_3127_equation_0 = const()[name = tensor<string, []>("op_3127_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3127_cast = einsum(equation = var_3127_equation_0, values = (var_2955_cast, var_3102_cast))[name = tensor<string, []>("op_3127_cast")];
            tensor<string, []> var_3129_equation_0 = const()[name = tensor<string, []>("op_3129_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3129_cast = einsum(equation = var_3129_equation_0, values = (var_2959_cast, var_3103_cast))[name = tensor<string, []>("op_3129_cast")];
            tensor<string, []> var_3131_equation_0 = const()[name = tensor<string, []>("op_3131_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3131_cast = einsum(equation = var_3131_equation_0, values = (var_2963_cast, var_3104_cast))[name = tensor<string, []>("op_3131_cast")];
            tensor<string, []> var_3133_equation_0 = const()[name = tensor<string, []>("op_3133_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3133_cast = einsum(equation = var_3133_equation_0, values = (var_2967_cast, var_3105_cast))[name = tensor<string, []>("op_3133_cast")];
            tensor<string, []> var_3135_equation_0 = const()[name = tensor<string, []>("op_3135_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3135_cast = einsum(equation = var_3135_equation_0, values = (var_2971_cast, var_3106_cast))[name = tensor<string, []>("op_3135_cast")];
            tensor<string, []> var_3137_equation_0 = const()[name = tensor<string, []>("op_3137_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3137_cast = einsum(equation = var_3137_equation_0, values = (var_2975_cast, var_3107_cast))[name = tensor<string, []>("op_3137_cast")];
            tensor<string, []> var_3139_equation_0 = const()[name = tensor<string, []>("op_3139_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3139_cast = einsum(equation = var_3139_equation_0, values = (var_2979_cast, var_3108_cast))[name = tensor<string, []>("op_3139_cast")];
            tensor<string, []> var_3141_equation_0 = const()[name = tensor<string, []>("op_3141_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3141_cast = einsum(equation = var_3141_equation_0, values = (var_2983_cast, var_3109_cast))[name = tensor<string, []>("op_3141_cast")];
            tensor<string, []> var_3143_equation_0 = const()[name = tensor<string, []>("op_3143_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3143_cast = einsum(equation = var_3143_equation_0, values = (var_2987_cast, var_3110_cast))[name = tensor<string, []>("op_3143_cast")];
            tensor<string, []> var_3145_equation_0 = const()[name = tensor<string, []>("op_3145_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3145_cast = einsum(equation = var_3145_equation_0, values = (var_2991_cast, var_3111_cast))[name = tensor<string, []>("op_3145_cast")];
            tensor<string, []> var_3147_equation_0 = const()[name = tensor<string, []>("op_3147_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3147_cast = einsum(equation = var_3147_equation_0, values = (var_2995_cast, var_3112_cast))[name = tensor<string, []>("op_3147_cast")];
            tensor<string, []> var_3149_equation_0 = const()[name = tensor<string, []>("op_3149_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3149_cast = einsum(equation = var_3149_equation_0, values = (var_2999_cast, var_3113_cast))[name = tensor<string, []>("op_3149_cast")];
            tensor<string, []> var_3151_equation_0 = const()[name = tensor<string, []>("op_3151_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3151_cast = einsum(equation = var_3151_equation_0, values = (var_3003_cast, var_3114_cast))[name = tensor<string, []>("op_3151_cast")];
            tensor<string, []> var_3153_equation_0 = const()[name = tensor<string, []>("op_3153_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3153_cast = einsum(equation = var_3153_equation_0, values = (var_3007_cast, var_3115_cast))[name = tensor<string, []>("op_3153_cast")];
            tensor<string, []> var_3155_equation_0 = const()[name = tensor<string, []>("op_3155_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3155_cast = einsum(equation = var_3155_equation_0, values = (var_3011_cast, var_3116_cast))[name = tensor<string, []>("op_3155_cast")];
            tensor<string, []> var_3157_equation_0 = const()[name = tensor<string, []>("op_3157_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3157_cast = einsum(equation = var_3157_equation_0, values = (var_3015_cast, var_3117_cast))[name = tensor<string, []>("op_3157_cast")];
            tensor<bool, []> input_135_interleave_0 = const()[name = tensor<string, []>("input_135_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_135_cast = concat(axis = var_2634, interleave = input_135_interleave_0, values = (var_3119_cast, var_3121_cast, var_3123_cast, var_3125_cast, var_3127_cast, var_3129_cast, var_3131_cast, var_3133_cast, var_3135_cast, var_3137_cast, var_3139_cast, var_3141_cast, var_3143_cast, var_3145_cast, var_3147_cast, var_3149_cast, var_3151_cast, var_3153_cast, var_3155_cast, var_3157_cast))[name = tensor<string, []>("input_135_cast")];
            tensor<int32, [2]> var_3163 = const()[name = tensor<string, []>("op_3163"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3165 = const()[name = tensor<string, []>("op_3165"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3167_pad_type_0 = const()[name = tensor<string, []>("op_3167_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3167_pad_0 = const()[name = tensor<string, []>("op_3167_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(208001728)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211278592)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_3167_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_3165, groups = var_2634, pad = var_3167_pad_0, pad_type = var_3167_pad_type_0, strides = var_3163, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_135_cast)[name = tensor<string, []>("op_3167_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_27_cast = add(x = var_3167_cast, y = inputs_25_cast)[name = tensor<string, []>("inputs_27_cast")];
            tensor<int32, [1]> var_3171 = const()[name = tensor<string, []>("op_3171"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_27_cast = reduce_mean(axes = var_3171, keep_dims = var_2629, x = inputs_27_cast)[name = tensor<string, []>("channels_mean_27_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_27_cast = sub(x = inputs_27_cast, y = channels_mean_27_cast)[name = tensor<string, []>("zero_mean_27_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_27_cast = mul(x = zero_mean_27_cast, y = zero_mean_27_cast)[name = tensor<string, []>("zero_mean_sq_27_cast")];
            tensor<int32, [1]> var_3175 = const()[name = tensor<string, []>("op_3175"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_3176_cast = reduce_mean(axes = var_3175, keep_dims = var_2629, x = zero_mean_sq_27_cast)[name = tensor<string, []>("op_3176_cast")];
            tensor<fp16, []> var_3177_to_fp16 = const()[name = tensor<string, []>("op_3177_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_3178_cast = add(x = var_3176_cast, y = var_3177_to_fp16)[name = tensor<string, []>("op_3178_cast")];
            tensor<fp16, []> denom_27_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_27_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_27_cast = rsqrt(epsilon = denom_27_epsilon_0_to_fp16, x = var_3178_cast)[name = tensor<string, []>("denom_27_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_27_cast = mul(x = zero_mean_27_cast, y = denom_27_cast)[name = tensor<string, []>("out_27_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_3182_to_fp16 = const()[name = tensor<string, []>("op_3182_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211281216)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_3183_cast = add(x = out_27_cast, y = var_3182_to_fp16)[name = tensor<string, []>("op_3183_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_3185_to_fp16 = const()[name = tensor<string, []>("op_3185_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211283840)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_67_cast = mul(x = var_3183_cast, y = var_3185_to_fp16)[name = tensor<string, []>("hidden_states_67_cast")];
            tensor<int32, [2]> var_3192 = const()[name = tensor<string, []>("op_3192"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3194 = const()[name = tensor<string, []>("op_3194"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_19_pad_type_0 = const()[name = tensor<string, []>("q_19_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_19_pad_0 = const()[name = tensor<string, []>("q_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211286464)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_19_cast = conv(dilations = var_3194, groups = var_2634, pad = q_19_pad_0, pad_type = q_19_pad_type_0, strides = var_3192, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_67_cast)[name = tensor<string, []>("q_19_cast")];
            tensor<int32, [2]> var_3198 = const()[name = tensor<string, []>("op_3198"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3200 = const()[name = tensor<string, []>("op_3200"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_37_pad_type_0 = const()[name = tensor<string, []>("k_37_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_37_pad_0 = const()[name = tensor<string, []>("k_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214563328)))];
            tensor<fp16, [2, 1280, 1, 77]> k_37_cast = conv(dilations = var_3200, groups = var_2634, pad = k_37_pad_0, pad_type = k_37_pad_type_0, strides = var_3198, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_37_cast")];
            tensor<int32, [2]> var_3204 = const()[name = tensor<string, []>("op_3204"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3206 = const()[name = tensor<string, []>("op_3206"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_19_pad_type_0 = const()[name = tensor<string, []>("v_19_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_19_pad_0 = const()[name = tensor<string, []>("v_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219806272)))];
            tensor<fp16, [2, 1280, 1, 77]> v_19_cast = conv(dilations = var_3206, groups = var_2634, pad = v_19_pad_0, pad_type = v_19_pad_type_0, strides = var_3204, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_19_cast")];
            tensor<int32, [4]> var_3210_begin_0 = const()[name = tensor<string, []>("op_3210_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3210_end_0 = const()[name = tensor<string, []>("op_3210_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3210_end_mask_0 = const()[name = tensor<string, []>("op_3210_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3210_cast = slice_by_index(begin = var_3210_begin_0, end = var_3210_end_0, end_mask = var_3210_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3210_cast")];
            tensor<int32, [4]> var_3214_begin_0 = const()[name = tensor<string, []>("op_3214_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_3214_end_0 = const()[name = tensor<string, []>("op_3214_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_3214_end_mask_0 = const()[name = tensor<string, []>("op_3214_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3214_cast = slice_by_index(begin = var_3214_begin_0, end = var_3214_end_0, end_mask = var_3214_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3214_cast")];
            tensor<int32, [4]> var_3218_begin_0 = const()[name = tensor<string, []>("op_3218_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_3218_end_0 = const()[name = tensor<string, []>("op_3218_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_3218_end_mask_0 = const()[name = tensor<string, []>("op_3218_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3218_cast = slice_by_index(begin = var_3218_begin_0, end = var_3218_end_0, end_mask = var_3218_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3218_cast")];
            tensor<int32, [4]> var_3222_begin_0 = const()[name = tensor<string, []>("op_3222_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_3222_end_0 = const()[name = tensor<string, []>("op_3222_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_3222_end_mask_0 = const()[name = tensor<string, []>("op_3222_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3222_cast = slice_by_index(begin = var_3222_begin_0, end = var_3222_end_0, end_mask = var_3222_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3222_cast")];
            tensor<int32, [4]> var_3226_begin_0 = const()[name = tensor<string, []>("op_3226_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_3226_end_0 = const()[name = tensor<string, []>("op_3226_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_3226_end_mask_0 = const()[name = tensor<string, []>("op_3226_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3226_cast = slice_by_index(begin = var_3226_begin_0, end = var_3226_end_0, end_mask = var_3226_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3226_cast")];
            tensor<int32, [4]> var_3230_begin_0 = const()[name = tensor<string, []>("op_3230_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_3230_end_0 = const()[name = tensor<string, []>("op_3230_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_3230_end_mask_0 = const()[name = tensor<string, []>("op_3230_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3230_cast = slice_by_index(begin = var_3230_begin_0, end = var_3230_end_0, end_mask = var_3230_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3230_cast")];
            tensor<int32, [4]> var_3234_begin_0 = const()[name = tensor<string, []>("op_3234_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_3234_end_0 = const()[name = tensor<string, []>("op_3234_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_3234_end_mask_0 = const()[name = tensor<string, []>("op_3234_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3234_cast = slice_by_index(begin = var_3234_begin_0, end = var_3234_end_0, end_mask = var_3234_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3234_cast")];
            tensor<int32, [4]> var_3238_begin_0 = const()[name = tensor<string, []>("op_3238_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_3238_end_0 = const()[name = tensor<string, []>("op_3238_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_3238_end_mask_0 = const()[name = tensor<string, []>("op_3238_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3238_cast = slice_by_index(begin = var_3238_begin_0, end = var_3238_end_0, end_mask = var_3238_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3238_cast")];
            tensor<int32, [4]> var_3242_begin_0 = const()[name = tensor<string, []>("op_3242_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_3242_end_0 = const()[name = tensor<string, []>("op_3242_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_3242_end_mask_0 = const()[name = tensor<string, []>("op_3242_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3242_cast = slice_by_index(begin = var_3242_begin_0, end = var_3242_end_0, end_mask = var_3242_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3242_cast")];
            tensor<int32, [4]> var_3246_begin_0 = const()[name = tensor<string, []>("op_3246_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_3246_end_0 = const()[name = tensor<string, []>("op_3246_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_3246_end_mask_0 = const()[name = tensor<string, []>("op_3246_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3246_cast = slice_by_index(begin = var_3246_begin_0, end = var_3246_end_0, end_mask = var_3246_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3246_cast")];
            tensor<int32, [4]> var_3250_begin_0 = const()[name = tensor<string, []>("op_3250_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_3250_end_0 = const()[name = tensor<string, []>("op_3250_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_3250_end_mask_0 = const()[name = tensor<string, []>("op_3250_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3250_cast = slice_by_index(begin = var_3250_begin_0, end = var_3250_end_0, end_mask = var_3250_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3250_cast")];
            tensor<int32, [4]> var_3254_begin_0 = const()[name = tensor<string, []>("op_3254_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_3254_end_0 = const()[name = tensor<string, []>("op_3254_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_3254_end_mask_0 = const()[name = tensor<string, []>("op_3254_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3254_cast = slice_by_index(begin = var_3254_begin_0, end = var_3254_end_0, end_mask = var_3254_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3254_cast")];
            tensor<int32, [4]> var_3258_begin_0 = const()[name = tensor<string, []>("op_3258_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_3258_end_0 = const()[name = tensor<string, []>("op_3258_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_3258_end_mask_0 = const()[name = tensor<string, []>("op_3258_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3258_cast = slice_by_index(begin = var_3258_begin_0, end = var_3258_end_0, end_mask = var_3258_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3258_cast")];
            tensor<int32, [4]> var_3262_begin_0 = const()[name = tensor<string, []>("op_3262_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_3262_end_0 = const()[name = tensor<string, []>("op_3262_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_3262_end_mask_0 = const()[name = tensor<string, []>("op_3262_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3262_cast = slice_by_index(begin = var_3262_begin_0, end = var_3262_end_0, end_mask = var_3262_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3262_cast")];
            tensor<int32, [4]> var_3266_begin_0 = const()[name = tensor<string, []>("op_3266_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_3266_end_0 = const()[name = tensor<string, []>("op_3266_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_3266_end_mask_0 = const()[name = tensor<string, []>("op_3266_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3266_cast = slice_by_index(begin = var_3266_begin_0, end = var_3266_end_0, end_mask = var_3266_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3266_cast")];
            tensor<int32, [4]> var_3270_begin_0 = const()[name = tensor<string, []>("op_3270_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_3270_end_0 = const()[name = tensor<string, []>("op_3270_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_3270_end_mask_0 = const()[name = tensor<string, []>("op_3270_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3270_cast = slice_by_index(begin = var_3270_begin_0, end = var_3270_end_0, end_mask = var_3270_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3270_cast")];
            tensor<int32, [4]> var_3274_begin_0 = const()[name = tensor<string, []>("op_3274_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_3274_end_0 = const()[name = tensor<string, []>("op_3274_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_3274_end_mask_0 = const()[name = tensor<string, []>("op_3274_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3274_cast = slice_by_index(begin = var_3274_begin_0, end = var_3274_end_0, end_mask = var_3274_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3274_cast")];
            tensor<int32, [4]> var_3278_begin_0 = const()[name = tensor<string, []>("op_3278_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_3278_end_0 = const()[name = tensor<string, []>("op_3278_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_3278_end_mask_0 = const()[name = tensor<string, []>("op_3278_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3278_cast = slice_by_index(begin = var_3278_begin_0, end = var_3278_end_0, end_mask = var_3278_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3278_cast")];
            tensor<int32, [4]> var_3282_begin_0 = const()[name = tensor<string, []>("op_3282_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_3282_end_0 = const()[name = tensor<string, []>("op_3282_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_3282_end_mask_0 = const()[name = tensor<string, []>("op_3282_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3282_cast = slice_by_index(begin = var_3282_begin_0, end = var_3282_end_0, end_mask = var_3282_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3282_cast")];
            tensor<int32, [4]> var_3286_begin_0 = const()[name = tensor<string, []>("op_3286_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_3286_end_0 = const()[name = tensor<string, []>("op_3286_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_3286_end_mask_0 = const()[name = tensor<string, []>("op_3286_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3286_cast = slice_by_index(begin = var_3286_begin_0, end = var_3286_end_0, end_mask = var_3286_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_3286_cast")];
            tensor<int32, [4]> k_39_perm_0 = const()[name = tensor<string, []>("k_39_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_3293_begin_0 = const()[name = tensor<string, []>("op_3293_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3293_end_0 = const()[name = tensor<string, []>("op_3293_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_3293_end_mask_0 = const()[name = tensor<string, []>("op_3293_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_130 = transpose(perm = k_39_perm_0, x = k_37_cast)[name = tensor<string, []>("transpose_130")];
            tensor<fp16, [2, 77, 1, 64]> var_3293_cast = slice_by_index(begin = var_3293_begin_0, end = var_3293_end_0, end_mask = var_3293_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3293_cast")];
            tensor<int32, [4]> var_3297_begin_0 = const()[name = tensor<string, []>("op_3297_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_3297_end_0 = const()[name = tensor<string, []>("op_3297_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_3297_end_mask_0 = const()[name = tensor<string, []>("op_3297_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3297_cast = slice_by_index(begin = var_3297_begin_0, end = var_3297_end_0, end_mask = var_3297_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3297_cast")];
            tensor<int32, [4]> var_3301_begin_0 = const()[name = tensor<string, []>("op_3301_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_3301_end_0 = const()[name = tensor<string, []>("op_3301_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_3301_end_mask_0 = const()[name = tensor<string, []>("op_3301_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3301_cast = slice_by_index(begin = var_3301_begin_0, end = var_3301_end_0, end_mask = var_3301_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3301_cast")];
            tensor<int32, [4]> var_3305_begin_0 = const()[name = tensor<string, []>("op_3305_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_3305_end_0 = const()[name = tensor<string, []>("op_3305_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_3305_end_mask_0 = const()[name = tensor<string, []>("op_3305_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3305_cast = slice_by_index(begin = var_3305_begin_0, end = var_3305_end_0, end_mask = var_3305_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3305_cast")];
            tensor<int32, [4]> var_3309_begin_0 = const()[name = tensor<string, []>("op_3309_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_3309_end_0 = const()[name = tensor<string, []>("op_3309_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_3309_end_mask_0 = const()[name = tensor<string, []>("op_3309_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3309_cast = slice_by_index(begin = var_3309_begin_0, end = var_3309_end_0, end_mask = var_3309_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3309_cast")];
            tensor<int32, [4]> var_3313_begin_0 = const()[name = tensor<string, []>("op_3313_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_3313_end_0 = const()[name = tensor<string, []>("op_3313_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_3313_end_mask_0 = const()[name = tensor<string, []>("op_3313_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3313_cast = slice_by_index(begin = var_3313_begin_0, end = var_3313_end_0, end_mask = var_3313_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3313_cast")];
            tensor<int32, [4]> var_3317_begin_0 = const()[name = tensor<string, []>("op_3317_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_3317_end_0 = const()[name = tensor<string, []>("op_3317_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_3317_end_mask_0 = const()[name = tensor<string, []>("op_3317_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3317_cast = slice_by_index(begin = var_3317_begin_0, end = var_3317_end_0, end_mask = var_3317_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3317_cast")];
            tensor<int32, [4]> var_3321_begin_0 = const()[name = tensor<string, []>("op_3321_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_3321_end_0 = const()[name = tensor<string, []>("op_3321_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_3321_end_mask_0 = const()[name = tensor<string, []>("op_3321_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3321_cast = slice_by_index(begin = var_3321_begin_0, end = var_3321_end_0, end_mask = var_3321_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3321_cast")];
            tensor<int32, [4]> var_3325_begin_0 = const()[name = tensor<string, []>("op_3325_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3325_end_0 = const()[name = tensor<string, []>("op_3325_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_3325_end_mask_0 = const()[name = tensor<string, []>("op_3325_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3325_cast = slice_by_index(begin = var_3325_begin_0, end = var_3325_end_0, end_mask = var_3325_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3325_cast")];
            tensor<int32, [4]> var_3329_begin_0 = const()[name = tensor<string, []>("op_3329_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_3329_end_0 = const()[name = tensor<string, []>("op_3329_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_3329_end_mask_0 = const()[name = tensor<string, []>("op_3329_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3329_cast = slice_by_index(begin = var_3329_begin_0, end = var_3329_end_0, end_mask = var_3329_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3329_cast")];
            tensor<int32, [4]> var_3333_begin_0 = const()[name = tensor<string, []>("op_3333_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_3333_end_0 = const()[name = tensor<string, []>("op_3333_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_3333_end_mask_0 = const()[name = tensor<string, []>("op_3333_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3333_cast = slice_by_index(begin = var_3333_begin_0, end = var_3333_end_0, end_mask = var_3333_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3333_cast")];
            tensor<int32, [4]> var_3337_begin_0 = const()[name = tensor<string, []>("op_3337_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_3337_end_0 = const()[name = tensor<string, []>("op_3337_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_3337_end_mask_0 = const()[name = tensor<string, []>("op_3337_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3337_cast = slice_by_index(begin = var_3337_begin_0, end = var_3337_end_0, end_mask = var_3337_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3337_cast")];
            tensor<int32, [4]> var_3341_begin_0 = const()[name = tensor<string, []>("op_3341_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_3341_end_0 = const()[name = tensor<string, []>("op_3341_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_3341_end_mask_0 = const()[name = tensor<string, []>("op_3341_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3341_cast = slice_by_index(begin = var_3341_begin_0, end = var_3341_end_0, end_mask = var_3341_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3341_cast")];
            tensor<int32, [4]> var_3345_begin_0 = const()[name = tensor<string, []>("op_3345_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_3345_end_0 = const()[name = tensor<string, []>("op_3345_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_3345_end_mask_0 = const()[name = tensor<string, []>("op_3345_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3345_cast = slice_by_index(begin = var_3345_begin_0, end = var_3345_end_0, end_mask = var_3345_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3345_cast")];
            tensor<int32, [4]> var_3349_begin_0 = const()[name = tensor<string, []>("op_3349_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_3349_end_0 = const()[name = tensor<string, []>("op_3349_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_3349_end_mask_0 = const()[name = tensor<string, []>("op_3349_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3349_cast = slice_by_index(begin = var_3349_begin_0, end = var_3349_end_0, end_mask = var_3349_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3349_cast")];
            tensor<int32, [4]> var_3353_begin_0 = const()[name = tensor<string, []>("op_3353_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_3353_end_0 = const()[name = tensor<string, []>("op_3353_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_3353_end_mask_0 = const()[name = tensor<string, []>("op_3353_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3353_cast = slice_by_index(begin = var_3353_begin_0, end = var_3353_end_0, end_mask = var_3353_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3353_cast")];
            tensor<int32, [4]> var_3357_begin_0 = const()[name = tensor<string, []>("op_3357_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_3357_end_0 = const()[name = tensor<string, []>("op_3357_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_3357_end_mask_0 = const()[name = tensor<string, []>("op_3357_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3357_cast = slice_by_index(begin = var_3357_begin_0, end = var_3357_end_0, end_mask = var_3357_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3357_cast")];
            tensor<int32, [4]> var_3361_begin_0 = const()[name = tensor<string, []>("op_3361_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_3361_end_0 = const()[name = tensor<string, []>("op_3361_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_3361_end_mask_0 = const()[name = tensor<string, []>("op_3361_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3361_cast = slice_by_index(begin = var_3361_begin_0, end = var_3361_end_0, end_mask = var_3361_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3361_cast")];
            tensor<int32, [4]> var_3365_begin_0 = const()[name = tensor<string, []>("op_3365_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_3365_end_0 = const()[name = tensor<string, []>("op_3365_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_3365_end_mask_0 = const()[name = tensor<string, []>("op_3365_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3365_cast = slice_by_index(begin = var_3365_begin_0, end = var_3365_end_0, end_mask = var_3365_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3365_cast")];
            tensor<int32, [4]> var_3369_begin_0 = const()[name = tensor<string, []>("op_3369_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_3369_end_0 = const()[name = tensor<string, []>("op_3369_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_3369_end_mask_0 = const()[name = tensor<string, []>("op_3369_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3369_cast = slice_by_index(begin = var_3369_begin_0, end = var_3369_end_0, end_mask = var_3369_end_mask_0, x = transpose_130)[name = tensor<string, []>("op_3369_cast")];
            tensor<int32, [4]> var_3371_begin_0 = const()[name = tensor<string, []>("op_3371_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3371_end_0 = const()[name = tensor<string, []>("op_3371_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_3371_end_mask_0 = const()[name = tensor<string, []>("op_3371_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3371_cast = slice_by_index(begin = var_3371_begin_0, end = var_3371_end_0, end_mask = var_3371_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3371_cast")];
            tensor<int32, [4]> var_3375_begin_0 = const()[name = tensor<string, []>("op_3375_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_3375_end_0 = const()[name = tensor<string, []>("op_3375_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_3375_end_mask_0 = const()[name = tensor<string, []>("op_3375_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3375_cast = slice_by_index(begin = var_3375_begin_0, end = var_3375_end_0, end_mask = var_3375_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3375_cast")];
            tensor<int32, [4]> var_3379_begin_0 = const()[name = tensor<string, []>("op_3379_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_3379_end_0 = const()[name = tensor<string, []>("op_3379_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_3379_end_mask_0 = const()[name = tensor<string, []>("op_3379_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3379_cast = slice_by_index(begin = var_3379_begin_0, end = var_3379_end_0, end_mask = var_3379_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3379_cast")];
            tensor<int32, [4]> var_3383_begin_0 = const()[name = tensor<string, []>("op_3383_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_3383_end_0 = const()[name = tensor<string, []>("op_3383_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_3383_end_mask_0 = const()[name = tensor<string, []>("op_3383_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3383_cast = slice_by_index(begin = var_3383_begin_0, end = var_3383_end_0, end_mask = var_3383_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3383_cast")];
            tensor<int32, [4]> var_3387_begin_0 = const()[name = tensor<string, []>("op_3387_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_3387_end_0 = const()[name = tensor<string, []>("op_3387_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_3387_end_mask_0 = const()[name = tensor<string, []>("op_3387_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3387_cast = slice_by_index(begin = var_3387_begin_0, end = var_3387_end_0, end_mask = var_3387_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3387_cast")];
            tensor<int32, [4]> var_3391_begin_0 = const()[name = tensor<string, []>("op_3391_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_3391_end_0 = const()[name = tensor<string, []>("op_3391_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_3391_end_mask_0 = const()[name = tensor<string, []>("op_3391_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3391_cast = slice_by_index(begin = var_3391_begin_0, end = var_3391_end_0, end_mask = var_3391_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3391_cast")];
            tensor<int32, [4]> var_3395_begin_0 = const()[name = tensor<string, []>("op_3395_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_3395_end_0 = const()[name = tensor<string, []>("op_3395_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_3395_end_mask_0 = const()[name = tensor<string, []>("op_3395_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3395_cast = slice_by_index(begin = var_3395_begin_0, end = var_3395_end_0, end_mask = var_3395_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3395_cast")];
            tensor<int32, [4]> var_3399_begin_0 = const()[name = tensor<string, []>("op_3399_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_3399_end_0 = const()[name = tensor<string, []>("op_3399_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_3399_end_mask_0 = const()[name = tensor<string, []>("op_3399_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3399_cast = slice_by_index(begin = var_3399_begin_0, end = var_3399_end_0, end_mask = var_3399_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3399_cast")];
            tensor<int32, [4]> var_3403_begin_0 = const()[name = tensor<string, []>("op_3403_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_3403_end_0 = const()[name = tensor<string, []>("op_3403_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_3403_end_mask_0 = const()[name = tensor<string, []>("op_3403_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3403_cast = slice_by_index(begin = var_3403_begin_0, end = var_3403_end_0, end_mask = var_3403_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3403_cast")];
            tensor<int32, [4]> var_3407_begin_0 = const()[name = tensor<string, []>("op_3407_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_3407_end_0 = const()[name = tensor<string, []>("op_3407_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_3407_end_mask_0 = const()[name = tensor<string, []>("op_3407_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3407_cast = slice_by_index(begin = var_3407_begin_0, end = var_3407_end_0, end_mask = var_3407_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3407_cast")];
            tensor<int32, [4]> var_3411_begin_0 = const()[name = tensor<string, []>("op_3411_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_3411_end_0 = const()[name = tensor<string, []>("op_3411_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_3411_end_mask_0 = const()[name = tensor<string, []>("op_3411_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3411_cast = slice_by_index(begin = var_3411_begin_0, end = var_3411_end_0, end_mask = var_3411_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3411_cast")];
            tensor<int32, [4]> var_3415_begin_0 = const()[name = tensor<string, []>("op_3415_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_3415_end_0 = const()[name = tensor<string, []>("op_3415_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_3415_end_mask_0 = const()[name = tensor<string, []>("op_3415_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3415_cast = slice_by_index(begin = var_3415_begin_0, end = var_3415_end_0, end_mask = var_3415_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3415_cast")];
            tensor<int32, [4]> var_3419_begin_0 = const()[name = tensor<string, []>("op_3419_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_3419_end_0 = const()[name = tensor<string, []>("op_3419_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_3419_end_mask_0 = const()[name = tensor<string, []>("op_3419_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3419_cast = slice_by_index(begin = var_3419_begin_0, end = var_3419_end_0, end_mask = var_3419_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3419_cast")];
            tensor<int32, [4]> var_3423_begin_0 = const()[name = tensor<string, []>("op_3423_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_3423_end_0 = const()[name = tensor<string, []>("op_3423_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_3423_end_mask_0 = const()[name = tensor<string, []>("op_3423_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3423_cast = slice_by_index(begin = var_3423_begin_0, end = var_3423_end_0, end_mask = var_3423_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3423_cast")];
            tensor<int32, [4]> var_3427_begin_0 = const()[name = tensor<string, []>("op_3427_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_3427_end_0 = const()[name = tensor<string, []>("op_3427_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_3427_end_mask_0 = const()[name = tensor<string, []>("op_3427_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3427_cast = slice_by_index(begin = var_3427_begin_0, end = var_3427_end_0, end_mask = var_3427_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3427_cast")];
            tensor<int32, [4]> var_3431_begin_0 = const()[name = tensor<string, []>("op_3431_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_3431_end_0 = const()[name = tensor<string, []>("op_3431_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_3431_end_mask_0 = const()[name = tensor<string, []>("op_3431_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3431_cast = slice_by_index(begin = var_3431_begin_0, end = var_3431_end_0, end_mask = var_3431_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3431_cast")];
            tensor<int32, [4]> var_3435_begin_0 = const()[name = tensor<string, []>("op_3435_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_3435_end_0 = const()[name = tensor<string, []>("op_3435_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_3435_end_mask_0 = const()[name = tensor<string, []>("op_3435_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3435_cast = slice_by_index(begin = var_3435_begin_0, end = var_3435_end_0, end_mask = var_3435_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3435_cast")];
            tensor<int32, [4]> var_3439_begin_0 = const()[name = tensor<string, []>("op_3439_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_3439_end_0 = const()[name = tensor<string, []>("op_3439_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_3439_end_mask_0 = const()[name = tensor<string, []>("op_3439_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3439_cast = slice_by_index(begin = var_3439_begin_0, end = var_3439_end_0, end_mask = var_3439_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3439_cast")];
            tensor<int32, [4]> var_3443_begin_0 = const()[name = tensor<string, []>("op_3443_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_3443_end_0 = const()[name = tensor<string, []>("op_3443_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_3443_end_mask_0 = const()[name = tensor<string, []>("op_3443_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3443_cast = slice_by_index(begin = var_3443_begin_0, end = var_3443_end_0, end_mask = var_3443_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3443_cast")];
            tensor<int32, [4]> var_3447_begin_0 = const()[name = tensor<string, []>("op_3447_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_3447_end_0 = const()[name = tensor<string, []>("op_3447_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_3447_end_mask_0 = const()[name = tensor<string, []>("op_3447_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3447_cast = slice_by_index(begin = var_3447_begin_0, end = var_3447_end_0, end_mask = var_3447_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_3447_cast")];
            tensor<string, []> var_3451_equation_0 = const()[name = tensor<string, []>("op_3451_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3451_cast = einsum(equation = var_3451_equation_0, values = (var_3293_cast, var_3210_cast))[name = tensor<string, []>("op_3451_cast")];
            tensor<fp16, []> var_3452_to_fp16 = const()[name = tensor<string, []>("op_3452_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_201_cast = mul(x = var_3451_cast, y = var_3452_to_fp16)[name = tensor<string, []>("aw_201_cast")];
            tensor<string, []> var_3455_equation_0 = const()[name = tensor<string, []>("op_3455_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3455_cast = einsum(equation = var_3455_equation_0, values = (var_3297_cast, var_3214_cast))[name = tensor<string, []>("op_3455_cast")];
            tensor<fp16, []> var_3456_to_fp16 = const()[name = tensor<string, []>("op_3456_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_203_cast = mul(x = var_3455_cast, y = var_3456_to_fp16)[name = tensor<string, []>("aw_203_cast")];
            tensor<string, []> var_3459_equation_0 = const()[name = tensor<string, []>("op_3459_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3459_cast = einsum(equation = var_3459_equation_0, values = (var_3301_cast, var_3218_cast))[name = tensor<string, []>("op_3459_cast")];
            tensor<fp16, []> var_3460_to_fp16 = const()[name = tensor<string, []>("op_3460_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_205_cast = mul(x = var_3459_cast, y = var_3460_to_fp16)[name = tensor<string, []>("aw_205_cast")];
            tensor<string, []> var_3463_equation_0 = const()[name = tensor<string, []>("op_3463_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3463_cast = einsum(equation = var_3463_equation_0, values = (var_3305_cast, var_3222_cast))[name = tensor<string, []>("op_3463_cast")];
            tensor<fp16, []> var_3464_to_fp16 = const()[name = tensor<string, []>("op_3464_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_207_cast = mul(x = var_3463_cast, y = var_3464_to_fp16)[name = tensor<string, []>("aw_207_cast")];
            tensor<string, []> var_3467_equation_0 = const()[name = tensor<string, []>("op_3467_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3467_cast = einsum(equation = var_3467_equation_0, values = (var_3309_cast, var_3226_cast))[name = tensor<string, []>("op_3467_cast")];
            tensor<fp16, []> var_3468_to_fp16 = const()[name = tensor<string, []>("op_3468_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_209_cast = mul(x = var_3467_cast, y = var_3468_to_fp16)[name = tensor<string, []>("aw_209_cast")];
            tensor<string, []> var_3471_equation_0 = const()[name = tensor<string, []>("op_3471_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3471_cast = einsum(equation = var_3471_equation_0, values = (var_3313_cast, var_3230_cast))[name = tensor<string, []>("op_3471_cast")];
            tensor<fp16, []> var_3472_to_fp16 = const()[name = tensor<string, []>("op_3472_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_211_cast = mul(x = var_3471_cast, y = var_3472_to_fp16)[name = tensor<string, []>("aw_211_cast")];
            tensor<string, []> var_3475_equation_0 = const()[name = tensor<string, []>("op_3475_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3475_cast = einsum(equation = var_3475_equation_0, values = (var_3317_cast, var_3234_cast))[name = tensor<string, []>("op_3475_cast")];
            tensor<fp16, []> var_3476_to_fp16 = const()[name = tensor<string, []>("op_3476_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_213_cast = mul(x = var_3475_cast, y = var_3476_to_fp16)[name = tensor<string, []>("aw_213_cast")];
            tensor<string, []> var_3479_equation_0 = const()[name = tensor<string, []>("op_3479_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3479_cast = einsum(equation = var_3479_equation_0, values = (var_3321_cast, var_3238_cast))[name = tensor<string, []>("op_3479_cast")];
            tensor<fp16, []> var_3480_to_fp16 = const()[name = tensor<string, []>("op_3480_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_215_cast = mul(x = var_3479_cast, y = var_3480_to_fp16)[name = tensor<string, []>("aw_215_cast")];
            tensor<string, []> var_3483_equation_0 = const()[name = tensor<string, []>("op_3483_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3483_cast = einsum(equation = var_3483_equation_0, values = (var_3325_cast, var_3242_cast))[name = tensor<string, []>("op_3483_cast")];
            tensor<fp16, []> var_3484_to_fp16 = const()[name = tensor<string, []>("op_3484_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_217_cast = mul(x = var_3483_cast, y = var_3484_to_fp16)[name = tensor<string, []>("aw_217_cast")];
            tensor<string, []> var_3487_equation_0 = const()[name = tensor<string, []>("op_3487_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3487_cast = einsum(equation = var_3487_equation_0, values = (var_3329_cast, var_3246_cast))[name = tensor<string, []>("op_3487_cast")];
            tensor<fp16, []> var_3488_to_fp16 = const()[name = tensor<string, []>("op_3488_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_219_cast = mul(x = var_3487_cast, y = var_3488_to_fp16)[name = tensor<string, []>("aw_219_cast")];
            tensor<string, []> var_3491_equation_0 = const()[name = tensor<string, []>("op_3491_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3491_cast = einsum(equation = var_3491_equation_0, values = (var_3333_cast, var_3250_cast))[name = tensor<string, []>("op_3491_cast")];
            tensor<fp16, []> var_3492_to_fp16 = const()[name = tensor<string, []>("op_3492_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_221_cast = mul(x = var_3491_cast, y = var_3492_to_fp16)[name = tensor<string, []>("aw_221_cast")];
            tensor<string, []> var_3495_equation_0 = const()[name = tensor<string, []>("op_3495_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3495_cast = einsum(equation = var_3495_equation_0, values = (var_3337_cast, var_3254_cast))[name = tensor<string, []>("op_3495_cast")];
            tensor<fp16, []> var_3496_to_fp16 = const()[name = tensor<string, []>("op_3496_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_223_cast = mul(x = var_3495_cast, y = var_3496_to_fp16)[name = tensor<string, []>("aw_223_cast")];
            tensor<string, []> var_3499_equation_0 = const()[name = tensor<string, []>("op_3499_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3499_cast = einsum(equation = var_3499_equation_0, values = (var_3341_cast, var_3258_cast))[name = tensor<string, []>("op_3499_cast")];
            tensor<fp16, []> var_3500_to_fp16 = const()[name = tensor<string, []>("op_3500_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_225_cast = mul(x = var_3499_cast, y = var_3500_to_fp16)[name = tensor<string, []>("aw_225_cast")];
            tensor<string, []> var_3503_equation_0 = const()[name = tensor<string, []>("op_3503_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3503_cast = einsum(equation = var_3503_equation_0, values = (var_3345_cast, var_3262_cast))[name = tensor<string, []>("op_3503_cast")];
            tensor<fp16, []> var_3504_to_fp16 = const()[name = tensor<string, []>("op_3504_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_227_cast = mul(x = var_3503_cast, y = var_3504_to_fp16)[name = tensor<string, []>("aw_227_cast")];
            tensor<string, []> var_3507_equation_0 = const()[name = tensor<string, []>("op_3507_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3507_cast = einsum(equation = var_3507_equation_0, values = (var_3349_cast, var_3266_cast))[name = tensor<string, []>("op_3507_cast")];
            tensor<fp16, []> var_3508_to_fp16 = const()[name = tensor<string, []>("op_3508_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_229_cast = mul(x = var_3507_cast, y = var_3508_to_fp16)[name = tensor<string, []>("aw_229_cast")];
            tensor<string, []> var_3511_equation_0 = const()[name = tensor<string, []>("op_3511_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3511_cast = einsum(equation = var_3511_equation_0, values = (var_3353_cast, var_3270_cast))[name = tensor<string, []>("op_3511_cast")];
            tensor<fp16, []> var_3512_to_fp16 = const()[name = tensor<string, []>("op_3512_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_231_cast = mul(x = var_3511_cast, y = var_3512_to_fp16)[name = tensor<string, []>("aw_231_cast")];
            tensor<string, []> var_3515_equation_0 = const()[name = tensor<string, []>("op_3515_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3515_cast = einsum(equation = var_3515_equation_0, values = (var_3357_cast, var_3274_cast))[name = tensor<string, []>("op_3515_cast")];
            tensor<fp16, []> var_3516_to_fp16 = const()[name = tensor<string, []>("op_3516_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_233_cast = mul(x = var_3515_cast, y = var_3516_to_fp16)[name = tensor<string, []>("aw_233_cast")];
            tensor<string, []> var_3519_equation_0 = const()[name = tensor<string, []>("op_3519_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3519_cast = einsum(equation = var_3519_equation_0, values = (var_3361_cast, var_3278_cast))[name = tensor<string, []>("op_3519_cast")];
            tensor<fp16, []> var_3520_to_fp16 = const()[name = tensor<string, []>("op_3520_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_235_cast = mul(x = var_3519_cast, y = var_3520_to_fp16)[name = tensor<string, []>("aw_235_cast")];
            tensor<string, []> var_3523_equation_0 = const()[name = tensor<string, []>("op_3523_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3523_cast = einsum(equation = var_3523_equation_0, values = (var_3365_cast, var_3282_cast))[name = tensor<string, []>("op_3523_cast")];
            tensor<fp16, []> var_3524_to_fp16 = const()[name = tensor<string, []>("op_3524_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_237_cast = mul(x = var_3523_cast, y = var_3524_to_fp16)[name = tensor<string, []>("aw_237_cast")];
            tensor<string, []> var_3527_equation_0 = const()[name = tensor<string, []>("op_3527_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_3527_cast = einsum(equation = var_3527_equation_0, values = (var_3369_cast, var_3286_cast))[name = tensor<string, []>("op_3527_cast")];
            tensor<fp16, []> var_3528_to_fp16 = const()[name = tensor<string, []>("op_3528_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_239_cast = mul(x = var_3527_cast, y = var_3528_to_fp16)[name = tensor<string, []>("aw_239_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3530_cast = softmax(axis = var_2634, x = aw_201_cast)[name = tensor<string, []>("op_3530_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3531_cast = softmax(axis = var_2634, x = aw_203_cast)[name = tensor<string, []>("op_3531_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3532_cast = softmax(axis = var_2634, x = aw_205_cast)[name = tensor<string, []>("op_3532_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3533_cast = softmax(axis = var_2634, x = aw_207_cast)[name = tensor<string, []>("op_3533_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3534_cast = softmax(axis = var_2634, x = aw_209_cast)[name = tensor<string, []>("op_3534_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3535_cast = softmax(axis = var_2634, x = aw_211_cast)[name = tensor<string, []>("op_3535_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3536_cast = softmax(axis = var_2634, x = aw_213_cast)[name = tensor<string, []>("op_3536_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3537_cast = softmax(axis = var_2634, x = aw_215_cast)[name = tensor<string, []>("op_3537_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3538_cast = softmax(axis = var_2634, x = aw_217_cast)[name = tensor<string, []>("op_3538_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3539_cast = softmax(axis = var_2634, x = aw_219_cast)[name = tensor<string, []>("op_3539_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3540_cast = softmax(axis = var_2634, x = aw_221_cast)[name = tensor<string, []>("op_3540_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3541_cast = softmax(axis = var_2634, x = aw_223_cast)[name = tensor<string, []>("op_3541_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3542_cast = softmax(axis = var_2634, x = aw_225_cast)[name = tensor<string, []>("op_3542_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3543_cast = softmax(axis = var_2634, x = aw_227_cast)[name = tensor<string, []>("op_3543_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3544_cast = softmax(axis = var_2634, x = aw_229_cast)[name = tensor<string, []>("op_3544_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3545_cast = softmax(axis = var_2634, x = aw_231_cast)[name = tensor<string, []>("op_3545_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3546_cast = softmax(axis = var_2634, x = aw_233_cast)[name = tensor<string, []>("op_3546_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3547_cast = softmax(axis = var_2634, x = aw_235_cast)[name = tensor<string, []>("op_3547_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3548_cast = softmax(axis = var_2634, x = aw_237_cast)[name = tensor<string, []>("op_3548_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_3549_cast = softmax(axis = var_2634, x = aw_239_cast)[name = tensor<string, []>("op_3549_cast")];
            tensor<string, []> var_3551_equation_0 = const()[name = tensor<string, []>("op_3551_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3551_cast = einsum(equation = var_3551_equation_0, values = (var_3371_cast, var_3530_cast))[name = tensor<string, []>("op_3551_cast")];
            tensor<string, []> var_3553_equation_0 = const()[name = tensor<string, []>("op_3553_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3553_cast = einsum(equation = var_3553_equation_0, values = (var_3375_cast, var_3531_cast))[name = tensor<string, []>("op_3553_cast")];
            tensor<string, []> var_3555_equation_0 = const()[name = tensor<string, []>("op_3555_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3555_cast = einsum(equation = var_3555_equation_0, values = (var_3379_cast, var_3532_cast))[name = tensor<string, []>("op_3555_cast")];
            tensor<string, []> var_3557_equation_0 = const()[name = tensor<string, []>("op_3557_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3557_cast = einsum(equation = var_3557_equation_0, values = (var_3383_cast, var_3533_cast))[name = tensor<string, []>("op_3557_cast")];
            tensor<string, []> var_3559_equation_0 = const()[name = tensor<string, []>("op_3559_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3559_cast = einsum(equation = var_3559_equation_0, values = (var_3387_cast, var_3534_cast))[name = tensor<string, []>("op_3559_cast")];
            tensor<string, []> var_3561_equation_0 = const()[name = tensor<string, []>("op_3561_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3561_cast = einsum(equation = var_3561_equation_0, values = (var_3391_cast, var_3535_cast))[name = tensor<string, []>("op_3561_cast")];
            tensor<string, []> var_3563_equation_0 = const()[name = tensor<string, []>("op_3563_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3563_cast = einsum(equation = var_3563_equation_0, values = (var_3395_cast, var_3536_cast))[name = tensor<string, []>("op_3563_cast")];
            tensor<string, []> var_3565_equation_0 = const()[name = tensor<string, []>("op_3565_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3565_cast = einsum(equation = var_3565_equation_0, values = (var_3399_cast, var_3537_cast))[name = tensor<string, []>("op_3565_cast")];
            tensor<string, []> var_3567_equation_0 = const()[name = tensor<string, []>("op_3567_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3567_cast = einsum(equation = var_3567_equation_0, values = (var_3403_cast, var_3538_cast))[name = tensor<string, []>("op_3567_cast")];
            tensor<string, []> var_3569_equation_0 = const()[name = tensor<string, []>("op_3569_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3569_cast = einsum(equation = var_3569_equation_0, values = (var_3407_cast, var_3539_cast))[name = tensor<string, []>("op_3569_cast")];
            tensor<string, []> var_3571_equation_0 = const()[name = tensor<string, []>("op_3571_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3571_cast = einsum(equation = var_3571_equation_0, values = (var_3411_cast, var_3540_cast))[name = tensor<string, []>("op_3571_cast")];
            tensor<string, []> var_3573_equation_0 = const()[name = tensor<string, []>("op_3573_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3573_cast = einsum(equation = var_3573_equation_0, values = (var_3415_cast, var_3541_cast))[name = tensor<string, []>("op_3573_cast")];
            tensor<string, []> var_3575_equation_0 = const()[name = tensor<string, []>("op_3575_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3575_cast = einsum(equation = var_3575_equation_0, values = (var_3419_cast, var_3542_cast))[name = tensor<string, []>("op_3575_cast")];
            tensor<string, []> var_3577_equation_0 = const()[name = tensor<string, []>("op_3577_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3577_cast = einsum(equation = var_3577_equation_0, values = (var_3423_cast, var_3543_cast))[name = tensor<string, []>("op_3577_cast")];
            tensor<string, []> var_3579_equation_0 = const()[name = tensor<string, []>("op_3579_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3579_cast = einsum(equation = var_3579_equation_0, values = (var_3427_cast, var_3544_cast))[name = tensor<string, []>("op_3579_cast")];
            tensor<string, []> var_3581_equation_0 = const()[name = tensor<string, []>("op_3581_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3581_cast = einsum(equation = var_3581_equation_0, values = (var_3431_cast, var_3545_cast))[name = tensor<string, []>("op_3581_cast")];
            tensor<string, []> var_3583_equation_0 = const()[name = tensor<string, []>("op_3583_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3583_cast = einsum(equation = var_3583_equation_0, values = (var_3435_cast, var_3546_cast))[name = tensor<string, []>("op_3583_cast")];
            tensor<string, []> var_3585_equation_0 = const()[name = tensor<string, []>("op_3585_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3585_cast = einsum(equation = var_3585_equation_0, values = (var_3439_cast, var_3547_cast))[name = tensor<string, []>("op_3585_cast")];
            tensor<string, []> var_3587_equation_0 = const()[name = tensor<string, []>("op_3587_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3587_cast = einsum(equation = var_3587_equation_0, values = (var_3443_cast, var_3548_cast))[name = tensor<string, []>("op_3587_cast")];
            tensor<string, []> var_3589_equation_0 = const()[name = tensor<string, []>("op_3589_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_3589_cast = einsum(equation = var_3589_equation_0, values = (var_3447_cast, var_3549_cast))[name = tensor<string, []>("op_3589_cast")];
            tensor<bool, []> input_137_interleave_0 = const()[name = tensor<string, []>("input_137_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_137_cast = concat(axis = var_2634, interleave = input_137_interleave_0, values = (var_3551_cast, var_3553_cast, var_3555_cast, var_3557_cast, var_3559_cast, var_3561_cast, var_3563_cast, var_3565_cast, var_3567_cast, var_3569_cast, var_3571_cast, var_3573_cast, var_3575_cast, var_3577_cast, var_3579_cast, var_3581_cast, var_3583_cast, var_3585_cast, var_3587_cast, var_3589_cast))[name = tensor<string, []>("input_137_cast")];
            tensor<int32, [2]> var_3595 = const()[name = tensor<string, []>("op_3595"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3597 = const()[name = tensor<string, []>("op_3597"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3599_pad_type_0 = const()[name = tensor<string, []>("op_3599_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3599_pad_0 = const()[name = tensor<string, []>("op_3599_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(225049216)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(228326080)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_3599_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_3597, groups = var_2634, pad = var_3599_pad_0, pad_type = var_3599_pad_type_0, strides = var_3595, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_137_cast)[name = tensor<string, []>("op_3599_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_29_cast = add(x = var_3599_cast, y = inputs_27_cast)[name = tensor<string, []>("inputs_29_cast")];
            tensor<int32, [1]> var_3603 = const()[name = tensor<string, []>("op_3603"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_29_cast = reduce_mean(axes = var_3603, keep_dims = var_2629, x = inputs_29_cast)[name = tensor<string, []>("channels_mean_29_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_29_cast = sub(x = inputs_29_cast, y = channels_mean_29_cast)[name = tensor<string, []>("zero_mean_29_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_29_cast = mul(x = zero_mean_29_cast, y = zero_mean_29_cast)[name = tensor<string, []>("zero_mean_sq_29_cast")];
            tensor<int32, [1]> var_3607 = const()[name = tensor<string, []>("op_3607"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_3608_cast = reduce_mean(axes = var_3607, keep_dims = var_2629, x = zero_mean_sq_29_cast)[name = tensor<string, []>("op_3608_cast")];
            tensor<fp16, []> var_3609_to_fp16 = const()[name = tensor<string, []>("op_3609_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_3610_cast = add(x = var_3608_cast, y = var_3609_to_fp16)[name = tensor<string, []>("op_3610_cast")];
            tensor<fp16, []> denom_29_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_29_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_29_cast = rsqrt(epsilon = denom_29_epsilon_0_to_fp16, x = var_3610_cast)[name = tensor<string, []>("denom_29_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_29_cast = mul(x = zero_mean_29_cast, y = denom_29_cast)[name = tensor<string, []>("out_29_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_3614_to_fp16 = const()[name = tensor<string, []>("op_3614_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(228328704)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_3615_cast = add(x = out_29_cast, y = var_3614_to_fp16)[name = tensor<string, []>("op_3615_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_3617_to_fp16 = const()[name = tensor<string, []>("op_3617_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(228331328)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_139_cast = mul(x = var_3615_cast, y = var_3617_to_fp16)[name = tensor<string, []>("input_139_cast")];
            tensor<int32, [2]> var_3625 = const()[name = tensor<string, []>("op_3625"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3627 = const()[name = tensor<string, []>("op_3627"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3629_pad_type_0 = const()[name = tensor<string, []>("op_3629_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3629_pad_0 = const()[name = tensor<string, []>("op_3629_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(228333952)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(254548416)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_3629_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_3627, groups = var_2634, pad = var_3629_pad_0, pad_type = var_3629_pad_type_0, strides = var_3625, weight = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_139_cast)[name = tensor<string, []>("op_3629_cast")];
            tensor<int32, [2]> var_3630_split_sizes_0 = const()[name = tensor<string, []>("op_3630_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_3630_axis_0 = const()[name = tensor<string, []>("op_3630_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_3630_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_3630_cast_1 = split(axis = var_3630_axis_0, split_sizes = var_3630_split_sizes_0, x = var_3629_cast)[name = tensor<string, []>("op_3630_cast")];
            tensor<string, []> var_3632_mode_0 = const()[name = tensor<string, []>("op_3632_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_3632_cast = gelu(mode = var_3632_mode_0, x = var_3630_cast_1)[name = tensor<string, []>("op_3632_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_141_cast = mul(x = var_3630_cast_0, y = var_3632_cast)[name = tensor<string, []>("input_141_cast")];
            tensor<int32, [2]> var_3636 = const()[name = tensor<string, []>("op_3636"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3638 = const()[name = tensor<string, []>("op_3638"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3640_pad_type_0 = const()[name = tensor<string, []>("op_3640_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3640_pad_0 = const()[name = tensor<string, []>("op_3640_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(254568960)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(267676224)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_3640_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_3638, groups = var_2634, pad = var_3640_pad_0, pad_type = var_3640_pad_type_0, strides = var_3636, weight = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_141_cast)[name = tensor<string, []>("op_3640_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_31_cast = add(x = var_3640_cast, y = inputs_29_cast)[name = tensor<string, []>("inputs_31_cast")];
            tensor<int32, [1]> var_3650 = const()[name = tensor<string, []>("op_3650"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_31_cast = reduce_mean(axes = var_3650, keep_dims = var_2629, x = inputs_31_cast)[name = tensor<string, []>("channels_mean_31_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_31_cast = sub(x = inputs_31_cast, y = channels_mean_31_cast)[name = tensor<string, []>("zero_mean_31_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_31_cast = mul(x = zero_mean_31_cast, y = zero_mean_31_cast)[name = tensor<string, []>("zero_mean_sq_31_cast")];
            tensor<int32, [1]> var_3654 = const()[name = tensor<string, []>("op_3654"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_3655_cast = reduce_mean(axes = var_3654, keep_dims = var_2629, x = zero_mean_sq_31_cast)[name = tensor<string, []>("op_3655_cast")];
            tensor<fp16, []> var_3656_to_fp16 = const()[name = tensor<string, []>("op_3656_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_3657_cast = add(x = var_3655_cast, y = var_3656_to_fp16)[name = tensor<string, []>("op_3657_cast")];
            tensor<fp16, []> denom_31_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_31_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_31_cast = rsqrt(epsilon = denom_31_epsilon_0_to_fp16, x = var_3657_cast)[name = tensor<string, []>("denom_31_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_31_cast = mul(x = zero_mean_31_cast, y = denom_31_cast)[name = tensor<string, []>("out_31_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_3661_to_fp16 = const()[name = tensor<string, []>("op_3661_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(267678848)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_3662_cast = add(x = out_31_cast, y = var_3661_to_fp16)[name = tensor<string, []>("op_3662_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_3664_to_fp16 = const()[name = tensor<string, []>("op_3664_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(267681472)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_71_cast = mul(x = var_3662_cast, y = var_3664_to_fp16)[name = tensor<string, []>("hidden_states_71_cast")];
            tensor<int32, [2]> var_3671 = const()[name = tensor<string, []>("op_3671"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3673 = const()[name = tensor<string, []>("op_3673"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_21_pad_type_0 = const()[name = tensor<string, []>("q_21_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_21_pad_0 = const()[name = tensor<string, []>("q_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(267684096)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_21_cast = conv(dilations = var_3673, groups = var_2634, pad = q_21_pad_0, pad_type = q_21_pad_type_0, strides = var_3671, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_71_cast)[name = tensor<string, []>("q_21_cast")];
            tensor<int32, [2]> var_3677 = const()[name = tensor<string, []>("op_3677"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3679 = const()[name = tensor<string, []>("op_3679"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_41_pad_type_0 = const()[name = tensor<string, []>("k_41_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_41_pad_0 = const()[name = tensor<string, []>("k_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(270960960)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_41_cast = conv(dilations = var_3679, groups = var_2634, pad = k_41_pad_0, pad_type = k_41_pad_type_0, strides = var_3677, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_71_cast)[name = tensor<string, []>("k_41_cast")];
            tensor<int32, [2]> var_3683 = const()[name = tensor<string, []>("op_3683"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3685 = const()[name = tensor<string, []>("op_3685"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_21_pad_type_0 = const()[name = tensor<string, []>("v_21_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_21_pad_0 = const()[name = tensor<string, []>("v_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(274237824)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_21_cast = conv(dilations = var_3685, groups = var_2634, pad = v_21_pad_0, pad_type = v_21_pad_type_0, strides = var_3683, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_71_cast)[name = tensor<string, []>("v_21_cast")];
            tensor<int32, [4]> var_3689_begin_0 = const()[name = tensor<string, []>("op_3689_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3689_end_0 = const()[name = tensor<string, []>("op_3689_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3689_end_mask_0 = const()[name = tensor<string, []>("op_3689_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3689_cast = slice_by_index(begin = var_3689_begin_0, end = var_3689_end_0, end_mask = var_3689_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3689_cast")];
            tensor<int32, [4]> var_3693_begin_0 = const()[name = tensor<string, []>("op_3693_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_3693_end_0 = const()[name = tensor<string, []>("op_3693_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_3693_end_mask_0 = const()[name = tensor<string, []>("op_3693_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3693_cast = slice_by_index(begin = var_3693_begin_0, end = var_3693_end_0, end_mask = var_3693_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3693_cast")];
            tensor<int32, [4]> var_3697_begin_0 = const()[name = tensor<string, []>("op_3697_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_3697_end_0 = const()[name = tensor<string, []>("op_3697_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_3697_end_mask_0 = const()[name = tensor<string, []>("op_3697_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3697_cast = slice_by_index(begin = var_3697_begin_0, end = var_3697_end_0, end_mask = var_3697_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3697_cast")];
            tensor<int32, [4]> var_3701_begin_0 = const()[name = tensor<string, []>("op_3701_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_3701_end_0 = const()[name = tensor<string, []>("op_3701_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_3701_end_mask_0 = const()[name = tensor<string, []>("op_3701_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3701_cast = slice_by_index(begin = var_3701_begin_0, end = var_3701_end_0, end_mask = var_3701_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3701_cast")];
            tensor<int32, [4]> var_3705_begin_0 = const()[name = tensor<string, []>("op_3705_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_3705_end_0 = const()[name = tensor<string, []>("op_3705_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_3705_end_mask_0 = const()[name = tensor<string, []>("op_3705_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3705_cast = slice_by_index(begin = var_3705_begin_0, end = var_3705_end_0, end_mask = var_3705_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3705_cast")];
            tensor<int32, [4]> var_3709_begin_0 = const()[name = tensor<string, []>("op_3709_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_3709_end_0 = const()[name = tensor<string, []>("op_3709_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_3709_end_mask_0 = const()[name = tensor<string, []>("op_3709_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3709_cast = slice_by_index(begin = var_3709_begin_0, end = var_3709_end_0, end_mask = var_3709_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3709_cast")];
            tensor<int32, [4]> var_3713_begin_0 = const()[name = tensor<string, []>("op_3713_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_3713_end_0 = const()[name = tensor<string, []>("op_3713_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_3713_end_mask_0 = const()[name = tensor<string, []>("op_3713_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3713_cast = slice_by_index(begin = var_3713_begin_0, end = var_3713_end_0, end_mask = var_3713_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3713_cast")];
            tensor<int32, [4]> var_3717_begin_0 = const()[name = tensor<string, []>("op_3717_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_3717_end_0 = const()[name = tensor<string, []>("op_3717_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_3717_end_mask_0 = const()[name = tensor<string, []>("op_3717_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3717_cast = slice_by_index(begin = var_3717_begin_0, end = var_3717_end_0, end_mask = var_3717_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3717_cast")];
            tensor<int32, [4]> var_3721_begin_0 = const()[name = tensor<string, []>("op_3721_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_3721_end_0 = const()[name = tensor<string, []>("op_3721_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_3721_end_mask_0 = const()[name = tensor<string, []>("op_3721_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3721_cast = slice_by_index(begin = var_3721_begin_0, end = var_3721_end_0, end_mask = var_3721_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3721_cast")];
            tensor<int32, [4]> var_3725_begin_0 = const()[name = tensor<string, []>("op_3725_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_3725_end_0 = const()[name = tensor<string, []>("op_3725_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_3725_end_mask_0 = const()[name = tensor<string, []>("op_3725_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3725_cast = slice_by_index(begin = var_3725_begin_0, end = var_3725_end_0, end_mask = var_3725_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3725_cast")];
            tensor<int32, [4]> var_3729_begin_0 = const()[name = tensor<string, []>("op_3729_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_3729_end_0 = const()[name = tensor<string, []>("op_3729_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_3729_end_mask_0 = const()[name = tensor<string, []>("op_3729_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3729_cast = slice_by_index(begin = var_3729_begin_0, end = var_3729_end_0, end_mask = var_3729_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3729_cast")];
            tensor<int32, [4]> var_3733_begin_0 = const()[name = tensor<string, []>("op_3733_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_3733_end_0 = const()[name = tensor<string, []>("op_3733_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_3733_end_mask_0 = const()[name = tensor<string, []>("op_3733_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3733_cast = slice_by_index(begin = var_3733_begin_0, end = var_3733_end_0, end_mask = var_3733_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3733_cast")];
            tensor<int32, [4]> var_3737_begin_0 = const()[name = tensor<string, []>("op_3737_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_3737_end_0 = const()[name = tensor<string, []>("op_3737_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_3737_end_mask_0 = const()[name = tensor<string, []>("op_3737_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3737_cast = slice_by_index(begin = var_3737_begin_0, end = var_3737_end_0, end_mask = var_3737_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3737_cast")];
            tensor<int32, [4]> var_3741_begin_0 = const()[name = tensor<string, []>("op_3741_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_3741_end_0 = const()[name = tensor<string, []>("op_3741_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_3741_end_mask_0 = const()[name = tensor<string, []>("op_3741_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3741_cast = slice_by_index(begin = var_3741_begin_0, end = var_3741_end_0, end_mask = var_3741_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3741_cast")];
            tensor<int32, [4]> var_3745_begin_0 = const()[name = tensor<string, []>("op_3745_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_3745_end_0 = const()[name = tensor<string, []>("op_3745_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_3745_end_mask_0 = const()[name = tensor<string, []>("op_3745_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3745_cast = slice_by_index(begin = var_3745_begin_0, end = var_3745_end_0, end_mask = var_3745_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3745_cast")];
            tensor<int32, [4]> var_3749_begin_0 = const()[name = tensor<string, []>("op_3749_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_3749_end_0 = const()[name = tensor<string, []>("op_3749_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_3749_end_mask_0 = const()[name = tensor<string, []>("op_3749_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3749_cast = slice_by_index(begin = var_3749_begin_0, end = var_3749_end_0, end_mask = var_3749_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3749_cast")];
            tensor<int32, [4]> var_3753_begin_0 = const()[name = tensor<string, []>("op_3753_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_3753_end_0 = const()[name = tensor<string, []>("op_3753_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_3753_end_mask_0 = const()[name = tensor<string, []>("op_3753_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3753_cast = slice_by_index(begin = var_3753_begin_0, end = var_3753_end_0, end_mask = var_3753_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3753_cast")];
            tensor<int32, [4]> var_3757_begin_0 = const()[name = tensor<string, []>("op_3757_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_3757_end_0 = const()[name = tensor<string, []>("op_3757_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_3757_end_mask_0 = const()[name = tensor<string, []>("op_3757_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3757_cast = slice_by_index(begin = var_3757_begin_0, end = var_3757_end_0, end_mask = var_3757_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3757_cast")];
            tensor<int32, [4]> var_3761_begin_0 = const()[name = tensor<string, []>("op_3761_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_3761_end_0 = const()[name = tensor<string, []>("op_3761_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_3761_end_mask_0 = const()[name = tensor<string, []>("op_3761_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3761_cast = slice_by_index(begin = var_3761_begin_0, end = var_3761_end_0, end_mask = var_3761_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3761_cast")];
            tensor<int32, [4]> var_3765_begin_0 = const()[name = tensor<string, []>("op_3765_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_3765_end_0 = const()[name = tensor<string, []>("op_3765_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_3765_end_mask_0 = const()[name = tensor<string, []>("op_3765_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3765_cast = slice_by_index(begin = var_3765_begin_0, end = var_3765_end_0, end_mask = var_3765_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_3765_cast")];
            tensor<int32, [4]> k_43_perm_0 = const()[name = tensor<string, []>("k_43_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_3772_begin_0 = const()[name = tensor<string, []>("op_3772_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3772_end_0 = const()[name = tensor<string, []>("op_3772_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_3772_end_mask_0 = const()[name = tensor<string, []>("op_3772_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_129 = transpose(perm = k_43_perm_0, x = k_41_cast)[name = tensor<string, []>("transpose_129")];
            tensor<fp16, [2, 1024, 1, 64]> var_3772_cast = slice_by_index(begin = var_3772_begin_0, end = var_3772_end_0, end_mask = var_3772_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3772_cast")];
            tensor<int32, [4]> var_3776_begin_0 = const()[name = tensor<string, []>("op_3776_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_3776_end_0 = const()[name = tensor<string, []>("op_3776_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_3776_end_mask_0 = const()[name = tensor<string, []>("op_3776_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3776_cast = slice_by_index(begin = var_3776_begin_0, end = var_3776_end_0, end_mask = var_3776_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3776_cast")];
            tensor<int32, [4]> var_3780_begin_0 = const()[name = tensor<string, []>("op_3780_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_3780_end_0 = const()[name = tensor<string, []>("op_3780_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_3780_end_mask_0 = const()[name = tensor<string, []>("op_3780_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3780_cast = slice_by_index(begin = var_3780_begin_0, end = var_3780_end_0, end_mask = var_3780_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3780_cast")];
            tensor<int32, [4]> var_3784_begin_0 = const()[name = tensor<string, []>("op_3784_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_3784_end_0 = const()[name = tensor<string, []>("op_3784_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_3784_end_mask_0 = const()[name = tensor<string, []>("op_3784_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3784_cast = slice_by_index(begin = var_3784_begin_0, end = var_3784_end_0, end_mask = var_3784_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3784_cast")];
            tensor<int32, [4]> var_3788_begin_0 = const()[name = tensor<string, []>("op_3788_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_3788_end_0 = const()[name = tensor<string, []>("op_3788_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_3788_end_mask_0 = const()[name = tensor<string, []>("op_3788_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3788_cast = slice_by_index(begin = var_3788_begin_0, end = var_3788_end_0, end_mask = var_3788_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3788_cast")];
            tensor<int32, [4]> var_3792_begin_0 = const()[name = tensor<string, []>("op_3792_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_3792_end_0 = const()[name = tensor<string, []>("op_3792_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_3792_end_mask_0 = const()[name = tensor<string, []>("op_3792_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3792_cast = slice_by_index(begin = var_3792_begin_0, end = var_3792_end_0, end_mask = var_3792_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3792_cast")];
            tensor<int32, [4]> var_3796_begin_0 = const()[name = tensor<string, []>("op_3796_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_3796_end_0 = const()[name = tensor<string, []>("op_3796_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_3796_end_mask_0 = const()[name = tensor<string, []>("op_3796_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3796_cast = slice_by_index(begin = var_3796_begin_0, end = var_3796_end_0, end_mask = var_3796_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3796_cast")];
            tensor<int32, [4]> var_3800_begin_0 = const()[name = tensor<string, []>("op_3800_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_3800_end_0 = const()[name = tensor<string, []>("op_3800_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_3800_end_mask_0 = const()[name = tensor<string, []>("op_3800_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3800_cast = slice_by_index(begin = var_3800_begin_0, end = var_3800_end_0, end_mask = var_3800_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3800_cast")];
            tensor<int32, [4]> var_3804_begin_0 = const()[name = tensor<string, []>("op_3804_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3804_end_0 = const()[name = tensor<string, []>("op_3804_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_3804_end_mask_0 = const()[name = tensor<string, []>("op_3804_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3804_cast = slice_by_index(begin = var_3804_begin_0, end = var_3804_end_0, end_mask = var_3804_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3804_cast")];
            tensor<int32, [4]> var_3808_begin_0 = const()[name = tensor<string, []>("op_3808_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_3808_end_0 = const()[name = tensor<string, []>("op_3808_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_3808_end_mask_0 = const()[name = tensor<string, []>("op_3808_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3808_cast = slice_by_index(begin = var_3808_begin_0, end = var_3808_end_0, end_mask = var_3808_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3808_cast")];
            tensor<int32, [4]> var_3812_begin_0 = const()[name = tensor<string, []>("op_3812_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_3812_end_0 = const()[name = tensor<string, []>("op_3812_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_3812_end_mask_0 = const()[name = tensor<string, []>("op_3812_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3812_cast = slice_by_index(begin = var_3812_begin_0, end = var_3812_end_0, end_mask = var_3812_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3812_cast")];
            tensor<int32, [4]> var_3816_begin_0 = const()[name = tensor<string, []>("op_3816_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_3816_end_0 = const()[name = tensor<string, []>("op_3816_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_3816_end_mask_0 = const()[name = tensor<string, []>("op_3816_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3816_cast = slice_by_index(begin = var_3816_begin_0, end = var_3816_end_0, end_mask = var_3816_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3816_cast")];
            tensor<int32, [4]> var_3820_begin_0 = const()[name = tensor<string, []>("op_3820_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_3820_end_0 = const()[name = tensor<string, []>("op_3820_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_3820_end_mask_0 = const()[name = tensor<string, []>("op_3820_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3820_cast = slice_by_index(begin = var_3820_begin_0, end = var_3820_end_0, end_mask = var_3820_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3820_cast")];
            tensor<int32, [4]> var_3824_begin_0 = const()[name = tensor<string, []>("op_3824_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_3824_end_0 = const()[name = tensor<string, []>("op_3824_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_3824_end_mask_0 = const()[name = tensor<string, []>("op_3824_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3824_cast = slice_by_index(begin = var_3824_begin_0, end = var_3824_end_0, end_mask = var_3824_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3824_cast")];
            tensor<int32, [4]> var_3828_begin_0 = const()[name = tensor<string, []>("op_3828_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_3828_end_0 = const()[name = tensor<string, []>("op_3828_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_3828_end_mask_0 = const()[name = tensor<string, []>("op_3828_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3828_cast = slice_by_index(begin = var_3828_begin_0, end = var_3828_end_0, end_mask = var_3828_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3828_cast")];
            tensor<int32, [4]> var_3832_begin_0 = const()[name = tensor<string, []>("op_3832_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_3832_end_0 = const()[name = tensor<string, []>("op_3832_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_3832_end_mask_0 = const()[name = tensor<string, []>("op_3832_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3832_cast = slice_by_index(begin = var_3832_begin_0, end = var_3832_end_0, end_mask = var_3832_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3832_cast")];
            tensor<int32, [4]> var_3836_begin_0 = const()[name = tensor<string, []>("op_3836_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_3836_end_0 = const()[name = tensor<string, []>("op_3836_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_3836_end_mask_0 = const()[name = tensor<string, []>("op_3836_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3836_cast = slice_by_index(begin = var_3836_begin_0, end = var_3836_end_0, end_mask = var_3836_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3836_cast")];
            tensor<int32, [4]> var_3840_begin_0 = const()[name = tensor<string, []>("op_3840_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_3840_end_0 = const()[name = tensor<string, []>("op_3840_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_3840_end_mask_0 = const()[name = tensor<string, []>("op_3840_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3840_cast = slice_by_index(begin = var_3840_begin_0, end = var_3840_end_0, end_mask = var_3840_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3840_cast")];
            tensor<int32, [4]> var_3844_begin_0 = const()[name = tensor<string, []>("op_3844_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_3844_end_0 = const()[name = tensor<string, []>("op_3844_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_3844_end_mask_0 = const()[name = tensor<string, []>("op_3844_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3844_cast = slice_by_index(begin = var_3844_begin_0, end = var_3844_end_0, end_mask = var_3844_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3844_cast")];
            tensor<int32, [4]> var_3848_begin_0 = const()[name = tensor<string, []>("op_3848_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_3848_end_0 = const()[name = tensor<string, []>("op_3848_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_3848_end_mask_0 = const()[name = tensor<string, []>("op_3848_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3848_cast = slice_by_index(begin = var_3848_begin_0, end = var_3848_end_0, end_mask = var_3848_end_mask_0, x = transpose_129)[name = tensor<string, []>("op_3848_cast")];
            tensor<int32, [4]> var_3850_begin_0 = const()[name = tensor<string, []>("op_3850_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3850_end_0 = const()[name = tensor<string, []>("op_3850_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3850_end_mask_0 = const()[name = tensor<string, []>("op_3850_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3850_cast = slice_by_index(begin = var_3850_begin_0, end = var_3850_end_0, end_mask = var_3850_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3850_cast")];
            tensor<int32, [4]> var_3854_begin_0 = const()[name = tensor<string, []>("op_3854_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_3854_end_0 = const()[name = tensor<string, []>("op_3854_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_3854_end_mask_0 = const()[name = tensor<string, []>("op_3854_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3854_cast = slice_by_index(begin = var_3854_begin_0, end = var_3854_end_0, end_mask = var_3854_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3854_cast")];
            tensor<int32, [4]> var_3858_begin_0 = const()[name = tensor<string, []>("op_3858_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_3858_end_0 = const()[name = tensor<string, []>("op_3858_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_3858_end_mask_0 = const()[name = tensor<string, []>("op_3858_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3858_cast = slice_by_index(begin = var_3858_begin_0, end = var_3858_end_0, end_mask = var_3858_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3858_cast")];
            tensor<int32, [4]> var_3862_begin_0 = const()[name = tensor<string, []>("op_3862_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_3862_end_0 = const()[name = tensor<string, []>("op_3862_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_3862_end_mask_0 = const()[name = tensor<string, []>("op_3862_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3862_cast = slice_by_index(begin = var_3862_begin_0, end = var_3862_end_0, end_mask = var_3862_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3862_cast")];
            tensor<int32, [4]> var_3866_begin_0 = const()[name = tensor<string, []>("op_3866_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_3866_end_0 = const()[name = tensor<string, []>("op_3866_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_3866_end_mask_0 = const()[name = tensor<string, []>("op_3866_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3866_cast = slice_by_index(begin = var_3866_begin_0, end = var_3866_end_0, end_mask = var_3866_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3866_cast")];
            tensor<int32, [4]> var_3870_begin_0 = const()[name = tensor<string, []>("op_3870_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_3870_end_0 = const()[name = tensor<string, []>("op_3870_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_3870_end_mask_0 = const()[name = tensor<string, []>("op_3870_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3870_cast = slice_by_index(begin = var_3870_begin_0, end = var_3870_end_0, end_mask = var_3870_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3870_cast")];
            tensor<int32, [4]> var_3874_begin_0 = const()[name = tensor<string, []>("op_3874_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_3874_end_0 = const()[name = tensor<string, []>("op_3874_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_3874_end_mask_0 = const()[name = tensor<string, []>("op_3874_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3874_cast = slice_by_index(begin = var_3874_begin_0, end = var_3874_end_0, end_mask = var_3874_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3874_cast")];
            tensor<int32, [4]> var_3878_begin_0 = const()[name = tensor<string, []>("op_3878_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_3878_end_0 = const()[name = tensor<string, []>("op_3878_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_3878_end_mask_0 = const()[name = tensor<string, []>("op_3878_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3878_cast = slice_by_index(begin = var_3878_begin_0, end = var_3878_end_0, end_mask = var_3878_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3878_cast")];
            tensor<int32, [4]> var_3882_begin_0 = const()[name = tensor<string, []>("op_3882_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_3882_end_0 = const()[name = tensor<string, []>("op_3882_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_3882_end_mask_0 = const()[name = tensor<string, []>("op_3882_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3882_cast = slice_by_index(begin = var_3882_begin_0, end = var_3882_end_0, end_mask = var_3882_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3882_cast")];
            tensor<int32, [4]> var_3886_begin_0 = const()[name = tensor<string, []>("op_3886_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_3886_end_0 = const()[name = tensor<string, []>("op_3886_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_3886_end_mask_0 = const()[name = tensor<string, []>("op_3886_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3886_cast = slice_by_index(begin = var_3886_begin_0, end = var_3886_end_0, end_mask = var_3886_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3886_cast")];
            tensor<int32, [4]> var_3890_begin_0 = const()[name = tensor<string, []>("op_3890_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_3890_end_0 = const()[name = tensor<string, []>("op_3890_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_3890_end_mask_0 = const()[name = tensor<string, []>("op_3890_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3890_cast = slice_by_index(begin = var_3890_begin_0, end = var_3890_end_0, end_mask = var_3890_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3890_cast")];
            tensor<int32, [4]> var_3894_begin_0 = const()[name = tensor<string, []>("op_3894_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_3894_end_0 = const()[name = tensor<string, []>("op_3894_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_3894_end_mask_0 = const()[name = tensor<string, []>("op_3894_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3894_cast = slice_by_index(begin = var_3894_begin_0, end = var_3894_end_0, end_mask = var_3894_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3894_cast")];
            tensor<int32, [4]> var_3898_begin_0 = const()[name = tensor<string, []>("op_3898_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_3898_end_0 = const()[name = tensor<string, []>("op_3898_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_3898_end_mask_0 = const()[name = tensor<string, []>("op_3898_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3898_cast = slice_by_index(begin = var_3898_begin_0, end = var_3898_end_0, end_mask = var_3898_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3898_cast")];
            tensor<int32, [4]> var_3902_begin_0 = const()[name = tensor<string, []>("op_3902_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_3902_end_0 = const()[name = tensor<string, []>("op_3902_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_3902_end_mask_0 = const()[name = tensor<string, []>("op_3902_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3902_cast = slice_by_index(begin = var_3902_begin_0, end = var_3902_end_0, end_mask = var_3902_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3902_cast")];
            tensor<int32, [4]> var_3906_begin_0 = const()[name = tensor<string, []>("op_3906_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_3906_end_0 = const()[name = tensor<string, []>("op_3906_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_3906_end_mask_0 = const()[name = tensor<string, []>("op_3906_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3906_cast = slice_by_index(begin = var_3906_begin_0, end = var_3906_end_0, end_mask = var_3906_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3906_cast")];
            tensor<int32, [4]> var_3910_begin_0 = const()[name = tensor<string, []>("op_3910_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_3910_end_0 = const()[name = tensor<string, []>("op_3910_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_3910_end_mask_0 = const()[name = tensor<string, []>("op_3910_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3910_cast = slice_by_index(begin = var_3910_begin_0, end = var_3910_end_0, end_mask = var_3910_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3910_cast")];
            tensor<int32, [4]> var_3914_begin_0 = const()[name = tensor<string, []>("op_3914_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_3914_end_0 = const()[name = tensor<string, []>("op_3914_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_3914_end_mask_0 = const()[name = tensor<string, []>("op_3914_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3914_cast = slice_by_index(begin = var_3914_begin_0, end = var_3914_end_0, end_mask = var_3914_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3914_cast")];
            tensor<int32, [4]> var_3918_begin_0 = const()[name = tensor<string, []>("op_3918_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_3918_end_0 = const()[name = tensor<string, []>("op_3918_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_3918_end_mask_0 = const()[name = tensor<string, []>("op_3918_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3918_cast = slice_by_index(begin = var_3918_begin_0, end = var_3918_end_0, end_mask = var_3918_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3918_cast")];
            tensor<int32, [4]> var_3922_begin_0 = const()[name = tensor<string, []>("op_3922_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_3922_end_0 = const()[name = tensor<string, []>("op_3922_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_3922_end_mask_0 = const()[name = tensor<string, []>("op_3922_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3922_cast = slice_by_index(begin = var_3922_begin_0, end = var_3922_end_0, end_mask = var_3922_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3922_cast")];
            tensor<int32, [4]> var_3926_begin_0 = const()[name = tensor<string, []>("op_3926_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_3926_end_0 = const()[name = tensor<string, []>("op_3926_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_3926_end_mask_0 = const()[name = tensor<string, []>("op_3926_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3926_cast = slice_by_index(begin = var_3926_begin_0, end = var_3926_end_0, end_mask = var_3926_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_3926_cast")];
            tensor<string, []> var_3930_equation_0 = const()[name = tensor<string, []>("op_3930_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3930_cast = einsum(equation = var_3930_equation_0, values = (var_3772_cast, var_3689_cast))[name = tensor<string, []>("op_3930_cast")];
            tensor<fp16, []> var_3931_to_fp16 = const()[name = tensor<string, []>("op_3931_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_241_cast = mul(x = var_3930_cast, y = var_3931_to_fp16)[name = tensor<string, []>("aw_241_cast")];
            tensor<string, []> var_3934_equation_0 = const()[name = tensor<string, []>("op_3934_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3934_cast = einsum(equation = var_3934_equation_0, values = (var_3776_cast, var_3693_cast))[name = tensor<string, []>("op_3934_cast")];
            tensor<fp16, []> var_3935_to_fp16 = const()[name = tensor<string, []>("op_3935_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_243_cast = mul(x = var_3934_cast, y = var_3935_to_fp16)[name = tensor<string, []>("aw_243_cast")];
            tensor<string, []> var_3938_equation_0 = const()[name = tensor<string, []>("op_3938_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3938_cast = einsum(equation = var_3938_equation_0, values = (var_3780_cast, var_3697_cast))[name = tensor<string, []>("op_3938_cast")];
            tensor<fp16, []> var_3939_to_fp16 = const()[name = tensor<string, []>("op_3939_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_245_cast = mul(x = var_3938_cast, y = var_3939_to_fp16)[name = tensor<string, []>("aw_245_cast")];
            tensor<string, []> var_3942_equation_0 = const()[name = tensor<string, []>("op_3942_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3942_cast = einsum(equation = var_3942_equation_0, values = (var_3784_cast, var_3701_cast))[name = tensor<string, []>("op_3942_cast")];
            tensor<fp16, []> var_3943_to_fp16 = const()[name = tensor<string, []>("op_3943_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_247_cast = mul(x = var_3942_cast, y = var_3943_to_fp16)[name = tensor<string, []>("aw_247_cast")];
            tensor<string, []> var_3946_equation_0 = const()[name = tensor<string, []>("op_3946_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3946_cast = einsum(equation = var_3946_equation_0, values = (var_3788_cast, var_3705_cast))[name = tensor<string, []>("op_3946_cast")];
            tensor<fp16, []> var_3947_to_fp16 = const()[name = tensor<string, []>("op_3947_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_249_cast = mul(x = var_3946_cast, y = var_3947_to_fp16)[name = tensor<string, []>("aw_249_cast")];
            tensor<string, []> var_3950_equation_0 = const()[name = tensor<string, []>("op_3950_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3950_cast = einsum(equation = var_3950_equation_0, values = (var_3792_cast, var_3709_cast))[name = tensor<string, []>("op_3950_cast")];
            tensor<fp16, []> var_3951_to_fp16 = const()[name = tensor<string, []>("op_3951_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_251_cast = mul(x = var_3950_cast, y = var_3951_to_fp16)[name = tensor<string, []>("aw_251_cast")];
            tensor<string, []> var_3954_equation_0 = const()[name = tensor<string, []>("op_3954_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3954_cast = einsum(equation = var_3954_equation_0, values = (var_3796_cast, var_3713_cast))[name = tensor<string, []>("op_3954_cast")];
            tensor<fp16, []> var_3955_to_fp16 = const()[name = tensor<string, []>("op_3955_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_253_cast = mul(x = var_3954_cast, y = var_3955_to_fp16)[name = tensor<string, []>("aw_253_cast")];
            tensor<string, []> var_3958_equation_0 = const()[name = tensor<string, []>("op_3958_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3958_cast = einsum(equation = var_3958_equation_0, values = (var_3800_cast, var_3717_cast))[name = tensor<string, []>("op_3958_cast")];
            tensor<fp16, []> var_3959_to_fp16 = const()[name = tensor<string, []>("op_3959_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_255_cast = mul(x = var_3958_cast, y = var_3959_to_fp16)[name = tensor<string, []>("aw_255_cast")];
            tensor<string, []> var_3962_equation_0 = const()[name = tensor<string, []>("op_3962_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3962_cast = einsum(equation = var_3962_equation_0, values = (var_3804_cast, var_3721_cast))[name = tensor<string, []>("op_3962_cast")];
            tensor<fp16, []> var_3963_to_fp16 = const()[name = tensor<string, []>("op_3963_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_257_cast = mul(x = var_3962_cast, y = var_3963_to_fp16)[name = tensor<string, []>("aw_257_cast")];
            tensor<string, []> var_3966_equation_0 = const()[name = tensor<string, []>("op_3966_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3966_cast = einsum(equation = var_3966_equation_0, values = (var_3808_cast, var_3725_cast))[name = tensor<string, []>("op_3966_cast")];
            tensor<fp16, []> var_3967_to_fp16 = const()[name = tensor<string, []>("op_3967_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_259_cast = mul(x = var_3966_cast, y = var_3967_to_fp16)[name = tensor<string, []>("aw_259_cast")];
            tensor<string, []> var_3970_equation_0 = const()[name = tensor<string, []>("op_3970_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3970_cast = einsum(equation = var_3970_equation_0, values = (var_3812_cast, var_3729_cast))[name = tensor<string, []>("op_3970_cast")];
            tensor<fp16, []> var_3971_to_fp16 = const()[name = tensor<string, []>("op_3971_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_261_cast = mul(x = var_3970_cast, y = var_3971_to_fp16)[name = tensor<string, []>("aw_261_cast")];
            tensor<string, []> var_3974_equation_0 = const()[name = tensor<string, []>("op_3974_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3974_cast = einsum(equation = var_3974_equation_0, values = (var_3816_cast, var_3733_cast))[name = tensor<string, []>("op_3974_cast")];
            tensor<fp16, []> var_3975_to_fp16 = const()[name = tensor<string, []>("op_3975_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_263_cast = mul(x = var_3974_cast, y = var_3975_to_fp16)[name = tensor<string, []>("aw_263_cast")];
            tensor<string, []> var_3978_equation_0 = const()[name = tensor<string, []>("op_3978_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3978_cast = einsum(equation = var_3978_equation_0, values = (var_3820_cast, var_3737_cast))[name = tensor<string, []>("op_3978_cast")];
            tensor<fp16, []> var_3979_to_fp16 = const()[name = tensor<string, []>("op_3979_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_265_cast = mul(x = var_3978_cast, y = var_3979_to_fp16)[name = tensor<string, []>("aw_265_cast")];
            tensor<string, []> var_3982_equation_0 = const()[name = tensor<string, []>("op_3982_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3982_cast = einsum(equation = var_3982_equation_0, values = (var_3824_cast, var_3741_cast))[name = tensor<string, []>("op_3982_cast")];
            tensor<fp16, []> var_3983_to_fp16 = const()[name = tensor<string, []>("op_3983_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_267_cast = mul(x = var_3982_cast, y = var_3983_to_fp16)[name = tensor<string, []>("aw_267_cast")];
            tensor<string, []> var_3986_equation_0 = const()[name = tensor<string, []>("op_3986_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3986_cast = einsum(equation = var_3986_equation_0, values = (var_3828_cast, var_3745_cast))[name = tensor<string, []>("op_3986_cast")];
            tensor<fp16, []> var_3987_to_fp16 = const()[name = tensor<string, []>("op_3987_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_269_cast = mul(x = var_3986_cast, y = var_3987_to_fp16)[name = tensor<string, []>("aw_269_cast")];
            tensor<string, []> var_3990_equation_0 = const()[name = tensor<string, []>("op_3990_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3990_cast = einsum(equation = var_3990_equation_0, values = (var_3832_cast, var_3749_cast))[name = tensor<string, []>("op_3990_cast")];
            tensor<fp16, []> var_3991_to_fp16 = const()[name = tensor<string, []>("op_3991_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_271_cast = mul(x = var_3990_cast, y = var_3991_to_fp16)[name = tensor<string, []>("aw_271_cast")];
            tensor<string, []> var_3994_equation_0 = const()[name = tensor<string, []>("op_3994_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3994_cast = einsum(equation = var_3994_equation_0, values = (var_3836_cast, var_3753_cast))[name = tensor<string, []>("op_3994_cast")];
            tensor<fp16, []> var_3995_to_fp16 = const()[name = tensor<string, []>("op_3995_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_273_cast = mul(x = var_3994_cast, y = var_3995_to_fp16)[name = tensor<string, []>("aw_273_cast")];
            tensor<string, []> var_3998_equation_0 = const()[name = tensor<string, []>("op_3998_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_3998_cast = einsum(equation = var_3998_equation_0, values = (var_3840_cast, var_3757_cast))[name = tensor<string, []>("op_3998_cast")];
            tensor<fp16, []> var_3999_to_fp16 = const()[name = tensor<string, []>("op_3999_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_275_cast = mul(x = var_3998_cast, y = var_3999_to_fp16)[name = tensor<string, []>("aw_275_cast")];
            tensor<string, []> var_4002_equation_0 = const()[name = tensor<string, []>("op_4002_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4002_cast = einsum(equation = var_4002_equation_0, values = (var_3844_cast, var_3761_cast))[name = tensor<string, []>("op_4002_cast")];
            tensor<fp16, []> var_4003_to_fp16 = const()[name = tensor<string, []>("op_4003_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_277_cast = mul(x = var_4002_cast, y = var_4003_to_fp16)[name = tensor<string, []>("aw_277_cast")];
            tensor<string, []> var_4006_equation_0 = const()[name = tensor<string, []>("op_4006_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4006_cast = einsum(equation = var_4006_equation_0, values = (var_3848_cast, var_3765_cast))[name = tensor<string, []>("op_4006_cast")];
            tensor<fp16, []> var_4007_to_fp16 = const()[name = tensor<string, []>("op_4007_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_279_cast = mul(x = var_4006_cast, y = var_4007_to_fp16)[name = tensor<string, []>("aw_279_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4009_cast = softmax(axis = var_2634, x = aw_241_cast)[name = tensor<string, []>("op_4009_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4010_cast = softmax(axis = var_2634, x = aw_243_cast)[name = tensor<string, []>("op_4010_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4011_cast = softmax(axis = var_2634, x = aw_245_cast)[name = tensor<string, []>("op_4011_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4012_cast = softmax(axis = var_2634, x = aw_247_cast)[name = tensor<string, []>("op_4012_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4013_cast = softmax(axis = var_2634, x = aw_249_cast)[name = tensor<string, []>("op_4013_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4014_cast = softmax(axis = var_2634, x = aw_251_cast)[name = tensor<string, []>("op_4014_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4015_cast = softmax(axis = var_2634, x = aw_253_cast)[name = tensor<string, []>("op_4015_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4016_cast = softmax(axis = var_2634, x = aw_255_cast)[name = tensor<string, []>("op_4016_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4017_cast = softmax(axis = var_2634, x = aw_257_cast)[name = tensor<string, []>("op_4017_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4018_cast = softmax(axis = var_2634, x = aw_259_cast)[name = tensor<string, []>("op_4018_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4019_cast = softmax(axis = var_2634, x = aw_261_cast)[name = tensor<string, []>("op_4019_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4020_cast = softmax(axis = var_2634, x = aw_263_cast)[name = tensor<string, []>("op_4020_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4021_cast = softmax(axis = var_2634, x = aw_265_cast)[name = tensor<string, []>("op_4021_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4022_cast = softmax(axis = var_2634, x = aw_267_cast)[name = tensor<string, []>("op_4022_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4023_cast = softmax(axis = var_2634, x = aw_269_cast)[name = tensor<string, []>("op_4023_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4024_cast = softmax(axis = var_2634, x = aw_271_cast)[name = tensor<string, []>("op_4024_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4025_cast = softmax(axis = var_2634, x = aw_273_cast)[name = tensor<string, []>("op_4025_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4026_cast = softmax(axis = var_2634, x = aw_275_cast)[name = tensor<string, []>("op_4026_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4027_cast = softmax(axis = var_2634, x = aw_277_cast)[name = tensor<string, []>("op_4027_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4028_cast = softmax(axis = var_2634, x = aw_279_cast)[name = tensor<string, []>("op_4028_cast")];
            tensor<string, []> var_4030_equation_0 = const()[name = tensor<string, []>("op_4030_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4030_cast = einsum(equation = var_4030_equation_0, values = (var_3850_cast, var_4009_cast))[name = tensor<string, []>("op_4030_cast")];
            tensor<string, []> var_4032_equation_0 = const()[name = tensor<string, []>("op_4032_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4032_cast = einsum(equation = var_4032_equation_0, values = (var_3854_cast, var_4010_cast))[name = tensor<string, []>("op_4032_cast")];
            tensor<string, []> var_4034_equation_0 = const()[name = tensor<string, []>("op_4034_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4034_cast = einsum(equation = var_4034_equation_0, values = (var_3858_cast, var_4011_cast))[name = tensor<string, []>("op_4034_cast")];
            tensor<string, []> var_4036_equation_0 = const()[name = tensor<string, []>("op_4036_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4036_cast = einsum(equation = var_4036_equation_0, values = (var_3862_cast, var_4012_cast))[name = tensor<string, []>("op_4036_cast")];
            tensor<string, []> var_4038_equation_0 = const()[name = tensor<string, []>("op_4038_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4038_cast = einsum(equation = var_4038_equation_0, values = (var_3866_cast, var_4013_cast))[name = tensor<string, []>("op_4038_cast")];
            tensor<string, []> var_4040_equation_0 = const()[name = tensor<string, []>("op_4040_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4040_cast = einsum(equation = var_4040_equation_0, values = (var_3870_cast, var_4014_cast))[name = tensor<string, []>("op_4040_cast")];
            tensor<string, []> var_4042_equation_0 = const()[name = tensor<string, []>("op_4042_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4042_cast = einsum(equation = var_4042_equation_0, values = (var_3874_cast, var_4015_cast))[name = tensor<string, []>("op_4042_cast")];
            tensor<string, []> var_4044_equation_0 = const()[name = tensor<string, []>("op_4044_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4044_cast = einsum(equation = var_4044_equation_0, values = (var_3878_cast, var_4016_cast))[name = tensor<string, []>("op_4044_cast")];
            tensor<string, []> var_4046_equation_0 = const()[name = tensor<string, []>("op_4046_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4046_cast = einsum(equation = var_4046_equation_0, values = (var_3882_cast, var_4017_cast))[name = tensor<string, []>("op_4046_cast")];
            tensor<string, []> var_4048_equation_0 = const()[name = tensor<string, []>("op_4048_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4048_cast = einsum(equation = var_4048_equation_0, values = (var_3886_cast, var_4018_cast))[name = tensor<string, []>("op_4048_cast")];
            tensor<string, []> var_4050_equation_0 = const()[name = tensor<string, []>("op_4050_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4050_cast = einsum(equation = var_4050_equation_0, values = (var_3890_cast, var_4019_cast))[name = tensor<string, []>("op_4050_cast")];
            tensor<string, []> var_4052_equation_0 = const()[name = tensor<string, []>("op_4052_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4052_cast = einsum(equation = var_4052_equation_0, values = (var_3894_cast, var_4020_cast))[name = tensor<string, []>("op_4052_cast")];
            tensor<string, []> var_4054_equation_0 = const()[name = tensor<string, []>("op_4054_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4054_cast = einsum(equation = var_4054_equation_0, values = (var_3898_cast, var_4021_cast))[name = tensor<string, []>("op_4054_cast")];
            tensor<string, []> var_4056_equation_0 = const()[name = tensor<string, []>("op_4056_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4056_cast = einsum(equation = var_4056_equation_0, values = (var_3902_cast, var_4022_cast))[name = tensor<string, []>("op_4056_cast")];
            tensor<string, []> var_4058_equation_0 = const()[name = tensor<string, []>("op_4058_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4058_cast = einsum(equation = var_4058_equation_0, values = (var_3906_cast, var_4023_cast))[name = tensor<string, []>("op_4058_cast")];
            tensor<string, []> var_4060_equation_0 = const()[name = tensor<string, []>("op_4060_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4060_cast = einsum(equation = var_4060_equation_0, values = (var_3910_cast, var_4024_cast))[name = tensor<string, []>("op_4060_cast")];
            tensor<string, []> var_4062_equation_0 = const()[name = tensor<string, []>("op_4062_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4062_cast = einsum(equation = var_4062_equation_0, values = (var_3914_cast, var_4025_cast))[name = tensor<string, []>("op_4062_cast")];
            tensor<string, []> var_4064_equation_0 = const()[name = tensor<string, []>("op_4064_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4064_cast = einsum(equation = var_4064_equation_0, values = (var_3918_cast, var_4026_cast))[name = tensor<string, []>("op_4064_cast")];
            tensor<string, []> var_4066_equation_0 = const()[name = tensor<string, []>("op_4066_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4066_cast = einsum(equation = var_4066_equation_0, values = (var_3922_cast, var_4027_cast))[name = tensor<string, []>("op_4066_cast")];
            tensor<string, []> var_4068_equation_0 = const()[name = tensor<string, []>("op_4068_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4068_cast = einsum(equation = var_4068_equation_0, values = (var_3926_cast, var_4028_cast))[name = tensor<string, []>("op_4068_cast")];
            tensor<bool, []> input_143_interleave_0 = const()[name = tensor<string, []>("input_143_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_143_cast = concat(axis = var_2634, interleave = input_143_interleave_0, values = (var_4030_cast, var_4032_cast, var_4034_cast, var_4036_cast, var_4038_cast, var_4040_cast, var_4042_cast, var_4044_cast, var_4046_cast, var_4048_cast, var_4050_cast, var_4052_cast, var_4054_cast, var_4056_cast, var_4058_cast, var_4060_cast, var_4062_cast, var_4064_cast, var_4066_cast, var_4068_cast))[name = tensor<string, []>("input_143_cast")];
            tensor<int32, [2]> var_4074 = const()[name = tensor<string, []>("op_4074"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4076 = const()[name = tensor<string, []>("op_4076"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_4078_pad_type_0 = const()[name = tensor<string, []>("op_4078_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_4078_pad_0 = const()[name = tensor<string, []>("op_4078_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(277514688)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(280791552)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_4078_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_4076, groups = var_2634, pad = var_4078_pad_0, pad_type = var_4078_pad_type_0, strides = var_4074, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_143_cast)[name = tensor<string, []>("op_4078_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_33_cast = add(x = var_4078_cast, y = inputs_31_cast)[name = tensor<string, []>("inputs_33_cast")];
            tensor<int32, [1]> var_4082 = const()[name = tensor<string, []>("op_4082"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_33_cast = reduce_mean(axes = var_4082, keep_dims = var_2629, x = inputs_33_cast)[name = tensor<string, []>("channels_mean_33_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_33_cast = sub(x = inputs_33_cast, y = channels_mean_33_cast)[name = tensor<string, []>("zero_mean_33_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_33_cast = mul(x = zero_mean_33_cast, y = zero_mean_33_cast)[name = tensor<string, []>("zero_mean_sq_33_cast")];
            tensor<int32, [1]> var_4086 = const()[name = tensor<string, []>("op_4086"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_4087_cast = reduce_mean(axes = var_4086, keep_dims = var_2629, x = zero_mean_sq_33_cast)[name = tensor<string, []>("op_4087_cast")];
            tensor<fp16, []> var_4088_to_fp16 = const()[name = tensor<string, []>("op_4088_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_4089_cast = add(x = var_4087_cast, y = var_4088_to_fp16)[name = tensor<string, []>("op_4089_cast")];
            tensor<fp16, []> denom_33_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_33_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_33_cast = rsqrt(epsilon = denom_33_epsilon_0_to_fp16, x = var_4089_cast)[name = tensor<string, []>("denom_33_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_33_cast = mul(x = zero_mean_33_cast, y = denom_33_cast)[name = tensor<string, []>("out_33_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_4093_to_fp16 = const()[name = tensor<string, []>("op_4093_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(280794176)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_4094_cast = add(x = out_33_cast, y = var_4093_to_fp16)[name = tensor<string, []>("op_4094_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_4096_to_fp16 = const()[name = tensor<string, []>("op_4096_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(280796800)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_73_cast = mul(x = var_4094_cast, y = var_4096_to_fp16)[name = tensor<string, []>("hidden_states_73_cast")];
            tensor<int32, [2]> var_4103 = const()[name = tensor<string, []>("op_4103"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4105 = const()[name = tensor<string, []>("op_4105"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_23_pad_type_0 = const()[name = tensor<string, []>("q_23_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_23_pad_0 = const()[name = tensor<string, []>("q_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(280799424)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_23_cast = conv(dilations = var_4105, groups = var_2634, pad = q_23_pad_0, pad_type = q_23_pad_type_0, strides = var_4103, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_73_cast)[name = tensor<string, []>("q_23_cast")];
            tensor<int32, [2]> var_4109 = const()[name = tensor<string, []>("op_4109"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4111 = const()[name = tensor<string, []>("op_4111"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_45_pad_type_0 = const()[name = tensor<string, []>("k_45_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_45_pad_0 = const()[name = tensor<string, []>("k_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(284076288)))];
            tensor<fp16, [2, 1280, 1, 77]> k_45_cast = conv(dilations = var_4111, groups = var_2634, pad = k_45_pad_0, pad_type = k_45_pad_type_0, strides = var_4109, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_45_cast")];
            tensor<int32, [2]> var_4115 = const()[name = tensor<string, []>("op_4115"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4117 = const()[name = tensor<string, []>("op_4117"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_23_pad_type_0 = const()[name = tensor<string, []>("v_23_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_23_pad_0 = const()[name = tensor<string, []>("v_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(289319232)))];
            tensor<fp16, [2, 1280, 1, 77]> v_23_cast = conv(dilations = var_4117, groups = var_2634, pad = v_23_pad_0, pad_type = v_23_pad_type_0, strides = var_4115, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_23_cast")];
            tensor<int32, [4]> var_4121_begin_0 = const()[name = tensor<string, []>("op_4121_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4121_end_0 = const()[name = tensor<string, []>("op_4121_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_4121_end_mask_0 = const()[name = tensor<string, []>("op_4121_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4121_cast = slice_by_index(begin = var_4121_begin_0, end = var_4121_end_0, end_mask = var_4121_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4121_cast")];
            tensor<int32, [4]> var_4125_begin_0 = const()[name = tensor<string, []>("op_4125_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_4125_end_0 = const()[name = tensor<string, []>("op_4125_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_4125_end_mask_0 = const()[name = tensor<string, []>("op_4125_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4125_cast = slice_by_index(begin = var_4125_begin_0, end = var_4125_end_0, end_mask = var_4125_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4125_cast")];
            tensor<int32, [4]> var_4129_begin_0 = const()[name = tensor<string, []>("op_4129_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_4129_end_0 = const()[name = tensor<string, []>("op_4129_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_4129_end_mask_0 = const()[name = tensor<string, []>("op_4129_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4129_cast = slice_by_index(begin = var_4129_begin_0, end = var_4129_end_0, end_mask = var_4129_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4129_cast")];
            tensor<int32, [4]> var_4133_begin_0 = const()[name = tensor<string, []>("op_4133_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_4133_end_0 = const()[name = tensor<string, []>("op_4133_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_4133_end_mask_0 = const()[name = tensor<string, []>("op_4133_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4133_cast = slice_by_index(begin = var_4133_begin_0, end = var_4133_end_0, end_mask = var_4133_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4133_cast")];
            tensor<int32, [4]> var_4137_begin_0 = const()[name = tensor<string, []>("op_4137_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_4137_end_0 = const()[name = tensor<string, []>("op_4137_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_4137_end_mask_0 = const()[name = tensor<string, []>("op_4137_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4137_cast = slice_by_index(begin = var_4137_begin_0, end = var_4137_end_0, end_mask = var_4137_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4137_cast")];
            tensor<int32, [4]> var_4141_begin_0 = const()[name = tensor<string, []>("op_4141_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_4141_end_0 = const()[name = tensor<string, []>("op_4141_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_4141_end_mask_0 = const()[name = tensor<string, []>("op_4141_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4141_cast = slice_by_index(begin = var_4141_begin_0, end = var_4141_end_0, end_mask = var_4141_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4141_cast")];
            tensor<int32, [4]> var_4145_begin_0 = const()[name = tensor<string, []>("op_4145_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_4145_end_0 = const()[name = tensor<string, []>("op_4145_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_4145_end_mask_0 = const()[name = tensor<string, []>("op_4145_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4145_cast = slice_by_index(begin = var_4145_begin_0, end = var_4145_end_0, end_mask = var_4145_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4145_cast")];
            tensor<int32, [4]> var_4149_begin_0 = const()[name = tensor<string, []>("op_4149_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_4149_end_0 = const()[name = tensor<string, []>("op_4149_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_4149_end_mask_0 = const()[name = tensor<string, []>("op_4149_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4149_cast = slice_by_index(begin = var_4149_begin_0, end = var_4149_end_0, end_mask = var_4149_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4149_cast")];
            tensor<int32, [4]> var_4153_begin_0 = const()[name = tensor<string, []>("op_4153_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_4153_end_0 = const()[name = tensor<string, []>("op_4153_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_4153_end_mask_0 = const()[name = tensor<string, []>("op_4153_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4153_cast = slice_by_index(begin = var_4153_begin_0, end = var_4153_end_0, end_mask = var_4153_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4153_cast")];
            tensor<int32, [4]> var_4157_begin_0 = const()[name = tensor<string, []>("op_4157_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_4157_end_0 = const()[name = tensor<string, []>("op_4157_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_4157_end_mask_0 = const()[name = tensor<string, []>("op_4157_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4157_cast = slice_by_index(begin = var_4157_begin_0, end = var_4157_end_0, end_mask = var_4157_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4157_cast")];
            tensor<int32, [4]> var_4161_begin_0 = const()[name = tensor<string, []>("op_4161_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_4161_end_0 = const()[name = tensor<string, []>("op_4161_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_4161_end_mask_0 = const()[name = tensor<string, []>("op_4161_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4161_cast = slice_by_index(begin = var_4161_begin_0, end = var_4161_end_0, end_mask = var_4161_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4161_cast")];
            tensor<int32, [4]> var_4165_begin_0 = const()[name = tensor<string, []>("op_4165_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_4165_end_0 = const()[name = tensor<string, []>("op_4165_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_4165_end_mask_0 = const()[name = tensor<string, []>("op_4165_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4165_cast = slice_by_index(begin = var_4165_begin_0, end = var_4165_end_0, end_mask = var_4165_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4165_cast")];
            tensor<int32, [4]> var_4169_begin_0 = const()[name = tensor<string, []>("op_4169_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_4169_end_0 = const()[name = tensor<string, []>("op_4169_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_4169_end_mask_0 = const()[name = tensor<string, []>("op_4169_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4169_cast = slice_by_index(begin = var_4169_begin_0, end = var_4169_end_0, end_mask = var_4169_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4169_cast")];
            tensor<int32, [4]> var_4173_begin_0 = const()[name = tensor<string, []>("op_4173_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_4173_end_0 = const()[name = tensor<string, []>("op_4173_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_4173_end_mask_0 = const()[name = tensor<string, []>("op_4173_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4173_cast = slice_by_index(begin = var_4173_begin_0, end = var_4173_end_0, end_mask = var_4173_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4173_cast")];
            tensor<int32, [4]> var_4177_begin_0 = const()[name = tensor<string, []>("op_4177_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_4177_end_0 = const()[name = tensor<string, []>("op_4177_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_4177_end_mask_0 = const()[name = tensor<string, []>("op_4177_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4177_cast = slice_by_index(begin = var_4177_begin_0, end = var_4177_end_0, end_mask = var_4177_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4177_cast")];
            tensor<int32, [4]> var_4181_begin_0 = const()[name = tensor<string, []>("op_4181_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_4181_end_0 = const()[name = tensor<string, []>("op_4181_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_4181_end_mask_0 = const()[name = tensor<string, []>("op_4181_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4181_cast = slice_by_index(begin = var_4181_begin_0, end = var_4181_end_0, end_mask = var_4181_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4181_cast")];
            tensor<int32, [4]> var_4185_begin_0 = const()[name = tensor<string, []>("op_4185_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_4185_end_0 = const()[name = tensor<string, []>("op_4185_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_4185_end_mask_0 = const()[name = tensor<string, []>("op_4185_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4185_cast = slice_by_index(begin = var_4185_begin_0, end = var_4185_end_0, end_mask = var_4185_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4185_cast")];
            tensor<int32, [4]> var_4189_begin_0 = const()[name = tensor<string, []>("op_4189_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_4189_end_0 = const()[name = tensor<string, []>("op_4189_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_4189_end_mask_0 = const()[name = tensor<string, []>("op_4189_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4189_cast = slice_by_index(begin = var_4189_begin_0, end = var_4189_end_0, end_mask = var_4189_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4189_cast")];
            tensor<int32, [4]> var_4193_begin_0 = const()[name = tensor<string, []>("op_4193_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_4193_end_0 = const()[name = tensor<string, []>("op_4193_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_4193_end_mask_0 = const()[name = tensor<string, []>("op_4193_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4193_cast = slice_by_index(begin = var_4193_begin_0, end = var_4193_end_0, end_mask = var_4193_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4193_cast")];
            tensor<int32, [4]> var_4197_begin_0 = const()[name = tensor<string, []>("op_4197_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_4197_end_0 = const()[name = tensor<string, []>("op_4197_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_4197_end_mask_0 = const()[name = tensor<string, []>("op_4197_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4197_cast = slice_by_index(begin = var_4197_begin_0, end = var_4197_end_0, end_mask = var_4197_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_4197_cast")];
            tensor<int32, [4]> k_47_perm_0 = const()[name = tensor<string, []>("k_47_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_4204_begin_0 = const()[name = tensor<string, []>("op_4204_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4204_end_0 = const()[name = tensor<string, []>("op_4204_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_4204_end_mask_0 = const()[name = tensor<string, []>("op_4204_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_128 = transpose(perm = k_47_perm_0, x = k_45_cast)[name = tensor<string, []>("transpose_128")];
            tensor<fp16, [2, 77, 1, 64]> var_4204_cast = slice_by_index(begin = var_4204_begin_0, end = var_4204_end_0, end_mask = var_4204_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4204_cast")];
            tensor<int32, [4]> var_4208_begin_0 = const()[name = tensor<string, []>("op_4208_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_4208_end_0 = const()[name = tensor<string, []>("op_4208_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_4208_end_mask_0 = const()[name = tensor<string, []>("op_4208_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4208_cast = slice_by_index(begin = var_4208_begin_0, end = var_4208_end_0, end_mask = var_4208_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4208_cast")];
            tensor<int32, [4]> var_4212_begin_0 = const()[name = tensor<string, []>("op_4212_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_4212_end_0 = const()[name = tensor<string, []>("op_4212_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_4212_end_mask_0 = const()[name = tensor<string, []>("op_4212_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4212_cast = slice_by_index(begin = var_4212_begin_0, end = var_4212_end_0, end_mask = var_4212_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4212_cast")];
            tensor<int32, [4]> var_4216_begin_0 = const()[name = tensor<string, []>("op_4216_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_4216_end_0 = const()[name = tensor<string, []>("op_4216_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_4216_end_mask_0 = const()[name = tensor<string, []>("op_4216_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4216_cast = slice_by_index(begin = var_4216_begin_0, end = var_4216_end_0, end_mask = var_4216_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4216_cast")];
            tensor<int32, [4]> var_4220_begin_0 = const()[name = tensor<string, []>("op_4220_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_4220_end_0 = const()[name = tensor<string, []>("op_4220_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_4220_end_mask_0 = const()[name = tensor<string, []>("op_4220_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4220_cast = slice_by_index(begin = var_4220_begin_0, end = var_4220_end_0, end_mask = var_4220_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4220_cast")];
            tensor<int32, [4]> var_4224_begin_0 = const()[name = tensor<string, []>("op_4224_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_4224_end_0 = const()[name = tensor<string, []>("op_4224_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_4224_end_mask_0 = const()[name = tensor<string, []>("op_4224_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4224_cast = slice_by_index(begin = var_4224_begin_0, end = var_4224_end_0, end_mask = var_4224_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4224_cast")];
            tensor<int32, [4]> var_4228_begin_0 = const()[name = tensor<string, []>("op_4228_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_4228_end_0 = const()[name = tensor<string, []>("op_4228_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_4228_end_mask_0 = const()[name = tensor<string, []>("op_4228_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4228_cast = slice_by_index(begin = var_4228_begin_0, end = var_4228_end_0, end_mask = var_4228_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4228_cast")];
            tensor<int32, [4]> var_4232_begin_0 = const()[name = tensor<string, []>("op_4232_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_4232_end_0 = const()[name = tensor<string, []>("op_4232_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_4232_end_mask_0 = const()[name = tensor<string, []>("op_4232_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4232_cast = slice_by_index(begin = var_4232_begin_0, end = var_4232_end_0, end_mask = var_4232_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4232_cast")];
            tensor<int32, [4]> var_4236_begin_0 = const()[name = tensor<string, []>("op_4236_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_4236_end_0 = const()[name = tensor<string, []>("op_4236_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_4236_end_mask_0 = const()[name = tensor<string, []>("op_4236_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4236_cast = slice_by_index(begin = var_4236_begin_0, end = var_4236_end_0, end_mask = var_4236_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4236_cast")];
            tensor<int32, [4]> var_4240_begin_0 = const()[name = tensor<string, []>("op_4240_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_4240_end_0 = const()[name = tensor<string, []>("op_4240_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_4240_end_mask_0 = const()[name = tensor<string, []>("op_4240_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4240_cast = slice_by_index(begin = var_4240_begin_0, end = var_4240_end_0, end_mask = var_4240_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4240_cast")];
            tensor<int32, [4]> var_4244_begin_0 = const()[name = tensor<string, []>("op_4244_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_4244_end_0 = const()[name = tensor<string, []>("op_4244_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_4244_end_mask_0 = const()[name = tensor<string, []>("op_4244_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4244_cast = slice_by_index(begin = var_4244_begin_0, end = var_4244_end_0, end_mask = var_4244_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4244_cast")];
            tensor<int32, [4]> var_4248_begin_0 = const()[name = tensor<string, []>("op_4248_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_4248_end_0 = const()[name = tensor<string, []>("op_4248_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_4248_end_mask_0 = const()[name = tensor<string, []>("op_4248_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4248_cast = slice_by_index(begin = var_4248_begin_0, end = var_4248_end_0, end_mask = var_4248_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4248_cast")];
            tensor<int32, [4]> var_4252_begin_0 = const()[name = tensor<string, []>("op_4252_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_4252_end_0 = const()[name = tensor<string, []>("op_4252_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_4252_end_mask_0 = const()[name = tensor<string, []>("op_4252_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4252_cast = slice_by_index(begin = var_4252_begin_0, end = var_4252_end_0, end_mask = var_4252_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4252_cast")];
            tensor<int32, [4]> var_4256_begin_0 = const()[name = tensor<string, []>("op_4256_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_4256_end_0 = const()[name = tensor<string, []>("op_4256_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_4256_end_mask_0 = const()[name = tensor<string, []>("op_4256_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4256_cast = slice_by_index(begin = var_4256_begin_0, end = var_4256_end_0, end_mask = var_4256_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4256_cast")];
            tensor<int32, [4]> var_4260_begin_0 = const()[name = tensor<string, []>("op_4260_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_4260_end_0 = const()[name = tensor<string, []>("op_4260_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_4260_end_mask_0 = const()[name = tensor<string, []>("op_4260_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4260_cast = slice_by_index(begin = var_4260_begin_0, end = var_4260_end_0, end_mask = var_4260_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4260_cast")];
            tensor<int32, [4]> var_4264_begin_0 = const()[name = tensor<string, []>("op_4264_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_4264_end_0 = const()[name = tensor<string, []>("op_4264_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_4264_end_mask_0 = const()[name = tensor<string, []>("op_4264_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4264_cast = slice_by_index(begin = var_4264_begin_0, end = var_4264_end_0, end_mask = var_4264_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4264_cast")];
            tensor<int32, [4]> var_4268_begin_0 = const()[name = tensor<string, []>("op_4268_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_4268_end_0 = const()[name = tensor<string, []>("op_4268_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_4268_end_mask_0 = const()[name = tensor<string, []>("op_4268_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4268_cast = slice_by_index(begin = var_4268_begin_0, end = var_4268_end_0, end_mask = var_4268_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4268_cast")];
            tensor<int32, [4]> var_4272_begin_0 = const()[name = tensor<string, []>("op_4272_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_4272_end_0 = const()[name = tensor<string, []>("op_4272_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_4272_end_mask_0 = const()[name = tensor<string, []>("op_4272_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4272_cast = slice_by_index(begin = var_4272_begin_0, end = var_4272_end_0, end_mask = var_4272_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4272_cast")];
            tensor<int32, [4]> var_4276_begin_0 = const()[name = tensor<string, []>("op_4276_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_4276_end_0 = const()[name = tensor<string, []>("op_4276_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_4276_end_mask_0 = const()[name = tensor<string, []>("op_4276_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4276_cast = slice_by_index(begin = var_4276_begin_0, end = var_4276_end_0, end_mask = var_4276_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4276_cast")];
            tensor<int32, [4]> var_4280_begin_0 = const()[name = tensor<string, []>("op_4280_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_4280_end_0 = const()[name = tensor<string, []>("op_4280_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_4280_end_mask_0 = const()[name = tensor<string, []>("op_4280_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4280_cast = slice_by_index(begin = var_4280_begin_0, end = var_4280_end_0, end_mask = var_4280_end_mask_0, x = transpose_128)[name = tensor<string, []>("op_4280_cast")];
            tensor<int32, [4]> var_4282_begin_0 = const()[name = tensor<string, []>("op_4282_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4282_end_0 = const()[name = tensor<string, []>("op_4282_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_4282_end_mask_0 = const()[name = tensor<string, []>("op_4282_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4282_cast = slice_by_index(begin = var_4282_begin_0, end = var_4282_end_0, end_mask = var_4282_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4282_cast")];
            tensor<int32, [4]> var_4286_begin_0 = const()[name = tensor<string, []>("op_4286_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_4286_end_0 = const()[name = tensor<string, []>("op_4286_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_4286_end_mask_0 = const()[name = tensor<string, []>("op_4286_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4286_cast = slice_by_index(begin = var_4286_begin_0, end = var_4286_end_0, end_mask = var_4286_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4286_cast")];
            tensor<int32, [4]> var_4290_begin_0 = const()[name = tensor<string, []>("op_4290_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_4290_end_0 = const()[name = tensor<string, []>("op_4290_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_4290_end_mask_0 = const()[name = tensor<string, []>("op_4290_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4290_cast = slice_by_index(begin = var_4290_begin_0, end = var_4290_end_0, end_mask = var_4290_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4290_cast")];
            tensor<int32, [4]> var_4294_begin_0 = const()[name = tensor<string, []>("op_4294_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_4294_end_0 = const()[name = tensor<string, []>("op_4294_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_4294_end_mask_0 = const()[name = tensor<string, []>("op_4294_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4294_cast = slice_by_index(begin = var_4294_begin_0, end = var_4294_end_0, end_mask = var_4294_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4294_cast")];
            tensor<int32, [4]> var_4298_begin_0 = const()[name = tensor<string, []>("op_4298_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_4298_end_0 = const()[name = tensor<string, []>("op_4298_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_4298_end_mask_0 = const()[name = tensor<string, []>("op_4298_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4298_cast = slice_by_index(begin = var_4298_begin_0, end = var_4298_end_0, end_mask = var_4298_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4298_cast")];
            tensor<int32, [4]> var_4302_begin_0 = const()[name = tensor<string, []>("op_4302_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_4302_end_0 = const()[name = tensor<string, []>("op_4302_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_4302_end_mask_0 = const()[name = tensor<string, []>("op_4302_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4302_cast = slice_by_index(begin = var_4302_begin_0, end = var_4302_end_0, end_mask = var_4302_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4302_cast")];
            tensor<int32, [4]> var_4306_begin_0 = const()[name = tensor<string, []>("op_4306_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_4306_end_0 = const()[name = tensor<string, []>("op_4306_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_4306_end_mask_0 = const()[name = tensor<string, []>("op_4306_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4306_cast = slice_by_index(begin = var_4306_begin_0, end = var_4306_end_0, end_mask = var_4306_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4306_cast")];
            tensor<int32, [4]> var_4310_begin_0 = const()[name = tensor<string, []>("op_4310_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_4310_end_0 = const()[name = tensor<string, []>("op_4310_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_4310_end_mask_0 = const()[name = tensor<string, []>("op_4310_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4310_cast = slice_by_index(begin = var_4310_begin_0, end = var_4310_end_0, end_mask = var_4310_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4310_cast")];
            tensor<int32, [4]> var_4314_begin_0 = const()[name = tensor<string, []>("op_4314_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_4314_end_0 = const()[name = tensor<string, []>("op_4314_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_4314_end_mask_0 = const()[name = tensor<string, []>("op_4314_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4314_cast = slice_by_index(begin = var_4314_begin_0, end = var_4314_end_0, end_mask = var_4314_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4314_cast")];
            tensor<int32, [4]> var_4318_begin_0 = const()[name = tensor<string, []>("op_4318_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_4318_end_0 = const()[name = tensor<string, []>("op_4318_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_4318_end_mask_0 = const()[name = tensor<string, []>("op_4318_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4318_cast = slice_by_index(begin = var_4318_begin_0, end = var_4318_end_0, end_mask = var_4318_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4318_cast")];
            tensor<int32, [4]> var_4322_begin_0 = const()[name = tensor<string, []>("op_4322_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_4322_end_0 = const()[name = tensor<string, []>("op_4322_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_4322_end_mask_0 = const()[name = tensor<string, []>("op_4322_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4322_cast = slice_by_index(begin = var_4322_begin_0, end = var_4322_end_0, end_mask = var_4322_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4322_cast")];
            tensor<int32, [4]> var_4326_begin_0 = const()[name = tensor<string, []>("op_4326_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_4326_end_0 = const()[name = tensor<string, []>("op_4326_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_4326_end_mask_0 = const()[name = tensor<string, []>("op_4326_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4326_cast = slice_by_index(begin = var_4326_begin_0, end = var_4326_end_0, end_mask = var_4326_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4326_cast")];
            tensor<int32, [4]> var_4330_begin_0 = const()[name = tensor<string, []>("op_4330_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_4330_end_0 = const()[name = tensor<string, []>("op_4330_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_4330_end_mask_0 = const()[name = tensor<string, []>("op_4330_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4330_cast = slice_by_index(begin = var_4330_begin_0, end = var_4330_end_0, end_mask = var_4330_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4330_cast")];
            tensor<int32, [4]> var_4334_begin_0 = const()[name = tensor<string, []>("op_4334_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_4334_end_0 = const()[name = tensor<string, []>("op_4334_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_4334_end_mask_0 = const()[name = tensor<string, []>("op_4334_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4334_cast = slice_by_index(begin = var_4334_begin_0, end = var_4334_end_0, end_mask = var_4334_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4334_cast")];
            tensor<int32, [4]> var_4338_begin_0 = const()[name = tensor<string, []>("op_4338_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_4338_end_0 = const()[name = tensor<string, []>("op_4338_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_4338_end_mask_0 = const()[name = tensor<string, []>("op_4338_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4338_cast = slice_by_index(begin = var_4338_begin_0, end = var_4338_end_0, end_mask = var_4338_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4338_cast")];
            tensor<int32, [4]> var_4342_begin_0 = const()[name = tensor<string, []>("op_4342_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_4342_end_0 = const()[name = tensor<string, []>("op_4342_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_4342_end_mask_0 = const()[name = tensor<string, []>("op_4342_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4342_cast = slice_by_index(begin = var_4342_begin_0, end = var_4342_end_0, end_mask = var_4342_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4342_cast")];
            tensor<int32, [4]> var_4346_begin_0 = const()[name = tensor<string, []>("op_4346_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_4346_end_0 = const()[name = tensor<string, []>("op_4346_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_4346_end_mask_0 = const()[name = tensor<string, []>("op_4346_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4346_cast = slice_by_index(begin = var_4346_begin_0, end = var_4346_end_0, end_mask = var_4346_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4346_cast")];
            tensor<int32, [4]> var_4350_begin_0 = const()[name = tensor<string, []>("op_4350_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_4350_end_0 = const()[name = tensor<string, []>("op_4350_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_4350_end_mask_0 = const()[name = tensor<string, []>("op_4350_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4350_cast = slice_by_index(begin = var_4350_begin_0, end = var_4350_end_0, end_mask = var_4350_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4350_cast")];
            tensor<int32, [4]> var_4354_begin_0 = const()[name = tensor<string, []>("op_4354_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_4354_end_0 = const()[name = tensor<string, []>("op_4354_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_4354_end_mask_0 = const()[name = tensor<string, []>("op_4354_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4354_cast = slice_by_index(begin = var_4354_begin_0, end = var_4354_end_0, end_mask = var_4354_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4354_cast")];
            tensor<int32, [4]> var_4358_begin_0 = const()[name = tensor<string, []>("op_4358_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_4358_end_0 = const()[name = tensor<string, []>("op_4358_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_4358_end_mask_0 = const()[name = tensor<string, []>("op_4358_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4358_cast = slice_by_index(begin = var_4358_begin_0, end = var_4358_end_0, end_mask = var_4358_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_4358_cast")];
            tensor<string, []> var_4362_equation_0 = const()[name = tensor<string, []>("op_4362_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4362_cast = einsum(equation = var_4362_equation_0, values = (var_4204_cast, var_4121_cast))[name = tensor<string, []>("op_4362_cast")];
            tensor<fp16, []> var_4363_to_fp16 = const()[name = tensor<string, []>("op_4363_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_281_cast = mul(x = var_4362_cast, y = var_4363_to_fp16)[name = tensor<string, []>("aw_281_cast")];
            tensor<string, []> var_4366_equation_0 = const()[name = tensor<string, []>("op_4366_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4366_cast = einsum(equation = var_4366_equation_0, values = (var_4208_cast, var_4125_cast))[name = tensor<string, []>("op_4366_cast")];
            tensor<fp16, []> var_4367_to_fp16 = const()[name = tensor<string, []>("op_4367_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_283_cast = mul(x = var_4366_cast, y = var_4367_to_fp16)[name = tensor<string, []>("aw_283_cast")];
            tensor<string, []> var_4370_equation_0 = const()[name = tensor<string, []>("op_4370_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4370_cast = einsum(equation = var_4370_equation_0, values = (var_4212_cast, var_4129_cast))[name = tensor<string, []>("op_4370_cast")];
            tensor<fp16, []> var_4371_to_fp16 = const()[name = tensor<string, []>("op_4371_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_285_cast = mul(x = var_4370_cast, y = var_4371_to_fp16)[name = tensor<string, []>("aw_285_cast")];
            tensor<string, []> var_4374_equation_0 = const()[name = tensor<string, []>("op_4374_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4374_cast = einsum(equation = var_4374_equation_0, values = (var_4216_cast, var_4133_cast))[name = tensor<string, []>("op_4374_cast")];
            tensor<fp16, []> var_4375_to_fp16 = const()[name = tensor<string, []>("op_4375_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_287_cast = mul(x = var_4374_cast, y = var_4375_to_fp16)[name = tensor<string, []>("aw_287_cast")];
            tensor<string, []> var_4378_equation_0 = const()[name = tensor<string, []>("op_4378_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4378_cast = einsum(equation = var_4378_equation_0, values = (var_4220_cast, var_4137_cast))[name = tensor<string, []>("op_4378_cast")];
            tensor<fp16, []> var_4379_to_fp16 = const()[name = tensor<string, []>("op_4379_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_289_cast = mul(x = var_4378_cast, y = var_4379_to_fp16)[name = tensor<string, []>("aw_289_cast")];
            tensor<string, []> var_4382_equation_0 = const()[name = tensor<string, []>("op_4382_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4382_cast = einsum(equation = var_4382_equation_0, values = (var_4224_cast, var_4141_cast))[name = tensor<string, []>("op_4382_cast")];
            tensor<fp16, []> var_4383_to_fp16 = const()[name = tensor<string, []>("op_4383_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_291_cast = mul(x = var_4382_cast, y = var_4383_to_fp16)[name = tensor<string, []>("aw_291_cast")];
            tensor<string, []> var_4386_equation_0 = const()[name = tensor<string, []>("op_4386_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4386_cast = einsum(equation = var_4386_equation_0, values = (var_4228_cast, var_4145_cast))[name = tensor<string, []>("op_4386_cast")];
            tensor<fp16, []> var_4387_to_fp16 = const()[name = tensor<string, []>("op_4387_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_293_cast = mul(x = var_4386_cast, y = var_4387_to_fp16)[name = tensor<string, []>("aw_293_cast")];
            tensor<string, []> var_4390_equation_0 = const()[name = tensor<string, []>("op_4390_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4390_cast = einsum(equation = var_4390_equation_0, values = (var_4232_cast, var_4149_cast))[name = tensor<string, []>("op_4390_cast")];
            tensor<fp16, []> var_4391_to_fp16 = const()[name = tensor<string, []>("op_4391_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_295_cast = mul(x = var_4390_cast, y = var_4391_to_fp16)[name = tensor<string, []>("aw_295_cast")];
            tensor<string, []> var_4394_equation_0 = const()[name = tensor<string, []>("op_4394_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4394_cast = einsum(equation = var_4394_equation_0, values = (var_4236_cast, var_4153_cast))[name = tensor<string, []>("op_4394_cast")];
            tensor<fp16, []> var_4395_to_fp16 = const()[name = tensor<string, []>("op_4395_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_297_cast = mul(x = var_4394_cast, y = var_4395_to_fp16)[name = tensor<string, []>("aw_297_cast")];
            tensor<string, []> var_4398_equation_0 = const()[name = tensor<string, []>("op_4398_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4398_cast = einsum(equation = var_4398_equation_0, values = (var_4240_cast, var_4157_cast))[name = tensor<string, []>("op_4398_cast")];
            tensor<fp16, []> var_4399_to_fp16 = const()[name = tensor<string, []>("op_4399_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_299_cast = mul(x = var_4398_cast, y = var_4399_to_fp16)[name = tensor<string, []>("aw_299_cast")];
            tensor<string, []> var_4402_equation_0 = const()[name = tensor<string, []>("op_4402_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4402_cast = einsum(equation = var_4402_equation_0, values = (var_4244_cast, var_4161_cast))[name = tensor<string, []>("op_4402_cast")];
            tensor<fp16, []> var_4403_to_fp16 = const()[name = tensor<string, []>("op_4403_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_301_cast = mul(x = var_4402_cast, y = var_4403_to_fp16)[name = tensor<string, []>("aw_301_cast")];
            tensor<string, []> var_4406_equation_0 = const()[name = tensor<string, []>("op_4406_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4406_cast = einsum(equation = var_4406_equation_0, values = (var_4248_cast, var_4165_cast))[name = tensor<string, []>("op_4406_cast")];
            tensor<fp16, []> var_4407_to_fp16 = const()[name = tensor<string, []>("op_4407_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_303_cast = mul(x = var_4406_cast, y = var_4407_to_fp16)[name = tensor<string, []>("aw_303_cast")];
            tensor<string, []> var_4410_equation_0 = const()[name = tensor<string, []>("op_4410_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4410_cast = einsum(equation = var_4410_equation_0, values = (var_4252_cast, var_4169_cast))[name = tensor<string, []>("op_4410_cast")];
            tensor<fp16, []> var_4411_to_fp16 = const()[name = tensor<string, []>("op_4411_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_305_cast = mul(x = var_4410_cast, y = var_4411_to_fp16)[name = tensor<string, []>("aw_305_cast")];
            tensor<string, []> var_4414_equation_0 = const()[name = tensor<string, []>("op_4414_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4414_cast = einsum(equation = var_4414_equation_0, values = (var_4256_cast, var_4173_cast))[name = tensor<string, []>("op_4414_cast")];
            tensor<fp16, []> var_4415_to_fp16 = const()[name = tensor<string, []>("op_4415_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_307_cast = mul(x = var_4414_cast, y = var_4415_to_fp16)[name = tensor<string, []>("aw_307_cast")];
            tensor<string, []> var_4418_equation_0 = const()[name = tensor<string, []>("op_4418_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4418_cast = einsum(equation = var_4418_equation_0, values = (var_4260_cast, var_4177_cast))[name = tensor<string, []>("op_4418_cast")];
            tensor<fp16, []> var_4419_to_fp16 = const()[name = tensor<string, []>("op_4419_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_309_cast = mul(x = var_4418_cast, y = var_4419_to_fp16)[name = tensor<string, []>("aw_309_cast")];
            tensor<string, []> var_4422_equation_0 = const()[name = tensor<string, []>("op_4422_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4422_cast = einsum(equation = var_4422_equation_0, values = (var_4264_cast, var_4181_cast))[name = tensor<string, []>("op_4422_cast")];
            tensor<fp16, []> var_4423_to_fp16 = const()[name = tensor<string, []>("op_4423_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_311_cast = mul(x = var_4422_cast, y = var_4423_to_fp16)[name = tensor<string, []>("aw_311_cast")];
            tensor<string, []> var_4426_equation_0 = const()[name = tensor<string, []>("op_4426_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4426_cast = einsum(equation = var_4426_equation_0, values = (var_4268_cast, var_4185_cast))[name = tensor<string, []>("op_4426_cast")];
            tensor<fp16, []> var_4427_to_fp16 = const()[name = tensor<string, []>("op_4427_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_313_cast = mul(x = var_4426_cast, y = var_4427_to_fp16)[name = tensor<string, []>("aw_313_cast")];
            tensor<string, []> var_4430_equation_0 = const()[name = tensor<string, []>("op_4430_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4430_cast = einsum(equation = var_4430_equation_0, values = (var_4272_cast, var_4189_cast))[name = tensor<string, []>("op_4430_cast")];
            tensor<fp16, []> var_4431_to_fp16 = const()[name = tensor<string, []>("op_4431_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_315_cast = mul(x = var_4430_cast, y = var_4431_to_fp16)[name = tensor<string, []>("aw_315_cast")];
            tensor<string, []> var_4434_equation_0 = const()[name = tensor<string, []>("op_4434_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4434_cast = einsum(equation = var_4434_equation_0, values = (var_4276_cast, var_4193_cast))[name = tensor<string, []>("op_4434_cast")];
            tensor<fp16, []> var_4435_to_fp16 = const()[name = tensor<string, []>("op_4435_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_317_cast = mul(x = var_4434_cast, y = var_4435_to_fp16)[name = tensor<string, []>("aw_317_cast")];
            tensor<string, []> var_4438_equation_0 = const()[name = tensor<string, []>("op_4438_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_4438_cast = einsum(equation = var_4438_equation_0, values = (var_4280_cast, var_4197_cast))[name = tensor<string, []>("op_4438_cast")];
            tensor<fp16, []> var_4439_to_fp16 = const()[name = tensor<string, []>("op_4439_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_319_cast = mul(x = var_4438_cast, y = var_4439_to_fp16)[name = tensor<string, []>("aw_319_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4441_cast = softmax(axis = var_2634, x = aw_281_cast)[name = tensor<string, []>("op_4441_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4442_cast = softmax(axis = var_2634, x = aw_283_cast)[name = tensor<string, []>("op_4442_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4443_cast = softmax(axis = var_2634, x = aw_285_cast)[name = tensor<string, []>("op_4443_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4444_cast = softmax(axis = var_2634, x = aw_287_cast)[name = tensor<string, []>("op_4444_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4445_cast = softmax(axis = var_2634, x = aw_289_cast)[name = tensor<string, []>("op_4445_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4446_cast = softmax(axis = var_2634, x = aw_291_cast)[name = tensor<string, []>("op_4446_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4447_cast = softmax(axis = var_2634, x = aw_293_cast)[name = tensor<string, []>("op_4447_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4448_cast = softmax(axis = var_2634, x = aw_295_cast)[name = tensor<string, []>("op_4448_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4449_cast = softmax(axis = var_2634, x = aw_297_cast)[name = tensor<string, []>("op_4449_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4450_cast = softmax(axis = var_2634, x = aw_299_cast)[name = tensor<string, []>("op_4450_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4451_cast = softmax(axis = var_2634, x = aw_301_cast)[name = tensor<string, []>("op_4451_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4452_cast = softmax(axis = var_2634, x = aw_303_cast)[name = tensor<string, []>("op_4452_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4453_cast = softmax(axis = var_2634, x = aw_305_cast)[name = tensor<string, []>("op_4453_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4454_cast = softmax(axis = var_2634, x = aw_307_cast)[name = tensor<string, []>("op_4454_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4455_cast = softmax(axis = var_2634, x = aw_309_cast)[name = tensor<string, []>("op_4455_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4456_cast = softmax(axis = var_2634, x = aw_311_cast)[name = tensor<string, []>("op_4456_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4457_cast = softmax(axis = var_2634, x = aw_313_cast)[name = tensor<string, []>("op_4457_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4458_cast = softmax(axis = var_2634, x = aw_315_cast)[name = tensor<string, []>("op_4458_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4459_cast = softmax(axis = var_2634, x = aw_317_cast)[name = tensor<string, []>("op_4459_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_4460_cast = softmax(axis = var_2634, x = aw_319_cast)[name = tensor<string, []>("op_4460_cast")];
            tensor<string, []> var_4462_equation_0 = const()[name = tensor<string, []>("op_4462_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4462_cast = einsum(equation = var_4462_equation_0, values = (var_4282_cast, var_4441_cast))[name = tensor<string, []>("op_4462_cast")];
            tensor<string, []> var_4464_equation_0 = const()[name = tensor<string, []>("op_4464_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4464_cast = einsum(equation = var_4464_equation_0, values = (var_4286_cast, var_4442_cast))[name = tensor<string, []>("op_4464_cast")];
            tensor<string, []> var_4466_equation_0 = const()[name = tensor<string, []>("op_4466_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4466_cast = einsum(equation = var_4466_equation_0, values = (var_4290_cast, var_4443_cast))[name = tensor<string, []>("op_4466_cast")];
            tensor<string, []> var_4468_equation_0 = const()[name = tensor<string, []>("op_4468_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4468_cast = einsum(equation = var_4468_equation_0, values = (var_4294_cast, var_4444_cast))[name = tensor<string, []>("op_4468_cast")];
            tensor<string, []> var_4470_equation_0 = const()[name = tensor<string, []>("op_4470_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4470_cast = einsum(equation = var_4470_equation_0, values = (var_4298_cast, var_4445_cast))[name = tensor<string, []>("op_4470_cast")];
            tensor<string, []> var_4472_equation_0 = const()[name = tensor<string, []>("op_4472_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4472_cast = einsum(equation = var_4472_equation_0, values = (var_4302_cast, var_4446_cast))[name = tensor<string, []>("op_4472_cast")];
            tensor<string, []> var_4474_equation_0 = const()[name = tensor<string, []>("op_4474_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4474_cast = einsum(equation = var_4474_equation_0, values = (var_4306_cast, var_4447_cast))[name = tensor<string, []>("op_4474_cast")];
            tensor<string, []> var_4476_equation_0 = const()[name = tensor<string, []>("op_4476_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4476_cast = einsum(equation = var_4476_equation_0, values = (var_4310_cast, var_4448_cast))[name = tensor<string, []>("op_4476_cast")];
            tensor<string, []> var_4478_equation_0 = const()[name = tensor<string, []>("op_4478_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4478_cast = einsum(equation = var_4478_equation_0, values = (var_4314_cast, var_4449_cast))[name = tensor<string, []>("op_4478_cast")];
            tensor<string, []> var_4480_equation_0 = const()[name = tensor<string, []>("op_4480_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4480_cast = einsum(equation = var_4480_equation_0, values = (var_4318_cast, var_4450_cast))[name = tensor<string, []>("op_4480_cast")];
            tensor<string, []> var_4482_equation_0 = const()[name = tensor<string, []>("op_4482_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4482_cast = einsum(equation = var_4482_equation_0, values = (var_4322_cast, var_4451_cast))[name = tensor<string, []>("op_4482_cast")];
            tensor<string, []> var_4484_equation_0 = const()[name = tensor<string, []>("op_4484_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4484_cast = einsum(equation = var_4484_equation_0, values = (var_4326_cast, var_4452_cast))[name = tensor<string, []>("op_4484_cast")];
            tensor<string, []> var_4486_equation_0 = const()[name = tensor<string, []>("op_4486_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4486_cast = einsum(equation = var_4486_equation_0, values = (var_4330_cast, var_4453_cast))[name = tensor<string, []>("op_4486_cast")];
            tensor<string, []> var_4488_equation_0 = const()[name = tensor<string, []>("op_4488_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4488_cast = einsum(equation = var_4488_equation_0, values = (var_4334_cast, var_4454_cast))[name = tensor<string, []>("op_4488_cast")];
            tensor<string, []> var_4490_equation_0 = const()[name = tensor<string, []>("op_4490_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4490_cast = einsum(equation = var_4490_equation_0, values = (var_4338_cast, var_4455_cast))[name = tensor<string, []>("op_4490_cast")];
            tensor<string, []> var_4492_equation_0 = const()[name = tensor<string, []>("op_4492_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4492_cast = einsum(equation = var_4492_equation_0, values = (var_4342_cast, var_4456_cast))[name = tensor<string, []>("op_4492_cast")];
            tensor<string, []> var_4494_equation_0 = const()[name = tensor<string, []>("op_4494_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4494_cast = einsum(equation = var_4494_equation_0, values = (var_4346_cast, var_4457_cast))[name = tensor<string, []>("op_4494_cast")];
            tensor<string, []> var_4496_equation_0 = const()[name = tensor<string, []>("op_4496_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4496_cast = einsum(equation = var_4496_equation_0, values = (var_4350_cast, var_4458_cast))[name = tensor<string, []>("op_4496_cast")];
            tensor<string, []> var_4498_equation_0 = const()[name = tensor<string, []>("op_4498_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4498_cast = einsum(equation = var_4498_equation_0, values = (var_4354_cast, var_4459_cast))[name = tensor<string, []>("op_4498_cast")];
            tensor<string, []> var_4500_equation_0 = const()[name = tensor<string, []>("op_4500_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4500_cast = einsum(equation = var_4500_equation_0, values = (var_4358_cast, var_4460_cast))[name = tensor<string, []>("op_4500_cast")];
            tensor<bool, []> input_145_interleave_0 = const()[name = tensor<string, []>("input_145_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_145_cast = concat(axis = var_2634, interleave = input_145_interleave_0, values = (var_4462_cast, var_4464_cast, var_4466_cast, var_4468_cast, var_4470_cast, var_4472_cast, var_4474_cast, var_4476_cast, var_4478_cast, var_4480_cast, var_4482_cast, var_4484_cast, var_4486_cast, var_4488_cast, var_4490_cast, var_4492_cast, var_4494_cast, var_4496_cast, var_4498_cast, var_4500_cast))[name = tensor<string, []>("input_145_cast")];
            tensor<int32, [2]> var_4506 = const()[name = tensor<string, []>("op_4506"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4508 = const()[name = tensor<string, []>("op_4508"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_4510_pad_type_0 = const()[name = tensor<string, []>("op_4510_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_4510_pad_0 = const()[name = tensor<string, []>("op_4510_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(294562176)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297839040)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_4510_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_4508, groups = var_2634, pad = var_4510_pad_0, pad_type = var_4510_pad_type_0, strides = var_4506, weight = down_blocks_2_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_145_cast)[name = tensor<string, []>("op_4510_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_35_cast = add(x = var_4510_cast, y = inputs_33_cast)[name = tensor<string, []>("inputs_35_cast")];
            tensor<int32, [1]> var_4514 = const()[name = tensor<string, []>("op_4514"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_35_cast = reduce_mean(axes = var_4514, keep_dims = var_2629, x = inputs_35_cast)[name = tensor<string, []>("channels_mean_35_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_35_cast = sub(x = inputs_35_cast, y = channels_mean_35_cast)[name = tensor<string, []>("zero_mean_35_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_35_cast = mul(x = zero_mean_35_cast, y = zero_mean_35_cast)[name = tensor<string, []>("zero_mean_sq_35_cast")];
            tensor<int32, [1]> var_4518 = const()[name = tensor<string, []>("op_4518"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_4519_cast = reduce_mean(axes = var_4518, keep_dims = var_2629, x = zero_mean_sq_35_cast)[name = tensor<string, []>("op_4519_cast")];
            tensor<fp16, []> var_4520_to_fp16 = const()[name = tensor<string, []>("op_4520_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_4521_cast = add(x = var_4519_cast, y = var_4520_to_fp16)[name = tensor<string, []>("op_4521_cast")];
            tensor<fp16, []> denom_35_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_35_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_35_cast = rsqrt(epsilon = denom_35_epsilon_0_to_fp16, x = var_4521_cast)[name = tensor<string, []>("denom_35_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_35_cast = mul(x = zero_mean_35_cast, y = denom_35_cast)[name = tensor<string, []>("out_35_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_4525_to_fp16 = const()[name = tensor<string, []>("op_4525_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297841664)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_4526_cast = add(x = out_35_cast, y = var_4525_to_fp16)[name = tensor<string, []>("op_4526_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_4528_to_fp16 = const()[name = tensor<string, []>("op_4528_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297844288)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_147_cast = mul(x = var_4526_cast, y = var_4528_to_fp16)[name = tensor<string, []>("input_147_cast")];
            tensor<int32, [2]> var_4536 = const()[name = tensor<string, []>("op_4536"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4538 = const()[name = tensor<string, []>("op_4538"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_4540_pad_type_0 = const()[name = tensor<string, []>("op_4540_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_4540_pad_0 = const()[name = tensor<string, []>("op_4540_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297846912)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(324061376)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_4540_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_4538, groups = var_2634, pad = var_4540_pad_0, pad_type = var_4540_pad_type_0, strides = var_4536, weight = down_blocks_2_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_147_cast)[name = tensor<string, []>("op_4540_cast")];
            tensor<int32, [2]> var_4541_split_sizes_0 = const()[name = tensor<string, []>("op_4541_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_4541_axis_0 = const()[name = tensor<string, []>("op_4541_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_4541_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_4541_cast_1 = split(axis = var_4541_axis_0, split_sizes = var_4541_split_sizes_0, x = var_4540_cast)[name = tensor<string, []>("op_4541_cast")];
            tensor<string, []> var_4543_mode_0 = const()[name = tensor<string, []>("op_4543_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_4543_cast = gelu(mode = var_4543_mode_0, x = var_4541_cast_1)[name = tensor<string, []>("op_4543_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_149_cast = mul(x = var_4541_cast_0, y = var_4543_cast)[name = tensor<string, []>("input_149_cast")];
            tensor<int32, [2]> var_4547 = const()[name = tensor<string, []>("op_4547"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4549 = const()[name = tensor<string, []>("op_4549"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_4551_pad_type_0 = const()[name = tensor<string, []>("op_4551_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_4551_pad_0 = const()[name = tensor<string, []>("op_4551_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(324081920)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(337189184)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_4551_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_4549, groups = var_2634, pad = var_4551_pad_0, pad_type = var_4551_pad_type_0, strides = var_4547, weight = down_blocks_2_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_149_cast)[name = tensor<string, []>("op_4551_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_37_cast = add(x = var_4551_cast, y = inputs_35_cast)[name = tensor<string, []>("inputs_37_cast")];
            tensor<int32, [1]> var_4561 = const()[name = tensor<string, []>("op_4561"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_37_cast = reduce_mean(axes = var_4561, keep_dims = var_2629, x = inputs_37_cast)[name = tensor<string, []>("channels_mean_37_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_37_cast = sub(x = inputs_37_cast, y = channels_mean_37_cast)[name = tensor<string, []>("zero_mean_37_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_37_cast = mul(x = zero_mean_37_cast, y = zero_mean_37_cast)[name = tensor<string, []>("zero_mean_sq_37_cast")];
            tensor<int32, [1]> var_4565 = const()[name = tensor<string, []>("op_4565"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_4566_cast = reduce_mean(axes = var_4565, keep_dims = var_2629, x = zero_mean_sq_37_cast)[name = tensor<string, []>("op_4566_cast")];
            tensor<fp16, []> var_4567_to_fp16 = const()[name = tensor<string, []>("op_4567_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_4568_cast = add(x = var_4566_cast, y = var_4567_to_fp16)[name = tensor<string, []>("op_4568_cast")];
            tensor<fp16, []> denom_37_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_37_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_37_cast = rsqrt(epsilon = denom_37_epsilon_0_to_fp16, x = var_4568_cast)[name = tensor<string, []>("denom_37_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_37_cast = mul(x = zero_mean_37_cast, y = denom_37_cast)[name = tensor<string, []>("out_37_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_4572_to_fp16 = const()[name = tensor<string, []>("op_4572_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(337191808)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_4573_cast = add(x = out_37_cast, y = var_4572_to_fp16)[name = tensor<string, []>("op_4573_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_4575_to_fp16 = const()[name = tensor<string, []>("op_4575_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(337194432)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_77_cast = mul(x = var_4573_cast, y = var_4575_to_fp16)[name = tensor<string, []>("hidden_states_77_cast")];
            tensor<int32, [2]> var_4582 = const()[name = tensor<string, []>("op_4582"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4584 = const()[name = tensor<string, []>("op_4584"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_25_pad_type_0 = const()[name = tensor<string, []>("q_25_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_25_pad_0 = const()[name = tensor<string, []>("q_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(337197056)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_25_cast = conv(dilations = var_4584, groups = var_2634, pad = q_25_pad_0, pad_type = q_25_pad_type_0, strides = var_4582, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16, x = hidden_states_77_cast)[name = tensor<string, []>("q_25_cast")];
            tensor<int32, [2]> var_4588 = const()[name = tensor<string, []>("op_4588"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4590 = const()[name = tensor<string, []>("op_4590"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_49_pad_type_0 = const()[name = tensor<string, []>("k_49_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_49_pad_0 = const()[name = tensor<string, []>("k_49_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(340473920)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_49_cast = conv(dilations = var_4590, groups = var_2634, pad = k_49_pad_0, pad_type = k_49_pad_type_0, strides = var_4588, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16, x = hidden_states_77_cast)[name = tensor<string, []>("k_49_cast")];
            tensor<int32, [2]> var_4594 = const()[name = tensor<string, []>("op_4594"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4596 = const()[name = tensor<string, []>("op_4596"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_25_pad_type_0 = const()[name = tensor<string, []>("v_25_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_25_pad_0 = const()[name = tensor<string, []>("v_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(343750784)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_25_cast = conv(dilations = var_4596, groups = var_2634, pad = v_25_pad_0, pad_type = v_25_pad_type_0, strides = var_4594, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16, x = hidden_states_77_cast)[name = tensor<string, []>("v_25_cast")];
            tensor<int32, [4]> var_4600_begin_0 = const()[name = tensor<string, []>("op_4600_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4600_end_0 = const()[name = tensor<string, []>("op_4600_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_4600_end_mask_0 = const()[name = tensor<string, []>("op_4600_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4600_cast = slice_by_index(begin = var_4600_begin_0, end = var_4600_end_0, end_mask = var_4600_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4600_cast")];
            tensor<int32, [4]> var_4604_begin_0 = const()[name = tensor<string, []>("op_4604_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_4604_end_0 = const()[name = tensor<string, []>("op_4604_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_4604_end_mask_0 = const()[name = tensor<string, []>("op_4604_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4604_cast = slice_by_index(begin = var_4604_begin_0, end = var_4604_end_0, end_mask = var_4604_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4604_cast")];
            tensor<int32, [4]> var_4608_begin_0 = const()[name = tensor<string, []>("op_4608_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_4608_end_0 = const()[name = tensor<string, []>("op_4608_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_4608_end_mask_0 = const()[name = tensor<string, []>("op_4608_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4608_cast = slice_by_index(begin = var_4608_begin_0, end = var_4608_end_0, end_mask = var_4608_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4608_cast")];
            tensor<int32, [4]> var_4612_begin_0 = const()[name = tensor<string, []>("op_4612_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_4612_end_0 = const()[name = tensor<string, []>("op_4612_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_4612_end_mask_0 = const()[name = tensor<string, []>("op_4612_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4612_cast = slice_by_index(begin = var_4612_begin_0, end = var_4612_end_0, end_mask = var_4612_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4612_cast")];
            tensor<int32, [4]> var_4616_begin_0 = const()[name = tensor<string, []>("op_4616_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_4616_end_0 = const()[name = tensor<string, []>("op_4616_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_4616_end_mask_0 = const()[name = tensor<string, []>("op_4616_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4616_cast = slice_by_index(begin = var_4616_begin_0, end = var_4616_end_0, end_mask = var_4616_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4616_cast")];
            tensor<int32, [4]> var_4620_begin_0 = const()[name = tensor<string, []>("op_4620_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_4620_end_0 = const()[name = tensor<string, []>("op_4620_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_4620_end_mask_0 = const()[name = tensor<string, []>("op_4620_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4620_cast = slice_by_index(begin = var_4620_begin_0, end = var_4620_end_0, end_mask = var_4620_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4620_cast")];
            tensor<int32, [4]> var_4624_begin_0 = const()[name = tensor<string, []>("op_4624_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_4624_end_0 = const()[name = tensor<string, []>("op_4624_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_4624_end_mask_0 = const()[name = tensor<string, []>("op_4624_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4624_cast = slice_by_index(begin = var_4624_begin_0, end = var_4624_end_0, end_mask = var_4624_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4624_cast")];
            tensor<int32, [4]> var_4628_begin_0 = const()[name = tensor<string, []>("op_4628_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_4628_end_0 = const()[name = tensor<string, []>("op_4628_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_4628_end_mask_0 = const()[name = tensor<string, []>("op_4628_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4628_cast = slice_by_index(begin = var_4628_begin_0, end = var_4628_end_0, end_mask = var_4628_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4628_cast")];
            tensor<int32, [4]> var_4632_begin_0 = const()[name = tensor<string, []>("op_4632_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_4632_end_0 = const()[name = tensor<string, []>("op_4632_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_4632_end_mask_0 = const()[name = tensor<string, []>("op_4632_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4632_cast = slice_by_index(begin = var_4632_begin_0, end = var_4632_end_0, end_mask = var_4632_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4632_cast")];
            tensor<int32, [4]> var_4636_begin_0 = const()[name = tensor<string, []>("op_4636_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_4636_end_0 = const()[name = tensor<string, []>("op_4636_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_4636_end_mask_0 = const()[name = tensor<string, []>("op_4636_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4636_cast = slice_by_index(begin = var_4636_begin_0, end = var_4636_end_0, end_mask = var_4636_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4636_cast")];
            tensor<int32, [4]> var_4640_begin_0 = const()[name = tensor<string, []>("op_4640_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_4640_end_0 = const()[name = tensor<string, []>("op_4640_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_4640_end_mask_0 = const()[name = tensor<string, []>("op_4640_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4640_cast = slice_by_index(begin = var_4640_begin_0, end = var_4640_end_0, end_mask = var_4640_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4640_cast")];
            tensor<int32, [4]> var_4644_begin_0 = const()[name = tensor<string, []>("op_4644_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_4644_end_0 = const()[name = tensor<string, []>("op_4644_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_4644_end_mask_0 = const()[name = tensor<string, []>("op_4644_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4644_cast = slice_by_index(begin = var_4644_begin_0, end = var_4644_end_0, end_mask = var_4644_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4644_cast")];
            tensor<int32, [4]> var_4648_begin_0 = const()[name = tensor<string, []>("op_4648_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_4648_end_0 = const()[name = tensor<string, []>("op_4648_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_4648_end_mask_0 = const()[name = tensor<string, []>("op_4648_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4648_cast = slice_by_index(begin = var_4648_begin_0, end = var_4648_end_0, end_mask = var_4648_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4648_cast")];
            tensor<int32, [4]> var_4652_begin_0 = const()[name = tensor<string, []>("op_4652_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_4652_end_0 = const()[name = tensor<string, []>("op_4652_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_4652_end_mask_0 = const()[name = tensor<string, []>("op_4652_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4652_cast = slice_by_index(begin = var_4652_begin_0, end = var_4652_end_0, end_mask = var_4652_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4652_cast")];
            tensor<int32, [4]> var_4656_begin_0 = const()[name = tensor<string, []>("op_4656_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_4656_end_0 = const()[name = tensor<string, []>("op_4656_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_4656_end_mask_0 = const()[name = tensor<string, []>("op_4656_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4656_cast = slice_by_index(begin = var_4656_begin_0, end = var_4656_end_0, end_mask = var_4656_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4656_cast")];
            tensor<int32, [4]> var_4660_begin_0 = const()[name = tensor<string, []>("op_4660_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_4660_end_0 = const()[name = tensor<string, []>("op_4660_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_4660_end_mask_0 = const()[name = tensor<string, []>("op_4660_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4660_cast = slice_by_index(begin = var_4660_begin_0, end = var_4660_end_0, end_mask = var_4660_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4660_cast")];
            tensor<int32, [4]> var_4664_begin_0 = const()[name = tensor<string, []>("op_4664_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_4664_end_0 = const()[name = tensor<string, []>("op_4664_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_4664_end_mask_0 = const()[name = tensor<string, []>("op_4664_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4664_cast = slice_by_index(begin = var_4664_begin_0, end = var_4664_end_0, end_mask = var_4664_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4664_cast")];
            tensor<int32, [4]> var_4668_begin_0 = const()[name = tensor<string, []>("op_4668_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_4668_end_0 = const()[name = tensor<string, []>("op_4668_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_4668_end_mask_0 = const()[name = tensor<string, []>("op_4668_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4668_cast = slice_by_index(begin = var_4668_begin_0, end = var_4668_end_0, end_mask = var_4668_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4668_cast")];
            tensor<int32, [4]> var_4672_begin_0 = const()[name = tensor<string, []>("op_4672_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_4672_end_0 = const()[name = tensor<string, []>("op_4672_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_4672_end_mask_0 = const()[name = tensor<string, []>("op_4672_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4672_cast = slice_by_index(begin = var_4672_begin_0, end = var_4672_end_0, end_mask = var_4672_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4672_cast")];
            tensor<int32, [4]> var_4676_begin_0 = const()[name = tensor<string, []>("op_4676_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_4676_end_0 = const()[name = tensor<string, []>("op_4676_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_4676_end_mask_0 = const()[name = tensor<string, []>("op_4676_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4676_cast = slice_by_index(begin = var_4676_begin_0, end = var_4676_end_0, end_mask = var_4676_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_4676_cast")];
            tensor<int32, [4]> k_51_perm_0 = const()[name = tensor<string, []>("k_51_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_4683_begin_0 = const()[name = tensor<string, []>("op_4683_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4683_end_0 = const()[name = tensor<string, []>("op_4683_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_4683_end_mask_0 = const()[name = tensor<string, []>("op_4683_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_127 = transpose(perm = k_51_perm_0, x = k_49_cast)[name = tensor<string, []>("transpose_127")];
            tensor<fp16, [2, 1024, 1, 64]> var_4683_cast = slice_by_index(begin = var_4683_begin_0, end = var_4683_end_0, end_mask = var_4683_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4683_cast")];
            tensor<int32, [4]> var_4687_begin_0 = const()[name = tensor<string, []>("op_4687_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_4687_end_0 = const()[name = tensor<string, []>("op_4687_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_4687_end_mask_0 = const()[name = tensor<string, []>("op_4687_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_4687_cast = slice_by_index(begin = var_4687_begin_0, end = var_4687_end_0, end_mask = var_4687_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4687_cast")];
            tensor<int32, [4]> var_4691_begin_0 = const()[name = tensor<string, []>("op_4691_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_4691_end_0 = const()[name = tensor<string, []>("op_4691_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_4691_end_mask_0 = const()[name = tensor<string, []>("op_4691_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_4691_cast = slice_by_index(begin = var_4691_begin_0, end = var_4691_end_0, end_mask = var_4691_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4691_cast")];
            tensor<int32, [4]> var_4695_begin_0 = const()[name = tensor<string, []>("op_4695_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_4695_end_0 = const()[name = tensor<string, []>("op_4695_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_4695_end_mask_0 = const()[name = tensor<string, []>("op_4695_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_4695_cast = slice_by_index(begin = var_4695_begin_0, end = var_4695_end_0, end_mask = var_4695_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4695_cast")];
            tensor<int32, [4]> var_4699_begin_0 = const()[name = tensor<string, []>("op_4699_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_4699_end_0 = const()[name = tensor<string, []>("op_4699_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_4699_end_mask_0 = const()[name = tensor<string, []>("op_4699_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_4699_cast = slice_by_index(begin = var_4699_begin_0, end = var_4699_end_0, end_mask = var_4699_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4699_cast")];
            tensor<int32, [4]> var_4703_begin_0 = const()[name = tensor<string, []>("op_4703_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_4703_end_0 = const()[name = tensor<string, []>("op_4703_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_4703_end_mask_0 = const()[name = tensor<string, []>("op_4703_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_4703_cast = slice_by_index(begin = var_4703_begin_0, end = var_4703_end_0, end_mask = var_4703_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4703_cast")];
            tensor<int32, [4]> var_4707_begin_0 = const()[name = tensor<string, []>("op_4707_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_4707_end_0 = const()[name = tensor<string, []>("op_4707_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_4707_end_mask_0 = const()[name = tensor<string, []>("op_4707_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_4707_cast = slice_by_index(begin = var_4707_begin_0, end = var_4707_end_0, end_mask = var_4707_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4707_cast")];
            tensor<int32, [4]> var_4711_begin_0 = const()[name = tensor<string, []>("op_4711_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_4711_end_0 = const()[name = tensor<string, []>("op_4711_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_4711_end_mask_0 = const()[name = tensor<string, []>("op_4711_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_4711_cast = slice_by_index(begin = var_4711_begin_0, end = var_4711_end_0, end_mask = var_4711_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4711_cast")];
            tensor<int32, [4]> var_4715_begin_0 = const()[name = tensor<string, []>("op_4715_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_4715_end_0 = const()[name = tensor<string, []>("op_4715_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_4715_end_mask_0 = const()[name = tensor<string, []>("op_4715_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_4715_cast = slice_by_index(begin = var_4715_begin_0, end = var_4715_end_0, end_mask = var_4715_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4715_cast")];
            tensor<int32, [4]> var_4719_begin_0 = const()[name = tensor<string, []>("op_4719_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_4719_end_0 = const()[name = tensor<string, []>("op_4719_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_4719_end_mask_0 = const()[name = tensor<string, []>("op_4719_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_4719_cast = slice_by_index(begin = var_4719_begin_0, end = var_4719_end_0, end_mask = var_4719_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4719_cast")];
            tensor<int32, [4]> var_4723_begin_0 = const()[name = tensor<string, []>("op_4723_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_4723_end_0 = const()[name = tensor<string, []>("op_4723_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_4723_end_mask_0 = const()[name = tensor<string, []>("op_4723_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_4723_cast = slice_by_index(begin = var_4723_begin_0, end = var_4723_end_0, end_mask = var_4723_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4723_cast")];
            tensor<int32, [4]> var_4727_begin_0 = const()[name = tensor<string, []>("op_4727_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_4727_end_0 = const()[name = tensor<string, []>("op_4727_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_4727_end_mask_0 = const()[name = tensor<string, []>("op_4727_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_4727_cast = slice_by_index(begin = var_4727_begin_0, end = var_4727_end_0, end_mask = var_4727_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4727_cast")];
            tensor<int32, [4]> var_4731_begin_0 = const()[name = tensor<string, []>("op_4731_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_4731_end_0 = const()[name = tensor<string, []>("op_4731_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_4731_end_mask_0 = const()[name = tensor<string, []>("op_4731_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_4731_cast = slice_by_index(begin = var_4731_begin_0, end = var_4731_end_0, end_mask = var_4731_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4731_cast")];
            tensor<int32, [4]> var_4735_begin_0 = const()[name = tensor<string, []>("op_4735_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_4735_end_0 = const()[name = tensor<string, []>("op_4735_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_4735_end_mask_0 = const()[name = tensor<string, []>("op_4735_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_4735_cast = slice_by_index(begin = var_4735_begin_0, end = var_4735_end_0, end_mask = var_4735_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4735_cast")];
            tensor<int32, [4]> var_4739_begin_0 = const()[name = tensor<string, []>("op_4739_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_4739_end_0 = const()[name = tensor<string, []>("op_4739_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_4739_end_mask_0 = const()[name = tensor<string, []>("op_4739_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_4739_cast = slice_by_index(begin = var_4739_begin_0, end = var_4739_end_0, end_mask = var_4739_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4739_cast")];
            tensor<int32, [4]> var_4743_begin_0 = const()[name = tensor<string, []>("op_4743_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_4743_end_0 = const()[name = tensor<string, []>("op_4743_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_4743_end_mask_0 = const()[name = tensor<string, []>("op_4743_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_4743_cast = slice_by_index(begin = var_4743_begin_0, end = var_4743_end_0, end_mask = var_4743_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4743_cast")];
            tensor<int32, [4]> var_4747_begin_0 = const()[name = tensor<string, []>("op_4747_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_4747_end_0 = const()[name = tensor<string, []>("op_4747_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_4747_end_mask_0 = const()[name = tensor<string, []>("op_4747_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_4747_cast = slice_by_index(begin = var_4747_begin_0, end = var_4747_end_0, end_mask = var_4747_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4747_cast")];
            tensor<int32, [4]> var_4751_begin_0 = const()[name = tensor<string, []>("op_4751_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_4751_end_0 = const()[name = tensor<string, []>("op_4751_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_4751_end_mask_0 = const()[name = tensor<string, []>("op_4751_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_4751_cast = slice_by_index(begin = var_4751_begin_0, end = var_4751_end_0, end_mask = var_4751_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4751_cast")];
            tensor<int32, [4]> var_4755_begin_0 = const()[name = tensor<string, []>("op_4755_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_4755_end_0 = const()[name = tensor<string, []>("op_4755_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_4755_end_mask_0 = const()[name = tensor<string, []>("op_4755_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_4755_cast = slice_by_index(begin = var_4755_begin_0, end = var_4755_end_0, end_mask = var_4755_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4755_cast")];
            tensor<int32, [4]> var_4759_begin_0 = const()[name = tensor<string, []>("op_4759_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_4759_end_0 = const()[name = tensor<string, []>("op_4759_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_4759_end_mask_0 = const()[name = tensor<string, []>("op_4759_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_4759_cast = slice_by_index(begin = var_4759_begin_0, end = var_4759_end_0, end_mask = var_4759_end_mask_0, x = transpose_127)[name = tensor<string, []>("op_4759_cast")];
            tensor<int32, [4]> var_4761_begin_0 = const()[name = tensor<string, []>("op_4761_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4761_end_0 = const()[name = tensor<string, []>("op_4761_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_4761_end_mask_0 = const()[name = tensor<string, []>("op_4761_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4761_cast = slice_by_index(begin = var_4761_begin_0, end = var_4761_end_0, end_mask = var_4761_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4761_cast")];
            tensor<int32, [4]> var_4765_begin_0 = const()[name = tensor<string, []>("op_4765_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_4765_end_0 = const()[name = tensor<string, []>("op_4765_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_4765_end_mask_0 = const()[name = tensor<string, []>("op_4765_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4765_cast = slice_by_index(begin = var_4765_begin_0, end = var_4765_end_0, end_mask = var_4765_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4765_cast")];
            tensor<int32, [4]> var_4769_begin_0 = const()[name = tensor<string, []>("op_4769_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_4769_end_0 = const()[name = tensor<string, []>("op_4769_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_4769_end_mask_0 = const()[name = tensor<string, []>("op_4769_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4769_cast = slice_by_index(begin = var_4769_begin_0, end = var_4769_end_0, end_mask = var_4769_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4769_cast")];
            tensor<int32, [4]> var_4773_begin_0 = const()[name = tensor<string, []>("op_4773_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_4773_end_0 = const()[name = tensor<string, []>("op_4773_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_4773_end_mask_0 = const()[name = tensor<string, []>("op_4773_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4773_cast = slice_by_index(begin = var_4773_begin_0, end = var_4773_end_0, end_mask = var_4773_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4773_cast")];
            tensor<int32, [4]> var_4777_begin_0 = const()[name = tensor<string, []>("op_4777_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_4777_end_0 = const()[name = tensor<string, []>("op_4777_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_4777_end_mask_0 = const()[name = tensor<string, []>("op_4777_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4777_cast = slice_by_index(begin = var_4777_begin_0, end = var_4777_end_0, end_mask = var_4777_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4777_cast")];
            tensor<int32, [4]> var_4781_begin_0 = const()[name = tensor<string, []>("op_4781_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_4781_end_0 = const()[name = tensor<string, []>("op_4781_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_4781_end_mask_0 = const()[name = tensor<string, []>("op_4781_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4781_cast = slice_by_index(begin = var_4781_begin_0, end = var_4781_end_0, end_mask = var_4781_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4781_cast")];
            tensor<int32, [4]> var_4785_begin_0 = const()[name = tensor<string, []>("op_4785_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_4785_end_0 = const()[name = tensor<string, []>("op_4785_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_4785_end_mask_0 = const()[name = tensor<string, []>("op_4785_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4785_cast = slice_by_index(begin = var_4785_begin_0, end = var_4785_end_0, end_mask = var_4785_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4785_cast")];
            tensor<int32, [4]> var_4789_begin_0 = const()[name = tensor<string, []>("op_4789_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_4789_end_0 = const()[name = tensor<string, []>("op_4789_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_4789_end_mask_0 = const()[name = tensor<string, []>("op_4789_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4789_cast = slice_by_index(begin = var_4789_begin_0, end = var_4789_end_0, end_mask = var_4789_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4789_cast")];
            tensor<int32, [4]> var_4793_begin_0 = const()[name = tensor<string, []>("op_4793_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_4793_end_0 = const()[name = tensor<string, []>("op_4793_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_4793_end_mask_0 = const()[name = tensor<string, []>("op_4793_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4793_cast = slice_by_index(begin = var_4793_begin_0, end = var_4793_end_0, end_mask = var_4793_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4793_cast")];
            tensor<int32, [4]> var_4797_begin_0 = const()[name = tensor<string, []>("op_4797_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_4797_end_0 = const()[name = tensor<string, []>("op_4797_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_4797_end_mask_0 = const()[name = tensor<string, []>("op_4797_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4797_cast = slice_by_index(begin = var_4797_begin_0, end = var_4797_end_0, end_mask = var_4797_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4797_cast")];
            tensor<int32, [4]> var_4801_begin_0 = const()[name = tensor<string, []>("op_4801_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_4801_end_0 = const()[name = tensor<string, []>("op_4801_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_4801_end_mask_0 = const()[name = tensor<string, []>("op_4801_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4801_cast = slice_by_index(begin = var_4801_begin_0, end = var_4801_end_0, end_mask = var_4801_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4801_cast")];
            tensor<int32, [4]> var_4805_begin_0 = const()[name = tensor<string, []>("op_4805_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_4805_end_0 = const()[name = tensor<string, []>("op_4805_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_4805_end_mask_0 = const()[name = tensor<string, []>("op_4805_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4805_cast = slice_by_index(begin = var_4805_begin_0, end = var_4805_end_0, end_mask = var_4805_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4805_cast")];
            tensor<int32, [4]> var_4809_begin_0 = const()[name = tensor<string, []>("op_4809_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_4809_end_0 = const()[name = tensor<string, []>("op_4809_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_4809_end_mask_0 = const()[name = tensor<string, []>("op_4809_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4809_cast = slice_by_index(begin = var_4809_begin_0, end = var_4809_end_0, end_mask = var_4809_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4809_cast")];
            tensor<int32, [4]> var_4813_begin_0 = const()[name = tensor<string, []>("op_4813_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_4813_end_0 = const()[name = tensor<string, []>("op_4813_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_4813_end_mask_0 = const()[name = tensor<string, []>("op_4813_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4813_cast = slice_by_index(begin = var_4813_begin_0, end = var_4813_end_0, end_mask = var_4813_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4813_cast")];
            tensor<int32, [4]> var_4817_begin_0 = const()[name = tensor<string, []>("op_4817_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_4817_end_0 = const()[name = tensor<string, []>("op_4817_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_4817_end_mask_0 = const()[name = tensor<string, []>("op_4817_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4817_cast = slice_by_index(begin = var_4817_begin_0, end = var_4817_end_0, end_mask = var_4817_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4817_cast")];
            tensor<int32, [4]> var_4821_begin_0 = const()[name = tensor<string, []>("op_4821_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_4821_end_0 = const()[name = tensor<string, []>("op_4821_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_4821_end_mask_0 = const()[name = tensor<string, []>("op_4821_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4821_cast = slice_by_index(begin = var_4821_begin_0, end = var_4821_end_0, end_mask = var_4821_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4821_cast")];
            tensor<int32, [4]> var_4825_begin_0 = const()[name = tensor<string, []>("op_4825_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_4825_end_0 = const()[name = tensor<string, []>("op_4825_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_4825_end_mask_0 = const()[name = tensor<string, []>("op_4825_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4825_cast = slice_by_index(begin = var_4825_begin_0, end = var_4825_end_0, end_mask = var_4825_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4825_cast")];
            tensor<int32, [4]> var_4829_begin_0 = const()[name = tensor<string, []>("op_4829_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_4829_end_0 = const()[name = tensor<string, []>("op_4829_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_4829_end_mask_0 = const()[name = tensor<string, []>("op_4829_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4829_cast = slice_by_index(begin = var_4829_begin_0, end = var_4829_end_0, end_mask = var_4829_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4829_cast")];
            tensor<int32, [4]> var_4833_begin_0 = const()[name = tensor<string, []>("op_4833_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_4833_end_0 = const()[name = tensor<string, []>("op_4833_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_4833_end_mask_0 = const()[name = tensor<string, []>("op_4833_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4833_cast = slice_by_index(begin = var_4833_begin_0, end = var_4833_end_0, end_mask = var_4833_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4833_cast")];
            tensor<int32, [4]> var_4837_begin_0 = const()[name = tensor<string, []>("op_4837_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_4837_end_0 = const()[name = tensor<string, []>("op_4837_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_4837_end_mask_0 = const()[name = tensor<string, []>("op_4837_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_4837_cast = slice_by_index(begin = var_4837_begin_0, end = var_4837_end_0, end_mask = var_4837_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_4837_cast")];
            tensor<string, []> var_4841_equation_0 = const()[name = tensor<string, []>("op_4841_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4841_cast = einsum(equation = var_4841_equation_0, values = (var_4683_cast, var_4600_cast))[name = tensor<string, []>("op_4841_cast")];
            tensor<fp16, []> var_4842_to_fp16 = const()[name = tensor<string, []>("op_4842_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_321_cast = mul(x = var_4841_cast, y = var_4842_to_fp16)[name = tensor<string, []>("aw_321_cast")];
            tensor<string, []> var_4845_equation_0 = const()[name = tensor<string, []>("op_4845_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4845_cast = einsum(equation = var_4845_equation_0, values = (var_4687_cast, var_4604_cast))[name = tensor<string, []>("op_4845_cast")];
            tensor<fp16, []> var_4846_to_fp16 = const()[name = tensor<string, []>("op_4846_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_323_cast = mul(x = var_4845_cast, y = var_4846_to_fp16)[name = tensor<string, []>("aw_323_cast")];
            tensor<string, []> var_4849_equation_0 = const()[name = tensor<string, []>("op_4849_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4849_cast = einsum(equation = var_4849_equation_0, values = (var_4691_cast, var_4608_cast))[name = tensor<string, []>("op_4849_cast")];
            tensor<fp16, []> var_4850_to_fp16 = const()[name = tensor<string, []>("op_4850_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_325_cast = mul(x = var_4849_cast, y = var_4850_to_fp16)[name = tensor<string, []>("aw_325_cast")];
            tensor<string, []> var_4853_equation_0 = const()[name = tensor<string, []>("op_4853_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4853_cast = einsum(equation = var_4853_equation_0, values = (var_4695_cast, var_4612_cast))[name = tensor<string, []>("op_4853_cast")];
            tensor<fp16, []> var_4854_to_fp16 = const()[name = tensor<string, []>("op_4854_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_327_cast = mul(x = var_4853_cast, y = var_4854_to_fp16)[name = tensor<string, []>("aw_327_cast")];
            tensor<string, []> var_4857_equation_0 = const()[name = tensor<string, []>("op_4857_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4857_cast = einsum(equation = var_4857_equation_0, values = (var_4699_cast, var_4616_cast))[name = tensor<string, []>("op_4857_cast")];
            tensor<fp16, []> var_4858_to_fp16 = const()[name = tensor<string, []>("op_4858_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_329_cast = mul(x = var_4857_cast, y = var_4858_to_fp16)[name = tensor<string, []>("aw_329_cast")];
            tensor<string, []> var_4861_equation_0 = const()[name = tensor<string, []>("op_4861_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4861_cast = einsum(equation = var_4861_equation_0, values = (var_4703_cast, var_4620_cast))[name = tensor<string, []>("op_4861_cast")];
            tensor<fp16, []> var_4862_to_fp16 = const()[name = tensor<string, []>("op_4862_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_331_cast = mul(x = var_4861_cast, y = var_4862_to_fp16)[name = tensor<string, []>("aw_331_cast")];
            tensor<string, []> var_4865_equation_0 = const()[name = tensor<string, []>("op_4865_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4865_cast = einsum(equation = var_4865_equation_0, values = (var_4707_cast, var_4624_cast))[name = tensor<string, []>("op_4865_cast")];
            tensor<fp16, []> var_4866_to_fp16 = const()[name = tensor<string, []>("op_4866_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_333_cast = mul(x = var_4865_cast, y = var_4866_to_fp16)[name = tensor<string, []>("aw_333_cast")];
            tensor<string, []> var_4869_equation_0 = const()[name = tensor<string, []>("op_4869_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4869_cast = einsum(equation = var_4869_equation_0, values = (var_4711_cast, var_4628_cast))[name = tensor<string, []>("op_4869_cast")];
            tensor<fp16, []> var_4870_to_fp16 = const()[name = tensor<string, []>("op_4870_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_335_cast = mul(x = var_4869_cast, y = var_4870_to_fp16)[name = tensor<string, []>("aw_335_cast")];
            tensor<string, []> var_4873_equation_0 = const()[name = tensor<string, []>("op_4873_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4873_cast = einsum(equation = var_4873_equation_0, values = (var_4715_cast, var_4632_cast))[name = tensor<string, []>("op_4873_cast")];
            tensor<fp16, []> var_4874_to_fp16 = const()[name = tensor<string, []>("op_4874_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_337_cast = mul(x = var_4873_cast, y = var_4874_to_fp16)[name = tensor<string, []>("aw_337_cast")];
            tensor<string, []> var_4877_equation_0 = const()[name = tensor<string, []>("op_4877_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4877_cast = einsum(equation = var_4877_equation_0, values = (var_4719_cast, var_4636_cast))[name = tensor<string, []>("op_4877_cast")];
            tensor<fp16, []> var_4878_to_fp16 = const()[name = tensor<string, []>("op_4878_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_339_cast = mul(x = var_4877_cast, y = var_4878_to_fp16)[name = tensor<string, []>("aw_339_cast")];
            tensor<string, []> var_4881_equation_0 = const()[name = tensor<string, []>("op_4881_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4881_cast = einsum(equation = var_4881_equation_0, values = (var_4723_cast, var_4640_cast))[name = tensor<string, []>("op_4881_cast")];
            tensor<fp16, []> var_4882_to_fp16 = const()[name = tensor<string, []>("op_4882_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_341_cast = mul(x = var_4881_cast, y = var_4882_to_fp16)[name = tensor<string, []>("aw_341_cast")];
            tensor<string, []> var_4885_equation_0 = const()[name = tensor<string, []>("op_4885_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4885_cast = einsum(equation = var_4885_equation_0, values = (var_4727_cast, var_4644_cast))[name = tensor<string, []>("op_4885_cast")];
            tensor<fp16, []> var_4886_to_fp16 = const()[name = tensor<string, []>("op_4886_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_343_cast = mul(x = var_4885_cast, y = var_4886_to_fp16)[name = tensor<string, []>("aw_343_cast")];
            tensor<string, []> var_4889_equation_0 = const()[name = tensor<string, []>("op_4889_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4889_cast = einsum(equation = var_4889_equation_0, values = (var_4731_cast, var_4648_cast))[name = tensor<string, []>("op_4889_cast")];
            tensor<fp16, []> var_4890_to_fp16 = const()[name = tensor<string, []>("op_4890_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_345_cast = mul(x = var_4889_cast, y = var_4890_to_fp16)[name = tensor<string, []>("aw_345_cast")];
            tensor<string, []> var_4893_equation_0 = const()[name = tensor<string, []>("op_4893_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4893_cast = einsum(equation = var_4893_equation_0, values = (var_4735_cast, var_4652_cast))[name = tensor<string, []>("op_4893_cast")];
            tensor<fp16, []> var_4894_to_fp16 = const()[name = tensor<string, []>("op_4894_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_347_cast = mul(x = var_4893_cast, y = var_4894_to_fp16)[name = tensor<string, []>("aw_347_cast")];
            tensor<string, []> var_4897_equation_0 = const()[name = tensor<string, []>("op_4897_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4897_cast = einsum(equation = var_4897_equation_0, values = (var_4739_cast, var_4656_cast))[name = tensor<string, []>("op_4897_cast")];
            tensor<fp16, []> var_4898_to_fp16 = const()[name = tensor<string, []>("op_4898_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_349_cast = mul(x = var_4897_cast, y = var_4898_to_fp16)[name = tensor<string, []>("aw_349_cast")];
            tensor<string, []> var_4901_equation_0 = const()[name = tensor<string, []>("op_4901_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4901_cast = einsum(equation = var_4901_equation_0, values = (var_4743_cast, var_4660_cast))[name = tensor<string, []>("op_4901_cast")];
            tensor<fp16, []> var_4902_to_fp16 = const()[name = tensor<string, []>("op_4902_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_351_cast = mul(x = var_4901_cast, y = var_4902_to_fp16)[name = tensor<string, []>("aw_351_cast")];
            tensor<string, []> var_4905_equation_0 = const()[name = tensor<string, []>("op_4905_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4905_cast = einsum(equation = var_4905_equation_0, values = (var_4747_cast, var_4664_cast))[name = tensor<string, []>("op_4905_cast")];
            tensor<fp16, []> var_4906_to_fp16 = const()[name = tensor<string, []>("op_4906_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_353_cast = mul(x = var_4905_cast, y = var_4906_to_fp16)[name = tensor<string, []>("aw_353_cast")];
            tensor<string, []> var_4909_equation_0 = const()[name = tensor<string, []>("op_4909_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4909_cast = einsum(equation = var_4909_equation_0, values = (var_4751_cast, var_4668_cast))[name = tensor<string, []>("op_4909_cast")];
            tensor<fp16, []> var_4910_to_fp16 = const()[name = tensor<string, []>("op_4910_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_355_cast = mul(x = var_4909_cast, y = var_4910_to_fp16)[name = tensor<string, []>("aw_355_cast")];
            tensor<string, []> var_4913_equation_0 = const()[name = tensor<string, []>("op_4913_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4913_cast = einsum(equation = var_4913_equation_0, values = (var_4755_cast, var_4672_cast))[name = tensor<string, []>("op_4913_cast")];
            tensor<fp16, []> var_4914_to_fp16 = const()[name = tensor<string, []>("op_4914_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_357_cast = mul(x = var_4913_cast, y = var_4914_to_fp16)[name = tensor<string, []>("aw_357_cast")];
            tensor<string, []> var_4917_equation_0 = const()[name = tensor<string, []>("op_4917_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4917_cast = einsum(equation = var_4917_equation_0, values = (var_4759_cast, var_4676_cast))[name = tensor<string, []>("op_4917_cast")];
            tensor<fp16, []> var_4918_to_fp16 = const()[name = tensor<string, []>("op_4918_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_359_cast = mul(x = var_4917_cast, y = var_4918_to_fp16)[name = tensor<string, []>("aw_359_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4920_cast = softmax(axis = var_2634, x = aw_321_cast)[name = tensor<string, []>("op_4920_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4921_cast = softmax(axis = var_2634, x = aw_323_cast)[name = tensor<string, []>("op_4921_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4922_cast = softmax(axis = var_2634, x = aw_325_cast)[name = tensor<string, []>("op_4922_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4923_cast = softmax(axis = var_2634, x = aw_327_cast)[name = tensor<string, []>("op_4923_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4924_cast = softmax(axis = var_2634, x = aw_329_cast)[name = tensor<string, []>("op_4924_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4925_cast = softmax(axis = var_2634, x = aw_331_cast)[name = tensor<string, []>("op_4925_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4926_cast = softmax(axis = var_2634, x = aw_333_cast)[name = tensor<string, []>("op_4926_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4927_cast = softmax(axis = var_2634, x = aw_335_cast)[name = tensor<string, []>("op_4927_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4928_cast = softmax(axis = var_2634, x = aw_337_cast)[name = tensor<string, []>("op_4928_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4929_cast = softmax(axis = var_2634, x = aw_339_cast)[name = tensor<string, []>("op_4929_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4930_cast = softmax(axis = var_2634, x = aw_341_cast)[name = tensor<string, []>("op_4930_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4931_cast = softmax(axis = var_2634, x = aw_343_cast)[name = tensor<string, []>("op_4931_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4932_cast = softmax(axis = var_2634, x = aw_345_cast)[name = tensor<string, []>("op_4932_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4933_cast = softmax(axis = var_2634, x = aw_347_cast)[name = tensor<string, []>("op_4933_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4934_cast = softmax(axis = var_2634, x = aw_349_cast)[name = tensor<string, []>("op_4934_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4935_cast = softmax(axis = var_2634, x = aw_351_cast)[name = tensor<string, []>("op_4935_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4936_cast = softmax(axis = var_2634, x = aw_353_cast)[name = tensor<string, []>("op_4936_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4937_cast = softmax(axis = var_2634, x = aw_355_cast)[name = tensor<string, []>("op_4937_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4938_cast = softmax(axis = var_2634, x = aw_357_cast)[name = tensor<string, []>("op_4938_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_4939_cast = softmax(axis = var_2634, x = aw_359_cast)[name = tensor<string, []>("op_4939_cast")];
            tensor<string, []> var_4941_equation_0 = const()[name = tensor<string, []>("op_4941_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4941_cast = einsum(equation = var_4941_equation_0, values = (var_4761_cast, var_4920_cast))[name = tensor<string, []>("op_4941_cast")];
            tensor<string, []> var_4943_equation_0 = const()[name = tensor<string, []>("op_4943_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4943_cast = einsum(equation = var_4943_equation_0, values = (var_4765_cast, var_4921_cast))[name = tensor<string, []>("op_4943_cast")];
            tensor<string, []> var_4945_equation_0 = const()[name = tensor<string, []>("op_4945_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4945_cast = einsum(equation = var_4945_equation_0, values = (var_4769_cast, var_4922_cast))[name = tensor<string, []>("op_4945_cast")];
            tensor<string, []> var_4947_equation_0 = const()[name = tensor<string, []>("op_4947_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4947_cast = einsum(equation = var_4947_equation_0, values = (var_4773_cast, var_4923_cast))[name = tensor<string, []>("op_4947_cast")];
            tensor<string, []> var_4949_equation_0 = const()[name = tensor<string, []>("op_4949_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4949_cast = einsum(equation = var_4949_equation_0, values = (var_4777_cast, var_4924_cast))[name = tensor<string, []>("op_4949_cast")];
            tensor<string, []> var_4951_equation_0 = const()[name = tensor<string, []>("op_4951_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4951_cast = einsum(equation = var_4951_equation_0, values = (var_4781_cast, var_4925_cast))[name = tensor<string, []>("op_4951_cast")];
            tensor<string, []> var_4953_equation_0 = const()[name = tensor<string, []>("op_4953_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4953_cast = einsum(equation = var_4953_equation_0, values = (var_4785_cast, var_4926_cast))[name = tensor<string, []>("op_4953_cast")];
            tensor<string, []> var_4955_equation_0 = const()[name = tensor<string, []>("op_4955_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4955_cast = einsum(equation = var_4955_equation_0, values = (var_4789_cast, var_4927_cast))[name = tensor<string, []>("op_4955_cast")];
            tensor<string, []> var_4957_equation_0 = const()[name = tensor<string, []>("op_4957_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4957_cast = einsum(equation = var_4957_equation_0, values = (var_4793_cast, var_4928_cast))[name = tensor<string, []>("op_4957_cast")];
            tensor<string, []> var_4959_equation_0 = const()[name = tensor<string, []>("op_4959_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4959_cast = einsum(equation = var_4959_equation_0, values = (var_4797_cast, var_4929_cast))[name = tensor<string, []>("op_4959_cast")];
            tensor<string, []> var_4961_equation_0 = const()[name = tensor<string, []>("op_4961_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4961_cast = einsum(equation = var_4961_equation_0, values = (var_4801_cast, var_4930_cast))[name = tensor<string, []>("op_4961_cast")];
            tensor<string, []> var_4963_equation_0 = const()[name = tensor<string, []>("op_4963_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4963_cast = einsum(equation = var_4963_equation_0, values = (var_4805_cast, var_4931_cast))[name = tensor<string, []>("op_4963_cast")];
            tensor<string, []> var_4965_equation_0 = const()[name = tensor<string, []>("op_4965_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4965_cast = einsum(equation = var_4965_equation_0, values = (var_4809_cast, var_4932_cast))[name = tensor<string, []>("op_4965_cast")];
            tensor<string, []> var_4967_equation_0 = const()[name = tensor<string, []>("op_4967_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4967_cast = einsum(equation = var_4967_equation_0, values = (var_4813_cast, var_4933_cast))[name = tensor<string, []>("op_4967_cast")];
            tensor<string, []> var_4969_equation_0 = const()[name = tensor<string, []>("op_4969_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4969_cast = einsum(equation = var_4969_equation_0, values = (var_4817_cast, var_4934_cast))[name = tensor<string, []>("op_4969_cast")];
            tensor<string, []> var_4971_equation_0 = const()[name = tensor<string, []>("op_4971_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4971_cast = einsum(equation = var_4971_equation_0, values = (var_4821_cast, var_4935_cast))[name = tensor<string, []>("op_4971_cast")];
            tensor<string, []> var_4973_equation_0 = const()[name = tensor<string, []>("op_4973_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4973_cast = einsum(equation = var_4973_equation_0, values = (var_4825_cast, var_4936_cast))[name = tensor<string, []>("op_4973_cast")];
            tensor<string, []> var_4975_equation_0 = const()[name = tensor<string, []>("op_4975_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4975_cast = einsum(equation = var_4975_equation_0, values = (var_4829_cast, var_4937_cast))[name = tensor<string, []>("op_4975_cast")];
            tensor<string, []> var_4977_equation_0 = const()[name = tensor<string, []>("op_4977_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4977_cast = einsum(equation = var_4977_equation_0, values = (var_4833_cast, var_4938_cast))[name = tensor<string, []>("op_4977_cast")];
            tensor<string, []> var_4979_equation_0 = const()[name = tensor<string, []>("op_4979_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_4979_cast = einsum(equation = var_4979_equation_0, values = (var_4837_cast, var_4939_cast))[name = tensor<string, []>("op_4979_cast")];
            tensor<bool, []> input_151_interleave_0 = const()[name = tensor<string, []>("input_151_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_151_cast = concat(axis = var_2634, interleave = input_151_interleave_0, values = (var_4941_cast, var_4943_cast, var_4945_cast, var_4947_cast, var_4949_cast, var_4951_cast, var_4953_cast, var_4955_cast, var_4957_cast, var_4959_cast, var_4961_cast, var_4963_cast, var_4965_cast, var_4967_cast, var_4969_cast, var_4971_cast, var_4973_cast, var_4975_cast, var_4977_cast, var_4979_cast))[name = tensor<string, []>("input_151_cast")];
            tensor<int32, [2]> var_4985 = const()[name = tensor<string, []>("op_4985"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4987 = const()[name = tensor<string, []>("op_4987"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_4989_pad_type_0 = const()[name = tensor<string, []>("op_4989_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_4989_pad_0 = const()[name = tensor<string, []>("op_4989_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(347027648)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(350304512)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_4989_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16, dilations = var_4987, groups = var_2634, pad = var_4989_pad_0, pad_type = var_4989_pad_type_0, strides = var_4985, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16, x = input_151_cast)[name = tensor<string, []>("op_4989_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_39_cast = add(x = var_4989_cast, y = inputs_37_cast)[name = tensor<string, []>("inputs_39_cast")];
            tensor<int32, [1]> var_4993 = const()[name = tensor<string, []>("op_4993"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_39_cast = reduce_mean(axes = var_4993, keep_dims = var_2629, x = inputs_39_cast)[name = tensor<string, []>("channels_mean_39_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_39_cast = sub(x = inputs_39_cast, y = channels_mean_39_cast)[name = tensor<string, []>("zero_mean_39_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_39_cast = mul(x = zero_mean_39_cast, y = zero_mean_39_cast)[name = tensor<string, []>("zero_mean_sq_39_cast")];
            tensor<int32, [1]> var_4997 = const()[name = tensor<string, []>("op_4997"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_4998_cast = reduce_mean(axes = var_4997, keep_dims = var_2629, x = zero_mean_sq_39_cast)[name = tensor<string, []>("op_4998_cast")];
            tensor<fp16, []> var_4999_to_fp16 = const()[name = tensor<string, []>("op_4999_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_5000_cast = add(x = var_4998_cast, y = var_4999_to_fp16)[name = tensor<string, []>("op_5000_cast")];
            tensor<fp16, []> denom_39_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_39_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_39_cast = rsqrt(epsilon = denom_39_epsilon_0_to_fp16, x = var_5000_cast)[name = tensor<string, []>("denom_39_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_39_cast = mul(x = zero_mean_39_cast, y = denom_39_cast)[name = tensor<string, []>("out_39_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5004_to_fp16 = const()[name = tensor<string, []>("op_5004_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(350307136)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_5005_cast = add(x = out_39_cast, y = var_5004_to_fp16)[name = tensor<string, []>("op_5005_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5007_to_fp16 = const()[name = tensor<string, []>("op_5007_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(350309760)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_79_cast = mul(x = var_5005_cast, y = var_5007_to_fp16)[name = tensor<string, []>("hidden_states_79_cast")];
            tensor<int32, [2]> var_5014 = const()[name = tensor<string, []>("op_5014"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5016 = const()[name = tensor<string, []>("op_5016"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_27_pad_type_0 = const()[name = tensor<string, []>("q_27_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_27_pad_0 = const()[name = tensor<string, []>("q_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(350312384)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_27_cast = conv(dilations = var_5016, groups = var_2634, pad = q_27_pad_0, pad_type = q_27_pad_type_0, strides = var_5014, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16, x = hidden_states_79_cast)[name = tensor<string, []>("q_27_cast")];
            tensor<int32, [2]> var_5020 = const()[name = tensor<string, []>("op_5020"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5022 = const()[name = tensor<string, []>("op_5022"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_53_pad_type_0 = const()[name = tensor<string, []>("k_53_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_53_pad_0 = const()[name = tensor<string, []>("k_53_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(353589248)))];
            tensor<fp16, [2, 1280, 1, 77]> k_53_cast = conv(dilations = var_5022, groups = var_2634, pad = k_53_pad_0, pad_type = k_53_pad_type_0, strides = var_5020, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_53_cast")];
            tensor<int32, [2]> var_5026 = const()[name = tensor<string, []>("op_5026"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5028 = const()[name = tensor<string, []>("op_5028"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_27_pad_type_0 = const()[name = tensor<string, []>("v_27_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_27_pad_0 = const()[name = tensor<string, []>("v_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(358832192)))];
            tensor<fp16, [2, 1280, 1, 77]> v_27_cast = conv(dilations = var_5028, groups = var_2634, pad = v_27_pad_0, pad_type = v_27_pad_type_0, strides = var_5026, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_27_cast")];
            tensor<int32, [4]> var_5032_begin_0 = const()[name = tensor<string, []>("op_5032_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5032_end_0 = const()[name = tensor<string, []>("op_5032_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_5032_end_mask_0 = const()[name = tensor<string, []>("op_5032_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5032_cast = slice_by_index(begin = var_5032_begin_0, end = var_5032_end_0, end_mask = var_5032_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5032_cast")];
            tensor<int32, [4]> var_5036_begin_0 = const()[name = tensor<string, []>("op_5036_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_5036_end_0 = const()[name = tensor<string, []>("op_5036_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_5036_end_mask_0 = const()[name = tensor<string, []>("op_5036_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5036_cast = slice_by_index(begin = var_5036_begin_0, end = var_5036_end_0, end_mask = var_5036_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5036_cast")];
            tensor<int32, [4]> var_5040_begin_0 = const()[name = tensor<string, []>("op_5040_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_5040_end_0 = const()[name = tensor<string, []>("op_5040_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_5040_end_mask_0 = const()[name = tensor<string, []>("op_5040_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5040_cast = slice_by_index(begin = var_5040_begin_0, end = var_5040_end_0, end_mask = var_5040_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5040_cast")];
            tensor<int32, [4]> var_5044_begin_0 = const()[name = tensor<string, []>("op_5044_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_5044_end_0 = const()[name = tensor<string, []>("op_5044_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_5044_end_mask_0 = const()[name = tensor<string, []>("op_5044_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5044_cast = slice_by_index(begin = var_5044_begin_0, end = var_5044_end_0, end_mask = var_5044_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5044_cast")];
            tensor<int32, [4]> var_5048_begin_0 = const()[name = tensor<string, []>("op_5048_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_5048_end_0 = const()[name = tensor<string, []>("op_5048_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_5048_end_mask_0 = const()[name = tensor<string, []>("op_5048_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5048_cast = slice_by_index(begin = var_5048_begin_0, end = var_5048_end_0, end_mask = var_5048_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5048_cast")];
            tensor<int32, [4]> var_5052_begin_0 = const()[name = tensor<string, []>("op_5052_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_5052_end_0 = const()[name = tensor<string, []>("op_5052_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_5052_end_mask_0 = const()[name = tensor<string, []>("op_5052_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5052_cast = slice_by_index(begin = var_5052_begin_0, end = var_5052_end_0, end_mask = var_5052_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5052_cast")];
            tensor<int32, [4]> var_5056_begin_0 = const()[name = tensor<string, []>("op_5056_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_5056_end_0 = const()[name = tensor<string, []>("op_5056_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_5056_end_mask_0 = const()[name = tensor<string, []>("op_5056_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5056_cast = slice_by_index(begin = var_5056_begin_0, end = var_5056_end_0, end_mask = var_5056_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5056_cast")];
            tensor<int32, [4]> var_5060_begin_0 = const()[name = tensor<string, []>("op_5060_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_5060_end_0 = const()[name = tensor<string, []>("op_5060_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_5060_end_mask_0 = const()[name = tensor<string, []>("op_5060_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5060_cast = slice_by_index(begin = var_5060_begin_0, end = var_5060_end_0, end_mask = var_5060_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5060_cast")];
            tensor<int32, [4]> var_5064_begin_0 = const()[name = tensor<string, []>("op_5064_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_5064_end_0 = const()[name = tensor<string, []>("op_5064_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_5064_end_mask_0 = const()[name = tensor<string, []>("op_5064_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5064_cast = slice_by_index(begin = var_5064_begin_0, end = var_5064_end_0, end_mask = var_5064_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5064_cast")];
            tensor<int32, [4]> var_5068_begin_0 = const()[name = tensor<string, []>("op_5068_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_5068_end_0 = const()[name = tensor<string, []>("op_5068_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_5068_end_mask_0 = const()[name = tensor<string, []>("op_5068_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5068_cast = slice_by_index(begin = var_5068_begin_0, end = var_5068_end_0, end_mask = var_5068_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5068_cast")];
            tensor<int32, [4]> var_5072_begin_0 = const()[name = tensor<string, []>("op_5072_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_5072_end_0 = const()[name = tensor<string, []>("op_5072_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_5072_end_mask_0 = const()[name = tensor<string, []>("op_5072_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5072_cast = slice_by_index(begin = var_5072_begin_0, end = var_5072_end_0, end_mask = var_5072_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5072_cast")];
            tensor<int32, [4]> var_5076_begin_0 = const()[name = tensor<string, []>("op_5076_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_5076_end_0 = const()[name = tensor<string, []>("op_5076_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_5076_end_mask_0 = const()[name = tensor<string, []>("op_5076_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5076_cast = slice_by_index(begin = var_5076_begin_0, end = var_5076_end_0, end_mask = var_5076_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5076_cast")];
            tensor<int32, [4]> var_5080_begin_0 = const()[name = tensor<string, []>("op_5080_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_5080_end_0 = const()[name = tensor<string, []>("op_5080_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_5080_end_mask_0 = const()[name = tensor<string, []>("op_5080_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5080_cast = slice_by_index(begin = var_5080_begin_0, end = var_5080_end_0, end_mask = var_5080_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5080_cast")];
            tensor<int32, [4]> var_5084_begin_0 = const()[name = tensor<string, []>("op_5084_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_5084_end_0 = const()[name = tensor<string, []>("op_5084_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_5084_end_mask_0 = const()[name = tensor<string, []>("op_5084_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5084_cast = slice_by_index(begin = var_5084_begin_0, end = var_5084_end_0, end_mask = var_5084_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5084_cast")];
            tensor<int32, [4]> var_5088_begin_0 = const()[name = tensor<string, []>("op_5088_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_5088_end_0 = const()[name = tensor<string, []>("op_5088_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_5088_end_mask_0 = const()[name = tensor<string, []>("op_5088_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5088_cast = slice_by_index(begin = var_5088_begin_0, end = var_5088_end_0, end_mask = var_5088_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5088_cast")];
            tensor<int32, [4]> var_5092_begin_0 = const()[name = tensor<string, []>("op_5092_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_5092_end_0 = const()[name = tensor<string, []>("op_5092_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_5092_end_mask_0 = const()[name = tensor<string, []>("op_5092_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5092_cast = slice_by_index(begin = var_5092_begin_0, end = var_5092_end_0, end_mask = var_5092_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5092_cast")];
            tensor<int32, [4]> var_5096_begin_0 = const()[name = tensor<string, []>("op_5096_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_5096_end_0 = const()[name = tensor<string, []>("op_5096_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_5096_end_mask_0 = const()[name = tensor<string, []>("op_5096_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5096_cast = slice_by_index(begin = var_5096_begin_0, end = var_5096_end_0, end_mask = var_5096_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5096_cast")];
            tensor<int32, [4]> var_5100_begin_0 = const()[name = tensor<string, []>("op_5100_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_5100_end_0 = const()[name = tensor<string, []>("op_5100_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_5100_end_mask_0 = const()[name = tensor<string, []>("op_5100_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5100_cast = slice_by_index(begin = var_5100_begin_0, end = var_5100_end_0, end_mask = var_5100_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5100_cast")];
            tensor<int32, [4]> var_5104_begin_0 = const()[name = tensor<string, []>("op_5104_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_5104_end_0 = const()[name = tensor<string, []>("op_5104_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_5104_end_mask_0 = const()[name = tensor<string, []>("op_5104_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5104_cast = slice_by_index(begin = var_5104_begin_0, end = var_5104_end_0, end_mask = var_5104_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5104_cast")];
            tensor<int32, [4]> var_5108_begin_0 = const()[name = tensor<string, []>("op_5108_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_5108_end_0 = const()[name = tensor<string, []>("op_5108_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_5108_end_mask_0 = const()[name = tensor<string, []>("op_5108_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5108_cast = slice_by_index(begin = var_5108_begin_0, end = var_5108_end_0, end_mask = var_5108_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_5108_cast")];
            tensor<int32, [4]> k_55_perm_0 = const()[name = tensor<string, []>("k_55_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_5115_begin_0 = const()[name = tensor<string, []>("op_5115_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5115_end_0 = const()[name = tensor<string, []>("op_5115_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_5115_end_mask_0 = const()[name = tensor<string, []>("op_5115_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_126 = transpose(perm = k_55_perm_0, x = k_53_cast)[name = tensor<string, []>("transpose_126")];
            tensor<fp16, [2, 77, 1, 64]> var_5115_cast = slice_by_index(begin = var_5115_begin_0, end = var_5115_end_0, end_mask = var_5115_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5115_cast")];
            tensor<int32, [4]> var_5119_begin_0 = const()[name = tensor<string, []>("op_5119_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_5119_end_0 = const()[name = tensor<string, []>("op_5119_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_5119_end_mask_0 = const()[name = tensor<string, []>("op_5119_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5119_cast = slice_by_index(begin = var_5119_begin_0, end = var_5119_end_0, end_mask = var_5119_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5119_cast")];
            tensor<int32, [4]> var_5123_begin_0 = const()[name = tensor<string, []>("op_5123_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_5123_end_0 = const()[name = tensor<string, []>("op_5123_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_5123_end_mask_0 = const()[name = tensor<string, []>("op_5123_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5123_cast = slice_by_index(begin = var_5123_begin_0, end = var_5123_end_0, end_mask = var_5123_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5123_cast")];
            tensor<int32, [4]> var_5127_begin_0 = const()[name = tensor<string, []>("op_5127_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_5127_end_0 = const()[name = tensor<string, []>("op_5127_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_5127_end_mask_0 = const()[name = tensor<string, []>("op_5127_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5127_cast = slice_by_index(begin = var_5127_begin_0, end = var_5127_end_0, end_mask = var_5127_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5127_cast")];
            tensor<int32, [4]> var_5131_begin_0 = const()[name = tensor<string, []>("op_5131_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_5131_end_0 = const()[name = tensor<string, []>("op_5131_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_5131_end_mask_0 = const()[name = tensor<string, []>("op_5131_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5131_cast = slice_by_index(begin = var_5131_begin_0, end = var_5131_end_0, end_mask = var_5131_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5131_cast")];
            tensor<int32, [4]> var_5135_begin_0 = const()[name = tensor<string, []>("op_5135_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_5135_end_0 = const()[name = tensor<string, []>("op_5135_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_5135_end_mask_0 = const()[name = tensor<string, []>("op_5135_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5135_cast = slice_by_index(begin = var_5135_begin_0, end = var_5135_end_0, end_mask = var_5135_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5135_cast")];
            tensor<int32, [4]> var_5139_begin_0 = const()[name = tensor<string, []>("op_5139_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_5139_end_0 = const()[name = tensor<string, []>("op_5139_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_5139_end_mask_0 = const()[name = tensor<string, []>("op_5139_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5139_cast = slice_by_index(begin = var_5139_begin_0, end = var_5139_end_0, end_mask = var_5139_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5139_cast")];
            tensor<int32, [4]> var_5143_begin_0 = const()[name = tensor<string, []>("op_5143_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_5143_end_0 = const()[name = tensor<string, []>("op_5143_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_5143_end_mask_0 = const()[name = tensor<string, []>("op_5143_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5143_cast = slice_by_index(begin = var_5143_begin_0, end = var_5143_end_0, end_mask = var_5143_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5143_cast")];
            tensor<int32, [4]> var_5147_begin_0 = const()[name = tensor<string, []>("op_5147_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_5147_end_0 = const()[name = tensor<string, []>("op_5147_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_5147_end_mask_0 = const()[name = tensor<string, []>("op_5147_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5147_cast = slice_by_index(begin = var_5147_begin_0, end = var_5147_end_0, end_mask = var_5147_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5147_cast")];
            tensor<int32, [4]> var_5151_begin_0 = const()[name = tensor<string, []>("op_5151_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_5151_end_0 = const()[name = tensor<string, []>("op_5151_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_5151_end_mask_0 = const()[name = tensor<string, []>("op_5151_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5151_cast = slice_by_index(begin = var_5151_begin_0, end = var_5151_end_0, end_mask = var_5151_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5151_cast")];
            tensor<int32, [4]> var_5155_begin_0 = const()[name = tensor<string, []>("op_5155_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_5155_end_0 = const()[name = tensor<string, []>("op_5155_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_5155_end_mask_0 = const()[name = tensor<string, []>("op_5155_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5155_cast = slice_by_index(begin = var_5155_begin_0, end = var_5155_end_0, end_mask = var_5155_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5155_cast")];
            tensor<int32, [4]> var_5159_begin_0 = const()[name = tensor<string, []>("op_5159_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_5159_end_0 = const()[name = tensor<string, []>("op_5159_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_5159_end_mask_0 = const()[name = tensor<string, []>("op_5159_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5159_cast = slice_by_index(begin = var_5159_begin_0, end = var_5159_end_0, end_mask = var_5159_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5159_cast")];
            tensor<int32, [4]> var_5163_begin_0 = const()[name = tensor<string, []>("op_5163_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_5163_end_0 = const()[name = tensor<string, []>("op_5163_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_5163_end_mask_0 = const()[name = tensor<string, []>("op_5163_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5163_cast = slice_by_index(begin = var_5163_begin_0, end = var_5163_end_0, end_mask = var_5163_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5163_cast")];
            tensor<int32, [4]> var_5167_begin_0 = const()[name = tensor<string, []>("op_5167_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_5167_end_0 = const()[name = tensor<string, []>("op_5167_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_5167_end_mask_0 = const()[name = tensor<string, []>("op_5167_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5167_cast = slice_by_index(begin = var_5167_begin_0, end = var_5167_end_0, end_mask = var_5167_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5167_cast")];
            tensor<int32, [4]> var_5171_begin_0 = const()[name = tensor<string, []>("op_5171_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_5171_end_0 = const()[name = tensor<string, []>("op_5171_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_5171_end_mask_0 = const()[name = tensor<string, []>("op_5171_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5171_cast = slice_by_index(begin = var_5171_begin_0, end = var_5171_end_0, end_mask = var_5171_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5171_cast")];
            tensor<int32, [4]> var_5175_begin_0 = const()[name = tensor<string, []>("op_5175_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_5175_end_0 = const()[name = tensor<string, []>("op_5175_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_5175_end_mask_0 = const()[name = tensor<string, []>("op_5175_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5175_cast = slice_by_index(begin = var_5175_begin_0, end = var_5175_end_0, end_mask = var_5175_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5175_cast")];
            tensor<int32, [4]> var_5179_begin_0 = const()[name = tensor<string, []>("op_5179_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_5179_end_0 = const()[name = tensor<string, []>("op_5179_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_5179_end_mask_0 = const()[name = tensor<string, []>("op_5179_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5179_cast = slice_by_index(begin = var_5179_begin_0, end = var_5179_end_0, end_mask = var_5179_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5179_cast")];
            tensor<int32, [4]> var_5183_begin_0 = const()[name = tensor<string, []>("op_5183_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_5183_end_0 = const()[name = tensor<string, []>("op_5183_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_5183_end_mask_0 = const()[name = tensor<string, []>("op_5183_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5183_cast = slice_by_index(begin = var_5183_begin_0, end = var_5183_end_0, end_mask = var_5183_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5183_cast")];
            tensor<int32, [4]> var_5187_begin_0 = const()[name = tensor<string, []>("op_5187_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_5187_end_0 = const()[name = tensor<string, []>("op_5187_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_5187_end_mask_0 = const()[name = tensor<string, []>("op_5187_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5187_cast = slice_by_index(begin = var_5187_begin_0, end = var_5187_end_0, end_mask = var_5187_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5187_cast")];
            tensor<int32, [4]> var_5191_begin_0 = const()[name = tensor<string, []>("op_5191_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_5191_end_0 = const()[name = tensor<string, []>("op_5191_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_5191_end_mask_0 = const()[name = tensor<string, []>("op_5191_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5191_cast = slice_by_index(begin = var_5191_begin_0, end = var_5191_end_0, end_mask = var_5191_end_mask_0, x = transpose_126)[name = tensor<string, []>("op_5191_cast")];
            tensor<int32, [4]> var_5193_begin_0 = const()[name = tensor<string, []>("op_5193_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5193_end_0 = const()[name = tensor<string, []>("op_5193_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_5193_end_mask_0 = const()[name = tensor<string, []>("op_5193_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5193_cast = slice_by_index(begin = var_5193_begin_0, end = var_5193_end_0, end_mask = var_5193_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5193_cast")];
            tensor<int32, [4]> var_5197_begin_0 = const()[name = tensor<string, []>("op_5197_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_5197_end_0 = const()[name = tensor<string, []>("op_5197_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_5197_end_mask_0 = const()[name = tensor<string, []>("op_5197_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5197_cast = slice_by_index(begin = var_5197_begin_0, end = var_5197_end_0, end_mask = var_5197_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5197_cast")];
            tensor<int32, [4]> var_5201_begin_0 = const()[name = tensor<string, []>("op_5201_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_5201_end_0 = const()[name = tensor<string, []>("op_5201_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_5201_end_mask_0 = const()[name = tensor<string, []>("op_5201_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5201_cast = slice_by_index(begin = var_5201_begin_0, end = var_5201_end_0, end_mask = var_5201_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5201_cast")];
            tensor<int32, [4]> var_5205_begin_0 = const()[name = tensor<string, []>("op_5205_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_5205_end_0 = const()[name = tensor<string, []>("op_5205_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_5205_end_mask_0 = const()[name = tensor<string, []>("op_5205_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5205_cast = slice_by_index(begin = var_5205_begin_0, end = var_5205_end_0, end_mask = var_5205_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5205_cast")];
            tensor<int32, [4]> var_5209_begin_0 = const()[name = tensor<string, []>("op_5209_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_5209_end_0 = const()[name = tensor<string, []>("op_5209_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_5209_end_mask_0 = const()[name = tensor<string, []>("op_5209_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5209_cast = slice_by_index(begin = var_5209_begin_0, end = var_5209_end_0, end_mask = var_5209_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5209_cast")];
            tensor<int32, [4]> var_5213_begin_0 = const()[name = tensor<string, []>("op_5213_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_5213_end_0 = const()[name = tensor<string, []>("op_5213_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_5213_end_mask_0 = const()[name = tensor<string, []>("op_5213_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5213_cast = slice_by_index(begin = var_5213_begin_0, end = var_5213_end_0, end_mask = var_5213_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5213_cast")];
            tensor<int32, [4]> var_5217_begin_0 = const()[name = tensor<string, []>("op_5217_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_5217_end_0 = const()[name = tensor<string, []>("op_5217_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_5217_end_mask_0 = const()[name = tensor<string, []>("op_5217_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5217_cast = slice_by_index(begin = var_5217_begin_0, end = var_5217_end_0, end_mask = var_5217_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5217_cast")];
            tensor<int32, [4]> var_5221_begin_0 = const()[name = tensor<string, []>("op_5221_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_5221_end_0 = const()[name = tensor<string, []>("op_5221_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_5221_end_mask_0 = const()[name = tensor<string, []>("op_5221_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5221_cast = slice_by_index(begin = var_5221_begin_0, end = var_5221_end_0, end_mask = var_5221_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5221_cast")];
            tensor<int32, [4]> var_5225_begin_0 = const()[name = tensor<string, []>("op_5225_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_5225_end_0 = const()[name = tensor<string, []>("op_5225_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_5225_end_mask_0 = const()[name = tensor<string, []>("op_5225_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5225_cast = slice_by_index(begin = var_5225_begin_0, end = var_5225_end_0, end_mask = var_5225_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5225_cast")];
            tensor<int32, [4]> var_5229_begin_0 = const()[name = tensor<string, []>("op_5229_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_5229_end_0 = const()[name = tensor<string, []>("op_5229_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_5229_end_mask_0 = const()[name = tensor<string, []>("op_5229_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5229_cast = slice_by_index(begin = var_5229_begin_0, end = var_5229_end_0, end_mask = var_5229_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5229_cast")];
            tensor<int32, [4]> var_5233_begin_0 = const()[name = tensor<string, []>("op_5233_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_5233_end_0 = const()[name = tensor<string, []>("op_5233_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_5233_end_mask_0 = const()[name = tensor<string, []>("op_5233_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5233_cast = slice_by_index(begin = var_5233_begin_0, end = var_5233_end_0, end_mask = var_5233_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5233_cast")];
            tensor<int32, [4]> var_5237_begin_0 = const()[name = tensor<string, []>("op_5237_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_5237_end_0 = const()[name = tensor<string, []>("op_5237_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_5237_end_mask_0 = const()[name = tensor<string, []>("op_5237_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5237_cast = slice_by_index(begin = var_5237_begin_0, end = var_5237_end_0, end_mask = var_5237_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5237_cast")];
            tensor<int32, [4]> var_5241_begin_0 = const()[name = tensor<string, []>("op_5241_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_5241_end_0 = const()[name = tensor<string, []>("op_5241_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_5241_end_mask_0 = const()[name = tensor<string, []>("op_5241_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5241_cast = slice_by_index(begin = var_5241_begin_0, end = var_5241_end_0, end_mask = var_5241_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5241_cast")];
            tensor<int32, [4]> var_5245_begin_0 = const()[name = tensor<string, []>("op_5245_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_5245_end_0 = const()[name = tensor<string, []>("op_5245_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_5245_end_mask_0 = const()[name = tensor<string, []>("op_5245_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5245_cast = slice_by_index(begin = var_5245_begin_0, end = var_5245_end_0, end_mask = var_5245_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5245_cast")];
            tensor<int32, [4]> var_5249_begin_0 = const()[name = tensor<string, []>("op_5249_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_5249_end_0 = const()[name = tensor<string, []>("op_5249_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_5249_end_mask_0 = const()[name = tensor<string, []>("op_5249_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5249_cast = slice_by_index(begin = var_5249_begin_0, end = var_5249_end_0, end_mask = var_5249_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5249_cast")];
            tensor<int32, [4]> var_5253_begin_0 = const()[name = tensor<string, []>("op_5253_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_5253_end_0 = const()[name = tensor<string, []>("op_5253_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_5253_end_mask_0 = const()[name = tensor<string, []>("op_5253_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5253_cast = slice_by_index(begin = var_5253_begin_0, end = var_5253_end_0, end_mask = var_5253_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5253_cast")];
            tensor<int32, [4]> var_5257_begin_0 = const()[name = tensor<string, []>("op_5257_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_5257_end_0 = const()[name = tensor<string, []>("op_5257_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_5257_end_mask_0 = const()[name = tensor<string, []>("op_5257_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5257_cast = slice_by_index(begin = var_5257_begin_0, end = var_5257_end_0, end_mask = var_5257_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5257_cast")];
            tensor<int32, [4]> var_5261_begin_0 = const()[name = tensor<string, []>("op_5261_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_5261_end_0 = const()[name = tensor<string, []>("op_5261_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_5261_end_mask_0 = const()[name = tensor<string, []>("op_5261_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5261_cast = slice_by_index(begin = var_5261_begin_0, end = var_5261_end_0, end_mask = var_5261_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5261_cast")];
            tensor<int32, [4]> var_5265_begin_0 = const()[name = tensor<string, []>("op_5265_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_5265_end_0 = const()[name = tensor<string, []>("op_5265_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_5265_end_mask_0 = const()[name = tensor<string, []>("op_5265_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5265_cast = slice_by_index(begin = var_5265_begin_0, end = var_5265_end_0, end_mask = var_5265_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5265_cast")];
            tensor<int32, [4]> var_5269_begin_0 = const()[name = tensor<string, []>("op_5269_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_5269_end_0 = const()[name = tensor<string, []>("op_5269_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_5269_end_mask_0 = const()[name = tensor<string, []>("op_5269_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5269_cast = slice_by_index(begin = var_5269_begin_0, end = var_5269_end_0, end_mask = var_5269_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_5269_cast")];
            tensor<string, []> var_5273_equation_0 = const()[name = tensor<string, []>("op_5273_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5273_cast = einsum(equation = var_5273_equation_0, values = (var_5115_cast, var_5032_cast))[name = tensor<string, []>("op_5273_cast")];
            tensor<fp16, []> var_5274_to_fp16 = const()[name = tensor<string, []>("op_5274_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_361_cast = mul(x = var_5273_cast, y = var_5274_to_fp16)[name = tensor<string, []>("aw_361_cast")];
            tensor<string, []> var_5277_equation_0 = const()[name = tensor<string, []>("op_5277_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5277_cast = einsum(equation = var_5277_equation_0, values = (var_5119_cast, var_5036_cast))[name = tensor<string, []>("op_5277_cast")];
            tensor<fp16, []> var_5278_to_fp16 = const()[name = tensor<string, []>("op_5278_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_363_cast = mul(x = var_5277_cast, y = var_5278_to_fp16)[name = tensor<string, []>("aw_363_cast")];
            tensor<string, []> var_5281_equation_0 = const()[name = tensor<string, []>("op_5281_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5281_cast = einsum(equation = var_5281_equation_0, values = (var_5123_cast, var_5040_cast))[name = tensor<string, []>("op_5281_cast")];
            tensor<fp16, []> var_5282_to_fp16 = const()[name = tensor<string, []>("op_5282_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_365_cast = mul(x = var_5281_cast, y = var_5282_to_fp16)[name = tensor<string, []>("aw_365_cast")];
            tensor<string, []> var_5285_equation_0 = const()[name = tensor<string, []>("op_5285_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5285_cast = einsum(equation = var_5285_equation_0, values = (var_5127_cast, var_5044_cast))[name = tensor<string, []>("op_5285_cast")];
            tensor<fp16, []> var_5286_to_fp16 = const()[name = tensor<string, []>("op_5286_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_367_cast = mul(x = var_5285_cast, y = var_5286_to_fp16)[name = tensor<string, []>("aw_367_cast")];
            tensor<string, []> var_5289_equation_0 = const()[name = tensor<string, []>("op_5289_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5289_cast = einsum(equation = var_5289_equation_0, values = (var_5131_cast, var_5048_cast))[name = tensor<string, []>("op_5289_cast")];
            tensor<fp16, []> var_5290_to_fp16 = const()[name = tensor<string, []>("op_5290_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_369_cast = mul(x = var_5289_cast, y = var_5290_to_fp16)[name = tensor<string, []>("aw_369_cast")];
            tensor<string, []> var_5293_equation_0 = const()[name = tensor<string, []>("op_5293_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5293_cast = einsum(equation = var_5293_equation_0, values = (var_5135_cast, var_5052_cast))[name = tensor<string, []>("op_5293_cast")];
            tensor<fp16, []> var_5294_to_fp16 = const()[name = tensor<string, []>("op_5294_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_371_cast = mul(x = var_5293_cast, y = var_5294_to_fp16)[name = tensor<string, []>("aw_371_cast")];
            tensor<string, []> var_5297_equation_0 = const()[name = tensor<string, []>("op_5297_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5297_cast = einsum(equation = var_5297_equation_0, values = (var_5139_cast, var_5056_cast))[name = tensor<string, []>("op_5297_cast")];
            tensor<fp16, []> var_5298_to_fp16 = const()[name = tensor<string, []>("op_5298_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_373_cast = mul(x = var_5297_cast, y = var_5298_to_fp16)[name = tensor<string, []>("aw_373_cast")];
            tensor<string, []> var_5301_equation_0 = const()[name = tensor<string, []>("op_5301_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5301_cast = einsum(equation = var_5301_equation_0, values = (var_5143_cast, var_5060_cast))[name = tensor<string, []>("op_5301_cast")];
            tensor<fp16, []> var_5302_to_fp16 = const()[name = tensor<string, []>("op_5302_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_375_cast = mul(x = var_5301_cast, y = var_5302_to_fp16)[name = tensor<string, []>("aw_375_cast")];
            tensor<string, []> var_5305_equation_0 = const()[name = tensor<string, []>("op_5305_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5305_cast = einsum(equation = var_5305_equation_0, values = (var_5147_cast, var_5064_cast))[name = tensor<string, []>("op_5305_cast")];
            tensor<fp16, []> var_5306_to_fp16 = const()[name = tensor<string, []>("op_5306_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_377_cast = mul(x = var_5305_cast, y = var_5306_to_fp16)[name = tensor<string, []>("aw_377_cast")];
            tensor<string, []> var_5309_equation_0 = const()[name = tensor<string, []>("op_5309_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5309_cast = einsum(equation = var_5309_equation_0, values = (var_5151_cast, var_5068_cast))[name = tensor<string, []>("op_5309_cast")];
            tensor<fp16, []> var_5310_to_fp16 = const()[name = tensor<string, []>("op_5310_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_379_cast = mul(x = var_5309_cast, y = var_5310_to_fp16)[name = tensor<string, []>("aw_379_cast")];
            tensor<string, []> var_5313_equation_0 = const()[name = tensor<string, []>("op_5313_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5313_cast = einsum(equation = var_5313_equation_0, values = (var_5155_cast, var_5072_cast))[name = tensor<string, []>("op_5313_cast")];
            tensor<fp16, []> var_5314_to_fp16 = const()[name = tensor<string, []>("op_5314_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_381_cast = mul(x = var_5313_cast, y = var_5314_to_fp16)[name = tensor<string, []>("aw_381_cast")];
            tensor<string, []> var_5317_equation_0 = const()[name = tensor<string, []>("op_5317_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5317_cast = einsum(equation = var_5317_equation_0, values = (var_5159_cast, var_5076_cast))[name = tensor<string, []>("op_5317_cast")];
            tensor<fp16, []> var_5318_to_fp16 = const()[name = tensor<string, []>("op_5318_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_383_cast = mul(x = var_5317_cast, y = var_5318_to_fp16)[name = tensor<string, []>("aw_383_cast")];
            tensor<string, []> var_5321_equation_0 = const()[name = tensor<string, []>("op_5321_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5321_cast = einsum(equation = var_5321_equation_0, values = (var_5163_cast, var_5080_cast))[name = tensor<string, []>("op_5321_cast")];
            tensor<fp16, []> var_5322_to_fp16 = const()[name = tensor<string, []>("op_5322_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_385_cast = mul(x = var_5321_cast, y = var_5322_to_fp16)[name = tensor<string, []>("aw_385_cast")];
            tensor<string, []> var_5325_equation_0 = const()[name = tensor<string, []>("op_5325_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5325_cast = einsum(equation = var_5325_equation_0, values = (var_5167_cast, var_5084_cast))[name = tensor<string, []>("op_5325_cast")];
            tensor<fp16, []> var_5326_to_fp16 = const()[name = tensor<string, []>("op_5326_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_387_cast = mul(x = var_5325_cast, y = var_5326_to_fp16)[name = tensor<string, []>("aw_387_cast")];
            tensor<string, []> var_5329_equation_0 = const()[name = tensor<string, []>("op_5329_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5329_cast = einsum(equation = var_5329_equation_0, values = (var_5171_cast, var_5088_cast))[name = tensor<string, []>("op_5329_cast")];
            tensor<fp16, []> var_5330_to_fp16 = const()[name = tensor<string, []>("op_5330_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_389_cast = mul(x = var_5329_cast, y = var_5330_to_fp16)[name = tensor<string, []>("aw_389_cast")];
            tensor<string, []> var_5333_equation_0 = const()[name = tensor<string, []>("op_5333_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5333_cast = einsum(equation = var_5333_equation_0, values = (var_5175_cast, var_5092_cast))[name = tensor<string, []>("op_5333_cast")];
            tensor<fp16, []> var_5334_to_fp16 = const()[name = tensor<string, []>("op_5334_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_391_cast = mul(x = var_5333_cast, y = var_5334_to_fp16)[name = tensor<string, []>("aw_391_cast")];
            tensor<string, []> var_5337_equation_0 = const()[name = tensor<string, []>("op_5337_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5337_cast = einsum(equation = var_5337_equation_0, values = (var_5179_cast, var_5096_cast))[name = tensor<string, []>("op_5337_cast")];
            tensor<fp16, []> var_5338_to_fp16 = const()[name = tensor<string, []>("op_5338_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_393_cast = mul(x = var_5337_cast, y = var_5338_to_fp16)[name = tensor<string, []>("aw_393_cast")];
            tensor<string, []> var_5341_equation_0 = const()[name = tensor<string, []>("op_5341_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5341_cast = einsum(equation = var_5341_equation_0, values = (var_5183_cast, var_5100_cast))[name = tensor<string, []>("op_5341_cast")];
            tensor<fp16, []> var_5342_to_fp16 = const()[name = tensor<string, []>("op_5342_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_395_cast = mul(x = var_5341_cast, y = var_5342_to_fp16)[name = tensor<string, []>("aw_395_cast")];
            tensor<string, []> var_5345_equation_0 = const()[name = tensor<string, []>("op_5345_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5345_cast = einsum(equation = var_5345_equation_0, values = (var_5187_cast, var_5104_cast))[name = tensor<string, []>("op_5345_cast")];
            tensor<fp16, []> var_5346_to_fp16 = const()[name = tensor<string, []>("op_5346_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_397_cast = mul(x = var_5345_cast, y = var_5346_to_fp16)[name = tensor<string, []>("aw_397_cast")];
            tensor<string, []> var_5349_equation_0 = const()[name = tensor<string, []>("op_5349_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_5349_cast = einsum(equation = var_5349_equation_0, values = (var_5191_cast, var_5108_cast))[name = tensor<string, []>("op_5349_cast")];
            tensor<fp16, []> var_5350_to_fp16 = const()[name = tensor<string, []>("op_5350_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_399_cast = mul(x = var_5349_cast, y = var_5350_to_fp16)[name = tensor<string, []>("aw_399_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5352_cast = softmax(axis = var_2634, x = aw_361_cast)[name = tensor<string, []>("op_5352_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5353_cast = softmax(axis = var_2634, x = aw_363_cast)[name = tensor<string, []>("op_5353_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5354_cast = softmax(axis = var_2634, x = aw_365_cast)[name = tensor<string, []>("op_5354_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5355_cast = softmax(axis = var_2634, x = aw_367_cast)[name = tensor<string, []>("op_5355_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5356_cast = softmax(axis = var_2634, x = aw_369_cast)[name = tensor<string, []>("op_5356_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5357_cast = softmax(axis = var_2634, x = aw_371_cast)[name = tensor<string, []>("op_5357_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5358_cast = softmax(axis = var_2634, x = aw_373_cast)[name = tensor<string, []>("op_5358_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5359_cast = softmax(axis = var_2634, x = aw_375_cast)[name = tensor<string, []>("op_5359_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5360_cast = softmax(axis = var_2634, x = aw_377_cast)[name = tensor<string, []>("op_5360_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5361_cast = softmax(axis = var_2634, x = aw_379_cast)[name = tensor<string, []>("op_5361_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5362_cast = softmax(axis = var_2634, x = aw_381_cast)[name = tensor<string, []>("op_5362_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5363_cast = softmax(axis = var_2634, x = aw_383_cast)[name = tensor<string, []>("op_5363_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5364_cast = softmax(axis = var_2634, x = aw_385_cast)[name = tensor<string, []>("op_5364_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5365_cast = softmax(axis = var_2634, x = aw_387_cast)[name = tensor<string, []>("op_5365_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5366_cast = softmax(axis = var_2634, x = aw_389_cast)[name = tensor<string, []>("op_5366_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5367_cast = softmax(axis = var_2634, x = aw_391_cast)[name = tensor<string, []>("op_5367_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5368_cast = softmax(axis = var_2634, x = aw_393_cast)[name = tensor<string, []>("op_5368_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5369_cast = softmax(axis = var_2634, x = aw_395_cast)[name = tensor<string, []>("op_5369_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5370_cast = softmax(axis = var_2634, x = aw_397_cast)[name = tensor<string, []>("op_5370_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_5371_cast = softmax(axis = var_2634, x = aw_399_cast)[name = tensor<string, []>("op_5371_cast")];
            tensor<string, []> var_5373_equation_0 = const()[name = tensor<string, []>("op_5373_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5373_cast = einsum(equation = var_5373_equation_0, values = (var_5193_cast, var_5352_cast))[name = tensor<string, []>("op_5373_cast")];
            tensor<string, []> var_5375_equation_0 = const()[name = tensor<string, []>("op_5375_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5375_cast = einsum(equation = var_5375_equation_0, values = (var_5197_cast, var_5353_cast))[name = tensor<string, []>("op_5375_cast")];
            tensor<string, []> var_5377_equation_0 = const()[name = tensor<string, []>("op_5377_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5377_cast = einsum(equation = var_5377_equation_0, values = (var_5201_cast, var_5354_cast))[name = tensor<string, []>("op_5377_cast")];
            tensor<string, []> var_5379_equation_0 = const()[name = tensor<string, []>("op_5379_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5379_cast = einsum(equation = var_5379_equation_0, values = (var_5205_cast, var_5355_cast))[name = tensor<string, []>("op_5379_cast")];
            tensor<string, []> var_5381_equation_0 = const()[name = tensor<string, []>("op_5381_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5381_cast = einsum(equation = var_5381_equation_0, values = (var_5209_cast, var_5356_cast))[name = tensor<string, []>("op_5381_cast")];
            tensor<string, []> var_5383_equation_0 = const()[name = tensor<string, []>("op_5383_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5383_cast = einsum(equation = var_5383_equation_0, values = (var_5213_cast, var_5357_cast))[name = tensor<string, []>("op_5383_cast")];
            tensor<string, []> var_5385_equation_0 = const()[name = tensor<string, []>("op_5385_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5385_cast = einsum(equation = var_5385_equation_0, values = (var_5217_cast, var_5358_cast))[name = tensor<string, []>("op_5385_cast")];
            tensor<string, []> var_5387_equation_0 = const()[name = tensor<string, []>("op_5387_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5387_cast = einsum(equation = var_5387_equation_0, values = (var_5221_cast, var_5359_cast))[name = tensor<string, []>("op_5387_cast")];
            tensor<string, []> var_5389_equation_0 = const()[name = tensor<string, []>("op_5389_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5389_cast = einsum(equation = var_5389_equation_0, values = (var_5225_cast, var_5360_cast))[name = tensor<string, []>("op_5389_cast")];
            tensor<string, []> var_5391_equation_0 = const()[name = tensor<string, []>("op_5391_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5391_cast = einsum(equation = var_5391_equation_0, values = (var_5229_cast, var_5361_cast))[name = tensor<string, []>("op_5391_cast")];
            tensor<string, []> var_5393_equation_0 = const()[name = tensor<string, []>("op_5393_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5393_cast = einsum(equation = var_5393_equation_0, values = (var_5233_cast, var_5362_cast))[name = tensor<string, []>("op_5393_cast")];
            tensor<string, []> var_5395_equation_0 = const()[name = tensor<string, []>("op_5395_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5395_cast = einsum(equation = var_5395_equation_0, values = (var_5237_cast, var_5363_cast))[name = tensor<string, []>("op_5395_cast")];
            tensor<string, []> var_5397_equation_0 = const()[name = tensor<string, []>("op_5397_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5397_cast = einsum(equation = var_5397_equation_0, values = (var_5241_cast, var_5364_cast))[name = tensor<string, []>("op_5397_cast")];
            tensor<string, []> var_5399_equation_0 = const()[name = tensor<string, []>("op_5399_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5399_cast = einsum(equation = var_5399_equation_0, values = (var_5245_cast, var_5365_cast))[name = tensor<string, []>("op_5399_cast")];
            tensor<string, []> var_5401_equation_0 = const()[name = tensor<string, []>("op_5401_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5401_cast = einsum(equation = var_5401_equation_0, values = (var_5249_cast, var_5366_cast))[name = tensor<string, []>("op_5401_cast")];
            tensor<string, []> var_5403_equation_0 = const()[name = tensor<string, []>("op_5403_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5403_cast = einsum(equation = var_5403_equation_0, values = (var_5253_cast, var_5367_cast))[name = tensor<string, []>("op_5403_cast")];
            tensor<string, []> var_5405_equation_0 = const()[name = tensor<string, []>("op_5405_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5405_cast = einsum(equation = var_5405_equation_0, values = (var_5257_cast, var_5368_cast))[name = tensor<string, []>("op_5405_cast")];
            tensor<string, []> var_5407_equation_0 = const()[name = tensor<string, []>("op_5407_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5407_cast = einsum(equation = var_5407_equation_0, values = (var_5261_cast, var_5369_cast))[name = tensor<string, []>("op_5407_cast")];
            tensor<string, []> var_5409_equation_0 = const()[name = tensor<string, []>("op_5409_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5409_cast = einsum(equation = var_5409_equation_0, values = (var_5265_cast, var_5370_cast))[name = tensor<string, []>("op_5409_cast")];
            tensor<string, []> var_5411_equation_0 = const()[name = tensor<string, []>("op_5411_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5411_cast = einsum(equation = var_5411_equation_0, values = (var_5269_cast, var_5371_cast))[name = tensor<string, []>("op_5411_cast")];
            tensor<bool, []> input_153_interleave_0 = const()[name = tensor<string, []>("input_153_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_153_cast = concat(axis = var_2634, interleave = input_153_interleave_0, values = (var_5373_cast, var_5375_cast, var_5377_cast, var_5379_cast, var_5381_cast, var_5383_cast, var_5385_cast, var_5387_cast, var_5389_cast, var_5391_cast, var_5393_cast, var_5395_cast, var_5397_cast, var_5399_cast, var_5401_cast, var_5403_cast, var_5405_cast, var_5407_cast, var_5409_cast, var_5411_cast))[name = tensor<string, []>("input_153_cast")];
            tensor<int32, [2]> var_5417 = const()[name = tensor<string, []>("op_5417"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5419 = const()[name = tensor<string, []>("op_5419"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_5421_pad_type_0 = const()[name = tensor<string, []>("op_5421_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_5421_pad_0 = const()[name = tensor<string, []>("op_5421_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(364075136)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(367352000)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_5421_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16, dilations = var_5419, groups = var_2634, pad = var_5421_pad_0, pad_type = var_5421_pad_type_0, strides = var_5417, weight = down_blocks_2_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16, x = input_153_cast)[name = tensor<string, []>("op_5421_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_41_cast = add(x = var_5421_cast, y = inputs_39_cast)[name = tensor<string, []>("inputs_41_cast")];
            tensor<int32, [1]> var_5425 = const()[name = tensor<string, []>("op_5425"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_41_cast = reduce_mean(axes = var_5425, keep_dims = var_2629, x = inputs_41_cast)[name = tensor<string, []>("channels_mean_41_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_41_cast = sub(x = inputs_41_cast, y = channels_mean_41_cast)[name = tensor<string, []>("zero_mean_41_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_41_cast = mul(x = zero_mean_41_cast, y = zero_mean_41_cast)[name = tensor<string, []>("zero_mean_sq_41_cast")];
            tensor<int32, [1]> var_5429 = const()[name = tensor<string, []>("op_5429"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_5430_cast = reduce_mean(axes = var_5429, keep_dims = var_2629, x = zero_mean_sq_41_cast)[name = tensor<string, []>("op_5430_cast")];
            tensor<fp16, []> var_5431_to_fp16 = const()[name = tensor<string, []>("op_5431_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_5432_cast = add(x = var_5430_cast, y = var_5431_to_fp16)[name = tensor<string, []>("op_5432_cast")];
            tensor<fp16, []> denom_41_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_41_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_41_cast = rsqrt(epsilon = denom_41_epsilon_0_to_fp16, x = var_5432_cast)[name = tensor<string, []>("denom_41_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_41_cast = mul(x = zero_mean_41_cast, y = denom_41_cast)[name = tensor<string, []>("out_41_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5436_to_fp16 = const()[name = tensor<string, []>("op_5436_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(367354624)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_5437_cast = add(x = out_41_cast, y = var_5436_to_fp16)[name = tensor<string, []>("op_5437_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5439_to_fp16 = const()[name = tensor<string, []>("op_5439_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(367357248)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_155_cast = mul(x = var_5437_cast, y = var_5439_to_fp16)[name = tensor<string, []>("input_155_cast")];
            tensor<int32, [2]> var_5447 = const()[name = tensor<string, []>("op_5447"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5449 = const()[name = tensor<string, []>("op_5449"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_5451_pad_type_0 = const()[name = tensor<string, []>("op_5451_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_5451_pad_0 = const()[name = tensor<string, []>("op_5451_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(367359872)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(393574336)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_5451_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16, dilations = var_5449, groups = var_2634, pad = var_5451_pad_0, pad_type = var_5451_pad_type_0, strides = var_5447, weight = down_blocks_2_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16, x = input_155_cast)[name = tensor<string, []>("op_5451_cast")];
            tensor<int32, [2]> var_5452_split_sizes_0 = const()[name = tensor<string, []>("op_5452_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_5452_axis_0 = const()[name = tensor<string, []>("op_5452_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_5452_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_5452_cast_1 = split(axis = var_5452_axis_0, split_sizes = var_5452_split_sizes_0, x = var_5451_cast)[name = tensor<string, []>("op_5452_cast")];
            tensor<string, []> var_5454_mode_0 = const()[name = tensor<string, []>("op_5454_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_5454_cast = gelu(mode = var_5454_mode_0, x = var_5452_cast_1)[name = tensor<string, []>("op_5454_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_157_cast = mul(x = var_5452_cast_0, y = var_5454_cast)[name = tensor<string, []>("input_157_cast")];
            tensor<int32, [2]> var_5458 = const()[name = tensor<string, []>("op_5458"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5460 = const()[name = tensor<string, []>("op_5460"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_5462_pad_type_0 = const()[name = tensor<string, []>("op_5462_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_5462_pad_0 = const()[name = tensor<string, []>("op_5462_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(393594880)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(406702144)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_5462_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16, dilations = var_5460, groups = var_2634, pad = var_5462_pad_0, pad_type = var_5462_pad_type_0, strides = var_5458, weight = down_blocks_2_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16, x = input_157_cast)[name = tensor<string, []>("op_5462_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_43_cast = add(x = var_5462_cast, y = inputs_41_cast)[name = tensor<string, []>("inputs_43_cast")];
            tensor<int32, [1]> var_5472 = const()[name = tensor<string, []>("op_5472"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_43_cast = reduce_mean(axes = var_5472, keep_dims = var_2629, x = inputs_43_cast)[name = tensor<string, []>("channels_mean_43_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_43_cast = sub(x = inputs_43_cast, y = channels_mean_43_cast)[name = tensor<string, []>("zero_mean_43_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_43_cast = mul(x = zero_mean_43_cast, y = zero_mean_43_cast)[name = tensor<string, []>("zero_mean_sq_43_cast")];
            tensor<int32, [1]> var_5476 = const()[name = tensor<string, []>("op_5476"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_5477_cast = reduce_mean(axes = var_5476, keep_dims = var_2629, x = zero_mean_sq_43_cast)[name = tensor<string, []>("op_5477_cast")];
            tensor<fp16, []> var_5478_to_fp16 = const()[name = tensor<string, []>("op_5478_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_5479_cast = add(x = var_5477_cast, y = var_5478_to_fp16)[name = tensor<string, []>("op_5479_cast")];
            tensor<fp16, []> denom_43_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_43_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_43_cast = rsqrt(epsilon = denom_43_epsilon_0_to_fp16, x = var_5479_cast)[name = tensor<string, []>("denom_43_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_43_cast = mul(x = zero_mean_43_cast, y = denom_43_cast)[name = tensor<string, []>("out_43_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5483_to_fp16 = const()[name = tensor<string, []>("op_5483_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(406704768)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_5484_cast = add(x = out_43_cast, y = var_5483_to_fp16)[name = tensor<string, []>("op_5484_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5486_to_fp16 = const()[name = tensor<string, []>("op_5486_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(406707392)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_83_cast = mul(x = var_5484_cast, y = var_5486_to_fp16)[name = tensor<string, []>("hidden_states_83_cast")];
            tensor<int32, [2]> var_5493 = const()[name = tensor<string, []>("op_5493"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5495 = const()[name = tensor<string, []>("op_5495"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_29_pad_type_0 = const()[name = tensor<string, []>("q_29_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_29_pad_0 = const()[name = tensor<string, []>("q_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(406710016)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_29_cast = conv(dilations = var_5495, groups = var_2634, pad = q_29_pad_0, pad_type = q_29_pad_type_0, strides = var_5493, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16, x = hidden_states_83_cast)[name = tensor<string, []>("q_29_cast")];
            tensor<int32, [2]> var_5499 = const()[name = tensor<string, []>("op_5499"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5501 = const()[name = tensor<string, []>("op_5501"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_57_pad_type_0 = const()[name = tensor<string, []>("k_57_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_57_pad_0 = const()[name = tensor<string, []>("k_57_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(409986880)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_57_cast = conv(dilations = var_5501, groups = var_2634, pad = k_57_pad_0, pad_type = k_57_pad_type_0, strides = var_5499, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16, x = hidden_states_83_cast)[name = tensor<string, []>("k_57_cast")];
            tensor<int32, [2]> var_5505 = const()[name = tensor<string, []>("op_5505"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5507 = const()[name = tensor<string, []>("op_5507"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_29_pad_type_0 = const()[name = tensor<string, []>("v_29_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_29_pad_0 = const()[name = tensor<string, []>("v_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(413263744)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_29_cast = conv(dilations = var_5507, groups = var_2634, pad = v_29_pad_0, pad_type = v_29_pad_type_0, strides = var_5505, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16, x = hidden_states_83_cast)[name = tensor<string, []>("v_29_cast")];
            tensor<int32, [4]> var_5511_begin_0 = const()[name = tensor<string, []>("op_5511_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5511_end_0 = const()[name = tensor<string, []>("op_5511_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_5511_end_mask_0 = const()[name = tensor<string, []>("op_5511_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5511_cast = slice_by_index(begin = var_5511_begin_0, end = var_5511_end_0, end_mask = var_5511_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5511_cast")];
            tensor<int32, [4]> var_5515_begin_0 = const()[name = tensor<string, []>("op_5515_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_5515_end_0 = const()[name = tensor<string, []>("op_5515_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_5515_end_mask_0 = const()[name = tensor<string, []>("op_5515_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5515_cast = slice_by_index(begin = var_5515_begin_0, end = var_5515_end_0, end_mask = var_5515_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5515_cast")];
            tensor<int32, [4]> var_5519_begin_0 = const()[name = tensor<string, []>("op_5519_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_5519_end_0 = const()[name = tensor<string, []>("op_5519_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_5519_end_mask_0 = const()[name = tensor<string, []>("op_5519_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5519_cast = slice_by_index(begin = var_5519_begin_0, end = var_5519_end_0, end_mask = var_5519_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5519_cast")];
            tensor<int32, [4]> var_5523_begin_0 = const()[name = tensor<string, []>("op_5523_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_5523_end_0 = const()[name = tensor<string, []>("op_5523_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_5523_end_mask_0 = const()[name = tensor<string, []>("op_5523_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5523_cast = slice_by_index(begin = var_5523_begin_0, end = var_5523_end_0, end_mask = var_5523_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5523_cast")];
            tensor<int32, [4]> var_5527_begin_0 = const()[name = tensor<string, []>("op_5527_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_5527_end_0 = const()[name = tensor<string, []>("op_5527_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_5527_end_mask_0 = const()[name = tensor<string, []>("op_5527_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5527_cast = slice_by_index(begin = var_5527_begin_0, end = var_5527_end_0, end_mask = var_5527_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5527_cast")];
            tensor<int32, [4]> var_5531_begin_0 = const()[name = tensor<string, []>("op_5531_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_5531_end_0 = const()[name = tensor<string, []>("op_5531_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_5531_end_mask_0 = const()[name = tensor<string, []>("op_5531_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5531_cast = slice_by_index(begin = var_5531_begin_0, end = var_5531_end_0, end_mask = var_5531_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5531_cast")];
            tensor<int32, [4]> var_5535_begin_0 = const()[name = tensor<string, []>("op_5535_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_5535_end_0 = const()[name = tensor<string, []>("op_5535_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_5535_end_mask_0 = const()[name = tensor<string, []>("op_5535_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5535_cast = slice_by_index(begin = var_5535_begin_0, end = var_5535_end_0, end_mask = var_5535_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5535_cast")];
            tensor<int32, [4]> var_5539_begin_0 = const()[name = tensor<string, []>("op_5539_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_5539_end_0 = const()[name = tensor<string, []>("op_5539_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_5539_end_mask_0 = const()[name = tensor<string, []>("op_5539_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5539_cast = slice_by_index(begin = var_5539_begin_0, end = var_5539_end_0, end_mask = var_5539_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5539_cast")];
            tensor<int32, [4]> var_5543_begin_0 = const()[name = tensor<string, []>("op_5543_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_5543_end_0 = const()[name = tensor<string, []>("op_5543_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_5543_end_mask_0 = const()[name = tensor<string, []>("op_5543_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5543_cast = slice_by_index(begin = var_5543_begin_0, end = var_5543_end_0, end_mask = var_5543_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5543_cast")];
            tensor<int32, [4]> var_5547_begin_0 = const()[name = tensor<string, []>("op_5547_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_5547_end_0 = const()[name = tensor<string, []>("op_5547_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_5547_end_mask_0 = const()[name = tensor<string, []>("op_5547_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5547_cast = slice_by_index(begin = var_5547_begin_0, end = var_5547_end_0, end_mask = var_5547_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5547_cast")];
            tensor<int32, [4]> var_5551_begin_0 = const()[name = tensor<string, []>("op_5551_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_5551_end_0 = const()[name = tensor<string, []>("op_5551_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_5551_end_mask_0 = const()[name = tensor<string, []>("op_5551_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5551_cast = slice_by_index(begin = var_5551_begin_0, end = var_5551_end_0, end_mask = var_5551_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5551_cast")];
            tensor<int32, [4]> var_5555_begin_0 = const()[name = tensor<string, []>("op_5555_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_5555_end_0 = const()[name = tensor<string, []>("op_5555_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_5555_end_mask_0 = const()[name = tensor<string, []>("op_5555_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5555_cast = slice_by_index(begin = var_5555_begin_0, end = var_5555_end_0, end_mask = var_5555_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5555_cast")];
            tensor<int32, [4]> var_5559_begin_0 = const()[name = tensor<string, []>("op_5559_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_5559_end_0 = const()[name = tensor<string, []>("op_5559_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_5559_end_mask_0 = const()[name = tensor<string, []>("op_5559_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5559_cast = slice_by_index(begin = var_5559_begin_0, end = var_5559_end_0, end_mask = var_5559_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5559_cast")];
            tensor<int32, [4]> var_5563_begin_0 = const()[name = tensor<string, []>("op_5563_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_5563_end_0 = const()[name = tensor<string, []>("op_5563_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_5563_end_mask_0 = const()[name = tensor<string, []>("op_5563_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5563_cast = slice_by_index(begin = var_5563_begin_0, end = var_5563_end_0, end_mask = var_5563_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5563_cast")];
            tensor<int32, [4]> var_5567_begin_0 = const()[name = tensor<string, []>("op_5567_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_5567_end_0 = const()[name = tensor<string, []>("op_5567_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_5567_end_mask_0 = const()[name = tensor<string, []>("op_5567_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5567_cast = slice_by_index(begin = var_5567_begin_0, end = var_5567_end_0, end_mask = var_5567_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5567_cast")];
            tensor<int32, [4]> var_5571_begin_0 = const()[name = tensor<string, []>("op_5571_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_5571_end_0 = const()[name = tensor<string, []>("op_5571_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_5571_end_mask_0 = const()[name = tensor<string, []>("op_5571_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5571_cast = slice_by_index(begin = var_5571_begin_0, end = var_5571_end_0, end_mask = var_5571_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5571_cast")];
            tensor<int32, [4]> var_5575_begin_0 = const()[name = tensor<string, []>("op_5575_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_5575_end_0 = const()[name = tensor<string, []>("op_5575_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_5575_end_mask_0 = const()[name = tensor<string, []>("op_5575_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5575_cast = slice_by_index(begin = var_5575_begin_0, end = var_5575_end_0, end_mask = var_5575_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5575_cast")];
            tensor<int32, [4]> var_5579_begin_0 = const()[name = tensor<string, []>("op_5579_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_5579_end_0 = const()[name = tensor<string, []>("op_5579_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_5579_end_mask_0 = const()[name = tensor<string, []>("op_5579_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5579_cast = slice_by_index(begin = var_5579_begin_0, end = var_5579_end_0, end_mask = var_5579_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5579_cast")];
            tensor<int32, [4]> var_5583_begin_0 = const()[name = tensor<string, []>("op_5583_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_5583_end_0 = const()[name = tensor<string, []>("op_5583_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_5583_end_mask_0 = const()[name = tensor<string, []>("op_5583_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5583_cast = slice_by_index(begin = var_5583_begin_0, end = var_5583_end_0, end_mask = var_5583_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5583_cast")];
            tensor<int32, [4]> var_5587_begin_0 = const()[name = tensor<string, []>("op_5587_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_5587_end_0 = const()[name = tensor<string, []>("op_5587_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_5587_end_mask_0 = const()[name = tensor<string, []>("op_5587_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5587_cast = slice_by_index(begin = var_5587_begin_0, end = var_5587_end_0, end_mask = var_5587_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_5587_cast")];
            tensor<int32, [4]> k_59_perm_0 = const()[name = tensor<string, []>("k_59_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_5594_begin_0 = const()[name = tensor<string, []>("op_5594_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5594_end_0 = const()[name = tensor<string, []>("op_5594_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_5594_end_mask_0 = const()[name = tensor<string, []>("op_5594_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_125 = transpose(perm = k_59_perm_0, x = k_57_cast)[name = tensor<string, []>("transpose_125")];
            tensor<fp16, [2, 1024, 1, 64]> var_5594_cast = slice_by_index(begin = var_5594_begin_0, end = var_5594_end_0, end_mask = var_5594_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5594_cast")];
            tensor<int32, [4]> var_5598_begin_0 = const()[name = tensor<string, []>("op_5598_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_5598_end_0 = const()[name = tensor<string, []>("op_5598_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_5598_end_mask_0 = const()[name = tensor<string, []>("op_5598_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_5598_cast = slice_by_index(begin = var_5598_begin_0, end = var_5598_end_0, end_mask = var_5598_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5598_cast")];
            tensor<int32, [4]> var_5602_begin_0 = const()[name = tensor<string, []>("op_5602_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_5602_end_0 = const()[name = tensor<string, []>("op_5602_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_5602_end_mask_0 = const()[name = tensor<string, []>("op_5602_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_5602_cast = slice_by_index(begin = var_5602_begin_0, end = var_5602_end_0, end_mask = var_5602_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5602_cast")];
            tensor<int32, [4]> var_5606_begin_0 = const()[name = tensor<string, []>("op_5606_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_5606_end_0 = const()[name = tensor<string, []>("op_5606_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_5606_end_mask_0 = const()[name = tensor<string, []>("op_5606_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_5606_cast = slice_by_index(begin = var_5606_begin_0, end = var_5606_end_0, end_mask = var_5606_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5606_cast")];
            tensor<int32, [4]> var_5610_begin_0 = const()[name = tensor<string, []>("op_5610_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_5610_end_0 = const()[name = tensor<string, []>("op_5610_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_5610_end_mask_0 = const()[name = tensor<string, []>("op_5610_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_5610_cast = slice_by_index(begin = var_5610_begin_0, end = var_5610_end_0, end_mask = var_5610_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5610_cast")];
            tensor<int32, [4]> var_5614_begin_0 = const()[name = tensor<string, []>("op_5614_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_5614_end_0 = const()[name = tensor<string, []>("op_5614_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_5614_end_mask_0 = const()[name = tensor<string, []>("op_5614_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_5614_cast = slice_by_index(begin = var_5614_begin_0, end = var_5614_end_0, end_mask = var_5614_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5614_cast")];
            tensor<int32, [4]> var_5618_begin_0 = const()[name = tensor<string, []>("op_5618_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_5618_end_0 = const()[name = tensor<string, []>("op_5618_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_5618_end_mask_0 = const()[name = tensor<string, []>("op_5618_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_5618_cast = slice_by_index(begin = var_5618_begin_0, end = var_5618_end_0, end_mask = var_5618_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5618_cast")];
            tensor<int32, [4]> var_5622_begin_0 = const()[name = tensor<string, []>("op_5622_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_5622_end_0 = const()[name = tensor<string, []>("op_5622_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_5622_end_mask_0 = const()[name = tensor<string, []>("op_5622_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_5622_cast = slice_by_index(begin = var_5622_begin_0, end = var_5622_end_0, end_mask = var_5622_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5622_cast")];
            tensor<int32, [4]> var_5626_begin_0 = const()[name = tensor<string, []>("op_5626_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_5626_end_0 = const()[name = tensor<string, []>("op_5626_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_5626_end_mask_0 = const()[name = tensor<string, []>("op_5626_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_5626_cast = slice_by_index(begin = var_5626_begin_0, end = var_5626_end_0, end_mask = var_5626_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5626_cast")];
            tensor<int32, [4]> var_5630_begin_0 = const()[name = tensor<string, []>("op_5630_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_5630_end_0 = const()[name = tensor<string, []>("op_5630_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_5630_end_mask_0 = const()[name = tensor<string, []>("op_5630_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_5630_cast = slice_by_index(begin = var_5630_begin_0, end = var_5630_end_0, end_mask = var_5630_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5630_cast")];
            tensor<int32, [4]> var_5634_begin_0 = const()[name = tensor<string, []>("op_5634_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_5634_end_0 = const()[name = tensor<string, []>("op_5634_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_5634_end_mask_0 = const()[name = tensor<string, []>("op_5634_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_5634_cast = slice_by_index(begin = var_5634_begin_0, end = var_5634_end_0, end_mask = var_5634_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5634_cast")];
            tensor<int32, [4]> var_5638_begin_0 = const()[name = tensor<string, []>("op_5638_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_5638_end_0 = const()[name = tensor<string, []>("op_5638_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_5638_end_mask_0 = const()[name = tensor<string, []>("op_5638_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_5638_cast = slice_by_index(begin = var_5638_begin_0, end = var_5638_end_0, end_mask = var_5638_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5638_cast")];
            tensor<int32, [4]> var_5642_begin_0 = const()[name = tensor<string, []>("op_5642_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_5642_end_0 = const()[name = tensor<string, []>("op_5642_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_5642_end_mask_0 = const()[name = tensor<string, []>("op_5642_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_5642_cast = slice_by_index(begin = var_5642_begin_0, end = var_5642_end_0, end_mask = var_5642_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5642_cast")];
            tensor<int32, [4]> var_5646_begin_0 = const()[name = tensor<string, []>("op_5646_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_5646_end_0 = const()[name = tensor<string, []>("op_5646_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_5646_end_mask_0 = const()[name = tensor<string, []>("op_5646_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_5646_cast = slice_by_index(begin = var_5646_begin_0, end = var_5646_end_0, end_mask = var_5646_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5646_cast")];
            tensor<int32, [4]> var_5650_begin_0 = const()[name = tensor<string, []>("op_5650_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_5650_end_0 = const()[name = tensor<string, []>("op_5650_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_5650_end_mask_0 = const()[name = tensor<string, []>("op_5650_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_5650_cast = slice_by_index(begin = var_5650_begin_0, end = var_5650_end_0, end_mask = var_5650_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5650_cast")];
            tensor<int32, [4]> var_5654_begin_0 = const()[name = tensor<string, []>("op_5654_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_5654_end_0 = const()[name = tensor<string, []>("op_5654_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_5654_end_mask_0 = const()[name = tensor<string, []>("op_5654_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_5654_cast = slice_by_index(begin = var_5654_begin_0, end = var_5654_end_0, end_mask = var_5654_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5654_cast")];
            tensor<int32, [4]> var_5658_begin_0 = const()[name = tensor<string, []>("op_5658_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_5658_end_0 = const()[name = tensor<string, []>("op_5658_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_5658_end_mask_0 = const()[name = tensor<string, []>("op_5658_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_5658_cast = slice_by_index(begin = var_5658_begin_0, end = var_5658_end_0, end_mask = var_5658_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5658_cast")];
            tensor<int32, [4]> var_5662_begin_0 = const()[name = tensor<string, []>("op_5662_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_5662_end_0 = const()[name = tensor<string, []>("op_5662_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_5662_end_mask_0 = const()[name = tensor<string, []>("op_5662_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_5662_cast = slice_by_index(begin = var_5662_begin_0, end = var_5662_end_0, end_mask = var_5662_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5662_cast")];
            tensor<int32, [4]> var_5666_begin_0 = const()[name = tensor<string, []>("op_5666_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_5666_end_0 = const()[name = tensor<string, []>("op_5666_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_5666_end_mask_0 = const()[name = tensor<string, []>("op_5666_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_5666_cast = slice_by_index(begin = var_5666_begin_0, end = var_5666_end_0, end_mask = var_5666_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5666_cast")];
            tensor<int32, [4]> var_5670_begin_0 = const()[name = tensor<string, []>("op_5670_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_5670_end_0 = const()[name = tensor<string, []>("op_5670_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_5670_end_mask_0 = const()[name = tensor<string, []>("op_5670_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_5670_cast = slice_by_index(begin = var_5670_begin_0, end = var_5670_end_0, end_mask = var_5670_end_mask_0, x = transpose_125)[name = tensor<string, []>("op_5670_cast")];
            tensor<int32, [4]> var_5672_begin_0 = const()[name = tensor<string, []>("op_5672_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5672_end_0 = const()[name = tensor<string, []>("op_5672_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_5672_end_mask_0 = const()[name = tensor<string, []>("op_5672_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5672_cast = slice_by_index(begin = var_5672_begin_0, end = var_5672_end_0, end_mask = var_5672_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5672_cast")];
            tensor<int32, [4]> var_5676_begin_0 = const()[name = tensor<string, []>("op_5676_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_5676_end_0 = const()[name = tensor<string, []>("op_5676_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_5676_end_mask_0 = const()[name = tensor<string, []>("op_5676_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5676_cast = slice_by_index(begin = var_5676_begin_0, end = var_5676_end_0, end_mask = var_5676_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5676_cast")];
            tensor<int32, [4]> var_5680_begin_0 = const()[name = tensor<string, []>("op_5680_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_5680_end_0 = const()[name = tensor<string, []>("op_5680_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_5680_end_mask_0 = const()[name = tensor<string, []>("op_5680_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5680_cast = slice_by_index(begin = var_5680_begin_0, end = var_5680_end_0, end_mask = var_5680_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5680_cast")];
            tensor<int32, [4]> var_5684_begin_0 = const()[name = tensor<string, []>("op_5684_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_5684_end_0 = const()[name = tensor<string, []>("op_5684_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_5684_end_mask_0 = const()[name = tensor<string, []>("op_5684_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5684_cast = slice_by_index(begin = var_5684_begin_0, end = var_5684_end_0, end_mask = var_5684_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5684_cast")];
            tensor<int32, [4]> var_5688_begin_0 = const()[name = tensor<string, []>("op_5688_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_5688_end_0 = const()[name = tensor<string, []>("op_5688_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_5688_end_mask_0 = const()[name = tensor<string, []>("op_5688_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5688_cast = slice_by_index(begin = var_5688_begin_0, end = var_5688_end_0, end_mask = var_5688_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5688_cast")];
            tensor<int32, [4]> var_5692_begin_0 = const()[name = tensor<string, []>("op_5692_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_5692_end_0 = const()[name = tensor<string, []>("op_5692_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_5692_end_mask_0 = const()[name = tensor<string, []>("op_5692_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5692_cast = slice_by_index(begin = var_5692_begin_0, end = var_5692_end_0, end_mask = var_5692_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5692_cast")];
            tensor<int32, [4]> var_5696_begin_0 = const()[name = tensor<string, []>("op_5696_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_5696_end_0 = const()[name = tensor<string, []>("op_5696_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_5696_end_mask_0 = const()[name = tensor<string, []>("op_5696_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5696_cast = slice_by_index(begin = var_5696_begin_0, end = var_5696_end_0, end_mask = var_5696_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5696_cast")];
            tensor<int32, [4]> var_5700_begin_0 = const()[name = tensor<string, []>("op_5700_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_5700_end_0 = const()[name = tensor<string, []>("op_5700_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_5700_end_mask_0 = const()[name = tensor<string, []>("op_5700_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5700_cast = slice_by_index(begin = var_5700_begin_0, end = var_5700_end_0, end_mask = var_5700_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5700_cast")];
            tensor<int32, [4]> var_5704_begin_0 = const()[name = tensor<string, []>("op_5704_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_5704_end_0 = const()[name = tensor<string, []>("op_5704_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_5704_end_mask_0 = const()[name = tensor<string, []>("op_5704_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5704_cast = slice_by_index(begin = var_5704_begin_0, end = var_5704_end_0, end_mask = var_5704_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5704_cast")];
            tensor<int32, [4]> var_5708_begin_0 = const()[name = tensor<string, []>("op_5708_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_5708_end_0 = const()[name = tensor<string, []>("op_5708_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_5708_end_mask_0 = const()[name = tensor<string, []>("op_5708_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5708_cast = slice_by_index(begin = var_5708_begin_0, end = var_5708_end_0, end_mask = var_5708_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5708_cast")];
            tensor<int32, [4]> var_5712_begin_0 = const()[name = tensor<string, []>("op_5712_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_5712_end_0 = const()[name = tensor<string, []>("op_5712_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_5712_end_mask_0 = const()[name = tensor<string, []>("op_5712_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5712_cast = slice_by_index(begin = var_5712_begin_0, end = var_5712_end_0, end_mask = var_5712_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5712_cast")];
            tensor<int32, [4]> var_5716_begin_0 = const()[name = tensor<string, []>("op_5716_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_5716_end_0 = const()[name = tensor<string, []>("op_5716_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_5716_end_mask_0 = const()[name = tensor<string, []>("op_5716_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5716_cast = slice_by_index(begin = var_5716_begin_0, end = var_5716_end_0, end_mask = var_5716_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5716_cast")];
            tensor<int32, [4]> var_5720_begin_0 = const()[name = tensor<string, []>("op_5720_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_5720_end_0 = const()[name = tensor<string, []>("op_5720_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_5720_end_mask_0 = const()[name = tensor<string, []>("op_5720_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5720_cast = slice_by_index(begin = var_5720_begin_0, end = var_5720_end_0, end_mask = var_5720_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5720_cast")];
            tensor<int32, [4]> var_5724_begin_0 = const()[name = tensor<string, []>("op_5724_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_5724_end_0 = const()[name = tensor<string, []>("op_5724_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_5724_end_mask_0 = const()[name = tensor<string, []>("op_5724_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5724_cast = slice_by_index(begin = var_5724_begin_0, end = var_5724_end_0, end_mask = var_5724_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5724_cast")];
            tensor<int32, [4]> var_5728_begin_0 = const()[name = tensor<string, []>("op_5728_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_5728_end_0 = const()[name = tensor<string, []>("op_5728_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_5728_end_mask_0 = const()[name = tensor<string, []>("op_5728_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5728_cast = slice_by_index(begin = var_5728_begin_0, end = var_5728_end_0, end_mask = var_5728_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5728_cast")];
            tensor<int32, [4]> var_5732_begin_0 = const()[name = tensor<string, []>("op_5732_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_5732_end_0 = const()[name = tensor<string, []>("op_5732_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_5732_end_mask_0 = const()[name = tensor<string, []>("op_5732_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5732_cast = slice_by_index(begin = var_5732_begin_0, end = var_5732_end_0, end_mask = var_5732_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5732_cast")];
            tensor<int32, [4]> var_5736_begin_0 = const()[name = tensor<string, []>("op_5736_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_5736_end_0 = const()[name = tensor<string, []>("op_5736_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_5736_end_mask_0 = const()[name = tensor<string, []>("op_5736_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5736_cast = slice_by_index(begin = var_5736_begin_0, end = var_5736_end_0, end_mask = var_5736_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5736_cast")];
            tensor<int32, [4]> var_5740_begin_0 = const()[name = tensor<string, []>("op_5740_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_5740_end_0 = const()[name = tensor<string, []>("op_5740_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_5740_end_mask_0 = const()[name = tensor<string, []>("op_5740_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5740_cast = slice_by_index(begin = var_5740_begin_0, end = var_5740_end_0, end_mask = var_5740_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5740_cast")];
            tensor<int32, [4]> var_5744_begin_0 = const()[name = tensor<string, []>("op_5744_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_5744_end_0 = const()[name = tensor<string, []>("op_5744_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_5744_end_mask_0 = const()[name = tensor<string, []>("op_5744_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5744_cast = slice_by_index(begin = var_5744_begin_0, end = var_5744_end_0, end_mask = var_5744_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5744_cast")];
            tensor<int32, [4]> var_5748_begin_0 = const()[name = tensor<string, []>("op_5748_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_5748_end_0 = const()[name = tensor<string, []>("op_5748_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_5748_end_mask_0 = const()[name = tensor<string, []>("op_5748_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5748_cast = slice_by_index(begin = var_5748_begin_0, end = var_5748_end_0, end_mask = var_5748_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_5748_cast")];
            tensor<string, []> var_5752_equation_0 = const()[name = tensor<string, []>("op_5752_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5752_cast = einsum(equation = var_5752_equation_0, values = (var_5594_cast, var_5511_cast))[name = tensor<string, []>("op_5752_cast")];
            tensor<fp16, []> var_5753_to_fp16 = const()[name = tensor<string, []>("op_5753_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_401_cast = mul(x = var_5752_cast, y = var_5753_to_fp16)[name = tensor<string, []>("aw_401_cast")];
            tensor<string, []> var_5756_equation_0 = const()[name = tensor<string, []>("op_5756_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5756_cast = einsum(equation = var_5756_equation_0, values = (var_5598_cast, var_5515_cast))[name = tensor<string, []>("op_5756_cast")];
            tensor<fp16, []> var_5757_to_fp16 = const()[name = tensor<string, []>("op_5757_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_403_cast = mul(x = var_5756_cast, y = var_5757_to_fp16)[name = tensor<string, []>("aw_403_cast")];
            tensor<string, []> var_5760_equation_0 = const()[name = tensor<string, []>("op_5760_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5760_cast = einsum(equation = var_5760_equation_0, values = (var_5602_cast, var_5519_cast))[name = tensor<string, []>("op_5760_cast")];
            tensor<fp16, []> var_5761_to_fp16 = const()[name = tensor<string, []>("op_5761_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_405_cast = mul(x = var_5760_cast, y = var_5761_to_fp16)[name = tensor<string, []>("aw_405_cast")];
            tensor<string, []> var_5764_equation_0 = const()[name = tensor<string, []>("op_5764_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5764_cast = einsum(equation = var_5764_equation_0, values = (var_5606_cast, var_5523_cast))[name = tensor<string, []>("op_5764_cast")];
            tensor<fp16, []> var_5765_to_fp16 = const()[name = tensor<string, []>("op_5765_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_407_cast = mul(x = var_5764_cast, y = var_5765_to_fp16)[name = tensor<string, []>("aw_407_cast")];
            tensor<string, []> var_5768_equation_0 = const()[name = tensor<string, []>("op_5768_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5768_cast = einsum(equation = var_5768_equation_0, values = (var_5610_cast, var_5527_cast))[name = tensor<string, []>("op_5768_cast")];
            tensor<fp16, []> var_5769_to_fp16 = const()[name = tensor<string, []>("op_5769_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_409_cast = mul(x = var_5768_cast, y = var_5769_to_fp16)[name = tensor<string, []>("aw_409_cast")];
            tensor<string, []> var_5772_equation_0 = const()[name = tensor<string, []>("op_5772_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5772_cast = einsum(equation = var_5772_equation_0, values = (var_5614_cast, var_5531_cast))[name = tensor<string, []>("op_5772_cast")];
            tensor<fp16, []> var_5773_to_fp16 = const()[name = tensor<string, []>("op_5773_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_411_cast = mul(x = var_5772_cast, y = var_5773_to_fp16)[name = tensor<string, []>("aw_411_cast")];
            tensor<string, []> var_5776_equation_0 = const()[name = tensor<string, []>("op_5776_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5776_cast = einsum(equation = var_5776_equation_0, values = (var_5618_cast, var_5535_cast))[name = tensor<string, []>("op_5776_cast")];
            tensor<fp16, []> var_5777_to_fp16 = const()[name = tensor<string, []>("op_5777_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_413_cast = mul(x = var_5776_cast, y = var_5777_to_fp16)[name = tensor<string, []>("aw_413_cast")];
            tensor<string, []> var_5780_equation_0 = const()[name = tensor<string, []>("op_5780_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5780_cast = einsum(equation = var_5780_equation_0, values = (var_5622_cast, var_5539_cast))[name = tensor<string, []>("op_5780_cast")];
            tensor<fp16, []> var_5781_to_fp16 = const()[name = tensor<string, []>("op_5781_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_415_cast = mul(x = var_5780_cast, y = var_5781_to_fp16)[name = tensor<string, []>("aw_415_cast")];
            tensor<string, []> var_5784_equation_0 = const()[name = tensor<string, []>("op_5784_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5784_cast = einsum(equation = var_5784_equation_0, values = (var_5626_cast, var_5543_cast))[name = tensor<string, []>("op_5784_cast")];
            tensor<fp16, []> var_5785_to_fp16 = const()[name = tensor<string, []>("op_5785_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_417_cast = mul(x = var_5784_cast, y = var_5785_to_fp16)[name = tensor<string, []>("aw_417_cast")];
            tensor<string, []> var_5788_equation_0 = const()[name = tensor<string, []>("op_5788_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5788_cast = einsum(equation = var_5788_equation_0, values = (var_5630_cast, var_5547_cast))[name = tensor<string, []>("op_5788_cast")];
            tensor<fp16, []> var_5789_to_fp16 = const()[name = tensor<string, []>("op_5789_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_419_cast = mul(x = var_5788_cast, y = var_5789_to_fp16)[name = tensor<string, []>("aw_419_cast")];
            tensor<string, []> var_5792_equation_0 = const()[name = tensor<string, []>("op_5792_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5792_cast = einsum(equation = var_5792_equation_0, values = (var_5634_cast, var_5551_cast))[name = tensor<string, []>("op_5792_cast")];
            tensor<fp16, []> var_5793_to_fp16 = const()[name = tensor<string, []>("op_5793_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_421_cast = mul(x = var_5792_cast, y = var_5793_to_fp16)[name = tensor<string, []>("aw_421_cast")];
            tensor<string, []> var_5796_equation_0 = const()[name = tensor<string, []>("op_5796_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5796_cast = einsum(equation = var_5796_equation_0, values = (var_5638_cast, var_5555_cast))[name = tensor<string, []>("op_5796_cast")];
            tensor<fp16, []> var_5797_to_fp16 = const()[name = tensor<string, []>("op_5797_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_423_cast = mul(x = var_5796_cast, y = var_5797_to_fp16)[name = tensor<string, []>("aw_423_cast")];
            tensor<string, []> var_5800_equation_0 = const()[name = tensor<string, []>("op_5800_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5800_cast = einsum(equation = var_5800_equation_0, values = (var_5642_cast, var_5559_cast))[name = tensor<string, []>("op_5800_cast")];
            tensor<fp16, []> var_5801_to_fp16 = const()[name = tensor<string, []>("op_5801_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_425_cast = mul(x = var_5800_cast, y = var_5801_to_fp16)[name = tensor<string, []>("aw_425_cast")];
            tensor<string, []> var_5804_equation_0 = const()[name = tensor<string, []>("op_5804_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5804_cast = einsum(equation = var_5804_equation_0, values = (var_5646_cast, var_5563_cast))[name = tensor<string, []>("op_5804_cast")];
            tensor<fp16, []> var_5805_to_fp16 = const()[name = tensor<string, []>("op_5805_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_427_cast = mul(x = var_5804_cast, y = var_5805_to_fp16)[name = tensor<string, []>("aw_427_cast")];
            tensor<string, []> var_5808_equation_0 = const()[name = tensor<string, []>("op_5808_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5808_cast = einsum(equation = var_5808_equation_0, values = (var_5650_cast, var_5567_cast))[name = tensor<string, []>("op_5808_cast")];
            tensor<fp16, []> var_5809_to_fp16 = const()[name = tensor<string, []>("op_5809_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_429_cast = mul(x = var_5808_cast, y = var_5809_to_fp16)[name = tensor<string, []>("aw_429_cast")];
            tensor<string, []> var_5812_equation_0 = const()[name = tensor<string, []>("op_5812_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5812_cast = einsum(equation = var_5812_equation_0, values = (var_5654_cast, var_5571_cast))[name = tensor<string, []>("op_5812_cast")];
            tensor<fp16, []> var_5813_to_fp16 = const()[name = tensor<string, []>("op_5813_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_431_cast = mul(x = var_5812_cast, y = var_5813_to_fp16)[name = tensor<string, []>("aw_431_cast")];
            tensor<string, []> var_5816_equation_0 = const()[name = tensor<string, []>("op_5816_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5816_cast = einsum(equation = var_5816_equation_0, values = (var_5658_cast, var_5575_cast))[name = tensor<string, []>("op_5816_cast")];
            tensor<fp16, []> var_5817_to_fp16 = const()[name = tensor<string, []>("op_5817_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_433_cast = mul(x = var_5816_cast, y = var_5817_to_fp16)[name = tensor<string, []>("aw_433_cast")];
            tensor<string, []> var_5820_equation_0 = const()[name = tensor<string, []>("op_5820_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5820_cast = einsum(equation = var_5820_equation_0, values = (var_5662_cast, var_5579_cast))[name = tensor<string, []>("op_5820_cast")];
            tensor<fp16, []> var_5821_to_fp16 = const()[name = tensor<string, []>("op_5821_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_435_cast = mul(x = var_5820_cast, y = var_5821_to_fp16)[name = tensor<string, []>("aw_435_cast")];
            tensor<string, []> var_5824_equation_0 = const()[name = tensor<string, []>("op_5824_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5824_cast = einsum(equation = var_5824_equation_0, values = (var_5666_cast, var_5583_cast))[name = tensor<string, []>("op_5824_cast")];
            tensor<fp16, []> var_5825_to_fp16 = const()[name = tensor<string, []>("op_5825_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_437_cast = mul(x = var_5824_cast, y = var_5825_to_fp16)[name = tensor<string, []>("aw_437_cast")];
            tensor<string, []> var_5828_equation_0 = const()[name = tensor<string, []>("op_5828_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5828_cast = einsum(equation = var_5828_equation_0, values = (var_5670_cast, var_5587_cast))[name = tensor<string, []>("op_5828_cast")];
            tensor<fp16, []> var_5829_to_fp16 = const()[name = tensor<string, []>("op_5829_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_439_cast = mul(x = var_5828_cast, y = var_5829_to_fp16)[name = tensor<string, []>("aw_439_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5831_cast = softmax(axis = var_2634, x = aw_401_cast)[name = tensor<string, []>("op_5831_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5832_cast = softmax(axis = var_2634, x = aw_403_cast)[name = tensor<string, []>("op_5832_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5833_cast = softmax(axis = var_2634, x = aw_405_cast)[name = tensor<string, []>("op_5833_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5834_cast = softmax(axis = var_2634, x = aw_407_cast)[name = tensor<string, []>("op_5834_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5835_cast = softmax(axis = var_2634, x = aw_409_cast)[name = tensor<string, []>("op_5835_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5836_cast = softmax(axis = var_2634, x = aw_411_cast)[name = tensor<string, []>("op_5836_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5837_cast = softmax(axis = var_2634, x = aw_413_cast)[name = tensor<string, []>("op_5837_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5838_cast = softmax(axis = var_2634, x = aw_415_cast)[name = tensor<string, []>("op_5838_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5839_cast = softmax(axis = var_2634, x = aw_417_cast)[name = tensor<string, []>("op_5839_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5840_cast = softmax(axis = var_2634, x = aw_419_cast)[name = tensor<string, []>("op_5840_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5841_cast = softmax(axis = var_2634, x = aw_421_cast)[name = tensor<string, []>("op_5841_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5842_cast = softmax(axis = var_2634, x = aw_423_cast)[name = tensor<string, []>("op_5842_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5843_cast = softmax(axis = var_2634, x = aw_425_cast)[name = tensor<string, []>("op_5843_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5844_cast = softmax(axis = var_2634, x = aw_427_cast)[name = tensor<string, []>("op_5844_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5845_cast = softmax(axis = var_2634, x = aw_429_cast)[name = tensor<string, []>("op_5845_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5846_cast = softmax(axis = var_2634, x = aw_431_cast)[name = tensor<string, []>("op_5846_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5847_cast = softmax(axis = var_2634, x = aw_433_cast)[name = tensor<string, []>("op_5847_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5848_cast = softmax(axis = var_2634, x = aw_435_cast)[name = tensor<string, []>("op_5848_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5849_cast = softmax(axis = var_2634, x = aw_437_cast)[name = tensor<string, []>("op_5849_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_5850_cast = softmax(axis = var_2634, x = aw_439_cast)[name = tensor<string, []>("op_5850_cast")];
            tensor<string, []> var_5852_equation_0 = const()[name = tensor<string, []>("op_5852_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5852_cast = einsum(equation = var_5852_equation_0, values = (var_5672_cast, var_5831_cast))[name = tensor<string, []>("op_5852_cast")];
            tensor<string, []> var_5854_equation_0 = const()[name = tensor<string, []>("op_5854_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5854_cast = einsum(equation = var_5854_equation_0, values = (var_5676_cast, var_5832_cast))[name = tensor<string, []>("op_5854_cast")];
            tensor<string, []> var_5856_equation_0 = const()[name = tensor<string, []>("op_5856_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5856_cast = einsum(equation = var_5856_equation_0, values = (var_5680_cast, var_5833_cast))[name = tensor<string, []>("op_5856_cast")];
            tensor<string, []> var_5858_equation_0 = const()[name = tensor<string, []>("op_5858_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5858_cast = einsum(equation = var_5858_equation_0, values = (var_5684_cast, var_5834_cast))[name = tensor<string, []>("op_5858_cast")];
            tensor<string, []> var_5860_equation_0 = const()[name = tensor<string, []>("op_5860_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5860_cast = einsum(equation = var_5860_equation_0, values = (var_5688_cast, var_5835_cast))[name = tensor<string, []>("op_5860_cast")];
            tensor<string, []> var_5862_equation_0 = const()[name = tensor<string, []>("op_5862_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5862_cast = einsum(equation = var_5862_equation_0, values = (var_5692_cast, var_5836_cast))[name = tensor<string, []>("op_5862_cast")];
            tensor<string, []> var_5864_equation_0 = const()[name = tensor<string, []>("op_5864_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5864_cast = einsum(equation = var_5864_equation_0, values = (var_5696_cast, var_5837_cast))[name = tensor<string, []>("op_5864_cast")];
            tensor<string, []> var_5866_equation_0 = const()[name = tensor<string, []>("op_5866_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5866_cast = einsum(equation = var_5866_equation_0, values = (var_5700_cast, var_5838_cast))[name = tensor<string, []>("op_5866_cast")];
            tensor<string, []> var_5868_equation_0 = const()[name = tensor<string, []>("op_5868_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5868_cast = einsum(equation = var_5868_equation_0, values = (var_5704_cast, var_5839_cast))[name = tensor<string, []>("op_5868_cast")];
            tensor<string, []> var_5870_equation_0 = const()[name = tensor<string, []>("op_5870_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5870_cast = einsum(equation = var_5870_equation_0, values = (var_5708_cast, var_5840_cast))[name = tensor<string, []>("op_5870_cast")];
            tensor<string, []> var_5872_equation_0 = const()[name = tensor<string, []>("op_5872_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5872_cast = einsum(equation = var_5872_equation_0, values = (var_5712_cast, var_5841_cast))[name = tensor<string, []>("op_5872_cast")];
            tensor<string, []> var_5874_equation_0 = const()[name = tensor<string, []>("op_5874_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5874_cast = einsum(equation = var_5874_equation_0, values = (var_5716_cast, var_5842_cast))[name = tensor<string, []>("op_5874_cast")];
            tensor<string, []> var_5876_equation_0 = const()[name = tensor<string, []>("op_5876_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5876_cast = einsum(equation = var_5876_equation_0, values = (var_5720_cast, var_5843_cast))[name = tensor<string, []>("op_5876_cast")];
            tensor<string, []> var_5878_equation_0 = const()[name = tensor<string, []>("op_5878_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5878_cast = einsum(equation = var_5878_equation_0, values = (var_5724_cast, var_5844_cast))[name = tensor<string, []>("op_5878_cast")];
            tensor<string, []> var_5880_equation_0 = const()[name = tensor<string, []>("op_5880_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5880_cast = einsum(equation = var_5880_equation_0, values = (var_5728_cast, var_5845_cast))[name = tensor<string, []>("op_5880_cast")];
            tensor<string, []> var_5882_equation_0 = const()[name = tensor<string, []>("op_5882_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5882_cast = einsum(equation = var_5882_equation_0, values = (var_5732_cast, var_5846_cast))[name = tensor<string, []>("op_5882_cast")];
            tensor<string, []> var_5884_equation_0 = const()[name = tensor<string, []>("op_5884_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5884_cast = einsum(equation = var_5884_equation_0, values = (var_5736_cast, var_5847_cast))[name = tensor<string, []>("op_5884_cast")];
            tensor<string, []> var_5886_equation_0 = const()[name = tensor<string, []>("op_5886_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5886_cast = einsum(equation = var_5886_equation_0, values = (var_5740_cast, var_5848_cast))[name = tensor<string, []>("op_5886_cast")];
            tensor<string, []> var_5888_equation_0 = const()[name = tensor<string, []>("op_5888_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5888_cast = einsum(equation = var_5888_equation_0, values = (var_5744_cast, var_5849_cast))[name = tensor<string, []>("op_5888_cast")];
            tensor<string, []> var_5890_equation_0 = const()[name = tensor<string, []>("op_5890_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_5890_cast = einsum(equation = var_5890_equation_0, values = (var_5748_cast, var_5850_cast))[name = tensor<string, []>("op_5890_cast")];
            tensor<bool, []> input_159_interleave_0 = const()[name = tensor<string, []>("input_159_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_159_cast = concat(axis = var_2634, interleave = input_159_interleave_0, values = (var_5852_cast, var_5854_cast, var_5856_cast, var_5858_cast, var_5860_cast, var_5862_cast, var_5864_cast, var_5866_cast, var_5868_cast, var_5870_cast, var_5872_cast, var_5874_cast, var_5876_cast, var_5878_cast, var_5880_cast, var_5882_cast, var_5884_cast, var_5886_cast, var_5888_cast, var_5890_cast))[name = tensor<string, []>("input_159_cast")];
            tensor<int32, [2]> var_5896 = const()[name = tensor<string, []>("op_5896"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5898 = const()[name = tensor<string, []>("op_5898"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_5900_pad_type_0 = const()[name = tensor<string, []>("op_5900_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_5900_pad_0 = const()[name = tensor<string, []>("op_5900_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(416540608)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(419817472)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_5900_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16, dilations = var_5898, groups = var_2634, pad = var_5900_pad_0, pad_type = var_5900_pad_type_0, strides = var_5896, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16, x = input_159_cast)[name = tensor<string, []>("op_5900_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_45_cast = add(x = var_5900_cast, y = inputs_43_cast)[name = tensor<string, []>("inputs_45_cast")];
            tensor<int32, [1]> var_5904 = const()[name = tensor<string, []>("op_5904"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_45_cast = reduce_mean(axes = var_5904, keep_dims = var_2629, x = inputs_45_cast)[name = tensor<string, []>("channels_mean_45_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_45_cast = sub(x = inputs_45_cast, y = channels_mean_45_cast)[name = tensor<string, []>("zero_mean_45_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_45_cast = mul(x = zero_mean_45_cast, y = zero_mean_45_cast)[name = tensor<string, []>("zero_mean_sq_45_cast")];
            tensor<int32, [1]> var_5908 = const()[name = tensor<string, []>("op_5908"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_5909_cast = reduce_mean(axes = var_5908, keep_dims = var_2629, x = zero_mean_sq_45_cast)[name = tensor<string, []>("op_5909_cast")];
            tensor<fp16, []> var_5910_to_fp16 = const()[name = tensor<string, []>("op_5910_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_5911_cast = add(x = var_5909_cast, y = var_5910_to_fp16)[name = tensor<string, []>("op_5911_cast")];
            tensor<fp16, []> denom_45_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_45_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_45_cast = rsqrt(epsilon = denom_45_epsilon_0_to_fp16, x = var_5911_cast)[name = tensor<string, []>("denom_45_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_45_cast = mul(x = zero_mean_45_cast, y = denom_45_cast)[name = tensor<string, []>("out_45_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5915_to_fp16 = const()[name = tensor<string, []>("op_5915_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(419820096)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_5916_cast = add(x = out_45_cast, y = var_5915_to_fp16)[name = tensor<string, []>("op_5916_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5918_to_fp16 = const()[name = tensor<string, []>("op_5918_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(419822720)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_85_cast = mul(x = var_5916_cast, y = var_5918_to_fp16)[name = tensor<string, []>("hidden_states_85_cast")];
            tensor<int32, [2]> var_5925 = const()[name = tensor<string, []>("op_5925"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5927 = const()[name = tensor<string, []>("op_5927"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_31_pad_type_0 = const()[name = tensor<string, []>("q_31_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_31_pad_0 = const()[name = tensor<string, []>("q_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(419825344)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_31_cast = conv(dilations = var_5927, groups = var_2634, pad = q_31_pad_0, pad_type = q_31_pad_type_0, strides = var_5925, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16, x = hidden_states_85_cast)[name = tensor<string, []>("q_31_cast")];
            tensor<int32, [2]> var_5931 = const()[name = tensor<string, []>("op_5931"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5933 = const()[name = tensor<string, []>("op_5933"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_61_pad_type_0 = const()[name = tensor<string, []>("k_61_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_61_pad_0 = const()[name = tensor<string, []>("k_61_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(423102208)))];
            tensor<fp16, [2, 1280, 1, 77]> k_61_cast = conv(dilations = var_5933, groups = var_2634, pad = k_61_pad_0, pad_type = k_61_pad_type_0, strides = var_5931, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_61_cast")];
            tensor<int32, [2]> var_5937 = const()[name = tensor<string, []>("op_5937"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5939 = const()[name = tensor<string, []>("op_5939"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_31_pad_type_0 = const()[name = tensor<string, []>("v_31_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_31_pad_0 = const()[name = tensor<string, []>("v_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(428345152)))];
            tensor<fp16, [2, 1280, 1, 77]> v_31_cast = conv(dilations = var_5939, groups = var_2634, pad = v_31_pad_0, pad_type = v_31_pad_type_0, strides = var_5937, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_31_cast")];
            tensor<int32, [4]> var_5943_begin_0 = const()[name = tensor<string, []>("op_5943_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5943_end_0 = const()[name = tensor<string, []>("op_5943_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_5943_end_mask_0 = const()[name = tensor<string, []>("op_5943_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5943_cast = slice_by_index(begin = var_5943_begin_0, end = var_5943_end_0, end_mask = var_5943_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_5943_cast")];
            tensor<int32, [4]> var_5947_begin_0 = const()[name = tensor<string, []>("op_5947_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_5947_end_0 = const()[name = tensor<string, []>("op_5947_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_5947_end_mask_0 = const()[name = tensor<string, []>("op_5947_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5947_cast = slice_by_index(begin = var_5947_begin_0, end = var_5947_end_0, end_mask = var_5947_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_5947_cast")];
            tensor<int32, [4]> var_5951_begin_0 = const()[name = tensor<string, []>("op_5951_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_5951_end_0 = const()[name = tensor<string, []>("op_5951_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_5951_end_mask_0 = const()[name = tensor<string, []>("op_5951_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5951_cast = slice_by_index(begin = var_5951_begin_0, end = var_5951_end_0, end_mask = var_5951_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_5951_cast")];
            tensor<int32, [4]> var_5955_begin_0 = const()[name = tensor<string, []>("op_5955_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_5955_end_0 = const()[name = tensor<string, []>("op_5955_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_5955_end_mask_0 = const()[name = tensor<string, []>("op_5955_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5955_cast = slice_by_index(begin = var_5955_begin_0, end = var_5955_end_0, end_mask = var_5955_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_5955_cast")];
            tensor<int32, [4]> var_5959_begin_0 = const()[name = tensor<string, []>("op_5959_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_5959_end_0 = const()[name = tensor<string, []>("op_5959_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_5959_end_mask_0 = const()[name = tensor<string, []>("op_5959_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5959_cast = slice_by_index(begin = var_5959_begin_0, end = var_5959_end_0, end_mask = var_5959_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_5959_cast")];
            tensor<int32, [4]> var_5963_begin_0 = const()[name = tensor<string, []>("op_5963_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_5963_end_0 = const()[name = tensor<string, []>("op_5963_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_5963_end_mask_0 = const()[name = tensor<string, []>("op_5963_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5963_cast = slice_by_index(begin = var_5963_begin_0, end = var_5963_end_0, end_mask = var_5963_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_5963_cast")];
            tensor<int32, [4]> var_5967_begin_0 = const()[name = tensor<string, []>("op_5967_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_5967_end_0 = const()[name = tensor<string, []>("op_5967_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_5967_end_mask_0 = const()[name = tensor<string, []>("op_5967_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5967_cast = slice_by_index(begin = var_5967_begin_0, end = var_5967_end_0, end_mask = var_5967_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_5967_cast")];
            tensor<int32, [4]> var_5971_begin_0 = const()[name = tensor<string, []>("op_5971_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_5971_end_0 = const()[name = tensor<string, []>("op_5971_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_5971_end_mask_0 = const()[name = tensor<string, []>("op_5971_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5971_cast = slice_by_index(begin = var_5971_begin_0, end = var_5971_end_0, end_mask = var_5971_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_5971_cast")];
            tensor<int32, [4]> var_5975_begin_0 = const()[name = tensor<string, []>("op_5975_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_5975_end_0 = const()[name = tensor<string, []>("op_5975_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_5975_end_mask_0 = const()[name = tensor<string, []>("op_5975_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5975_cast = slice_by_index(begin = var_5975_begin_0, end = var_5975_end_0, end_mask = var_5975_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_5975_cast")];
            tensor<int32, [4]> var_5979_begin_0 = const()[name = tensor<string, []>("op_5979_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_5979_end_0 = const()[name = tensor<string, []>("op_5979_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_5979_end_mask_0 = const()[name = tensor<string, []>("op_5979_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5979_cast = slice_by_index(begin = var_5979_begin_0, end = var_5979_end_0, end_mask = var_5979_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_5979_cast")];
            tensor<int32, [4]> var_5983_begin_0 = const()[name = tensor<string, []>("op_5983_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_5983_end_0 = const()[name = tensor<string, []>("op_5983_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_5983_end_mask_0 = const()[name = tensor<string, []>("op_5983_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5983_cast = slice_by_index(begin = var_5983_begin_0, end = var_5983_end_0, end_mask = var_5983_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_5983_cast")];
            tensor<int32, [4]> var_5987_begin_0 = const()[name = tensor<string, []>("op_5987_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_5987_end_0 = const()[name = tensor<string, []>("op_5987_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_5987_end_mask_0 = const()[name = tensor<string, []>("op_5987_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5987_cast = slice_by_index(begin = var_5987_begin_0, end = var_5987_end_0, end_mask = var_5987_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_5987_cast")];
            tensor<int32, [4]> var_5991_begin_0 = const()[name = tensor<string, []>("op_5991_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_5991_end_0 = const()[name = tensor<string, []>("op_5991_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_5991_end_mask_0 = const()[name = tensor<string, []>("op_5991_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5991_cast = slice_by_index(begin = var_5991_begin_0, end = var_5991_end_0, end_mask = var_5991_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_5991_cast")];
            tensor<int32, [4]> var_5995_begin_0 = const()[name = tensor<string, []>("op_5995_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_5995_end_0 = const()[name = tensor<string, []>("op_5995_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_5995_end_mask_0 = const()[name = tensor<string, []>("op_5995_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5995_cast = slice_by_index(begin = var_5995_begin_0, end = var_5995_end_0, end_mask = var_5995_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_5995_cast")];
            tensor<int32, [4]> var_5999_begin_0 = const()[name = tensor<string, []>("op_5999_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_5999_end_0 = const()[name = tensor<string, []>("op_5999_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_5999_end_mask_0 = const()[name = tensor<string, []>("op_5999_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_5999_cast = slice_by_index(begin = var_5999_begin_0, end = var_5999_end_0, end_mask = var_5999_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_5999_cast")];
            tensor<int32, [4]> var_6003_begin_0 = const()[name = tensor<string, []>("op_6003_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_6003_end_0 = const()[name = tensor<string, []>("op_6003_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_6003_end_mask_0 = const()[name = tensor<string, []>("op_6003_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6003_cast = slice_by_index(begin = var_6003_begin_0, end = var_6003_end_0, end_mask = var_6003_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_6003_cast")];
            tensor<int32, [4]> var_6007_begin_0 = const()[name = tensor<string, []>("op_6007_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_6007_end_0 = const()[name = tensor<string, []>("op_6007_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_6007_end_mask_0 = const()[name = tensor<string, []>("op_6007_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6007_cast = slice_by_index(begin = var_6007_begin_0, end = var_6007_end_0, end_mask = var_6007_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_6007_cast")];
            tensor<int32, [4]> var_6011_begin_0 = const()[name = tensor<string, []>("op_6011_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_6011_end_0 = const()[name = tensor<string, []>("op_6011_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_6011_end_mask_0 = const()[name = tensor<string, []>("op_6011_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6011_cast = slice_by_index(begin = var_6011_begin_0, end = var_6011_end_0, end_mask = var_6011_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_6011_cast")];
            tensor<int32, [4]> var_6015_begin_0 = const()[name = tensor<string, []>("op_6015_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_6015_end_0 = const()[name = tensor<string, []>("op_6015_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_6015_end_mask_0 = const()[name = tensor<string, []>("op_6015_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6015_cast = slice_by_index(begin = var_6015_begin_0, end = var_6015_end_0, end_mask = var_6015_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_6015_cast")];
            tensor<int32, [4]> var_6019_begin_0 = const()[name = tensor<string, []>("op_6019_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_6019_end_0 = const()[name = tensor<string, []>("op_6019_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_6019_end_mask_0 = const()[name = tensor<string, []>("op_6019_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6019_cast = slice_by_index(begin = var_6019_begin_0, end = var_6019_end_0, end_mask = var_6019_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_6019_cast")];
            tensor<int32, [4]> k_63_perm_0 = const()[name = tensor<string, []>("k_63_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_6026_begin_0 = const()[name = tensor<string, []>("op_6026_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6026_end_0 = const()[name = tensor<string, []>("op_6026_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_6026_end_mask_0 = const()[name = tensor<string, []>("op_6026_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_124 = transpose(perm = k_63_perm_0, x = k_61_cast)[name = tensor<string, []>("transpose_124")];
            tensor<fp16, [2, 77, 1, 64]> var_6026_cast = slice_by_index(begin = var_6026_begin_0, end = var_6026_end_0, end_mask = var_6026_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6026_cast")];
            tensor<int32, [4]> var_6030_begin_0 = const()[name = tensor<string, []>("op_6030_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_6030_end_0 = const()[name = tensor<string, []>("op_6030_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_6030_end_mask_0 = const()[name = tensor<string, []>("op_6030_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6030_cast = slice_by_index(begin = var_6030_begin_0, end = var_6030_end_0, end_mask = var_6030_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6030_cast")];
            tensor<int32, [4]> var_6034_begin_0 = const()[name = tensor<string, []>("op_6034_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_6034_end_0 = const()[name = tensor<string, []>("op_6034_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_6034_end_mask_0 = const()[name = tensor<string, []>("op_6034_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6034_cast = slice_by_index(begin = var_6034_begin_0, end = var_6034_end_0, end_mask = var_6034_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6034_cast")];
            tensor<int32, [4]> var_6038_begin_0 = const()[name = tensor<string, []>("op_6038_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_6038_end_0 = const()[name = tensor<string, []>("op_6038_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_6038_end_mask_0 = const()[name = tensor<string, []>("op_6038_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6038_cast = slice_by_index(begin = var_6038_begin_0, end = var_6038_end_0, end_mask = var_6038_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6038_cast")];
            tensor<int32, [4]> var_6042_begin_0 = const()[name = tensor<string, []>("op_6042_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_6042_end_0 = const()[name = tensor<string, []>("op_6042_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_6042_end_mask_0 = const()[name = tensor<string, []>("op_6042_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6042_cast = slice_by_index(begin = var_6042_begin_0, end = var_6042_end_0, end_mask = var_6042_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6042_cast")];
            tensor<int32, [4]> var_6046_begin_0 = const()[name = tensor<string, []>("op_6046_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_6046_end_0 = const()[name = tensor<string, []>("op_6046_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_6046_end_mask_0 = const()[name = tensor<string, []>("op_6046_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6046_cast = slice_by_index(begin = var_6046_begin_0, end = var_6046_end_0, end_mask = var_6046_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6046_cast")];
            tensor<int32, [4]> var_6050_begin_0 = const()[name = tensor<string, []>("op_6050_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_6050_end_0 = const()[name = tensor<string, []>("op_6050_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_6050_end_mask_0 = const()[name = tensor<string, []>("op_6050_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6050_cast = slice_by_index(begin = var_6050_begin_0, end = var_6050_end_0, end_mask = var_6050_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6050_cast")];
            tensor<int32, [4]> var_6054_begin_0 = const()[name = tensor<string, []>("op_6054_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_6054_end_0 = const()[name = tensor<string, []>("op_6054_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_6054_end_mask_0 = const()[name = tensor<string, []>("op_6054_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6054_cast = slice_by_index(begin = var_6054_begin_0, end = var_6054_end_0, end_mask = var_6054_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6054_cast")];
            tensor<int32, [4]> var_6058_begin_0 = const()[name = tensor<string, []>("op_6058_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_6058_end_0 = const()[name = tensor<string, []>("op_6058_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_6058_end_mask_0 = const()[name = tensor<string, []>("op_6058_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6058_cast = slice_by_index(begin = var_6058_begin_0, end = var_6058_end_0, end_mask = var_6058_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6058_cast")];
            tensor<int32, [4]> var_6062_begin_0 = const()[name = tensor<string, []>("op_6062_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_6062_end_0 = const()[name = tensor<string, []>("op_6062_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_6062_end_mask_0 = const()[name = tensor<string, []>("op_6062_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6062_cast = slice_by_index(begin = var_6062_begin_0, end = var_6062_end_0, end_mask = var_6062_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6062_cast")];
            tensor<int32, [4]> var_6066_begin_0 = const()[name = tensor<string, []>("op_6066_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_6066_end_0 = const()[name = tensor<string, []>("op_6066_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_6066_end_mask_0 = const()[name = tensor<string, []>("op_6066_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6066_cast = slice_by_index(begin = var_6066_begin_0, end = var_6066_end_0, end_mask = var_6066_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6066_cast")];
            tensor<int32, [4]> var_6070_begin_0 = const()[name = tensor<string, []>("op_6070_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_6070_end_0 = const()[name = tensor<string, []>("op_6070_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_6070_end_mask_0 = const()[name = tensor<string, []>("op_6070_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6070_cast = slice_by_index(begin = var_6070_begin_0, end = var_6070_end_0, end_mask = var_6070_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6070_cast")];
            tensor<int32, [4]> var_6074_begin_0 = const()[name = tensor<string, []>("op_6074_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_6074_end_0 = const()[name = tensor<string, []>("op_6074_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_6074_end_mask_0 = const()[name = tensor<string, []>("op_6074_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6074_cast = slice_by_index(begin = var_6074_begin_0, end = var_6074_end_0, end_mask = var_6074_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6074_cast")];
            tensor<int32, [4]> var_6078_begin_0 = const()[name = tensor<string, []>("op_6078_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_6078_end_0 = const()[name = tensor<string, []>("op_6078_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_6078_end_mask_0 = const()[name = tensor<string, []>("op_6078_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6078_cast = slice_by_index(begin = var_6078_begin_0, end = var_6078_end_0, end_mask = var_6078_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6078_cast")];
            tensor<int32, [4]> var_6082_begin_0 = const()[name = tensor<string, []>("op_6082_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_6082_end_0 = const()[name = tensor<string, []>("op_6082_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_6082_end_mask_0 = const()[name = tensor<string, []>("op_6082_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6082_cast = slice_by_index(begin = var_6082_begin_0, end = var_6082_end_0, end_mask = var_6082_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6082_cast")];
            tensor<int32, [4]> var_6086_begin_0 = const()[name = tensor<string, []>("op_6086_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_6086_end_0 = const()[name = tensor<string, []>("op_6086_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_6086_end_mask_0 = const()[name = tensor<string, []>("op_6086_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6086_cast = slice_by_index(begin = var_6086_begin_0, end = var_6086_end_0, end_mask = var_6086_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6086_cast")];
            tensor<int32, [4]> var_6090_begin_0 = const()[name = tensor<string, []>("op_6090_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_6090_end_0 = const()[name = tensor<string, []>("op_6090_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_6090_end_mask_0 = const()[name = tensor<string, []>("op_6090_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6090_cast = slice_by_index(begin = var_6090_begin_0, end = var_6090_end_0, end_mask = var_6090_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6090_cast")];
            tensor<int32, [4]> var_6094_begin_0 = const()[name = tensor<string, []>("op_6094_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_6094_end_0 = const()[name = tensor<string, []>("op_6094_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_6094_end_mask_0 = const()[name = tensor<string, []>("op_6094_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6094_cast = slice_by_index(begin = var_6094_begin_0, end = var_6094_end_0, end_mask = var_6094_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6094_cast")];
            tensor<int32, [4]> var_6098_begin_0 = const()[name = tensor<string, []>("op_6098_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_6098_end_0 = const()[name = tensor<string, []>("op_6098_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_6098_end_mask_0 = const()[name = tensor<string, []>("op_6098_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6098_cast = slice_by_index(begin = var_6098_begin_0, end = var_6098_end_0, end_mask = var_6098_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6098_cast")];
            tensor<int32, [4]> var_6102_begin_0 = const()[name = tensor<string, []>("op_6102_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_6102_end_0 = const()[name = tensor<string, []>("op_6102_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_6102_end_mask_0 = const()[name = tensor<string, []>("op_6102_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6102_cast = slice_by_index(begin = var_6102_begin_0, end = var_6102_end_0, end_mask = var_6102_end_mask_0, x = transpose_124)[name = tensor<string, []>("op_6102_cast")];
            tensor<int32, [4]> var_6104_begin_0 = const()[name = tensor<string, []>("op_6104_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6104_end_0 = const()[name = tensor<string, []>("op_6104_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_6104_end_mask_0 = const()[name = tensor<string, []>("op_6104_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6104_cast = slice_by_index(begin = var_6104_begin_0, end = var_6104_end_0, end_mask = var_6104_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6104_cast")];
            tensor<int32, [4]> var_6108_begin_0 = const()[name = tensor<string, []>("op_6108_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_6108_end_0 = const()[name = tensor<string, []>("op_6108_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_6108_end_mask_0 = const()[name = tensor<string, []>("op_6108_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6108_cast = slice_by_index(begin = var_6108_begin_0, end = var_6108_end_0, end_mask = var_6108_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6108_cast")];
            tensor<int32, [4]> var_6112_begin_0 = const()[name = tensor<string, []>("op_6112_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_6112_end_0 = const()[name = tensor<string, []>("op_6112_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_6112_end_mask_0 = const()[name = tensor<string, []>("op_6112_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6112_cast = slice_by_index(begin = var_6112_begin_0, end = var_6112_end_0, end_mask = var_6112_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6112_cast")];
            tensor<int32, [4]> var_6116_begin_0 = const()[name = tensor<string, []>("op_6116_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_6116_end_0 = const()[name = tensor<string, []>("op_6116_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_6116_end_mask_0 = const()[name = tensor<string, []>("op_6116_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6116_cast = slice_by_index(begin = var_6116_begin_0, end = var_6116_end_0, end_mask = var_6116_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6116_cast")];
            tensor<int32, [4]> var_6120_begin_0 = const()[name = tensor<string, []>("op_6120_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_6120_end_0 = const()[name = tensor<string, []>("op_6120_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_6120_end_mask_0 = const()[name = tensor<string, []>("op_6120_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6120_cast = slice_by_index(begin = var_6120_begin_0, end = var_6120_end_0, end_mask = var_6120_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6120_cast")];
            tensor<int32, [4]> var_6124_begin_0 = const()[name = tensor<string, []>("op_6124_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_6124_end_0 = const()[name = tensor<string, []>("op_6124_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_6124_end_mask_0 = const()[name = tensor<string, []>("op_6124_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6124_cast = slice_by_index(begin = var_6124_begin_0, end = var_6124_end_0, end_mask = var_6124_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6124_cast")];
            tensor<int32, [4]> var_6128_begin_0 = const()[name = tensor<string, []>("op_6128_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_6128_end_0 = const()[name = tensor<string, []>("op_6128_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_6128_end_mask_0 = const()[name = tensor<string, []>("op_6128_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6128_cast = slice_by_index(begin = var_6128_begin_0, end = var_6128_end_0, end_mask = var_6128_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6128_cast")];
            tensor<int32, [4]> var_6132_begin_0 = const()[name = tensor<string, []>("op_6132_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_6132_end_0 = const()[name = tensor<string, []>("op_6132_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_6132_end_mask_0 = const()[name = tensor<string, []>("op_6132_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6132_cast = slice_by_index(begin = var_6132_begin_0, end = var_6132_end_0, end_mask = var_6132_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6132_cast")];
            tensor<int32, [4]> var_6136_begin_0 = const()[name = tensor<string, []>("op_6136_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_6136_end_0 = const()[name = tensor<string, []>("op_6136_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_6136_end_mask_0 = const()[name = tensor<string, []>("op_6136_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6136_cast = slice_by_index(begin = var_6136_begin_0, end = var_6136_end_0, end_mask = var_6136_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6136_cast")];
            tensor<int32, [4]> var_6140_begin_0 = const()[name = tensor<string, []>("op_6140_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_6140_end_0 = const()[name = tensor<string, []>("op_6140_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_6140_end_mask_0 = const()[name = tensor<string, []>("op_6140_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6140_cast = slice_by_index(begin = var_6140_begin_0, end = var_6140_end_0, end_mask = var_6140_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6140_cast")];
            tensor<int32, [4]> var_6144_begin_0 = const()[name = tensor<string, []>("op_6144_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_6144_end_0 = const()[name = tensor<string, []>("op_6144_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_6144_end_mask_0 = const()[name = tensor<string, []>("op_6144_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6144_cast = slice_by_index(begin = var_6144_begin_0, end = var_6144_end_0, end_mask = var_6144_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6144_cast")];
            tensor<int32, [4]> var_6148_begin_0 = const()[name = tensor<string, []>("op_6148_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_6148_end_0 = const()[name = tensor<string, []>("op_6148_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_6148_end_mask_0 = const()[name = tensor<string, []>("op_6148_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6148_cast = slice_by_index(begin = var_6148_begin_0, end = var_6148_end_0, end_mask = var_6148_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6148_cast")];
            tensor<int32, [4]> var_6152_begin_0 = const()[name = tensor<string, []>("op_6152_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_6152_end_0 = const()[name = tensor<string, []>("op_6152_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_6152_end_mask_0 = const()[name = tensor<string, []>("op_6152_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6152_cast = slice_by_index(begin = var_6152_begin_0, end = var_6152_end_0, end_mask = var_6152_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6152_cast")];
            tensor<int32, [4]> var_6156_begin_0 = const()[name = tensor<string, []>("op_6156_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_6156_end_0 = const()[name = tensor<string, []>("op_6156_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_6156_end_mask_0 = const()[name = tensor<string, []>("op_6156_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6156_cast = slice_by_index(begin = var_6156_begin_0, end = var_6156_end_0, end_mask = var_6156_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6156_cast")];
            tensor<int32, [4]> var_6160_begin_0 = const()[name = tensor<string, []>("op_6160_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_6160_end_0 = const()[name = tensor<string, []>("op_6160_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_6160_end_mask_0 = const()[name = tensor<string, []>("op_6160_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6160_cast = slice_by_index(begin = var_6160_begin_0, end = var_6160_end_0, end_mask = var_6160_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6160_cast")];
            tensor<int32, [4]> var_6164_begin_0 = const()[name = tensor<string, []>("op_6164_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_6164_end_0 = const()[name = tensor<string, []>("op_6164_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_6164_end_mask_0 = const()[name = tensor<string, []>("op_6164_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6164_cast = slice_by_index(begin = var_6164_begin_0, end = var_6164_end_0, end_mask = var_6164_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6164_cast")];
            tensor<int32, [4]> var_6168_begin_0 = const()[name = tensor<string, []>("op_6168_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_6168_end_0 = const()[name = tensor<string, []>("op_6168_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_6168_end_mask_0 = const()[name = tensor<string, []>("op_6168_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6168_cast = slice_by_index(begin = var_6168_begin_0, end = var_6168_end_0, end_mask = var_6168_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6168_cast")];
            tensor<int32, [4]> var_6172_begin_0 = const()[name = tensor<string, []>("op_6172_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_6172_end_0 = const()[name = tensor<string, []>("op_6172_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_6172_end_mask_0 = const()[name = tensor<string, []>("op_6172_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6172_cast = slice_by_index(begin = var_6172_begin_0, end = var_6172_end_0, end_mask = var_6172_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6172_cast")];
            tensor<int32, [4]> var_6176_begin_0 = const()[name = tensor<string, []>("op_6176_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_6176_end_0 = const()[name = tensor<string, []>("op_6176_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_6176_end_mask_0 = const()[name = tensor<string, []>("op_6176_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6176_cast = slice_by_index(begin = var_6176_begin_0, end = var_6176_end_0, end_mask = var_6176_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6176_cast")];
            tensor<int32, [4]> var_6180_begin_0 = const()[name = tensor<string, []>("op_6180_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_6180_end_0 = const()[name = tensor<string, []>("op_6180_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_6180_end_mask_0 = const()[name = tensor<string, []>("op_6180_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6180_cast = slice_by_index(begin = var_6180_begin_0, end = var_6180_end_0, end_mask = var_6180_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6180_cast")];
            tensor<string, []> var_6184_equation_0 = const()[name = tensor<string, []>("op_6184_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6184_cast = einsum(equation = var_6184_equation_0, values = (var_6026_cast, var_5943_cast))[name = tensor<string, []>("op_6184_cast")];
            tensor<fp16, []> var_6185_to_fp16 = const()[name = tensor<string, []>("op_6185_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_441_cast = mul(x = var_6184_cast, y = var_6185_to_fp16)[name = tensor<string, []>("aw_441_cast")];
            tensor<string, []> var_6188_equation_0 = const()[name = tensor<string, []>("op_6188_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6188_cast = einsum(equation = var_6188_equation_0, values = (var_6030_cast, var_5947_cast))[name = tensor<string, []>("op_6188_cast")];
            tensor<fp16, []> var_6189_to_fp16 = const()[name = tensor<string, []>("op_6189_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_443_cast = mul(x = var_6188_cast, y = var_6189_to_fp16)[name = tensor<string, []>("aw_443_cast")];
            tensor<string, []> var_6192_equation_0 = const()[name = tensor<string, []>("op_6192_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6192_cast = einsum(equation = var_6192_equation_0, values = (var_6034_cast, var_5951_cast))[name = tensor<string, []>("op_6192_cast")];
            tensor<fp16, []> var_6193_to_fp16 = const()[name = tensor<string, []>("op_6193_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_445_cast = mul(x = var_6192_cast, y = var_6193_to_fp16)[name = tensor<string, []>("aw_445_cast")];
            tensor<string, []> var_6196_equation_0 = const()[name = tensor<string, []>("op_6196_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6196_cast = einsum(equation = var_6196_equation_0, values = (var_6038_cast, var_5955_cast))[name = tensor<string, []>("op_6196_cast")];
            tensor<fp16, []> var_6197_to_fp16 = const()[name = tensor<string, []>("op_6197_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_447_cast = mul(x = var_6196_cast, y = var_6197_to_fp16)[name = tensor<string, []>("aw_447_cast")];
            tensor<string, []> var_6200_equation_0 = const()[name = tensor<string, []>("op_6200_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6200_cast = einsum(equation = var_6200_equation_0, values = (var_6042_cast, var_5959_cast))[name = tensor<string, []>("op_6200_cast")];
            tensor<fp16, []> var_6201_to_fp16 = const()[name = tensor<string, []>("op_6201_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_449_cast = mul(x = var_6200_cast, y = var_6201_to_fp16)[name = tensor<string, []>("aw_449_cast")];
            tensor<string, []> var_6204_equation_0 = const()[name = tensor<string, []>("op_6204_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6204_cast = einsum(equation = var_6204_equation_0, values = (var_6046_cast, var_5963_cast))[name = tensor<string, []>("op_6204_cast")];
            tensor<fp16, []> var_6205_to_fp16 = const()[name = tensor<string, []>("op_6205_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_451_cast = mul(x = var_6204_cast, y = var_6205_to_fp16)[name = tensor<string, []>("aw_451_cast")];
            tensor<string, []> var_6208_equation_0 = const()[name = tensor<string, []>("op_6208_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6208_cast = einsum(equation = var_6208_equation_0, values = (var_6050_cast, var_5967_cast))[name = tensor<string, []>("op_6208_cast")];
            tensor<fp16, []> var_6209_to_fp16 = const()[name = tensor<string, []>("op_6209_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_453_cast = mul(x = var_6208_cast, y = var_6209_to_fp16)[name = tensor<string, []>("aw_453_cast")];
            tensor<string, []> var_6212_equation_0 = const()[name = tensor<string, []>("op_6212_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6212_cast = einsum(equation = var_6212_equation_0, values = (var_6054_cast, var_5971_cast))[name = tensor<string, []>("op_6212_cast")];
            tensor<fp16, []> var_6213_to_fp16 = const()[name = tensor<string, []>("op_6213_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_455_cast = mul(x = var_6212_cast, y = var_6213_to_fp16)[name = tensor<string, []>("aw_455_cast")];
            tensor<string, []> var_6216_equation_0 = const()[name = tensor<string, []>("op_6216_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6216_cast = einsum(equation = var_6216_equation_0, values = (var_6058_cast, var_5975_cast))[name = tensor<string, []>("op_6216_cast")];
            tensor<fp16, []> var_6217_to_fp16 = const()[name = tensor<string, []>("op_6217_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_457_cast = mul(x = var_6216_cast, y = var_6217_to_fp16)[name = tensor<string, []>("aw_457_cast")];
            tensor<string, []> var_6220_equation_0 = const()[name = tensor<string, []>("op_6220_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6220_cast = einsum(equation = var_6220_equation_0, values = (var_6062_cast, var_5979_cast))[name = tensor<string, []>("op_6220_cast")];
            tensor<fp16, []> var_6221_to_fp16 = const()[name = tensor<string, []>("op_6221_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_459_cast = mul(x = var_6220_cast, y = var_6221_to_fp16)[name = tensor<string, []>("aw_459_cast")];
            tensor<string, []> var_6224_equation_0 = const()[name = tensor<string, []>("op_6224_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6224_cast = einsum(equation = var_6224_equation_0, values = (var_6066_cast, var_5983_cast))[name = tensor<string, []>("op_6224_cast")];
            tensor<fp16, []> var_6225_to_fp16 = const()[name = tensor<string, []>("op_6225_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_461_cast = mul(x = var_6224_cast, y = var_6225_to_fp16)[name = tensor<string, []>("aw_461_cast")];
            tensor<string, []> var_6228_equation_0 = const()[name = tensor<string, []>("op_6228_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6228_cast = einsum(equation = var_6228_equation_0, values = (var_6070_cast, var_5987_cast))[name = tensor<string, []>("op_6228_cast")];
            tensor<fp16, []> var_6229_to_fp16 = const()[name = tensor<string, []>("op_6229_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_463_cast = mul(x = var_6228_cast, y = var_6229_to_fp16)[name = tensor<string, []>("aw_463_cast")];
            tensor<string, []> var_6232_equation_0 = const()[name = tensor<string, []>("op_6232_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6232_cast = einsum(equation = var_6232_equation_0, values = (var_6074_cast, var_5991_cast))[name = tensor<string, []>("op_6232_cast")];
            tensor<fp16, []> var_6233_to_fp16 = const()[name = tensor<string, []>("op_6233_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_465_cast = mul(x = var_6232_cast, y = var_6233_to_fp16)[name = tensor<string, []>("aw_465_cast")];
            tensor<string, []> var_6236_equation_0 = const()[name = tensor<string, []>("op_6236_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6236_cast = einsum(equation = var_6236_equation_0, values = (var_6078_cast, var_5995_cast))[name = tensor<string, []>("op_6236_cast")];
            tensor<fp16, []> var_6237_to_fp16 = const()[name = tensor<string, []>("op_6237_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_467_cast = mul(x = var_6236_cast, y = var_6237_to_fp16)[name = tensor<string, []>("aw_467_cast")];
            tensor<string, []> var_6240_equation_0 = const()[name = tensor<string, []>("op_6240_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6240_cast = einsum(equation = var_6240_equation_0, values = (var_6082_cast, var_5999_cast))[name = tensor<string, []>("op_6240_cast")];
            tensor<fp16, []> var_6241_to_fp16 = const()[name = tensor<string, []>("op_6241_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_469_cast = mul(x = var_6240_cast, y = var_6241_to_fp16)[name = tensor<string, []>("aw_469_cast")];
            tensor<string, []> var_6244_equation_0 = const()[name = tensor<string, []>("op_6244_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6244_cast = einsum(equation = var_6244_equation_0, values = (var_6086_cast, var_6003_cast))[name = tensor<string, []>("op_6244_cast")];
            tensor<fp16, []> var_6245_to_fp16 = const()[name = tensor<string, []>("op_6245_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_471_cast = mul(x = var_6244_cast, y = var_6245_to_fp16)[name = tensor<string, []>("aw_471_cast")];
            tensor<string, []> var_6248_equation_0 = const()[name = tensor<string, []>("op_6248_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6248_cast = einsum(equation = var_6248_equation_0, values = (var_6090_cast, var_6007_cast))[name = tensor<string, []>("op_6248_cast")];
            tensor<fp16, []> var_6249_to_fp16 = const()[name = tensor<string, []>("op_6249_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_473_cast = mul(x = var_6248_cast, y = var_6249_to_fp16)[name = tensor<string, []>("aw_473_cast")];
            tensor<string, []> var_6252_equation_0 = const()[name = tensor<string, []>("op_6252_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6252_cast = einsum(equation = var_6252_equation_0, values = (var_6094_cast, var_6011_cast))[name = tensor<string, []>("op_6252_cast")];
            tensor<fp16, []> var_6253_to_fp16 = const()[name = tensor<string, []>("op_6253_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_475_cast = mul(x = var_6252_cast, y = var_6253_to_fp16)[name = tensor<string, []>("aw_475_cast")];
            tensor<string, []> var_6256_equation_0 = const()[name = tensor<string, []>("op_6256_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6256_cast = einsum(equation = var_6256_equation_0, values = (var_6098_cast, var_6015_cast))[name = tensor<string, []>("op_6256_cast")];
            tensor<fp16, []> var_6257_to_fp16 = const()[name = tensor<string, []>("op_6257_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_477_cast = mul(x = var_6256_cast, y = var_6257_to_fp16)[name = tensor<string, []>("aw_477_cast")];
            tensor<string, []> var_6260_equation_0 = const()[name = tensor<string, []>("op_6260_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_6260_cast = einsum(equation = var_6260_equation_0, values = (var_6102_cast, var_6019_cast))[name = tensor<string, []>("op_6260_cast")];
            tensor<fp16, []> var_6261_to_fp16 = const()[name = tensor<string, []>("op_6261_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_479_cast = mul(x = var_6260_cast, y = var_6261_to_fp16)[name = tensor<string, []>("aw_479_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6263_cast = softmax(axis = var_2634, x = aw_441_cast)[name = tensor<string, []>("op_6263_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6264_cast = softmax(axis = var_2634, x = aw_443_cast)[name = tensor<string, []>("op_6264_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6265_cast = softmax(axis = var_2634, x = aw_445_cast)[name = tensor<string, []>("op_6265_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6266_cast = softmax(axis = var_2634, x = aw_447_cast)[name = tensor<string, []>("op_6266_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6267_cast = softmax(axis = var_2634, x = aw_449_cast)[name = tensor<string, []>("op_6267_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6268_cast = softmax(axis = var_2634, x = aw_451_cast)[name = tensor<string, []>("op_6268_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6269_cast = softmax(axis = var_2634, x = aw_453_cast)[name = tensor<string, []>("op_6269_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6270_cast = softmax(axis = var_2634, x = aw_455_cast)[name = tensor<string, []>("op_6270_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6271_cast = softmax(axis = var_2634, x = aw_457_cast)[name = tensor<string, []>("op_6271_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6272_cast = softmax(axis = var_2634, x = aw_459_cast)[name = tensor<string, []>("op_6272_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6273_cast = softmax(axis = var_2634, x = aw_461_cast)[name = tensor<string, []>("op_6273_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6274_cast = softmax(axis = var_2634, x = aw_463_cast)[name = tensor<string, []>("op_6274_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6275_cast = softmax(axis = var_2634, x = aw_465_cast)[name = tensor<string, []>("op_6275_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6276_cast = softmax(axis = var_2634, x = aw_467_cast)[name = tensor<string, []>("op_6276_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6277_cast = softmax(axis = var_2634, x = aw_469_cast)[name = tensor<string, []>("op_6277_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6278_cast = softmax(axis = var_2634, x = aw_471_cast)[name = tensor<string, []>("op_6278_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6279_cast = softmax(axis = var_2634, x = aw_473_cast)[name = tensor<string, []>("op_6279_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6280_cast = softmax(axis = var_2634, x = aw_475_cast)[name = tensor<string, []>("op_6280_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6281_cast = softmax(axis = var_2634, x = aw_477_cast)[name = tensor<string, []>("op_6281_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_6282_cast = softmax(axis = var_2634, x = aw_479_cast)[name = tensor<string, []>("op_6282_cast")];
            tensor<string, []> var_6284_equation_0 = const()[name = tensor<string, []>("op_6284_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6284_cast = einsum(equation = var_6284_equation_0, values = (var_6104_cast, var_6263_cast))[name = tensor<string, []>("op_6284_cast")];
            tensor<string, []> var_6286_equation_0 = const()[name = tensor<string, []>("op_6286_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6286_cast = einsum(equation = var_6286_equation_0, values = (var_6108_cast, var_6264_cast))[name = tensor<string, []>("op_6286_cast")];
            tensor<string, []> var_6288_equation_0 = const()[name = tensor<string, []>("op_6288_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6288_cast = einsum(equation = var_6288_equation_0, values = (var_6112_cast, var_6265_cast))[name = tensor<string, []>("op_6288_cast")];
            tensor<string, []> var_6290_equation_0 = const()[name = tensor<string, []>("op_6290_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6290_cast = einsum(equation = var_6290_equation_0, values = (var_6116_cast, var_6266_cast))[name = tensor<string, []>("op_6290_cast")];
            tensor<string, []> var_6292_equation_0 = const()[name = tensor<string, []>("op_6292_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6292_cast = einsum(equation = var_6292_equation_0, values = (var_6120_cast, var_6267_cast))[name = tensor<string, []>("op_6292_cast")];
            tensor<string, []> var_6294_equation_0 = const()[name = tensor<string, []>("op_6294_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6294_cast = einsum(equation = var_6294_equation_0, values = (var_6124_cast, var_6268_cast))[name = tensor<string, []>("op_6294_cast")];
            tensor<string, []> var_6296_equation_0 = const()[name = tensor<string, []>("op_6296_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6296_cast = einsum(equation = var_6296_equation_0, values = (var_6128_cast, var_6269_cast))[name = tensor<string, []>("op_6296_cast")];
            tensor<string, []> var_6298_equation_0 = const()[name = tensor<string, []>("op_6298_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6298_cast = einsum(equation = var_6298_equation_0, values = (var_6132_cast, var_6270_cast))[name = tensor<string, []>("op_6298_cast")];
            tensor<string, []> var_6300_equation_0 = const()[name = tensor<string, []>("op_6300_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6300_cast = einsum(equation = var_6300_equation_0, values = (var_6136_cast, var_6271_cast))[name = tensor<string, []>("op_6300_cast")];
            tensor<string, []> var_6302_equation_0 = const()[name = tensor<string, []>("op_6302_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6302_cast = einsum(equation = var_6302_equation_0, values = (var_6140_cast, var_6272_cast))[name = tensor<string, []>("op_6302_cast")];
            tensor<string, []> var_6304_equation_0 = const()[name = tensor<string, []>("op_6304_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6304_cast = einsum(equation = var_6304_equation_0, values = (var_6144_cast, var_6273_cast))[name = tensor<string, []>("op_6304_cast")];
            tensor<string, []> var_6306_equation_0 = const()[name = tensor<string, []>("op_6306_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6306_cast = einsum(equation = var_6306_equation_0, values = (var_6148_cast, var_6274_cast))[name = tensor<string, []>("op_6306_cast")];
            tensor<string, []> var_6308_equation_0 = const()[name = tensor<string, []>("op_6308_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6308_cast = einsum(equation = var_6308_equation_0, values = (var_6152_cast, var_6275_cast))[name = tensor<string, []>("op_6308_cast")];
            tensor<string, []> var_6310_equation_0 = const()[name = tensor<string, []>("op_6310_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6310_cast = einsum(equation = var_6310_equation_0, values = (var_6156_cast, var_6276_cast))[name = tensor<string, []>("op_6310_cast")];
            tensor<string, []> var_6312_equation_0 = const()[name = tensor<string, []>("op_6312_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6312_cast = einsum(equation = var_6312_equation_0, values = (var_6160_cast, var_6277_cast))[name = tensor<string, []>("op_6312_cast")];
            tensor<string, []> var_6314_equation_0 = const()[name = tensor<string, []>("op_6314_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6314_cast = einsum(equation = var_6314_equation_0, values = (var_6164_cast, var_6278_cast))[name = tensor<string, []>("op_6314_cast")];
            tensor<string, []> var_6316_equation_0 = const()[name = tensor<string, []>("op_6316_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6316_cast = einsum(equation = var_6316_equation_0, values = (var_6168_cast, var_6279_cast))[name = tensor<string, []>("op_6316_cast")];
            tensor<string, []> var_6318_equation_0 = const()[name = tensor<string, []>("op_6318_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6318_cast = einsum(equation = var_6318_equation_0, values = (var_6172_cast, var_6280_cast))[name = tensor<string, []>("op_6318_cast")];
            tensor<string, []> var_6320_equation_0 = const()[name = tensor<string, []>("op_6320_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6320_cast = einsum(equation = var_6320_equation_0, values = (var_6176_cast, var_6281_cast))[name = tensor<string, []>("op_6320_cast")];
            tensor<string, []> var_6322_equation_0 = const()[name = tensor<string, []>("op_6322_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6322_cast = einsum(equation = var_6322_equation_0, values = (var_6180_cast, var_6282_cast))[name = tensor<string, []>("op_6322_cast")];
            tensor<bool, []> input_161_interleave_0 = const()[name = tensor<string, []>("input_161_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_161_cast = concat(axis = var_2634, interleave = input_161_interleave_0, values = (var_6284_cast, var_6286_cast, var_6288_cast, var_6290_cast, var_6292_cast, var_6294_cast, var_6296_cast, var_6298_cast, var_6300_cast, var_6302_cast, var_6304_cast, var_6306_cast, var_6308_cast, var_6310_cast, var_6312_cast, var_6314_cast, var_6316_cast, var_6318_cast, var_6320_cast, var_6322_cast))[name = tensor<string, []>("input_161_cast")];
            tensor<int32, [2]> var_6328 = const()[name = tensor<string, []>("op_6328"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6330 = const()[name = tensor<string, []>("op_6330"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_6332_pad_type_0 = const()[name = tensor<string, []>("op_6332_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_6332_pad_0 = const()[name = tensor<string, []>("op_6332_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(433588096)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(436864960)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_6332_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16, dilations = var_6330, groups = var_2634, pad = var_6332_pad_0, pad_type = var_6332_pad_type_0, strides = var_6328, weight = down_blocks_2_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16, x = input_161_cast)[name = tensor<string, []>("op_6332_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_47_cast = add(x = var_6332_cast, y = inputs_45_cast)[name = tensor<string, []>("inputs_47_cast")];
            tensor<int32, [1]> var_6336 = const()[name = tensor<string, []>("op_6336"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_47_cast = reduce_mean(axes = var_6336, keep_dims = var_2629, x = inputs_47_cast)[name = tensor<string, []>("channels_mean_47_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_47_cast = sub(x = inputs_47_cast, y = channels_mean_47_cast)[name = tensor<string, []>("zero_mean_47_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_47_cast = mul(x = zero_mean_47_cast, y = zero_mean_47_cast)[name = tensor<string, []>("zero_mean_sq_47_cast")];
            tensor<int32, [1]> var_6340 = const()[name = tensor<string, []>("op_6340"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_6341_cast = reduce_mean(axes = var_6340, keep_dims = var_2629, x = zero_mean_sq_47_cast)[name = tensor<string, []>("op_6341_cast")];
            tensor<fp16, []> var_6342_to_fp16 = const()[name = tensor<string, []>("op_6342_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_6343_cast = add(x = var_6341_cast, y = var_6342_to_fp16)[name = tensor<string, []>("op_6343_cast")];
            tensor<fp16, []> denom_47_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_47_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_47_cast = rsqrt(epsilon = denom_47_epsilon_0_to_fp16, x = var_6343_cast)[name = tensor<string, []>("denom_47_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_47_cast = mul(x = zero_mean_47_cast, y = denom_47_cast)[name = tensor<string, []>("out_47_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_6347_to_fp16 = const()[name = tensor<string, []>("op_6347_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(436867584)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_6348_cast = add(x = out_47_cast, y = var_6347_to_fp16)[name = tensor<string, []>("op_6348_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_6350_to_fp16 = const()[name = tensor<string, []>("op_6350_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(436870208)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_163_cast = mul(x = var_6348_cast, y = var_6350_to_fp16)[name = tensor<string, []>("input_163_cast")];
            tensor<int32, [2]> var_6358 = const()[name = tensor<string, []>("op_6358"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6360 = const()[name = tensor<string, []>("op_6360"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_6362_pad_type_0 = const()[name = tensor<string, []>("op_6362_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_6362_pad_0 = const()[name = tensor<string, []>("op_6362_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(436872832)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(463087296)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_6362_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16, dilations = var_6360, groups = var_2634, pad = var_6362_pad_0, pad_type = var_6362_pad_type_0, strides = var_6358, weight = down_blocks_2_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16, x = input_163_cast)[name = tensor<string, []>("op_6362_cast")];
            tensor<int32, [2]> var_6363_split_sizes_0 = const()[name = tensor<string, []>("op_6363_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_6363_axis_0 = const()[name = tensor<string, []>("op_6363_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_6363_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_6363_cast_1 = split(axis = var_6363_axis_0, split_sizes = var_6363_split_sizes_0, x = var_6362_cast)[name = tensor<string, []>("op_6363_cast")];
            tensor<string, []> var_6365_mode_0 = const()[name = tensor<string, []>("op_6365_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_6365_cast = gelu(mode = var_6365_mode_0, x = var_6363_cast_1)[name = tensor<string, []>("op_6365_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_165_cast = mul(x = var_6363_cast_0, y = var_6365_cast)[name = tensor<string, []>("input_165_cast")];
            tensor<int32, [2]> var_6369 = const()[name = tensor<string, []>("op_6369"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6371 = const()[name = tensor<string, []>("op_6371"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_6373_pad_type_0 = const()[name = tensor<string, []>("op_6373_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_6373_pad_0 = const()[name = tensor<string, []>("op_6373_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(463107840)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(476215104)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_6373_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16, dilations = var_6371, groups = var_2634, pad = var_6373_pad_0, pad_type = var_6373_pad_type_0, strides = var_6369, weight = down_blocks_2_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16, x = input_165_cast)[name = tensor<string, []>("op_6373_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_49_cast = add(x = var_6373_cast, y = inputs_47_cast)[name = tensor<string, []>("inputs_49_cast")];
            tensor<int32, [1]> var_6383 = const()[name = tensor<string, []>("op_6383"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_49_cast = reduce_mean(axes = var_6383, keep_dims = var_2629, x = inputs_49_cast)[name = tensor<string, []>("channels_mean_49_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_49_cast = sub(x = inputs_49_cast, y = channels_mean_49_cast)[name = tensor<string, []>("zero_mean_49_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_49_cast = mul(x = zero_mean_49_cast, y = zero_mean_49_cast)[name = tensor<string, []>("zero_mean_sq_49_cast")];
            tensor<int32, [1]> var_6387 = const()[name = tensor<string, []>("op_6387"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_6388_cast = reduce_mean(axes = var_6387, keep_dims = var_2629, x = zero_mean_sq_49_cast)[name = tensor<string, []>("op_6388_cast")];
            tensor<fp16, []> var_6389_to_fp16 = const()[name = tensor<string, []>("op_6389_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_6390_cast = add(x = var_6388_cast, y = var_6389_to_fp16)[name = tensor<string, []>("op_6390_cast")];
            tensor<fp16, []> denom_49_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_49_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_49_cast = rsqrt(epsilon = denom_49_epsilon_0_to_fp16, x = var_6390_cast)[name = tensor<string, []>("denom_49_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_49_cast = mul(x = zero_mean_49_cast, y = denom_49_cast)[name = tensor<string, []>("out_49_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_6394_to_fp16 = const()[name = tensor<string, []>("op_6394_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(476217728)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_6395_cast = add(x = out_49_cast, y = var_6394_to_fp16)[name = tensor<string, []>("op_6395_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_6397_to_fp16 = const()[name = tensor<string, []>("op_6397_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(476220352)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_89_cast = mul(x = var_6395_cast, y = var_6397_to_fp16)[name = tensor<string, []>("hidden_states_89_cast")];
            tensor<int32, [2]> var_6404 = const()[name = tensor<string, []>("op_6404"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6406 = const()[name = tensor<string, []>("op_6406"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_33_pad_type_0 = const()[name = tensor<string, []>("q_33_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_33_pad_0 = const()[name = tensor<string, []>("q_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(476222976)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_33_cast = conv(dilations = var_6406, groups = var_2634, pad = q_33_pad_0, pad_type = q_33_pad_type_0, strides = var_6404, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16, x = hidden_states_89_cast)[name = tensor<string, []>("q_33_cast")];
            tensor<int32, [2]> var_6410 = const()[name = tensor<string, []>("op_6410"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6412 = const()[name = tensor<string, []>("op_6412"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_65_pad_type_0 = const()[name = tensor<string, []>("k_65_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_65_pad_0 = const()[name = tensor<string, []>("k_65_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(479499840)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_65_cast = conv(dilations = var_6412, groups = var_2634, pad = k_65_pad_0, pad_type = k_65_pad_type_0, strides = var_6410, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16, x = hidden_states_89_cast)[name = tensor<string, []>("k_65_cast")];
            tensor<int32, [2]> var_6416 = const()[name = tensor<string, []>("op_6416"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6418 = const()[name = tensor<string, []>("op_6418"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_33_pad_type_0 = const()[name = tensor<string, []>("v_33_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_33_pad_0 = const()[name = tensor<string, []>("v_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(482776704)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_33_cast = conv(dilations = var_6418, groups = var_2634, pad = v_33_pad_0, pad_type = v_33_pad_type_0, strides = var_6416, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16, x = hidden_states_89_cast)[name = tensor<string, []>("v_33_cast")];
            tensor<int32, [4]> var_6422_begin_0 = const()[name = tensor<string, []>("op_6422_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6422_end_0 = const()[name = tensor<string, []>("op_6422_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_6422_end_mask_0 = const()[name = tensor<string, []>("op_6422_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6422_cast = slice_by_index(begin = var_6422_begin_0, end = var_6422_end_0, end_mask = var_6422_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6422_cast")];
            tensor<int32, [4]> var_6426_begin_0 = const()[name = tensor<string, []>("op_6426_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_6426_end_0 = const()[name = tensor<string, []>("op_6426_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_6426_end_mask_0 = const()[name = tensor<string, []>("op_6426_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6426_cast = slice_by_index(begin = var_6426_begin_0, end = var_6426_end_0, end_mask = var_6426_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6426_cast")];
            tensor<int32, [4]> var_6430_begin_0 = const()[name = tensor<string, []>("op_6430_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_6430_end_0 = const()[name = tensor<string, []>("op_6430_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_6430_end_mask_0 = const()[name = tensor<string, []>("op_6430_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6430_cast = slice_by_index(begin = var_6430_begin_0, end = var_6430_end_0, end_mask = var_6430_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6430_cast")];
            tensor<int32, [4]> var_6434_begin_0 = const()[name = tensor<string, []>("op_6434_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_6434_end_0 = const()[name = tensor<string, []>("op_6434_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_6434_end_mask_0 = const()[name = tensor<string, []>("op_6434_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6434_cast = slice_by_index(begin = var_6434_begin_0, end = var_6434_end_0, end_mask = var_6434_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6434_cast")];
            tensor<int32, [4]> var_6438_begin_0 = const()[name = tensor<string, []>("op_6438_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_6438_end_0 = const()[name = tensor<string, []>("op_6438_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_6438_end_mask_0 = const()[name = tensor<string, []>("op_6438_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6438_cast = slice_by_index(begin = var_6438_begin_0, end = var_6438_end_0, end_mask = var_6438_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6438_cast")];
            tensor<int32, [4]> var_6442_begin_0 = const()[name = tensor<string, []>("op_6442_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_6442_end_0 = const()[name = tensor<string, []>("op_6442_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_6442_end_mask_0 = const()[name = tensor<string, []>("op_6442_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6442_cast = slice_by_index(begin = var_6442_begin_0, end = var_6442_end_0, end_mask = var_6442_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6442_cast")];
            tensor<int32, [4]> var_6446_begin_0 = const()[name = tensor<string, []>("op_6446_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_6446_end_0 = const()[name = tensor<string, []>("op_6446_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_6446_end_mask_0 = const()[name = tensor<string, []>("op_6446_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6446_cast = slice_by_index(begin = var_6446_begin_0, end = var_6446_end_0, end_mask = var_6446_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6446_cast")];
            tensor<int32, [4]> var_6450_begin_0 = const()[name = tensor<string, []>("op_6450_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_6450_end_0 = const()[name = tensor<string, []>("op_6450_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_6450_end_mask_0 = const()[name = tensor<string, []>("op_6450_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6450_cast = slice_by_index(begin = var_6450_begin_0, end = var_6450_end_0, end_mask = var_6450_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6450_cast")];
            tensor<int32, [4]> var_6454_begin_0 = const()[name = tensor<string, []>("op_6454_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_6454_end_0 = const()[name = tensor<string, []>("op_6454_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_6454_end_mask_0 = const()[name = tensor<string, []>("op_6454_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6454_cast = slice_by_index(begin = var_6454_begin_0, end = var_6454_end_0, end_mask = var_6454_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6454_cast")];
            tensor<int32, [4]> var_6458_begin_0 = const()[name = tensor<string, []>("op_6458_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_6458_end_0 = const()[name = tensor<string, []>("op_6458_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_6458_end_mask_0 = const()[name = tensor<string, []>("op_6458_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6458_cast = slice_by_index(begin = var_6458_begin_0, end = var_6458_end_0, end_mask = var_6458_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6458_cast")];
            tensor<int32, [4]> var_6462_begin_0 = const()[name = tensor<string, []>("op_6462_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_6462_end_0 = const()[name = tensor<string, []>("op_6462_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_6462_end_mask_0 = const()[name = tensor<string, []>("op_6462_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6462_cast = slice_by_index(begin = var_6462_begin_0, end = var_6462_end_0, end_mask = var_6462_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6462_cast")];
            tensor<int32, [4]> var_6466_begin_0 = const()[name = tensor<string, []>("op_6466_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_6466_end_0 = const()[name = tensor<string, []>("op_6466_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_6466_end_mask_0 = const()[name = tensor<string, []>("op_6466_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6466_cast = slice_by_index(begin = var_6466_begin_0, end = var_6466_end_0, end_mask = var_6466_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6466_cast")];
            tensor<int32, [4]> var_6470_begin_0 = const()[name = tensor<string, []>("op_6470_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_6470_end_0 = const()[name = tensor<string, []>("op_6470_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_6470_end_mask_0 = const()[name = tensor<string, []>("op_6470_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6470_cast = slice_by_index(begin = var_6470_begin_0, end = var_6470_end_0, end_mask = var_6470_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6470_cast")];
            tensor<int32, [4]> var_6474_begin_0 = const()[name = tensor<string, []>("op_6474_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_6474_end_0 = const()[name = tensor<string, []>("op_6474_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_6474_end_mask_0 = const()[name = tensor<string, []>("op_6474_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6474_cast = slice_by_index(begin = var_6474_begin_0, end = var_6474_end_0, end_mask = var_6474_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6474_cast")];
            tensor<int32, [4]> var_6478_begin_0 = const()[name = tensor<string, []>("op_6478_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_6478_end_0 = const()[name = tensor<string, []>("op_6478_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_6478_end_mask_0 = const()[name = tensor<string, []>("op_6478_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6478_cast = slice_by_index(begin = var_6478_begin_0, end = var_6478_end_0, end_mask = var_6478_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6478_cast")];
            tensor<int32, [4]> var_6482_begin_0 = const()[name = tensor<string, []>("op_6482_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_6482_end_0 = const()[name = tensor<string, []>("op_6482_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_6482_end_mask_0 = const()[name = tensor<string, []>("op_6482_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6482_cast = slice_by_index(begin = var_6482_begin_0, end = var_6482_end_0, end_mask = var_6482_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6482_cast")];
            tensor<int32, [4]> var_6486_begin_0 = const()[name = tensor<string, []>("op_6486_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_6486_end_0 = const()[name = tensor<string, []>("op_6486_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_6486_end_mask_0 = const()[name = tensor<string, []>("op_6486_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6486_cast = slice_by_index(begin = var_6486_begin_0, end = var_6486_end_0, end_mask = var_6486_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6486_cast")];
            tensor<int32, [4]> var_6490_begin_0 = const()[name = tensor<string, []>("op_6490_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_6490_end_0 = const()[name = tensor<string, []>("op_6490_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_6490_end_mask_0 = const()[name = tensor<string, []>("op_6490_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6490_cast = slice_by_index(begin = var_6490_begin_0, end = var_6490_end_0, end_mask = var_6490_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6490_cast")];
            tensor<int32, [4]> var_6494_begin_0 = const()[name = tensor<string, []>("op_6494_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_6494_end_0 = const()[name = tensor<string, []>("op_6494_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_6494_end_mask_0 = const()[name = tensor<string, []>("op_6494_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6494_cast = slice_by_index(begin = var_6494_begin_0, end = var_6494_end_0, end_mask = var_6494_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6494_cast")];
            tensor<int32, [4]> var_6498_begin_0 = const()[name = tensor<string, []>("op_6498_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_6498_end_0 = const()[name = tensor<string, []>("op_6498_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_6498_end_mask_0 = const()[name = tensor<string, []>("op_6498_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6498_cast = slice_by_index(begin = var_6498_begin_0, end = var_6498_end_0, end_mask = var_6498_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_6498_cast")];
            tensor<int32, [4]> k_67_perm_0 = const()[name = tensor<string, []>("k_67_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_6505_begin_0 = const()[name = tensor<string, []>("op_6505_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6505_end_0 = const()[name = tensor<string, []>("op_6505_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_6505_end_mask_0 = const()[name = tensor<string, []>("op_6505_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_123 = transpose(perm = k_67_perm_0, x = k_65_cast)[name = tensor<string, []>("transpose_123")];
            tensor<fp16, [2, 1024, 1, 64]> var_6505_cast = slice_by_index(begin = var_6505_begin_0, end = var_6505_end_0, end_mask = var_6505_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6505_cast")];
            tensor<int32, [4]> var_6509_begin_0 = const()[name = tensor<string, []>("op_6509_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_6509_end_0 = const()[name = tensor<string, []>("op_6509_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_6509_end_mask_0 = const()[name = tensor<string, []>("op_6509_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_6509_cast = slice_by_index(begin = var_6509_begin_0, end = var_6509_end_0, end_mask = var_6509_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6509_cast")];
            tensor<int32, [4]> var_6513_begin_0 = const()[name = tensor<string, []>("op_6513_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_6513_end_0 = const()[name = tensor<string, []>("op_6513_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_6513_end_mask_0 = const()[name = tensor<string, []>("op_6513_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_6513_cast = slice_by_index(begin = var_6513_begin_0, end = var_6513_end_0, end_mask = var_6513_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6513_cast")];
            tensor<int32, [4]> var_6517_begin_0 = const()[name = tensor<string, []>("op_6517_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_6517_end_0 = const()[name = tensor<string, []>("op_6517_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_6517_end_mask_0 = const()[name = tensor<string, []>("op_6517_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_6517_cast = slice_by_index(begin = var_6517_begin_0, end = var_6517_end_0, end_mask = var_6517_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6517_cast")];
            tensor<int32, [4]> var_6521_begin_0 = const()[name = tensor<string, []>("op_6521_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_6521_end_0 = const()[name = tensor<string, []>("op_6521_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_6521_end_mask_0 = const()[name = tensor<string, []>("op_6521_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_6521_cast = slice_by_index(begin = var_6521_begin_0, end = var_6521_end_0, end_mask = var_6521_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6521_cast")];
            tensor<int32, [4]> var_6525_begin_0 = const()[name = tensor<string, []>("op_6525_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_6525_end_0 = const()[name = tensor<string, []>("op_6525_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_6525_end_mask_0 = const()[name = tensor<string, []>("op_6525_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_6525_cast = slice_by_index(begin = var_6525_begin_0, end = var_6525_end_0, end_mask = var_6525_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6525_cast")];
            tensor<int32, [4]> var_6529_begin_0 = const()[name = tensor<string, []>("op_6529_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_6529_end_0 = const()[name = tensor<string, []>("op_6529_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_6529_end_mask_0 = const()[name = tensor<string, []>("op_6529_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_6529_cast = slice_by_index(begin = var_6529_begin_0, end = var_6529_end_0, end_mask = var_6529_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6529_cast")];
            tensor<int32, [4]> var_6533_begin_0 = const()[name = tensor<string, []>("op_6533_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_6533_end_0 = const()[name = tensor<string, []>("op_6533_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_6533_end_mask_0 = const()[name = tensor<string, []>("op_6533_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_6533_cast = slice_by_index(begin = var_6533_begin_0, end = var_6533_end_0, end_mask = var_6533_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6533_cast")];
            tensor<int32, [4]> var_6537_begin_0 = const()[name = tensor<string, []>("op_6537_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_6537_end_0 = const()[name = tensor<string, []>("op_6537_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_6537_end_mask_0 = const()[name = tensor<string, []>("op_6537_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_6537_cast = slice_by_index(begin = var_6537_begin_0, end = var_6537_end_0, end_mask = var_6537_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6537_cast")];
            tensor<int32, [4]> var_6541_begin_0 = const()[name = tensor<string, []>("op_6541_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_6541_end_0 = const()[name = tensor<string, []>("op_6541_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_6541_end_mask_0 = const()[name = tensor<string, []>("op_6541_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_6541_cast = slice_by_index(begin = var_6541_begin_0, end = var_6541_end_0, end_mask = var_6541_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6541_cast")];
            tensor<int32, [4]> var_6545_begin_0 = const()[name = tensor<string, []>("op_6545_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_6545_end_0 = const()[name = tensor<string, []>("op_6545_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_6545_end_mask_0 = const()[name = tensor<string, []>("op_6545_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_6545_cast = slice_by_index(begin = var_6545_begin_0, end = var_6545_end_0, end_mask = var_6545_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6545_cast")];
            tensor<int32, [4]> var_6549_begin_0 = const()[name = tensor<string, []>("op_6549_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_6549_end_0 = const()[name = tensor<string, []>("op_6549_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_6549_end_mask_0 = const()[name = tensor<string, []>("op_6549_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_6549_cast = slice_by_index(begin = var_6549_begin_0, end = var_6549_end_0, end_mask = var_6549_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6549_cast")];
            tensor<int32, [4]> var_6553_begin_0 = const()[name = tensor<string, []>("op_6553_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_6553_end_0 = const()[name = tensor<string, []>("op_6553_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_6553_end_mask_0 = const()[name = tensor<string, []>("op_6553_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_6553_cast = slice_by_index(begin = var_6553_begin_0, end = var_6553_end_0, end_mask = var_6553_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6553_cast")];
            tensor<int32, [4]> var_6557_begin_0 = const()[name = tensor<string, []>("op_6557_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_6557_end_0 = const()[name = tensor<string, []>("op_6557_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_6557_end_mask_0 = const()[name = tensor<string, []>("op_6557_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_6557_cast = slice_by_index(begin = var_6557_begin_0, end = var_6557_end_0, end_mask = var_6557_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6557_cast")];
            tensor<int32, [4]> var_6561_begin_0 = const()[name = tensor<string, []>("op_6561_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_6561_end_0 = const()[name = tensor<string, []>("op_6561_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_6561_end_mask_0 = const()[name = tensor<string, []>("op_6561_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_6561_cast = slice_by_index(begin = var_6561_begin_0, end = var_6561_end_0, end_mask = var_6561_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6561_cast")];
            tensor<int32, [4]> var_6565_begin_0 = const()[name = tensor<string, []>("op_6565_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_6565_end_0 = const()[name = tensor<string, []>("op_6565_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_6565_end_mask_0 = const()[name = tensor<string, []>("op_6565_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_6565_cast = slice_by_index(begin = var_6565_begin_0, end = var_6565_end_0, end_mask = var_6565_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6565_cast")];
            tensor<int32, [4]> var_6569_begin_0 = const()[name = tensor<string, []>("op_6569_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_6569_end_0 = const()[name = tensor<string, []>("op_6569_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_6569_end_mask_0 = const()[name = tensor<string, []>("op_6569_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_6569_cast = slice_by_index(begin = var_6569_begin_0, end = var_6569_end_0, end_mask = var_6569_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6569_cast")];
            tensor<int32, [4]> var_6573_begin_0 = const()[name = tensor<string, []>("op_6573_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_6573_end_0 = const()[name = tensor<string, []>("op_6573_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_6573_end_mask_0 = const()[name = tensor<string, []>("op_6573_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_6573_cast = slice_by_index(begin = var_6573_begin_0, end = var_6573_end_0, end_mask = var_6573_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6573_cast")];
            tensor<int32, [4]> var_6577_begin_0 = const()[name = tensor<string, []>("op_6577_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_6577_end_0 = const()[name = tensor<string, []>("op_6577_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_6577_end_mask_0 = const()[name = tensor<string, []>("op_6577_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_6577_cast = slice_by_index(begin = var_6577_begin_0, end = var_6577_end_0, end_mask = var_6577_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6577_cast")];
            tensor<int32, [4]> var_6581_begin_0 = const()[name = tensor<string, []>("op_6581_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_6581_end_0 = const()[name = tensor<string, []>("op_6581_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_6581_end_mask_0 = const()[name = tensor<string, []>("op_6581_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_6581_cast = slice_by_index(begin = var_6581_begin_0, end = var_6581_end_0, end_mask = var_6581_end_mask_0, x = transpose_123)[name = tensor<string, []>("op_6581_cast")];
            tensor<int32, [4]> var_6583_begin_0 = const()[name = tensor<string, []>("op_6583_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6583_end_0 = const()[name = tensor<string, []>("op_6583_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_6583_end_mask_0 = const()[name = tensor<string, []>("op_6583_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6583_cast = slice_by_index(begin = var_6583_begin_0, end = var_6583_end_0, end_mask = var_6583_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6583_cast")];
            tensor<int32, [4]> var_6587_begin_0 = const()[name = tensor<string, []>("op_6587_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_6587_end_0 = const()[name = tensor<string, []>("op_6587_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_6587_end_mask_0 = const()[name = tensor<string, []>("op_6587_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6587_cast = slice_by_index(begin = var_6587_begin_0, end = var_6587_end_0, end_mask = var_6587_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6587_cast")];
            tensor<int32, [4]> var_6591_begin_0 = const()[name = tensor<string, []>("op_6591_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_6591_end_0 = const()[name = tensor<string, []>("op_6591_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_6591_end_mask_0 = const()[name = tensor<string, []>("op_6591_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6591_cast = slice_by_index(begin = var_6591_begin_0, end = var_6591_end_0, end_mask = var_6591_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6591_cast")];
            tensor<int32, [4]> var_6595_begin_0 = const()[name = tensor<string, []>("op_6595_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_6595_end_0 = const()[name = tensor<string, []>("op_6595_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_6595_end_mask_0 = const()[name = tensor<string, []>("op_6595_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6595_cast = slice_by_index(begin = var_6595_begin_0, end = var_6595_end_0, end_mask = var_6595_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6595_cast")];
            tensor<int32, [4]> var_6599_begin_0 = const()[name = tensor<string, []>("op_6599_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_6599_end_0 = const()[name = tensor<string, []>("op_6599_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_6599_end_mask_0 = const()[name = tensor<string, []>("op_6599_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6599_cast = slice_by_index(begin = var_6599_begin_0, end = var_6599_end_0, end_mask = var_6599_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6599_cast")];
            tensor<int32, [4]> var_6603_begin_0 = const()[name = tensor<string, []>("op_6603_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_6603_end_0 = const()[name = tensor<string, []>("op_6603_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_6603_end_mask_0 = const()[name = tensor<string, []>("op_6603_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6603_cast = slice_by_index(begin = var_6603_begin_0, end = var_6603_end_0, end_mask = var_6603_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6603_cast")];
            tensor<int32, [4]> var_6607_begin_0 = const()[name = tensor<string, []>("op_6607_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_6607_end_0 = const()[name = tensor<string, []>("op_6607_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_6607_end_mask_0 = const()[name = tensor<string, []>("op_6607_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6607_cast = slice_by_index(begin = var_6607_begin_0, end = var_6607_end_0, end_mask = var_6607_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6607_cast")];
            tensor<int32, [4]> var_6611_begin_0 = const()[name = tensor<string, []>("op_6611_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_6611_end_0 = const()[name = tensor<string, []>("op_6611_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_6611_end_mask_0 = const()[name = tensor<string, []>("op_6611_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6611_cast = slice_by_index(begin = var_6611_begin_0, end = var_6611_end_0, end_mask = var_6611_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6611_cast")];
            tensor<int32, [4]> var_6615_begin_0 = const()[name = tensor<string, []>("op_6615_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_6615_end_0 = const()[name = tensor<string, []>("op_6615_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_6615_end_mask_0 = const()[name = tensor<string, []>("op_6615_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6615_cast = slice_by_index(begin = var_6615_begin_0, end = var_6615_end_0, end_mask = var_6615_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6615_cast")];
            tensor<int32, [4]> var_6619_begin_0 = const()[name = tensor<string, []>("op_6619_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_6619_end_0 = const()[name = tensor<string, []>("op_6619_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_6619_end_mask_0 = const()[name = tensor<string, []>("op_6619_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6619_cast = slice_by_index(begin = var_6619_begin_0, end = var_6619_end_0, end_mask = var_6619_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6619_cast")];
            tensor<int32, [4]> var_6623_begin_0 = const()[name = tensor<string, []>("op_6623_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_6623_end_0 = const()[name = tensor<string, []>("op_6623_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_6623_end_mask_0 = const()[name = tensor<string, []>("op_6623_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6623_cast = slice_by_index(begin = var_6623_begin_0, end = var_6623_end_0, end_mask = var_6623_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6623_cast")];
            tensor<int32, [4]> var_6627_begin_0 = const()[name = tensor<string, []>("op_6627_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_6627_end_0 = const()[name = tensor<string, []>("op_6627_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_6627_end_mask_0 = const()[name = tensor<string, []>("op_6627_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6627_cast = slice_by_index(begin = var_6627_begin_0, end = var_6627_end_0, end_mask = var_6627_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6627_cast")];
            tensor<int32, [4]> var_6631_begin_0 = const()[name = tensor<string, []>("op_6631_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_6631_end_0 = const()[name = tensor<string, []>("op_6631_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_6631_end_mask_0 = const()[name = tensor<string, []>("op_6631_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6631_cast = slice_by_index(begin = var_6631_begin_0, end = var_6631_end_0, end_mask = var_6631_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6631_cast")];
            tensor<int32, [4]> var_6635_begin_0 = const()[name = tensor<string, []>("op_6635_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_6635_end_0 = const()[name = tensor<string, []>("op_6635_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_6635_end_mask_0 = const()[name = tensor<string, []>("op_6635_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6635_cast = slice_by_index(begin = var_6635_begin_0, end = var_6635_end_0, end_mask = var_6635_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6635_cast")];
            tensor<int32, [4]> var_6639_begin_0 = const()[name = tensor<string, []>("op_6639_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_6639_end_0 = const()[name = tensor<string, []>("op_6639_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_6639_end_mask_0 = const()[name = tensor<string, []>("op_6639_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6639_cast = slice_by_index(begin = var_6639_begin_0, end = var_6639_end_0, end_mask = var_6639_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6639_cast")];
            tensor<int32, [4]> var_6643_begin_0 = const()[name = tensor<string, []>("op_6643_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_6643_end_0 = const()[name = tensor<string, []>("op_6643_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_6643_end_mask_0 = const()[name = tensor<string, []>("op_6643_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6643_cast = slice_by_index(begin = var_6643_begin_0, end = var_6643_end_0, end_mask = var_6643_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6643_cast")];
            tensor<int32, [4]> var_6647_begin_0 = const()[name = tensor<string, []>("op_6647_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_6647_end_0 = const()[name = tensor<string, []>("op_6647_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_6647_end_mask_0 = const()[name = tensor<string, []>("op_6647_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6647_cast = slice_by_index(begin = var_6647_begin_0, end = var_6647_end_0, end_mask = var_6647_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6647_cast")];
            tensor<int32, [4]> var_6651_begin_0 = const()[name = tensor<string, []>("op_6651_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_6651_end_0 = const()[name = tensor<string, []>("op_6651_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_6651_end_mask_0 = const()[name = tensor<string, []>("op_6651_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6651_cast = slice_by_index(begin = var_6651_begin_0, end = var_6651_end_0, end_mask = var_6651_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6651_cast")];
            tensor<int32, [4]> var_6655_begin_0 = const()[name = tensor<string, []>("op_6655_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_6655_end_0 = const()[name = tensor<string, []>("op_6655_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_6655_end_mask_0 = const()[name = tensor<string, []>("op_6655_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6655_cast = slice_by_index(begin = var_6655_begin_0, end = var_6655_end_0, end_mask = var_6655_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6655_cast")];
            tensor<int32, [4]> var_6659_begin_0 = const()[name = tensor<string, []>("op_6659_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_6659_end_0 = const()[name = tensor<string, []>("op_6659_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_6659_end_mask_0 = const()[name = tensor<string, []>("op_6659_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6659_cast = slice_by_index(begin = var_6659_begin_0, end = var_6659_end_0, end_mask = var_6659_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_6659_cast")];
            tensor<string, []> var_6663_equation_0 = const()[name = tensor<string, []>("op_6663_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6663_cast = einsum(equation = var_6663_equation_0, values = (var_6505_cast, var_6422_cast))[name = tensor<string, []>("op_6663_cast")];
            tensor<fp16, []> var_6664_to_fp16 = const()[name = tensor<string, []>("op_6664_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_481_cast = mul(x = var_6663_cast, y = var_6664_to_fp16)[name = tensor<string, []>("aw_481_cast")];
            tensor<string, []> var_6667_equation_0 = const()[name = tensor<string, []>("op_6667_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6667_cast = einsum(equation = var_6667_equation_0, values = (var_6509_cast, var_6426_cast))[name = tensor<string, []>("op_6667_cast")];
            tensor<fp16, []> var_6668_to_fp16 = const()[name = tensor<string, []>("op_6668_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_483_cast = mul(x = var_6667_cast, y = var_6668_to_fp16)[name = tensor<string, []>("aw_483_cast")];
            tensor<string, []> var_6671_equation_0 = const()[name = tensor<string, []>("op_6671_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6671_cast = einsum(equation = var_6671_equation_0, values = (var_6513_cast, var_6430_cast))[name = tensor<string, []>("op_6671_cast")];
            tensor<fp16, []> var_6672_to_fp16 = const()[name = tensor<string, []>("op_6672_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_485_cast = mul(x = var_6671_cast, y = var_6672_to_fp16)[name = tensor<string, []>("aw_485_cast")];
            tensor<string, []> var_6675_equation_0 = const()[name = tensor<string, []>("op_6675_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6675_cast = einsum(equation = var_6675_equation_0, values = (var_6517_cast, var_6434_cast))[name = tensor<string, []>("op_6675_cast")];
            tensor<fp16, []> var_6676_to_fp16 = const()[name = tensor<string, []>("op_6676_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_487_cast = mul(x = var_6675_cast, y = var_6676_to_fp16)[name = tensor<string, []>("aw_487_cast")];
            tensor<string, []> var_6679_equation_0 = const()[name = tensor<string, []>("op_6679_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6679_cast = einsum(equation = var_6679_equation_0, values = (var_6521_cast, var_6438_cast))[name = tensor<string, []>("op_6679_cast")];
            tensor<fp16, []> var_6680_to_fp16 = const()[name = tensor<string, []>("op_6680_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_489_cast = mul(x = var_6679_cast, y = var_6680_to_fp16)[name = tensor<string, []>("aw_489_cast")];
            tensor<string, []> var_6683_equation_0 = const()[name = tensor<string, []>("op_6683_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6683_cast = einsum(equation = var_6683_equation_0, values = (var_6525_cast, var_6442_cast))[name = tensor<string, []>("op_6683_cast")];
            tensor<fp16, []> var_6684_to_fp16 = const()[name = tensor<string, []>("op_6684_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_491_cast = mul(x = var_6683_cast, y = var_6684_to_fp16)[name = tensor<string, []>("aw_491_cast")];
            tensor<string, []> var_6687_equation_0 = const()[name = tensor<string, []>("op_6687_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6687_cast = einsum(equation = var_6687_equation_0, values = (var_6529_cast, var_6446_cast))[name = tensor<string, []>("op_6687_cast")];
            tensor<fp16, []> var_6688_to_fp16 = const()[name = tensor<string, []>("op_6688_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_493_cast = mul(x = var_6687_cast, y = var_6688_to_fp16)[name = tensor<string, []>("aw_493_cast")];
            tensor<string, []> var_6691_equation_0 = const()[name = tensor<string, []>("op_6691_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6691_cast = einsum(equation = var_6691_equation_0, values = (var_6533_cast, var_6450_cast))[name = tensor<string, []>("op_6691_cast")];
            tensor<fp16, []> var_6692_to_fp16 = const()[name = tensor<string, []>("op_6692_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_495_cast = mul(x = var_6691_cast, y = var_6692_to_fp16)[name = tensor<string, []>("aw_495_cast")];
            tensor<string, []> var_6695_equation_0 = const()[name = tensor<string, []>("op_6695_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6695_cast = einsum(equation = var_6695_equation_0, values = (var_6537_cast, var_6454_cast))[name = tensor<string, []>("op_6695_cast")];
            tensor<fp16, []> var_6696_to_fp16 = const()[name = tensor<string, []>("op_6696_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_497_cast = mul(x = var_6695_cast, y = var_6696_to_fp16)[name = tensor<string, []>("aw_497_cast")];
            tensor<string, []> var_6699_equation_0 = const()[name = tensor<string, []>("op_6699_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6699_cast = einsum(equation = var_6699_equation_0, values = (var_6541_cast, var_6458_cast))[name = tensor<string, []>("op_6699_cast")];
            tensor<fp16, []> var_6700_to_fp16 = const()[name = tensor<string, []>("op_6700_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_499_cast = mul(x = var_6699_cast, y = var_6700_to_fp16)[name = tensor<string, []>("aw_499_cast")];
            tensor<string, []> var_6703_equation_0 = const()[name = tensor<string, []>("op_6703_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6703_cast = einsum(equation = var_6703_equation_0, values = (var_6545_cast, var_6462_cast))[name = tensor<string, []>("op_6703_cast")];
            tensor<fp16, []> var_6704_to_fp16 = const()[name = tensor<string, []>("op_6704_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_501_cast = mul(x = var_6703_cast, y = var_6704_to_fp16)[name = tensor<string, []>("aw_501_cast")];
            tensor<string, []> var_6707_equation_0 = const()[name = tensor<string, []>("op_6707_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6707_cast = einsum(equation = var_6707_equation_0, values = (var_6549_cast, var_6466_cast))[name = tensor<string, []>("op_6707_cast")];
            tensor<fp16, []> var_6708_to_fp16 = const()[name = tensor<string, []>("op_6708_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_503_cast = mul(x = var_6707_cast, y = var_6708_to_fp16)[name = tensor<string, []>("aw_503_cast")];
            tensor<string, []> var_6711_equation_0 = const()[name = tensor<string, []>("op_6711_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6711_cast = einsum(equation = var_6711_equation_0, values = (var_6553_cast, var_6470_cast))[name = tensor<string, []>("op_6711_cast")];
            tensor<fp16, []> var_6712_to_fp16 = const()[name = tensor<string, []>("op_6712_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_505_cast = mul(x = var_6711_cast, y = var_6712_to_fp16)[name = tensor<string, []>("aw_505_cast")];
            tensor<string, []> var_6715_equation_0 = const()[name = tensor<string, []>("op_6715_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6715_cast = einsum(equation = var_6715_equation_0, values = (var_6557_cast, var_6474_cast))[name = tensor<string, []>("op_6715_cast")];
            tensor<fp16, []> var_6716_to_fp16 = const()[name = tensor<string, []>("op_6716_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_507_cast = mul(x = var_6715_cast, y = var_6716_to_fp16)[name = tensor<string, []>("aw_507_cast")];
            tensor<string, []> var_6719_equation_0 = const()[name = tensor<string, []>("op_6719_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6719_cast = einsum(equation = var_6719_equation_0, values = (var_6561_cast, var_6478_cast))[name = tensor<string, []>("op_6719_cast")];
            tensor<fp16, []> var_6720_to_fp16 = const()[name = tensor<string, []>("op_6720_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_509_cast = mul(x = var_6719_cast, y = var_6720_to_fp16)[name = tensor<string, []>("aw_509_cast")];
            tensor<string, []> var_6723_equation_0 = const()[name = tensor<string, []>("op_6723_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6723_cast = einsum(equation = var_6723_equation_0, values = (var_6565_cast, var_6482_cast))[name = tensor<string, []>("op_6723_cast")];
            tensor<fp16, []> var_6724_to_fp16 = const()[name = tensor<string, []>("op_6724_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_511_cast = mul(x = var_6723_cast, y = var_6724_to_fp16)[name = tensor<string, []>("aw_511_cast")];
            tensor<string, []> var_6727_equation_0 = const()[name = tensor<string, []>("op_6727_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6727_cast = einsum(equation = var_6727_equation_0, values = (var_6569_cast, var_6486_cast))[name = tensor<string, []>("op_6727_cast")];
            tensor<fp16, []> var_6728_to_fp16 = const()[name = tensor<string, []>("op_6728_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_513_cast = mul(x = var_6727_cast, y = var_6728_to_fp16)[name = tensor<string, []>("aw_513_cast")];
            tensor<string, []> var_6731_equation_0 = const()[name = tensor<string, []>("op_6731_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6731_cast = einsum(equation = var_6731_equation_0, values = (var_6573_cast, var_6490_cast))[name = tensor<string, []>("op_6731_cast")];
            tensor<fp16, []> var_6732_to_fp16 = const()[name = tensor<string, []>("op_6732_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_515_cast = mul(x = var_6731_cast, y = var_6732_to_fp16)[name = tensor<string, []>("aw_515_cast")];
            tensor<string, []> var_6735_equation_0 = const()[name = tensor<string, []>("op_6735_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6735_cast = einsum(equation = var_6735_equation_0, values = (var_6577_cast, var_6494_cast))[name = tensor<string, []>("op_6735_cast")];
            tensor<fp16, []> var_6736_to_fp16 = const()[name = tensor<string, []>("op_6736_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_517_cast = mul(x = var_6735_cast, y = var_6736_to_fp16)[name = tensor<string, []>("aw_517_cast")];
            tensor<string, []> var_6739_equation_0 = const()[name = tensor<string, []>("op_6739_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6739_cast = einsum(equation = var_6739_equation_0, values = (var_6581_cast, var_6498_cast))[name = tensor<string, []>("op_6739_cast")];
            tensor<fp16, []> var_6740_to_fp16 = const()[name = tensor<string, []>("op_6740_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_519_cast = mul(x = var_6739_cast, y = var_6740_to_fp16)[name = tensor<string, []>("aw_519_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6742_cast = softmax(axis = var_2634, x = aw_481_cast)[name = tensor<string, []>("op_6742_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6743_cast = softmax(axis = var_2634, x = aw_483_cast)[name = tensor<string, []>("op_6743_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6744_cast = softmax(axis = var_2634, x = aw_485_cast)[name = tensor<string, []>("op_6744_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6745_cast = softmax(axis = var_2634, x = aw_487_cast)[name = tensor<string, []>("op_6745_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6746_cast = softmax(axis = var_2634, x = aw_489_cast)[name = tensor<string, []>("op_6746_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6747_cast = softmax(axis = var_2634, x = aw_491_cast)[name = tensor<string, []>("op_6747_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6748_cast = softmax(axis = var_2634, x = aw_493_cast)[name = tensor<string, []>("op_6748_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6749_cast = softmax(axis = var_2634, x = aw_495_cast)[name = tensor<string, []>("op_6749_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6750_cast = softmax(axis = var_2634, x = aw_497_cast)[name = tensor<string, []>("op_6750_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6751_cast = softmax(axis = var_2634, x = aw_499_cast)[name = tensor<string, []>("op_6751_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6752_cast = softmax(axis = var_2634, x = aw_501_cast)[name = tensor<string, []>("op_6752_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6753_cast = softmax(axis = var_2634, x = aw_503_cast)[name = tensor<string, []>("op_6753_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6754_cast = softmax(axis = var_2634, x = aw_505_cast)[name = tensor<string, []>("op_6754_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6755_cast = softmax(axis = var_2634, x = aw_507_cast)[name = tensor<string, []>("op_6755_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6756_cast = softmax(axis = var_2634, x = aw_509_cast)[name = tensor<string, []>("op_6756_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6757_cast = softmax(axis = var_2634, x = aw_511_cast)[name = tensor<string, []>("op_6757_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6758_cast = softmax(axis = var_2634, x = aw_513_cast)[name = tensor<string, []>("op_6758_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6759_cast = softmax(axis = var_2634, x = aw_515_cast)[name = tensor<string, []>("op_6759_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6760_cast = softmax(axis = var_2634, x = aw_517_cast)[name = tensor<string, []>("op_6760_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_6761_cast = softmax(axis = var_2634, x = aw_519_cast)[name = tensor<string, []>("op_6761_cast")];
            tensor<string, []> var_6763_equation_0 = const()[name = tensor<string, []>("op_6763_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6763_cast = einsum(equation = var_6763_equation_0, values = (var_6583_cast, var_6742_cast))[name = tensor<string, []>("op_6763_cast")];
            tensor<string, []> var_6765_equation_0 = const()[name = tensor<string, []>("op_6765_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6765_cast = einsum(equation = var_6765_equation_0, values = (var_6587_cast, var_6743_cast))[name = tensor<string, []>("op_6765_cast")];
            tensor<string, []> var_6767_equation_0 = const()[name = tensor<string, []>("op_6767_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6767_cast = einsum(equation = var_6767_equation_0, values = (var_6591_cast, var_6744_cast))[name = tensor<string, []>("op_6767_cast")];
            tensor<string, []> var_6769_equation_0 = const()[name = tensor<string, []>("op_6769_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6769_cast = einsum(equation = var_6769_equation_0, values = (var_6595_cast, var_6745_cast))[name = tensor<string, []>("op_6769_cast")];
            tensor<string, []> var_6771_equation_0 = const()[name = tensor<string, []>("op_6771_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6771_cast = einsum(equation = var_6771_equation_0, values = (var_6599_cast, var_6746_cast))[name = tensor<string, []>("op_6771_cast")];
            tensor<string, []> var_6773_equation_0 = const()[name = tensor<string, []>("op_6773_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6773_cast = einsum(equation = var_6773_equation_0, values = (var_6603_cast, var_6747_cast))[name = tensor<string, []>("op_6773_cast")];
            tensor<string, []> var_6775_equation_0 = const()[name = tensor<string, []>("op_6775_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6775_cast = einsum(equation = var_6775_equation_0, values = (var_6607_cast, var_6748_cast))[name = tensor<string, []>("op_6775_cast")];
            tensor<string, []> var_6777_equation_0 = const()[name = tensor<string, []>("op_6777_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6777_cast = einsum(equation = var_6777_equation_0, values = (var_6611_cast, var_6749_cast))[name = tensor<string, []>("op_6777_cast")];
            tensor<string, []> var_6779_equation_0 = const()[name = tensor<string, []>("op_6779_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6779_cast = einsum(equation = var_6779_equation_0, values = (var_6615_cast, var_6750_cast))[name = tensor<string, []>("op_6779_cast")];
            tensor<string, []> var_6781_equation_0 = const()[name = tensor<string, []>("op_6781_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6781_cast = einsum(equation = var_6781_equation_0, values = (var_6619_cast, var_6751_cast))[name = tensor<string, []>("op_6781_cast")];
            tensor<string, []> var_6783_equation_0 = const()[name = tensor<string, []>("op_6783_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6783_cast = einsum(equation = var_6783_equation_0, values = (var_6623_cast, var_6752_cast))[name = tensor<string, []>("op_6783_cast")];
            tensor<string, []> var_6785_equation_0 = const()[name = tensor<string, []>("op_6785_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6785_cast = einsum(equation = var_6785_equation_0, values = (var_6627_cast, var_6753_cast))[name = tensor<string, []>("op_6785_cast")];
            tensor<string, []> var_6787_equation_0 = const()[name = tensor<string, []>("op_6787_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6787_cast = einsum(equation = var_6787_equation_0, values = (var_6631_cast, var_6754_cast))[name = tensor<string, []>("op_6787_cast")];
            tensor<string, []> var_6789_equation_0 = const()[name = tensor<string, []>("op_6789_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6789_cast = einsum(equation = var_6789_equation_0, values = (var_6635_cast, var_6755_cast))[name = tensor<string, []>("op_6789_cast")];
            tensor<string, []> var_6791_equation_0 = const()[name = tensor<string, []>("op_6791_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6791_cast = einsum(equation = var_6791_equation_0, values = (var_6639_cast, var_6756_cast))[name = tensor<string, []>("op_6791_cast")];
            tensor<string, []> var_6793_equation_0 = const()[name = tensor<string, []>("op_6793_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6793_cast = einsum(equation = var_6793_equation_0, values = (var_6643_cast, var_6757_cast))[name = tensor<string, []>("op_6793_cast")];
            tensor<string, []> var_6795_equation_0 = const()[name = tensor<string, []>("op_6795_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6795_cast = einsum(equation = var_6795_equation_0, values = (var_6647_cast, var_6758_cast))[name = tensor<string, []>("op_6795_cast")];
            tensor<string, []> var_6797_equation_0 = const()[name = tensor<string, []>("op_6797_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6797_cast = einsum(equation = var_6797_equation_0, values = (var_6651_cast, var_6759_cast))[name = tensor<string, []>("op_6797_cast")];
            tensor<string, []> var_6799_equation_0 = const()[name = tensor<string, []>("op_6799_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6799_cast = einsum(equation = var_6799_equation_0, values = (var_6655_cast, var_6760_cast))[name = tensor<string, []>("op_6799_cast")];
            tensor<string, []> var_6801_equation_0 = const()[name = tensor<string, []>("op_6801_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_6801_cast = einsum(equation = var_6801_equation_0, values = (var_6659_cast, var_6761_cast))[name = tensor<string, []>("op_6801_cast")];
            tensor<bool, []> input_167_interleave_0 = const()[name = tensor<string, []>("input_167_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_167_cast = concat(axis = var_2634, interleave = input_167_interleave_0, values = (var_6763_cast, var_6765_cast, var_6767_cast, var_6769_cast, var_6771_cast, var_6773_cast, var_6775_cast, var_6777_cast, var_6779_cast, var_6781_cast, var_6783_cast, var_6785_cast, var_6787_cast, var_6789_cast, var_6791_cast, var_6793_cast, var_6795_cast, var_6797_cast, var_6799_cast, var_6801_cast))[name = tensor<string, []>("input_167_cast")];
            tensor<int32, [2]> var_6807 = const()[name = tensor<string, []>("op_6807"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6809 = const()[name = tensor<string, []>("op_6809"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_6811_pad_type_0 = const()[name = tensor<string, []>("op_6811_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_6811_pad_0 = const()[name = tensor<string, []>("op_6811_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486053568)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(489330432)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_6811_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16, dilations = var_6809, groups = var_2634, pad = var_6811_pad_0, pad_type = var_6811_pad_type_0, strides = var_6807, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16, x = input_167_cast)[name = tensor<string, []>("op_6811_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_51_cast = add(x = var_6811_cast, y = inputs_49_cast)[name = tensor<string, []>("inputs_51_cast")];
            tensor<int32, [1]> var_6815 = const()[name = tensor<string, []>("op_6815"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_51_cast = reduce_mean(axes = var_6815, keep_dims = var_2629, x = inputs_51_cast)[name = tensor<string, []>("channels_mean_51_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_51_cast = sub(x = inputs_51_cast, y = channels_mean_51_cast)[name = tensor<string, []>("zero_mean_51_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_51_cast = mul(x = zero_mean_51_cast, y = zero_mean_51_cast)[name = tensor<string, []>("zero_mean_sq_51_cast")];
            tensor<int32, [1]> var_6819 = const()[name = tensor<string, []>("op_6819"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_6820_cast = reduce_mean(axes = var_6819, keep_dims = var_2629, x = zero_mean_sq_51_cast)[name = tensor<string, []>("op_6820_cast")];
            tensor<fp16, []> var_6821_to_fp16 = const()[name = tensor<string, []>("op_6821_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_6822_cast = add(x = var_6820_cast, y = var_6821_to_fp16)[name = tensor<string, []>("op_6822_cast")];
            tensor<fp16, []> denom_51_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_51_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_51_cast = rsqrt(epsilon = denom_51_epsilon_0_to_fp16, x = var_6822_cast)[name = tensor<string, []>("denom_51_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_51_cast = mul(x = zero_mean_51_cast, y = denom_51_cast)[name = tensor<string, []>("out_51_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_6826_to_fp16 = const()[name = tensor<string, []>("op_6826_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(489333056)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_6827_cast = add(x = out_51_cast, y = var_6826_to_fp16)[name = tensor<string, []>("op_6827_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_6829_to_fp16 = const()[name = tensor<string, []>("op_6829_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(489335680)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_91_cast = mul(x = var_6827_cast, y = var_6829_to_fp16)[name = tensor<string, []>("hidden_states_91_cast")];
            tensor<int32, [2]> var_6836 = const()[name = tensor<string, []>("op_6836"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6838 = const()[name = tensor<string, []>("op_6838"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_35_pad_type_0 = const()[name = tensor<string, []>("q_35_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_35_pad_0 = const()[name = tensor<string, []>("q_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(489338304)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_35_cast = conv(dilations = var_6838, groups = var_2634, pad = q_35_pad_0, pad_type = q_35_pad_type_0, strides = var_6836, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16, x = hidden_states_91_cast)[name = tensor<string, []>("q_35_cast")];
            tensor<int32, [2]> var_6842 = const()[name = tensor<string, []>("op_6842"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6844 = const()[name = tensor<string, []>("op_6844"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_69_pad_type_0 = const()[name = tensor<string, []>("k_69_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_69_pad_0 = const()[name = tensor<string, []>("k_69_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(492615168)))];
            tensor<fp16, [2, 1280, 1, 77]> k_69_cast = conv(dilations = var_6844, groups = var_2634, pad = k_69_pad_0, pad_type = k_69_pad_type_0, strides = var_6842, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_69_cast")];
            tensor<int32, [2]> var_6848 = const()[name = tensor<string, []>("op_6848"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6850 = const()[name = tensor<string, []>("op_6850"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_35_pad_type_0 = const()[name = tensor<string, []>("v_35_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_35_pad_0 = const()[name = tensor<string, []>("v_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(497858112)))];
            tensor<fp16, [2, 1280, 1, 77]> v_35_cast = conv(dilations = var_6850, groups = var_2634, pad = v_35_pad_0, pad_type = v_35_pad_type_0, strides = var_6848, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_35_cast")];
            tensor<int32, [4]> var_6854_begin_0 = const()[name = tensor<string, []>("op_6854_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6854_end_0 = const()[name = tensor<string, []>("op_6854_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_6854_end_mask_0 = const()[name = tensor<string, []>("op_6854_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6854_cast = slice_by_index(begin = var_6854_begin_0, end = var_6854_end_0, end_mask = var_6854_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6854_cast")];
            tensor<int32, [4]> var_6858_begin_0 = const()[name = tensor<string, []>("op_6858_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_6858_end_0 = const()[name = tensor<string, []>("op_6858_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_6858_end_mask_0 = const()[name = tensor<string, []>("op_6858_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6858_cast = slice_by_index(begin = var_6858_begin_0, end = var_6858_end_0, end_mask = var_6858_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6858_cast")];
            tensor<int32, [4]> var_6862_begin_0 = const()[name = tensor<string, []>("op_6862_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_6862_end_0 = const()[name = tensor<string, []>("op_6862_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_6862_end_mask_0 = const()[name = tensor<string, []>("op_6862_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6862_cast = slice_by_index(begin = var_6862_begin_0, end = var_6862_end_0, end_mask = var_6862_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6862_cast")];
            tensor<int32, [4]> var_6866_begin_0 = const()[name = tensor<string, []>("op_6866_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_6866_end_0 = const()[name = tensor<string, []>("op_6866_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_6866_end_mask_0 = const()[name = tensor<string, []>("op_6866_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6866_cast = slice_by_index(begin = var_6866_begin_0, end = var_6866_end_0, end_mask = var_6866_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6866_cast")];
            tensor<int32, [4]> var_6870_begin_0 = const()[name = tensor<string, []>("op_6870_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_6870_end_0 = const()[name = tensor<string, []>("op_6870_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_6870_end_mask_0 = const()[name = tensor<string, []>("op_6870_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6870_cast = slice_by_index(begin = var_6870_begin_0, end = var_6870_end_0, end_mask = var_6870_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6870_cast")];
            tensor<int32, [4]> var_6874_begin_0 = const()[name = tensor<string, []>("op_6874_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_6874_end_0 = const()[name = tensor<string, []>("op_6874_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_6874_end_mask_0 = const()[name = tensor<string, []>("op_6874_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6874_cast = slice_by_index(begin = var_6874_begin_0, end = var_6874_end_0, end_mask = var_6874_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6874_cast")];
            tensor<int32, [4]> var_6878_begin_0 = const()[name = tensor<string, []>("op_6878_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_6878_end_0 = const()[name = tensor<string, []>("op_6878_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_6878_end_mask_0 = const()[name = tensor<string, []>("op_6878_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6878_cast = slice_by_index(begin = var_6878_begin_0, end = var_6878_end_0, end_mask = var_6878_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6878_cast")];
            tensor<int32, [4]> var_6882_begin_0 = const()[name = tensor<string, []>("op_6882_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_6882_end_0 = const()[name = tensor<string, []>("op_6882_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_6882_end_mask_0 = const()[name = tensor<string, []>("op_6882_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6882_cast = slice_by_index(begin = var_6882_begin_0, end = var_6882_end_0, end_mask = var_6882_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6882_cast")];
            tensor<int32, [4]> var_6886_begin_0 = const()[name = tensor<string, []>("op_6886_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_6886_end_0 = const()[name = tensor<string, []>("op_6886_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_6886_end_mask_0 = const()[name = tensor<string, []>("op_6886_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6886_cast = slice_by_index(begin = var_6886_begin_0, end = var_6886_end_0, end_mask = var_6886_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6886_cast")];
            tensor<int32, [4]> var_6890_begin_0 = const()[name = tensor<string, []>("op_6890_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_6890_end_0 = const()[name = tensor<string, []>("op_6890_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_6890_end_mask_0 = const()[name = tensor<string, []>("op_6890_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6890_cast = slice_by_index(begin = var_6890_begin_0, end = var_6890_end_0, end_mask = var_6890_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6890_cast")];
            tensor<int32, [4]> var_6894_begin_0 = const()[name = tensor<string, []>("op_6894_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_6894_end_0 = const()[name = tensor<string, []>("op_6894_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_6894_end_mask_0 = const()[name = tensor<string, []>("op_6894_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6894_cast = slice_by_index(begin = var_6894_begin_0, end = var_6894_end_0, end_mask = var_6894_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6894_cast")];
            tensor<int32, [4]> var_6898_begin_0 = const()[name = tensor<string, []>("op_6898_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_6898_end_0 = const()[name = tensor<string, []>("op_6898_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_6898_end_mask_0 = const()[name = tensor<string, []>("op_6898_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6898_cast = slice_by_index(begin = var_6898_begin_0, end = var_6898_end_0, end_mask = var_6898_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6898_cast")];
            tensor<int32, [4]> var_6902_begin_0 = const()[name = tensor<string, []>("op_6902_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_6902_end_0 = const()[name = tensor<string, []>("op_6902_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_6902_end_mask_0 = const()[name = tensor<string, []>("op_6902_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6902_cast = slice_by_index(begin = var_6902_begin_0, end = var_6902_end_0, end_mask = var_6902_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6902_cast")];
            tensor<int32, [4]> var_6906_begin_0 = const()[name = tensor<string, []>("op_6906_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_6906_end_0 = const()[name = tensor<string, []>("op_6906_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_6906_end_mask_0 = const()[name = tensor<string, []>("op_6906_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6906_cast = slice_by_index(begin = var_6906_begin_0, end = var_6906_end_0, end_mask = var_6906_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6906_cast")];
            tensor<int32, [4]> var_6910_begin_0 = const()[name = tensor<string, []>("op_6910_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_6910_end_0 = const()[name = tensor<string, []>("op_6910_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_6910_end_mask_0 = const()[name = tensor<string, []>("op_6910_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6910_cast = slice_by_index(begin = var_6910_begin_0, end = var_6910_end_0, end_mask = var_6910_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6910_cast")];
            tensor<int32, [4]> var_6914_begin_0 = const()[name = tensor<string, []>("op_6914_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_6914_end_0 = const()[name = tensor<string, []>("op_6914_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_6914_end_mask_0 = const()[name = tensor<string, []>("op_6914_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6914_cast = slice_by_index(begin = var_6914_begin_0, end = var_6914_end_0, end_mask = var_6914_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6914_cast")];
            tensor<int32, [4]> var_6918_begin_0 = const()[name = tensor<string, []>("op_6918_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_6918_end_0 = const()[name = tensor<string, []>("op_6918_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_6918_end_mask_0 = const()[name = tensor<string, []>("op_6918_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6918_cast = slice_by_index(begin = var_6918_begin_0, end = var_6918_end_0, end_mask = var_6918_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6918_cast")];
            tensor<int32, [4]> var_6922_begin_0 = const()[name = tensor<string, []>("op_6922_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_6922_end_0 = const()[name = tensor<string, []>("op_6922_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_6922_end_mask_0 = const()[name = tensor<string, []>("op_6922_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6922_cast = slice_by_index(begin = var_6922_begin_0, end = var_6922_end_0, end_mask = var_6922_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6922_cast")];
            tensor<int32, [4]> var_6926_begin_0 = const()[name = tensor<string, []>("op_6926_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_6926_end_0 = const()[name = tensor<string, []>("op_6926_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_6926_end_mask_0 = const()[name = tensor<string, []>("op_6926_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6926_cast = slice_by_index(begin = var_6926_begin_0, end = var_6926_end_0, end_mask = var_6926_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6926_cast")];
            tensor<int32, [4]> var_6930_begin_0 = const()[name = tensor<string, []>("op_6930_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_6930_end_0 = const()[name = tensor<string, []>("op_6930_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_6930_end_mask_0 = const()[name = tensor<string, []>("op_6930_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_6930_cast = slice_by_index(begin = var_6930_begin_0, end = var_6930_end_0, end_mask = var_6930_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_6930_cast")];
            tensor<int32, [4]> k_71_perm_0 = const()[name = tensor<string, []>("k_71_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_6937_begin_0 = const()[name = tensor<string, []>("op_6937_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6937_end_0 = const()[name = tensor<string, []>("op_6937_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_6937_end_mask_0 = const()[name = tensor<string, []>("op_6937_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_122 = transpose(perm = k_71_perm_0, x = k_69_cast)[name = tensor<string, []>("transpose_122")];
            tensor<fp16, [2, 77, 1, 64]> var_6937_cast = slice_by_index(begin = var_6937_begin_0, end = var_6937_end_0, end_mask = var_6937_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_6937_cast")];
            tensor<int32, [4]> var_6941_begin_0 = const()[name = tensor<string, []>("op_6941_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_6941_end_0 = const()[name = tensor<string, []>("op_6941_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_6941_end_mask_0 = const()[name = tensor<string, []>("op_6941_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6941_cast = slice_by_index(begin = var_6941_begin_0, end = var_6941_end_0, end_mask = var_6941_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_6941_cast")];
            tensor<int32, [4]> var_6945_begin_0 = const()[name = tensor<string, []>("op_6945_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_6945_end_0 = const()[name = tensor<string, []>("op_6945_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_6945_end_mask_0 = const()[name = tensor<string, []>("op_6945_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6945_cast = slice_by_index(begin = var_6945_begin_0, end = var_6945_end_0, end_mask = var_6945_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_6945_cast")];
            tensor<int32, [4]> var_6949_begin_0 = const()[name = tensor<string, []>("op_6949_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_6949_end_0 = const()[name = tensor<string, []>("op_6949_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_6949_end_mask_0 = const()[name = tensor<string, []>("op_6949_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6949_cast = slice_by_index(begin = var_6949_begin_0, end = var_6949_end_0, end_mask = var_6949_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_6949_cast")];
            tensor<int32, [4]> var_6953_begin_0 = const()[name = tensor<string, []>("op_6953_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_6953_end_0 = const()[name = tensor<string, []>("op_6953_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_6953_end_mask_0 = const()[name = tensor<string, []>("op_6953_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6953_cast = slice_by_index(begin = var_6953_begin_0, end = var_6953_end_0, end_mask = var_6953_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_6953_cast")];
            tensor<int32, [4]> var_6957_begin_0 = const()[name = tensor<string, []>("op_6957_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_6957_end_0 = const()[name = tensor<string, []>("op_6957_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_6957_end_mask_0 = const()[name = tensor<string, []>("op_6957_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6957_cast = slice_by_index(begin = var_6957_begin_0, end = var_6957_end_0, end_mask = var_6957_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_6957_cast")];
            tensor<int32, [4]> var_6961_begin_0 = const()[name = tensor<string, []>("op_6961_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_6961_end_0 = const()[name = tensor<string, []>("op_6961_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_6961_end_mask_0 = const()[name = tensor<string, []>("op_6961_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6961_cast = slice_by_index(begin = var_6961_begin_0, end = var_6961_end_0, end_mask = var_6961_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_6961_cast")];
            tensor<int32, [4]> var_6965_begin_0 = const()[name = tensor<string, []>("op_6965_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_6965_end_0 = const()[name = tensor<string, []>("op_6965_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_6965_end_mask_0 = const()[name = tensor<string, []>("op_6965_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6965_cast = slice_by_index(begin = var_6965_begin_0, end = var_6965_end_0, end_mask = var_6965_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_6965_cast")];
            tensor<int32, [4]> var_6969_begin_0 = const()[name = tensor<string, []>("op_6969_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_6969_end_0 = const()[name = tensor<string, []>("op_6969_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_6969_end_mask_0 = const()[name = tensor<string, []>("op_6969_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6969_cast = slice_by_index(begin = var_6969_begin_0, end = var_6969_end_0, end_mask = var_6969_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_6969_cast")];
            tensor<int32, [4]> var_6973_begin_0 = const()[name = tensor<string, []>("op_6973_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_6973_end_0 = const()[name = tensor<string, []>("op_6973_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_6973_end_mask_0 = const()[name = tensor<string, []>("op_6973_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6973_cast = slice_by_index(begin = var_6973_begin_0, end = var_6973_end_0, end_mask = var_6973_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_6973_cast")];
            tensor<int32, [4]> var_6977_begin_0 = const()[name = tensor<string, []>("op_6977_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_6977_end_0 = const()[name = tensor<string, []>("op_6977_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_6977_end_mask_0 = const()[name = tensor<string, []>("op_6977_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6977_cast = slice_by_index(begin = var_6977_begin_0, end = var_6977_end_0, end_mask = var_6977_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_6977_cast")];
            tensor<int32, [4]> var_6981_begin_0 = const()[name = tensor<string, []>("op_6981_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_6981_end_0 = const()[name = tensor<string, []>("op_6981_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_6981_end_mask_0 = const()[name = tensor<string, []>("op_6981_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6981_cast = slice_by_index(begin = var_6981_begin_0, end = var_6981_end_0, end_mask = var_6981_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_6981_cast")];
            tensor<int32, [4]> var_6985_begin_0 = const()[name = tensor<string, []>("op_6985_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_6985_end_0 = const()[name = tensor<string, []>("op_6985_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_6985_end_mask_0 = const()[name = tensor<string, []>("op_6985_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6985_cast = slice_by_index(begin = var_6985_begin_0, end = var_6985_end_0, end_mask = var_6985_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_6985_cast")];
            tensor<int32, [4]> var_6989_begin_0 = const()[name = tensor<string, []>("op_6989_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_6989_end_0 = const()[name = tensor<string, []>("op_6989_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_6989_end_mask_0 = const()[name = tensor<string, []>("op_6989_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6989_cast = slice_by_index(begin = var_6989_begin_0, end = var_6989_end_0, end_mask = var_6989_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_6989_cast")];
            tensor<int32, [4]> var_6993_begin_0 = const()[name = tensor<string, []>("op_6993_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_6993_end_0 = const()[name = tensor<string, []>("op_6993_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_6993_end_mask_0 = const()[name = tensor<string, []>("op_6993_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6993_cast = slice_by_index(begin = var_6993_begin_0, end = var_6993_end_0, end_mask = var_6993_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_6993_cast")];
            tensor<int32, [4]> var_6997_begin_0 = const()[name = tensor<string, []>("op_6997_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_6997_end_0 = const()[name = tensor<string, []>("op_6997_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_6997_end_mask_0 = const()[name = tensor<string, []>("op_6997_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6997_cast = slice_by_index(begin = var_6997_begin_0, end = var_6997_end_0, end_mask = var_6997_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_6997_cast")];
            tensor<int32, [4]> var_7001_begin_0 = const()[name = tensor<string, []>("op_7001_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_7001_end_0 = const()[name = tensor<string, []>("op_7001_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_7001_end_mask_0 = const()[name = tensor<string, []>("op_7001_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7001_cast = slice_by_index(begin = var_7001_begin_0, end = var_7001_end_0, end_mask = var_7001_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_7001_cast")];
            tensor<int32, [4]> var_7005_begin_0 = const()[name = tensor<string, []>("op_7005_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_7005_end_0 = const()[name = tensor<string, []>("op_7005_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_7005_end_mask_0 = const()[name = tensor<string, []>("op_7005_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7005_cast = slice_by_index(begin = var_7005_begin_0, end = var_7005_end_0, end_mask = var_7005_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_7005_cast")];
            tensor<int32, [4]> var_7009_begin_0 = const()[name = tensor<string, []>("op_7009_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_7009_end_0 = const()[name = tensor<string, []>("op_7009_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_7009_end_mask_0 = const()[name = tensor<string, []>("op_7009_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7009_cast = slice_by_index(begin = var_7009_begin_0, end = var_7009_end_0, end_mask = var_7009_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_7009_cast")];
            tensor<int32, [4]> var_7013_begin_0 = const()[name = tensor<string, []>("op_7013_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_7013_end_0 = const()[name = tensor<string, []>("op_7013_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_7013_end_mask_0 = const()[name = tensor<string, []>("op_7013_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7013_cast = slice_by_index(begin = var_7013_begin_0, end = var_7013_end_0, end_mask = var_7013_end_mask_0, x = transpose_122)[name = tensor<string, []>("op_7013_cast")];
            tensor<int32, [4]> var_7015_begin_0 = const()[name = tensor<string, []>("op_7015_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7015_end_0 = const()[name = tensor<string, []>("op_7015_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_7015_end_mask_0 = const()[name = tensor<string, []>("op_7015_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7015_cast = slice_by_index(begin = var_7015_begin_0, end = var_7015_end_0, end_mask = var_7015_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7015_cast")];
            tensor<int32, [4]> var_7019_begin_0 = const()[name = tensor<string, []>("op_7019_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_7019_end_0 = const()[name = tensor<string, []>("op_7019_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_7019_end_mask_0 = const()[name = tensor<string, []>("op_7019_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7019_cast = slice_by_index(begin = var_7019_begin_0, end = var_7019_end_0, end_mask = var_7019_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7019_cast")];
            tensor<int32, [4]> var_7023_begin_0 = const()[name = tensor<string, []>("op_7023_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_7023_end_0 = const()[name = tensor<string, []>("op_7023_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_7023_end_mask_0 = const()[name = tensor<string, []>("op_7023_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7023_cast = slice_by_index(begin = var_7023_begin_0, end = var_7023_end_0, end_mask = var_7023_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7023_cast")];
            tensor<int32, [4]> var_7027_begin_0 = const()[name = tensor<string, []>("op_7027_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_7027_end_0 = const()[name = tensor<string, []>("op_7027_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_7027_end_mask_0 = const()[name = tensor<string, []>("op_7027_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7027_cast = slice_by_index(begin = var_7027_begin_0, end = var_7027_end_0, end_mask = var_7027_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7027_cast")];
            tensor<int32, [4]> var_7031_begin_0 = const()[name = tensor<string, []>("op_7031_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_7031_end_0 = const()[name = tensor<string, []>("op_7031_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_7031_end_mask_0 = const()[name = tensor<string, []>("op_7031_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7031_cast = slice_by_index(begin = var_7031_begin_0, end = var_7031_end_0, end_mask = var_7031_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7031_cast")];
            tensor<int32, [4]> var_7035_begin_0 = const()[name = tensor<string, []>("op_7035_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7035_end_0 = const()[name = tensor<string, []>("op_7035_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_7035_end_mask_0 = const()[name = tensor<string, []>("op_7035_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7035_cast = slice_by_index(begin = var_7035_begin_0, end = var_7035_end_0, end_mask = var_7035_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7035_cast")];
            tensor<int32, [4]> var_7039_begin_0 = const()[name = tensor<string, []>("op_7039_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_7039_end_0 = const()[name = tensor<string, []>("op_7039_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_7039_end_mask_0 = const()[name = tensor<string, []>("op_7039_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7039_cast = slice_by_index(begin = var_7039_begin_0, end = var_7039_end_0, end_mask = var_7039_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7039_cast")];
            tensor<int32, [4]> var_7043_begin_0 = const()[name = tensor<string, []>("op_7043_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_7043_end_0 = const()[name = tensor<string, []>("op_7043_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_7043_end_mask_0 = const()[name = tensor<string, []>("op_7043_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7043_cast = slice_by_index(begin = var_7043_begin_0, end = var_7043_end_0, end_mask = var_7043_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7043_cast")];
            tensor<int32, [4]> var_7047_begin_0 = const()[name = tensor<string, []>("op_7047_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_7047_end_0 = const()[name = tensor<string, []>("op_7047_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_7047_end_mask_0 = const()[name = tensor<string, []>("op_7047_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7047_cast = slice_by_index(begin = var_7047_begin_0, end = var_7047_end_0, end_mask = var_7047_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7047_cast")];
            tensor<int32, [4]> var_7051_begin_0 = const()[name = tensor<string, []>("op_7051_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_7051_end_0 = const()[name = tensor<string, []>("op_7051_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_7051_end_mask_0 = const()[name = tensor<string, []>("op_7051_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7051_cast = slice_by_index(begin = var_7051_begin_0, end = var_7051_end_0, end_mask = var_7051_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7051_cast")];
            tensor<int32, [4]> var_7055_begin_0 = const()[name = tensor<string, []>("op_7055_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7055_end_0 = const()[name = tensor<string, []>("op_7055_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_7055_end_mask_0 = const()[name = tensor<string, []>("op_7055_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7055_cast = slice_by_index(begin = var_7055_begin_0, end = var_7055_end_0, end_mask = var_7055_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7055_cast")];
            tensor<int32, [4]> var_7059_begin_0 = const()[name = tensor<string, []>("op_7059_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_7059_end_0 = const()[name = tensor<string, []>("op_7059_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_7059_end_mask_0 = const()[name = tensor<string, []>("op_7059_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7059_cast = slice_by_index(begin = var_7059_begin_0, end = var_7059_end_0, end_mask = var_7059_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7059_cast")];
            tensor<int32, [4]> var_7063_begin_0 = const()[name = tensor<string, []>("op_7063_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_7063_end_0 = const()[name = tensor<string, []>("op_7063_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_7063_end_mask_0 = const()[name = tensor<string, []>("op_7063_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7063_cast = slice_by_index(begin = var_7063_begin_0, end = var_7063_end_0, end_mask = var_7063_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7063_cast")];
            tensor<int32, [4]> var_7067_begin_0 = const()[name = tensor<string, []>("op_7067_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_7067_end_0 = const()[name = tensor<string, []>("op_7067_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_7067_end_mask_0 = const()[name = tensor<string, []>("op_7067_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7067_cast = slice_by_index(begin = var_7067_begin_0, end = var_7067_end_0, end_mask = var_7067_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7067_cast")];
            tensor<int32, [4]> var_7071_begin_0 = const()[name = tensor<string, []>("op_7071_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_7071_end_0 = const()[name = tensor<string, []>("op_7071_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_7071_end_mask_0 = const()[name = tensor<string, []>("op_7071_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7071_cast = slice_by_index(begin = var_7071_begin_0, end = var_7071_end_0, end_mask = var_7071_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7071_cast")];
            tensor<int32, [4]> var_7075_begin_0 = const()[name = tensor<string, []>("op_7075_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_7075_end_0 = const()[name = tensor<string, []>("op_7075_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_7075_end_mask_0 = const()[name = tensor<string, []>("op_7075_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7075_cast = slice_by_index(begin = var_7075_begin_0, end = var_7075_end_0, end_mask = var_7075_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7075_cast")];
            tensor<int32, [4]> var_7079_begin_0 = const()[name = tensor<string, []>("op_7079_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_7079_end_0 = const()[name = tensor<string, []>("op_7079_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_7079_end_mask_0 = const()[name = tensor<string, []>("op_7079_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7079_cast = slice_by_index(begin = var_7079_begin_0, end = var_7079_end_0, end_mask = var_7079_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7079_cast")];
            tensor<int32, [4]> var_7083_begin_0 = const()[name = tensor<string, []>("op_7083_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_7083_end_0 = const()[name = tensor<string, []>("op_7083_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_7083_end_mask_0 = const()[name = tensor<string, []>("op_7083_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7083_cast = slice_by_index(begin = var_7083_begin_0, end = var_7083_end_0, end_mask = var_7083_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7083_cast")];
            tensor<int32, [4]> var_7087_begin_0 = const()[name = tensor<string, []>("op_7087_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_7087_end_0 = const()[name = tensor<string, []>("op_7087_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_7087_end_mask_0 = const()[name = tensor<string, []>("op_7087_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7087_cast = slice_by_index(begin = var_7087_begin_0, end = var_7087_end_0, end_mask = var_7087_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7087_cast")];
            tensor<int32, [4]> var_7091_begin_0 = const()[name = tensor<string, []>("op_7091_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_7091_end_0 = const()[name = tensor<string, []>("op_7091_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_7091_end_mask_0 = const()[name = tensor<string, []>("op_7091_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7091_cast = slice_by_index(begin = var_7091_begin_0, end = var_7091_end_0, end_mask = var_7091_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7091_cast")];
            tensor<string, []> var_7095_equation_0 = const()[name = tensor<string, []>("op_7095_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7095_cast = einsum(equation = var_7095_equation_0, values = (var_6937_cast, var_6854_cast))[name = tensor<string, []>("op_7095_cast")];
            tensor<fp16, []> var_7096_to_fp16 = const()[name = tensor<string, []>("op_7096_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_521_cast = mul(x = var_7095_cast, y = var_7096_to_fp16)[name = tensor<string, []>("aw_521_cast")];
            tensor<string, []> var_7099_equation_0 = const()[name = tensor<string, []>("op_7099_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7099_cast = einsum(equation = var_7099_equation_0, values = (var_6941_cast, var_6858_cast))[name = tensor<string, []>("op_7099_cast")];
            tensor<fp16, []> var_7100_to_fp16 = const()[name = tensor<string, []>("op_7100_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_523_cast = mul(x = var_7099_cast, y = var_7100_to_fp16)[name = tensor<string, []>("aw_523_cast")];
            tensor<string, []> var_7103_equation_0 = const()[name = tensor<string, []>("op_7103_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7103_cast = einsum(equation = var_7103_equation_0, values = (var_6945_cast, var_6862_cast))[name = tensor<string, []>("op_7103_cast")];
            tensor<fp16, []> var_7104_to_fp16 = const()[name = tensor<string, []>("op_7104_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_525_cast = mul(x = var_7103_cast, y = var_7104_to_fp16)[name = tensor<string, []>("aw_525_cast")];
            tensor<string, []> var_7107_equation_0 = const()[name = tensor<string, []>("op_7107_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7107_cast = einsum(equation = var_7107_equation_0, values = (var_6949_cast, var_6866_cast))[name = tensor<string, []>("op_7107_cast")];
            tensor<fp16, []> var_7108_to_fp16 = const()[name = tensor<string, []>("op_7108_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_527_cast = mul(x = var_7107_cast, y = var_7108_to_fp16)[name = tensor<string, []>("aw_527_cast")];
            tensor<string, []> var_7111_equation_0 = const()[name = tensor<string, []>("op_7111_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7111_cast = einsum(equation = var_7111_equation_0, values = (var_6953_cast, var_6870_cast))[name = tensor<string, []>("op_7111_cast")];
            tensor<fp16, []> var_7112_to_fp16 = const()[name = tensor<string, []>("op_7112_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_529_cast = mul(x = var_7111_cast, y = var_7112_to_fp16)[name = tensor<string, []>("aw_529_cast")];
            tensor<string, []> var_7115_equation_0 = const()[name = tensor<string, []>("op_7115_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7115_cast = einsum(equation = var_7115_equation_0, values = (var_6957_cast, var_6874_cast))[name = tensor<string, []>("op_7115_cast")];
            tensor<fp16, []> var_7116_to_fp16 = const()[name = tensor<string, []>("op_7116_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_531_cast = mul(x = var_7115_cast, y = var_7116_to_fp16)[name = tensor<string, []>("aw_531_cast")];
            tensor<string, []> var_7119_equation_0 = const()[name = tensor<string, []>("op_7119_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7119_cast = einsum(equation = var_7119_equation_0, values = (var_6961_cast, var_6878_cast))[name = tensor<string, []>("op_7119_cast")];
            tensor<fp16, []> var_7120_to_fp16 = const()[name = tensor<string, []>("op_7120_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_533_cast = mul(x = var_7119_cast, y = var_7120_to_fp16)[name = tensor<string, []>("aw_533_cast")];
            tensor<string, []> var_7123_equation_0 = const()[name = tensor<string, []>("op_7123_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7123_cast = einsum(equation = var_7123_equation_0, values = (var_6965_cast, var_6882_cast))[name = tensor<string, []>("op_7123_cast")];
            tensor<fp16, []> var_7124_to_fp16 = const()[name = tensor<string, []>("op_7124_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_535_cast = mul(x = var_7123_cast, y = var_7124_to_fp16)[name = tensor<string, []>("aw_535_cast")];
            tensor<string, []> var_7127_equation_0 = const()[name = tensor<string, []>("op_7127_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7127_cast = einsum(equation = var_7127_equation_0, values = (var_6969_cast, var_6886_cast))[name = tensor<string, []>("op_7127_cast")];
            tensor<fp16, []> var_7128_to_fp16 = const()[name = tensor<string, []>("op_7128_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_537_cast = mul(x = var_7127_cast, y = var_7128_to_fp16)[name = tensor<string, []>("aw_537_cast")];
            tensor<string, []> var_7131_equation_0 = const()[name = tensor<string, []>("op_7131_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7131_cast = einsum(equation = var_7131_equation_0, values = (var_6973_cast, var_6890_cast))[name = tensor<string, []>("op_7131_cast")];
            tensor<fp16, []> var_7132_to_fp16 = const()[name = tensor<string, []>("op_7132_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_539_cast = mul(x = var_7131_cast, y = var_7132_to_fp16)[name = tensor<string, []>("aw_539_cast")];
            tensor<string, []> var_7135_equation_0 = const()[name = tensor<string, []>("op_7135_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7135_cast = einsum(equation = var_7135_equation_0, values = (var_6977_cast, var_6894_cast))[name = tensor<string, []>("op_7135_cast")];
            tensor<fp16, []> var_7136_to_fp16 = const()[name = tensor<string, []>("op_7136_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_541_cast = mul(x = var_7135_cast, y = var_7136_to_fp16)[name = tensor<string, []>("aw_541_cast")];
            tensor<string, []> var_7139_equation_0 = const()[name = tensor<string, []>("op_7139_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7139_cast = einsum(equation = var_7139_equation_0, values = (var_6981_cast, var_6898_cast))[name = tensor<string, []>("op_7139_cast")];
            tensor<fp16, []> var_7140_to_fp16 = const()[name = tensor<string, []>("op_7140_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_543_cast = mul(x = var_7139_cast, y = var_7140_to_fp16)[name = tensor<string, []>("aw_543_cast")];
            tensor<string, []> var_7143_equation_0 = const()[name = tensor<string, []>("op_7143_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7143_cast = einsum(equation = var_7143_equation_0, values = (var_6985_cast, var_6902_cast))[name = tensor<string, []>("op_7143_cast")];
            tensor<fp16, []> var_7144_to_fp16 = const()[name = tensor<string, []>("op_7144_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_545_cast = mul(x = var_7143_cast, y = var_7144_to_fp16)[name = tensor<string, []>("aw_545_cast")];
            tensor<string, []> var_7147_equation_0 = const()[name = tensor<string, []>("op_7147_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7147_cast = einsum(equation = var_7147_equation_0, values = (var_6989_cast, var_6906_cast))[name = tensor<string, []>("op_7147_cast")];
            tensor<fp16, []> var_7148_to_fp16 = const()[name = tensor<string, []>("op_7148_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_547_cast = mul(x = var_7147_cast, y = var_7148_to_fp16)[name = tensor<string, []>("aw_547_cast")];
            tensor<string, []> var_7151_equation_0 = const()[name = tensor<string, []>("op_7151_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7151_cast = einsum(equation = var_7151_equation_0, values = (var_6993_cast, var_6910_cast))[name = tensor<string, []>("op_7151_cast")];
            tensor<fp16, []> var_7152_to_fp16 = const()[name = tensor<string, []>("op_7152_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_549_cast = mul(x = var_7151_cast, y = var_7152_to_fp16)[name = tensor<string, []>("aw_549_cast")];
            tensor<string, []> var_7155_equation_0 = const()[name = tensor<string, []>("op_7155_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7155_cast = einsum(equation = var_7155_equation_0, values = (var_6997_cast, var_6914_cast))[name = tensor<string, []>("op_7155_cast")];
            tensor<fp16, []> var_7156_to_fp16 = const()[name = tensor<string, []>("op_7156_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_551_cast = mul(x = var_7155_cast, y = var_7156_to_fp16)[name = tensor<string, []>("aw_551_cast")];
            tensor<string, []> var_7159_equation_0 = const()[name = tensor<string, []>("op_7159_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7159_cast = einsum(equation = var_7159_equation_0, values = (var_7001_cast, var_6918_cast))[name = tensor<string, []>("op_7159_cast")];
            tensor<fp16, []> var_7160_to_fp16 = const()[name = tensor<string, []>("op_7160_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_553_cast = mul(x = var_7159_cast, y = var_7160_to_fp16)[name = tensor<string, []>("aw_553_cast")];
            tensor<string, []> var_7163_equation_0 = const()[name = tensor<string, []>("op_7163_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7163_cast = einsum(equation = var_7163_equation_0, values = (var_7005_cast, var_6922_cast))[name = tensor<string, []>("op_7163_cast")];
            tensor<fp16, []> var_7164_to_fp16 = const()[name = tensor<string, []>("op_7164_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_555_cast = mul(x = var_7163_cast, y = var_7164_to_fp16)[name = tensor<string, []>("aw_555_cast")];
            tensor<string, []> var_7167_equation_0 = const()[name = tensor<string, []>("op_7167_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7167_cast = einsum(equation = var_7167_equation_0, values = (var_7009_cast, var_6926_cast))[name = tensor<string, []>("op_7167_cast")];
            tensor<fp16, []> var_7168_to_fp16 = const()[name = tensor<string, []>("op_7168_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_557_cast = mul(x = var_7167_cast, y = var_7168_to_fp16)[name = tensor<string, []>("aw_557_cast")];
            tensor<string, []> var_7171_equation_0 = const()[name = tensor<string, []>("op_7171_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_7171_cast = einsum(equation = var_7171_equation_0, values = (var_7013_cast, var_6930_cast))[name = tensor<string, []>("op_7171_cast")];
            tensor<fp16, []> var_7172_to_fp16 = const()[name = tensor<string, []>("op_7172_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_559_cast = mul(x = var_7171_cast, y = var_7172_to_fp16)[name = tensor<string, []>("aw_559_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7174_cast = softmax(axis = var_2634, x = aw_521_cast)[name = tensor<string, []>("op_7174_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7175_cast = softmax(axis = var_2634, x = aw_523_cast)[name = tensor<string, []>("op_7175_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7176_cast = softmax(axis = var_2634, x = aw_525_cast)[name = tensor<string, []>("op_7176_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7177_cast = softmax(axis = var_2634, x = aw_527_cast)[name = tensor<string, []>("op_7177_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7178_cast = softmax(axis = var_2634, x = aw_529_cast)[name = tensor<string, []>("op_7178_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7179_cast = softmax(axis = var_2634, x = aw_531_cast)[name = tensor<string, []>("op_7179_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7180_cast = softmax(axis = var_2634, x = aw_533_cast)[name = tensor<string, []>("op_7180_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7181_cast = softmax(axis = var_2634, x = aw_535_cast)[name = tensor<string, []>("op_7181_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7182_cast = softmax(axis = var_2634, x = aw_537_cast)[name = tensor<string, []>("op_7182_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7183_cast = softmax(axis = var_2634, x = aw_539_cast)[name = tensor<string, []>("op_7183_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7184_cast = softmax(axis = var_2634, x = aw_541_cast)[name = tensor<string, []>("op_7184_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7185_cast = softmax(axis = var_2634, x = aw_543_cast)[name = tensor<string, []>("op_7185_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7186_cast = softmax(axis = var_2634, x = aw_545_cast)[name = tensor<string, []>("op_7186_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7187_cast = softmax(axis = var_2634, x = aw_547_cast)[name = tensor<string, []>("op_7187_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7188_cast = softmax(axis = var_2634, x = aw_549_cast)[name = tensor<string, []>("op_7188_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7189_cast = softmax(axis = var_2634, x = aw_551_cast)[name = tensor<string, []>("op_7189_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7190_cast = softmax(axis = var_2634, x = aw_553_cast)[name = tensor<string, []>("op_7190_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7191_cast = softmax(axis = var_2634, x = aw_555_cast)[name = tensor<string, []>("op_7191_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7192_cast = softmax(axis = var_2634, x = aw_557_cast)[name = tensor<string, []>("op_7192_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_7193_cast = softmax(axis = var_2634, x = aw_559_cast)[name = tensor<string, []>("op_7193_cast")];
            tensor<string, []> var_7195_equation_0 = const()[name = tensor<string, []>("op_7195_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7195_cast = einsum(equation = var_7195_equation_0, values = (var_7015_cast, var_7174_cast))[name = tensor<string, []>("op_7195_cast")];
            tensor<string, []> var_7197_equation_0 = const()[name = tensor<string, []>("op_7197_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7197_cast = einsum(equation = var_7197_equation_0, values = (var_7019_cast, var_7175_cast))[name = tensor<string, []>("op_7197_cast")];
            tensor<string, []> var_7199_equation_0 = const()[name = tensor<string, []>("op_7199_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7199_cast = einsum(equation = var_7199_equation_0, values = (var_7023_cast, var_7176_cast))[name = tensor<string, []>("op_7199_cast")];
            tensor<string, []> var_7201_equation_0 = const()[name = tensor<string, []>("op_7201_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7201_cast = einsum(equation = var_7201_equation_0, values = (var_7027_cast, var_7177_cast))[name = tensor<string, []>("op_7201_cast")];
            tensor<string, []> var_7203_equation_0 = const()[name = tensor<string, []>("op_7203_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7203_cast = einsum(equation = var_7203_equation_0, values = (var_7031_cast, var_7178_cast))[name = tensor<string, []>("op_7203_cast")];
            tensor<string, []> var_7205_equation_0 = const()[name = tensor<string, []>("op_7205_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7205_cast = einsum(equation = var_7205_equation_0, values = (var_7035_cast, var_7179_cast))[name = tensor<string, []>("op_7205_cast")];
            tensor<string, []> var_7207_equation_0 = const()[name = tensor<string, []>("op_7207_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7207_cast = einsum(equation = var_7207_equation_0, values = (var_7039_cast, var_7180_cast))[name = tensor<string, []>("op_7207_cast")];
            tensor<string, []> var_7209_equation_0 = const()[name = tensor<string, []>("op_7209_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7209_cast = einsum(equation = var_7209_equation_0, values = (var_7043_cast, var_7181_cast))[name = tensor<string, []>("op_7209_cast")];
            tensor<string, []> var_7211_equation_0 = const()[name = tensor<string, []>("op_7211_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7211_cast = einsum(equation = var_7211_equation_0, values = (var_7047_cast, var_7182_cast))[name = tensor<string, []>("op_7211_cast")];
            tensor<string, []> var_7213_equation_0 = const()[name = tensor<string, []>("op_7213_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7213_cast = einsum(equation = var_7213_equation_0, values = (var_7051_cast, var_7183_cast))[name = tensor<string, []>("op_7213_cast")];
            tensor<string, []> var_7215_equation_0 = const()[name = tensor<string, []>("op_7215_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7215_cast = einsum(equation = var_7215_equation_0, values = (var_7055_cast, var_7184_cast))[name = tensor<string, []>("op_7215_cast")];
            tensor<string, []> var_7217_equation_0 = const()[name = tensor<string, []>("op_7217_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7217_cast = einsum(equation = var_7217_equation_0, values = (var_7059_cast, var_7185_cast))[name = tensor<string, []>("op_7217_cast")];
            tensor<string, []> var_7219_equation_0 = const()[name = tensor<string, []>("op_7219_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7219_cast = einsum(equation = var_7219_equation_0, values = (var_7063_cast, var_7186_cast))[name = tensor<string, []>("op_7219_cast")];
            tensor<string, []> var_7221_equation_0 = const()[name = tensor<string, []>("op_7221_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7221_cast = einsum(equation = var_7221_equation_0, values = (var_7067_cast, var_7187_cast))[name = tensor<string, []>("op_7221_cast")];
            tensor<string, []> var_7223_equation_0 = const()[name = tensor<string, []>("op_7223_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7223_cast = einsum(equation = var_7223_equation_0, values = (var_7071_cast, var_7188_cast))[name = tensor<string, []>("op_7223_cast")];
            tensor<string, []> var_7225_equation_0 = const()[name = tensor<string, []>("op_7225_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7225_cast = einsum(equation = var_7225_equation_0, values = (var_7075_cast, var_7189_cast))[name = tensor<string, []>("op_7225_cast")];
            tensor<string, []> var_7227_equation_0 = const()[name = tensor<string, []>("op_7227_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7227_cast = einsum(equation = var_7227_equation_0, values = (var_7079_cast, var_7190_cast))[name = tensor<string, []>("op_7227_cast")];
            tensor<string, []> var_7229_equation_0 = const()[name = tensor<string, []>("op_7229_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7229_cast = einsum(equation = var_7229_equation_0, values = (var_7083_cast, var_7191_cast))[name = tensor<string, []>("op_7229_cast")];
            tensor<string, []> var_7231_equation_0 = const()[name = tensor<string, []>("op_7231_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7231_cast = einsum(equation = var_7231_equation_0, values = (var_7087_cast, var_7192_cast))[name = tensor<string, []>("op_7231_cast")];
            tensor<string, []> var_7233_equation_0 = const()[name = tensor<string, []>("op_7233_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7233_cast = einsum(equation = var_7233_equation_0, values = (var_7091_cast, var_7193_cast))[name = tensor<string, []>("op_7233_cast")];
            tensor<bool, []> input_169_interleave_0 = const()[name = tensor<string, []>("input_169_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_169_cast = concat(axis = var_2634, interleave = input_169_interleave_0, values = (var_7195_cast, var_7197_cast, var_7199_cast, var_7201_cast, var_7203_cast, var_7205_cast, var_7207_cast, var_7209_cast, var_7211_cast, var_7213_cast, var_7215_cast, var_7217_cast, var_7219_cast, var_7221_cast, var_7223_cast, var_7225_cast, var_7227_cast, var_7229_cast, var_7231_cast, var_7233_cast))[name = tensor<string, []>("input_169_cast")];
            tensor<int32, [2]> var_7239 = const()[name = tensor<string, []>("op_7239"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7241 = const()[name = tensor<string, []>("op_7241"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_7243_pad_type_0 = const()[name = tensor<string, []>("op_7243_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_7243_pad_0 = const()[name = tensor<string, []>("op_7243_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(503101056)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(506377920)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_7243_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16, dilations = var_7241, groups = var_2634, pad = var_7243_pad_0, pad_type = var_7243_pad_type_0, strides = var_7239, weight = down_blocks_2_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16, x = input_169_cast)[name = tensor<string, []>("op_7243_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_53_cast = add(x = var_7243_cast, y = inputs_51_cast)[name = tensor<string, []>("inputs_53_cast")];
            tensor<int32, [1]> var_7247 = const()[name = tensor<string, []>("op_7247"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_53_cast = reduce_mean(axes = var_7247, keep_dims = var_2629, x = inputs_53_cast)[name = tensor<string, []>("channels_mean_53_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_53_cast = sub(x = inputs_53_cast, y = channels_mean_53_cast)[name = tensor<string, []>("zero_mean_53_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_53_cast = mul(x = zero_mean_53_cast, y = zero_mean_53_cast)[name = tensor<string, []>("zero_mean_sq_53_cast")];
            tensor<int32, [1]> var_7251 = const()[name = tensor<string, []>("op_7251"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_7252_cast = reduce_mean(axes = var_7251, keep_dims = var_2629, x = zero_mean_sq_53_cast)[name = tensor<string, []>("op_7252_cast")];
            tensor<fp16, []> var_7253_to_fp16 = const()[name = tensor<string, []>("op_7253_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_7254_cast = add(x = var_7252_cast, y = var_7253_to_fp16)[name = tensor<string, []>("op_7254_cast")];
            tensor<fp16, []> denom_53_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_53_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_53_cast = rsqrt(epsilon = denom_53_epsilon_0_to_fp16, x = var_7254_cast)[name = tensor<string, []>("denom_53_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_53_cast = mul(x = zero_mean_53_cast, y = denom_53_cast)[name = tensor<string, []>("out_53_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7258_to_fp16 = const()[name = tensor<string, []>("op_7258_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(506380544)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_7259_cast = add(x = out_53_cast, y = var_7258_to_fp16)[name = tensor<string, []>("op_7259_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7261_to_fp16 = const()[name = tensor<string, []>("op_7261_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(506383168)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_171_cast = mul(x = var_7259_cast, y = var_7261_to_fp16)[name = tensor<string, []>("input_171_cast")];
            tensor<int32, [2]> var_7269 = const()[name = tensor<string, []>("op_7269"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7271 = const()[name = tensor<string, []>("op_7271"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_7273_pad_type_0 = const()[name = tensor<string, []>("op_7273_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_7273_pad_0 = const()[name = tensor<string, []>("op_7273_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(506385792)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(532600256)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_7273_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16, dilations = var_7271, groups = var_2634, pad = var_7273_pad_0, pad_type = var_7273_pad_type_0, strides = var_7269, weight = down_blocks_2_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16, x = input_171_cast)[name = tensor<string, []>("op_7273_cast")];
            tensor<int32, [2]> var_7274_split_sizes_0 = const()[name = tensor<string, []>("op_7274_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_7274_axis_0 = const()[name = tensor<string, []>("op_7274_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_7274_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_7274_cast_1 = split(axis = var_7274_axis_0, split_sizes = var_7274_split_sizes_0, x = var_7273_cast)[name = tensor<string, []>("op_7274_cast")];
            tensor<string, []> var_7276_mode_0 = const()[name = tensor<string, []>("op_7276_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_7276_cast = gelu(mode = var_7276_mode_0, x = var_7274_cast_1)[name = tensor<string, []>("op_7276_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_173_cast = mul(x = var_7274_cast_0, y = var_7276_cast)[name = tensor<string, []>("input_173_cast")];
            tensor<int32, [2]> var_7280 = const()[name = tensor<string, []>("op_7280"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7282 = const()[name = tensor<string, []>("op_7282"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_7284_pad_type_0 = const()[name = tensor<string, []>("op_7284_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_7284_pad_0 = const()[name = tensor<string, []>("op_7284_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(532620800)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545728064)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_7284_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16, dilations = var_7282, groups = var_2634, pad = var_7284_pad_0, pad_type = var_7284_pad_type_0, strides = var_7280, weight = down_blocks_2_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16, x = input_173_cast)[name = tensor<string, []>("op_7284_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_55_cast = add(x = var_7284_cast, y = inputs_53_cast)[name = tensor<string, []>("inputs_55_cast")];
            tensor<int32, [1]> var_7294 = const()[name = tensor<string, []>("op_7294"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_55_cast = reduce_mean(axes = var_7294, keep_dims = var_2629, x = inputs_55_cast)[name = tensor<string, []>("channels_mean_55_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_55_cast = sub(x = inputs_55_cast, y = channels_mean_55_cast)[name = tensor<string, []>("zero_mean_55_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_55_cast = mul(x = zero_mean_55_cast, y = zero_mean_55_cast)[name = tensor<string, []>("zero_mean_sq_55_cast")];
            tensor<int32, [1]> var_7298 = const()[name = tensor<string, []>("op_7298"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_7299_cast = reduce_mean(axes = var_7298, keep_dims = var_2629, x = zero_mean_sq_55_cast)[name = tensor<string, []>("op_7299_cast")];
            tensor<fp16, []> var_7300_to_fp16 = const()[name = tensor<string, []>("op_7300_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_7301_cast = add(x = var_7299_cast, y = var_7300_to_fp16)[name = tensor<string, []>("op_7301_cast")];
            tensor<fp16, []> denom_55_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_55_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_55_cast = rsqrt(epsilon = denom_55_epsilon_0_to_fp16, x = var_7301_cast)[name = tensor<string, []>("denom_55_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_55_cast = mul(x = zero_mean_55_cast, y = denom_55_cast)[name = tensor<string, []>("out_55_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7305_to_fp16 = const()[name = tensor<string, []>("op_7305_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545730688)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_7306_cast = add(x = out_55_cast, y = var_7305_to_fp16)[name = tensor<string, []>("op_7306_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7308_to_fp16 = const()[name = tensor<string, []>("op_7308_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545733312)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_95_cast = mul(x = var_7306_cast, y = var_7308_to_fp16)[name = tensor<string, []>("hidden_states_95_cast")];
            tensor<int32, [2]> var_7315 = const()[name = tensor<string, []>("op_7315"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7317 = const()[name = tensor<string, []>("op_7317"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_37_pad_type_0 = const()[name = tensor<string, []>("q_37_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_37_pad_0 = const()[name = tensor<string, []>("q_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545735936)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_37_cast = conv(dilations = var_7317, groups = var_2634, pad = q_37_pad_0, pad_type = q_37_pad_type_0, strides = var_7315, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16, x = hidden_states_95_cast)[name = tensor<string, []>("q_37_cast")];
            tensor<int32, [2]> var_7321 = const()[name = tensor<string, []>("op_7321"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7323 = const()[name = tensor<string, []>("op_7323"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_73_pad_type_0 = const()[name = tensor<string, []>("k_73_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_73_pad_0 = const()[name = tensor<string, []>("k_73_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(549012800)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_73_cast = conv(dilations = var_7323, groups = var_2634, pad = k_73_pad_0, pad_type = k_73_pad_type_0, strides = var_7321, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16, x = hidden_states_95_cast)[name = tensor<string, []>("k_73_cast")];
            tensor<int32, [2]> var_7327 = const()[name = tensor<string, []>("op_7327"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7329 = const()[name = tensor<string, []>("op_7329"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_37_pad_type_0 = const()[name = tensor<string, []>("v_37_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_37_pad_0 = const()[name = tensor<string, []>("v_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(552289664)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_37_cast = conv(dilations = var_7329, groups = var_2634, pad = v_37_pad_0, pad_type = v_37_pad_type_0, strides = var_7327, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16, x = hidden_states_95_cast)[name = tensor<string, []>("v_37_cast")];
            tensor<int32, [4]> var_7333_begin_0 = const()[name = tensor<string, []>("op_7333_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7333_end_0 = const()[name = tensor<string, []>("op_7333_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_7333_end_mask_0 = const()[name = tensor<string, []>("op_7333_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7333_cast = slice_by_index(begin = var_7333_begin_0, end = var_7333_end_0, end_mask = var_7333_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7333_cast")];
            tensor<int32, [4]> var_7337_begin_0 = const()[name = tensor<string, []>("op_7337_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_7337_end_0 = const()[name = tensor<string, []>("op_7337_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_7337_end_mask_0 = const()[name = tensor<string, []>("op_7337_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7337_cast = slice_by_index(begin = var_7337_begin_0, end = var_7337_end_0, end_mask = var_7337_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7337_cast")];
            tensor<int32, [4]> var_7341_begin_0 = const()[name = tensor<string, []>("op_7341_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_7341_end_0 = const()[name = tensor<string, []>("op_7341_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_7341_end_mask_0 = const()[name = tensor<string, []>("op_7341_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7341_cast = slice_by_index(begin = var_7341_begin_0, end = var_7341_end_0, end_mask = var_7341_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7341_cast")];
            tensor<int32, [4]> var_7345_begin_0 = const()[name = tensor<string, []>("op_7345_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_7345_end_0 = const()[name = tensor<string, []>("op_7345_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_7345_end_mask_0 = const()[name = tensor<string, []>("op_7345_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7345_cast = slice_by_index(begin = var_7345_begin_0, end = var_7345_end_0, end_mask = var_7345_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7345_cast")];
            tensor<int32, [4]> var_7349_begin_0 = const()[name = tensor<string, []>("op_7349_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_7349_end_0 = const()[name = tensor<string, []>("op_7349_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_7349_end_mask_0 = const()[name = tensor<string, []>("op_7349_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7349_cast = slice_by_index(begin = var_7349_begin_0, end = var_7349_end_0, end_mask = var_7349_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7349_cast")];
            tensor<int32, [4]> var_7353_begin_0 = const()[name = tensor<string, []>("op_7353_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7353_end_0 = const()[name = tensor<string, []>("op_7353_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_7353_end_mask_0 = const()[name = tensor<string, []>("op_7353_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7353_cast = slice_by_index(begin = var_7353_begin_0, end = var_7353_end_0, end_mask = var_7353_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7353_cast")];
            tensor<int32, [4]> var_7357_begin_0 = const()[name = tensor<string, []>("op_7357_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_7357_end_0 = const()[name = tensor<string, []>("op_7357_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_7357_end_mask_0 = const()[name = tensor<string, []>("op_7357_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7357_cast = slice_by_index(begin = var_7357_begin_0, end = var_7357_end_0, end_mask = var_7357_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7357_cast")];
            tensor<int32, [4]> var_7361_begin_0 = const()[name = tensor<string, []>("op_7361_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_7361_end_0 = const()[name = tensor<string, []>("op_7361_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_7361_end_mask_0 = const()[name = tensor<string, []>("op_7361_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7361_cast = slice_by_index(begin = var_7361_begin_0, end = var_7361_end_0, end_mask = var_7361_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7361_cast")];
            tensor<int32, [4]> var_7365_begin_0 = const()[name = tensor<string, []>("op_7365_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_7365_end_0 = const()[name = tensor<string, []>("op_7365_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_7365_end_mask_0 = const()[name = tensor<string, []>("op_7365_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7365_cast = slice_by_index(begin = var_7365_begin_0, end = var_7365_end_0, end_mask = var_7365_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7365_cast")];
            tensor<int32, [4]> var_7369_begin_0 = const()[name = tensor<string, []>("op_7369_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_7369_end_0 = const()[name = tensor<string, []>("op_7369_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_7369_end_mask_0 = const()[name = tensor<string, []>("op_7369_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7369_cast = slice_by_index(begin = var_7369_begin_0, end = var_7369_end_0, end_mask = var_7369_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7369_cast")];
            tensor<int32, [4]> var_7373_begin_0 = const()[name = tensor<string, []>("op_7373_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7373_end_0 = const()[name = tensor<string, []>("op_7373_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_7373_end_mask_0 = const()[name = tensor<string, []>("op_7373_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7373_cast = slice_by_index(begin = var_7373_begin_0, end = var_7373_end_0, end_mask = var_7373_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7373_cast")];
            tensor<int32, [4]> var_7377_begin_0 = const()[name = tensor<string, []>("op_7377_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_7377_end_0 = const()[name = tensor<string, []>("op_7377_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_7377_end_mask_0 = const()[name = tensor<string, []>("op_7377_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7377_cast = slice_by_index(begin = var_7377_begin_0, end = var_7377_end_0, end_mask = var_7377_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7377_cast")];
            tensor<int32, [4]> var_7381_begin_0 = const()[name = tensor<string, []>("op_7381_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_7381_end_0 = const()[name = tensor<string, []>("op_7381_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_7381_end_mask_0 = const()[name = tensor<string, []>("op_7381_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7381_cast = slice_by_index(begin = var_7381_begin_0, end = var_7381_end_0, end_mask = var_7381_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7381_cast")];
            tensor<int32, [4]> var_7385_begin_0 = const()[name = tensor<string, []>("op_7385_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_7385_end_0 = const()[name = tensor<string, []>("op_7385_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_7385_end_mask_0 = const()[name = tensor<string, []>("op_7385_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7385_cast = slice_by_index(begin = var_7385_begin_0, end = var_7385_end_0, end_mask = var_7385_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7385_cast")];
            tensor<int32, [4]> var_7389_begin_0 = const()[name = tensor<string, []>("op_7389_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_7389_end_0 = const()[name = tensor<string, []>("op_7389_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_7389_end_mask_0 = const()[name = tensor<string, []>("op_7389_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7389_cast = slice_by_index(begin = var_7389_begin_0, end = var_7389_end_0, end_mask = var_7389_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7389_cast")];
            tensor<int32, [4]> var_7393_begin_0 = const()[name = tensor<string, []>("op_7393_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_7393_end_0 = const()[name = tensor<string, []>("op_7393_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_7393_end_mask_0 = const()[name = tensor<string, []>("op_7393_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7393_cast = slice_by_index(begin = var_7393_begin_0, end = var_7393_end_0, end_mask = var_7393_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7393_cast")];
            tensor<int32, [4]> var_7397_begin_0 = const()[name = tensor<string, []>("op_7397_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_7397_end_0 = const()[name = tensor<string, []>("op_7397_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_7397_end_mask_0 = const()[name = tensor<string, []>("op_7397_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7397_cast = slice_by_index(begin = var_7397_begin_0, end = var_7397_end_0, end_mask = var_7397_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7397_cast")];
            tensor<int32, [4]> var_7401_begin_0 = const()[name = tensor<string, []>("op_7401_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_7401_end_0 = const()[name = tensor<string, []>("op_7401_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_7401_end_mask_0 = const()[name = tensor<string, []>("op_7401_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7401_cast = slice_by_index(begin = var_7401_begin_0, end = var_7401_end_0, end_mask = var_7401_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7401_cast")];
            tensor<int32, [4]> var_7405_begin_0 = const()[name = tensor<string, []>("op_7405_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_7405_end_0 = const()[name = tensor<string, []>("op_7405_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_7405_end_mask_0 = const()[name = tensor<string, []>("op_7405_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7405_cast = slice_by_index(begin = var_7405_begin_0, end = var_7405_end_0, end_mask = var_7405_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7405_cast")];
            tensor<int32, [4]> var_7409_begin_0 = const()[name = tensor<string, []>("op_7409_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_7409_end_0 = const()[name = tensor<string, []>("op_7409_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_7409_end_mask_0 = const()[name = tensor<string, []>("op_7409_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7409_cast = slice_by_index(begin = var_7409_begin_0, end = var_7409_end_0, end_mask = var_7409_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7409_cast")];
            tensor<int32, [4]> k_75_perm_0 = const()[name = tensor<string, []>("k_75_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_7416_begin_0 = const()[name = tensor<string, []>("op_7416_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7416_end_0 = const()[name = tensor<string, []>("op_7416_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_7416_end_mask_0 = const()[name = tensor<string, []>("op_7416_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_121 = transpose(perm = k_75_perm_0, x = k_73_cast)[name = tensor<string, []>("transpose_121")];
            tensor<fp16, [2, 1024, 1, 64]> var_7416_cast = slice_by_index(begin = var_7416_begin_0, end = var_7416_end_0, end_mask = var_7416_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7416_cast")];
            tensor<int32, [4]> var_7420_begin_0 = const()[name = tensor<string, []>("op_7420_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_7420_end_0 = const()[name = tensor<string, []>("op_7420_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_7420_end_mask_0 = const()[name = tensor<string, []>("op_7420_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_7420_cast = slice_by_index(begin = var_7420_begin_0, end = var_7420_end_0, end_mask = var_7420_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7420_cast")];
            tensor<int32, [4]> var_7424_begin_0 = const()[name = tensor<string, []>("op_7424_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_7424_end_0 = const()[name = tensor<string, []>("op_7424_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_7424_end_mask_0 = const()[name = tensor<string, []>("op_7424_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_7424_cast = slice_by_index(begin = var_7424_begin_0, end = var_7424_end_0, end_mask = var_7424_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7424_cast")];
            tensor<int32, [4]> var_7428_begin_0 = const()[name = tensor<string, []>("op_7428_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_7428_end_0 = const()[name = tensor<string, []>("op_7428_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_7428_end_mask_0 = const()[name = tensor<string, []>("op_7428_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_7428_cast = slice_by_index(begin = var_7428_begin_0, end = var_7428_end_0, end_mask = var_7428_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7428_cast")];
            tensor<int32, [4]> var_7432_begin_0 = const()[name = tensor<string, []>("op_7432_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_7432_end_0 = const()[name = tensor<string, []>("op_7432_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_7432_end_mask_0 = const()[name = tensor<string, []>("op_7432_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_7432_cast = slice_by_index(begin = var_7432_begin_0, end = var_7432_end_0, end_mask = var_7432_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7432_cast")];
            tensor<int32, [4]> var_7436_begin_0 = const()[name = tensor<string, []>("op_7436_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_7436_end_0 = const()[name = tensor<string, []>("op_7436_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_7436_end_mask_0 = const()[name = tensor<string, []>("op_7436_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_7436_cast = slice_by_index(begin = var_7436_begin_0, end = var_7436_end_0, end_mask = var_7436_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7436_cast")];
            tensor<int32, [4]> var_7440_begin_0 = const()[name = tensor<string, []>("op_7440_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_7440_end_0 = const()[name = tensor<string, []>("op_7440_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_7440_end_mask_0 = const()[name = tensor<string, []>("op_7440_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_7440_cast = slice_by_index(begin = var_7440_begin_0, end = var_7440_end_0, end_mask = var_7440_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7440_cast")];
            tensor<int32, [4]> var_7444_begin_0 = const()[name = tensor<string, []>("op_7444_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_7444_end_0 = const()[name = tensor<string, []>("op_7444_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_7444_end_mask_0 = const()[name = tensor<string, []>("op_7444_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_7444_cast = slice_by_index(begin = var_7444_begin_0, end = var_7444_end_0, end_mask = var_7444_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7444_cast")];
            tensor<int32, [4]> var_7448_begin_0 = const()[name = tensor<string, []>("op_7448_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_7448_end_0 = const()[name = tensor<string, []>("op_7448_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_7448_end_mask_0 = const()[name = tensor<string, []>("op_7448_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_7448_cast = slice_by_index(begin = var_7448_begin_0, end = var_7448_end_0, end_mask = var_7448_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7448_cast")];
            tensor<int32, [4]> var_7452_begin_0 = const()[name = tensor<string, []>("op_7452_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_7452_end_0 = const()[name = tensor<string, []>("op_7452_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_7452_end_mask_0 = const()[name = tensor<string, []>("op_7452_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_7452_cast = slice_by_index(begin = var_7452_begin_0, end = var_7452_end_0, end_mask = var_7452_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7452_cast")];
            tensor<int32, [4]> var_7456_begin_0 = const()[name = tensor<string, []>("op_7456_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_7456_end_0 = const()[name = tensor<string, []>("op_7456_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_7456_end_mask_0 = const()[name = tensor<string, []>("op_7456_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_7456_cast = slice_by_index(begin = var_7456_begin_0, end = var_7456_end_0, end_mask = var_7456_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7456_cast")];
            tensor<int32, [4]> var_7460_begin_0 = const()[name = tensor<string, []>("op_7460_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_7460_end_0 = const()[name = tensor<string, []>("op_7460_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_7460_end_mask_0 = const()[name = tensor<string, []>("op_7460_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_7460_cast = slice_by_index(begin = var_7460_begin_0, end = var_7460_end_0, end_mask = var_7460_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7460_cast")];
            tensor<int32, [4]> var_7464_begin_0 = const()[name = tensor<string, []>("op_7464_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_7464_end_0 = const()[name = tensor<string, []>("op_7464_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_7464_end_mask_0 = const()[name = tensor<string, []>("op_7464_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_7464_cast = slice_by_index(begin = var_7464_begin_0, end = var_7464_end_0, end_mask = var_7464_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7464_cast")];
            tensor<int32, [4]> var_7468_begin_0 = const()[name = tensor<string, []>("op_7468_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_7468_end_0 = const()[name = tensor<string, []>("op_7468_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_7468_end_mask_0 = const()[name = tensor<string, []>("op_7468_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_7468_cast = slice_by_index(begin = var_7468_begin_0, end = var_7468_end_0, end_mask = var_7468_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7468_cast")];
            tensor<int32, [4]> var_7472_begin_0 = const()[name = tensor<string, []>("op_7472_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_7472_end_0 = const()[name = tensor<string, []>("op_7472_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_7472_end_mask_0 = const()[name = tensor<string, []>("op_7472_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_7472_cast = slice_by_index(begin = var_7472_begin_0, end = var_7472_end_0, end_mask = var_7472_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7472_cast")];
            tensor<int32, [4]> var_7476_begin_0 = const()[name = tensor<string, []>("op_7476_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_7476_end_0 = const()[name = tensor<string, []>("op_7476_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_7476_end_mask_0 = const()[name = tensor<string, []>("op_7476_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_7476_cast = slice_by_index(begin = var_7476_begin_0, end = var_7476_end_0, end_mask = var_7476_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7476_cast")];
            tensor<int32, [4]> var_7480_begin_0 = const()[name = tensor<string, []>("op_7480_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_7480_end_0 = const()[name = tensor<string, []>("op_7480_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_7480_end_mask_0 = const()[name = tensor<string, []>("op_7480_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_7480_cast = slice_by_index(begin = var_7480_begin_0, end = var_7480_end_0, end_mask = var_7480_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7480_cast")];
            tensor<int32, [4]> var_7484_begin_0 = const()[name = tensor<string, []>("op_7484_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_7484_end_0 = const()[name = tensor<string, []>("op_7484_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_7484_end_mask_0 = const()[name = tensor<string, []>("op_7484_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_7484_cast = slice_by_index(begin = var_7484_begin_0, end = var_7484_end_0, end_mask = var_7484_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7484_cast")];
            tensor<int32, [4]> var_7488_begin_0 = const()[name = tensor<string, []>("op_7488_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_7488_end_0 = const()[name = tensor<string, []>("op_7488_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_7488_end_mask_0 = const()[name = tensor<string, []>("op_7488_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_7488_cast = slice_by_index(begin = var_7488_begin_0, end = var_7488_end_0, end_mask = var_7488_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7488_cast")];
            tensor<int32, [4]> var_7492_begin_0 = const()[name = tensor<string, []>("op_7492_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_7492_end_0 = const()[name = tensor<string, []>("op_7492_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_7492_end_mask_0 = const()[name = tensor<string, []>("op_7492_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_7492_cast = slice_by_index(begin = var_7492_begin_0, end = var_7492_end_0, end_mask = var_7492_end_mask_0, x = transpose_121)[name = tensor<string, []>("op_7492_cast")];
            tensor<int32, [4]> var_7494_begin_0 = const()[name = tensor<string, []>("op_7494_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7494_end_0 = const()[name = tensor<string, []>("op_7494_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_7494_end_mask_0 = const()[name = tensor<string, []>("op_7494_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7494_cast = slice_by_index(begin = var_7494_begin_0, end = var_7494_end_0, end_mask = var_7494_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7494_cast")];
            tensor<int32, [4]> var_7498_begin_0 = const()[name = tensor<string, []>("op_7498_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_7498_end_0 = const()[name = tensor<string, []>("op_7498_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_7498_end_mask_0 = const()[name = tensor<string, []>("op_7498_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7498_cast = slice_by_index(begin = var_7498_begin_0, end = var_7498_end_0, end_mask = var_7498_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7498_cast")];
            tensor<int32, [4]> var_7502_begin_0 = const()[name = tensor<string, []>("op_7502_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_7502_end_0 = const()[name = tensor<string, []>("op_7502_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_7502_end_mask_0 = const()[name = tensor<string, []>("op_7502_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7502_cast = slice_by_index(begin = var_7502_begin_0, end = var_7502_end_0, end_mask = var_7502_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7502_cast")];
            tensor<int32, [4]> var_7506_begin_0 = const()[name = tensor<string, []>("op_7506_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_7506_end_0 = const()[name = tensor<string, []>("op_7506_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_7506_end_mask_0 = const()[name = tensor<string, []>("op_7506_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7506_cast = slice_by_index(begin = var_7506_begin_0, end = var_7506_end_0, end_mask = var_7506_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7506_cast")];
            tensor<int32, [4]> var_7510_begin_0 = const()[name = tensor<string, []>("op_7510_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_7510_end_0 = const()[name = tensor<string, []>("op_7510_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_7510_end_mask_0 = const()[name = tensor<string, []>("op_7510_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7510_cast = slice_by_index(begin = var_7510_begin_0, end = var_7510_end_0, end_mask = var_7510_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7510_cast")];
            tensor<int32, [4]> var_7514_begin_0 = const()[name = tensor<string, []>("op_7514_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7514_end_0 = const()[name = tensor<string, []>("op_7514_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_7514_end_mask_0 = const()[name = tensor<string, []>("op_7514_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7514_cast = slice_by_index(begin = var_7514_begin_0, end = var_7514_end_0, end_mask = var_7514_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7514_cast")];
            tensor<int32, [4]> var_7518_begin_0 = const()[name = tensor<string, []>("op_7518_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_7518_end_0 = const()[name = tensor<string, []>("op_7518_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_7518_end_mask_0 = const()[name = tensor<string, []>("op_7518_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7518_cast = slice_by_index(begin = var_7518_begin_0, end = var_7518_end_0, end_mask = var_7518_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7518_cast")];
            tensor<int32, [4]> var_7522_begin_0 = const()[name = tensor<string, []>("op_7522_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_7522_end_0 = const()[name = tensor<string, []>("op_7522_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_7522_end_mask_0 = const()[name = tensor<string, []>("op_7522_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7522_cast = slice_by_index(begin = var_7522_begin_0, end = var_7522_end_0, end_mask = var_7522_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7522_cast")];
            tensor<int32, [4]> var_7526_begin_0 = const()[name = tensor<string, []>("op_7526_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_7526_end_0 = const()[name = tensor<string, []>("op_7526_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_7526_end_mask_0 = const()[name = tensor<string, []>("op_7526_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7526_cast = slice_by_index(begin = var_7526_begin_0, end = var_7526_end_0, end_mask = var_7526_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7526_cast")];
            tensor<int32, [4]> var_7530_begin_0 = const()[name = tensor<string, []>("op_7530_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_7530_end_0 = const()[name = tensor<string, []>("op_7530_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_7530_end_mask_0 = const()[name = tensor<string, []>("op_7530_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7530_cast = slice_by_index(begin = var_7530_begin_0, end = var_7530_end_0, end_mask = var_7530_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7530_cast")];
            tensor<int32, [4]> var_7534_begin_0 = const()[name = tensor<string, []>("op_7534_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7534_end_0 = const()[name = tensor<string, []>("op_7534_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_7534_end_mask_0 = const()[name = tensor<string, []>("op_7534_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7534_cast = slice_by_index(begin = var_7534_begin_0, end = var_7534_end_0, end_mask = var_7534_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7534_cast")];
            tensor<int32, [4]> var_7538_begin_0 = const()[name = tensor<string, []>("op_7538_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_7538_end_0 = const()[name = tensor<string, []>("op_7538_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_7538_end_mask_0 = const()[name = tensor<string, []>("op_7538_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7538_cast = slice_by_index(begin = var_7538_begin_0, end = var_7538_end_0, end_mask = var_7538_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7538_cast")];
            tensor<int32, [4]> var_7542_begin_0 = const()[name = tensor<string, []>("op_7542_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_7542_end_0 = const()[name = tensor<string, []>("op_7542_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_7542_end_mask_0 = const()[name = tensor<string, []>("op_7542_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7542_cast = slice_by_index(begin = var_7542_begin_0, end = var_7542_end_0, end_mask = var_7542_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7542_cast")];
            tensor<int32, [4]> var_7546_begin_0 = const()[name = tensor<string, []>("op_7546_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_7546_end_0 = const()[name = tensor<string, []>("op_7546_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_7546_end_mask_0 = const()[name = tensor<string, []>("op_7546_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7546_cast = slice_by_index(begin = var_7546_begin_0, end = var_7546_end_0, end_mask = var_7546_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7546_cast")];
            tensor<int32, [4]> var_7550_begin_0 = const()[name = tensor<string, []>("op_7550_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_7550_end_0 = const()[name = tensor<string, []>("op_7550_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_7550_end_mask_0 = const()[name = tensor<string, []>("op_7550_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7550_cast = slice_by_index(begin = var_7550_begin_0, end = var_7550_end_0, end_mask = var_7550_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7550_cast")];
            tensor<int32, [4]> var_7554_begin_0 = const()[name = tensor<string, []>("op_7554_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_7554_end_0 = const()[name = tensor<string, []>("op_7554_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_7554_end_mask_0 = const()[name = tensor<string, []>("op_7554_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7554_cast = slice_by_index(begin = var_7554_begin_0, end = var_7554_end_0, end_mask = var_7554_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7554_cast")];
            tensor<int32, [4]> var_7558_begin_0 = const()[name = tensor<string, []>("op_7558_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_7558_end_0 = const()[name = tensor<string, []>("op_7558_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_7558_end_mask_0 = const()[name = tensor<string, []>("op_7558_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7558_cast = slice_by_index(begin = var_7558_begin_0, end = var_7558_end_0, end_mask = var_7558_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7558_cast")];
            tensor<int32, [4]> var_7562_begin_0 = const()[name = tensor<string, []>("op_7562_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_7562_end_0 = const()[name = tensor<string, []>("op_7562_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_7562_end_mask_0 = const()[name = tensor<string, []>("op_7562_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7562_cast = slice_by_index(begin = var_7562_begin_0, end = var_7562_end_0, end_mask = var_7562_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7562_cast")];
            tensor<int32, [4]> var_7566_begin_0 = const()[name = tensor<string, []>("op_7566_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_7566_end_0 = const()[name = tensor<string, []>("op_7566_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_7566_end_mask_0 = const()[name = tensor<string, []>("op_7566_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7566_cast = slice_by_index(begin = var_7566_begin_0, end = var_7566_end_0, end_mask = var_7566_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7566_cast")];
            tensor<int32, [4]> var_7570_begin_0 = const()[name = tensor<string, []>("op_7570_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_7570_end_0 = const()[name = tensor<string, []>("op_7570_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_7570_end_mask_0 = const()[name = tensor<string, []>("op_7570_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7570_cast = slice_by_index(begin = var_7570_begin_0, end = var_7570_end_0, end_mask = var_7570_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7570_cast")];
            tensor<string, []> var_7574_equation_0 = const()[name = tensor<string, []>("op_7574_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7574_cast = einsum(equation = var_7574_equation_0, values = (var_7416_cast, var_7333_cast))[name = tensor<string, []>("op_7574_cast")];
            tensor<fp16, []> var_7575_to_fp16 = const()[name = tensor<string, []>("op_7575_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_561_cast = mul(x = var_7574_cast, y = var_7575_to_fp16)[name = tensor<string, []>("aw_561_cast")];
            tensor<string, []> var_7578_equation_0 = const()[name = tensor<string, []>("op_7578_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7578_cast = einsum(equation = var_7578_equation_0, values = (var_7420_cast, var_7337_cast))[name = tensor<string, []>("op_7578_cast")];
            tensor<fp16, []> var_7579_to_fp16 = const()[name = tensor<string, []>("op_7579_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_563_cast = mul(x = var_7578_cast, y = var_7579_to_fp16)[name = tensor<string, []>("aw_563_cast")];
            tensor<string, []> var_7582_equation_0 = const()[name = tensor<string, []>("op_7582_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7582_cast = einsum(equation = var_7582_equation_0, values = (var_7424_cast, var_7341_cast))[name = tensor<string, []>("op_7582_cast")];
            tensor<fp16, []> var_7583_to_fp16 = const()[name = tensor<string, []>("op_7583_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_565_cast = mul(x = var_7582_cast, y = var_7583_to_fp16)[name = tensor<string, []>("aw_565_cast")];
            tensor<string, []> var_7586_equation_0 = const()[name = tensor<string, []>("op_7586_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7586_cast = einsum(equation = var_7586_equation_0, values = (var_7428_cast, var_7345_cast))[name = tensor<string, []>("op_7586_cast")];
            tensor<fp16, []> var_7587_to_fp16 = const()[name = tensor<string, []>("op_7587_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_567_cast = mul(x = var_7586_cast, y = var_7587_to_fp16)[name = tensor<string, []>("aw_567_cast")];
            tensor<string, []> var_7590_equation_0 = const()[name = tensor<string, []>("op_7590_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7590_cast = einsum(equation = var_7590_equation_0, values = (var_7432_cast, var_7349_cast))[name = tensor<string, []>("op_7590_cast")];
            tensor<fp16, []> var_7591_to_fp16 = const()[name = tensor<string, []>("op_7591_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_569_cast = mul(x = var_7590_cast, y = var_7591_to_fp16)[name = tensor<string, []>("aw_569_cast")];
            tensor<string, []> var_7594_equation_0 = const()[name = tensor<string, []>("op_7594_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7594_cast = einsum(equation = var_7594_equation_0, values = (var_7436_cast, var_7353_cast))[name = tensor<string, []>("op_7594_cast")];
            tensor<fp16, []> var_7595_to_fp16 = const()[name = tensor<string, []>("op_7595_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_571_cast = mul(x = var_7594_cast, y = var_7595_to_fp16)[name = tensor<string, []>("aw_571_cast")];
            tensor<string, []> var_7598_equation_0 = const()[name = tensor<string, []>("op_7598_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7598_cast = einsum(equation = var_7598_equation_0, values = (var_7440_cast, var_7357_cast))[name = tensor<string, []>("op_7598_cast")];
            tensor<fp16, []> var_7599_to_fp16 = const()[name = tensor<string, []>("op_7599_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_573_cast = mul(x = var_7598_cast, y = var_7599_to_fp16)[name = tensor<string, []>("aw_573_cast")];
            tensor<string, []> var_7602_equation_0 = const()[name = tensor<string, []>("op_7602_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7602_cast = einsum(equation = var_7602_equation_0, values = (var_7444_cast, var_7361_cast))[name = tensor<string, []>("op_7602_cast")];
            tensor<fp16, []> var_7603_to_fp16 = const()[name = tensor<string, []>("op_7603_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_575_cast = mul(x = var_7602_cast, y = var_7603_to_fp16)[name = tensor<string, []>("aw_575_cast")];
            tensor<string, []> var_7606_equation_0 = const()[name = tensor<string, []>("op_7606_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7606_cast = einsum(equation = var_7606_equation_0, values = (var_7448_cast, var_7365_cast))[name = tensor<string, []>("op_7606_cast")];
            tensor<fp16, []> var_7607_to_fp16 = const()[name = tensor<string, []>("op_7607_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_577_cast = mul(x = var_7606_cast, y = var_7607_to_fp16)[name = tensor<string, []>("aw_577_cast")];
            tensor<string, []> var_7610_equation_0 = const()[name = tensor<string, []>("op_7610_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7610_cast = einsum(equation = var_7610_equation_0, values = (var_7452_cast, var_7369_cast))[name = tensor<string, []>("op_7610_cast")];
            tensor<fp16, []> var_7611_to_fp16 = const()[name = tensor<string, []>("op_7611_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_579_cast = mul(x = var_7610_cast, y = var_7611_to_fp16)[name = tensor<string, []>("aw_579_cast")];
            tensor<string, []> var_7614_equation_0 = const()[name = tensor<string, []>("op_7614_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7614_cast = einsum(equation = var_7614_equation_0, values = (var_7456_cast, var_7373_cast))[name = tensor<string, []>("op_7614_cast")];
            tensor<fp16, []> var_7615_to_fp16 = const()[name = tensor<string, []>("op_7615_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_581_cast = mul(x = var_7614_cast, y = var_7615_to_fp16)[name = tensor<string, []>("aw_581_cast")];
            tensor<string, []> var_7618_equation_0 = const()[name = tensor<string, []>("op_7618_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7618_cast = einsum(equation = var_7618_equation_0, values = (var_7460_cast, var_7377_cast))[name = tensor<string, []>("op_7618_cast")];
            tensor<fp16, []> var_7619_to_fp16 = const()[name = tensor<string, []>("op_7619_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_583_cast = mul(x = var_7618_cast, y = var_7619_to_fp16)[name = tensor<string, []>("aw_583_cast")];
            tensor<string, []> var_7622_equation_0 = const()[name = tensor<string, []>("op_7622_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7622_cast = einsum(equation = var_7622_equation_0, values = (var_7464_cast, var_7381_cast))[name = tensor<string, []>("op_7622_cast")];
            tensor<fp16, []> var_7623_to_fp16 = const()[name = tensor<string, []>("op_7623_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_585_cast = mul(x = var_7622_cast, y = var_7623_to_fp16)[name = tensor<string, []>("aw_585_cast")];
            tensor<string, []> var_7626_equation_0 = const()[name = tensor<string, []>("op_7626_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7626_cast = einsum(equation = var_7626_equation_0, values = (var_7468_cast, var_7385_cast))[name = tensor<string, []>("op_7626_cast")];
            tensor<fp16, []> var_7627_to_fp16 = const()[name = tensor<string, []>("op_7627_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_587_cast = mul(x = var_7626_cast, y = var_7627_to_fp16)[name = tensor<string, []>("aw_587_cast")];
            tensor<string, []> var_7630_equation_0 = const()[name = tensor<string, []>("op_7630_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7630_cast = einsum(equation = var_7630_equation_0, values = (var_7472_cast, var_7389_cast))[name = tensor<string, []>("op_7630_cast")];
            tensor<fp16, []> var_7631_to_fp16 = const()[name = tensor<string, []>("op_7631_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_589_cast = mul(x = var_7630_cast, y = var_7631_to_fp16)[name = tensor<string, []>("aw_589_cast")];
            tensor<string, []> var_7634_equation_0 = const()[name = tensor<string, []>("op_7634_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7634_cast = einsum(equation = var_7634_equation_0, values = (var_7476_cast, var_7393_cast))[name = tensor<string, []>("op_7634_cast")];
            tensor<fp16, []> var_7635_to_fp16 = const()[name = tensor<string, []>("op_7635_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_591_cast = mul(x = var_7634_cast, y = var_7635_to_fp16)[name = tensor<string, []>("aw_591_cast")];
            tensor<string, []> var_7638_equation_0 = const()[name = tensor<string, []>("op_7638_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7638_cast = einsum(equation = var_7638_equation_0, values = (var_7480_cast, var_7397_cast))[name = tensor<string, []>("op_7638_cast")];
            tensor<fp16, []> var_7639_to_fp16 = const()[name = tensor<string, []>("op_7639_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_593_cast = mul(x = var_7638_cast, y = var_7639_to_fp16)[name = tensor<string, []>("aw_593_cast")];
            tensor<string, []> var_7642_equation_0 = const()[name = tensor<string, []>("op_7642_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7642_cast = einsum(equation = var_7642_equation_0, values = (var_7484_cast, var_7401_cast))[name = tensor<string, []>("op_7642_cast")];
            tensor<fp16, []> var_7643_to_fp16 = const()[name = tensor<string, []>("op_7643_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_595_cast = mul(x = var_7642_cast, y = var_7643_to_fp16)[name = tensor<string, []>("aw_595_cast")];
            tensor<string, []> var_7646_equation_0 = const()[name = tensor<string, []>("op_7646_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7646_cast = einsum(equation = var_7646_equation_0, values = (var_7488_cast, var_7405_cast))[name = tensor<string, []>("op_7646_cast")];
            tensor<fp16, []> var_7647_to_fp16 = const()[name = tensor<string, []>("op_7647_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_597_cast = mul(x = var_7646_cast, y = var_7647_to_fp16)[name = tensor<string, []>("aw_597_cast")];
            tensor<string, []> var_7650_equation_0 = const()[name = tensor<string, []>("op_7650_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7650_cast = einsum(equation = var_7650_equation_0, values = (var_7492_cast, var_7409_cast))[name = tensor<string, []>("op_7650_cast")];
            tensor<fp16, []> var_7651_to_fp16 = const()[name = tensor<string, []>("op_7651_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_599_cast = mul(x = var_7650_cast, y = var_7651_to_fp16)[name = tensor<string, []>("aw_599_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7653_cast = softmax(axis = var_2634, x = aw_561_cast)[name = tensor<string, []>("op_7653_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7654_cast = softmax(axis = var_2634, x = aw_563_cast)[name = tensor<string, []>("op_7654_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7655_cast = softmax(axis = var_2634, x = aw_565_cast)[name = tensor<string, []>("op_7655_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7656_cast = softmax(axis = var_2634, x = aw_567_cast)[name = tensor<string, []>("op_7656_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7657_cast = softmax(axis = var_2634, x = aw_569_cast)[name = tensor<string, []>("op_7657_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7658_cast = softmax(axis = var_2634, x = aw_571_cast)[name = tensor<string, []>("op_7658_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7659_cast = softmax(axis = var_2634, x = aw_573_cast)[name = tensor<string, []>("op_7659_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7660_cast = softmax(axis = var_2634, x = aw_575_cast)[name = tensor<string, []>("op_7660_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7661_cast = softmax(axis = var_2634, x = aw_577_cast)[name = tensor<string, []>("op_7661_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7662_cast = softmax(axis = var_2634, x = aw_579_cast)[name = tensor<string, []>("op_7662_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7663_cast = softmax(axis = var_2634, x = aw_581_cast)[name = tensor<string, []>("op_7663_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7664_cast = softmax(axis = var_2634, x = aw_583_cast)[name = tensor<string, []>("op_7664_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7665_cast = softmax(axis = var_2634, x = aw_585_cast)[name = tensor<string, []>("op_7665_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7666_cast = softmax(axis = var_2634, x = aw_587_cast)[name = tensor<string, []>("op_7666_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7667_cast = softmax(axis = var_2634, x = aw_589_cast)[name = tensor<string, []>("op_7667_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7668_cast = softmax(axis = var_2634, x = aw_591_cast)[name = tensor<string, []>("op_7668_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7669_cast = softmax(axis = var_2634, x = aw_593_cast)[name = tensor<string, []>("op_7669_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7670_cast = softmax(axis = var_2634, x = aw_595_cast)[name = tensor<string, []>("op_7670_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7671_cast = softmax(axis = var_2634, x = aw_597_cast)[name = tensor<string, []>("op_7671_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_7672_cast = softmax(axis = var_2634, x = aw_599_cast)[name = tensor<string, []>("op_7672_cast")];
            tensor<string, []> var_7674_equation_0 = const()[name = tensor<string, []>("op_7674_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7674_cast = einsum(equation = var_7674_equation_0, values = (var_7494_cast, var_7653_cast))[name = tensor<string, []>("op_7674_cast")];
            tensor<string, []> var_7676_equation_0 = const()[name = tensor<string, []>("op_7676_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7676_cast = einsum(equation = var_7676_equation_0, values = (var_7498_cast, var_7654_cast))[name = tensor<string, []>("op_7676_cast")];
            tensor<string, []> var_7678_equation_0 = const()[name = tensor<string, []>("op_7678_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7678_cast = einsum(equation = var_7678_equation_0, values = (var_7502_cast, var_7655_cast))[name = tensor<string, []>("op_7678_cast")];
            tensor<string, []> var_7680_equation_0 = const()[name = tensor<string, []>("op_7680_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7680_cast = einsum(equation = var_7680_equation_0, values = (var_7506_cast, var_7656_cast))[name = tensor<string, []>("op_7680_cast")];
            tensor<string, []> var_7682_equation_0 = const()[name = tensor<string, []>("op_7682_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7682_cast = einsum(equation = var_7682_equation_0, values = (var_7510_cast, var_7657_cast))[name = tensor<string, []>("op_7682_cast")];
            tensor<string, []> var_7684_equation_0 = const()[name = tensor<string, []>("op_7684_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7684_cast = einsum(equation = var_7684_equation_0, values = (var_7514_cast, var_7658_cast))[name = tensor<string, []>("op_7684_cast")];
            tensor<string, []> var_7686_equation_0 = const()[name = tensor<string, []>("op_7686_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7686_cast = einsum(equation = var_7686_equation_0, values = (var_7518_cast, var_7659_cast))[name = tensor<string, []>("op_7686_cast")];
            tensor<string, []> var_7688_equation_0 = const()[name = tensor<string, []>("op_7688_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7688_cast = einsum(equation = var_7688_equation_0, values = (var_7522_cast, var_7660_cast))[name = tensor<string, []>("op_7688_cast")];
            tensor<string, []> var_7690_equation_0 = const()[name = tensor<string, []>("op_7690_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7690_cast = einsum(equation = var_7690_equation_0, values = (var_7526_cast, var_7661_cast))[name = tensor<string, []>("op_7690_cast")];
            tensor<string, []> var_7692_equation_0 = const()[name = tensor<string, []>("op_7692_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7692_cast = einsum(equation = var_7692_equation_0, values = (var_7530_cast, var_7662_cast))[name = tensor<string, []>("op_7692_cast")];
            tensor<string, []> var_7694_equation_0 = const()[name = tensor<string, []>("op_7694_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7694_cast = einsum(equation = var_7694_equation_0, values = (var_7534_cast, var_7663_cast))[name = tensor<string, []>("op_7694_cast")];
            tensor<string, []> var_7696_equation_0 = const()[name = tensor<string, []>("op_7696_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7696_cast = einsum(equation = var_7696_equation_0, values = (var_7538_cast, var_7664_cast))[name = tensor<string, []>("op_7696_cast")];
            tensor<string, []> var_7698_equation_0 = const()[name = tensor<string, []>("op_7698_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7698_cast = einsum(equation = var_7698_equation_0, values = (var_7542_cast, var_7665_cast))[name = tensor<string, []>("op_7698_cast")];
            tensor<string, []> var_7700_equation_0 = const()[name = tensor<string, []>("op_7700_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7700_cast = einsum(equation = var_7700_equation_0, values = (var_7546_cast, var_7666_cast))[name = tensor<string, []>("op_7700_cast")];
            tensor<string, []> var_7702_equation_0 = const()[name = tensor<string, []>("op_7702_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7702_cast = einsum(equation = var_7702_equation_0, values = (var_7550_cast, var_7667_cast))[name = tensor<string, []>("op_7702_cast")];
            tensor<string, []> var_7704_equation_0 = const()[name = tensor<string, []>("op_7704_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7704_cast = einsum(equation = var_7704_equation_0, values = (var_7554_cast, var_7668_cast))[name = tensor<string, []>("op_7704_cast")];
            tensor<string, []> var_7706_equation_0 = const()[name = tensor<string, []>("op_7706_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7706_cast = einsum(equation = var_7706_equation_0, values = (var_7558_cast, var_7669_cast))[name = tensor<string, []>("op_7706_cast")];
            tensor<string, []> var_7708_equation_0 = const()[name = tensor<string, []>("op_7708_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7708_cast = einsum(equation = var_7708_equation_0, values = (var_7562_cast, var_7670_cast))[name = tensor<string, []>("op_7708_cast")];
            tensor<string, []> var_7710_equation_0 = const()[name = tensor<string, []>("op_7710_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7710_cast = einsum(equation = var_7710_equation_0, values = (var_7566_cast, var_7671_cast))[name = tensor<string, []>("op_7710_cast")];
            tensor<string, []> var_7712_equation_0 = const()[name = tensor<string, []>("op_7712_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_7712_cast = einsum(equation = var_7712_equation_0, values = (var_7570_cast, var_7672_cast))[name = tensor<string, []>("op_7712_cast")];
            tensor<bool, []> input_175_interleave_0 = const()[name = tensor<string, []>("input_175_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_175_cast = concat(axis = var_2634, interleave = input_175_interleave_0, values = (var_7674_cast, var_7676_cast, var_7678_cast, var_7680_cast, var_7682_cast, var_7684_cast, var_7686_cast, var_7688_cast, var_7690_cast, var_7692_cast, var_7694_cast, var_7696_cast, var_7698_cast, var_7700_cast, var_7702_cast, var_7704_cast, var_7706_cast, var_7708_cast, var_7710_cast, var_7712_cast))[name = tensor<string, []>("input_175_cast")];
            tensor<int32, [2]> var_7718 = const()[name = tensor<string, []>("op_7718"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7720 = const()[name = tensor<string, []>("op_7720"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_7722_pad_type_0 = const()[name = tensor<string, []>("op_7722_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_7722_pad_0 = const()[name = tensor<string, []>("op_7722_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(555566528)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(558843392)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_7722_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16, dilations = var_7720, groups = var_2634, pad = var_7722_pad_0, pad_type = var_7722_pad_type_0, strides = var_7718, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16, x = input_175_cast)[name = tensor<string, []>("op_7722_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_57_cast = add(x = var_7722_cast, y = inputs_55_cast)[name = tensor<string, []>("inputs_57_cast")];
            tensor<int32, [1]> var_7726 = const()[name = tensor<string, []>("op_7726"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_57_cast = reduce_mean(axes = var_7726, keep_dims = var_2629, x = inputs_57_cast)[name = tensor<string, []>("channels_mean_57_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_57_cast = sub(x = inputs_57_cast, y = channels_mean_57_cast)[name = tensor<string, []>("zero_mean_57_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_57_cast = mul(x = zero_mean_57_cast, y = zero_mean_57_cast)[name = tensor<string, []>("zero_mean_sq_57_cast")];
            tensor<int32, [1]> var_7730 = const()[name = tensor<string, []>("op_7730"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_7731_cast = reduce_mean(axes = var_7730, keep_dims = var_2629, x = zero_mean_sq_57_cast)[name = tensor<string, []>("op_7731_cast")];
            tensor<fp16, []> var_7732_to_fp16 = const()[name = tensor<string, []>("op_7732_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_7733_cast = add(x = var_7731_cast, y = var_7732_to_fp16)[name = tensor<string, []>("op_7733_cast")];
            tensor<fp16, []> denom_57_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_57_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_57_cast = rsqrt(epsilon = denom_57_epsilon_0_to_fp16, x = var_7733_cast)[name = tensor<string, []>("denom_57_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_57_cast = mul(x = zero_mean_57_cast, y = denom_57_cast)[name = tensor<string, []>("out_57_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7737_to_fp16 = const()[name = tensor<string, []>("op_7737_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(558846016)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_7738_cast = add(x = out_57_cast, y = var_7737_to_fp16)[name = tensor<string, []>("op_7738_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7740_to_fp16 = const()[name = tensor<string, []>("op_7740_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(558848640)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_97_cast = mul(x = var_7738_cast, y = var_7740_to_fp16)[name = tensor<string, []>("hidden_states_97_cast")];
            tensor<int32, [2]> var_7747 = const()[name = tensor<string, []>("op_7747"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7749 = const()[name = tensor<string, []>("op_7749"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_39_pad_type_0 = const()[name = tensor<string, []>("q_39_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_39_pad_0 = const()[name = tensor<string, []>("q_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(558851264)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_39_cast = conv(dilations = var_7749, groups = var_2634, pad = q_39_pad_0, pad_type = q_39_pad_type_0, strides = var_7747, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16, x = hidden_states_97_cast)[name = tensor<string, []>("q_39_cast")];
            tensor<int32, [2]> var_7753 = const()[name = tensor<string, []>("op_7753"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7755 = const()[name = tensor<string, []>("op_7755"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_77_pad_type_0 = const()[name = tensor<string, []>("k_77_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_77_pad_0 = const()[name = tensor<string, []>("k_77_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(562128128)))];
            tensor<fp16, [2, 1280, 1, 77]> k_77_cast = conv(dilations = var_7755, groups = var_2634, pad = k_77_pad_0, pad_type = k_77_pad_type_0, strides = var_7753, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_77_cast")];
            tensor<int32, [2]> var_7759 = const()[name = tensor<string, []>("op_7759"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7761 = const()[name = tensor<string, []>("op_7761"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_39_pad_type_0 = const()[name = tensor<string, []>("v_39_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_39_pad_0 = const()[name = tensor<string, []>("v_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(567371072)))];
            tensor<fp16, [2, 1280, 1, 77]> v_39_cast = conv(dilations = var_7761, groups = var_2634, pad = v_39_pad_0, pad_type = v_39_pad_type_0, strides = var_7759, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_39_cast")];
            tensor<int32, [4]> var_7765_begin_0 = const()[name = tensor<string, []>("op_7765_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7765_end_0 = const()[name = tensor<string, []>("op_7765_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_7765_end_mask_0 = const()[name = tensor<string, []>("op_7765_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7765_cast = slice_by_index(begin = var_7765_begin_0, end = var_7765_end_0, end_mask = var_7765_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7765_cast")];
            tensor<int32, [4]> var_7769_begin_0 = const()[name = tensor<string, []>("op_7769_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_7769_end_0 = const()[name = tensor<string, []>("op_7769_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_7769_end_mask_0 = const()[name = tensor<string, []>("op_7769_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7769_cast = slice_by_index(begin = var_7769_begin_0, end = var_7769_end_0, end_mask = var_7769_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7769_cast")];
            tensor<int32, [4]> var_7773_begin_0 = const()[name = tensor<string, []>("op_7773_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_7773_end_0 = const()[name = tensor<string, []>("op_7773_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_7773_end_mask_0 = const()[name = tensor<string, []>("op_7773_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7773_cast = slice_by_index(begin = var_7773_begin_0, end = var_7773_end_0, end_mask = var_7773_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7773_cast")];
            tensor<int32, [4]> var_7777_begin_0 = const()[name = tensor<string, []>("op_7777_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_7777_end_0 = const()[name = tensor<string, []>("op_7777_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_7777_end_mask_0 = const()[name = tensor<string, []>("op_7777_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7777_cast = slice_by_index(begin = var_7777_begin_0, end = var_7777_end_0, end_mask = var_7777_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7777_cast")];
            tensor<int32, [4]> var_7781_begin_0 = const()[name = tensor<string, []>("op_7781_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_7781_end_0 = const()[name = tensor<string, []>("op_7781_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_7781_end_mask_0 = const()[name = tensor<string, []>("op_7781_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7781_cast = slice_by_index(begin = var_7781_begin_0, end = var_7781_end_0, end_mask = var_7781_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7781_cast")];
            tensor<int32, [4]> var_7785_begin_0 = const()[name = tensor<string, []>("op_7785_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7785_end_0 = const()[name = tensor<string, []>("op_7785_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_7785_end_mask_0 = const()[name = tensor<string, []>("op_7785_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7785_cast = slice_by_index(begin = var_7785_begin_0, end = var_7785_end_0, end_mask = var_7785_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7785_cast")];
            tensor<int32, [4]> var_7789_begin_0 = const()[name = tensor<string, []>("op_7789_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_7789_end_0 = const()[name = tensor<string, []>("op_7789_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_7789_end_mask_0 = const()[name = tensor<string, []>("op_7789_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7789_cast = slice_by_index(begin = var_7789_begin_0, end = var_7789_end_0, end_mask = var_7789_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7789_cast")];
            tensor<int32, [4]> var_7793_begin_0 = const()[name = tensor<string, []>("op_7793_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_7793_end_0 = const()[name = tensor<string, []>("op_7793_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_7793_end_mask_0 = const()[name = tensor<string, []>("op_7793_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7793_cast = slice_by_index(begin = var_7793_begin_0, end = var_7793_end_0, end_mask = var_7793_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7793_cast")];
            tensor<int32, [4]> var_7797_begin_0 = const()[name = tensor<string, []>("op_7797_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_7797_end_0 = const()[name = tensor<string, []>("op_7797_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_7797_end_mask_0 = const()[name = tensor<string, []>("op_7797_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7797_cast = slice_by_index(begin = var_7797_begin_0, end = var_7797_end_0, end_mask = var_7797_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7797_cast")];
            tensor<int32, [4]> var_7801_begin_0 = const()[name = tensor<string, []>("op_7801_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_7801_end_0 = const()[name = tensor<string, []>("op_7801_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_7801_end_mask_0 = const()[name = tensor<string, []>("op_7801_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7801_cast = slice_by_index(begin = var_7801_begin_0, end = var_7801_end_0, end_mask = var_7801_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7801_cast")];
            tensor<int32, [4]> var_7805_begin_0 = const()[name = tensor<string, []>("op_7805_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7805_end_0 = const()[name = tensor<string, []>("op_7805_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_7805_end_mask_0 = const()[name = tensor<string, []>("op_7805_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7805_cast = slice_by_index(begin = var_7805_begin_0, end = var_7805_end_0, end_mask = var_7805_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7805_cast")];
            tensor<int32, [4]> var_7809_begin_0 = const()[name = tensor<string, []>("op_7809_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_7809_end_0 = const()[name = tensor<string, []>("op_7809_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_7809_end_mask_0 = const()[name = tensor<string, []>("op_7809_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7809_cast = slice_by_index(begin = var_7809_begin_0, end = var_7809_end_0, end_mask = var_7809_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7809_cast")];
            tensor<int32, [4]> var_7813_begin_0 = const()[name = tensor<string, []>("op_7813_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_7813_end_0 = const()[name = tensor<string, []>("op_7813_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_7813_end_mask_0 = const()[name = tensor<string, []>("op_7813_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7813_cast = slice_by_index(begin = var_7813_begin_0, end = var_7813_end_0, end_mask = var_7813_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7813_cast")];
            tensor<int32, [4]> var_7817_begin_0 = const()[name = tensor<string, []>("op_7817_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_7817_end_0 = const()[name = tensor<string, []>("op_7817_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_7817_end_mask_0 = const()[name = tensor<string, []>("op_7817_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7817_cast = slice_by_index(begin = var_7817_begin_0, end = var_7817_end_0, end_mask = var_7817_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7817_cast")];
            tensor<int32, [4]> var_7821_begin_0 = const()[name = tensor<string, []>("op_7821_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_7821_end_0 = const()[name = tensor<string, []>("op_7821_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_7821_end_mask_0 = const()[name = tensor<string, []>("op_7821_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7821_cast = slice_by_index(begin = var_7821_begin_0, end = var_7821_end_0, end_mask = var_7821_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7821_cast")];
            tensor<int32, [4]> var_7825_begin_0 = const()[name = tensor<string, []>("op_7825_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_7825_end_0 = const()[name = tensor<string, []>("op_7825_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_7825_end_mask_0 = const()[name = tensor<string, []>("op_7825_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7825_cast = slice_by_index(begin = var_7825_begin_0, end = var_7825_end_0, end_mask = var_7825_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7825_cast")];
            tensor<int32, [4]> var_7829_begin_0 = const()[name = tensor<string, []>("op_7829_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_7829_end_0 = const()[name = tensor<string, []>("op_7829_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_7829_end_mask_0 = const()[name = tensor<string, []>("op_7829_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7829_cast = slice_by_index(begin = var_7829_begin_0, end = var_7829_end_0, end_mask = var_7829_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7829_cast")];
            tensor<int32, [4]> var_7833_begin_0 = const()[name = tensor<string, []>("op_7833_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_7833_end_0 = const()[name = tensor<string, []>("op_7833_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_7833_end_mask_0 = const()[name = tensor<string, []>("op_7833_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7833_cast = slice_by_index(begin = var_7833_begin_0, end = var_7833_end_0, end_mask = var_7833_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7833_cast")];
            tensor<int32, [4]> var_7837_begin_0 = const()[name = tensor<string, []>("op_7837_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_7837_end_0 = const()[name = tensor<string, []>("op_7837_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_7837_end_mask_0 = const()[name = tensor<string, []>("op_7837_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7837_cast = slice_by_index(begin = var_7837_begin_0, end = var_7837_end_0, end_mask = var_7837_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7837_cast")];
            tensor<int32, [4]> var_7841_begin_0 = const()[name = tensor<string, []>("op_7841_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_7841_end_0 = const()[name = tensor<string, []>("op_7841_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_7841_end_mask_0 = const()[name = tensor<string, []>("op_7841_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_7841_cast = slice_by_index(begin = var_7841_begin_0, end = var_7841_end_0, end_mask = var_7841_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7841_cast")];
            tensor<int32, [4]> k_79_perm_0 = const()[name = tensor<string, []>("k_79_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_7848_begin_0 = const()[name = tensor<string, []>("op_7848_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7848_end_0 = const()[name = tensor<string, []>("op_7848_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_7848_end_mask_0 = const()[name = tensor<string, []>("op_7848_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_120 = transpose(perm = k_79_perm_0, x = k_77_cast)[name = tensor<string, []>("transpose_120")];
            tensor<fp16, [2, 77, 1, 64]> var_7848_cast = slice_by_index(begin = var_7848_begin_0, end = var_7848_end_0, end_mask = var_7848_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7848_cast")];
            tensor<int32, [4]> var_7852_begin_0 = const()[name = tensor<string, []>("op_7852_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_7852_end_0 = const()[name = tensor<string, []>("op_7852_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_7852_end_mask_0 = const()[name = tensor<string, []>("op_7852_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7852_cast = slice_by_index(begin = var_7852_begin_0, end = var_7852_end_0, end_mask = var_7852_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7852_cast")];
            tensor<int32, [4]> var_7856_begin_0 = const()[name = tensor<string, []>("op_7856_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_7856_end_0 = const()[name = tensor<string, []>("op_7856_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_7856_end_mask_0 = const()[name = tensor<string, []>("op_7856_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7856_cast = slice_by_index(begin = var_7856_begin_0, end = var_7856_end_0, end_mask = var_7856_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7856_cast")];
            tensor<int32, [4]> var_7860_begin_0 = const()[name = tensor<string, []>("op_7860_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_7860_end_0 = const()[name = tensor<string, []>("op_7860_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_7860_end_mask_0 = const()[name = tensor<string, []>("op_7860_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7860_cast = slice_by_index(begin = var_7860_begin_0, end = var_7860_end_0, end_mask = var_7860_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7860_cast")];
            tensor<int32, [4]> var_7864_begin_0 = const()[name = tensor<string, []>("op_7864_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_7864_end_0 = const()[name = tensor<string, []>("op_7864_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_7864_end_mask_0 = const()[name = tensor<string, []>("op_7864_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7864_cast = slice_by_index(begin = var_7864_begin_0, end = var_7864_end_0, end_mask = var_7864_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7864_cast")];
            tensor<int32, [4]> var_7868_begin_0 = const()[name = tensor<string, []>("op_7868_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_7868_end_0 = const()[name = tensor<string, []>("op_7868_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_7868_end_mask_0 = const()[name = tensor<string, []>("op_7868_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7868_cast = slice_by_index(begin = var_7868_begin_0, end = var_7868_end_0, end_mask = var_7868_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7868_cast")];
            tensor<int32, [4]> var_7872_begin_0 = const()[name = tensor<string, []>("op_7872_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_7872_end_0 = const()[name = tensor<string, []>("op_7872_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_7872_end_mask_0 = const()[name = tensor<string, []>("op_7872_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7872_cast = slice_by_index(begin = var_7872_begin_0, end = var_7872_end_0, end_mask = var_7872_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7872_cast")];
            tensor<int32, [4]> var_7876_begin_0 = const()[name = tensor<string, []>("op_7876_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_7876_end_0 = const()[name = tensor<string, []>("op_7876_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_7876_end_mask_0 = const()[name = tensor<string, []>("op_7876_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7876_cast = slice_by_index(begin = var_7876_begin_0, end = var_7876_end_0, end_mask = var_7876_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7876_cast")];
            tensor<int32, [4]> var_7880_begin_0 = const()[name = tensor<string, []>("op_7880_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_7880_end_0 = const()[name = tensor<string, []>("op_7880_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_7880_end_mask_0 = const()[name = tensor<string, []>("op_7880_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7880_cast = slice_by_index(begin = var_7880_begin_0, end = var_7880_end_0, end_mask = var_7880_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7880_cast")];
            tensor<int32, [4]> var_7884_begin_0 = const()[name = tensor<string, []>("op_7884_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_7884_end_0 = const()[name = tensor<string, []>("op_7884_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_7884_end_mask_0 = const()[name = tensor<string, []>("op_7884_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7884_cast = slice_by_index(begin = var_7884_begin_0, end = var_7884_end_0, end_mask = var_7884_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7884_cast")];
            tensor<int32, [4]> var_7888_begin_0 = const()[name = tensor<string, []>("op_7888_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_7888_end_0 = const()[name = tensor<string, []>("op_7888_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_7888_end_mask_0 = const()[name = tensor<string, []>("op_7888_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7888_cast = slice_by_index(begin = var_7888_begin_0, end = var_7888_end_0, end_mask = var_7888_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7888_cast")];
            tensor<int32, [4]> var_7892_begin_0 = const()[name = tensor<string, []>("op_7892_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_7892_end_0 = const()[name = tensor<string, []>("op_7892_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_7892_end_mask_0 = const()[name = tensor<string, []>("op_7892_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7892_cast = slice_by_index(begin = var_7892_begin_0, end = var_7892_end_0, end_mask = var_7892_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7892_cast")];
            tensor<int32, [4]> var_7896_begin_0 = const()[name = tensor<string, []>("op_7896_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_7896_end_0 = const()[name = tensor<string, []>("op_7896_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_7896_end_mask_0 = const()[name = tensor<string, []>("op_7896_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7896_cast = slice_by_index(begin = var_7896_begin_0, end = var_7896_end_0, end_mask = var_7896_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7896_cast")];
            tensor<int32, [4]> var_7900_begin_0 = const()[name = tensor<string, []>("op_7900_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_7900_end_0 = const()[name = tensor<string, []>("op_7900_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_7900_end_mask_0 = const()[name = tensor<string, []>("op_7900_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7900_cast = slice_by_index(begin = var_7900_begin_0, end = var_7900_end_0, end_mask = var_7900_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7900_cast")];
            tensor<int32, [4]> var_7904_begin_0 = const()[name = tensor<string, []>("op_7904_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_7904_end_0 = const()[name = tensor<string, []>("op_7904_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_7904_end_mask_0 = const()[name = tensor<string, []>("op_7904_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7904_cast = slice_by_index(begin = var_7904_begin_0, end = var_7904_end_0, end_mask = var_7904_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7904_cast")];
            tensor<int32, [4]> var_7908_begin_0 = const()[name = tensor<string, []>("op_7908_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_7908_end_0 = const()[name = tensor<string, []>("op_7908_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_7908_end_mask_0 = const()[name = tensor<string, []>("op_7908_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7908_cast = slice_by_index(begin = var_7908_begin_0, end = var_7908_end_0, end_mask = var_7908_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7908_cast")];
            tensor<int32, [4]> var_7912_begin_0 = const()[name = tensor<string, []>("op_7912_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_7912_end_0 = const()[name = tensor<string, []>("op_7912_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_7912_end_mask_0 = const()[name = tensor<string, []>("op_7912_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7912_cast = slice_by_index(begin = var_7912_begin_0, end = var_7912_end_0, end_mask = var_7912_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7912_cast")];
            tensor<int32, [4]> var_7916_begin_0 = const()[name = tensor<string, []>("op_7916_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_7916_end_0 = const()[name = tensor<string, []>("op_7916_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_7916_end_mask_0 = const()[name = tensor<string, []>("op_7916_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7916_cast = slice_by_index(begin = var_7916_begin_0, end = var_7916_end_0, end_mask = var_7916_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7916_cast")];
            tensor<int32, [4]> var_7920_begin_0 = const()[name = tensor<string, []>("op_7920_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_7920_end_0 = const()[name = tensor<string, []>("op_7920_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_7920_end_mask_0 = const()[name = tensor<string, []>("op_7920_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7920_cast = slice_by_index(begin = var_7920_begin_0, end = var_7920_end_0, end_mask = var_7920_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7920_cast")];
            tensor<int32, [4]> var_7924_begin_0 = const()[name = tensor<string, []>("op_7924_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_7924_end_0 = const()[name = tensor<string, []>("op_7924_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_7924_end_mask_0 = const()[name = tensor<string, []>("op_7924_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7924_cast = slice_by_index(begin = var_7924_begin_0, end = var_7924_end_0, end_mask = var_7924_end_mask_0, x = transpose_120)[name = tensor<string, []>("op_7924_cast")];
            tensor<int32, [4]> var_7926_begin_0 = const()[name = tensor<string, []>("op_7926_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7926_end_0 = const()[name = tensor<string, []>("op_7926_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_7926_end_mask_0 = const()[name = tensor<string, []>("op_7926_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7926_cast = slice_by_index(begin = var_7926_begin_0, end = var_7926_end_0, end_mask = var_7926_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7926_cast")];
            tensor<int32, [4]> var_7930_begin_0 = const()[name = tensor<string, []>("op_7930_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_7930_end_0 = const()[name = tensor<string, []>("op_7930_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_7930_end_mask_0 = const()[name = tensor<string, []>("op_7930_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7930_cast = slice_by_index(begin = var_7930_begin_0, end = var_7930_end_0, end_mask = var_7930_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7930_cast")];
            tensor<int32, [4]> var_7934_begin_0 = const()[name = tensor<string, []>("op_7934_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_7934_end_0 = const()[name = tensor<string, []>("op_7934_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_7934_end_mask_0 = const()[name = tensor<string, []>("op_7934_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7934_cast = slice_by_index(begin = var_7934_begin_0, end = var_7934_end_0, end_mask = var_7934_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7934_cast")];
            tensor<int32, [4]> var_7938_begin_0 = const()[name = tensor<string, []>("op_7938_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_7938_end_0 = const()[name = tensor<string, []>("op_7938_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_7938_end_mask_0 = const()[name = tensor<string, []>("op_7938_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7938_cast = slice_by_index(begin = var_7938_begin_0, end = var_7938_end_0, end_mask = var_7938_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7938_cast")];
            tensor<int32, [4]> var_7942_begin_0 = const()[name = tensor<string, []>("op_7942_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_7942_end_0 = const()[name = tensor<string, []>("op_7942_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_7942_end_mask_0 = const()[name = tensor<string, []>("op_7942_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7942_cast = slice_by_index(begin = var_7942_begin_0, end = var_7942_end_0, end_mask = var_7942_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7942_cast")];
            tensor<int32, [4]> var_7946_begin_0 = const()[name = tensor<string, []>("op_7946_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7946_end_0 = const()[name = tensor<string, []>("op_7946_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_7946_end_mask_0 = const()[name = tensor<string, []>("op_7946_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7946_cast = slice_by_index(begin = var_7946_begin_0, end = var_7946_end_0, end_mask = var_7946_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7946_cast")];
            tensor<int32, [4]> var_7950_begin_0 = const()[name = tensor<string, []>("op_7950_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_7950_end_0 = const()[name = tensor<string, []>("op_7950_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_7950_end_mask_0 = const()[name = tensor<string, []>("op_7950_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7950_cast = slice_by_index(begin = var_7950_begin_0, end = var_7950_end_0, end_mask = var_7950_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7950_cast")];
            tensor<int32, [4]> var_7954_begin_0 = const()[name = tensor<string, []>("op_7954_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_7954_end_0 = const()[name = tensor<string, []>("op_7954_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_7954_end_mask_0 = const()[name = tensor<string, []>("op_7954_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7954_cast = slice_by_index(begin = var_7954_begin_0, end = var_7954_end_0, end_mask = var_7954_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7954_cast")];
            tensor<int32, [4]> var_7958_begin_0 = const()[name = tensor<string, []>("op_7958_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_7958_end_0 = const()[name = tensor<string, []>("op_7958_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_7958_end_mask_0 = const()[name = tensor<string, []>("op_7958_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7958_cast = slice_by_index(begin = var_7958_begin_0, end = var_7958_end_0, end_mask = var_7958_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7958_cast")];
            tensor<int32, [4]> var_7962_begin_0 = const()[name = tensor<string, []>("op_7962_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_7962_end_0 = const()[name = tensor<string, []>("op_7962_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_7962_end_mask_0 = const()[name = tensor<string, []>("op_7962_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7962_cast = slice_by_index(begin = var_7962_begin_0, end = var_7962_end_0, end_mask = var_7962_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7962_cast")];
            tensor<int32, [4]> var_7966_begin_0 = const()[name = tensor<string, []>("op_7966_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7966_end_0 = const()[name = tensor<string, []>("op_7966_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_7966_end_mask_0 = const()[name = tensor<string, []>("op_7966_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7966_cast = slice_by_index(begin = var_7966_begin_0, end = var_7966_end_0, end_mask = var_7966_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7966_cast")];
            tensor<int32, [4]> var_7970_begin_0 = const()[name = tensor<string, []>("op_7970_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_7970_end_0 = const()[name = tensor<string, []>("op_7970_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_7970_end_mask_0 = const()[name = tensor<string, []>("op_7970_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7970_cast = slice_by_index(begin = var_7970_begin_0, end = var_7970_end_0, end_mask = var_7970_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7970_cast")];
            tensor<int32, [4]> var_7974_begin_0 = const()[name = tensor<string, []>("op_7974_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_7974_end_0 = const()[name = tensor<string, []>("op_7974_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_7974_end_mask_0 = const()[name = tensor<string, []>("op_7974_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7974_cast = slice_by_index(begin = var_7974_begin_0, end = var_7974_end_0, end_mask = var_7974_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7974_cast")];
            tensor<int32, [4]> var_7978_begin_0 = const()[name = tensor<string, []>("op_7978_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_7978_end_0 = const()[name = tensor<string, []>("op_7978_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_7978_end_mask_0 = const()[name = tensor<string, []>("op_7978_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7978_cast = slice_by_index(begin = var_7978_begin_0, end = var_7978_end_0, end_mask = var_7978_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7978_cast")];
            tensor<int32, [4]> var_7982_begin_0 = const()[name = tensor<string, []>("op_7982_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_7982_end_0 = const()[name = tensor<string, []>("op_7982_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_7982_end_mask_0 = const()[name = tensor<string, []>("op_7982_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7982_cast = slice_by_index(begin = var_7982_begin_0, end = var_7982_end_0, end_mask = var_7982_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7982_cast")];
            tensor<int32, [4]> var_7986_begin_0 = const()[name = tensor<string, []>("op_7986_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_7986_end_0 = const()[name = tensor<string, []>("op_7986_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_7986_end_mask_0 = const()[name = tensor<string, []>("op_7986_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7986_cast = slice_by_index(begin = var_7986_begin_0, end = var_7986_end_0, end_mask = var_7986_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7986_cast")];
            tensor<int32, [4]> var_7990_begin_0 = const()[name = tensor<string, []>("op_7990_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_7990_end_0 = const()[name = tensor<string, []>("op_7990_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_7990_end_mask_0 = const()[name = tensor<string, []>("op_7990_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7990_cast = slice_by_index(begin = var_7990_begin_0, end = var_7990_end_0, end_mask = var_7990_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7990_cast")];
            tensor<int32, [4]> var_7994_begin_0 = const()[name = tensor<string, []>("op_7994_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_7994_end_0 = const()[name = tensor<string, []>("op_7994_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_7994_end_mask_0 = const()[name = tensor<string, []>("op_7994_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7994_cast = slice_by_index(begin = var_7994_begin_0, end = var_7994_end_0, end_mask = var_7994_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7994_cast")];
            tensor<int32, [4]> var_7998_begin_0 = const()[name = tensor<string, []>("op_7998_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_7998_end_0 = const()[name = tensor<string, []>("op_7998_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_7998_end_mask_0 = const()[name = tensor<string, []>("op_7998_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7998_cast = slice_by_index(begin = var_7998_begin_0, end = var_7998_end_0, end_mask = var_7998_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7998_cast")];
            tensor<int32, [4]> var_8002_begin_0 = const()[name = tensor<string, []>("op_8002_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_8002_end_0 = const()[name = tensor<string, []>("op_8002_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_8002_end_mask_0 = const()[name = tensor<string, []>("op_8002_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8002_cast = slice_by_index(begin = var_8002_begin_0, end = var_8002_end_0, end_mask = var_8002_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_8002_cast")];
            tensor<string, []> var_8006_equation_0 = const()[name = tensor<string, []>("op_8006_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8006_cast = einsum(equation = var_8006_equation_0, values = (var_7848_cast, var_7765_cast))[name = tensor<string, []>("op_8006_cast")];
            tensor<fp16, []> var_8007_to_fp16 = const()[name = tensor<string, []>("op_8007_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_601_cast = mul(x = var_8006_cast, y = var_8007_to_fp16)[name = tensor<string, []>("aw_601_cast")];
            tensor<string, []> var_8010_equation_0 = const()[name = tensor<string, []>("op_8010_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8010_cast = einsum(equation = var_8010_equation_0, values = (var_7852_cast, var_7769_cast))[name = tensor<string, []>("op_8010_cast")];
            tensor<fp16, []> var_8011_to_fp16 = const()[name = tensor<string, []>("op_8011_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_603_cast = mul(x = var_8010_cast, y = var_8011_to_fp16)[name = tensor<string, []>("aw_603_cast")];
            tensor<string, []> var_8014_equation_0 = const()[name = tensor<string, []>("op_8014_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8014_cast = einsum(equation = var_8014_equation_0, values = (var_7856_cast, var_7773_cast))[name = tensor<string, []>("op_8014_cast")];
            tensor<fp16, []> var_8015_to_fp16 = const()[name = tensor<string, []>("op_8015_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_605_cast = mul(x = var_8014_cast, y = var_8015_to_fp16)[name = tensor<string, []>("aw_605_cast")];
            tensor<string, []> var_8018_equation_0 = const()[name = tensor<string, []>("op_8018_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8018_cast = einsum(equation = var_8018_equation_0, values = (var_7860_cast, var_7777_cast))[name = tensor<string, []>("op_8018_cast")];
            tensor<fp16, []> var_8019_to_fp16 = const()[name = tensor<string, []>("op_8019_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_607_cast = mul(x = var_8018_cast, y = var_8019_to_fp16)[name = tensor<string, []>("aw_607_cast")];
            tensor<string, []> var_8022_equation_0 = const()[name = tensor<string, []>("op_8022_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8022_cast = einsum(equation = var_8022_equation_0, values = (var_7864_cast, var_7781_cast))[name = tensor<string, []>("op_8022_cast")];
            tensor<fp16, []> var_8023_to_fp16 = const()[name = tensor<string, []>("op_8023_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_609_cast = mul(x = var_8022_cast, y = var_8023_to_fp16)[name = tensor<string, []>("aw_609_cast")];
            tensor<string, []> var_8026_equation_0 = const()[name = tensor<string, []>("op_8026_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8026_cast = einsum(equation = var_8026_equation_0, values = (var_7868_cast, var_7785_cast))[name = tensor<string, []>("op_8026_cast")];
            tensor<fp16, []> var_8027_to_fp16 = const()[name = tensor<string, []>("op_8027_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_611_cast = mul(x = var_8026_cast, y = var_8027_to_fp16)[name = tensor<string, []>("aw_611_cast")];
            tensor<string, []> var_8030_equation_0 = const()[name = tensor<string, []>("op_8030_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8030_cast = einsum(equation = var_8030_equation_0, values = (var_7872_cast, var_7789_cast))[name = tensor<string, []>("op_8030_cast")];
            tensor<fp16, []> var_8031_to_fp16 = const()[name = tensor<string, []>("op_8031_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_613_cast = mul(x = var_8030_cast, y = var_8031_to_fp16)[name = tensor<string, []>("aw_613_cast")];
            tensor<string, []> var_8034_equation_0 = const()[name = tensor<string, []>("op_8034_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8034_cast = einsum(equation = var_8034_equation_0, values = (var_7876_cast, var_7793_cast))[name = tensor<string, []>("op_8034_cast")];
            tensor<fp16, []> var_8035_to_fp16 = const()[name = tensor<string, []>("op_8035_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_615_cast = mul(x = var_8034_cast, y = var_8035_to_fp16)[name = tensor<string, []>("aw_615_cast")];
            tensor<string, []> var_8038_equation_0 = const()[name = tensor<string, []>("op_8038_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8038_cast = einsum(equation = var_8038_equation_0, values = (var_7880_cast, var_7797_cast))[name = tensor<string, []>("op_8038_cast")];
            tensor<fp16, []> var_8039_to_fp16 = const()[name = tensor<string, []>("op_8039_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_617_cast = mul(x = var_8038_cast, y = var_8039_to_fp16)[name = tensor<string, []>("aw_617_cast")];
            tensor<string, []> var_8042_equation_0 = const()[name = tensor<string, []>("op_8042_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8042_cast = einsum(equation = var_8042_equation_0, values = (var_7884_cast, var_7801_cast))[name = tensor<string, []>("op_8042_cast")];
            tensor<fp16, []> var_8043_to_fp16 = const()[name = tensor<string, []>("op_8043_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_619_cast = mul(x = var_8042_cast, y = var_8043_to_fp16)[name = tensor<string, []>("aw_619_cast")];
            tensor<string, []> var_8046_equation_0 = const()[name = tensor<string, []>("op_8046_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8046_cast = einsum(equation = var_8046_equation_0, values = (var_7888_cast, var_7805_cast))[name = tensor<string, []>("op_8046_cast")];
            tensor<fp16, []> var_8047_to_fp16 = const()[name = tensor<string, []>("op_8047_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_621_cast = mul(x = var_8046_cast, y = var_8047_to_fp16)[name = tensor<string, []>("aw_621_cast")];
            tensor<string, []> var_8050_equation_0 = const()[name = tensor<string, []>("op_8050_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8050_cast = einsum(equation = var_8050_equation_0, values = (var_7892_cast, var_7809_cast))[name = tensor<string, []>("op_8050_cast")];
            tensor<fp16, []> var_8051_to_fp16 = const()[name = tensor<string, []>("op_8051_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_623_cast = mul(x = var_8050_cast, y = var_8051_to_fp16)[name = tensor<string, []>("aw_623_cast")];
            tensor<string, []> var_8054_equation_0 = const()[name = tensor<string, []>("op_8054_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8054_cast = einsum(equation = var_8054_equation_0, values = (var_7896_cast, var_7813_cast))[name = tensor<string, []>("op_8054_cast")];
            tensor<fp16, []> var_8055_to_fp16 = const()[name = tensor<string, []>("op_8055_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_625_cast = mul(x = var_8054_cast, y = var_8055_to_fp16)[name = tensor<string, []>("aw_625_cast")];
            tensor<string, []> var_8058_equation_0 = const()[name = tensor<string, []>("op_8058_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8058_cast = einsum(equation = var_8058_equation_0, values = (var_7900_cast, var_7817_cast))[name = tensor<string, []>("op_8058_cast")];
            tensor<fp16, []> var_8059_to_fp16 = const()[name = tensor<string, []>("op_8059_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_627_cast = mul(x = var_8058_cast, y = var_8059_to_fp16)[name = tensor<string, []>("aw_627_cast")];
            tensor<string, []> var_8062_equation_0 = const()[name = tensor<string, []>("op_8062_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8062_cast = einsum(equation = var_8062_equation_0, values = (var_7904_cast, var_7821_cast))[name = tensor<string, []>("op_8062_cast")];
            tensor<fp16, []> var_8063_to_fp16 = const()[name = tensor<string, []>("op_8063_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_629_cast = mul(x = var_8062_cast, y = var_8063_to_fp16)[name = tensor<string, []>("aw_629_cast")];
            tensor<string, []> var_8066_equation_0 = const()[name = tensor<string, []>("op_8066_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8066_cast = einsum(equation = var_8066_equation_0, values = (var_7908_cast, var_7825_cast))[name = tensor<string, []>("op_8066_cast")];
            tensor<fp16, []> var_8067_to_fp16 = const()[name = tensor<string, []>("op_8067_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_631_cast = mul(x = var_8066_cast, y = var_8067_to_fp16)[name = tensor<string, []>("aw_631_cast")];
            tensor<string, []> var_8070_equation_0 = const()[name = tensor<string, []>("op_8070_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8070_cast = einsum(equation = var_8070_equation_0, values = (var_7912_cast, var_7829_cast))[name = tensor<string, []>("op_8070_cast")];
            tensor<fp16, []> var_8071_to_fp16 = const()[name = tensor<string, []>("op_8071_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_633_cast = mul(x = var_8070_cast, y = var_8071_to_fp16)[name = tensor<string, []>("aw_633_cast")];
            tensor<string, []> var_8074_equation_0 = const()[name = tensor<string, []>("op_8074_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8074_cast = einsum(equation = var_8074_equation_0, values = (var_7916_cast, var_7833_cast))[name = tensor<string, []>("op_8074_cast")];
            tensor<fp16, []> var_8075_to_fp16 = const()[name = tensor<string, []>("op_8075_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_635_cast = mul(x = var_8074_cast, y = var_8075_to_fp16)[name = tensor<string, []>("aw_635_cast")];
            tensor<string, []> var_8078_equation_0 = const()[name = tensor<string, []>("op_8078_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8078_cast = einsum(equation = var_8078_equation_0, values = (var_7920_cast, var_7837_cast))[name = tensor<string, []>("op_8078_cast")];
            tensor<fp16, []> var_8079_to_fp16 = const()[name = tensor<string, []>("op_8079_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_637_cast = mul(x = var_8078_cast, y = var_8079_to_fp16)[name = tensor<string, []>("aw_637_cast")];
            tensor<string, []> var_8082_equation_0 = const()[name = tensor<string, []>("op_8082_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8082_cast = einsum(equation = var_8082_equation_0, values = (var_7924_cast, var_7841_cast))[name = tensor<string, []>("op_8082_cast")];
            tensor<fp16, []> var_8083_to_fp16 = const()[name = tensor<string, []>("op_8083_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_639_cast = mul(x = var_8082_cast, y = var_8083_to_fp16)[name = tensor<string, []>("aw_639_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8085_cast = softmax(axis = var_2634, x = aw_601_cast)[name = tensor<string, []>("op_8085_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8086_cast = softmax(axis = var_2634, x = aw_603_cast)[name = tensor<string, []>("op_8086_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8087_cast = softmax(axis = var_2634, x = aw_605_cast)[name = tensor<string, []>("op_8087_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8088_cast = softmax(axis = var_2634, x = aw_607_cast)[name = tensor<string, []>("op_8088_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8089_cast = softmax(axis = var_2634, x = aw_609_cast)[name = tensor<string, []>("op_8089_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8090_cast = softmax(axis = var_2634, x = aw_611_cast)[name = tensor<string, []>("op_8090_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8091_cast = softmax(axis = var_2634, x = aw_613_cast)[name = tensor<string, []>("op_8091_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8092_cast = softmax(axis = var_2634, x = aw_615_cast)[name = tensor<string, []>("op_8092_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8093_cast = softmax(axis = var_2634, x = aw_617_cast)[name = tensor<string, []>("op_8093_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8094_cast = softmax(axis = var_2634, x = aw_619_cast)[name = tensor<string, []>("op_8094_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8095_cast = softmax(axis = var_2634, x = aw_621_cast)[name = tensor<string, []>("op_8095_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8096_cast = softmax(axis = var_2634, x = aw_623_cast)[name = tensor<string, []>("op_8096_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8097_cast = softmax(axis = var_2634, x = aw_625_cast)[name = tensor<string, []>("op_8097_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8098_cast = softmax(axis = var_2634, x = aw_627_cast)[name = tensor<string, []>("op_8098_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8099_cast = softmax(axis = var_2634, x = aw_629_cast)[name = tensor<string, []>("op_8099_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8100_cast = softmax(axis = var_2634, x = aw_631_cast)[name = tensor<string, []>("op_8100_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8101_cast = softmax(axis = var_2634, x = aw_633_cast)[name = tensor<string, []>("op_8101_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8102_cast = softmax(axis = var_2634, x = aw_635_cast)[name = tensor<string, []>("op_8102_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8103_cast = softmax(axis = var_2634, x = aw_637_cast)[name = tensor<string, []>("op_8103_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8104_cast = softmax(axis = var_2634, x = aw_639_cast)[name = tensor<string, []>("op_8104_cast")];
            tensor<string, []> var_8106_equation_0 = const()[name = tensor<string, []>("op_8106_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8106_cast = einsum(equation = var_8106_equation_0, values = (var_7926_cast, var_8085_cast))[name = tensor<string, []>("op_8106_cast")];
            tensor<string, []> var_8108_equation_0 = const()[name = tensor<string, []>("op_8108_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8108_cast = einsum(equation = var_8108_equation_0, values = (var_7930_cast, var_8086_cast))[name = tensor<string, []>("op_8108_cast")];
            tensor<string, []> var_8110_equation_0 = const()[name = tensor<string, []>("op_8110_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8110_cast = einsum(equation = var_8110_equation_0, values = (var_7934_cast, var_8087_cast))[name = tensor<string, []>("op_8110_cast")];
            tensor<string, []> var_8112_equation_0 = const()[name = tensor<string, []>("op_8112_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8112_cast = einsum(equation = var_8112_equation_0, values = (var_7938_cast, var_8088_cast))[name = tensor<string, []>("op_8112_cast")];
            tensor<string, []> var_8114_equation_0 = const()[name = tensor<string, []>("op_8114_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8114_cast = einsum(equation = var_8114_equation_0, values = (var_7942_cast, var_8089_cast))[name = tensor<string, []>("op_8114_cast")];
            tensor<string, []> var_8116_equation_0 = const()[name = tensor<string, []>("op_8116_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8116_cast = einsum(equation = var_8116_equation_0, values = (var_7946_cast, var_8090_cast))[name = tensor<string, []>("op_8116_cast")];
            tensor<string, []> var_8118_equation_0 = const()[name = tensor<string, []>("op_8118_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8118_cast = einsum(equation = var_8118_equation_0, values = (var_7950_cast, var_8091_cast))[name = tensor<string, []>("op_8118_cast")];
            tensor<string, []> var_8120_equation_0 = const()[name = tensor<string, []>("op_8120_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8120_cast = einsum(equation = var_8120_equation_0, values = (var_7954_cast, var_8092_cast))[name = tensor<string, []>("op_8120_cast")];
            tensor<string, []> var_8122_equation_0 = const()[name = tensor<string, []>("op_8122_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8122_cast = einsum(equation = var_8122_equation_0, values = (var_7958_cast, var_8093_cast))[name = tensor<string, []>("op_8122_cast")];
            tensor<string, []> var_8124_equation_0 = const()[name = tensor<string, []>("op_8124_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8124_cast = einsum(equation = var_8124_equation_0, values = (var_7962_cast, var_8094_cast))[name = tensor<string, []>("op_8124_cast")];
            tensor<string, []> var_8126_equation_0 = const()[name = tensor<string, []>("op_8126_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8126_cast = einsum(equation = var_8126_equation_0, values = (var_7966_cast, var_8095_cast))[name = tensor<string, []>("op_8126_cast")];
            tensor<string, []> var_8128_equation_0 = const()[name = tensor<string, []>("op_8128_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8128_cast = einsum(equation = var_8128_equation_0, values = (var_7970_cast, var_8096_cast))[name = tensor<string, []>("op_8128_cast")];
            tensor<string, []> var_8130_equation_0 = const()[name = tensor<string, []>("op_8130_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8130_cast = einsum(equation = var_8130_equation_0, values = (var_7974_cast, var_8097_cast))[name = tensor<string, []>("op_8130_cast")];
            tensor<string, []> var_8132_equation_0 = const()[name = tensor<string, []>("op_8132_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8132_cast = einsum(equation = var_8132_equation_0, values = (var_7978_cast, var_8098_cast))[name = tensor<string, []>("op_8132_cast")];
            tensor<string, []> var_8134_equation_0 = const()[name = tensor<string, []>("op_8134_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8134_cast = einsum(equation = var_8134_equation_0, values = (var_7982_cast, var_8099_cast))[name = tensor<string, []>("op_8134_cast")];
            tensor<string, []> var_8136_equation_0 = const()[name = tensor<string, []>("op_8136_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8136_cast = einsum(equation = var_8136_equation_0, values = (var_7986_cast, var_8100_cast))[name = tensor<string, []>("op_8136_cast")];
            tensor<string, []> var_8138_equation_0 = const()[name = tensor<string, []>("op_8138_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8138_cast = einsum(equation = var_8138_equation_0, values = (var_7990_cast, var_8101_cast))[name = tensor<string, []>("op_8138_cast")];
            tensor<string, []> var_8140_equation_0 = const()[name = tensor<string, []>("op_8140_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8140_cast = einsum(equation = var_8140_equation_0, values = (var_7994_cast, var_8102_cast))[name = tensor<string, []>("op_8140_cast")];
            tensor<string, []> var_8142_equation_0 = const()[name = tensor<string, []>("op_8142_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8142_cast = einsum(equation = var_8142_equation_0, values = (var_7998_cast, var_8103_cast))[name = tensor<string, []>("op_8142_cast")];
            tensor<string, []> var_8144_equation_0 = const()[name = tensor<string, []>("op_8144_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8144_cast = einsum(equation = var_8144_equation_0, values = (var_8002_cast, var_8104_cast))[name = tensor<string, []>("op_8144_cast")];
            tensor<bool, []> input_177_interleave_0 = const()[name = tensor<string, []>("input_177_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_177_cast = concat(axis = var_2634, interleave = input_177_interleave_0, values = (var_8106_cast, var_8108_cast, var_8110_cast, var_8112_cast, var_8114_cast, var_8116_cast, var_8118_cast, var_8120_cast, var_8122_cast, var_8124_cast, var_8126_cast, var_8128_cast, var_8130_cast, var_8132_cast, var_8134_cast, var_8136_cast, var_8138_cast, var_8140_cast, var_8142_cast, var_8144_cast))[name = tensor<string, []>("input_177_cast")];
            tensor<int32, [2]> var_8150 = const()[name = tensor<string, []>("op_8150"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8152 = const()[name = tensor<string, []>("op_8152"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8154_pad_type_0 = const()[name = tensor<string, []>("op_8154_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8154_pad_0 = const()[name = tensor<string, []>("op_8154_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572614016)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(575890880)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_8154_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16, dilations = var_8152, groups = var_2634, pad = var_8154_pad_0, pad_type = var_8154_pad_type_0, strides = var_8150, weight = down_blocks_2_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16, x = input_177_cast)[name = tensor<string, []>("op_8154_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_59_cast = add(x = var_8154_cast, y = inputs_57_cast)[name = tensor<string, []>("inputs_59_cast")];
            tensor<int32, [1]> var_8158 = const()[name = tensor<string, []>("op_8158"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_59_cast = reduce_mean(axes = var_8158, keep_dims = var_2629, x = inputs_59_cast)[name = tensor<string, []>("channels_mean_59_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_59_cast = sub(x = inputs_59_cast, y = channels_mean_59_cast)[name = tensor<string, []>("zero_mean_59_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_59_cast = mul(x = zero_mean_59_cast, y = zero_mean_59_cast)[name = tensor<string, []>("zero_mean_sq_59_cast")];
            tensor<int32, [1]> var_8162 = const()[name = tensor<string, []>("op_8162"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_8163_cast = reduce_mean(axes = var_8162, keep_dims = var_2629, x = zero_mean_sq_59_cast)[name = tensor<string, []>("op_8163_cast")];
            tensor<fp16, []> var_8164_to_fp16 = const()[name = tensor<string, []>("op_8164_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_8165_cast = add(x = var_8163_cast, y = var_8164_to_fp16)[name = tensor<string, []>("op_8165_cast")];
            tensor<fp16, []> denom_59_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_59_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_59_cast = rsqrt(epsilon = denom_59_epsilon_0_to_fp16, x = var_8165_cast)[name = tensor<string, []>("denom_59_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_59_cast = mul(x = zero_mean_59_cast, y = denom_59_cast)[name = tensor<string, []>("out_59_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_8169_to_fp16 = const()[name = tensor<string, []>("op_8169_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(575893504)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_8170_cast = add(x = out_59_cast, y = var_8169_to_fp16)[name = tensor<string, []>("op_8170_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_8172_to_fp16 = const()[name = tensor<string, []>("op_8172_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(575896128)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_179_cast = mul(x = var_8170_cast, y = var_8172_to_fp16)[name = tensor<string, []>("input_179_cast")];
            tensor<int32, [2]> var_8180 = const()[name = tensor<string, []>("op_8180"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8182 = const()[name = tensor<string, []>("op_8182"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8184_pad_type_0 = const()[name = tensor<string, []>("op_8184_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8184_pad_0 = const()[name = tensor<string, []>("op_8184_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(575898752)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(602113216)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_8184_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16, dilations = var_8182, groups = var_2634, pad = var_8184_pad_0, pad_type = var_8184_pad_type_0, strides = var_8180, weight = down_blocks_2_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16, x = input_179_cast)[name = tensor<string, []>("op_8184_cast")];
            tensor<int32, [2]> var_8185_split_sizes_0 = const()[name = tensor<string, []>("op_8185_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_8185_axis_0 = const()[name = tensor<string, []>("op_8185_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_8185_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_8185_cast_1 = split(axis = var_8185_axis_0, split_sizes = var_8185_split_sizes_0, x = var_8184_cast)[name = tensor<string, []>("op_8185_cast")];
            tensor<string, []> var_8187_mode_0 = const()[name = tensor<string, []>("op_8187_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_8187_cast = gelu(mode = var_8187_mode_0, x = var_8185_cast_1)[name = tensor<string, []>("op_8187_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_181_cast = mul(x = var_8185_cast_0, y = var_8187_cast)[name = tensor<string, []>("input_181_cast")];
            tensor<int32, [2]> var_8191 = const()[name = tensor<string, []>("op_8191"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8193 = const()[name = tensor<string, []>("op_8193"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8195_pad_type_0 = const()[name = tensor<string, []>("op_8195_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8195_pad_0 = const()[name = tensor<string, []>("op_8195_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(602133760)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(615241024)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_8195_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16, dilations = var_8193, groups = var_2634, pad = var_8195_pad_0, pad_type = var_8195_pad_type_0, strides = var_8191, weight = down_blocks_2_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16, x = input_181_cast)[name = tensor<string, []>("op_8195_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_61_cast = add(x = var_8195_cast, y = inputs_59_cast)[name = tensor<string, []>("inputs_61_cast")];
            tensor<int32, [1]> var_8205 = const()[name = tensor<string, []>("op_8205"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_61_cast = reduce_mean(axes = var_8205, keep_dims = var_2629, x = inputs_61_cast)[name = tensor<string, []>("channels_mean_61_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_61_cast = sub(x = inputs_61_cast, y = channels_mean_61_cast)[name = tensor<string, []>("zero_mean_61_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_61_cast = mul(x = zero_mean_61_cast, y = zero_mean_61_cast)[name = tensor<string, []>("zero_mean_sq_61_cast")];
            tensor<int32, [1]> var_8209 = const()[name = tensor<string, []>("op_8209"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_8210_cast = reduce_mean(axes = var_8209, keep_dims = var_2629, x = zero_mean_sq_61_cast)[name = tensor<string, []>("op_8210_cast")];
            tensor<fp16, []> var_8211_to_fp16 = const()[name = tensor<string, []>("op_8211_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_8212_cast = add(x = var_8210_cast, y = var_8211_to_fp16)[name = tensor<string, []>("op_8212_cast")];
            tensor<fp16, []> denom_61_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_61_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_61_cast = rsqrt(epsilon = denom_61_epsilon_0_to_fp16, x = var_8212_cast)[name = tensor<string, []>("denom_61_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_61_cast = mul(x = zero_mean_61_cast, y = denom_61_cast)[name = tensor<string, []>("out_61_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_8216_to_fp16 = const()[name = tensor<string, []>("op_8216_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(615243648)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_8217_cast = add(x = out_61_cast, y = var_8216_to_fp16)[name = tensor<string, []>("op_8217_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_8219_to_fp16 = const()[name = tensor<string, []>("op_8219_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(615246272)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_101_cast = mul(x = var_8217_cast, y = var_8219_to_fp16)[name = tensor<string, []>("hidden_states_101_cast")];
            tensor<int32, [2]> var_8226 = const()[name = tensor<string, []>("op_8226"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8228 = const()[name = tensor<string, []>("op_8228"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_41_pad_type_0 = const()[name = tensor<string, []>("q_41_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_41_pad_0 = const()[name = tensor<string, []>("q_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(615248896)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_41_cast = conv(dilations = var_8228, groups = var_2634, pad = q_41_pad_0, pad_type = q_41_pad_type_0, strides = var_8226, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16, x = hidden_states_101_cast)[name = tensor<string, []>("q_41_cast")];
            tensor<int32, [2]> var_8232 = const()[name = tensor<string, []>("op_8232"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8234 = const()[name = tensor<string, []>("op_8234"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_81_pad_type_0 = const()[name = tensor<string, []>("k_81_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_81_pad_0 = const()[name = tensor<string, []>("k_81_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(618525760)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_81_cast = conv(dilations = var_8234, groups = var_2634, pad = k_81_pad_0, pad_type = k_81_pad_type_0, strides = var_8232, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16, x = hidden_states_101_cast)[name = tensor<string, []>("k_81_cast")];
            tensor<int32, [2]> var_8238 = const()[name = tensor<string, []>("op_8238"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8240 = const()[name = tensor<string, []>("op_8240"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_41_pad_type_0 = const()[name = tensor<string, []>("v_41_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_41_pad_0 = const()[name = tensor<string, []>("v_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(621802624)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_41_cast = conv(dilations = var_8240, groups = var_2634, pad = v_41_pad_0, pad_type = v_41_pad_type_0, strides = var_8238, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16, x = hidden_states_101_cast)[name = tensor<string, []>("v_41_cast")];
            tensor<int32, [4]> var_8244_begin_0 = const()[name = tensor<string, []>("op_8244_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8244_end_0 = const()[name = tensor<string, []>("op_8244_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_8244_end_mask_0 = const()[name = tensor<string, []>("op_8244_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8244_cast = slice_by_index(begin = var_8244_begin_0, end = var_8244_end_0, end_mask = var_8244_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8244_cast")];
            tensor<int32, [4]> var_8248_begin_0 = const()[name = tensor<string, []>("op_8248_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_8248_end_0 = const()[name = tensor<string, []>("op_8248_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_8248_end_mask_0 = const()[name = tensor<string, []>("op_8248_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8248_cast = slice_by_index(begin = var_8248_begin_0, end = var_8248_end_0, end_mask = var_8248_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8248_cast")];
            tensor<int32, [4]> var_8252_begin_0 = const()[name = tensor<string, []>("op_8252_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_8252_end_0 = const()[name = tensor<string, []>("op_8252_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_8252_end_mask_0 = const()[name = tensor<string, []>("op_8252_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8252_cast = slice_by_index(begin = var_8252_begin_0, end = var_8252_end_0, end_mask = var_8252_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8252_cast")];
            tensor<int32, [4]> var_8256_begin_0 = const()[name = tensor<string, []>("op_8256_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_8256_end_0 = const()[name = tensor<string, []>("op_8256_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_8256_end_mask_0 = const()[name = tensor<string, []>("op_8256_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8256_cast = slice_by_index(begin = var_8256_begin_0, end = var_8256_end_0, end_mask = var_8256_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8256_cast")];
            tensor<int32, [4]> var_8260_begin_0 = const()[name = tensor<string, []>("op_8260_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_8260_end_0 = const()[name = tensor<string, []>("op_8260_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_8260_end_mask_0 = const()[name = tensor<string, []>("op_8260_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8260_cast = slice_by_index(begin = var_8260_begin_0, end = var_8260_end_0, end_mask = var_8260_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8260_cast")];
            tensor<int32, [4]> var_8264_begin_0 = const()[name = tensor<string, []>("op_8264_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_8264_end_0 = const()[name = tensor<string, []>("op_8264_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_8264_end_mask_0 = const()[name = tensor<string, []>("op_8264_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8264_cast = slice_by_index(begin = var_8264_begin_0, end = var_8264_end_0, end_mask = var_8264_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8264_cast")];
            tensor<int32, [4]> var_8268_begin_0 = const()[name = tensor<string, []>("op_8268_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_8268_end_0 = const()[name = tensor<string, []>("op_8268_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_8268_end_mask_0 = const()[name = tensor<string, []>("op_8268_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8268_cast = slice_by_index(begin = var_8268_begin_0, end = var_8268_end_0, end_mask = var_8268_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8268_cast")];
            tensor<int32, [4]> var_8272_begin_0 = const()[name = tensor<string, []>("op_8272_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_8272_end_0 = const()[name = tensor<string, []>("op_8272_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_8272_end_mask_0 = const()[name = tensor<string, []>("op_8272_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8272_cast = slice_by_index(begin = var_8272_begin_0, end = var_8272_end_0, end_mask = var_8272_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8272_cast")];
            tensor<int32, [4]> var_8276_begin_0 = const()[name = tensor<string, []>("op_8276_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_8276_end_0 = const()[name = tensor<string, []>("op_8276_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_8276_end_mask_0 = const()[name = tensor<string, []>("op_8276_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8276_cast = slice_by_index(begin = var_8276_begin_0, end = var_8276_end_0, end_mask = var_8276_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8276_cast")];
            tensor<int32, [4]> var_8280_begin_0 = const()[name = tensor<string, []>("op_8280_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_8280_end_0 = const()[name = tensor<string, []>("op_8280_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_8280_end_mask_0 = const()[name = tensor<string, []>("op_8280_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8280_cast = slice_by_index(begin = var_8280_begin_0, end = var_8280_end_0, end_mask = var_8280_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8280_cast")];
            tensor<int32, [4]> var_8284_begin_0 = const()[name = tensor<string, []>("op_8284_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_8284_end_0 = const()[name = tensor<string, []>("op_8284_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_8284_end_mask_0 = const()[name = tensor<string, []>("op_8284_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8284_cast = slice_by_index(begin = var_8284_begin_0, end = var_8284_end_0, end_mask = var_8284_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8284_cast")];
            tensor<int32, [4]> var_8288_begin_0 = const()[name = tensor<string, []>("op_8288_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_8288_end_0 = const()[name = tensor<string, []>("op_8288_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_8288_end_mask_0 = const()[name = tensor<string, []>("op_8288_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8288_cast = slice_by_index(begin = var_8288_begin_0, end = var_8288_end_0, end_mask = var_8288_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8288_cast")];
            tensor<int32, [4]> var_8292_begin_0 = const()[name = tensor<string, []>("op_8292_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_8292_end_0 = const()[name = tensor<string, []>("op_8292_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_8292_end_mask_0 = const()[name = tensor<string, []>("op_8292_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8292_cast = slice_by_index(begin = var_8292_begin_0, end = var_8292_end_0, end_mask = var_8292_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8292_cast")];
            tensor<int32, [4]> var_8296_begin_0 = const()[name = tensor<string, []>("op_8296_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_8296_end_0 = const()[name = tensor<string, []>("op_8296_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_8296_end_mask_0 = const()[name = tensor<string, []>("op_8296_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8296_cast = slice_by_index(begin = var_8296_begin_0, end = var_8296_end_0, end_mask = var_8296_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8296_cast")];
            tensor<int32, [4]> var_8300_begin_0 = const()[name = tensor<string, []>("op_8300_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_8300_end_0 = const()[name = tensor<string, []>("op_8300_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_8300_end_mask_0 = const()[name = tensor<string, []>("op_8300_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8300_cast = slice_by_index(begin = var_8300_begin_0, end = var_8300_end_0, end_mask = var_8300_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8300_cast")];
            tensor<int32, [4]> var_8304_begin_0 = const()[name = tensor<string, []>("op_8304_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_8304_end_0 = const()[name = tensor<string, []>("op_8304_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_8304_end_mask_0 = const()[name = tensor<string, []>("op_8304_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8304_cast = slice_by_index(begin = var_8304_begin_0, end = var_8304_end_0, end_mask = var_8304_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8304_cast")];
            tensor<int32, [4]> var_8308_begin_0 = const()[name = tensor<string, []>("op_8308_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_8308_end_0 = const()[name = tensor<string, []>("op_8308_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_8308_end_mask_0 = const()[name = tensor<string, []>("op_8308_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8308_cast = slice_by_index(begin = var_8308_begin_0, end = var_8308_end_0, end_mask = var_8308_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8308_cast")];
            tensor<int32, [4]> var_8312_begin_0 = const()[name = tensor<string, []>("op_8312_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_8312_end_0 = const()[name = tensor<string, []>("op_8312_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_8312_end_mask_0 = const()[name = tensor<string, []>("op_8312_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8312_cast = slice_by_index(begin = var_8312_begin_0, end = var_8312_end_0, end_mask = var_8312_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8312_cast")];
            tensor<int32, [4]> var_8316_begin_0 = const()[name = tensor<string, []>("op_8316_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_8316_end_0 = const()[name = tensor<string, []>("op_8316_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_8316_end_mask_0 = const()[name = tensor<string, []>("op_8316_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8316_cast = slice_by_index(begin = var_8316_begin_0, end = var_8316_end_0, end_mask = var_8316_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8316_cast")];
            tensor<int32, [4]> var_8320_begin_0 = const()[name = tensor<string, []>("op_8320_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_8320_end_0 = const()[name = tensor<string, []>("op_8320_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_8320_end_mask_0 = const()[name = tensor<string, []>("op_8320_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8320_cast = slice_by_index(begin = var_8320_begin_0, end = var_8320_end_0, end_mask = var_8320_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8320_cast")];
            tensor<int32, [4]> k_83_perm_0 = const()[name = tensor<string, []>("k_83_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_8327_begin_0 = const()[name = tensor<string, []>("op_8327_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8327_end_0 = const()[name = tensor<string, []>("op_8327_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_8327_end_mask_0 = const()[name = tensor<string, []>("op_8327_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_119 = transpose(perm = k_83_perm_0, x = k_81_cast)[name = tensor<string, []>("transpose_119")];
            tensor<fp16, [2, 1024, 1, 64]> var_8327_cast = slice_by_index(begin = var_8327_begin_0, end = var_8327_end_0, end_mask = var_8327_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8327_cast")];
            tensor<int32, [4]> var_8331_begin_0 = const()[name = tensor<string, []>("op_8331_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_8331_end_0 = const()[name = tensor<string, []>("op_8331_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_8331_end_mask_0 = const()[name = tensor<string, []>("op_8331_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8331_cast = slice_by_index(begin = var_8331_begin_0, end = var_8331_end_0, end_mask = var_8331_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8331_cast")];
            tensor<int32, [4]> var_8335_begin_0 = const()[name = tensor<string, []>("op_8335_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_8335_end_0 = const()[name = tensor<string, []>("op_8335_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_8335_end_mask_0 = const()[name = tensor<string, []>("op_8335_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8335_cast = slice_by_index(begin = var_8335_begin_0, end = var_8335_end_0, end_mask = var_8335_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8335_cast")];
            tensor<int32, [4]> var_8339_begin_0 = const()[name = tensor<string, []>("op_8339_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_8339_end_0 = const()[name = tensor<string, []>("op_8339_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_8339_end_mask_0 = const()[name = tensor<string, []>("op_8339_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8339_cast = slice_by_index(begin = var_8339_begin_0, end = var_8339_end_0, end_mask = var_8339_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8339_cast")];
            tensor<int32, [4]> var_8343_begin_0 = const()[name = tensor<string, []>("op_8343_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_8343_end_0 = const()[name = tensor<string, []>("op_8343_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_8343_end_mask_0 = const()[name = tensor<string, []>("op_8343_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8343_cast = slice_by_index(begin = var_8343_begin_0, end = var_8343_end_0, end_mask = var_8343_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8343_cast")];
            tensor<int32, [4]> var_8347_begin_0 = const()[name = tensor<string, []>("op_8347_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_8347_end_0 = const()[name = tensor<string, []>("op_8347_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_8347_end_mask_0 = const()[name = tensor<string, []>("op_8347_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8347_cast = slice_by_index(begin = var_8347_begin_0, end = var_8347_end_0, end_mask = var_8347_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8347_cast")];
            tensor<int32, [4]> var_8351_begin_0 = const()[name = tensor<string, []>("op_8351_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_8351_end_0 = const()[name = tensor<string, []>("op_8351_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_8351_end_mask_0 = const()[name = tensor<string, []>("op_8351_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8351_cast = slice_by_index(begin = var_8351_begin_0, end = var_8351_end_0, end_mask = var_8351_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8351_cast")];
            tensor<int32, [4]> var_8355_begin_0 = const()[name = tensor<string, []>("op_8355_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_8355_end_0 = const()[name = tensor<string, []>("op_8355_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_8355_end_mask_0 = const()[name = tensor<string, []>("op_8355_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8355_cast = slice_by_index(begin = var_8355_begin_0, end = var_8355_end_0, end_mask = var_8355_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8355_cast")];
            tensor<int32, [4]> var_8359_begin_0 = const()[name = tensor<string, []>("op_8359_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8359_end_0 = const()[name = tensor<string, []>("op_8359_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_8359_end_mask_0 = const()[name = tensor<string, []>("op_8359_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8359_cast = slice_by_index(begin = var_8359_begin_0, end = var_8359_end_0, end_mask = var_8359_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8359_cast")];
            tensor<int32, [4]> var_8363_begin_0 = const()[name = tensor<string, []>("op_8363_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_8363_end_0 = const()[name = tensor<string, []>("op_8363_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_8363_end_mask_0 = const()[name = tensor<string, []>("op_8363_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8363_cast = slice_by_index(begin = var_8363_begin_0, end = var_8363_end_0, end_mask = var_8363_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8363_cast")];
            tensor<int32, [4]> var_8367_begin_0 = const()[name = tensor<string, []>("op_8367_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_8367_end_0 = const()[name = tensor<string, []>("op_8367_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_8367_end_mask_0 = const()[name = tensor<string, []>("op_8367_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8367_cast = slice_by_index(begin = var_8367_begin_0, end = var_8367_end_0, end_mask = var_8367_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8367_cast")];
            tensor<int32, [4]> var_8371_begin_0 = const()[name = tensor<string, []>("op_8371_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_8371_end_0 = const()[name = tensor<string, []>("op_8371_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_8371_end_mask_0 = const()[name = tensor<string, []>("op_8371_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8371_cast = slice_by_index(begin = var_8371_begin_0, end = var_8371_end_0, end_mask = var_8371_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8371_cast")];
            tensor<int32, [4]> var_8375_begin_0 = const()[name = tensor<string, []>("op_8375_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_8375_end_0 = const()[name = tensor<string, []>("op_8375_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_8375_end_mask_0 = const()[name = tensor<string, []>("op_8375_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8375_cast = slice_by_index(begin = var_8375_begin_0, end = var_8375_end_0, end_mask = var_8375_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8375_cast")];
            tensor<int32, [4]> var_8379_begin_0 = const()[name = tensor<string, []>("op_8379_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_8379_end_0 = const()[name = tensor<string, []>("op_8379_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_8379_end_mask_0 = const()[name = tensor<string, []>("op_8379_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8379_cast = slice_by_index(begin = var_8379_begin_0, end = var_8379_end_0, end_mask = var_8379_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8379_cast")];
            tensor<int32, [4]> var_8383_begin_0 = const()[name = tensor<string, []>("op_8383_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_8383_end_0 = const()[name = tensor<string, []>("op_8383_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_8383_end_mask_0 = const()[name = tensor<string, []>("op_8383_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8383_cast = slice_by_index(begin = var_8383_begin_0, end = var_8383_end_0, end_mask = var_8383_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8383_cast")];
            tensor<int32, [4]> var_8387_begin_0 = const()[name = tensor<string, []>("op_8387_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_8387_end_0 = const()[name = tensor<string, []>("op_8387_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_8387_end_mask_0 = const()[name = tensor<string, []>("op_8387_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8387_cast = slice_by_index(begin = var_8387_begin_0, end = var_8387_end_0, end_mask = var_8387_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8387_cast")];
            tensor<int32, [4]> var_8391_begin_0 = const()[name = tensor<string, []>("op_8391_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_8391_end_0 = const()[name = tensor<string, []>("op_8391_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_8391_end_mask_0 = const()[name = tensor<string, []>("op_8391_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8391_cast = slice_by_index(begin = var_8391_begin_0, end = var_8391_end_0, end_mask = var_8391_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8391_cast")];
            tensor<int32, [4]> var_8395_begin_0 = const()[name = tensor<string, []>("op_8395_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_8395_end_0 = const()[name = tensor<string, []>("op_8395_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_8395_end_mask_0 = const()[name = tensor<string, []>("op_8395_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8395_cast = slice_by_index(begin = var_8395_begin_0, end = var_8395_end_0, end_mask = var_8395_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8395_cast")];
            tensor<int32, [4]> var_8399_begin_0 = const()[name = tensor<string, []>("op_8399_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_8399_end_0 = const()[name = tensor<string, []>("op_8399_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_8399_end_mask_0 = const()[name = tensor<string, []>("op_8399_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8399_cast = slice_by_index(begin = var_8399_begin_0, end = var_8399_end_0, end_mask = var_8399_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8399_cast")];
            tensor<int32, [4]> var_8403_begin_0 = const()[name = tensor<string, []>("op_8403_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_8403_end_0 = const()[name = tensor<string, []>("op_8403_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_8403_end_mask_0 = const()[name = tensor<string, []>("op_8403_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8403_cast = slice_by_index(begin = var_8403_begin_0, end = var_8403_end_0, end_mask = var_8403_end_mask_0, x = transpose_119)[name = tensor<string, []>("op_8403_cast")];
            tensor<int32, [4]> var_8405_begin_0 = const()[name = tensor<string, []>("op_8405_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8405_end_0 = const()[name = tensor<string, []>("op_8405_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_8405_end_mask_0 = const()[name = tensor<string, []>("op_8405_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8405_cast = slice_by_index(begin = var_8405_begin_0, end = var_8405_end_0, end_mask = var_8405_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8405_cast")];
            tensor<int32, [4]> var_8409_begin_0 = const()[name = tensor<string, []>("op_8409_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_8409_end_0 = const()[name = tensor<string, []>("op_8409_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_8409_end_mask_0 = const()[name = tensor<string, []>("op_8409_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8409_cast = slice_by_index(begin = var_8409_begin_0, end = var_8409_end_0, end_mask = var_8409_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8409_cast")];
            tensor<int32, [4]> var_8413_begin_0 = const()[name = tensor<string, []>("op_8413_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_8413_end_0 = const()[name = tensor<string, []>("op_8413_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_8413_end_mask_0 = const()[name = tensor<string, []>("op_8413_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8413_cast = slice_by_index(begin = var_8413_begin_0, end = var_8413_end_0, end_mask = var_8413_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8413_cast")];
            tensor<int32, [4]> var_8417_begin_0 = const()[name = tensor<string, []>("op_8417_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_8417_end_0 = const()[name = tensor<string, []>("op_8417_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_8417_end_mask_0 = const()[name = tensor<string, []>("op_8417_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8417_cast = slice_by_index(begin = var_8417_begin_0, end = var_8417_end_0, end_mask = var_8417_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8417_cast")];
            tensor<int32, [4]> var_8421_begin_0 = const()[name = tensor<string, []>("op_8421_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_8421_end_0 = const()[name = tensor<string, []>("op_8421_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_8421_end_mask_0 = const()[name = tensor<string, []>("op_8421_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8421_cast = slice_by_index(begin = var_8421_begin_0, end = var_8421_end_0, end_mask = var_8421_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8421_cast")];
            tensor<int32, [4]> var_8425_begin_0 = const()[name = tensor<string, []>("op_8425_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_8425_end_0 = const()[name = tensor<string, []>("op_8425_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_8425_end_mask_0 = const()[name = tensor<string, []>("op_8425_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8425_cast = slice_by_index(begin = var_8425_begin_0, end = var_8425_end_0, end_mask = var_8425_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8425_cast")];
            tensor<int32, [4]> var_8429_begin_0 = const()[name = tensor<string, []>("op_8429_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_8429_end_0 = const()[name = tensor<string, []>("op_8429_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_8429_end_mask_0 = const()[name = tensor<string, []>("op_8429_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8429_cast = slice_by_index(begin = var_8429_begin_0, end = var_8429_end_0, end_mask = var_8429_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8429_cast")];
            tensor<int32, [4]> var_8433_begin_0 = const()[name = tensor<string, []>("op_8433_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_8433_end_0 = const()[name = tensor<string, []>("op_8433_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_8433_end_mask_0 = const()[name = tensor<string, []>("op_8433_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8433_cast = slice_by_index(begin = var_8433_begin_0, end = var_8433_end_0, end_mask = var_8433_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8433_cast")];
            tensor<int32, [4]> var_8437_begin_0 = const()[name = tensor<string, []>("op_8437_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_8437_end_0 = const()[name = tensor<string, []>("op_8437_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_8437_end_mask_0 = const()[name = tensor<string, []>("op_8437_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8437_cast = slice_by_index(begin = var_8437_begin_0, end = var_8437_end_0, end_mask = var_8437_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8437_cast")];
            tensor<int32, [4]> var_8441_begin_0 = const()[name = tensor<string, []>("op_8441_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_8441_end_0 = const()[name = tensor<string, []>("op_8441_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_8441_end_mask_0 = const()[name = tensor<string, []>("op_8441_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8441_cast = slice_by_index(begin = var_8441_begin_0, end = var_8441_end_0, end_mask = var_8441_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8441_cast")];
            tensor<int32, [4]> var_8445_begin_0 = const()[name = tensor<string, []>("op_8445_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_8445_end_0 = const()[name = tensor<string, []>("op_8445_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_8445_end_mask_0 = const()[name = tensor<string, []>("op_8445_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8445_cast = slice_by_index(begin = var_8445_begin_0, end = var_8445_end_0, end_mask = var_8445_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8445_cast")];
            tensor<int32, [4]> var_8449_begin_0 = const()[name = tensor<string, []>("op_8449_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_8449_end_0 = const()[name = tensor<string, []>("op_8449_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_8449_end_mask_0 = const()[name = tensor<string, []>("op_8449_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8449_cast = slice_by_index(begin = var_8449_begin_0, end = var_8449_end_0, end_mask = var_8449_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8449_cast")];
            tensor<int32, [4]> var_8453_begin_0 = const()[name = tensor<string, []>("op_8453_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_8453_end_0 = const()[name = tensor<string, []>("op_8453_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_8453_end_mask_0 = const()[name = tensor<string, []>("op_8453_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8453_cast = slice_by_index(begin = var_8453_begin_0, end = var_8453_end_0, end_mask = var_8453_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8453_cast")];
            tensor<int32, [4]> var_8457_begin_0 = const()[name = tensor<string, []>("op_8457_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_8457_end_0 = const()[name = tensor<string, []>("op_8457_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_8457_end_mask_0 = const()[name = tensor<string, []>("op_8457_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8457_cast = slice_by_index(begin = var_8457_begin_0, end = var_8457_end_0, end_mask = var_8457_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8457_cast")];
            tensor<int32, [4]> var_8461_begin_0 = const()[name = tensor<string, []>("op_8461_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_8461_end_0 = const()[name = tensor<string, []>("op_8461_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_8461_end_mask_0 = const()[name = tensor<string, []>("op_8461_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8461_cast = slice_by_index(begin = var_8461_begin_0, end = var_8461_end_0, end_mask = var_8461_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8461_cast")];
            tensor<int32, [4]> var_8465_begin_0 = const()[name = tensor<string, []>("op_8465_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_8465_end_0 = const()[name = tensor<string, []>("op_8465_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_8465_end_mask_0 = const()[name = tensor<string, []>("op_8465_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8465_cast = slice_by_index(begin = var_8465_begin_0, end = var_8465_end_0, end_mask = var_8465_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8465_cast")];
            tensor<int32, [4]> var_8469_begin_0 = const()[name = tensor<string, []>("op_8469_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_8469_end_0 = const()[name = tensor<string, []>("op_8469_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_8469_end_mask_0 = const()[name = tensor<string, []>("op_8469_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8469_cast = slice_by_index(begin = var_8469_begin_0, end = var_8469_end_0, end_mask = var_8469_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8469_cast")];
            tensor<int32, [4]> var_8473_begin_0 = const()[name = tensor<string, []>("op_8473_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_8473_end_0 = const()[name = tensor<string, []>("op_8473_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_8473_end_mask_0 = const()[name = tensor<string, []>("op_8473_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8473_cast = slice_by_index(begin = var_8473_begin_0, end = var_8473_end_0, end_mask = var_8473_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8473_cast")];
            tensor<int32, [4]> var_8477_begin_0 = const()[name = tensor<string, []>("op_8477_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_8477_end_0 = const()[name = tensor<string, []>("op_8477_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_8477_end_mask_0 = const()[name = tensor<string, []>("op_8477_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8477_cast = slice_by_index(begin = var_8477_begin_0, end = var_8477_end_0, end_mask = var_8477_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8477_cast")];
            tensor<int32, [4]> var_8481_begin_0 = const()[name = tensor<string, []>("op_8481_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_8481_end_0 = const()[name = tensor<string, []>("op_8481_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_8481_end_mask_0 = const()[name = tensor<string, []>("op_8481_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8481_cast = slice_by_index(begin = var_8481_begin_0, end = var_8481_end_0, end_mask = var_8481_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8481_cast")];
            tensor<string, []> var_8485_equation_0 = const()[name = tensor<string, []>("op_8485_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8485_cast = einsum(equation = var_8485_equation_0, values = (var_8327_cast, var_8244_cast))[name = tensor<string, []>("op_8485_cast")];
            tensor<fp16, []> var_8486_to_fp16 = const()[name = tensor<string, []>("op_8486_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_641_cast = mul(x = var_8485_cast, y = var_8486_to_fp16)[name = tensor<string, []>("aw_641_cast")];
            tensor<string, []> var_8489_equation_0 = const()[name = tensor<string, []>("op_8489_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8489_cast = einsum(equation = var_8489_equation_0, values = (var_8331_cast, var_8248_cast))[name = tensor<string, []>("op_8489_cast")];
            tensor<fp16, []> var_8490_to_fp16 = const()[name = tensor<string, []>("op_8490_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_643_cast = mul(x = var_8489_cast, y = var_8490_to_fp16)[name = tensor<string, []>("aw_643_cast")];
            tensor<string, []> var_8493_equation_0 = const()[name = tensor<string, []>("op_8493_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8493_cast = einsum(equation = var_8493_equation_0, values = (var_8335_cast, var_8252_cast))[name = tensor<string, []>("op_8493_cast")];
            tensor<fp16, []> var_8494_to_fp16 = const()[name = tensor<string, []>("op_8494_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_645_cast = mul(x = var_8493_cast, y = var_8494_to_fp16)[name = tensor<string, []>("aw_645_cast")];
            tensor<string, []> var_8497_equation_0 = const()[name = tensor<string, []>("op_8497_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8497_cast = einsum(equation = var_8497_equation_0, values = (var_8339_cast, var_8256_cast))[name = tensor<string, []>("op_8497_cast")];
            tensor<fp16, []> var_8498_to_fp16 = const()[name = tensor<string, []>("op_8498_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_647_cast = mul(x = var_8497_cast, y = var_8498_to_fp16)[name = tensor<string, []>("aw_647_cast")];
            tensor<string, []> var_8501_equation_0 = const()[name = tensor<string, []>("op_8501_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8501_cast = einsum(equation = var_8501_equation_0, values = (var_8343_cast, var_8260_cast))[name = tensor<string, []>("op_8501_cast")];
            tensor<fp16, []> var_8502_to_fp16 = const()[name = tensor<string, []>("op_8502_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_649_cast = mul(x = var_8501_cast, y = var_8502_to_fp16)[name = tensor<string, []>("aw_649_cast")];
            tensor<string, []> var_8505_equation_0 = const()[name = tensor<string, []>("op_8505_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8505_cast = einsum(equation = var_8505_equation_0, values = (var_8347_cast, var_8264_cast))[name = tensor<string, []>("op_8505_cast")];
            tensor<fp16, []> var_8506_to_fp16 = const()[name = tensor<string, []>("op_8506_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_651_cast = mul(x = var_8505_cast, y = var_8506_to_fp16)[name = tensor<string, []>("aw_651_cast")];
            tensor<string, []> var_8509_equation_0 = const()[name = tensor<string, []>("op_8509_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8509_cast = einsum(equation = var_8509_equation_0, values = (var_8351_cast, var_8268_cast))[name = tensor<string, []>("op_8509_cast")];
            tensor<fp16, []> var_8510_to_fp16 = const()[name = tensor<string, []>("op_8510_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_653_cast = mul(x = var_8509_cast, y = var_8510_to_fp16)[name = tensor<string, []>("aw_653_cast")];
            tensor<string, []> var_8513_equation_0 = const()[name = tensor<string, []>("op_8513_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8513_cast = einsum(equation = var_8513_equation_0, values = (var_8355_cast, var_8272_cast))[name = tensor<string, []>("op_8513_cast")];
            tensor<fp16, []> var_8514_to_fp16 = const()[name = tensor<string, []>("op_8514_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_655_cast = mul(x = var_8513_cast, y = var_8514_to_fp16)[name = tensor<string, []>("aw_655_cast")];
            tensor<string, []> var_8517_equation_0 = const()[name = tensor<string, []>("op_8517_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8517_cast = einsum(equation = var_8517_equation_0, values = (var_8359_cast, var_8276_cast))[name = tensor<string, []>("op_8517_cast")];
            tensor<fp16, []> var_8518_to_fp16 = const()[name = tensor<string, []>("op_8518_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_657_cast = mul(x = var_8517_cast, y = var_8518_to_fp16)[name = tensor<string, []>("aw_657_cast")];
            tensor<string, []> var_8521_equation_0 = const()[name = tensor<string, []>("op_8521_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8521_cast = einsum(equation = var_8521_equation_0, values = (var_8363_cast, var_8280_cast))[name = tensor<string, []>("op_8521_cast")];
            tensor<fp16, []> var_8522_to_fp16 = const()[name = tensor<string, []>("op_8522_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_659_cast = mul(x = var_8521_cast, y = var_8522_to_fp16)[name = tensor<string, []>("aw_659_cast")];
            tensor<string, []> var_8525_equation_0 = const()[name = tensor<string, []>("op_8525_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8525_cast = einsum(equation = var_8525_equation_0, values = (var_8367_cast, var_8284_cast))[name = tensor<string, []>("op_8525_cast")];
            tensor<fp16, []> var_8526_to_fp16 = const()[name = tensor<string, []>("op_8526_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_661_cast = mul(x = var_8525_cast, y = var_8526_to_fp16)[name = tensor<string, []>("aw_661_cast")];
            tensor<string, []> var_8529_equation_0 = const()[name = tensor<string, []>("op_8529_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8529_cast = einsum(equation = var_8529_equation_0, values = (var_8371_cast, var_8288_cast))[name = tensor<string, []>("op_8529_cast")];
            tensor<fp16, []> var_8530_to_fp16 = const()[name = tensor<string, []>("op_8530_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_663_cast = mul(x = var_8529_cast, y = var_8530_to_fp16)[name = tensor<string, []>("aw_663_cast")];
            tensor<string, []> var_8533_equation_0 = const()[name = tensor<string, []>("op_8533_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8533_cast = einsum(equation = var_8533_equation_0, values = (var_8375_cast, var_8292_cast))[name = tensor<string, []>("op_8533_cast")];
            tensor<fp16, []> var_8534_to_fp16 = const()[name = tensor<string, []>("op_8534_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_665_cast = mul(x = var_8533_cast, y = var_8534_to_fp16)[name = tensor<string, []>("aw_665_cast")];
            tensor<string, []> var_8537_equation_0 = const()[name = tensor<string, []>("op_8537_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8537_cast = einsum(equation = var_8537_equation_0, values = (var_8379_cast, var_8296_cast))[name = tensor<string, []>("op_8537_cast")];
            tensor<fp16, []> var_8538_to_fp16 = const()[name = tensor<string, []>("op_8538_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_667_cast = mul(x = var_8537_cast, y = var_8538_to_fp16)[name = tensor<string, []>("aw_667_cast")];
            tensor<string, []> var_8541_equation_0 = const()[name = tensor<string, []>("op_8541_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8541_cast = einsum(equation = var_8541_equation_0, values = (var_8383_cast, var_8300_cast))[name = tensor<string, []>("op_8541_cast")];
            tensor<fp16, []> var_8542_to_fp16 = const()[name = tensor<string, []>("op_8542_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_669_cast = mul(x = var_8541_cast, y = var_8542_to_fp16)[name = tensor<string, []>("aw_669_cast")];
            tensor<string, []> var_8545_equation_0 = const()[name = tensor<string, []>("op_8545_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8545_cast = einsum(equation = var_8545_equation_0, values = (var_8387_cast, var_8304_cast))[name = tensor<string, []>("op_8545_cast")];
            tensor<fp16, []> var_8546_to_fp16 = const()[name = tensor<string, []>("op_8546_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_671_cast = mul(x = var_8545_cast, y = var_8546_to_fp16)[name = tensor<string, []>("aw_671_cast")];
            tensor<string, []> var_8549_equation_0 = const()[name = tensor<string, []>("op_8549_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8549_cast = einsum(equation = var_8549_equation_0, values = (var_8391_cast, var_8308_cast))[name = tensor<string, []>("op_8549_cast")];
            tensor<fp16, []> var_8550_to_fp16 = const()[name = tensor<string, []>("op_8550_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_673_cast = mul(x = var_8549_cast, y = var_8550_to_fp16)[name = tensor<string, []>("aw_673_cast")];
            tensor<string, []> var_8553_equation_0 = const()[name = tensor<string, []>("op_8553_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8553_cast = einsum(equation = var_8553_equation_0, values = (var_8395_cast, var_8312_cast))[name = tensor<string, []>("op_8553_cast")];
            tensor<fp16, []> var_8554_to_fp16 = const()[name = tensor<string, []>("op_8554_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_675_cast = mul(x = var_8553_cast, y = var_8554_to_fp16)[name = tensor<string, []>("aw_675_cast")];
            tensor<string, []> var_8557_equation_0 = const()[name = tensor<string, []>("op_8557_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8557_cast = einsum(equation = var_8557_equation_0, values = (var_8399_cast, var_8316_cast))[name = tensor<string, []>("op_8557_cast")];
            tensor<fp16, []> var_8558_to_fp16 = const()[name = tensor<string, []>("op_8558_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_677_cast = mul(x = var_8557_cast, y = var_8558_to_fp16)[name = tensor<string, []>("aw_677_cast")];
            tensor<string, []> var_8561_equation_0 = const()[name = tensor<string, []>("op_8561_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8561_cast = einsum(equation = var_8561_equation_0, values = (var_8403_cast, var_8320_cast))[name = tensor<string, []>("op_8561_cast")];
            tensor<fp16, []> var_8562_to_fp16 = const()[name = tensor<string, []>("op_8562_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_679_cast = mul(x = var_8561_cast, y = var_8562_to_fp16)[name = tensor<string, []>("aw_679_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8564_cast = softmax(axis = var_2634, x = aw_641_cast)[name = tensor<string, []>("op_8564_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8565_cast = softmax(axis = var_2634, x = aw_643_cast)[name = tensor<string, []>("op_8565_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8566_cast = softmax(axis = var_2634, x = aw_645_cast)[name = tensor<string, []>("op_8566_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8567_cast = softmax(axis = var_2634, x = aw_647_cast)[name = tensor<string, []>("op_8567_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8568_cast = softmax(axis = var_2634, x = aw_649_cast)[name = tensor<string, []>("op_8568_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8569_cast = softmax(axis = var_2634, x = aw_651_cast)[name = tensor<string, []>("op_8569_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8570_cast = softmax(axis = var_2634, x = aw_653_cast)[name = tensor<string, []>("op_8570_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8571_cast = softmax(axis = var_2634, x = aw_655_cast)[name = tensor<string, []>("op_8571_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8572_cast = softmax(axis = var_2634, x = aw_657_cast)[name = tensor<string, []>("op_8572_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8573_cast = softmax(axis = var_2634, x = aw_659_cast)[name = tensor<string, []>("op_8573_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8574_cast = softmax(axis = var_2634, x = aw_661_cast)[name = tensor<string, []>("op_8574_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8575_cast = softmax(axis = var_2634, x = aw_663_cast)[name = tensor<string, []>("op_8575_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8576_cast = softmax(axis = var_2634, x = aw_665_cast)[name = tensor<string, []>("op_8576_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8577_cast = softmax(axis = var_2634, x = aw_667_cast)[name = tensor<string, []>("op_8577_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8578_cast = softmax(axis = var_2634, x = aw_669_cast)[name = tensor<string, []>("op_8578_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8579_cast = softmax(axis = var_2634, x = aw_671_cast)[name = tensor<string, []>("op_8579_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8580_cast = softmax(axis = var_2634, x = aw_673_cast)[name = tensor<string, []>("op_8580_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8581_cast = softmax(axis = var_2634, x = aw_675_cast)[name = tensor<string, []>("op_8581_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8582_cast = softmax(axis = var_2634, x = aw_677_cast)[name = tensor<string, []>("op_8582_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8583_cast = softmax(axis = var_2634, x = aw_679_cast)[name = tensor<string, []>("op_8583_cast")];
            tensor<string, []> var_8585_equation_0 = const()[name = tensor<string, []>("op_8585_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8585_cast = einsum(equation = var_8585_equation_0, values = (var_8405_cast, var_8564_cast))[name = tensor<string, []>("op_8585_cast")];
            tensor<string, []> var_8587_equation_0 = const()[name = tensor<string, []>("op_8587_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8587_cast = einsum(equation = var_8587_equation_0, values = (var_8409_cast, var_8565_cast))[name = tensor<string, []>("op_8587_cast")];
            tensor<string, []> var_8589_equation_0 = const()[name = tensor<string, []>("op_8589_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8589_cast = einsum(equation = var_8589_equation_0, values = (var_8413_cast, var_8566_cast))[name = tensor<string, []>("op_8589_cast")];
            tensor<string, []> var_8591_equation_0 = const()[name = tensor<string, []>("op_8591_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8591_cast = einsum(equation = var_8591_equation_0, values = (var_8417_cast, var_8567_cast))[name = tensor<string, []>("op_8591_cast")];
            tensor<string, []> var_8593_equation_0 = const()[name = tensor<string, []>("op_8593_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8593_cast = einsum(equation = var_8593_equation_0, values = (var_8421_cast, var_8568_cast))[name = tensor<string, []>("op_8593_cast")];
            tensor<string, []> var_8595_equation_0 = const()[name = tensor<string, []>("op_8595_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8595_cast = einsum(equation = var_8595_equation_0, values = (var_8425_cast, var_8569_cast))[name = tensor<string, []>("op_8595_cast")];
            tensor<string, []> var_8597_equation_0 = const()[name = tensor<string, []>("op_8597_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8597_cast = einsum(equation = var_8597_equation_0, values = (var_8429_cast, var_8570_cast))[name = tensor<string, []>("op_8597_cast")];
            tensor<string, []> var_8599_equation_0 = const()[name = tensor<string, []>("op_8599_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8599_cast = einsum(equation = var_8599_equation_0, values = (var_8433_cast, var_8571_cast))[name = tensor<string, []>("op_8599_cast")];
            tensor<string, []> var_8601_equation_0 = const()[name = tensor<string, []>("op_8601_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8601_cast = einsum(equation = var_8601_equation_0, values = (var_8437_cast, var_8572_cast))[name = tensor<string, []>("op_8601_cast")];
            tensor<string, []> var_8603_equation_0 = const()[name = tensor<string, []>("op_8603_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8603_cast = einsum(equation = var_8603_equation_0, values = (var_8441_cast, var_8573_cast))[name = tensor<string, []>("op_8603_cast")];
            tensor<string, []> var_8605_equation_0 = const()[name = tensor<string, []>("op_8605_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8605_cast = einsum(equation = var_8605_equation_0, values = (var_8445_cast, var_8574_cast))[name = tensor<string, []>("op_8605_cast")];
            tensor<string, []> var_8607_equation_0 = const()[name = tensor<string, []>("op_8607_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8607_cast = einsum(equation = var_8607_equation_0, values = (var_8449_cast, var_8575_cast))[name = tensor<string, []>("op_8607_cast")];
            tensor<string, []> var_8609_equation_0 = const()[name = tensor<string, []>("op_8609_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8609_cast = einsum(equation = var_8609_equation_0, values = (var_8453_cast, var_8576_cast))[name = tensor<string, []>("op_8609_cast")];
            tensor<string, []> var_8611_equation_0 = const()[name = tensor<string, []>("op_8611_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8611_cast = einsum(equation = var_8611_equation_0, values = (var_8457_cast, var_8577_cast))[name = tensor<string, []>("op_8611_cast")];
            tensor<string, []> var_8613_equation_0 = const()[name = tensor<string, []>("op_8613_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8613_cast = einsum(equation = var_8613_equation_0, values = (var_8461_cast, var_8578_cast))[name = tensor<string, []>("op_8613_cast")];
            tensor<string, []> var_8615_equation_0 = const()[name = tensor<string, []>("op_8615_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8615_cast = einsum(equation = var_8615_equation_0, values = (var_8465_cast, var_8579_cast))[name = tensor<string, []>("op_8615_cast")];
            tensor<string, []> var_8617_equation_0 = const()[name = tensor<string, []>("op_8617_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8617_cast = einsum(equation = var_8617_equation_0, values = (var_8469_cast, var_8580_cast))[name = tensor<string, []>("op_8617_cast")];
            tensor<string, []> var_8619_equation_0 = const()[name = tensor<string, []>("op_8619_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8619_cast = einsum(equation = var_8619_equation_0, values = (var_8473_cast, var_8581_cast))[name = tensor<string, []>("op_8619_cast")];
            tensor<string, []> var_8621_equation_0 = const()[name = tensor<string, []>("op_8621_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8621_cast = einsum(equation = var_8621_equation_0, values = (var_8477_cast, var_8582_cast))[name = tensor<string, []>("op_8621_cast")];
            tensor<string, []> var_8623_equation_0 = const()[name = tensor<string, []>("op_8623_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8623_cast = einsum(equation = var_8623_equation_0, values = (var_8481_cast, var_8583_cast))[name = tensor<string, []>("op_8623_cast")];
            tensor<bool, []> input_183_interleave_0 = const()[name = tensor<string, []>("input_183_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_183_cast = concat(axis = var_2634, interleave = input_183_interleave_0, values = (var_8585_cast, var_8587_cast, var_8589_cast, var_8591_cast, var_8593_cast, var_8595_cast, var_8597_cast, var_8599_cast, var_8601_cast, var_8603_cast, var_8605_cast, var_8607_cast, var_8609_cast, var_8611_cast, var_8613_cast, var_8615_cast, var_8617_cast, var_8619_cast, var_8621_cast, var_8623_cast))[name = tensor<string, []>("input_183_cast")];
            tensor<int32, [2]> var_8629 = const()[name = tensor<string, []>("op_8629"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8631 = const()[name = tensor<string, []>("op_8631"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8633_pad_type_0 = const()[name = tensor<string, []>("op_8633_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8633_pad_0 = const()[name = tensor<string, []>("op_8633_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(625079488)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(628356352)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_8633_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16, dilations = var_8631, groups = var_2634, pad = var_8633_pad_0, pad_type = var_8633_pad_type_0, strides = var_8629, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16, x = input_183_cast)[name = tensor<string, []>("op_8633_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_63_cast = add(x = var_8633_cast, y = inputs_61_cast)[name = tensor<string, []>("inputs_63_cast")];
            tensor<int32, [1]> var_8637 = const()[name = tensor<string, []>("op_8637"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_63_cast = reduce_mean(axes = var_8637, keep_dims = var_2629, x = inputs_63_cast)[name = tensor<string, []>("channels_mean_63_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_63_cast = sub(x = inputs_63_cast, y = channels_mean_63_cast)[name = tensor<string, []>("zero_mean_63_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_63_cast = mul(x = zero_mean_63_cast, y = zero_mean_63_cast)[name = tensor<string, []>("zero_mean_sq_63_cast")];
            tensor<int32, [1]> var_8641 = const()[name = tensor<string, []>("op_8641"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_8642_cast = reduce_mean(axes = var_8641, keep_dims = var_2629, x = zero_mean_sq_63_cast)[name = tensor<string, []>("op_8642_cast")];
            tensor<fp16, []> var_8643_to_fp16 = const()[name = tensor<string, []>("op_8643_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_8644_cast = add(x = var_8642_cast, y = var_8643_to_fp16)[name = tensor<string, []>("op_8644_cast")];
            tensor<fp16, []> denom_63_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_63_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_63_cast = rsqrt(epsilon = denom_63_epsilon_0_to_fp16, x = var_8644_cast)[name = tensor<string, []>("denom_63_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_63_cast = mul(x = zero_mean_63_cast, y = denom_63_cast)[name = tensor<string, []>("out_63_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_8648_to_fp16 = const()[name = tensor<string, []>("op_8648_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(628358976)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_8649_cast = add(x = out_63_cast, y = var_8648_to_fp16)[name = tensor<string, []>("op_8649_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_8651_to_fp16 = const()[name = tensor<string, []>("op_8651_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(628361600)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_103_cast = mul(x = var_8649_cast, y = var_8651_to_fp16)[name = tensor<string, []>("hidden_states_103_cast")];
            tensor<int32, [2]> var_8658 = const()[name = tensor<string, []>("op_8658"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8660 = const()[name = tensor<string, []>("op_8660"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_43_pad_type_0 = const()[name = tensor<string, []>("q_43_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_43_pad_0 = const()[name = tensor<string, []>("q_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(628364224)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_43_cast = conv(dilations = var_8660, groups = var_2634, pad = q_43_pad_0, pad_type = q_43_pad_type_0, strides = var_8658, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16, x = hidden_states_103_cast)[name = tensor<string, []>("q_43_cast")];
            tensor<int32, [2]> var_8664 = const()[name = tensor<string, []>("op_8664"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8666 = const()[name = tensor<string, []>("op_8666"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_85_pad_type_0 = const()[name = tensor<string, []>("k_85_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_85_pad_0 = const()[name = tensor<string, []>("k_85_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(631641088)))];
            tensor<fp16, [2, 1280, 1, 77]> k_85_cast = conv(dilations = var_8666, groups = var_2634, pad = k_85_pad_0, pad_type = k_85_pad_type_0, strides = var_8664, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_85_cast")];
            tensor<int32, [2]> var_8670 = const()[name = tensor<string, []>("op_8670"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8672 = const()[name = tensor<string, []>("op_8672"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_43_pad_type_0 = const()[name = tensor<string, []>("v_43_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_43_pad_0 = const()[name = tensor<string, []>("v_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(636884032)))];
            tensor<fp16, [2, 1280, 1, 77]> v_43_cast = conv(dilations = var_8672, groups = var_2634, pad = v_43_pad_0, pad_type = v_43_pad_type_0, strides = var_8670, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_43_cast")];
            tensor<int32, [4]> var_8676_begin_0 = const()[name = tensor<string, []>("op_8676_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8676_end_0 = const()[name = tensor<string, []>("op_8676_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_8676_end_mask_0 = const()[name = tensor<string, []>("op_8676_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8676_cast = slice_by_index(begin = var_8676_begin_0, end = var_8676_end_0, end_mask = var_8676_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8676_cast")];
            tensor<int32, [4]> var_8680_begin_0 = const()[name = tensor<string, []>("op_8680_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_8680_end_0 = const()[name = tensor<string, []>("op_8680_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_8680_end_mask_0 = const()[name = tensor<string, []>("op_8680_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8680_cast = slice_by_index(begin = var_8680_begin_0, end = var_8680_end_0, end_mask = var_8680_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8680_cast")];
            tensor<int32, [4]> var_8684_begin_0 = const()[name = tensor<string, []>("op_8684_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_8684_end_0 = const()[name = tensor<string, []>("op_8684_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_8684_end_mask_0 = const()[name = tensor<string, []>("op_8684_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8684_cast = slice_by_index(begin = var_8684_begin_0, end = var_8684_end_0, end_mask = var_8684_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8684_cast")];
            tensor<int32, [4]> var_8688_begin_0 = const()[name = tensor<string, []>("op_8688_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_8688_end_0 = const()[name = tensor<string, []>("op_8688_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_8688_end_mask_0 = const()[name = tensor<string, []>("op_8688_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8688_cast = slice_by_index(begin = var_8688_begin_0, end = var_8688_end_0, end_mask = var_8688_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8688_cast")];
            tensor<int32, [4]> var_8692_begin_0 = const()[name = tensor<string, []>("op_8692_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_8692_end_0 = const()[name = tensor<string, []>("op_8692_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_8692_end_mask_0 = const()[name = tensor<string, []>("op_8692_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8692_cast = slice_by_index(begin = var_8692_begin_0, end = var_8692_end_0, end_mask = var_8692_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8692_cast")];
            tensor<int32, [4]> var_8696_begin_0 = const()[name = tensor<string, []>("op_8696_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_8696_end_0 = const()[name = tensor<string, []>("op_8696_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_8696_end_mask_0 = const()[name = tensor<string, []>("op_8696_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8696_cast = slice_by_index(begin = var_8696_begin_0, end = var_8696_end_0, end_mask = var_8696_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8696_cast")];
            tensor<int32, [4]> var_8700_begin_0 = const()[name = tensor<string, []>("op_8700_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_8700_end_0 = const()[name = tensor<string, []>("op_8700_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_8700_end_mask_0 = const()[name = tensor<string, []>("op_8700_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8700_cast = slice_by_index(begin = var_8700_begin_0, end = var_8700_end_0, end_mask = var_8700_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8700_cast")];
            tensor<int32, [4]> var_8704_begin_0 = const()[name = tensor<string, []>("op_8704_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_8704_end_0 = const()[name = tensor<string, []>("op_8704_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_8704_end_mask_0 = const()[name = tensor<string, []>("op_8704_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8704_cast = slice_by_index(begin = var_8704_begin_0, end = var_8704_end_0, end_mask = var_8704_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8704_cast")];
            tensor<int32, [4]> var_8708_begin_0 = const()[name = tensor<string, []>("op_8708_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_8708_end_0 = const()[name = tensor<string, []>("op_8708_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_8708_end_mask_0 = const()[name = tensor<string, []>("op_8708_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8708_cast = slice_by_index(begin = var_8708_begin_0, end = var_8708_end_0, end_mask = var_8708_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8708_cast")];
            tensor<int32, [4]> var_8712_begin_0 = const()[name = tensor<string, []>("op_8712_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_8712_end_0 = const()[name = tensor<string, []>("op_8712_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_8712_end_mask_0 = const()[name = tensor<string, []>("op_8712_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8712_cast = slice_by_index(begin = var_8712_begin_0, end = var_8712_end_0, end_mask = var_8712_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8712_cast")];
            tensor<int32, [4]> var_8716_begin_0 = const()[name = tensor<string, []>("op_8716_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_8716_end_0 = const()[name = tensor<string, []>("op_8716_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_8716_end_mask_0 = const()[name = tensor<string, []>("op_8716_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8716_cast = slice_by_index(begin = var_8716_begin_0, end = var_8716_end_0, end_mask = var_8716_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8716_cast")];
            tensor<int32, [4]> var_8720_begin_0 = const()[name = tensor<string, []>("op_8720_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_8720_end_0 = const()[name = tensor<string, []>("op_8720_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_8720_end_mask_0 = const()[name = tensor<string, []>("op_8720_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8720_cast = slice_by_index(begin = var_8720_begin_0, end = var_8720_end_0, end_mask = var_8720_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8720_cast")];
            tensor<int32, [4]> var_8724_begin_0 = const()[name = tensor<string, []>("op_8724_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_8724_end_0 = const()[name = tensor<string, []>("op_8724_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_8724_end_mask_0 = const()[name = tensor<string, []>("op_8724_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8724_cast = slice_by_index(begin = var_8724_begin_0, end = var_8724_end_0, end_mask = var_8724_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8724_cast")];
            tensor<int32, [4]> var_8728_begin_0 = const()[name = tensor<string, []>("op_8728_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_8728_end_0 = const()[name = tensor<string, []>("op_8728_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_8728_end_mask_0 = const()[name = tensor<string, []>("op_8728_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8728_cast = slice_by_index(begin = var_8728_begin_0, end = var_8728_end_0, end_mask = var_8728_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8728_cast")];
            tensor<int32, [4]> var_8732_begin_0 = const()[name = tensor<string, []>("op_8732_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_8732_end_0 = const()[name = tensor<string, []>("op_8732_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_8732_end_mask_0 = const()[name = tensor<string, []>("op_8732_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8732_cast = slice_by_index(begin = var_8732_begin_0, end = var_8732_end_0, end_mask = var_8732_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8732_cast")];
            tensor<int32, [4]> var_8736_begin_0 = const()[name = tensor<string, []>("op_8736_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_8736_end_0 = const()[name = tensor<string, []>("op_8736_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_8736_end_mask_0 = const()[name = tensor<string, []>("op_8736_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8736_cast = slice_by_index(begin = var_8736_begin_0, end = var_8736_end_0, end_mask = var_8736_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8736_cast")];
            tensor<int32, [4]> var_8740_begin_0 = const()[name = tensor<string, []>("op_8740_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_8740_end_0 = const()[name = tensor<string, []>("op_8740_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_8740_end_mask_0 = const()[name = tensor<string, []>("op_8740_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8740_cast = slice_by_index(begin = var_8740_begin_0, end = var_8740_end_0, end_mask = var_8740_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8740_cast")];
            tensor<int32, [4]> var_8744_begin_0 = const()[name = tensor<string, []>("op_8744_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_8744_end_0 = const()[name = tensor<string, []>("op_8744_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_8744_end_mask_0 = const()[name = tensor<string, []>("op_8744_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8744_cast = slice_by_index(begin = var_8744_begin_0, end = var_8744_end_0, end_mask = var_8744_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8744_cast")];
            tensor<int32, [4]> var_8748_begin_0 = const()[name = tensor<string, []>("op_8748_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_8748_end_0 = const()[name = tensor<string, []>("op_8748_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_8748_end_mask_0 = const()[name = tensor<string, []>("op_8748_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8748_cast = slice_by_index(begin = var_8748_begin_0, end = var_8748_end_0, end_mask = var_8748_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8748_cast")];
            tensor<int32, [4]> var_8752_begin_0 = const()[name = tensor<string, []>("op_8752_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_8752_end_0 = const()[name = tensor<string, []>("op_8752_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_8752_end_mask_0 = const()[name = tensor<string, []>("op_8752_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8752_cast = slice_by_index(begin = var_8752_begin_0, end = var_8752_end_0, end_mask = var_8752_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8752_cast")];
            tensor<int32, [4]> k_87_perm_0 = const()[name = tensor<string, []>("k_87_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_8759_begin_0 = const()[name = tensor<string, []>("op_8759_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8759_end_0 = const()[name = tensor<string, []>("op_8759_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_8759_end_mask_0 = const()[name = tensor<string, []>("op_8759_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_118 = transpose(perm = k_87_perm_0, x = k_85_cast)[name = tensor<string, []>("transpose_118")];
            tensor<fp16, [2, 77, 1, 64]> var_8759_cast = slice_by_index(begin = var_8759_begin_0, end = var_8759_end_0, end_mask = var_8759_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8759_cast")];
            tensor<int32, [4]> var_8763_begin_0 = const()[name = tensor<string, []>("op_8763_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_8763_end_0 = const()[name = tensor<string, []>("op_8763_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_8763_end_mask_0 = const()[name = tensor<string, []>("op_8763_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8763_cast = slice_by_index(begin = var_8763_begin_0, end = var_8763_end_0, end_mask = var_8763_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8763_cast")];
            tensor<int32, [4]> var_8767_begin_0 = const()[name = tensor<string, []>("op_8767_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_8767_end_0 = const()[name = tensor<string, []>("op_8767_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_8767_end_mask_0 = const()[name = tensor<string, []>("op_8767_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8767_cast = slice_by_index(begin = var_8767_begin_0, end = var_8767_end_0, end_mask = var_8767_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8767_cast")];
            tensor<int32, [4]> var_8771_begin_0 = const()[name = tensor<string, []>("op_8771_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_8771_end_0 = const()[name = tensor<string, []>("op_8771_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_8771_end_mask_0 = const()[name = tensor<string, []>("op_8771_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8771_cast = slice_by_index(begin = var_8771_begin_0, end = var_8771_end_0, end_mask = var_8771_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8771_cast")];
            tensor<int32, [4]> var_8775_begin_0 = const()[name = tensor<string, []>("op_8775_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_8775_end_0 = const()[name = tensor<string, []>("op_8775_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_8775_end_mask_0 = const()[name = tensor<string, []>("op_8775_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8775_cast = slice_by_index(begin = var_8775_begin_0, end = var_8775_end_0, end_mask = var_8775_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8775_cast")];
            tensor<int32, [4]> var_8779_begin_0 = const()[name = tensor<string, []>("op_8779_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_8779_end_0 = const()[name = tensor<string, []>("op_8779_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_8779_end_mask_0 = const()[name = tensor<string, []>("op_8779_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8779_cast = slice_by_index(begin = var_8779_begin_0, end = var_8779_end_0, end_mask = var_8779_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8779_cast")];
            tensor<int32, [4]> var_8783_begin_0 = const()[name = tensor<string, []>("op_8783_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_8783_end_0 = const()[name = tensor<string, []>("op_8783_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_8783_end_mask_0 = const()[name = tensor<string, []>("op_8783_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8783_cast = slice_by_index(begin = var_8783_begin_0, end = var_8783_end_0, end_mask = var_8783_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8783_cast")];
            tensor<int32, [4]> var_8787_begin_0 = const()[name = tensor<string, []>("op_8787_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_8787_end_0 = const()[name = tensor<string, []>("op_8787_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_8787_end_mask_0 = const()[name = tensor<string, []>("op_8787_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8787_cast = slice_by_index(begin = var_8787_begin_0, end = var_8787_end_0, end_mask = var_8787_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8787_cast")];
            tensor<int32, [4]> var_8791_begin_0 = const()[name = tensor<string, []>("op_8791_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8791_end_0 = const()[name = tensor<string, []>("op_8791_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_8791_end_mask_0 = const()[name = tensor<string, []>("op_8791_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8791_cast = slice_by_index(begin = var_8791_begin_0, end = var_8791_end_0, end_mask = var_8791_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8791_cast")];
            tensor<int32, [4]> var_8795_begin_0 = const()[name = tensor<string, []>("op_8795_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_8795_end_0 = const()[name = tensor<string, []>("op_8795_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_8795_end_mask_0 = const()[name = tensor<string, []>("op_8795_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8795_cast = slice_by_index(begin = var_8795_begin_0, end = var_8795_end_0, end_mask = var_8795_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8795_cast")];
            tensor<int32, [4]> var_8799_begin_0 = const()[name = tensor<string, []>("op_8799_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_8799_end_0 = const()[name = tensor<string, []>("op_8799_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_8799_end_mask_0 = const()[name = tensor<string, []>("op_8799_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8799_cast = slice_by_index(begin = var_8799_begin_0, end = var_8799_end_0, end_mask = var_8799_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8799_cast")];
            tensor<int32, [4]> var_8803_begin_0 = const()[name = tensor<string, []>("op_8803_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_8803_end_0 = const()[name = tensor<string, []>("op_8803_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_8803_end_mask_0 = const()[name = tensor<string, []>("op_8803_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8803_cast = slice_by_index(begin = var_8803_begin_0, end = var_8803_end_0, end_mask = var_8803_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8803_cast")];
            tensor<int32, [4]> var_8807_begin_0 = const()[name = tensor<string, []>("op_8807_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_8807_end_0 = const()[name = tensor<string, []>("op_8807_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_8807_end_mask_0 = const()[name = tensor<string, []>("op_8807_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8807_cast = slice_by_index(begin = var_8807_begin_0, end = var_8807_end_0, end_mask = var_8807_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8807_cast")];
            tensor<int32, [4]> var_8811_begin_0 = const()[name = tensor<string, []>("op_8811_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_8811_end_0 = const()[name = tensor<string, []>("op_8811_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_8811_end_mask_0 = const()[name = tensor<string, []>("op_8811_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8811_cast = slice_by_index(begin = var_8811_begin_0, end = var_8811_end_0, end_mask = var_8811_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8811_cast")];
            tensor<int32, [4]> var_8815_begin_0 = const()[name = tensor<string, []>("op_8815_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_8815_end_0 = const()[name = tensor<string, []>("op_8815_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_8815_end_mask_0 = const()[name = tensor<string, []>("op_8815_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8815_cast = slice_by_index(begin = var_8815_begin_0, end = var_8815_end_0, end_mask = var_8815_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8815_cast")];
            tensor<int32, [4]> var_8819_begin_0 = const()[name = tensor<string, []>("op_8819_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_8819_end_0 = const()[name = tensor<string, []>("op_8819_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_8819_end_mask_0 = const()[name = tensor<string, []>("op_8819_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8819_cast = slice_by_index(begin = var_8819_begin_0, end = var_8819_end_0, end_mask = var_8819_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8819_cast")];
            tensor<int32, [4]> var_8823_begin_0 = const()[name = tensor<string, []>("op_8823_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_8823_end_0 = const()[name = tensor<string, []>("op_8823_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_8823_end_mask_0 = const()[name = tensor<string, []>("op_8823_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8823_cast = slice_by_index(begin = var_8823_begin_0, end = var_8823_end_0, end_mask = var_8823_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8823_cast")];
            tensor<int32, [4]> var_8827_begin_0 = const()[name = tensor<string, []>("op_8827_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_8827_end_0 = const()[name = tensor<string, []>("op_8827_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_8827_end_mask_0 = const()[name = tensor<string, []>("op_8827_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8827_cast = slice_by_index(begin = var_8827_begin_0, end = var_8827_end_0, end_mask = var_8827_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8827_cast")];
            tensor<int32, [4]> var_8831_begin_0 = const()[name = tensor<string, []>("op_8831_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_8831_end_0 = const()[name = tensor<string, []>("op_8831_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_8831_end_mask_0 = const()[name = tensor<string, []>("op_8831_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8831_cast = slice_by_index(begin = var_8831_begin_0, end = var_8831_end_0, end_mask = var_8831_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8831_cast")];
            tensor<int32, [4]> var_8835_begin_0 = const()[name = tensor<string, []>("op_8835_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_8835_end_0 = const()[name = tensor<string, []>("op_8835_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_8835_end_mask_0 = const()[name = tensor<string, []>("op_8835_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8835_cast = slice_by_index(begin = var_8835_begin_0, end = var_8835_end_0, end_mask = var_8835_end_mask_0, x = transpose_118)[name = tensor<string, []>("op_8835_cast")];
            tensor<int32, [4]> var_8837_begin_0 = const()[name = tensor<string, []>("op_8837_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8837_end_0 = const()[name = tensor<string, []>("op_8837_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_8837_end_mask_0 = const()[name = tensor<string, []>("op_8837_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8837_cast = slice_by_index(begin = var_8837_begin_0, end = var_8837_end_0, end_mask = var_8837_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8837_cast")];
            tensor<int32, [4]> var_8841_begin_0 = const()[name = tensor<string, []>("op_8841_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_8841_end_0 = const()[name = tensor<string, []>("op_8841_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_8841_end_mask_0 = const()[name = tensor<string, []>("op_8841_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8841_cast = slice_by_index(begin = var_8841_begin_0, end = var_8841_end_0, end_mask = var_8841_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8841_cast")];
            tensor<int32, [4]> var_8845_begin_0 = const()[name = tensor<string, []>("op_8845_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_8845_end_0 = const()[name = tensor<string, []>("op_8845_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_8845_end_mask_0 = const()[name = tensor<string, []>("op_8845_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8845_cast = slice_by_index(begin = var_8845_begin_0, end = var_8845_end_0, end_mask = var_8845_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8845_cast")];
            tensor<int32, [4]> var_8849_begin_0 = const()[name = tensor<string, []>("op_8849_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_8849_end_0 = const()[name = tensor<string, []>("op_8849_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_8849_end_mask_0 = const()[name = tensor<string, []>("op_8849_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8849_cast = slice_by_index(begin = var_8849_begin_0, end = var_8849_end_0, end_mask = var_8849_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8849_cast")];
            tensor<int32, [4]> var_8853_begin_0 = const()[name = tensor<string, []>("op_8853_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_8853_end_0 = const()[name = tensor<string, []>("op_8853_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_8853_end_mask_0 = const()[name = tensor<string, []>("op_8853_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8853_cast = slice_by_index(begin = var_8853_begin_0, end = var_8853_end_0, end_mask = var_8853_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8853_cast")];
            tensor<int32, [4]> var_8857_begin_0 = const()[name = tensor<string, []>("op_8857_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_8857_end_0 = const()[name = tensor<string, []>("op_8857_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_8857_end_mask_0 = const()[name = tensor<string, []>("op_8857_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8857_cast = slice_by_index(begin = var_8857_begin_0, end = var_8857_end_0, end_mask = var_8857_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8857_cast")];
            tensor<int32, [4]> var_8861_begin_0 = const()[name = tensor<string, []>("op_8861_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_8861_end_0 = const()[name = tensor<string, []>("op_8861_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_8861_end_mask_0 = const()[name = tensor<string, []>("op_8861_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8861_cast = slice_by_index(begin = var_8861_begin_0, end = var_8861_end_0, end_mask = var_8861_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8861_cast")];
            tensor<int32, [4]> var_8865_begin_0 = const()[name = tensor<string, []>("op_8865_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_8865_end_0 = const()[name = tensor<string, []>("op_8865_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_8865_end_mask_0 = const()[name = tensor<string, []>("op_8865_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8865_cast = slice_by_index(begin = var_8865_begin_0, end = var_8865_end_0, end_mask = var_8865_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8865_cast")];
            tensor<int32, [4]> var_8869_begin_0 = const()[name = tensor<string, []>("op_8869_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_8869_end_0 = const()[name = tensor<string, []>("op_8869_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_8869_end_mask_0 = const()[name = tensor<string, []>("op_8869_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8869_cast = slice_by_index(begin = var_8869_begin_0, end = var_8869_end_0, end_mask = var_8869_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8869_cast")];
            tensor<int32, [4]> var_8873_begin_0 = const()[name = tensor<string, []>("op_8873_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_8873_end_0 = const()[name = tensor<string, []>("op_8873_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_8873_end_mask_0 = const()[name = tensor<string, []>("op_8873_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8873_cast = slice_by_index(begin = var_8873_begin_0, end = var_8873_end_0, end_mask = var_8873_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8873_cast")];
            tensor<int32, [4]> var_8877_begin_0 = const()[name = tensor<string, []>("op_8877_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_8877_end_0 = const()[name = tensor<string, []>("op_8877_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_8877_end_mask_0 = const()[name = tensor<string, []>("op_8877_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8877_cast = slice_by_index(begin = var_8877_begin_0, end = var_8877_end_0, end_mask = var_8877_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8877_cast")];
            tensor<int32, [4]> var_8881_begin_0 = const()[name = tensor<string, []>("op_8881_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_8881_end_0 = const()[name = tensor<string, []>("op_8881_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_8881_end_mask_0 = const()[name = tensor<string, []>("op_8881_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8881_cast = slice_by_index(begin = var_8881_begin_0, end = var_8881_end_0, end_mask = var_8881_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8881_cast")];
            tensor<int32, [4]> var_8885_begin_0 = const()[name = tensor<string, []>("op_8885_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_8885_end_0 = const()[name = tensor<string, []>("op_8885_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_8885_end_mask_0 = const()[name = tensor<string, []>("op_8885_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8885_cast = slice_by_index(begin = var_8885_begin_0, end = var_8885_end_0, end_mask = var_8885_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8885_cast")];
            tensor<int32, [4]> var_8889_begin_0 = const()[name = tensor<string, []>("op_8889_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_8889_end_0 = const()[name = tensor<string, []>("op_8889_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_8889_end_mask_0 = const()[name = tensor<string, []>("op_8889_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8889_cast = slice_by_index(begin = var_8889_begin_0, end = var_8889_end_0, end_mask = var_8889_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8889_cast")];
            tensor<int32, [4]> var_8893_begin_0 = const()[name = tensor<string, []>("op_8893_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_8893_end_0 = const()[name = tensor<string, []>("op_8893_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_8893_end_mask_0 = const()[name = tensor<string, []>("op_8893_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8893_cast = slice_by_index(begin = var_8893_begin_0, end = var_8893_end_0, end_mask = var_8893_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8893_cast")];
            tensor<int32, [4]> var_8897_begin_0 = const()[name = tensor<string, []>("op_8897_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_8897_end_0 = const()[name = tensor<string, []>("op_8897_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_8897_end_mask_0 = const()[name = tensor<string, []>("op_8897_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8897_cast = slice_by_index(begin = var_8897_begin_0, end = var_8897_end_0, end_mask = var_8897_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8897_cast")];
            tensor<int32, [4]> var_8901_begin_0 = const()[name = tensor<string, []>("op_8901_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_8901_end_0 = const()[name = tensor<string, []>("op_8901_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_8901_end_mask_0 = const()[name = tensor<string, []>("op_8901_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8901_cast = slice_by_index(begin = var_8901_begin_0, end = var_8901_end_0, end_mask = var_8901_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8901_cast")];
            tensor<int32, [4]> var_8905_begin_0 = const()[name = tensor<string, []>("op_8905_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_8905_end_0 = const()[name = tensor<string, []>("op_8905_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_8905_end_mask_0 = const()[name = tensor<string, []>("op_8905_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8905_cast = slice_by_index(begin = var_8905_begin_0, end = var_8905_end_0, end_mask = var_8905_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8905_cast")];
            tensor<int32, [4]> var_8909_begin_0 = const()[name = tensor<string, []>("op_8909_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_8909_end_0 = const()[name = tensor<string, []>("op_8909_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_8909_end_mask_0 = const()[name = tensor<string, []>("op_8909_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8909_cast = slice_by_index(begin = var_8909_begin_0, end = var_8909_end_0, end_mask = var_8909_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8909_cast")];
            tensor<int32, [4]> var_8913_begin_0 = const()[name = tensor<string, []>("op_8913_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_8913_end_0 = const()[name = tensor<string, []>("op_8913_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_8913_end_mask_0 = const()[name = tensor<string, []>("op_8913_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8913_cast = slice_by_index(begin = var_8913_begin_0, end = var_8913_end_0, end_mask = var_8913_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8913_cast")];
            tensor<string, []> var_8917_equation_0 = const()[name = tensor<string, []>("op_8917_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8917_cast = einsum(equation = var_8917_equation_0, values = (var_8759_cast, var_8676_cast))[name = tensor<string, []>("op_8917_cast")];
            tensor<fp16, []> var_8918_to_fp16 = const()[name = tensor<string, []>("op_8918_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_681_cast = mul(x = var_8917_cast, y = var_8918_to_fp16)[name = tensor<string, []>("aw_681_cast")];
            tensor<string, []> var_8921_equation_0 = const()[name = tensor<string, []>("op_8921_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8921_cast = einsum(equation = var_8921_equation_0, values = (var_8763_cast, var_8680_cast))[name = tensor<string, []>("op_8921_cast")];
            tensor<fp16, []> var_8922_to_fp16 = const()[name = tensor<string, []>("op_8922_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_683_cast = mul(x = var_8921_cast, y = var_8922_to_fp16)[name = tensor<string, []>("aw_683_cast")];
            tensor<string, []> var_8925_equation_0 = const()[name = tensor<string, []>("op_8925_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8925_cast = einsum(equation = var_8925_equation_0, values = (var_8767_cast, var_8684_cast))[name = tensor<string, []>("op_8925_cast")];
            tensor<fp16, []> var_8926_to_fp16 = const()[name = tensor<string, []>("op_8926_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_685_cast = mul(x = var_8925_cast, y = var_8926_to_fp16)[name = tensor<string, []>("aw_685_cast")];
            tensor<string, []> var_8929_equation_0 = const()[name = tensor<string, []>("op_8929_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8929_cast = einsum(equation = var_8929_equation_0, values = (var_8771_cast, var_8688_cast))[name = tensor<string, []>("op_8929_cast")];
            tensor<fp16, []> var_8930_to_fp16 = const()[name = tensor<string, []>("op_8930_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_687_cast = mul(x = var_8929_cast, y = var_8930_to_fp16)[name = tensor<string, []>("aw_687_cast")];
            tensor<string, []> var_8933_equation_0 = const()[name = tensor<string, []>("op_8933_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8933_cast = einsum(equation = var_8933_equation_0, values = (var_8775_cast, var_8692_cast))[name = tensor<string, []>("op_8933_cast")];
            tensor<fp16, []> var_8934_to_fp16 = const()[name = tensor<string, []>("op_8934_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_689_cast = mul(x = var_8933_cast, y = var_8934_to_fp16)[name = tensor<string, []>("aw_689_cast")];
            tensor<string, []> var_8937_equation_0 = const()[name = tensor<string, []>("op_8937_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8937_cast = einsum(equation = var_8937_equation_0, values = (var_8779_cast, var_8696_cast))[name = tensor<string, []>("op_8937_cast")];
            tensor<fp16, []> var_8938_to_fp16 = const()[name = tensor<string, []>("op_8938_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_691_cast = mul(x = var_8937_cast, y = var_8938_to_fp16)[name = tensor<string, []>("aw_691_cast")];
            tensor<string, []> var_8941_equation_0 = const()[name = tensor<string, []>("op_8941_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8941_cast = einsum(equation = var_8941_equation_0, values = (var_8783_cast, var_8700_cast))[name = tensor<string, []>("op_8941_cast")];
            tensor<fp16, []> var_8942_to_fp16 = const()[name = tensor<string, []>("op_8942_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_693_cast = mul(x = var_8941_cast, y = var_8942_to_fp16)[name = tensor<string, []>("aw_693_cast")];
            tensor<string, []> var_8945_equation_0 = const()[name = tensor<string, []>("op_8945_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8945_cast = einsum(equation = var_8945_equation_0, values = (var_8787_cast, var_8704_cast))[name = tensor<string, []>("op_8945_cast")];
            tensor<fp16, []> var_8946_to_fp16 = const()[name = tensor<string, []>("op_8946_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_695_cast = mul(x = var_8945_cast, y = var_8946_to_fp16)[name = tensor<string, []>("aw_695_cast")];
            tensor<string, []> var_8949_equation_0 = const()[name = tensor<string, []>("op_8949_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8949_cast = einsum(equation = var_8949_equation_0, values = (var_8791_cast, var_8708_cast))[name = tensor<string, []>("op_8949_cast")];
            tensor<fp16, []> var_8950_to_fp16 = const()[name = tensor<string, []>("op_8950_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_697_cast = mul(x = var_8949_cast, y = var_8950_to_fp16)[name = tensor<string, []>("aw_697_cast")];
            tensor<string, []> var_8953_equation_0 = const()[name = tensor<string, []>("op_8953_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8953_cast = einsum(equation = var_8953_equation_0, values = (var_8795_cast, var_8712_cast))[name = tensor<string, []>("op_8953_cast")];
            tensor<fp16, []> var_8954_to_fp16 = const()[name = tensor<string, []>("op_8954_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_699_cast = mul(x = var_8953_cast, y = var_8954_to_fp16)[name = tensor<string, []>("aw_699_cast")];
            tensor<string, []> var_8957_equation_0 = const()[name = tensor<string, []>("op_8957_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8957_cast = einsum(equation = var_8957_equation_0, values = (var_8799_cast, var_8716_cast))[name = tensor<string, []>("op_8957_cast")];
            tensor<fp16, []> var_8958_to_fp16 = const()[name = tensor<string, []>("op_8958_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_701_cast = mul(x = var_8957_cast, y = var_8958_to_fp16)[name = tensor<string, []>("aw_701_cast")];
            tensor<string, []> var_8961_equation_0 = const()[name = tensor<string, []>("op_8961_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8961_cast = einsum(equation = var_8961_equation_0, values = (var_8803_cast, var_8720_cast))[name = tensor<string, []>("op_8961_cast")];
            tensor<fp16, []> var_8962_to_fp16 = const()[name = tensor<string, []>("op_8962_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_703_cast = mul(x = var_8961_cast, y = var_8962_to_fp16)[name = tensor<string, []>("aw_703_cast")];
            tensor<string, []> var_8965_equation_0 = const()[name = tensor<string, []>("op_8965_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8965_cast = einsum(equation = var_8965_equation_0, values = (var_8807_cast, var_8724_cast))[name = tensor<string, []>("op_8965_cast")];
            tensor<fp16, []> var_8966_to_fp16 = const()[name = tensor<string, []>("op_8966_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_705_cast = mul(x = var_8965_cast, y = var_8966_to_fp16)[name = tensor<string, []>("aw_705_cast")];
            tensor<string, []> var_8969_equation_0 = const()[name = tensor<string, []>("op_8969_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8969_cast = einsum(equation = var_8969_equation_0, values = (var_8811_cast, var_8728_cast))[name = tensor<string, []>("op_8969_cast")];
            tensor<fp16, []> var_8970_to_fp16 = const()[name = tensor<string, []>("op_8970_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_707_cast = mul(x = var_8969_cast, y = var_8970_to_fp16)[name = tensor<string, []>("aw_707_cast")];
            tensor<string, []> var_8973_equation_0 = const()[name = tensor<string, []>("op_8973_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8973_cast = einsum(equation = var_8973_equation_0, values = (var_8815_cast, var_8732_cast))[name = tensor<string, []>("op_8973_cast")];
            tensor<fp16, []> var_8974_to_fp16 = const()[name = tensor<string, []>("op_8974_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_709_cast = mul(x = var_8973_cast, y = var_8974_to_fp16)[name = tensor<string, []>("aw_709_cast")];
            tensor<string, []> var_8977_equation_0 = const()[name = tensor<string, []>("op_8977_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8977_cast = einsum(equation = var_8977_equation_0, values = (var_8819_cast, var_8736_cast))[name = tensor<string, []>("op_8977_cast")];
            tensor<fp16, []> var_8978_to_fp16 = const()[name = tensor<string, []>("op_8978_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_711_cast = mul(x = var_8977_cast, y = var_8978_to_fp16)[name = tensor<string, []>("aw_711_cast")];
            tensor<string, []> var_8981_equation_0 = const()[name = tensor<string, []>("op_8981_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8981_cast = einsum(equation = var_8981_equation_0, values = (var_8823_cast, var_8740_cast))[name = tensor<string, []>("op_8981_cast")];
            tensor<fp16, []> var_8982_to_fp16 = const()[name = tensor<string, []>("op_8982_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_713_cast = mul(x = var_8981_cast, y = var_8982_to_fp16)[name = tensor<string, []>("aw_713_cast")];
            tensor<string, []> var_8985_equation_0 = const()[name = tensor<string, []>("op_8985_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8985_cast = einsum(equation = var_8985_equation_0, values = (var_8827_cast, var_8744_cast))[name = tensor<string, []>("op_8985_cast")];
            tensor<fp16, []> var_8986_to_fp16 = const()[name = tensor<string, []>("op_8986_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_715_cast = mul(x = var_8985_cast, y = var_8986_to_fp16)[name = tensor<string, []>("aw_715_cast")];
            tensor<string, []> var_8989_equation_0 = const()[name = tensor<string, []>("op_8989_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8989_cast = einsum(equation = var_8989_equation_0, values = (var_8831_cast, var_8748_cast))[name = tensor<string, []>("op_8989_cast")];
            tensor<fp16, []> var_8990_to_fp16 = const()[name = tensor<string, []>("op_8990_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_717_cast = mul(x = var_8989_cast, y = var_8990_to_fp16)[name = tensor<string, []>("aw_717_cast")];
            tensor<string, []> var_8993_equation_0 = const()[name = tensor<string, []>("op_8993_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_8993_cast = einsum(equation = var_8993_equation_0, values = (var_8835_cast, var_8752_cast))[name = tensor<string, []>("op_8993_cast")];
            tensor<fp16, []> var_8994_to_fp16 = const()[name = tensor<string, []>("op_8994_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_719_cast = mul(x = var_8993_cast, y = var_8994_to_fp16)[name = tensor<string, []>("aw_719_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8996_cast = softmax(axis = var_2634, x = aw_681_cast)[name = tensor<string, []>("op_8996_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8997_cast = softmax(axis = var_2634, x = aw_683_cast)[name = tensor<string, []>("op_8997_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8998_cast = softmax(axis = var_2634, x = aw_685_cast)[name = tensor<string, []>("op_8998_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_8999_cast = softmax(axis = var_2634, x = aw_687_cast)[name = tensor<string, []>("op_8999_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9000_cast = softmax(axis = var_2634, x = aw_689_cast)[name = tensor<string, []>("op_9000_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9001_cast = softmax(axis = var_2634, x = aw_691_cast)[name = tensor<string, []>("op_9001_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9002_cast = softmax(axis = var_2634, x = aw_693_cast)[name = tensor<string, []>("op_9002_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9003_cast = softmax(axis = var_2634, x = aw_695_cast)[name = tensor<string, []>("op_9003_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9004_cast = softmax(axis = var_2634, x = aw_697_cast)[name = tensor<string, []>("op_9004_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9005_cast = softmax(axis = var_2634, x = aw_699_cast)[name = tensor<string, []>("op_9005_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9006_cast = softmax(axis = var_2634, x = aw_701_cast)[name = tensor<string, []>("op_9006_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9007_cast = softmax(axis = var_2634, x = aw_703_cast)[name = tensor<string, []>("op_9007_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9008_cast = softmax(axis = var_2634, x = aw_705_cast)[name = tensor<string, []>("op_9008_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9009_cast = softmax(axis = var_2634, x = aw_707_cast)[name = tensor<string, []>("op_9009_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9010_cast = softmax(axis = var_2634, x = aw_709_cast)[name = tensor<string, []>("op_9010_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9011_cast = softmax(axis = var_2634, x = aw_711_cast)[name = tensor<string, []>("op_9011_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9012_cast = softmax(axis = var_2634, x = aw_713_cast)[name = tensor<string, []>("op_9012_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9013_cast = softmax(axis = var_2634, x = aw_715_cast)[name = tensor<string, []>("op_9013_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9014_cast = softmax(axis = var_2634, x = aw_717_cast)[name = tensor<string, []>("op_9014_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9015_cast = softmax(axis = var_2634, x = aw_719_cast)[name = tensor<string, []>("op_9015_cast")];
            tensor<string, []> var_9017_equation_0 = const()[name = tensor<string, []>("op_9017_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9017_cast = einsum(equation = var_9017_equation_0, values = (var_8837_cast, var_8996_cast))[name = tensor<string, []>("op_9017_cast")];
            tensor<string, []> var_9019_equation_0 = const()[name = tensor<string, []>("op_9019_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9019_cast = einsum(equation = var_9019_equation_0, values = (var_8841_cast, var_8997_cast))[name = tensor<string, []>("op_9019_cast")];
            tensor<string, []> var_9021_equation_0 = const()[name = tensor<string, []>("op_9021_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9021_cast = einsum(equation = var_9021_equation_0, values = (var_8845_cast, var_8998_cast))[name = tensor<string, []>("op_9021_cast")];
            tensor<string, []> var_9023_equation_0 = const()[name = tensor<string, []>("op_9023_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9023_cast = einsum(equation = var_9023_equation_0, values = (var_8849_cast, var_8999_cast))[name = tensor<string, []>("op_9023_cast")];
            tensor<string, []> var_9025_equation_0 = const()[name = tensor<string, []>("op_9025_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9025_cast = einsum(equation = var_9025_equation_0, values = (var_8853_cast, var_9000_cast))[name = tensor<string, []>("op_9025_cast")];
            tensor<string, []> var_9027_equation_0 = const()[name = tensor<string, []>("op_9027_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9027_cast = einsum(equation = var_9027_equation_0, values = (var_8857_cast, var_9001_cast))[name = tensor<string, []>("op_9027_cast")];
            tensor<string, []> var_9029_equation_0 = const()[name = tensor<string, []>("op_9029_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9029_cast = einsum(equation = var_9029_equation_0, values = (var_8861_cast, var_9002_cast))[name = tensor<string, []>("op_9029_cast")];
            tensor<string, []> var_9031_equation_0 = const()[name = tensor<string, []>("op_9031_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9031_cast = einsum(equation = var_9031_equation_0, values = (var_8865_cast, var_9003_cast))[name = tensor<string, []>("op_9031_cast")];
            tensor<string, []> var_9033_equation_0 = const()[name = tensor<string, []>("op_9033_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9033_cast = einsum(equation = var_9033_equation_0, values = (var_8869_cast, var_9004_cast))[name = tensor<string, []>("op_9033_cast")];
            tensor<string, []> var_9035_equation_0 = const()[name = tensor<string, []>("op_9035_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9035_cast = einsum(equation = var_9035_equation_0, values = (var_8873_cast, var_9005_cast))[name = tensor<string, []>("op_9035_cast")];
            tensor<string, []> var_9037_equation_0 = const()[name = tensor<string, []>("op_9037_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9037_cast = einsum(equation = var_9037_equation_0, values = (var_8877_cast, var_9006_cast))[name = tensor<string, []>("op_9037_cast")];
            tensor<string, []> var_9039_equation_0 = const()[name = tensor<string, []>("op_9039_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9039_cast = einsum(equation = var_9039_equation_0, values = (var_8881_cast, var_9007_cast))[name = tensor<string, []>("op_9039_cast")];
            tensor<string, []> var_9041_equation_0 = const()[name = tensor<string, []>("op_9041_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9041_cast = einsum(equation = var_9041_equation_0, values = (var_8885_cast, var_9008_cast))[name = tensor<string, []>("op_9041_cast")];
            tensor<string, []> var_9043_equation_0 = const()[name = tensor<string, []>("op_9043_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9043_cast = einsum(equation = var_9043_equation_0, values = (var_8889_cast, var_9009_cast))[name = tensor<string, []>("op_9043_cast")];
            tensor<string, []> var_9045_equation_0 = const()[name = tensor<string, []>("op_9045_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9045_cast = einsum(equation = var_9045_equation_0, values = (var_8893_cast, var_9010_cast))[name = tensor<string, []>("op_9045_cast")];
            tensor<string, []> var_9047_equation_0 = const()[name = tensor<string, []>("op_9047_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9047_cast = einsum(equation = var_9047_equation_0, values = (var_8897_cast, var_9011_cast))[name = tensor<string, []>("op_9047_cast")];
            tensor<string, []> var_9049_equation_0 = const()[name = tensor<string, []>("op_9049_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9049_cast = einsum(equation = var_9049_equation_0, values = (var_8901_cast, var_9012_cast))[name = tensor<string, []>("op_9049_cast")];
            tensor<string, []> var_9051_equation_0 = const()[name = tensor<string, []>("op_9051_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9051_cast = einsum(equation = var_9051_equation_0, values = (var_8905_cast, var_9013_cast))[name = tensor<string, []>("op_9051_cast")];
            tensor<string, []> var_9053_equation_0 = const()[name = tensor<string, []>("op_9053_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9053_cast = einsum(equation = var_9053_equation_0, values = (var_8909_cast, var_9014_cast))[name = tensor<string, []>("op_9053_cast")];
            tensor<string, []> var_9055_equation_0 = const()[name = tensor<string, []>("op_9055_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9055_cast = einsum(equation = var_9055_equation_0, values = (var_8913_cast, var_9015_cast))[name = tensor<string, []>("op_9055_cast")];
            tensor<bool, []> input_185_interleave_0 = const()[name = tensor<string, []>("input_185_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_185_cast = concat(axis = var_2634, interleave = input_185_interleave_0, values = (var_9017_cast, var_9019_cast, var_9021_cast, var_9023_cast, var_9025_cast, var_9027_cast, var_9029_cast, var_9031_cast, var_9033_cast, var_9035_cast, var_9037_cast, var_9039_cast, var_9041_cast, var_9043_cast, var_9045_cast, var_9047_cast, var_9049_cast, var_9051_cast, var_9053_cast, var_9055_cast))[name = tensor<string, []>("input_185_cast")];
            tensor<int32, [2]> var_9061 = const()[name = tensor<string, []>("op_9061"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9063 = const()[name = tensor<string, []>("op_9063"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9065_pad_type_0 = const()[name = tensor<string, []>("op_9065_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9065_pad_0 = const()[name = tensor<string, []>("op_9065_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(642126976)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(645403840)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_9065_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16, dilations = var_9063, groups = var_2634, pad = var_9065_pad_0, pad_type = var_9065_pad_type_0, strides = var_9061, weight = down_blocks_2_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16, x = input_185_cast)[name = tensor<string, []>("op_9065_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_65_cast = add(x = var_9065_cast, y = inputs_63_cast)[name = tensor<string, []>("inputs_65_cast")];
            tensor<int32, [1]> var_9069 = const()[name = tensor<string, []>("op_9069"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_65_cast = reduce_mean(axes = var_9069, keep_dims = var_2629, x = inputs_65_cast)[name = tensor<string, []>("channels_mean_65_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_65_cast = sub(x = inputs_65_cast, y = channels_mean_65_cast)[name = tensor<string, []>("zero_mean_65_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_65_cast = mul(x = zero_mean_65_cast, y = zero_mean_65_cast)[name = tensor<string, []>("zero_mean_sq_65_cast")];
            tensor<int32, [1]> var_9073 = const()[name = tensor<string, []>("op_9073"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_9074_cast = reduce_mean(axes = var_9073, keep_dims = var_2629, x = zero_mean_sq_65_cast)[name = tensor<string, []>("op_9074_cast")];
            tensor<fp16, []> var_9075_to_fp16 = const()[name = tensor<string, []>("op_9075_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_9076_cast = add(x = var_9074_cast, y = var_9075_to_fp16)[name = tensor<string, []>("op_9076_cast")];
            tensor<fp16, []> denom_65_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_65_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_65_cast = rsqrt(epsilon = denom_65_epsilon_0_to_fp16, x = var_9076_cast)[name = tensor<string, []>("denom_65_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_65_cast = mul(x = zero_mean_65_cast, y = denom_65_cast)[name = tensor<string, []>("out_65_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_9080_to_fp16 = const()[name = tensor<string, []>("op_9080_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(645406464)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_9081_cast = add(x = out_65_cast, y = var_9080_to_fp16)[name = tensor<string, []>("op_9081_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_9083_to_fp16 = const()[name = tensor<string, []>("op_9083_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(645409088)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_187_cast = mul(x = var_9081_cast, y = var_9083_to_fp16)[name = tensor<string, []>("input_187_cast")];
            tensor<int32, [2]> var_9091 = const()[name = tensor<string, []>("op_9091"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9093 = const()[name = tensor<string, []>("op_9093"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9095_pad_type_0 = const()[name = tensor<string, []>("op_9095_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9095_pad_0 = const()[name = tensor<string, []>("op_9095_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(645411712)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(671626176)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_9095_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16, dilations = var_9093, groups = var_2634, pad = var_9095_pad_0, pad_type = var_9095_pad_type_0, strides = var_9091, weight = down_blocks_2_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16, x = input_187_cast)[name = tensor<string, []>("op_9095_cast")];
            tensor<int32, [2]> var_9096_split_sizes_0 = const()[name = tensor<string, []>("op_9096_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_9096_axis_0 = const()[name = tensor<string, []>("op_9096_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_9096_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_9096_cast_1 = split(axis = var_9096_axis_0, split_sizes = var_9096_split_sizes_0, x = var_9095_cast)[name = tensor<string, []>("op_9096_cast")];
            tensor<string, []> var_9098_mode_0 = const()[name = tensor<string, []>("op_9098_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_9098_cast = gelu(mode = var_9098_mode_0, x = var_9096_cast_1)[name = tensor<string, []>("op_9098_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_189_cast = mul(x = var_9096_cast_0, y = var_9098_cast)[name = tensor<string, []>("input_189_cast")];
            tensor<int32, [2]> var_9102 = const()[name = tensor<string, []>("op_9102"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9104 = const()[name = tensor<string, []>("op_9104"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9106_pad_type_0 = const()[name = tensor<string, []>("op_9106_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9106_pad_0 = const()[name = tensor<string, []>("op_9106_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(671646720)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(684753984)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_9106_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16, dilations = var_9104, groups = var_2634, pad = var_9106_pad_0, pad_type = var_9106_pad_type_0, strides = var_9102, weight = down_blocks_2_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16, x = input_189_cast)[name = tensor<string, []>("op_9106_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_67_cast = add(x = var_9106_cast, y = inputs_65_cast)[name = tensor<string, []>("inputs_67_cast")];
            tensor<int32, [1]> var_9116 = const()[name = tensor<string, []>("op_9116"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_67_cast = reduce_mean(axes = var_9116, keep_dims = var_2629, x = inputs_67_cast)[name = tensor<string, []>("channels_mean_67_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_67_cast = sub(x = inputs_67_cast, y = channels_mean_67_cast)[name = tensor<string, []>("zero_mean_67_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_67_cast = mul(x = zero_mean_67_cast, y = zero_mean_67_cast)[name = tensor<string, []>("zero_mean_sq_67_cast")];
            tensor<int32, [1]> var_9120 = const()[name = tensor<string, []>("op_9120"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_9121_cast = reduce_mean(axes = var_9120, keep_dims = var_2629, x = zero_mean_sq_67_cast)[name = tensor<string, []>("op_9121_cast")];
            tensor<fp16, []> var_9122_to_fp16 = const()[name = tensor<string, []>("op_9122_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_9123_cast = add(x = var_9121_cast, y = var_9122_to_fp16)[name = tensor<string, []>("op_9123_cast")];
            tensor<fp16, []> denom_67_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_67_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_67_cast = rsqrt(epsilon = denom_67_epsilon_0_to_fp16, x = var_9123_cast)[name = tensor<string, []>("denom_67_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_67_cast = mul(x = zero_mean_67_cast, y = denom_67_cast)[name = tensor<string, []>("out_67_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_9127_to_fp16 = const()[name = tensor<string, []>("op_9127_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(684756608)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_9128_cast = add(x = out_67_cast, y = var_9127_to_fp16)[name = tensor<string, []>("op_9128_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_9130_to_fp16 = const()[name = tensor<string, []>("op_9130_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(684759232)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_107_cast = mul(x = var_9128_cast, y = var_9130_to_fp16)[name = tensor<string, []>("hidden_states_107_cast")];
            tensor<int32, [2]> var_9137 = const()[name = tensor<string, []>("op_9137"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9139 = const()[name = tensor<string, []>("op_9139"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_45_pad_type_0 = const()[name = tensor<string, []>("q_45_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_45_pad_0 = const()[name = tensor<string, []>("q_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(684761856)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_45_cast = conv(dilations = var_9139, groups = var_2634, pad = q_45_pad_0, pad_type = q_45_pad_type_0, strides = var_9137, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16, x = hidden_states_107_cast)[name = tensor<string, []>("q_45_cast")];
            tensor<int32, [2]> var_9143 = const()[name = tensor<string, []>("op_9143"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9145 = const()[name = tensor<string, []>("op_9145"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_89_pad_type_0 = const()[name = tensor<string, []>("k_89_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_89_pad_0 = const()[name = tensor<string, []>("k_89_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(688038720)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_89_cast = conv(dilations = var_9145, groups = var_2634, pad = k_89_pad_0, pad_type = k_89_pad_type_0, strides = var_9143, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16, x = hidden_states_107_cast)[name = tensor<string, []>("k_89_cast")];
            tensor<int32, [2]> var_9149 = const()[name = tensor<string, []>("op_9149"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9151 = const()[name = tensor<string, []>("op_9151"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_45_pad_type_0 = const()[name = tensor<string, []>("v_45_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_45_pad_0 = const()[name = tensor<string, []>("v_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(691315584)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_45_cast = conv(dilations = var_9151, groups = var_2634, pad = v_45_pad_0, pad_type = v_45_pad_type_0, strides = var_9149, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16, x = hidden_states_107_cast)[name = tensor<string, []>("v_45_cast")];
            tensor<int32, [4]> var_9155_begin_0 = const()[name = tensor<string, []>("op_9155_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9155_end_0 = const()[name = tensor<string, []>("op_9155_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_9155_end_mask_0 = const()[name = tensor<string, []>("op_9155_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9155_cast = slice_by_index(begin = var_9155_begin_0, end = var_9155_end_0, end_mask = var_9155_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9155_cast")];
            tensor<int32, [4]> var_9159_begin_0 = const()[name = tensor<string, []>("op_9159_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_9159_end_0 = const()[name = tensor<string, []>("op_9159_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_9159_end_mask_0 = const()[name = tensor<string, []>("op_9159_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9159_cast = slice_by_index(begin = var_9159_begin_0, end = var_9159_end_0, end_mask = var_9159_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9159_cast")];
            tensor<int32, [4]> var_9163_begin_0 = const()[name = tensor<string, []>("op_9163_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_9163_end_0 = const()[name = tensor<string, []>("op_9163_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_9163_end_mask_0 = const()[name = tensor<string, []>("op_9163_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9163_cast = slice_by_index(begin = var_9163_begin_0, end = var_9163_end_0, end_mask = var_9163_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9163_cast")];
            tensor<int32, [4]> var_9167_begin_0 = const()[name = tensor<string, []>("op_9167_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_9167_end_0 = const()[name = tensor<string, []>("op_9167_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_9167_end_mask_0 = const()[name = tensor<string, []>("op_9167_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9167_cast = slice_by_index(begin = var_9167_begin_0, end = var_9167_end_0, end_mask = var_9167_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9167_cast")];
            tensor<int32, [4]> var_9171_begin_0 = const()[name = tensor<string, []>("op_9171_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_9171_end_0 = const()[name = tensor<string, []>("op_9171_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_9171_end_mask_0 = const()[name = tensor<string, []>("op_9171_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9171_cast = slice_by_index(begin = var_9171_begin_0, end = var_9171_end_0, end_mask = var_9171_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9171_cast")];
            tensor<int32, [4]> var_9175_begin_0 = const()[name = tensor<string, []>("op_9175_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9175_end_0 = const()[name = tensor<string, []>("op_9175_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_9175_end_mask_0 = const()[name = tensor<string, []>("op_9175_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9175_cast = slice_by_index(begin = var_9175_begin_0, end = var_9175_end_0, end_mask = var_9175_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9175_cast")];
            tensor<int32, [4]> var_9179_begin_0 = const()[name = tensor<string, []>("op_9179_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_9179_end_0 = const()[name = tensor<string, []>("op_9179_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_9179_end_mask_0 = const()[name = tensor<string, []>("op_9179_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9179_cast = slice_by_index(begin = var_9179_begin_0, end = var_9179_end_0, end_mask = var_9179_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9179_cast")];
            tensor<int32, [4]> var_9183_begin_0 = const()[name = tensor<string, []>("op_9183_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_9183_end_0 = const()[name = tensor<string, []>("op_9183_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_9183_end_mask_0 = const()[name = tensor<string, []>("op_9183_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9183_cast = slice_by_index(begin = var_9183_begin_0, end = var_9183_end_0, end_mask = var_9183_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9183_cast")];
            tensor<int32, [4]> var_9187_begin_0 = const()[name = tensor<string, []>("op_9187_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_9187_end_0 = const()[name = tensor<string, []>("op_9187_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_9187_end_mask_0 = const()[name = tensor<string, []>("op_9187_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9187_cast = slice_by_index(begin = var_9187_begin_0, end = var_9187_end_0, end_mask = var_9187_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9187_cast")];
            tensor<int32, [4]> var_9191_begin_0 = const()[name = tensor<string, []>("op_9191_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_9191_end_0 = const()[name = tensor<string, []>("op_9191_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_9191_end_mask_0 = const()[name = tensor<string, []>("op_9191_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9191_cast = slice_by_index(begin = var_9191_begin_0, end = var_9191_end_0, end_mask = var_9191_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9191_cast")];
            tensor<int32, [4]> var_9195_begin_0 = const()[name = tensor<string, []>("op_9195_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_9195_end_0 = const()[name = tensor<string, []>("op_9195_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_9195_end_mask_0 = const()[name = tensor<string, []>("op_9195_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9195_cast = slice_by_index(begin = var_9195_begin_0, end = var_9195_end_0, end_mask = var_9195_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9195_cast")];
            tensor<int32, [4]> var_9199_begin_0 = const()[name = tensor<string, []>("op_9199_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_9199_end_0 = const()[name = tensor<string, []>("op_9199_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_9199_end_mask_0 = const()[name = tensor<string, []>("op_9199_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9199_cast = slice_by_index(begin = var_9199_begin_0, end = var_9199_end_0, end_mask = var_9199_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9199_cast")];
            tensor<int32, [4]> var_9203_begin_0 = const()[name = tensor<string, []>("op_9203_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_9203_end_0 = const()[name = tensor<string, []>("op_9203_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_9203_end_mask_0 = const()[name = tensor<string, []>("op_9203_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9203_cast = slice_by_index(begin = var_9203_begin_0, end = var_9203_end_0, end_mask = var_9203_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9203_cast")];
            tensor<int32, [4]> var_9207_begin_0 = const()[name = tensor<string, []>("op_9207_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_9207_end_0 = const()[name = tensor<string, []>("op_9207_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_9207_end_mask_0 = const()[name = tensor<string, []>("op_9207_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9207_cast = slice_by_index(begin = var_9207_begin_0, end = var_9207_end_0, end_mask = var_9207_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9207_cast")];
            tensor<int32, [4]> var_9211_begin_0 = const()[name = tensor<string, []>("op_9211_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_9211_end_0 = const()[name = tensor<string, []>("op_9211_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_9211_end_mask_0 = const()[name = tensor<string, []>("op_9211_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9211_cast = slice_by_index(begin = var_9211_begin_0, end = var_9211_end_0, end_mask = var_9211_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9211_cast")];
            tensor<int32, [4]> var_9215_begin_0 = const()[name = tensor<string, []>("op_9215_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_9215_end_0 = const()[name = tensor<string, []>("op_9215_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_9215_end_mask_0 = const()[name = tensor<string, []>("op_9215_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9215_cast = slice_by_index(begin = var_9215_begin_0, end = var_9215_end_0, end_mask = var_9215_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9215_cast")];
            tensor<int32, [4]> var_9219_begin_0 = const()[name = tensor<string, []>("op_9219_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_9219_end_0 = const()[name = tensor<string, []>("op_9219_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_9219_end_mask_0 = const()[name = tensor<string, []>("op_9219_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9219_cast = slice_by_index(begin = var_9219_begin_0, end = var_9219_end_0, end_mask = var_9219_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9219_cast")];
            tensor<int32, [4]> var_9223_begin_0 = const()[name = tensor<string, []>("op_9223_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_9223_end_0 = const()[name = tensor<string, []>("op_9223_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_9223_end_mask_0 = const()[name = tensor<string, []>("op_9223_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9223_cast = slice_by_index(begin = var_9223_begin_0, end = var_9223_end_0, end_mask = var_9223_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9223_cast")];
            tensor<int32, [4]> var_9227_begin_0 = const()[name = tensor<string, []>("op_9227_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_9227_end_0 = const()[name = tensor<string, []>("op_9227_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_9227_end_mask_0 = const()[name = tensor<string, []>("op_9227_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9227_cast = slice_by_index(begin = var_9227_begin_0, end = var_9227_end_0, end_mask = var_9227_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9227_cast")];
            tensor<int32, [4]> var_9231_begin_0 = const()[name = tensor<string, []>("op_9231_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_9231_end_0 = const()[name = tensor<string, []>("op_9231_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_9231_end_mask_0 = const()[name = tensor<string, []>("op_9231_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9231_cast = slice_by_index(begin = var_9231_begin_0, end = var_9231_end_0, end_mask = var_9231_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9231_cast")];
            tensor<int32, [4]> k_91_perm_0 = const()[name = tensor<string, []>("k_91_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_9238_begin_0 = const()[name = tensor<string, []>("op_9238_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9238_end_0 = const()[name = tensor<string, []>("op_9238_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_9238_end_mask_0 = const()[name = tensor<string, []>("op_9238_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_117 = transpose(perm = k_91_perm_0, x = k_89_cast)[name = tensor<string, []>("transpose_117")];
            tensor<fp16, [2, 1024, 1, 64]> var_9238_cast = slice_by_index(begin = var_9238_begin_0, end = var_9238_end_0, end_mask = var_9238_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9238_cast")];
            tensor<int32, [4]> var_9242_begin_0 = const()[name = tensor<string, []>("op_9242_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_9242_end_0 = const()[name = tensor<string, []>("op_9242_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_9242_end_mask_0 = const()[name = tensor<string, []>("op_9242_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9242_cast = slice_by_index(begin = var_9242_begin_0, end = var_9242_end_0, end_mask = var_9242_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9242_cast")];
            tensor<int32, [4]> var_9246_begin_0 = const()[name = tensor<string, []>("op_9246_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_9246_end_0 = const()[name = tensor<string, []>("op_9246_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_9246_end_mask_0 = const()[name = tensor<string, []>("op_9246_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9246_cast = slice_by_index(begin = var_9246_begin_0, end = var_9246_end_0, end_mask = var_9246_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9246_cast")];
            tensor<int32, [4]> var_9250_begin_0 = const()[name = tensor<string, []>("op_9250_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_9250_end_0 = const()[name = tensor<string, []>("op_9250_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_9250_end_mask_0 = const()[name = tensor<string, []>("op_9250_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9250_cast = slice_by_index(begin = var_9250_begin_0, end = var_9250_end_0, end_mask = var_9250_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9250_cast")];
            tensor<int32, [4]> var_9254_begin_0 = const()[name = tensor<string, []>("op_9254_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_9254_end_0 = const()[name = tensor<string, []>("op_9254_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_9254_end_mask_0 = const()[name = tensor<string, []>("op_9254_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9254_cast = slice_by_index(begin = var_9254_begin_0, end = var_9254_end_0, end_mask = var_9254_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9254_cast")];
            tensor<int32, [4]> var_9258_begin_0 = const()[name = tensor<string, []>("op_9258_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_9258_end_0 = const()[name = tensor<string, []>("op_9258_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_9258_end_mask_0 = const()[name = tensor<string, []>("op_9258_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9258_cast = slice_by_index(begin = var_9258_begin_0, end = var_9258_end_0, end_mask = var_9258_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9258_cast")];
            tensor<int32, [4]> var_9262_begin_0 = const()[name = tensor<string, []>("op_9262_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_9262_end_0 = const()[name = tensor<string, []>("op_9262_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_9262_end_mask_0 = const()[name = tensor<string, []>("op_9262_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9262_cast = slice_by_index(begin = var_9262_begin_0, end = var_9262_end_0, end_mask = var_9262_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9262_cast")];
            tensor<int32, [4]> var_9266_begin_0 = const()[name = tensor<string, []>("op_9266_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_9266_end_0 = const()[name = tensor<string, []>("op_9266_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_9266_end_mask_0 = const()[name = tensor<string, []>("op_9266_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9266_cast = slice_by_index(begin = var_9266_begin_0, end = var_9266_end_0, end_mask = var_9266_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9266_cast")];
            tensor<int32, [4]> var_9270_begin_0 = const()[name = tensor<string, []>("op_9270_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9270_end_0 = const()[name = tensor<string, []>("op_9270_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_9270_end_mask_0 = const()[name = tensor<string, []>("op_9270_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9270_cast = slice_by_index(begin = var_9270_begin_0, end = var_9270_end_0, end_mask = var_9270_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9270_cast")];
            tensor<int32, [4]> var_9274_begin_0 = const()[name = tensor<string, []>("op_9274_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_9274_end_0 = const()[name = tensor<string, []>("op_9274_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_9274_end_mask_0 = const()[name = tensor<string, []>("op_9274_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9274_cast = slice_by_index(begin = var_9274_begin_0, end = var_9274_end_0, end_mask = var_9274_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9274_cast")];
            tensor<int32, [4]> var_9278_begin_0 = const()[name = tensor<string, []>("op_9278_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_9278_end_0 = const()[name = tensor<string, []>("op_9278_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_9278_end_mask_0 = const()[name = tensor<string, []>("op_9278_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9278_cast = slice_by_index(begin = var_9278_begin_0, end = var_9278_end_0, end_mask = var_9278_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9278_cast")];
            tensor<int32, [4]> var_9282_begin_0 = const()[name = tensor<string, []>("op_9282_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_9282_end_0 = const()[name = tensor<string, []>("op_9282_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_9282_end_mask_0 = const()[name = tensor<string, []>("op_9282_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9282_cast = slice_by_index(begin = var_9282_begin_0, end = var_9282_end_0, end_mask = var_9282_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9282_cast")];
            tensor<int32, [4]> var_9286_begin_0 = const()[name = tensor<string, []>("op_9286_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_9286_end_0 = const()[name = tensor<string, []>("op_9286_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_9286_end_mask_0 = const()[name = tensor<string, []>("op_9286_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9286_cast = slice_by_index(begin = var_9286_begin_0, end = var_9286_end_0, end_mask = var_9286_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9286_cast")];
            tensor<int32, [4]> var_9290_begin_0 = const()[name = tensor<string, []>("op_9290_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_9290_end_0 = const()[name = tensor<string, []>("op_9290_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_9290_end_mask_0 = const()[name = tensor<string, []>("op_9290_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9290_cast = slice_by_index(begin = var_9290_begin_0, end = var_9290_end_0, end_mask = var_9290_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9290_cast")];
            tensor<int32, [4]> var_9294_begin_0 = const()[name = tensor<string, []>("op_9294_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_9294_end_0 = const()[name = tensor<string, []>("op_9294_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_9294_end_mask_0 = const()[name = tensor<string, []>("op_9294_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9294_cast = slice_by_index(begin = var_9294_begin_0, end = var_9294_end_0, end_mask = var_9294_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9294_cast")];
            tensor<int32, [4]> var_9298_begin_0 = const()[name = tensor<string, []>("op_9298_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_9298_end_0 = const()[name = tensor<string, []>("op_9298_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_9298_end_mask_0 = const()[name = tensor<string, []>("op_9298_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9298_cast = slice_by_index(begin = var_9298_begin_0, end = var_9298_end_0, end_mask = var_9298_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9298_cast")];
            tensor<int32, [4]> var_9302_begin_0 = const()[name = tensor<string, []>("op_9302_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_9302_end_0 = const()[name = tensor<string, []>("op_9302_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_9302_end_mask_0 = const()[name = tensor<string, []>("op_9302_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9302_cast = slice_by_index(begin = var_9302_begin_0, end = var_9302_end_0, end_mask = var_9302_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9302_cast")];
            tensor<int32, [4]> var_9306_begin_0 = const()[name = tensor<string, []>("op_9306_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_9306_end_0 = const()[name = tensor<string, []>("op_9306_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_9306_end_mask_0 = const()[name = tensor<string, []>("op_9306_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9306_cast = slice_by_index(begin = var_9306_begin_0, end = var_9306_end_0, end_mask = var_9306_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9306_cast")];
            tensor<int32, [4]> var_9310_begin_0 = const()[name = tensor<string, []>("op_9310_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_9310_end_0 = const()[name = tensor<string, []>("op_9310_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_9310_end_mask_0 = const()[name = tensor<string, []>("op_9310_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9310_cast = slice_by_index(begin = var_9310_begin_0, end = var_9310_end_0, end_mask = var_9310_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9310_cast")];
            tensor<int32, [4]> var_9314_begin_0 = const()[name = tensor<string, []>("op_9314_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_9314_end_0 = const()[name = tensor<string, []>("op_9314_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_9314_end_mask_0 = const()[name = tensor<string, []>("op_9314_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9314_cast = slice_by_index(begin = var_9314_begin_0, end = var_9314_end_0, end_mask = var_9314_end_mask_0, x = transpose_117)[name = tensor<string, []>("op_9314_cast")];
            tensor<int32, [4]> var_9316_begin_0 = const()[name = tensor<string, []>("op_9316_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9316_end_0 = const()[name = tensor<string, []>("op_9316_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_9316_end_mask_0 = const()[name = tensor<string, []>("op_9316_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9316_cast = slice_by_index(begin = var_9316_begin_0, end = var_9316_end_0, end_mask = var_9316_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9316_cast")];
            tensor<int32, [4]> var_9320_begin_0 = const()[name = tensor<string, []>("op_9320_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_9320_end_0 = const()[name = tensor<string, []>("op_9320_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_9320_end_mask_0 = const()[name = tensor<string, []>("op_9320_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9320_cast = slice_by_index(begin = var_9320_begin_0, end = var_9320_end_0, end_mask = var_9320_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9320_cast")];
            tensor<int32, [4]> var_9324_begin_0 = const()[name = tensor<string, []>("op_9324_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_9324_end_0 = const()[name = tensor<string, []>("op_9324_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_9324_end_mask_0 = const()[name = tensor<string, []>("op_9324_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9324_cast = slice_by_index(begin = var_9324_begin_0, end = var_9324_end_0, end_mask = var_9324_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9324_cast")];
            tensor<int32, [4]> var_9328_begin_0 = const()[name = tensor<string, []>("op_9328_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_9328_end_0 = const()[name = tensor<string, []>("op_9328_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_9328_end_mask_0 = const()[name = tensor<string, []>("op_9328_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9328_cast = slice_by_index(begin = var_9328_begin_0, end = var_9328_end_0, end_mask = var_9328_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9328_cast")];
            tensor<int32, [4]> var_9332_begin_0 = const()[name = tensor<string, []>("op_9332_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_9332_end_0 = const()[name = tensor<string, []>("op_9332_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_9332_end_mask_0 = const()[name = tensor<string, []>("op_9332_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9332_cast = slice_by_index(begin = var_9332_begin_0, end = var_9332_end_0, end_mask = var_9332_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9332_cast")];
            tensor<int32, [4]> var_9336_begin_0 = const()[name = tensor<string, []>("op_9336_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9336_end_0 = const()[name = tensor<string, []>("op_9336_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_9336_end_mask_0 = const()[name = tensor<string, []>("op_9336_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9336_cast = slice_by_index(begin = var_9336_begin_0, end = var_9336_end_0, end_mask = var_9336_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9336_cast")];
            tensor<int32, [4]> var_9340_begin_0 = const()[name = tensor<string, []>("op_9340_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_9340_end_0 = const()[name = tensor<string, []>("op_9340_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_9340_end_mask_0 = const()[name = tensor<string, []>("op_9340_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9340_cast = slice_by_index(begin = var_9340_begin_0, end = var_9340_end_0, end_mask = var_9340_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9340_cast")];
            tensor<int32, [4]> var_9344_begin_0 = const()[name = tensor<string, []>("op_9344_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_9344_end_0 = const()[name = tensor<string, []>("op_9344_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_9344_end_mask_0 = const()[name = tensor<string, []>("op_9344_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9344_cast = slice_by_index(begin = var_9344_begin_0, end = var_9344_end_0, end_mask = var_9344_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9344_cast")];
            tensor<int32, [4]> var_9348_begin_0 = const()[name = tensor<string, []>("op_9348_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_9348_end_0 = const()[name = tensor<string, []>("op_9348_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_9348_end_mask_0 = const()[name = tensor<string, []>("op_9348_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9348_cast = slice_by_index(begin = var_9348_begin_0, end = var_9348_end_0, end_mask = var_9348_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9348_cast")];
            tensor<int32, [4]> var_9352_begin_0 = const()[name = tensor<string, []>("op_9352_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_9352_end_0 = const()[name = tensor<string, []>("op_9352_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_9352_end_mask_0 = const()[name = tensor<string, []>("op_9352_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9352_cast = slice_by_index(begin = var_9352_begin_0, end = var_9352_end_0, end_mask = var_9352_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9352_cast")];
            tensor<int32, [4]> var_9356_begin_0 = const()[name = tensor<string, []>("op_9356_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_9356_end_0 = const()[name = tensor<string, []>("op_9356_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_9356_end_mask_0 = const()[name = tensor<string, []>("op_9356_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9356_cast = slice_by_index(begin = var_9356_begin_0, end = var_9356_end_0, end_mask = var_9356_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9356_cast")];
            tensor<int32, [4]> var_9360_begin_0 = const()[name = tensor<string, []>("op_9360_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_9360_end_0 = const()[name = tensor<string, []>("op_9360_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_9360_end_mask_0 = const()[name = tensor<string, []>("op_9360_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9360_cast = slice_by_index(begin = var_9360_begin_0, end = var_9360_end_0, end_mask = var_9360_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9360_cast")];
            tensor<int32, [4]> var_9364_begin_0 = const()[name = tensor<string, []>("op_9364_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_9364_end_0 = const()[name = tensor<string, []>("op_9364_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_9364_end_mask_0 = const()[name = tensor<string, []>("op_9364_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9364_cast = slice_by_index(begin = var_9364_begin_0, end = var_9364_end_0, end_mask = var_9364_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9364_cast")];
            tensor<int32, [4]> var_9368_begin_0 = const()[name = tensor<string, []>("op_9368_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_9368_end_0 = const()[name = tensor<string, []>("op_9368_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_9368_end_mask_0 = const()[name = tensor<string, []>("op_9368_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9368_cast = slice_by_index(begin = var_9368_begin_0, end = var_9368_end_0, end_mask = var_9368_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9368_cast")];
            tensor<int32, [4]> var_9372_begin_0 = const()[name = tensor<string, []>("op_9372_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_9372_end_0 = const()[name = tensor<string, []>("op_9372_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_9372_end_mask_0 = const()[name = tensor<string, []>("op_9372_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9372_cast = slice_by_index(begin = var_9372_begin_0, end = var_9372_end_0, end_mask = var_9372_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9372_cast")];
            tensor<int32, [4]> var_9376_begin_0 = const()[name = tensor<string, []>("op_9376_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_9376_end_0 = const()[name = tensor<string, []>("op_9376_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_9376_end_mask_0 = const()[name = tensor<string, []>("op_9376_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9376_cast = slice_by_index(begin = var_9376_begin_0, end = var_9376_end_0, end_mask = var_9376_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9376_cast")];
            tensor<int32, [4]> var_9380_begin_0 = const()[name = tensor<string, []>("op_9380_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_9380_end_0 = const()[name = tensor<string, []>("op_9380_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_9380_end_mask_0 = const()[name = tensor<string, []>("op_9380_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9380_cast = slice_by_index(begin = var_9380_begin_0, end = var_9380_end_0, end_mask = var_9380_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9380_cast")];
            tensor<int32, [4]> var_9384_begin_0 = const()[name = tensor<string, []>("op_9384_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_9384_end_0 = const()[name = tensor<string, []>("op_9384_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_9384_end_mask_0 = const()[name = tensor<string, []>("op_9384_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9384_cast = slice_by_index(begin = var_9384_begin_0, end = var_9384_end_0, end_mask = var_9384_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9384_cast")];
            tensor<int32, [4]> var_9388_begin_0 = const()[name = tensor<string, []>("op_9388_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_9388_end_0 = const()[name = tensor<string, []>("op_9388_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_9388_end_mask_0 = const()[name = tensor<string, []>("op_9388_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9388_cast = slice_by_index(begin = var_9388_begin_0, end = var_9388_end_0, end_mask = var_9388_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9388_cast")];
            tensor<int32, [4]> var_9392_begin_0 = const()[name = tensor<string, []>("op_9392_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_9392_end_0 = const()[name = tensor<string, []>("op_9392_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_9392_end_mask_0 = const()[name = tensor<string, []>("op_9392_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9392_cast = slice_by_index(begin = var_9392_begin_0, end = var_9392_end_0, end_mask = var_9392_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9392_cast")];
            tensor<string, []> var_9396_equation_0 = const()[name = tensor<string, []>("op_9396_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9396_cast = einsum(equation = var_9396_equation_0, values = (var_9238_cast, var_9155_cast))[name = tensor<string, []>("op_9396_cast")];
            tensor<fp16, []> var_9397_to_fp16 = const()[name = tensor<string, []>("op_9397_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_721_cast = mul(x = var_9396_cast, y = var_9397_to_fp16)[name = tensor<string, []>("aw_721_cast")];
            tensor<string, []> var_9400_equation_0 = const()[name = tensor<string, []>("op_9400_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9400_cast = einsum(equation = var_9400_equation_0, values = (var_9242_cast, var_9159_cast))[name = tensor<string, []>("op_9400_cast")];
            tensor<fp16, []> var_9401_to_fp16 = const()[name = tensor<string, []>("op_9401_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_723_cast = mul(x = var_9400_cast, y = var_9401_to_fp16)[name = tensor<string, []>("aw_723_cast")];
            tensor<string, []> var_9404_equation_0 = const()[name = tensor<string, []>("op_9404_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9404_cast = einsum(equation = var_9404_equation_0, values = (var_9246_cast, var_9163_cast))[name = tensor<string, []>("op_9404_cast")];
            tensor<fp16, []> var_9405_to_fp16 = const()[name = tensor<string, []>("op_9405_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_725_cast = mul(x = var_9404_cast, y = var_9405_to_fp16)[name = tensor<string, []>("aw_725_cast")];
            tensor<string, []> var_9408_equation_0 = const()[name = tensor<string, []>("op_9408_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9408_cast = einsum(equation = var_9408_equation_0, values = (var_9250_cast, var_9167_cast))[name = tensor<string, []>("op_9408_cast")];
            tensor<fp16, []> var_9409_to_fp16 = const()[name = tensor<string, []>("op_9409_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_727_cast = mul(x = var_9408_cast, y = var_9409_to_fp16)[name = tensor<string, []>("aw_727_cast")];
            tensor<string, []> var_9412_equation_0 = const()[name = tensor<string, []>("op_9412_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9412_cast = einsum(equation = var_9412_equation_0, values = (var_9254_cast, var_9171_cast))[name = tensor<string, []>("op_9412_cast")];
            tensor<fp16, []> var_9413_to_fp16 = const()[name = tensor<string, []>("op_9413_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_729_cast = mul(x = var_9412_cast, y = var_9413_to_fp16)[name = tensor<string, []>("aw_729_cast")];
            tensor<string, []> var_9416_equation_0 = const()[name = tensor<string, []>("op_9416_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9416_cast = einsum(equation = var_9416_equation_0, values = (var_9258_cast, var_9175_cast))[name = tensor<string, []>("op_9416_cast")];
            tensor<fp16, []> var_9417_to_fp16 = const()[name = tensor<string, []>("op_9417_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_731_cast = mul(x = var_9416_cast, y = var_9417_to_fp16)[name = tensor<string, []>("aw_731_cast")];
            tensor<string, []> var_9420_equation_0 = const()[name = tensor<string, []>("op_9420_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9420_cast = einsum(equation = var_9420_equation_0, values = (var_9262_cast, var_9179_cast))[name = tensor<string, []>("op_9420_cast")];
            tensor<fp16, []> var_9421_to_fp16 = const()[name = tensor<string, []>("op_9421_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_733_cast = mul(x = var_9420_cast, y = var_9421_to_fp16)[name = tensor<string, []>("aw_733_cast")];
            tensor<string, []> var_9424_equation_0 = const()[name = tensor<string, []>("op_9424_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9424_cast = einsum(equation = var_9424_equation_0, values = (var_9266_cast, var_9183_cast))[name = tensor<string, []>("op_9424_cast")];
            tensor<fp16, []> var_9425_to_fp16 = const()[name = tensor<string, []>("op_9425_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_735_cast = mul(x = var_9424_cast, y = var_9425_to_fp16)[name = tensor<string, []>("aw_735_cast")];
            tensor<string, []> var_9428_equation_0 = const()[name = tensor<string, []>("op_9428_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9428_cast = einsum(equation = var_9428_equation_0, values = (var_9270_cast, var_9187_cast))[name = tensor<string, []>("op_9428_cast")];
            tensor<fp16, []> var_9429_to_fp16 = const()[name = tensor<string, []>("op_9429_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_737_cast = mul(x = var_9428_cast, y = var_9429_to_fp16)[name = tensor<string, []>("aw_737_cast")];
            tensor<string, []> var_9432_equation_0 = const()[name = tensor<string, []>("op_9432_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9432_cast = einsum(equation = var_9432_equation_0, values = (var_9274_cast, var_9191_cast))[name = tensor<string, []>("op_9432_cast")];
            tensor<fp16, []> var_9433_to_fp16 = const()[name = tensor<string, []>("op_9433_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_739_cast = mul(x = var_9432_cast, y = var_9433_to_fp16)[name = tensor<string, []>("aw_739_cast")];
            tensor<string, []> var_9436_equation_0 = const()[name = tensor<string, []>("op_9436_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9436_cast = einsum(equation = var_9436_equation_0, values = (var_9278_cast, var_9195_cast))[name = tensor<string, []>("op_9436_cast")];
            tensor<fp16, []> var_9437_to_fp16 = const()[name = tensor<string, []>("op_9437_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_741_cast = mul(x = var_9436_cast, y = var_9437_to_fp16)[name = tensor<string, []>("aw_741_cast")];
            tensor<string, []> var_9440_equation_0 = const()[name = tensor<string, []>("op_9440_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9440_cast = einsum(equation = var_9440_equation_0, values = (var_9282_cast, var_9199_cast))[name = tensor<string, []>("op_9440_cast")];
            tensor<fp16, []> var_9441_to_fp16 = const()[name = tensor<string, []>("op_9441_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_743_cast = mul(x = var_9440_cast, y = var_9441_to_fp16)[name = tensor<string, []>("aw_743_cast")];
            tensor<string, []> var_9444_equation_0 = const()[name = tensor<string, []>("op_9444_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9444_cast = einsum(equation = var_9444_equation_0, values = (var_9286_cast, var_9203_cast))[name = tensor<string, []>("op_9444_cast")];
            tensor<fp16, []> var_9445_to_fp16 = const()[name = tensor<string, []>("op_9445_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_745_cast = mul(x = var_9444_cast, y = var_9445_to_fp16)[name = tensor<string, []>("aw_745_cast")];
            tensor<string, []> var_9448_equation_0 = const()[name = tensor<string, []>("op_9448_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9448_cast = einsum(equation = var_9448_equation_0, values = (var_9290_cast, var_9207_cast))[name = tensor<string, []>("op_9448_cast")];
            tensor<fp16, []> var_9449_to_fp16 = const()[name = tensor<string, []>("op_9449_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_747_cast = mul(x = var_9448_cast, y = var_9449_to_fp16)[name = tensor<string, []>("aw_747_cast")];
            tensor<string, []> var_9452_equation_0 = const()[name = tensor<string, []>("op_9452_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9452_cast = einsum(equation = var_9452_equation_0, values = (var_9294_cast, var_9211_cast))[name = tensor<string, []>("op_9452_cast")];
            tensor<fp16, []> var_9453_to_fp16 = const()[name = tensor<string, []>("op_9453_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_749_cast = mul(x = var_9452_cast, y = var_9453_to_fp16)[name = tensor<string, []>("aw_749_cast")];
            tensor<string, []> var_9456_equation_0 = const()[name = tensor<string, []>("op_9456_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9456_cast = einsum(equation = var_9456_equation_0, values = (var_9298_cast, var_9215_cast))[name = tensor<string, []>("op_9456_cast")];
            tensor<fp16, []> var_9457_to_fp16 = const()[name = tensor<string, []>("op_9457_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_751_cast = mul(x = var_9456_cast, y = var_9457_to_fp16)[name = tensor<string, []>("aw_751_cast")];
            tensor<string, []> var_9460_equation_0 = const()[name = tensor<string, []>("op_9460_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9460_cast = einsum(equation = var_9460_equation_0, values = (var_9302_cast, var_9219_cast))[name = tensor<string, []>("op_9460_cast")];
            tensor<fp16, []> var_9461_to_fp16 = const()[name = tensor<string, []>("op_9461_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_753_cast = mul(x = var_9460_cast, y = var_9461_to_fp16)[name = tensor<string, []>("aw_753_cast")];
            tensor<string, []> var_9464_equation_0 = const()[name = tensor<string, []>("op_9464_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9464_cast = einsum(equation = var_9464_equation_0, values = (var_9306_cast, var_9223_cast))[name = tensor<string, []>("op_9464_cast")];
            tensor<fp16, []> var_9465_to_fp16 = const()[name = tensor<string, []>("op_9465_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_755_cast = mul(x = var_9464_cast, y = var_9465_to_fp16)[name = tensor<string, []>("aw_755_cast")];
            tensor<string, []> var_9468_equation_0 = const()[name = tensor<string, []>("op_9468_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9468_cast = einsum(equation = var_9468_equation_0, values = (var_9310_cast, var_9227_cast))[name = tensor<string, []>("op_9468_cast")];
            tensor<fp16, []> var_9469_to_fp16 = const()[name = tensor<string, []>("op_9469_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_757_cast = mul(x = var_9468_cast, y = var_9469_to_fp16)[name = tensor<string, []>("aw_757_cast")];
            tensor<string, []> var_9472_equation_0 = const()[name = tensor<string, []>("op_9472_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9472_cast = einsum(equation = var_9472_equation_0, values = (var_9314_cast, var_9231_cast))[name = tensor<string, []>("op_9472_cast")];
            tensor<fp16, []> var_9473_to_fp16 = const()[name = tensor<string, []>("op_9473_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_759_cast = mul(x = var_9472_cast, y = var_9473_to_fp16)[name = tensor<string, []>("aw_759_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9475_cast = softmax(axis = var_2634, x = aw_721_cast)[name = tensor<string, []>("op_9475_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9476_cast = softmax(axis = var_2634, x = aw_723_cast)[name = tensor<string, []>("op_9476_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9477_cast = softmax(axis = var_2634, x = aw_725_cast)[name = tensor<string, []>("op_9477_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9478_cast = softmax(axis = var_2634, x = aw_727_cast)[name = tensor<string, []>("op_9478_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9479_cast = softmax(axis = var_2634, x = aw_729_cast)[name = tensor<string, []>("op_9479_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9480_cast = softmax(axis = var_2634, x = aw_731_cast)[name = tensor<string, []>("op_9480_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9481_cast = softmax(axis = var_2634, x = aw_733_cast)[name = tensor<string, []>("op_9481_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9482_cast = softmax(axis = var_2634, x = aw_735_cast)[name = tensor<string, []>("op_9482_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9483_cast = softmax(axis = var_2634, x = aw_737_cast)[name = tensor<string, []>("op_9483_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9484_cast = softmax(axis = var_2634, x = aw_739_cast)[name = tensor<string, []>("op_9484_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9485_cast = softmax(axis = var_2634, x = aw_741_cast)[name = tensor<string, []>("op_9485_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9486_cast = softmax(axis = var_2634, x = aw_743_cast)[name = tensor<string, []>("op_9486_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9487_cast = softmax(axis = var_2634, x = aw_745_cast)[name = tensor<string, []>("op_9487_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9488_cast = softmax(axis = var_2634, x = aw_747_cast)[name = tensor<string, []>("op_9488_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9489_cast = softmax(axis = var_2634, x = aw_749_cast)[name = tensor<string, []>("op_9489_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9490_cast = softmax(axis = var_2634, x = aw_751_cast)[name = tensor<string, []>("op_9490_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9491_cast = softmax(axis = var_2634, x = aw_753_cast)[name = tensor<string, []>("op_9491_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9492_cast = softmax(axis = var_2634, x = aw_755_cast)[name = tensor<string, []>("op_9492_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9493_cast = softmax(axis = var_2634, x = aw_757_cast)[name = tensor<string, []>("op_9493_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9494_cast = softmax(axis = var_2634, x = aw_759_cast)[name = tensor<string, []>("op_9494_cast")];
            tensor<string, []> var_9496_equation_0 = const()[name = tensor<string, []>("op_9496_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9496_cast = einsum(equation = var_9496_equation_0, values = (var_9316_cast, var_9475_cast))[name = tensor<string, []>("op_9496_cast")];
            tensor<string, []> var_9498_equation_0 = const()[name = tensor<string, []>("op_9498_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9498_cast = einsum(equation = var_9498_equation_0, values = (var_9320_cast, var_9476_cast))[name = tensor<string, []>("op_9498_cast")];
            tensor<string, []> var_9500_equation_0 = const()[name = tensor<string, []>("op_9500_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9500_cast = einsum(equation = var_9500_equation_0, values = (var_9324_cast, var_9477_cast))[name = tensor<string, []>("op_9500_cast")];
            tensor<string, []> var_9502_equation_0 = const()[name = tensor<string, []>("op_9502_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9502_cast = einsum(equation = var_9502_equation_0, values = (var_9328_cast, var_9478_cast))[name = tensor<string, []>("op_9502_cast")];
            tensor<string, []> var_9504_equation_0 = const()[name = tensor<string, []>("op_9504_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9504_cast = einsum(equation = var_9504_equation_0, values = (var_9332_cast, var_9479_cast))[name = tensor<string, []>("op_9504_cast")];
            tensor<string, []> var_9506_equation_0 = const()[name = tensor<string, []>("op_9506_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9506_cast = einsum(equation = var_9506_equation_0, values = (var_9336_cast, var_9480_cast))[name = tensor<string, []>("op_9506_cast")];
            tensor<string, []> var_9508_equation_0 = const()[name = tensor<string, []>("op_9508_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9508_cast = einsum(equation = var_9508_equation_0, values = (var_9340_cast, var_9481_cast))[name = tensor<string, []>("op_9508_cast")];
            tensor<string, []> var_9510_equation_0 = const()[name = tensor<string, []>("op_9510_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9510_cast = einsum(equation = var_9510_equation_0, values = (var_9344_cast, var_9482_cast))[name = tensor<string, []>("op_9510_cast")];
            tensor<string, []> var_9512_equation_0 = const()[name = tensor<string, []>("op_9512_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9512_cast = einsum(equation = var_9512_equation_0, values = (var_9348_cast, var_9483_cast))[name = tensor<string, []>("op_9512_cast")];
            tensor<string, []> var_9514_equation_0 = const()[name = tensor<string, []>("op_9514_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9514_cast = einsum(equation = var_9514_equation_0, values = (var_9352_cast, var_9484_cast))[name = tensor<string, []>("op_9514_cast")];
            tensor<string, []> var_9516_equation_0 = const()[name = tensor<string, []>("op_9516_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9516_cast = einsum(equation = var_9516_equation_0, values = (var_9356_cast, var_9485_cast))[name = tensor<string, []>("op_9516_cast")];
            tensor<string, []> var_9518_equation_0 = const()[name = tensor<string, []>("op_9518_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9518_cast = einsum(equation = var_9518_equation_0, values = (var_9360_cast, var_9486_cast))[name = tensor<string, []>("op_9518_cast")];
            tensor<string, []> var_9520_equation_0 = const()[name = tensor<string, []>("op_9520_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9520_cast = einsum(equation = var_9520_equation_0, values = (var_9364_cast, var_9487_cast))[name = tensor<string, []>("op_9520_cast")];
            tensor<string, []> var_9522_equation_0 = const()[name = tensor<string, []>("op_9522_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9522_cast = einsum(equation = var_9522_equation_0, values = (var_9368_cast, var_9488_cast))[name = tensor<string, []>("op_9522_cast")];
            tensor<string, []> var_9524_equation_0 = const()[name = tensor<string, []>("op_9524_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9524_cast = einsum(equation = var_9524_equation_0, values = (var_9372_cast, var_9489_cast))[name = tensor<string, []>("op_9524_cast")];
            tensor<string, []> var_9526_equation_0 = const()[name = tensor<string, []>("op_9526_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9526_cast = einsum(equation = var_9526_equation_0, values = (var_9376_cast, var_9490_cast))[name = tensor<string, []>("op_9526_cast")];
            tensor<string, []> var_9528_equation_0 = const()[name = tensor<string, []>("op_9528_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9528_cast = einsum(equation = var_9528_equation_0, values = (var_9380_cast, var_9491_cast))[name = tensor<string, []>("op_9528_cast")];
            tensor<string, []> var_9530_equation_0 = const()[name = tensor<string, []>("op_9530_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9530_cast = einsum(equation = var_9530_equation_0, values = (var_9384_cast, var_9492_cast))[name = tensor<string, []>("op_9530_cast")];
            tensor<string, []> var_9532_equation_0 = const()[name = tensor<string, []>("op_9532_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9532_cast = einsum(equation = var_9532_equation_0, values = (var_9388_cast, var_9493_cast))[name = tensor<string, []>("op_9532_cast")];
            tensor<string, []> var_9534_equation_0 = const()[name = tensor<string, []>("op_9534_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9534_cast = einsum(equation = var_9534_equation_0, values = (var_9392_cast, var_9494_cast))[name = tensor<string, []>("op_9534_cast")];
            tensor<bool, []> input_191_interleave_0 = const()[name = tensor<string, []>("input_191_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_191_cast = concat(axis = var_2634, interleave = input_191_interleave_0, values = (var_9496_cast, var_9498_cast, var_9500_cast, var_9502_cast, var_9504_cast, var_9506_cast, var_9508_cast, var_9510_cast, var_9512_cast, var_9514_cast, var_9516_cast, var_9518_cast, var_9520_cast, var_9522_cast, var_9524_cast, var_9526_cast, var_9528_cast, var_9530_cast, var_9532_cast, var_9534_cast))[name = tensor<string, []>("input_191_cast")];
            tensor<int32, [2]> var_9540 = const()[name = tensor<string, []>("op_9540"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9542 = const()[name = tensor<string, []>("op_9542"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9544_pad_type_0 = const()[name = tensor<string, []>("op_9544_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9544_pad_0 = const()[name = tensor<string, []>("op_9544_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(694592448)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(697869312)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_9544_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16, dilations = var_9542, groups = var_2634, pad = var_9544_pad_0, pad_type = var_9544_pad_type_0, strides = var_9540, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16, x = input_191_cast)[name = tensor<string, []>("op_9544_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_69_cast = add(x = var_9544_cast, y = inputs_67_cast)[name = tensor<string, []>("inputs_69_cast")];
            tensor<int32, [1]> var_9548 = const()[name = tensor<string, []>("op_9548"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_69_cast = reduce_mean(axes = var_9548, keep_dims = var_2629, x = inputs_69_cast)[name = tensor<string, []>("channels_mean_69_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_69_cast = sub(x = inputs_69_cast, y = channels_mean_69_cast)[name = tensor<string, []>("zero_mean_69_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_69_cast = mul(x = zero_mean_69_cast, y = zero_mean_69_cast)[name = tensor<string, []>("zero_mean_sq_69_cast")];
            tensor<int32, [1]> var_9552 = const()[name = tensor<string, []>("op_9552"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_9553_cast = reduce_mean(axes = var_9552, keep_dims = var_2629, x = zero_mean_sq_69_cast)[name = tensor<string, []>("op_9553_cast")];
            tensor<fp16, []> var_9554_to_fp16 = const()[name = tensor<string, []>("op_9554_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_9555_cast = add(x = var_9553_cast, y = var_9554_to_fp16)[name = tensor<string, []>("op_9555_cast")];
            tensor<fp16, []> denom_69_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_69_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_69_cast = rsqrt(epsilon = denom_69_epsilon_0_to_fp16, x = var_9555_cast)[name = tensor<string, []>("denom_69_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_69_cast = mul(x = zero_mean_69_cast, y = denom_69_cast)[name = tensor<string, []>("out_69_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_9559_to_fp16 = const()[name = tensor<string, []>("op_9559_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(697871936)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_9560_cast = add(x = out_69_cast, y = var_9559_to_fp16)[name = tensor<string, []>("op_9560_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_9562_to_fp16 = const()[name = tensor<string, []>("op_9562_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(697874560)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_109_cast = mul(x = var_9560_cast, y = var_9562_to_fp16)[name = tensor<string, []>("hidden_states_109_cast")];
            tensor<int32, [2]> var_9569 = const()[name = tensor<string, []>("op_9569"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9571 = const()[name = tensor<string, []>("op_9571"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_47_pad_type_0 = const()[name = tensor<string, []>("q_47_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_47_pad_0 = const()[name = tensor<string, []>("q_47_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(697877184)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_47_cast = conv(dilations = var_9571, groups = var_2634, pad = q_47_pad_0, pad_type = q_47_pad_type_0, strides = var_9569, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16, x = hidden_states_109_cast)[name = tensor<string, []>("q_47_cast")];
            tensor<int32, [2]> var_9575 = const()[name = tensor<string, []>("op_9575"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9577 = const()[name = tensor<string, []>("op_9577"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_93_pad_type_0 = const()[name = tensor<string, []>("k_93_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_93_pad_0 = const()[name = tensor<string, []>("k_93_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(701154048)))];
            tensor<fp16, [2, 1280, 1, 77]> k_93_cast = conv(dilations = var_9577, groups = var_2634, pad = k_93_pad_0, pad_type = k_93_pad_type_0, strides = var_9575, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_93_cast")];
            tensor<int32, [2]> var_9581 = const()[name = tensor<string, []>("op_9581"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9583 = const()[name = tensor<string, []>("op_9583"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_47_pad_type_0 = const()[name = tensor<string, []>("v_47_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_47_pad_0 = const()[name = tensor<string, []>("v_47_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(706396992)))];
            tensor<fp16, [2, 1280, 1, 77]> v_47_cast = conv(dilations = var_9583, groups = var_2634, pad = v_47_pad_0, pad_type = v_47_pad_type_0, strides = var_9581, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_47_cast")];
            tensor<int32, [4]> var_9587_begin_0 = const()[name = tensor<string, []>("op_9587_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9587_end_0 = const()[name = tensor<string, []>("op_9587_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_9587_end_mask_0 = const()[name = tensor<string, []>("op_9587_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9587_cast = slice_by_index(begin = var_9587_begin_0, end = var_9587_end_0, end_mask = var_9587_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9587_cast")];
            tensor<int32, [4]> var_9591_begin_0 = const()[name = tensor<string, []>("op_9591_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_9591_end_0 = const()[name = tensor<string, []>("op_9591_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_9591_end_mask_0 = const()[name = tensor<string, []>("op_9591_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9591_cast = slice_by_index(begin = var_9591_begin_0, end = var_9591_end_0, end_mask = var_9591_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9591_cast")];
            tensor<int32, [4]> var_9595_begin_0 = const()[name = tensor<string, []>("op_9595_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_9595_end_0 = const()[name = tensor<string, []>("op_9595_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_9595_end_mask_0 = const()[name = tensor<string, []>("op_9595_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9595_cast = slice_by_index(begin = var_9595_begin_0, end = var_9595_end_0, end_mask = var_9595_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9595_cast")];
            tensor<int32, [4]> var_9599_begin_0 = const()[name = tensor<string, []>("op_9599_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_9599_end_0 = const()[name = tensor<string, []>("op_9599_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_9599_end_mask_0 = const()[name = tensor<string, []>("op_9599_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9599_cast = slice_by_index(begin = var_9599_begin_0, end = var_9599_end_0, end_mask = var_9599_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9599_cast")];
            tensor<int32, [4]> var_9603_begin_0 = const()[name = tensor<string, []>("op_9603_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_9603_end_0 = const()[name = tensor<string, []>("op_9603_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_9603_end_mask_0 = const()[name = tensor<string, []>("op_9603_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9603_cast = slice_by_index(begin = var_9603_begin_0, end = var_9603_end_0, end_mask = var_9603_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9603_cast")];
            tensor<int32, [4]> var_9607_begin_0 = const()[name = tensor<string, []>("op_9607_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9607_end_0 = const()[name = tensor<string, []>("op_9607_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_9607_end_mask_0 = const()[name = tensor<string, []>("op_9607_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9607_cast = slice_by_index(begin = var_9607_begin_0, end = var_9607_end_0, end_mask = var_9607_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9607_cast")];
            tensor<int32, [4]> var_9611_begin_0 = const()[name = tensor<string, []>("op_9611_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_9611_end_0 = const()[name = tensor<string, []>("op_9611_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_9611_end_mask_0 = const()[name = tensor<string, []>("op_9611_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9611_cast = slice_by_index(begin = var_9611_begin_0, end = var_9611_end_0, end_mask = var_9611_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9611_cast")];
            tensor<int32, [4]> var_9615_begin_0 = const()[name = tensor<string, []>("op_9615_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_9615_end_0 = const()[name = tensor<string, []>("op_9615_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_9615_end_mask_0 = const()[name = tensor<string, []>("op_9615_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9615_cast = slice_by_index(begin = var_9615_begin_0, end = var_9615_end_0, end_mask = var_9615_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9615_cast")];
            tensor<int32, [4]> var_9619_begin_0 = const()[name = tensor<string, []>("op_9619_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_9619_end_0 = const()[name = tensor<string, []>("op_9619_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_9619_end_mask_0 = const()[name = tensor<string, []>("op_9619_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9619_cast = slice_by_index(begin = var_9619_begin_0, end = var_9619_end_0, end_mask = var_9619_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9619_cast")];
            tensor<int32, [4]> var_9623_begin_0 = const()[name = tensor<string, []>("op_9623_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_9623_end_0 = const()[name = tensor<string, []>("op_9623_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_9623_end_mask_0 = const()[name = tensor<string, []>("op_9623_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9623_cast = slice_by_index(begin = var_9623_begin_0, end = var_9623_end_0, end_mask = var_9623_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9623_cast")];
            tensor<int32, [4]> var_9627_begin_0 = const()[name = tensor<string, []>("op_9627_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_9627_end_0 = const()[name = tensor<string, []>("op_9627_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_9627_end_mask_0 = const()[name = tensor<string, []>("op_9627_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9627_cast = slice_by_index(begin = var_9627_begin_0, end = var_9627_end_0, end_mask = var_9627_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9627_cast")];
            tensor<int32, [4]> var_9631_begin_0 = const()[name = tensor<string, []>("op_9631_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_9631_end_0 = const()[name = tensor<string, []>("op_9631_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_9631_end_mask_0 = const()[name = tensor<string, []>("op_9631_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9631_cast = slice_by_index(begin = var_9631_begin_0, end = var_9631_end_0, end_mask = var_9631_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9631_cast")];
            tensor<int32, [4]> var_9635_begin_0 = const()[name = tensor<string, []>("op_9635_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_9635_end_0 = const()[name = tensor<string, []>("op_9635_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_9635_end_mask_0 = const()[name = tensor<string, []>("op_9635_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9635_cast = slice_by_index(begin = var_9635_begin_0, end = var_9635_end_0, end_mask = var_9635_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9635_cast")];
            tensor<int32, [4]> var_9639_begin_0 = const()[name = tensor<string, []>("op_9639_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_9639_end_0 = const()[name = tensor<string, []>("op_9639_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_9639_end_mask_0 = const()[name = tensor<string, []>("op_9639_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9639_cast = slice_by_index(begin = var_9639_begin_0, end = var_9639_end_0, end_mask = var_9639_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9639_cast")];
            tensor<int32, [4]> var_9643_begin_0 = const()[name = tensor<string, []>("op_9643_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_9643_end_0 = const()[name = tensor<string, []>("op_9643_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_9643_end_mask_0 = const()[name = tensor<string, []>("op_9643_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9643_cast = slice_by_index(begin = var_9643_begin_0, end = var_9643_end_0, end_mask = var_9643_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9643_cast")];
            tensor<int32, [4]> var_9647_begin_0 = const()[name = tensor<string, []>("op_9647_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_9647_end_0 = const()[name = tensor<string, []>("op_9647_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_9647_end_mask_0 = const()[name = tensor<string, []>("op_9647_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9647_cast = slice_by_index(begin = var_9647_begin_0, end = var_9647_end_0, end_mask = var_9647_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9647_cast")];
            tensor<int32, [4]> var_9651_begin_0 = const()[name = tensor<string, []>("op_9651_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_9651_end_0 = const()[name = tensor<string, []>("op_9651_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_9651_end_mask_0 = const()[name = tensor<string, []>("op_9651_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9651_cast = slice_by_index(begin = var_9651_begin_0, end = var_9651_end_0, end_mask = var_9651_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9651_cast")];
            tensor<int32, [4]> var_9655_begin_0 = const()[name = tensor<string, []>("op_9655_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_9655_end_0 = const()[name = tensor<string, []>("op_9655_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_9655_end_mask_0 = const()[name = tensor<string, []>("op_9655_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9655_cast = slice_by_index(begin = var_9655_begin_0, end = var_9655_end_0, end_mask = var_9655_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9655_cast")];
            tensor<int32, [4]> var_9659_begin_0 = const()[name = tensor<string, []>("op_9659_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_9659_end_0 = const()[name = tensor<string, []>("op_9659_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_9659_end_mask_0 = const()[name = tensor<string, []>("op_9659_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9659_cast = slice_by_index(begin = var_9659_begin_0, end = var_9659_end_0, end_mask = var_9659_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9659_cast")];
            tensor<int32, [4]> var_9663_begin_0 = const()[name = tensor<string, []>("op_9663_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_9663_end_0 = const()[name = tensor<string, []>("op_9663_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_9663_end_mask_0 = const()[name = tensor<string, []>("op_9663_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9663_cast = slice_by_index(begin = var_9663_begin_0, end = var_9663_end_0, end_mask = var_9663_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9663_cast")];
            tensor<int32, [4]> k_95_perm_0 = const()[name = tensor<string, []>("k_95_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_9670_begin_0 = const()[name = tensor<string, []>("op_9670_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9670_end_0 = const()[name = tensor<string, []>("op_9670_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_9670_end_mask_0 = const()[name = tensor<string, []>("op_9670_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_116 = transpose(perm = k_95_perm_0, x = k_93_cast)[name = tensor<string, []>("transpose_116")];
            tensor<fp16, [2, 77, 1, 64]> var_9670_cast = slice_by_index(begin = var_9670_begin_0, end = var_9670_end_0, end_mask = var_9670_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9670_cast")];
            tensor<int32, [4]> var_9674_begin_0 = const()[name = tensor<string, []>("op_9674_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_9674_end_0 = const()[name = tensor<string, []>("op_9674_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_9674_end_mask_0 = const()[name = tensor<string, []>("op_9674_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9674_cast = slice_by_index(begin = var_9674_begin_0, end = var_9674_end_0, end_mask = var_9674_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9674_cast")];
            tensor<int32, [4]> var_9678_begin_0 = const()[name = tensor<string, []>("op_9678_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_9678_end_0 = const()[name = tensor<string, []>("op_9678_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_9678_end_mask_0 = const()[name = tensor<string, []>("op_9678_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9678_cast = slice_by_index(begin = var_9678_begin_0, end = var_9678_end_0, end_mask = var_9678_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9678_cast")];
            tensor<int32, [4]> var_9682_begin_0 = const()[name = tensor<string, []>("op_9682_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_9682_end_0 = const()[name = tensor<string, []>("op_9682_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_9682_end_mask_0 = const()[name = tensor<string, []>("op_9682_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9682_cast = slice_by_index(begin = var_9682_begin_0, end = var_9682_end_0, end_mask = var_9682_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9682_cast")];
            tensor<int32, [4]> var_9686_begin_0 = const()[name = tensor<string, []>("op_9686_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_9686_end_0 = const()[name = tensor<string, []>("op_9686_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_9686_end_mask_0 = const()[name = tensor<string, []>("op_9686_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9686_cast = slice_by_index(begin = var_9686_begin_0, end = var_9686_end_0, end_mask = var_9686_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9686_cast")];
            tensor<int32, [4]> var_9690_begin_0 = const()[name = tensor<string, []>("op_9690_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_9690_end_0 = const()[name = tensor<string, []>("op_9690_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_9690_end_mask_0 = const()[name = tensor<string, []>("op_9690_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9690_cast = slice_by_index(begin = var_9690_begin_0, end = var_9690_end_0, end_mask = var_9690_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9690_cast")];
            tensor<int32, [4]> var_9694_begin_0 = const()[name = tensor<string, []>("op_9694_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_9694_end_0 = const()[name = tensor<string, []>("op_9694_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_9694_end_mask_0 = const()[name = tensor<string, []>("op_9694_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9694_cast = slice_by_index(begin = var_9694_begin_0, end = var_9694_end_0, end_mask = var_9694_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9694_cast")];
            tensor<int32, [4]> var_9698_begin_0 = const()[name = tensor<string, []>("op_9698_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_9698_end_0 = const()[name = tensor<string, []>("op_9698_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_9698_end_mask_0 = const()[name = tensor<string, []>("op_9698_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9698_cast = slice_by_index(begin = var_9698_begin_0, end = var_9698_end_0, end_mask = var_9698_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9698_cast")];
            tensor<int32, [4]> var_9702_begin_0 = const()[name = tensor<string, []>("op_9702_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9702_end_0 = const()[name = tensor<string, []>("op_9702_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_9702_end_mask_0 = const()[name = tensor<string, []>("op_9702_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9702_cast = slice_by_index(begin = var_9702_begin_0, end = var_9702_end_0, end_mask = var_9702_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9702_cast")];
            tensor<int32, [4]> var_9706_begin_0 = const()[name = tensor<string, []>("op_9706_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_9706_end_0 = const()[name = tensor<string, []>("op_9706_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_9706_end_mask_0 = const()[name = tensor<string, []>("op_9706_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9706_cast = slice_by_index(begin = var_9706_begin_0, end = var_9706_end_0, end_mask = var_9706_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9706_cast")];
            tensor<int32, [4]> var_9710_begin_0 = const()[name = tensor<string, []>("op_9710_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_9710_end_0 = const()[name = tensor<string, []>("op_9710_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_9710_end_mask_0 = const()[name = tensor<string, []>("op_9710_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9710_cast = slice_by_index(begin = var_9710_begin_0, end = var_9710_end_0, end_mask = var_9710_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9710_cast")];
            tensor<int32, [4]> var_9714_begin_0 = const()[name = tensor<string, []>("op_9714_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_9714_end_0 = const()[name = tensor<string, []>("op_9714_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_9714_end_mask_0 = const()[name = tensor<string, []>("op_9714_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9714_cast = slice_by_index(begin = var_9714_begin_0, end = var_9714_end_0, end_mask = var_9714_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9714_cast")];
            tensor<int32, [4]> var_9718_begin_0 = const()[name = tensor<string, []>("op_9718_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_9718_end_0 = const()[name = tensor<string, []>("op_9718_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_9718_end_mask_0 = const()[name = tensor<string, []>("op_9718_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9718_cast = slice_by_index(begin = var_9718_begin_0, end = var_9718_end_0, end_mask = var_9718_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9718_cast")];
            tensor<int32, [4]> var_9722_begin_0 = const()[name = tensor<string, []>("op_9722_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_9722_end_0 = const()[name = tensor<string, []>("op_9722_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_9722_end_mask_0 = const()[name = tensor<string, []>("op_9722_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9722_cast = slice_by_index(begin = var_9722_begin_0, end = var_9722_end_0, end_mask = var_9722_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9722_cast")];
            tensor<int32, [4]> var_9726_begin_0 = const()[name = tensor<string, []>("op_9726_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_9726_end_0 = const()[name = tensor<string, []>("op_9726_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_9726_end_mask_0 = const()[name = tensor<string, []>("op_9726_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9726_cast = slice_by_index(begin = var_9726_begin_0, end = var_9726_end_0, end_mask = var_9726_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9726_cast")];
            tensor<int32, [4]> var_9730_begin_0 = const()[name = tensor<string, []>("op_9730_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_9730_end_0 = const()[name = tensor<string, []>("op_9730_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_9730_end_mask_0 = const()[name = tensor<string, []>("op_9730_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9730_cast = slice_by_index(begin = var_9730_begin_0, end = var_9730_end_0, end_mask = var_9730_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9730_cast")];
            tensor<int32, [4]> var_9734_begin_0 = const()[name = tensor<string, []>("op_9734_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_9734_end_0 = const()[name = tensor<string, []>("op_9734_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_9734_end_mask_0 = const()[name = tensor<string, []>("op_9734_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9734_cast = slice_by_index(begin = var_9734_begin_0, end = var_9734_end_0, end_mask = var_9734_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9734_cast")];
            tensor<int32, [4]> var_9738_begin_0 = const()[name = tensor<string, []>("op_9738_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_9738_end_0 = const()[name = tensor<string, []>("op_9738_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_9738_end_mask_0 = const()[name = tensor<string, []>("op_9738_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9738_cast = slice_by_index(begin = var_9738_begin_0, end = var_9738_end_0, end_mask = var_9738_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9738_cast")];
            tensor<int32, [4]> var_9742_begin_0 = const()[name = tensor<string, []>("op_9742_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_9742_end_0 = const()[name = tensor<string, []>("op_9742_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_9742_end_mask_0 = const()[name = tensor<string, []>("op_9742_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9742_cast = slice_by_index(begin = var_9742_begin_0, end = var_9742_end_0, end_mask = var_9742_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9742_cast")];
            tensor<int32, [4]> var_9746_begin_0 = const()[name = tensor<string, []>("op_9746_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_9746_end_0 = const()[name = tensor<string, []>("op_9746_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_9746_end_mask_0 = const()[name = tensor<string, []>("op_9746_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9746_cast = slice_by_index(begin = var_9746_begin_0, end = var_9746_end_0, end_mask = var_9746_end_mask_0, x = transpose_116)[name = tensor<string, []>("op_9746_cast")];
            tensor<int32, [4]> var_9748_begin_0 = const()[name = tensor<string, []>("op_9748_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9748_end_0 = const()[name = tensor<string, []>("op_9748_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_9748_end_mask_0 = const()[name = tensor<string, []>("op_9748_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9748_cast = slice_by_index(begin = var_9748_begin_0, end = var_9748_end_0, end_mask = var_9748_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9748_cast")];
            tensor<int32, [4]> var_9752_begin_0 = const()[name = tensor<string, []>("op_9752_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_9752_end_0 = const()[name = tensor<string, []>("op_9752_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_9752_end_mask_0 = const()[name = tensor<string, []>("op_9752_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9752_cast = slice_by_index(begin = var_9752_begin_0, end = var_9752_end_0, end_mask = var_9752_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9752_cast")];
            tensor<int32, [4]> var_9756_begin_0 = const()[name = tensor<string, []>("op_9756_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_9756_end_0 = const()[name = tensor<string, []>("op_9756_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_9756_end_mask_0 = const()[name = tensor<string, []>("op_9756_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9756_cast = slice_by_index(begin = var_9756_begin_0, end = var_9756_end_0, end_mask = var_9756_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9756_cast")];
            tensor<int32, [4]> var_9760_begin_0 = const()[name = tensor<string, []>("op_9760_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_9760_end_0 = const()[name = tensor<string, []>("op_9760_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_9760_end_mask_0 = const()[name = tensor<string, []>("op_9760_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9760_cast = slice_by_index(begin = var_9760_begin_0, end = var_9760_end_0, end_mask = var_9760_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9760_cast")];
            tensor<int32, [4]> var_9764_begin_0 = const()[name = tensor<string, []>("op_9764_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_9764_end_0 = const()[name = tensor<string, []>("op_9764_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_9764_end_mask_0 = const()[name = tensor<string, []>("op_9764_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9764_cast = slice_by_index(begin = var_9764_begin_0, end = var_9764_end_0, end_mask = var_9764_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9764_cast")];
            tensor<int32, [4]> var_9768_begin_0 = const()[name = tensor<string, []>("op_9768_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9768_end_0 = const()[name = tensor<string, []>("op_9768_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_9768_end_mask_0 = const()[name = tensor<string, []>("op_9768_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9768_cast = slice_by_index(begin = var_9768_begin_0, end = var_9768_end_0, end_mask = var_9768_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9768_cast")];
            tensor<int32, [4]> var_9772_begin_0 = const()[name = tensor<string, []>("op_9772_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_9772_end_0 = const()[name = tensor<string, []>("op_9772_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_9772_end_mask_0 = const()[name = tensor<string, []>("op_9772_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9772_cast = slice_by_index(begin = var_9772_begin_0, end = var_9772_end_0, end_mask = var_9772_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9772_cast")];
            tensor<int32, [4]> var_9776_begin_0 = const()[name = tensor<string, []>("op_9776_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_9776_end_0 = const()[name = tensor<string, []>("op_9776_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_9776_end_mask_0 = const()[name = tensor<string, []>("op_9776_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9776_cast = slice_by_index(begin = var_9776_begin_0, end = var_9776_end_0, end_mask = var_9776_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9776_cast")];
            tensor<int32, [4]> var_9780_begin_0 = const()[name = tensor<string, []>("op_9780_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_9780_end_0 = const()[name = tensor<string, []>("op_9780_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_9780_end_mask_0 = const()[name = tensor<string, []>("op_9780_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9780_cast = slice_by_index(begin = var_9780_begin_0, end = var_9780_end_0, end_mask = var_9780_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9780_cast")];
            tensor<int32, [4]> var_9784_begin_0 = const()[name = tensor<string, []>("op_9784_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_9784_end_0 = const()[name = tensor<string, []>("op_9784_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_9784_end_mask_0 = const()[name = tensor<string, []>("op_9784_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9784_cast = slice_by_index(begin = var_9784_begin_0, end = var_9784_end_0, end_mask = var_9784_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9784_cast")];
            tensor<int32, [4]> var_9788_begin_0 = const()[name = tensor<string, []>("op_9788_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_9788_end_0 = const()[name = tensor<string, []>("op_9788_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_9788_end_mask_0 = const()[name = tensor<string, []>("op_9788_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9788_cast = slice_by_index(begin = var_9788_begin_0, end = var_9788_end_0, end_mask = var_9788_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9788_cast")];
            tensor<int32, [4]> var_9792_begin_0 = const()[name = tensor<string, []>("op_9792_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_9792_end_0 = const()[name = tensor<string, []>("op_9792_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_9792_end_mask_0 = const()[name = tensor<string, []>("op_9792_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9792_cast = slice_by_index(begin = var_9792_begin_0, end = var_9792_end_0, end_mask = var_9792_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9792_cast")];
            tensor<int32, [4]> var_9796_begin_0 = const()[name = tensor<string, []>("op_9796_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_9796_end_0 = const()[name = tensor<string, []>("op_9796_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_9796_end_mask_0 = const()[name = tensor<string, []>("op_9796_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9796_cast = slice_by_index(begin = var_9796_begin_0, end = var_9796_end_0, end_mask = var_9796_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9796_cast")];
            tensor<int32, [4]> var_9800_begin_0 = const()[name = tensor<string, []>("op_9800_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_9800_end_0 = const()[name = tensor<string, []>("op_9800_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_9800_end_mask_0 = const()[name = tensor<string, []>("op_9800_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9800_cast = slice_by_index(begin = var_9800_begin_0, end = var_9800_end_0, end_mask = var_9800_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9800_cast")];
            tensor<int32, [4]> var_9804_begin_0 = const()[name = tensor<string, []>("op_9804_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_9804_end_0 = const()[name = tensor<string, []>("op_9804_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_9804_end_mask_0 = const()[name = tensor<string, []>("op_9804_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9804_cast = slice_by_index(begin = var_9804_begin_0, end = var_9804_end_0, end_mask = var_9804_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9804_cast")];
            tensor<int32, [4]> var_9808_begin_0 = const()[name = tensor<string, []>("op_9808_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_9808_end_0 = const()[name = tensor<string, []>("op_9808_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_9808_end_mask_0 = const()[name = tensor<string, []>("op_9808_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9808_cast = slice_by_index(begin = var_9808_begin_0, end = var_9808_end_0, end_mask = var_9808_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9808_cast")];
            tensor<int32, [4]> var_9812_begin_0 = const()[name = tensor<string, []>("op_9812_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_9812_end_0 = const()[name = tensor<string, []>("op_9812_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_9812_end_mask_0 = const()[name = tensor<string, []>("op_9812_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9812_cast = slice_by_index(begin = var_9812_begin_0, end = var_9812_end_0, end_mask = var_9812_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9812_cast")];
            tensor<int32, [4]> var_9816_begin_0 = const()[name = tensor<string, []>("op_9816_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_9816_end_0 = const()[name = tensor<string, []>("op_9816_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_9816_end_mask_0 = const()[name = tensor<string, []>("op_9816_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9816_cast = slice_by_index(begin = var_9816_begin_0, end = var_9816_end_0, end_mask = var_9816_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9816_cast")];
            tensor<int32, [4]> var_9820_begin_0 = const()[name = tensor<string, []>("op_9820_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_9820_end_0 = const()[name = tensor<string, []>("op_9820_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_9820_end_mask_0 = const()[name = tensor<string, []>("op_9820_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9820_cast = slice_by_index(begin = var_9820_begin_0, end = var_9820_end_0, end_mask = var_9820_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9820_cast")];
            tensor<int32, [4]> var_9824_begin_0 = const()[name = tensor<string, []>("op_9824_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_9824_end_0 = const()[name = tensor<string, []>("op_9824_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_9824_end_mask_0 = const()[name = tensor<string, []>("op_9824_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9824_cast = slice_by_index(begin = var_9824_begin_0, end = var_9824_end_0, end_mask = var_9824_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9824_cast")];
            tensor<string, []> var_9828_equation_0 = const()[name = tensor<string, []>("op_9828_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9828_cast = einsum(equation = var_9828_equation_0, values = (var_9670_cast, var_9587_cast))[name = tensor<string, []>("op_9828_cast")];
            tensor<fp16, []> var_9829_to_fp16 = const()[name = tensor<string, []>("op_9829_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_761_cast = mul(x = var_9828_cast, y = var_9829_to_fp16)[name = tensor<string, []>("aw_761_cast")];
            tensor<string, []> var_9832_equation_0 = const()[name = tensor<string, []>("op_9832_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9832_cast = einsum(equation = var_9832_equation_0, values = (var_9674_cast, var_9591_cast))[name = tensor<string, []>("op_9832_cast")];
            tensor<fp16, []> var_9833_to_fp16 = const()[name = tensor<string, []>("op_9833_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_763_cast = mul(x = var_9832_cast, y = var_9833_to_fp16)[name = tensor<string, []>("aw_763_cast")];
            tensor<string, []> var_9836_equation_0 = const()[name = tensor<string, []>("op_9836_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9836_cast = einsum(equation = var_9836_equation_0, values = (var_9678_cast, var_9595_cast))[name = tensor<string, []>("op_9836_cast")];
            tensor<fp16, []> var_9837_to_fp16 = const()[name = tensor<string, []>("op_9837_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_765_cast = mul(x = var_9836_cast, y = var_9837_to_fp16)[name = tensor<string, []>("aw_765_cast")];
            tensor<string, []> var_9840_equation_0 = const()[name = tensor<string, []>("op_9840_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9840_cast = einsum(equation = var_9840_equation_0, values = (var_9682_cast, var_9599_cast))[name = tensor<string, []>("op_9840_cast")];
            tensor<fp16, []> var_9841_to_fp16 = const()[name = tensor<string, []>("op_9841_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_767_cast = mul(x = var_9840_cast, y = var_9841_to_fp16)[name = tensor<string, []>("aw_767_cast")];
            tensor<string, []> var_9844_equation_0 = const()[name = tensor<string, []>("op_9844_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9844_cast = einsum(equation = var_9844_equation_0, values = (var_9686_cast, var_9603_cast))[name = tensor<string, []>("op_9844_cast")];
            tensor<fp16, []> var_9845_to_fp16 = const()[name = tensor<string, []>("op_9845_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_769_cast = mul(x = var_9844_cast, y = var_9845_to_fp16)[name = tensor<string, []>("aw_769_cast")];
            tensor<string, []> var_9848_equation_0 = const()[name = tensor<string, []>("op_9848_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9848_cast = einsum(equation = var_9848_equation_0, values = (var_9690_cast, var_9607_cast))[name = tensor<string, []>("op_9848_cast")];
            tensor<fp16, []> var_9849_to_fp16 = const()[name = tensor<string, []>("op_9849_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_771_cast = mul(x = var_9848_cast, y = var_9849_to_fp16)[name = tensor<string, []>("aw_771_cast")];
            tensor<string, []> var_9852_equation_0 = const()[name = tensor<string, []>("op_9852_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9852_cast = einsum(equation = var_9852_equation_0, values = (var_9694_cast, var_9611_cast))[name = tensor<string, []>("op_9852_cast")];
            tensor<fp16, []> var_9853_to_fp16 = const()[name = tensor<string, []>("op_9853_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_773_cast = mul(x = var_9852_cast, y = var_9853_to_fp16)[name = tensor<string, []>("aw_773_cast")];
            tensor<string, []> var_9856_equation_0 = const()[name = tensor<string, []>("op_9856_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9856_cast = einsum(equation = var_9856_equation_0, values = (var_9698_cast, var_9615_cast))[name = tensor<string, []>("op_9856_cast")];
            tensor<fp16, []> var_9857_to_fp16 = const()[name = tensor<string, []>("op_9857_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_775_cast = mul(x = var_9856_cast, y = var_9857_to_fp16)[name = tensor<string, []>("aw_775_cast")];
            tensor<string, []> var_9860_equation_0 = const()[name = tensor<string, []>("op_9860_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9860_cast = einsum(equation = var_9860_equation_0, values = (var_9702_cast, var_9619_cast))[name = tensor<string, []>("op_9860_cast")];
            tensor<fp16, []> var_9861_to_fp16 = const()[name = tensor<string, []>("op_9861_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_777_cast = mul(x = var_9860_cast, y = var_9861_to_fp16)[name = tensor<string, []>("aw_777_cast")];
            tensor<string, []> var_9864_equation_0 = const()[name = tensor<string, []>("op_9864_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9864_cast = einsum(equation = var_9864_equation_0, values = (var_9706_cast, var_9623_cast))[name = tensor<string, []>("op_9864_cast")];
            tensor<fp16, []> var_9865_to_fp16 = const()[name = tensor<string, []>("op_9865_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_779_cast = mul(x = var_9864_cast, y = var_9865_to_fp16)[name = tensor<string, []>("aw_779_cast")];
            tensor<string, []> var_9868_equation_0 = const()[name = tensor<string, []>("op_9868_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9868_cast = einsum(equation = var_9868_equation_0, values = (var_9710_cast, var_9627_cast))[name = tensor<string, []>("op_9868_cast")];
            tensor<fp16, []> var_9869_to_fp16 = const()[name = tensor<string, []>("op_9869_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_781_cast = mul(x = var_9868_cast, y = var_9869_to_fp16)[name = tensor<string, []>("aw_781_cast")];
            tensor<string, []> var_9872_equation_0 = const()[name = tensor<string, []>("op_9872_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9872_cast = einsum(equation = var_9872_equation_0, values = (var_9714_cast, var_9631_cast))[name = tensor<string, []>("op_9872_cast")];
            tensor<fp16, []> var_9873_to_fp16 = const()[name = tensor<string, []>("op_9873_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_783_cast = mul(x = var_9872_cast, y = var_9873_to_fp16)[name = tensor<string, []>("aw_783_cast")];
            tensor<string, []> var_9876_equation_0 = const()[name = tensor<string, []>("op_9876_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9876_cast = einsum(equation = var_9876_equation_0, values = (var_9718_cast, var_9635_cast))[name = tensor<string, []>("op_9876_cast")];
            tensor<fp16, []> var_9877_to_fp16 = const()[name = tensor<string, []>("op_9877_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_785_cast = mul(x = var_9876_cast, y = var_9877_to_fp16)[name = tensor<string, []>("aw_785_cast")];
            tensor<string, []> var_9880_equation_0 = const()[name = tensor<string, []>("op_9880_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9880_cast = einsum(equation = var_9880_equation_0, values = (var_9722_cast, var_9639_cast))[name = tensor<string, []>("op_9880_cast")];
            tensor<fp16, []> var_9881_to_fp16 = const()[name = tensor<string, []>("op_9881_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_787_cast = mul(x = var_9880_cast, y = var_9881_to_fp16)[name = tensor<string, []>("aw_787_cast")];
            tensor<string, []> var_9884_equation_0 = const()[name = tensor<string, []>("op_9884_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9884_cast = einsum(equation = var_9884_equation_0, values = (var_9726_cast, var_9643_cast))[name = tensor<string, []>("op_9884_cast")];
            tensor<fp16, []> var_9885_to_fp16 = const()[name = tensor<string, []>("op_9885_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_789_cast = mul(x = var_9884_cast, y = var_9885_to_fp16)[name = tensor<string, []>("aw_789_cast")];
            tensor<string, []> var_9888_equation_0 = const()[name = tensor<string, []>("op_9888_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9888_cast = einsum(equation = var_9888_equation_0, values = (var_9730_cast, var_9647_cast))[name = tensor<string, []>("op_9888_cast")];
            tensor<fp16, []> var_9889_to_fp16 = const()[name = tensor<string, []>("op_9889_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_791_cast = mul(x = var_9888_cast, y = var_9889_to_fp16)[name = tensor<string, []>("aw_791_cast")];
            tensor<string, []> var_9892_equation_0 = const()[name = tensor<string, []>("op_9892_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9892_cast = einsum(equation = var_9892_equation_0, values = (var_9734_cast, var_9651_cast))[name = tensor<string, []>("op_9892_cast")];
            tensor<fp16, []> var_9893_to_fp16 = const()[name = tensor<string, []>("op_9893_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_793_cast = mul(x = var_9892_cast, y = var_9893_to_fp16)[name = tensor<string, []>("aw_793_cast")];
            tensor<string, []> var_9896_equation_0 = const()[name = tensor<string, []>("op_9896_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9896_cast = einsum(equation = var_9896_equation_0, values = (var_9738_cast, var_9655_cast))[name = tensor<string, []>("op_9896_cast")];
            tensor<fp16, []> var_9897_to_fp16 = const()[name = tensor<string, []>("op_9897_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_795_cast = mul(x = var_9896_cast, y = var_9897_to_fp16)[name = tensor<string, []>("aw_795_cast")];
            tensor<string, []> var_9900_equation_0 = const()[name = tensor<string, []>("op_9900_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9900_cast = einsum(equation = var_9900_equation_0, values = (var_9742_cast, var_9659_cast))[name = tensor<string, []>("op_9900_cast")];
            tensor<fp16, []> var_9901_to_fp16 = const()[name = tensor<string, []>("op_9901_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_797_cast = mul(x = var_9900_cast, y = var_9901_to_fp16)[name = tensor<string, []>("aw_797_cast")];
            tensor<string, []> var_9904_equation_0 = const()[name = tensor<string, []>("op_9904_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9904_cast = einsum(equation = var_9904_equation_0, values = (var_9746_cast, var_9663_cast))[name = tensor<string, []>("op_9904_cast")];
            tensor<fp16, []> var_9905_to_fp16 = const()[name = tensor<string, []>("op_9905_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_799_cast = mul(x = var_9904_cast, y = var_9905_to_fp16)[name = tensor<string, []>("aw_799_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9907_cast = softmax(axis = var_2634, x = aw_761_cast)[name = tensor<string, []>("op_9907_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9908_cast = softmax(axis = var_2634, x = aw_763_cast)[name = tensor<string, []>("op_9908_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9909_cast = softmax(axis = var_2634, x = aw_765_cast)[name = tensor<string, []>("op_9909_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9910_cast = softmax(axis = var_2634, x = aw_767_cast)[name = tensor<string, []>("op_9910_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9911_cast = softmax(axis = var_2634, x = aw_769_cast)[name = tensor<string, []>("op_9911_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9912_cast = softmax(axis = var_2634, x = aw_771_cast)[name = tensor<string, []>("op_9912_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9913_cast = softmax(axis = var_2634, x = aw_773_cast)[name = tensor<string, []>("op_9913_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9914_cast = softmax(axis = var_2634, x = aw_775_cast)[name = tensor<string, []>("op_9914_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9915_cast = softmax(axis = var_2634, x = aw_777_cast)[name = tensor<string, []>("op_9915_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9916_cast = softmax(axis = var_2634, x = aw_779_cast)[name = tensor<string, []>("op_9916_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9917_cast = softmax(axis = var_2634, x = aw_781_cast)[name = tensor<string, []>("op_9917_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9918_cast = softmax(axis = var_2634, x = aw_783_cast)[name = tensor<string, []>("op_9918_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9919_cast = softmax(axis = var_2634, x = aw_785_cast)[name = tensor<string, []>("op_9919_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9920_cast = softmax(axis = var_2634, x = aw_787_cast)[name = tensor<string, []>("op_9920_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9921_cast = softmax(axis = var_2634, x = aw_789_cast)[name = tensor<string, []>("op_9921_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9922_cast = softmax(axis = var_2634, x = aw_791_cast)[name = tensor<string, []>("op_9922_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9923_cast = softmax(axis = var_2634, x = aw_793_cast)[name = tensor<string, []>("op_9923_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9924_cast = softmax(axis = var_2634, x = aw_795_cast)[name = tensor<string, []>("op_9924_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9925_cast = softmax(axis = var_2634, x = aw_797_cast)[name = tensor<string, []>("op_9925_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_9926_cast = softmax(axis = var_2634, x = aw_799_cast)[name = tensor<string, []>("op_9926_cast")];
            tensor<string, []> var_9928_equation_0 = const()[name = tensor<string, []>("op_9928_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9928_cast = einsum(equation = var_9928_equation_0, values = (var_9748_cast, var_9907_cast))[name = tensor<string, []>("op_9928_cast")];
            tensor<string, []> var_9930_equation_0 = const()[name = tensor<string, []>("op_9930_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9930_cast = einsum(equation = var_9930_equation_0, values = (var_9752_cast, var_9908_cast))[name = tensor<string, []>("op_9930_cast")];
            tensor<string, []> var_9932_equation_0 = const()[name = tensor<string, []>("op_9932_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9932_cast = einsum(equation = var_9932_equation_0, values = (var_9756_cast, var_9909_cast))[name = tensor<string, []>("op_9932_cast")];
            tensor<string, []> var_9934_equation_0 = const()[name = tensor<string, []>("op_9934_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9934_cast = einsum(equation = var_9934_equation_0, values = (var_9760_cast, var_9910_cast))[name = tensor<string, []>("op_9934_cast")];
            tensor<string, []> var_9936_equation_0 = const()[name = tensor<string, []>("op_9936_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9936_cast = einsum(equation = var_9936_equation_0, values = (var_9764_cast, var_9911_cast))[name = tensor<string, []>("op_9936_cast")];
            tensor<string, []> var_9938_equation_0 = const()[name = tensor<string, []>("op_9938_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9938_cast = einsum(equation = var_9938_equation_0, values = (var_9768_cast, var_9912_cast))[name = tensor<string, []>("op_9938_cast")];
            tensor<string, []> var_9940_equation_0 = const()[name = tensor<string, []>("op_9940_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9940_cast = einsum(equation = var_9940_equation_0, values = (var_9772_cast, var_9913_cast))[name = tensor<string, []>("op_9940_cast")];
            tensor<string, []> var_9942_equation_0 = const()[name = tensor<string, []>("op_9942_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9942_cast = einsum(equation = var_9942_equation_0, values = (var_9776_cast, var_9914_cast))[name = tensor<string, []>("op_9942_cast")];
            tensor<string, []> var_9944_equation_0 = const()[name = tensor<string, []>("op_9944_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9944_cast = einsum(equation = var_9944_equation_0, values = (var_9780_cast, var_9915_cast))[name = tensor<string, []>("op_9944_cast")];
            tensor<string, []> var_9946_equation_0 = const()[name = tensor<string, []>("op_9946_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9946_cast = einsum(equation = var_9946_equation_0, values = (var_9784_cast, var_9916_cast))[name = tensor<string, []>("op_9946_cast")];
            tensor<string, []> var_9948_equation_0 = const()[name = tensor<string, []>("op_9948_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9948_cast = einsum(equation = var_9948_equation_0, values = (var_9788_cast, var_9917_cast))[name = tensor<string, []>("op_9948_cast")];
            tensor<string, []> var_9950_equation_0 = const()[name = tensor<string, []>("op_9950_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9950_cast = einsum(equation = var_9950_equation_0, values = (var_9792_cast, var_9918_cast))[name = tensor<string, []>("op_9950_cast")];
            tensor<string, []> var_9952_equation_0 = const()[name = tensor<string, []>("op_9952_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9952_cast = einsum(equation = var_9952_equation_0, values = (var_9796_cast, var_9919_cast))[name = tensor<string, []>("op_9952_cast")];
            tensor<string, []> var_9954_equation_0 = const()[name = tensor<string, []>("op_9954_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9954_cast = einsum(equation = var_9954_equation_0, values = (var_9800_cast, var_9920_cast))[name = tensor<string, []>("op_9954_cast")];
            tensor<string, []> var_9956_equation_0 = const()[name = tensor<string, []>("op_9956_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9956_cast = einsum(equation = var_9956_equation_0, values = (var_9804_cast, var_9921_cast))[name = tensor<string, []>("op_9956_cast")];
            tensor<string, []> var_9958_equation_0 = const()[name = tensor<string, []>("op_9958_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9958_cast = einsum(equation = var_9958_equation_0, values = (var_9808_cast, var_9922_cast))[name = tensor<string, []>("op_9958_cast")];
            tensor<string, []> var_9960_equation_0 = const()[name = tensor<string, []>("op_9960_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9960_cast = einsum(equation = var_9960_equation_0, values = (var_9812_cast, var_9923_cast))[name = tensor<string, []>("op_9960_cast")];
            tensor<string, []> var_9962_equation_0 = const()[name = tensor<string, []>("op_9962_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9962_cast = einsum(equation = var_9962_equation_0, values = (var_9816_cast, var_9924_cast))[name = tensor<string, []>("op_9962_cast")];
            tensor<string, []> var_9964_equation_0 = const()[name = tensor<string, []>("op_9964_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9964_cast = einsum(equation = var_9964_equation_0, values = (var_9820_cast, var_9925_cast))[name = tensor<string, []>("op_9964_cast")];
            tensor<string, []> var_9966_equation_0 = const()[name = tensor<string, []>("op_9966_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9966_cast = einsum(equation = var_9966_equation_0, values = (var_9824_cast, var_9926_cast))[name = tensor<string, []>("op_9966_cast")];
            tensor<bool, []> input_193_interleave_0 = const()[name = tensor<string, []>("input_193_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_193_cast = concat(axis = var_2634, interleave = input_193_interleave_0, values = (var_9928_cast, var_9930_cast, var_9932_cast, var_9934_cast, var_9936_cast, var_9938_cast, var_9940_cast, var_9942_cast, var_9944_cast, var_9946_cast, var_9948_cast, var_9950_cast, var_9952_cast, var_9954_cast, var_9956_cast, var_9958_cast, var_9960_cast, var_9962_cast, var_9964_cast, var_9966_cast))[name = tensor<string, []>("input_193_cast")];
            tensor<int32, [2]> var_9972 = const()[name = tensor<string, []>("op_9972"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9974 = const()[name = tensor<string, []>("op_9974"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9976_pad_type_0 = const()[name = tensor<string, []>("op_9976_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9976_pad_0 = const()[name = tensor<string, []>("op_9976_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(711639936)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(714916800)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_9976_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16, dilations = var_9974, groups = var_2634, pad = var_9976_pad_0, pad_type = var_9976_pad_type_0, strides = var_9972, weight = down_blocks_2_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16, x = input_193_cast)[name = tensor<string, []>("op_9976_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_71_cast = add(x = var_9976_cast, y = inputs_69_cast)[name = tensor<string, []>("inputs_71_cast")];
            tensor<int32, [1]> var_9980 = const()[name = tensor<string, []>("op_9980"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_71_cast = reduce_mean(axes = var_9980, keep_dims = var_2629, x = inputs_71_cast)[name = tensor<string, []>("channels_mean_71_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_71_cast = sub(x = inputs_71_cast, y = channels_mean_71_cast)[name = tensor<string, []>("zero_mean_71_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_71_cast = mul(x = zero_mean_71_cast, y = zero_mean_71_cast)[name = tensor<string, []>("zero_mean_sq_71_cast")];
            tensor<int32, [1]> var_9984 = const()[name = tensor<string, []>("op_9984"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_9985_cast = reduce_mean(axes = var_9984, keep_dims = var_2629, x = zero_mean_sq_71_cast)[name = tensor<string, []>("op_9985_cast")];
            tensor<fp16, []> var_9986_to_fp16 = const()[name = tensor<string, []>("op_9986_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_9987_cast = add(x = var_9985_cast, y = var_9986_to_fp16)[name = tensor<string, []>("op_9987_cast")];
            tensor<fp16, []> denom_71_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_71_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_71_cast = rsqrt(epsilon = denom_71_epsilon_0_to_fp16, x = var_9987_cast)[name = tensor<string, []>("denom_71_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_71_cast = mul(x = zero_mean_71_cast, y = denom_71_cast)[name = tensor<string, []>("out_71_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_9991_to_fp16 = const()[name = tensor<string, []>("op_9991_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(714919424)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_9992_cast = add(x = out_71_cast, y = var_9991_to_fp16)[name = tensor<string, []>("op_9992_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_9994_to_fp16 = const()[name = tensor<string, []>("op_9994_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(714922048)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_195_cast = mul(x = var_9992_cast, y = var_9994_to_fp16)[name = tensor<string, []>("input_195_cast")];
            tensor<int32, [2]> var_10002 = const()[name = tensor<string, []>("op_10002"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10004 = const()[name = tensor<string, []>("op_10004"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10006_pad_type_0 = const()[name = tensor<string, []>("op_10006_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10006_pad_0 = const()[name = tensor<string, []>("op_10006_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(714924672)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(741139136)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_10006_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16, dilations = var_10004, groups = var_2634, pad = var_10006_pad_0, pad_type = var_10006_pad_type_0, strides = var_10002, weight = down_blocks_2_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16, x = input_195_cast)[name = tensor<string, []>("op_10006_cast")];
            tensor<int32, [2]> var_10007_split_sizes_0 = const()[name = tensor<string, []>("op_10007_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_10007_axis_0 = const()[name = tensor<string, []>("op_10007_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_10007_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_10007_cast_1 = split(axis = var_10007_axis_0, split_sizes = var_10007_split_sizes_0, x = var_10006_cast)[name = tensor<string, []>("op_10007_cast")];
            tensor<string, []> var_10009_mode_0 = const()[name = tensor<string, []>("op_10009_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_10009_cast = gelu(mode = var_10009_mode_0, x = var_10007_cast_1)[name = tensor<string, []>("op_10009_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_197_cast = mul(x = var_10007_cast_0, y = var_10009_cast)[name = tensor<string, []>("input_197_cast")];
            tensor<int32, [2]> var_10013 = const()[name = tensor<string, []>("op_10013"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10015 = const()[name = tensor<string, []>("op_10015"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10017_pad_type_0 = const()[name = tensor<string, []>("op_10017_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10017_pad_0 = const()[name = tensor<string, []>("op_10017_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(741159680)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(754266944)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_10017_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16, dilations = var_10015, groups = var_2634, pad = var_10017_pad_0, pad_type = var_10017_pad_type_0, strides = var_10013, weight = down_blocks_2_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16, x = input_197_cast)[name = tensor<string, []>("op_10017_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_73_cast = add(x = var_10017_cast, y = inputs_71_cast)[name = tensor<string, []>("inputs_73_cast")];
            tensor<int32, [1]> var_10027 = const()[name = tensor<string, []>("op_10027"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_73_cast = reduce_mean(axes = var_10027, keep_dims = var_2629, x = inputs_73_cast)[name = tensor<string, []>("channels_mean_73_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_73_cast = sub(x = inputs_73_cast, y = channels_mean_73_cast)[name = tensor<string, []>("zero_mean_73_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_73_cast = mul(x = zero_mean_73_cast, y = zero_mean_73_cast)[name = tensor<string, []>("zero_mean_sq_73_cast")];
            tensor<int32, [1]> var_10031 = const()[name = tensor<string, []>("op_10031"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_10032_cast = reduce_mean(axes = var_10031, keep_dims = var_2629, x = zero_mean_sq_73_cast)[name = tensor<string, []>("op_10032_cast")];
            tensor<fp16, []> var_10033_to_fp16 = const()[name = tensor<string, []>("op_10033_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_10034_cast = add(x = var_10032_cast, y = var_10033_to_fp16)[name = tensor<string, []>("op_10034_cast")];
            tensor<fp16, []> denom_73_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_73_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_73_cast = rsqrt(epsilon = denom_73_epsilon_0_to_fp16, x = var_10034_cast)[name = tensor<string, []>("denom_73_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_73_cast = mul(x = zero_mean_73_cast, y = denom_73_cast)[name = tensor<string, []>("out_73_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_10038_to_fp16 = const()[name = tensor<string, []>("op_10038_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(754269568)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_10039_cast = add(x = out_73_cast, y = var_10038_to_fp16)[name = tensor<string, []>("op_10039_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_10041_to_fp16 = const()[name = tensor<string, []>("op_10041_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(754272192)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_113_cast = mul(x = var_10039_cast, y = var_10041_to_fp16)[name = tensor<string, []>("hidden_states_113_cast")];
            tensor<int32, [2]> var_10048 = const()[name = tensor<string, []>("op_10048"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10050 = const()[name = tensor<string, []>("op_10050"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_49_pad_type_0 = const()[name = tensor<string, []>("q_49_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_49_pad_0 = const()[name = tensor<string, []>("q_49_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(754274816)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_49_cast = conv(dilations = var_10050, groups = var_2634, pad = q_49_pad_0, pad_type = q_49_pad_type_0, strides = var_10048, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16, x = hidden_states_113_cast)[name = tensor<string, []>("q_49_cast")];
            tensor<int32, [2]> var_10054 = const()[name = tensor<string, []>("op_10054"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10056 = const()[name = tensor<string, []>("op_10056"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_97_pad_type_0 = const()[name = tensor<string, []>("k_97_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_97_pad_0 = const()[name = tensor<string, []>("k_97_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(757551680)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_97_cast = conv(dilations = var_10056, groups = var_2634, pad = k_97_pad_0, pad_type = k_97_pad_type_0, strides = var_10054, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16, x = hidden_states_113_cast)[name = tensor<string, []>("k_97_cast")];
            tensor<int32, [2]> var_10060 = const()[name = tensor<string, []>("op_10060"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10062 = const()[name = tensor<string, []>("op_10062"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_49_pad_type_0 = const()[name = tensor<string, []>("v_49_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_49_pad_0 = const()[name = tensor<string, []>("v_49_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(760828544)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_49_cast = conv(dilations = var_10062, groups = var_2634, pad = v_49_pad_0, pad_type = v_49_pad_type_0, strides = var_10060, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16, x = hidden_states_113_cast)[name = tensor<string, []>("v_49_cast")];
            tensor<int32, [4]> var_10066_begin_0 = const()[name = tensor<string, []>("op_10066_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10066_end_0 = const()[name = tensor<string, []>("op_10066_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10066_end_mask_0 = const()[name = tensor<string, []>("op_10066_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10066_cast = slice_by_index(begin = var_10066_begin_0, end = var_10066_end_0, end_mask = var_10066_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10066_cast")];
            tensor<int32, [4]> var_10070_begin_0 = const()[name = tensor<string, []>("op_10070_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_10070_end_0 = const()[name = tensor<string, []>("op_10070_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_10070_end_mask_0 = const()[name = tensor<string, []>("op_10070_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10070_cast = slice_by_index(begin = var_10070_begin_0, end = var_10070_end_0, end_mask = var_10070_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10070_cast")];
            tensor<int32, [4]> var_10074_begin_0 = const()[name = tensor<string, []>("op_10074_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_10074_end_0 = const()[name = tensor<string, []>("op_10074_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_10074_end_mask_0 = const()[name = tensor<string, []>("op_10074_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10074_cast = slice_by_index(begin = var_10074_begin_0, end = var_10074_end_0, end_mask = var_10074_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10074_cast")];
            tensor<int32, [4]> var_10078_begin_0 = const()[name = tensor<string, []>("op_10078_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_10078_end_0 = const()[name = tensor<string, []>("op_10078_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_10078_end_mask_0 = const()[name = tensor<string, []>("op_10078_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10078_cast = slice_by_index(begin = var_10078_begin_0, end = var_10078_end_0, end_mask = var_10078_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10078_cast")];
            tensor<int32, [4]> var_10082_begin_0 = const()[name = tensor<string, []>("op_10082_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_10082_end_0 = const()[name = tensor<string, []>("op_10082_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_10082_end_mask_0 = const()[name = tensor<string, []>("op_10082_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10082_cast = slice_by_index(begin = var_10082_begin_0, end = var_10082_end_0, end_mask = var_10082_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10082_cast")];
            tensor<int32, [4]> var_10086_begin_0 = const()[name = tensor<string, []>("op_10086_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_10086_end_0 = const()[name = tensor<string, []>("op_10086_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_10086_end_mask_0 = const()[name = tensor<string, []>("op_10086_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10086_cast = slice_by_index(begin = var_10086_begin_0, end = var_10086_end_0, end_mask = var_10086_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10086_cast")];
            tensor<int32, [4]> var_10090_begin_0 = const()[name = tensor<string, []>("op_10090_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_10090_end_0 = const()[name = tensor<string, []>("op_10090_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_10090_end_mask_0 = const()[name = tensor<string, []>("op_10090_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10090_cast = slice_by_index(begin = var_10090_begin_0, end = var_10090_end_0, end_mask = var_10090_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10090_cast")];
            tensor<int32, [4]> var_10094_begin_0 = const()[name = tensor<string, []>("op_10094_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_10094_end_0 = const()[name = tensor<string, []>("op_10094_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_10094_end_mask_0 = const()[name = tensor<string, []>("op_10094_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10094_cast = slice_by_index(begin = var_10094_begin_0, end = var_10094_end_0, end_mask = var_10094_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10094_cast")];
            tensor<int32, [4]> var_10098_begin_0 = const()[name = tensor<string, []>("op_10098_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_10098_end_0 = const()[name = tensor<string, []>("op_10098_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_10098_end_mask_0 = const()[name = tensor<string, []>("op_10098_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10098_cast = slice_by_index(begin = var_10098_begin_0, end = var_10098_end_0, end_mask = var_10098_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10098_cast")];
            tensor<int32, [4]> var_10102_begin_0 = const()[name = tensor<string, []>("op_10102_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_10102_end_0 = const()[name = tensor<string, []>("op_10102_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_10102_end_mask_0 = const()[name = tensor<string, []>("op_10102_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10102_cast = slice_by_index(begin = var_10102_begin_0, end = var_10102_end_0, end_mask = var_10102_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10102_cast")];
            tensor<int32, [4]> var_10106_begin_0 = const()[name = tensor<string, []>("op_10106_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_10106_end_0 = const()[name = tensor<string, []>("op_10106_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_10106_end_mask_0 = const()[name = tensor<string, []>("op_10106_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10106_cast = slice_by_index(begin = var_10106_begin_0, end = var_10106_end_0, end_mask = var_10106_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10106_cast")];
            tensor<int32, [4]> var_10110_begin_0 = const()[name = tensor<string, []>("op_10110_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_10110_end_0 = const()[name = tensor<string, []>("op_10110_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_10110_end_mask_0 = const()[name = tensor<string, []>("op_10110_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10110_cast = slice_by_index(begin = var_10110_begin_0, end = var_10110_end_0, end_mask = var_10110_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10110_cast")];
            tensor<int32, [4]> var_10114_begin_0 = const()[name = tensor<string, []>("op_10114_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_10114_end_0 = const()[name = tensor<string, []>("op_10114_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_10114_end_mask_0 = const()[name = tensor<string, []>("op_10114_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10114_cast = slice_by_index(begin = var_10114_begin_0, end = var_10114_end_0, end_mask = var_10114_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10114_cast")];
            tensor<int32, [4]> var_10118_begin_0 = const()[name = tensor<string, []>("op_10118_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_10118_end_0 = const()[name = tensor<string, []>("op_10118_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_10118_end_mask_0 = const()[name = tensor<string, []>("op_10118_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10118_cast = slice_by_index(begin = var_10118_begin_0, end = var_10118_end_0, end_mask = var_10118_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10118_cast")];
            tensor<int32, [4]> var_10122_begin_0 = const()[name = tensor<string, []>("op_10122_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_10122_end_0 = const()[name = tensor<string, []>("op_10122_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_10122_end_mask_0 = const()[name = tensor<string, []>("op_10122_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10122_cast = slice_by_index(begin = var_10122_begin_0, end = var_10122_end_0, end_mask = var_10122_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10122_cast")];
            tensor<int32, [4]> var_10126_begin_0 = const()[name = tensor<string, []>("op_10126_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_10126_end_0 = const()[name = tensor<string, []>("op_10126_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_10126_end_mask_0 = const()[name = tensor<string, []>("op_10126_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10126_cast = slice_by_index(begin = var_10126_begin_0, end = var_10126_end_0, end_mask = var_10126_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10126_cast")];
            tensor<int32, [4]> var_10130_begin_0 = const()[name = tensor<string, []>("op_10130_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_10130_end_0 = const()[name = tensor<string, []>("op_10130_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_10130_end_mask_0 = const()[name = tensor<string, []>("op_10130_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10130_cast = slice_by_index(begin = var_10130_begin_0, end = var_10130_end_0, end_mask = var_10130_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10130_cast")];
            tensor<int32, [4]> var_10134_begin_0 = const()[name = tensor<string, []>("op_10134_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_10134_end_0 = const()[name = tensor<string, []>("op_10134_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_10134_end_mask_0 = const()[name = tensor<string, []>("op_10134_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10134_cast = slice_by_index(begin = var_10134_begin_0, end = var_10134_end_0, end_mask = var_10134_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10134_cast")];
            tensor<int32, [4]> var_10138_begin_0 = const()[name = tensor<string, []>("op_10138_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_10138_end_0 = const()[name = tensor<string, []>("op_10138_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_10138_end_mask_0 = const()[name = tensor<string, []>("op_10138_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10138_cast = slice_by_index(begin = var_10138_begin_0, end = var_10138_end_0, end_mask = var_10138_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10138_cast")];
            tensor<int32, [4]> var_10142_begin_0 = const()[name = tensor<string, []>("op_10142_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_10142_end_0 = const()[name = tensor<string, []>("op_10142_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_10142_end_mask_0 = const()[name = tensor<string, []>("op_10142_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10142_cast = slice_by_index(begin = var_10142_begin_0, end = var_10142_end_0, end_mask = var_10142_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_10142_cast")];
            tensor<int32, [4]> k_99_perm_0 = const()[name = tensor<string, []>("k_99_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_10149_begin_0 = const()[name = tensor<string, []>("op_10149_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10149_end_0 = const()[name = tensor<string, []>("op_10149_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_10149_end_mask_0 = const()[name = tensor<string, []>("op_10149_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_115 = transpose(perm = k_99_perm_0, x = k_97_cast)[name = tensor<string, []>("transpose_115")];
            tensor<fp16, [2, 1024, 1, 64]> var_10149_cast = slice_by_index(begin = var_10149_begin_0, end = var_10149_end_0, end_mask = var_10149_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10149_cast")];
            tensor<int32, [4]> var_10153_begin_0 = const()[name = tensor<string, []>("op_10153_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_10153_end_0 = const()[name = tensor<string, []>("op_10153_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_10153_end_mask_0 = const()[name = tensor<string, []>("op_10153_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10153_cast = slice_by_index(begin = var_10153_begin_0, end = var_10153_end_0, end_mask = var_10153_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10153_cast")];
            tensor<int32, [4]> var_10157_begin_0 = const()[name = tensor<string, []>("op_10157_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_10157_end_0 = const()[name = tensor<string, []>("op_10157_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_10157_end_mask_0 = const()[name = tensor<string, []>("op_10157_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10157_cast = slice_by_index(begin = var_10157_begin_0, end = var_10157_end_0, end_mask = var_10157_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10157_cast")];
            tensor<int32, [4]> var_10161_begin_0 = const()[name = tensor<string, []>("op_10161_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_10161_end_0 = const()[name = tensor<string, []>("op_10161_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_10161_end_mask_0 = const()[name = tensor<string, []>("op_10161_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10161_cast = slice_by_index(begin = var_10161_begin_0, end = var_10161_end_0, end_mask = var_10161_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10161_cast")];
            tensor<int32, [4]> var_10165_begin_0 = const()[name = tensor<string, []>("op_10165_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_10165_end_0 = const()[name = tensor<string, []>("op_10165_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_10165_end_mask_0 = const()[name = tensor<string, []>("op_10165_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10165_cast = slice_by_index(begin = var_10165_begin_0, end = var_10165_end_0, end_mask = var_10165_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10165_cast")];
            tensor<int32, [4]> var_10169_begin_0 = const()[name = tensor<string, []>("op_10169_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_10169_end_0 = const()[name = tensor<string, []>("op_10169_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_10169_end_mask_0 = const()[name = tensor<string, []>("op_10169_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10169_cast = slice_by_index(begin = var_10169_begin_0, end = var_10169_end_0, end_mask = var_10169_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10169_cast")];
            tensor<int32, [4]> var_10173_begin_0 = const()[name = tensor<string, []>("op_10173_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_10173_end_0 = const()[name = tensor<string, []>("op_10173_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_10173_end_mask_0 = const()[name = tensor<string, []>("op_10173_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10173_cast = slice_by_index(begin = var_10173_begin_0, end = var_10173_end_0, end_mask = var_10173_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10173_cast")];
            tensor<int32, [4]> var_10177_begin_0 = const()[name = tensor<string, []>("op_10177_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_10177_end_0 = const()[name = tensor<string, []>("op_10177_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_10177_end_mask_0 = const()[name = tensor<string, []>("op_10177_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10177_cast = slice_by_index(begin = var_10177_begin_0, end = var_10177_end_0, end_mask = var_10177_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10177_cast")];
            tensor<int32, [4]> var_10181_begin_0 = const()[name = tensor<string, []>("op_10181_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10181_end_0 = const()[name = tensor<string, []>("op_10181_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_10181_end_mask_0 = const()[name = tensor<string, []>("op_10181_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10181_cast = slice_by_index(begin = var_10181_begin_0, end = var_10181_end_0, end_mask = var_10181_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10181_cast")];
            tensor<int32, [4]> var_10185_begin_0 = const()[name = tensor<string, []>("op_10185_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_10185_end_0 = const()[name = tensor<string, []>("op_10185_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_10185_end_mask_0 = const()[name = tensor<string, []>("op_10185_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10185_cast = slice_by_index(begin = var_10185_begin_0, end = var_10185_end_0, end_mask = var_10185_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10185_cast")];
            tensor<int32, [4]> var_10189_begin_0 = const()[name = tensor<string, []>("op_10189_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_10189_end_0 = const()[name = tensor<string, []>("op_10189_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_10189_end_mask_0 = const()[name = tensor<string, []>("op_10189_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10189_cast = slice_by_index(begin = var_10189_begin_0, end = var_10189_end_0, end_mask = var_10189_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10189_cast")];
            tensor<int32, [4]> var_10193_begin_0 = const()[name = tensor<string, []>("op_10193_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_10193_end_0 = const()[name = tensor<string, []>("op_10193_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_10193_end_mask_0 = const()[name = tensor<string, []>("op_10193_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10193_cast = slice_by_index(begin = var_10193_begin_0, end = var_10193_end_0, end_mask = var_10193_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10193_cast")];
            tensor<int32, [4]> var_10197_begin_0 = const()[name = tensor<string, []>("op_10197_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_10197_end_0 = const()[name = tensor<string, []>("op_10197_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_10197_end_mask_0 = const()[name = tensor<string, []>("op_10197_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10197_cast = slice_by_index(begin = var_10197_begin_0, end = var_10197_end_0, end_mask = var_10197_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10197_cast")];
            tensor<int32, [4]> var_10201_begin_0 = const()[name = tensor<string, []>("op_10201_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_10201_end_0 = const()[name = tensor<string, []>("op_10201_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_10201_end_mask_0 = const()[name = tensor<string, []>("op_10201_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10201_cast = slice_by_index(begin = var_10201_begin_0, end = var_10201_end_0, end_mask = var_10201_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10201_cast")];
            tensor<int32, [4]> var_10205_begin_0 = const()[name = tensor<string, []>("op_10205_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_10205_end_0 = const()[name = tensor<string, []>("op_10205_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_10205_end_mask_0 = const()[name = tensor<string, []>("op_10205_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10205_cast = slice_by_index(begin = var_10205_begin_0, end = var_10205_end_0, end_mask = var_10205_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10205_cast")];
            tensor<int32, [4]> var_10209_begin_0 = const()[name = tensor<string, []>("op_10209_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_10209_end_0 = const()[name = tensor<string, []>("op_10209_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_10209_end_mask_0 = const()[name = tensor<string, []>("op_10209_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10209_cast = slice_by_index(begin = var_10209_begin_0, end = var_10209_end_0, end_mask = var_10209_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10209_cast")];
            tensor<int32, [4]> var_10213_begin_0 = const()[name = tensor<string, []>("op_10213_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_10213_end_0 = const()[name = tensor<string, []>("op_10213_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_10213_end_mask_0 = const()[name = tensor<string, []>("op_10213_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10213_cast = slice_by_index(begin = var_10213_begin_0, end = var_10213_end_0, end_mask = var_10213_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10213_cast")];
            tensor<int32, [4]> var_10217_begin_0 = const()[name = tensor<string, []>("op_10217_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_10217_end_0 = const()[name = tensor<string, []>("op_10217_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_10217_end_mask_0 = const()[name = tensor<string, []>("op_10217_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10217_cast = slice_by_index(begin = var_10217_begin_0, end = var_10217_end_0, end_mask = var_10217_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10217_cast")];
            tensor<int32, [4]> var_10221_begin_0 = const()[name = tensor<string, []>("op_10221_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_10221_end_0 = const()[name = tensor<string, []>("op_10221_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_10221_end_mask_0 = const()[name = tensor<string, []>("op_10221_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10221_cast = slice_by_index(begin = var_10221_begin_0, end = var_10221_end_0, end_mask = var_10221_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10221_cast")];
            tensor<int32, [4]> var_10225_begin_0 = const()[name = tensor<string, []>("op_10225_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_10225_end_0 = const()[name = tensor<string, []>("op_10225_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_10225_end_mask_0 = const()[name = tensor<string, []>("op_10225_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10225_cast = slice_by_index(begin = var_10225_begin_0, end = var_10225_end_0, end_mask = var_10225_end_mask_0, x = transpose_115)[name = tensor<string, []>("op_10225_cast")];
            tensor<int32, [4]> var_10227_begin_0 = const()[name = tensor<string, []>("op_10227_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10227_end_0 = const()[name = tensor<string, []>("op_10227_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10227_end_mask_0 = const()[name = tensor<string, []>("op_10227_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10227_cast = slice_by_index(begin = var_10227_begin_0, end = var_10227_end_0, end_mask = var_10227_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10227_cast")];
            tensor<int32, [4]> var_10231_begin_0 = const()[name = tensor<string, []>("op_10231_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_10231_end_0 = const()[name = tensor<string, []>("op_10231_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_10231_end_mask_0 = const()[name = tensor<string, []>("op_10231_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10231_cast = slice_by_index(begin = var_10231_begin_0, end = var_10231_end_0, end_mask = var_10231_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10231_cast")];
            tensor<int32, [4]> var_10235_begin_0 = const()[name = tensor<string, []>("op_10235_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_10235_end_0 = const()[name = tensor<string, []>("op_10235_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_10235_end_mask_0 = const()[name = tensor<string, []>("op_10235_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10235_cast = slice_by_index(begin = var_10235_begin_0, end = var_10235_end_0, end_mask = var_10235_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10235_cast")];
            tensor<int32, [4]> var_10239_begin_0 = const()[name = tensor<string, []>("op_10239_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_10239_end_0 = const()[name = tensor<string, []>("op_10239_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_10239_end_mask_0 = const()[name = tensor<string, []>("op_10239_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10239_cast = slice_by_index(begin = var_10239_begin_0, end = var_10239_end_0, end_mask = var_10239_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10239_cast")];
            tensor<int32, [4]> var_10243_begin_0 = const()[name = tensor<string, []>("op_10243_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_10243_end_0 = const()[name = tensor<string, []>("op_10243_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_10243_end_mask_0 = const()[name = tensor<string, []>("op_10243_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10243_cast = slice_by_index(begin = var_10243_begin_0, end = var_10243_end_0, end_mask = var_10243_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10243_cast")];
            tensor<int32, [4]> var_10247_begin_0 = const()[name = tensor<string, []>("op_10247_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_10247_end_0 = const()[name = tensor<string, []>("op_10247_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_10247_end_mask_0 = const()[name = tensor<string, []>("op_10247_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10247_cast = slice_by_index(begin = var_10247_begin_0, end = var_10247_end_0, end_mask = var_10247_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10247_cast")];
            tensor<int32, [4]> var_10251_begin_0 = const()[name = tensor<string, []>("op_10251_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_10251_end_0 = const()[name = tensor<string, []>("op_10251_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_10251_end_mask_0 = const()[name = tensor<string, []>("op_10251_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10251_cast = slice_by_index(begin = var_10251_begin_0, end = var_10251_end_0, end_mask = var_10251_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10251_cast")];
            tensor<int32, [4]> var_10255_begin_0 = const()[name = tensor<string, []>("op_10255_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_10255_end_0 = const()[name = tensor<string, []>("op_10255_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_10255_end_mask_0 = const()[name = tensor<string, []>("op_10255_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10255_cast = slice_by_index(begin = var_10255_begin_0, end = var_10255_end_0, end_mask = var_10255_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10255_cast")];
            tensor<int32, [4]> var_10259_begin_0 = const()[name = tensor<string, []>("op_10259_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_10259_end_0 = const()[name = tensor<string, []>("op_10259_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_10259_end_mask_0 = const()[name = tensor<string, []>("op_10259_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10259_cast = slice_by_index(begin = var_10259_begin_0, end = var_10259_end_0, end_mask = var_10259_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10259_cast")];
            tensor<int32, [4]> var_10263_begin_0 = const()[name = tensor<string, []>("op_10263_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_10263_end_0 = const()[name = tensor<string, []>("op_10263_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_10263_end_mask_0 = const()[name = tensor<string, []>("op_10263_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10263_cast = slice_by_index(begin = var_10263_begin_0, end = var_10263_end_0, end_mask = var_10263_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10263_cast")];
            tensor<int32, [4]> var_10267_begin_0 = const()[name = tensor<string, []>("op_10267_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_10267_end_0 = const()[name = tensor<string, []>("op_10267_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_10267_end_mask_0 = const()[name = tensor<string, []>("op_10267_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10267_cast = slice_by_index(begin = var_10267_begin_0, end = var_10267_end_0, end_mask = var_10267_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10267_cast")];
            tensor<int32, [4]> var_10271_begin_0 = const()[name = tensor<string, []>("op_10271_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_10271_end_0 = const()[name = tensor<string, []>("op_10271_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_10271_end_mask_0 = const()[name = tensor<string, []>("op_10271_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10271_cast = slice_by_index(begin = var_10271_begin_0, end = var_10271_end_0, end_mask = var_10271_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10271_cast")];
            tensor<int32, [4]> var_10275_begin_0 = const()[name = tensor<string, []>("op_10275_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_10275_end_0 = const()[name = tensor<string, []>("op_10275_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_10275_end_mask_0 = const()[name = tensor<string, []>("op_10275_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10275_cast = slice_by_index(begin = var_10275_begin_0, end = var_10275_end_0, end_mask = var_10275_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10275_cast")];
            tensor<int32, [4]> var_10279_begin_0 = const()[name = tensor<string, []>("op_10279_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_10279_end_0 = const()[name = tensor<string, []>("op_10279_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_10279_end_mask_0 = const()[name = tensor<string, []>("op_10279_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10279_cast = slice_by_index(begin = var_10279_begin_0, end = var_10279_end_0, end_mask = var_10279_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10279_cast")];
            tensor<int32, [4]> var_10283_begin_0 = const()[name = tensor<string, []>("op_10283_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_10283_end_0 = const()[name = tensor<string, []>("op_10283_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_10283_end_mask_0 = const()[name = tensor<string, []>("op_10283_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10283_cast = slice_by_index(begin = var_10283_begin_0, end = var_10283_end_0, end_mask = var_10283_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10283_cast")];
            tensor<int32, [4]> var_10287_begin_0 = const()[name = tensor<string, []>("op_10287_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_10287_end_0 = const()[name = tensor<string, []>("op_10287_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_10287_end_mask_0 = const()[name = tensor<string, []>("op_10287_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10287_cast = slice_by_index(begin = var_10287_begin_0, end = var_10287_end_0, end_mask = var_10287_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10287_cast")];
            tensor<int32, [4]> var_10291_begin_0 = const()[name = tensor<string, []>("op_10291_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_10291_end_0 = const()[name = tensor<string, []>("op_10291_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_10291_end_mask_0 = const()[name = tensor<string, []>("op_10291_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10291_cast = slice_by_index(begin = var_10291_begin_0, end = var_10291_end_0, end_mask = var_10291_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10291_cast")];
            tensor<int32, [4]> var_10295_begin_0 = const()[name = tensor<string, []>("op_10295_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_10295_end_0 = const()[name = tensor<string, []>("op_10295_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_10295_end_mask_0 = const()[name = tensor<string, []>("op_10295_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10295_cast = slice_by_index(begin = var_10295_begin_0, end = var_10295_end_0, end_mask = var_10295_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10295_cast")];
            tensor<int32, [4]> var_10299_begin_0 = const()[name = tensor<string, []>("op_10299_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_10299_end_0 = const()[name = tensor<string, []>("op_10299_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_10299_end_mask_0 = const()[name = tensor<string, []>("op_10299_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10299_cast = slice_by_index(begin = var_10299_begin_0, end = var_10299_end_0, end_mask = var_10299_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10299_cast")];
            tensor<int32, [4]> var_10303_begin_0 = const()[name = tensor<string, []>("op_10303_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_10303_end_0 = const()[name = tensor<string, []>("op_10303_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_10303_end_mask_0 = const()[name = tensor<string, []>("op_10303_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10303_cast = slice_by_index(begin = var_10303_begin_0, end = var_10303_end_0, end_mask = var_10303_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_10303_cast")];
            tensor<string, []> var_10307_equation_0 = const()[name = tensor<string, []>("op_10307_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10307_cast = einsum(equation = var_10307_equation_0, values = (var_10149_cast, var_10066_cast))[name = tensor<string, []>("op_10307_cast")];
            tensor<fp16, []> var_10308_to_fp16 = const()[name = tensor<string, []>("op_10308_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_801_cast = mul(x = var_10307_cast, y = var_10308_to_fp16)[name = tensor<string, []>("aw_801_cast")];
            tensor<string, []> var_10311_equation_0 = const()[name = tensor<string, []>("op_10311_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10311_cast = einsum(equation = var_10311_equation_0, values = (var_10153_cast, var_10070_cast))[name = tensor<string, []>("op_10311_cast")];
            tensor<fp16, []> var_10312_to_fp16 = const()[name = tensor<string, []>("op_10312_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_803_cast = mul(x = var_10311_cast, y = var_10312_to_fp16)[name = tensor<string, []>("aw_803_cast")];
            tensor<string, []> var_10315_equation_0 = const()[name = tensor<string, []>("op_10315_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10315_cast = einsum(equation = var_10315_equation_0, values = (var_10157_cast, var_10074_cast))[name = tensor<string, []>("op_10315_cast")];
            tensor<fp16, []> var_10316_to_fp16 = const()[name = tensor<string, []>("op_10316_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_805_cast = mul(x = var_10315_cast, y = var_10316_to_fp16)[name = tensor<string, []>("aw_805_cast")];
            tensor<string, []> var_10319_equation_0 = const()[name = tensor<string, []>("op_10319_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10319_cast = einsum(equation = var_10319_equation_0, values = (var_10161_cast, var_10078_cast))[name = tensor<string, []>("op_10319_cast")];
            tensor<fp16, []> var_10320_to_fp16 = const()[name = tensor<string, []>("op_10320_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_807_cast = mul(x = var_10319_cast, y = var_10320_to_fp16)[name = tensor<string, []>("aw_807_cast")];
            tensor<string, []> var_10323_equation_0 = const()[name = tensor<string, []>("op_10323_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10323_cast = einsum(equation = var_10323_equation_0, values = (var_10165_cast, var_10082_cast))[name = tensor<string, []>("op_10323_cast")];
            tensor<fp16, []> var_10324_to_fp16 = const()[name = tensor<string, []>("op_10324_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_809_cast = mul(x = var_10323_cast, y = var_10324_to_fp16)[name = tensor<string, []>("aw_809_cast")];
            tensor<string, []> var_10327_equation_0 = const()[name = tensor<string, []>("op_10327_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10327_cast = einsum(equation = var_10327_equation_0, values = (var_10169_cast, var_10086_cast))[name = tensor<string, []>("op_10327_cast")];
            tensor<fp16, []> var_10328_to_fp16 = const()[name = tensor<string, []>("op_10328_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_811_cast = mul(x = var_10327_cast, y = var_10328_to_fp16)[name = tensor<string, []>("aw_811_cast")];
            tensor<string, []> var_10331_equation_0 = const()[name = tensor<string, []>("op_10331_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10331_cast = einsum(equation = var_10331_equation_0, values = (var_10173_cast, var_10090_cast))[name = tensor<string, []>("op_10331_cast")];
            tensor<fp16, []> var_10332_to_fp16 = const()[name = tensor<string, []>("op_10332_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_813_cast = mul(x = var_10331_cast, y = var_10332_to_fp16)[name = tensor<string, []>("aw_813_cast")];
            tensor<string, []> var_10335_equation_0 = const()[name = tensor<string, []>("op_10335_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10335_cast = einsum(equation = var_10335_equation_0, values = (var_10177_cast, var_10094_cast))[name = tensor<string, []>("op_10335_cast")];
            tensor<fp16, []> var_10336_to_fp16 = const()[name = tensor<string, []>("op_10336_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_815_cast = mul(x = var_10335_cast, y = var_10336_to_fp16)[name = tensor<string, []>("aw_815_cast")];
            tensor<string, []> var_10339_equation_0 = const()[name = tensor<string, []>("op_10339_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10339_cast = einsum(equation = var_10339_equation_0, values = (var_10181_cast, var_10098_cast))[name = tensor<string, []>("op_10339_cast")];
            tensor<fp16, []> var_10340_to_fp16 = const()[name = tensor<string, []>("op_10340_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_817_cast = mul(x = var_10339_cast, y = var_10340_to_fp16)[name = tensor<string, []>("aw_817_cast")];
            tensor<string, []> var_10343_equation_0 = const()[name = tensor<string, []>("op_10343_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10343_cast = einsum(equation = var_10343_equation_0, values = (var_10185_cast, var_10102_cast))[name = tensor<string, []>("op_10343_cast")];
            tensor<fp16, []> var_10344_to_fp16 = const()[name = tensor<string, []>("op_10344_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_819_cast = mul(x = var_10343_cast, y = var_10344_to_fp16)[name = tensor<string, []>("aw_819_cast")];
            tensor<string, []> var_10347_equation_0 = const()[name = tensor<string, []>("op_10347_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10347_cast = einsum(equation = var_10347_equation_0, values = (var_10189_cast, var_10106_cast))[name = tensor<string, []>("op_10347_cast")];
            tensor<fp16, []> var_10348_to_fp16 = const()[name = tensor<string, []>("op_10348_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_821_cast = mul(x = var_10347_cast, y = var_10348_to_fp16)[name = tensor<string, []>("aw_821_cast")];
            tensor<string, []> var_10351_equation_0 = const()[name = tensor<string, []>("op_10351_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10351_cast = einsum(equation = var_10351_equation_0, values = (var_10193_cast, var_10110_cast))[name = tensor<string, []>("op_10351_cast")];
            tensor<fp16, []> var_10352_to_fp16 = const()[name = tensor<string, []>("op_10352_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_823_cast = mul(x = var_10351_cast, y = var_10352_to_fp16)[name = tensor<string, []>("aw_823_cast")];
            tensor<string, []> var_10355_equation_0 = const()[name = tensor<string, []>("op_10355_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10355_cast = einsum(equation = var_10355_equation_0, values = (var_10197_cast, var_10114_cast))[name = tensor<string, []>("op_10355_cast")];
            tensor<fp16, []> var_10356_to_fp16 = const()[name = tensor<string, []>("op_10356_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_825_cast = mul(x = var_10355_cast, y = var_10356_to_fp16)[name = tensor<string, []>("aw_825_cast")];
            tensor<string, []> var_10359_equation_0 = const()[name = tensor<string, []>("op_10359_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10359_cast = einsum(equation = var_10359_equation_0, values = (var_10201_cast, var_10118_cast))[name = tensor<string, []>("op_10359_cast")];
            tensor<fp16, []> var_10360_to_fp16 = const()[name = tensor<string, []>("op_10360_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_827_cast = mul(x = var_10359_cast, y = var_10360_to_fp16)[name = tensor<string, []>("aw_827_cast")];
            tensor<string, []> var_10363_equation_0 = const()[name = tensor<string, []>("op_10363_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10363_cast = einsum(equation = var_10363_equation_0, values = (var_10205_cast, var_10122_cast))[name = tensor<string, []>("op_10363_cast")];
            tensor<fp16, []> var_10364_to_fp16 = const()[name = tensor<string, []>("op_10364_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_829_cast = mul(x = var_10363_cast, y = var_10364_to_fp16)[name = tensor<string, []>("aw_829_cast")];
            tensor<string, []> var_10367_equation_0 = const()[name = tensor<string, []>("op_10367_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10367_cast = einsum(equation = var_10367_equation_0, values = (var_10209_cast, var_10126_cast))[name = tensor<string, []>("op_10367_cast")];
            tensor<fp16, []> var_10368_to_fp16 = const()[name = tensor<string, []>("op_10368_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_831_cast = mul(x = var_10367_cast, y = var_10368_to_fp16)[name = tensor<string, []>("aw_831_cast")];
            tensor<string, []> var_10371_equation_0 = const()[name = tensor<string, []>("op_10371_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10371_cast = einsum(equation = var_10371_equation_0, values = (var_10213_cast, var_10130_cast))[name = tensor<string, []>("op_10371_cast")];
            tensor<fp16, []> var_10372_to_fp16 = const()[name = tensor<string, []>("op_10372_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_833_cast = mul(x = var_10371_cast, y = var_10372_to_fp16)[name = tensor<string, []>("aw_833_cast")];
            tensor<string, []> var_10375_equation_0 = const()[name = tensor<string, []>("op_10375_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10375_cast = einsum(equation = var_10375_equation_0, values = (var_10217_cast, var_10134_cast))[name = tensor<string, []>("op_10375_cast")];
            tensor<fp16, []> var_10376_to_fp16 = const()[name = tensor<string, []>("op_10376_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_835_cast = mul(x = var_10375_cast, y = var_10376_to_fp16)[name = tensor<string, []>("aw_835_cast")];
            tensor<string, []> var_10379_equation_0 = const()[name = tensor<string, []>("op_10379_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10379_cast = einsum(equation = var_10379_equation_0, values = (var_10221_cast, var_10138_cast))[name = tensor<string, []>("op_10379_cast")];
            tensor<fp16, []> var_10380_to_fp16 = const()[name = tensor<string, []>("op_10380_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_837_cast = mul(x = var_10379_cast, y = var_10380_to_fp16)[name = tensor<string, []>("aw_837_cast")];
            tensor<string, []> var_10383_equation_0 = const()[name = tensor<string, []>("op_10383_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10383_cast = einsum(equation = var_10383_equation_0, values = (var_10225_cast, var_10142_cast))[name = tensor<string, []>("op_10383_cast")];
            tensor<fp16, []> var_10384_to_fp16 = const()[name = tensor<string, []>("op_10384_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_839_cast = mul(x = var_10383_cast, y = var_10384_to_fp16)[name = tensor<string, []>("aw_839_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10386_cast = softmax(axis = var_2634, x = aw_801_cast)[name = tensor<string, []>("op_10386_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10387_cast = softmax(axis = var_2634, x = aw_803_cast)[name = tensor<string, []>("op_10387_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10388_cast = softmax(axis = var_2634, x = aw_805_cast)[name = tensor<string, []>("op_10388_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10389_cast = softmax(axis = var_2634, x = aw_807_cast)[name = tensor<string, []>("op_10389_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10390_cast = softmax(axis = var_2634, x = aw_809_cast)[name = tensor<string, []>("op_10390_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10391_cast = softmax(axis = var_2634, x = aw_811_cast)[name = tensor<string, []>("op_10391_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10392_cast = softmax(axis = var_2634, x = aw_813_cast)[name = tensor<string, []>("op_10392_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10393_cast = softmax(axis = var_2634, x = aw_815_cast)[name = tensor<string, []>("op_10393_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10394_cast = softmax(axis = var_2634, x = aw_817_cast)[name = tensor<string, []>("op_10394_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10395_cast = softmax(axis = var_2634, x = aw_819_cast)[name = tensor<string, []>("op_10395_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10396_cast = softmax(axis = var_2634, x = aw_821_cast)[name = tensor<string, []>("op_10396_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10397_cast = softmax(axis = var_2634, x = aw_823_cast)[name = tensor<string, []>("op_10397_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10398_cast = softmax(axis = var_2634, x = aw_825_cast)[name = tensor<string, []>("op_10398_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10399_cast = softmax(axis = var_2634, x = aw_827_cast)[name = tensor<string, []>("op_10399_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10400_cast = softmax(axis = var_2634, x = aw_829_cast)[name = tensor<string, []>("op_10400_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10401_cast = softmax(axis = var_2634, x = aw_831_cast)[name = tensor<string, []>("op_10401_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10402_cast = softmax(axis = var_2634, x = aw_833_cast)[name = tensor<string, []>("op_10402_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10403_cast = softmax(axis = var_2634, x = aw_835_cast)[name = tensor<string, []>("op_10403_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10404_cast = softmax(axis = var_2634, x = aw_837_cast)[name = tensor<string, []>("op_10404_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10405_cast = softmax(axis = var_2634, x = aw_839_cast)[name = tensor<string, []>("op_10405_cast")];
            tensor<string, []> var_10407_equation_0 = const()[name = tensor<string, []>("op_10407_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10407_cast = einsum(equation = var_10407_equation_0, values = (var_10227_cast, var_10386_cast))[name = tensor<string, []>("op_10407_cast")];
            tensor<string, []> var_10409_equation_0 = const()[name = tensor<string, []>("op_10409_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10409_cast = einsum(equation = var_10409_equation_0, values = (var_10231_cast, var_10387_cast))[name = tensor<string, []>("op_10409_cast")];
            tensor<string, []> var_10411_equation_0 = const()[name = tensor<string, []>("op_10411_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10411_cast = einsum(equation = var_10411_equation_0, values = (var_10235_cast, var_10388_cast))[name = tensor<string, []>("op_10411_cast")];
            tensor<string, []> var_10413_equation_0 = const()[name = tensor<string, []>("op_10413_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10413_cast = einsum(equation = var_10413_equation_0, values = (var_10239_cast, var_10389_cast))[name = tensor<string, []>("op_10413_cast")];
            tensor<string, []> var_10415_equation_0 = const()[name = tensor<string, []>("op_10415_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10415_cast = einsum(equation = var_10415_equation_0, values = (var_10243_cast, var_10390_cast))[name = tensor<string, []>("op_10415_cast")];
            tensor<string, []> var_10417_equation_0 = const()[name = tensor<string, []>("op_10417_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10417_cast = einsum(equation = var_10417_equation_0, values = (var_10247_cast, var_10391_cast))[name = tensor<string, []>("op_10417_cast")];
            tensor<string, []> var_10419_equation_0 = const()[name = tensor<string, []>("op_10419_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10419_cast = einsum(equation = var_10419_equation_0, values = (var_10251_cast, var_10392_cast))[name = tensor<string, []>("op_10419_cast")];
            tensor<string, []> var_10421_equation_0 = const()[name = tensor<string, []>("op_10421_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10421_cast = einsum(equation = var_10421_equation_0, values = (var_10255_cast, var_10393_cast))[name = tensor<string, []>("op_10421_cast")];
            tensor<string, []> var_10423_equation_0 = const()[name = tensor<string, []>("op_10423_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10423_cast = einsum(equation = var_10423_equation_0, values = (var_10259_cast, var_10394_cast))[name = tensor<string, []>("op_10423_cast")];
            tensor<string, []> var_10425_equation_0 = const()[name = tensor<string, []>("op_10425_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10425_cast = einsum(equation = var_10425_equation_0, values = (var_10263_cast, var_10395_cast))[name = tensor<string, []>("op_10425_cast")];
            tensor<string, []> var_10427_equation_0 = const()[name = tensor<string, []>("op_10427_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10427_cast = einsum(equation = var_10427_equation_0, values = (var_10267_cast, var_10396_cast))[name = tensor<string, []>("op_10427_cast")];
            tensor<string, []> var_10429_equation_0 = const()[name = tensor<string, []>("op_10429_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10429_cast = einsum(equation = var_10429_equation_0, values = (var_10271_cast, var_10397_cast))[name = tensor<string, []>("op_10429_cast")];
            tensor<string, []> var_10431_equation_0 = const()[name = tensor<string, []>("op_10431_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10431_cast = einsum(equation = var_10431_equation_0, values = (var_10275_cast, var_10398_cast))[name = tensor<string, []>("op_10431_cast")];
            tensor<string, []> var_10433_equation_0 = const()[name = tensor<string, []>("op_10433_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10433_cast = einsum(equation = var_10433_equation_0, values = (var_10279_cast, var_10399_cast))[name = tensor<string, []>("op_10433_cast")];
            tensor<string, []> var_10435_equation_0 = const()[name = tensor<string, []>("op_10435_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10435_cast = einsum(equation = var_10435_equation_0, values = (var_10283_cast, var_10400_cast))[name = tensor<string, []>("op_10435_cast")];
            tensor<string, []> var_10437_equation_0 = const()[name = tensor<string, []>("op_10437_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10437_cast = einsum(equation = var_10437_equation_0, values = (var_10287_cast, var_10401_cast))[name = tensor<string, []>("op_10437_cast")];
            tensor<string, []> var_10439_equation_0 = const()[name = tensor<string, []>("op_10439_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10439_cast = einsum(equation = var_10439_equation_0, values = (var_10291_cast, var_10402_cast))[name = tensor<string, []>("op_10439_cast")];
            tensor<string, []> var_10441_equation_0 = const()[name = tensor<string, []>("op_10441_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10441_cast = einsum(equation = var_10441_equation_0, values = (var_10295_cast, var_10403_cast))[name = tensor<string, []>("op_10441_cast")];
            tensor<string, []> var_10443_equation_0 = const()[name = tensor<string, []>("op_10443_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10443_cast = einsum(equation = var_10443_equation_0, values = (var_10299_cast, var_10404_cast))[name = tensor<string, []>("op_10443_cast")];
            tensor<string, []> var_10445_equation_0 = const()[name = tensor<string, []>("op_10445_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10445_cast = einsum(equation = var_10445_equation_0, values = (var_10303_cast, var_10405_cast))[name = tensor<string, []>("op_10445_cast")];
            tensor<bool, []> input_199_interleave_0 = const()[name = tensor<string, []>("input_199_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_199_cast = concat(axis = var_2634, interleave = input_199_interleave_0, values = (var_10407_cast, var_10409_cast, var_10411_cast, var_10413_cast, var_10415_cast, var_10417_cast, var_10419_cast, var_10421_cast, var_10423_cast, var_10425_cast, var_10427_cast, var_10429_cast, var_10431_cast, var_10433_cast, var_10435_cast, var_10437_cast, var_10439_cast, var_10441_cast, var_10443_cast, var_10445_cast))[name = tensor<string, []>("input_199_cast")];
            tensor<int32, [2]> var_10451 = const()[name = tensor<string, []>("op_10451"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10453 = const()[name = tensor<string, []>("op_10453"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10455_pad_type_0 = const()[name = tensor<string, []>("op_10455_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10455_pad_0 = const()[name = tensor<string, []>("op_10455_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(764105408)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(767382272)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_10455_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16, dilations = var_10453, groups = var_2634, pad = var_10455_pad_0, pad_type = var_10455_pad_type_0, strides = var_10451, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16, x = input_199_cast)[name = tensor<string, []>("op_10455_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_75_cast = add(x = var_10455_cast, y = inputs_73_cast)[name = tensor<string, []>("inputs_75_cast")];
            tensor<int32, [1]> var_10459 = const()[name = tensor<string, []>("op_10459"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_75_cast = reduce_mean(axes = var_10459, keep_dims = var_2629, x = inputs_75_cast)[name = tensor<string, []>("channels_mean_75_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_75_cast = sub(x = inputs_75_cast, y = channels_mean_75_cast)[name = tensor<string, []>("zero_mean_75_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_75_cast = mul(x = zero_mean_75_cast, y = zero_mean_75_cast)[name = tensor<string, []>("zero_mean_sq_75_cast")];
            tensor<int32, [1]> var_10463 = const()[name = tensor<string, []>("op_10463"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_10464_cast = reduce_mean(axes = var_10463, keep_dims = var_2629, x = zero_mean_sq_75_cast)[name = tensor<string, []>("op_10464_cast")];
            tensor<fp16, []> var_10465_to_fp16 = const()[name = tensor<string, []>("op_10465_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_10466_cast = add(x = var_10464_cast, y = var_10465_to_fp16)[name = tensor<string, []>("op_10466_cast")];
            tensor<fp16, []> denom_75_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_75_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_75_cast = rsqrt(epsilon = denom_75_epsilon_0_to_fp16, x = var_10466_cast)[name = tensor<string, []>("denom_75_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_75_cast = mul(x = zero_mean_75_cast, y = denom_75_cast)[name = tensor<string, []>("out_75_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_10470_to_fp16 = const()[name = tensor<string, []>("op_10470_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(767384896)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_10471_cast = add(x = out_75_cast, y = var_10470_to_fp16)[name = tensor<string, []>("op_10471_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_10473_to_fp16 = const()[name = tensor<string, []>("op_10473_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(767387520)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_115_cast = mul(x = var_10471_cast, y = var_10473_to_fp16)[name = tensor<string, []>("hidden_states_115_cast")];
            tensor<int32, [2]> var_10480 = const()[name = tensor<string, []>("op_10480"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10482 = const()[name = tensor<string, []>("op_10482"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_51_pad_type_0 = const()[name = tensor<string, []>("q_51_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_51_pad_0 = const()[name = tensor<string, []>("q_51_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(767390144)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_51_cast = conv(dilations = var_10482, groups = var_2634, pad = q_51_pad_0, pad_type = q_51_pad_type_0, strides = var_10480, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16, x = hidden_states_115_cast)[name = tensor<string, []>("q_51_cast")];
            tensor<int32, [2]> var_10486 = const()[name = tensor<string, []>("op_10486"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10488 = const()[name = tensor<string, []>("op_10488"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_101_pad_type_0 = const()[name = tensor<string, []>("k_101_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_101_pad_0 = const()[name = tensor<string, []>("k_101_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(770667008)))];
            tensor<fp16, [2, 1280, 1, 77]> k_101_cast = conv(dilations = var_10488, groups = var_2634, pad = k_101_pad_0, pad_type = k_101_pad_type_0, strides = var_10486, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_101_cast")];
            tensor<int32, [2]> var_10492 = const()[name = tensor<string, []>("op_10492"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10494 = const()[name = tensor<string, []>("op_10494"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_51_pad_type_0 = const()[name = tensor<string, []>("v_51_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_51_pad_0 = const()[name = tensor<string, []>("v_51_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(775909952)))];
            tensor<fp16, [2, 1280, 1, 77]> v_51_cast = conv(dilations = var_10494, groups = var_2634, pad = v_51_pad_0, pad_type = v_51_pad_type_0, strides = var_10492, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_51_cast")];
            tensor<int32, [4]> var_10498_begin_0 = const()[name = tensor<string, []>("op_10498_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10498_end_0 = const()[name = tensor<string, []>("op_10498_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10498_end_mask_0 = const()[name = tensor<string, []>("op_10498_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10498_cast = slice_by_index(begin = var_10498_begin_0, end = var_10498_end_0, end_mask = var_10498_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10498_cast")];
            tensor<int32, [4]> var_10502_begin_0 = const()[name = tensor<string, []>("op_10502_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_10502_end_0 = const()[name = tensor<string, []>("op_10502_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_10502_end_mask_0 = const()[name = tensor<string, []>("op_10502_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10502_cast = slice_by_index(begin = var_10502_begin_0, end = var_10502_end_0, end_mask = var_10502_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10502_cast")];
            tensor<int32, [4]> var_10506_begin_0 = const()[name = tensor<string, []>("op_10506_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_10506_end_0 = const()[name = tensor<string, []>("op_10506_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_10506_end_mask_0 = const()[name = tensor<string, []>("op_10506_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10506_cast = slice_by_index(begin = var_10506_begin_0, end = var_10506_end_0, end_mask = var_10506_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10506_cast")];
            tensor<int32, [4]> var_10510_begin_0 = const()[name = tensor<string, []>("op_10510_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_10510_end_0 = const()[name = tensor<string, []>("op_10510_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_10510_end_mask_0 = const()[name = tensor<string, []>("op_10510_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10510_cast = slice_by_index(begin = var_10510_begin_0, end = var_10510_end_0, end_mask = var_10510_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10510_cast")];
            tensor<int32, [4]> var_10514_begin_0 = const()[name = tensor<string, []>("op_10514_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_10514_end_0 = const()[name = tensor<string, []>("op_10514_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_10514_end_mask_0 = const()[name = tensor<string, []>("op_10514_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10514_cast = slice_by_index(begin = var_10514_begin_0, end = var_10514_end_0, end_mask = var_10514_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10514_cast")];
            tensor<int32, [4]> var_10518_begin_0 = const()[name = tensor<string, []>("op_10518_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_10518_end_0 = const()[name = tensor<string, []>("op_10518_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_10518_end_mask_0 = const()[name = tensor<string, []>("op_10518_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10518_cast = slice_by_index(begin = var_10518_begin_0, end = var_10518_end_0, end_mask = var_10518_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10518_cast")];
            tensor<int32, [4]> var_10522_begin_0 = const()[name = tensor<string, []>("op_10522_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_10522_end_0 = const()[name = tensor<string, []>("op_10522_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_10522_end_mask_0 = const()[name = tensor<string, []>("op_10522_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10522_cast = slice_by_index(begin = var_10522_begin_0, end = var_10522_end_0, end_mask = var_10522_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10522_cast")];
            tensor<int32, [4]> var_10526_begin_0 = const()[name = tensor<string, []>("op_10526_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_10526_end_0 = const()[name = tensor<string, []>("op_10526_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_10526_end_mask_0 = const()[name = tensor<string, []>("op_10526_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10526_cast = slice_by_index(begin = var_10526_begin_0, end = var_10526_end_0, end_mask = var_10526_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10526_cast")];
            tensor<int32, [4]> var_10530_begin_0 = const()[name = tensor<string, []>("op_10530_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_10530_end_0 = const()[name = tensor<string, []>("op_10530_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_10530_end_mask_0 = const()[name = tensor<string, []>("op_10530_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10530_cast = slice_by_index(begin = var_10530_begin_0, end = var_10530_end_0, end_mask = var_10530_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10530_cast")];
            tensor<int32, [4]> var_10534_begin_0 = const()[name = tensor<string, []>("op_10534_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_10534_end_0 = const()[name = tensor<string, []>("op_10534_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_10534_end_mask_0 = const()[name = tensor<string, []>("op_10534_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10534_cast = slice_by_index(begin = var_10534_begin_0, end = var_10534_end_0, end_mask = var_10534_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10534_cast")];
            tensor<int32, [4]> var_10538_begin_0 = const()[name = tensor<string, []>("op_10538_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_10538_end_0 = const()[name = tensor<string, []>("op_10538_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_10538_end_mask_0 = const()[name = tensor<string, []>("op_10538_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10538_cast = slice_by_index(begin = var_10538_begin_0, end = var_10538_end_0, end_mask = var_10538_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10538_cast")];
            tensor<int32, [4]> var_10542_begin_0 = const()[name = tensor<string, []>("op_10542_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_10542_end_0 = const()[name = tensor<string, []>("op_10542_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_10542_end_mask_0 = const()[name = tensor<string, []>("op_10542_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10542_cast = slice_by_index(begin = var_10542_begin_0, end = var_10542_end_0, end_mask = var_10542_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10542_cast")];
            tensor<int32, [4]> var_10546_begin_0 = const()[name = tensor<string, []>("op_10546_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_10546_end_0 = const()[name = tensor<string, []>("op_10546_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_10546_end_mask_0 = const()[name = tensor<string, []>("op_10546_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10546_cast = slice_by_index(begin = var_10546_begin_0, end = var_10546_end_0, end_mask = var_10546_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10546_cast")];
            tensor<int32, [4]> var_10550_begin_0 = const()[name = tensor<string, []>("op_10550_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_10550_end_0 = const()[name = tensor<string, []>("op_10550_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_10550_end_mask_0 = const()[name = tensor<string, []>("op_10550_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10550_cast = slice_by_index(begin = var_10550_begin_0, end = var_10550_end_0, end_mask = var_10550_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10550_cast")];
            tensor<int32, [4]> var_10554_begin_0 = const()[name = tensor<string, []>("op_10554_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_10554_end_0 = const()[name = tensor<string, []>("op_10554_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_10554_end_mask_0 = const()[name = tensor<string, []>("op_10554_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10554_cast = slice_by_index(begin = var_10554_begin_0, end = var_10554_end_0, end_mask = var_10554_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10554_cast")];
            tensor<int32, [4]> var_10558_begin_0 = const()[name = tensor<string, []>("op_10558_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_10558_end_0 = const()[name = tensor<string, []>("op_10558_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_10558_end_mask_0 = const()[name = tensor<string, []>("op_10558_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10558_cast = slice_by_index(begin = var_10558_begin_0, end = var_10558_end_0, end_mask = var_10558_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10558_cast")];
            tensor<int32, [4]> var_10562_begin_0 = const()[name = tensor<string, []>("op_10562_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_10562_end_0 = const()[name = tensor<string, []>("op_10562_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_10562_end_mask_0 = const()[name = tensor<string, []>("op_10562_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10562_cast = slice_by_index(begin = var_10562_begin_0, end = var_10562_end_0, end_mask = var_10562_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10562_cast")];
            tensor<int32, [4]> var_10566_begin_0 = const()[name = tensor<string, []>("op_10566_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_10566_end_0 = const()[name = tensor<string, []>("op_10566_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_10566_end_mask_0 = const()[name = tensor<string, []>("op_10566_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10566_cast = slice_by_index(begin = var_10566_begin_0, end = var_10566_end_0, end_mask = var_10566_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10566_cast")];
            tensor<int32, [4]> var_10570_begin_0 = const()[name = tensor<string, []>("op_10570_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_10570_end_0 = const()[name = tensor<string, []>("op_10570_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_10570_end_mask_0 = const()[name = tensor<string, []>("op_10570_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10570_cast = slice_by_index(begin = var_10570_begin_0, end = var_10570_end_0, end_mask = var_10570_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10570_cast")];
            tensor<int32, [4]> var_10574_begin_0 = const()[name = tensor<string, []>("op_10574_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_10574_end_0 = const()[name = tensor<string, []>("op_10574_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_10574_end_mask_0 = const()[name = tensor<string, []>("op_10574_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10574_cast = slice_by_index(begin = var_10574_begin_0, end = var_10574_end_0, end_mask = var_10574_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10574_cast")];
            tensor<int32, [4]> k_103_perm_0 = const()[name = tensor<string, []>("k_103_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_10581_begin_0 = const()[name = tensor<string, []>("op_10581_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10581_end_0 = const()[name = tensor<string, []>("op_10581_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_10581_end_mask_0 = const()[name = tensor<string, []>("op_10581_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_114 = transpose(perm = k_103_perm_0, x = k_101_cast)[name = tensor<string, []>("transpose_114")];
            tensor<fp16, [2, 77, 1, 64]> var_10581_cast = slice_by_index(begin = var_10581_begin_0, end = var_10581_end_0, end_mask = var_10581_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10581_cast")];
            tensor<int32, [4]> var_10585_begin_0 = const()[name = tensor<string, []>("op_10585_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_10585_end_0 = const()[name = tensor<string, []>("op_10585_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_10585_end_mask_0 = const()[name = tensor<string, []>("op_10585_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10585_cast = slice_by_index(begin = var_10585_begin_0, end = var_10585_end_0, end_mask = var_10585_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10585_cast")];
            tensor<int32, [4]> var_10589_begin_0 = const()[name = tensor<string, []>("op_10589_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_10589_end_0 = const()[name = tensor<string, []>("op_10589_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_10589_end_mask_0 = const()[name = tensor<string, []>("op_10589_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10589_cast = slice_by_index(begin = var_10589_begin_0, end = var_10589_end_0, end_mask = var_10589_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10589_cast")];
            tensor<int32, [4]> var_10593_begin_0 = const()[name = tensor<string, []>("op_10593_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_10593_end_0 = const()[name = tensor<string, []>("op_10593_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_10593_end_mask_0 = const()[name = tensor<string, []>("op_10593_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10593_cast = slice_by_index(begin = var_10593_begin_0, end = var_10593_end_0, end_mask = var_10593_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10593_cast")];
            tensor<int32, [4]> var_10597_begin_0 = const()[name = tensor<string, []>("op_10597_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_10597_end_0 = const()[name = tensor<string, []>("op_10597_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_10597_end_mask_0 = const()[name = tensor<string, []>("op_10597_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10597_cast = slice_by_index(begin = var_10597_begin_0, end = var_10597_end_0, end_mask = var_10597_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10597_cast")];
            tensor<int32, [4]> var_10601_begin_0 = const()[name = tensor<string, []>("op_10601_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_10601_end_0 = const()[name = tensor<string, []>("op_10601_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_10601_end_mask_0 = const()[name = tensor<string, []>("op_10601_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10601_cast = slice_by_index(begin = var_10601_begin_0, end = var_10601_end_0, end_mask = var_10601_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10601_cast")];
            tensor<int32, [4]> var_10605_begin_0 = const()[name = tensor<string, []>("op_10605_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_10605_end_0 = const()[name = tensor<string, []>("op_10605_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_10605_end_mask_0 = const()[name = tensor<string, []>("op_10605_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10605_cast = slice_by_index(begin = var_10605_begin_0, end = var_10605_end_0, end_mask = var_10605_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10605_cast")];
            tensor<int32, [4]> var_10609_begin_0 = const()[name = tensor<string, []>("op_10609_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_10609_end_0 = const()[name = tensor<string, []>("op_10609_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_10609_end_mask_0 = const()[name = tensor<string, []>("op_10609_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10609_cast = slice_by_index(begin = var_10609_begin_0, end = var_10609_end_0, end_mask = var_10609_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10609_cast")];
            tensor<int32, [4]> var_10613_begin_0 = const()[name = tensor<string, []>("op_10613_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10613_end_0 = const()[name = tensor<string, []>("op_10613_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_10613_end_mask_0 = const()[name = tensor<string, []>("op_10613_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10613_cast = slice_by_index(begin = var_10613_begin_0, end = var_10613_end_0, end_mask = var_10613_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10613_cast")];
            tensor<int32, [4]> var_10617_begin_0 = const()[name = tensor<string, []>("op_10617_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_10617_end_0 = const()[name = tensor<string, []>("op_10617_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_10617_end_mask_0 = const()[name = tensor<string, []>("op_10617_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10617_cast = slice_by_index(begin = var_10617_begin_0, end = var_10617_end_0, end_mask = var_10617_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10617_cast")];
            tensor<int32, [4]> var_10621_begin_0 = const()[name = tensor<string, []>("op_10621_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_10621_end_0 = const()[name = tensor<string, []>("op_10621_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_10621_end_mask_0 = const()[name = tensor<string, []>("op_10621_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10621_cast = slice_by_index(begin = var_10621_begin_0, end = var_10621_end_0, end_mask = var_10621_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10621_cast")];
            tensor<int32, [4]> var_10625_begin_0 = const()[name = tensor<string, []>("op_10625_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_10625_end_0 = const()[name = tensor<string, []>("op_10625_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_10625_end_mask_0 = const()[name = tensor<string, []>("op_10625_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10625_cast = slice_by_index(begin = var_10625_begin_0, end = var_10625_end_0, end_mask = var_10625_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10625_cast")];
            tensor<int32, [4]> var_10629_begin_0 = const()[name = tensor<string, []>("op_10629_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_10629_end_0 = const()[name = tensor<string, []>("op_10629_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_10629_end_mask_0 = const()[name = tensor<string, []>("op_10629_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10629_cast = slice_by_index(begin = var_10629_begin_0, end = var_10629_end_0, end_mask = var_10629_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10629_cast")];
            tensor<int32, [4]> var_10633_begin_0 = const()[name = tensor<string, []>("op_10633_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_10633_end_0 = const()[name = tensor<string, []>("op_10633_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_10633_end_mask_0 = const()[name = tensor<string, []>("op_10633_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10633_cast = slice_by_index(begin = var_10633_begin_0, end = var_10633_end_0, end_mask = var_10633_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10633_cast")];
            tensor<int32, [4]> var_10637_begin_0 = const()[name = tensor<string, []>("op_10637_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_10637_end_0 = const()[name = tensor<string, []>("op_10637_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_10637_end_mask_0 = const()[name = tensor<string, []>("op_10637_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10637_cast = slice_by_index(begin = var_10637_begin_0, end = var_10637_end_0, end_mask = var_10637_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10637_cast")];
            tensor<int32, [4]> var_10641_begin_0 = const()[name = tensor<string, []>("op_10641_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_10641_end_0 = const()[name = tensor<string, []>("op_10641_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_10641_end_mask_0 = const()[name = tensor<string, []>("op_10641_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10641_cast = slice_by_index(begin = var_10641_begin_0, end = var_10641_end_0, end_mask = var_10641_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10641_cast")];
            tensor<int32, [4]> var_10645_begin_0 = const()[name = tensor<string, []>("op_10645_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_10645_end_0 = const()[name = tensor<string, []>("op_10645_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_10645_end_mask_0 = const()[name = tensor<string, []>("op_10645_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10645_cast = slice_by_index(begin = var_10645_begin_0, end = var_10645_end_0, end_mask = var_10645_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10645_cast")];
            tensor<int32, [4]> var_10649_begin_0 = const()[name = tensor<string, []>("op_10649_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_10649_end_0 = const()[name = tensor<string, []>("op_10649_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_10649_end_mask_0 = const()[name = tensor<string, []>("op_10649_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10649_cast = slice_by_index(begin = var_10649_begin_0, end = var_10649_end_0, end_mask = var_10649_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10649_cast")];
            tensor<int32, [4]> var_10653_begin_0 = const()[name = tensor<string, []>("op_10653_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_10653_end_0 = const()[name = tensor<string, []>("op_10653_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_10653_end_mask_0 = const()[name = tensor<string, []>("op_10653_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10653_cast = slice_by_index(begin = var_10653_begin_0, end = var_10653_end_0, end_mask = var_10653_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10653_cast")];
            tensor<int32, [4]> var_10657_begin_0 = const()[name = tensor<string, []>("op_10657_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_10657_end_0 = const()[name = tensor<string, []>("op_10657_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_10657_end_mask_0 = const()[name = tensor<string, []>("op_10657_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10657_cast = slice_by_index(begin = var_10657_begin_0, end = var_10657_end_0, end_mask = var_10657_end_mask_0, x = transpose_114)[name = tensor<string, []>("op_10657_cast")];
            tensor<int32, [4]> var_10659_begin_0 = const()[name = tensor<string, []>("op_10659_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10659_end_0 = const()[name = tensor<string, []>("op_10659_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_10659_end_mask_0 = const()[name = tensor<string, []>("op_10659_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10659_cast = slice_by_index(begin = var_10659_begin_0, end = var_10659_end_0, end_mask = var_10659_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10659_cast")];
            tensor<int32, [4]> var_10663_begin_0 = const()[name = tensor<string, []>("op_10663_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_10663_end_0 = const()[name = tensor<string, []>("op_10663_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_10663_end_mask_0 = const()[name = tensor<string, []>("op_10663_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10663_cast = slice_by_index(begin = var_10663_begin_0, end = var_10663_end_0, end_mask = var_10663_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10663_cast")];
            tensor<int32, [4]> var_10667_begin_0 = const()[name = tensor<string, []>("op_10667_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_10667_end_0 = const()[name = tensor<string, []>("op_10667_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_10667_end_mask_0 = const()[name = tensor<string, []>("op_10667_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10667_cast = slice_by_index(begin = var_10667_begin_0, end = var_10667_end_0, end_mask = var_10667_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10667_cast")];
            tensor<int32, [4]> var_10671_begin_0 = const()[name = tensor<string, []>("op_10671_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_10671_end_0 = const()[name = tensor<string, []>("op_10671_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_10671_end_mask_0 = const()[name = tensor<string, []>("op_10671_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10671_cast = slice_by_index(begin = var_10671_begin_0, end = var_10671_end_0, end_mask = var_10671_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10671_cast")];
            tensor<int32, [4]> var_10675_begin_0 = const()[name = tensor<string, []>("op_10675_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_10675_end_0 = const()[name = tensor<string, []>("op_10675_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_10675_end_mask_0 = const()[name = tensor<string, []>("op_10675_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10675_cast = slice_by_index(begin = var_10675_begin_0, end = var_10675_end_0, end_mask = var_10675_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10675_cast")];
            tensor<int32, [4]> var_10679_begin_0 = const()[name = tensor<string, []>("op_10679_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_10679_end_0 = const()[name = tensor<string, []>("op_10679_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_10679_end_mask_0 = const()[name = tensor<string, []>("op_10679_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10679_cast = slice_by_index(begin = var_10679_begin_0, end = var_10679_end_0, end_mask = var_10679_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10679_cast")];
            tensor<int32, [4]> var_10683_begin_0 = const()[name = tensor<string, []>("op_10683_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_10683_end_0 = const()[name = tensor<string, []>("op_10683_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_10683_end_mask_0 = const()[name = tensor<string, []>("op_10683_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10683_cast = slice_by_index(begin = var_10683_begin_0, end = var_10683_end_0, end_mask = var_10683_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10683_cast")];
            tensor<int32, [4]> var_10687_begin_0 = const()[name = tensor<string, []>("op_10687_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_10687_end_0 = const()[name = tensor<string, []>("op_10687_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_10687_end_mask_0 = const()[name = tensor<string, []>("op_10687_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10687_cast = slice_by_index(begin = var_10687_begin_0, end = var_10687_end_0, end_mask = var_10687_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10687_cast")];
            tensor<int32, [4]> var_10691_begin_0 = const()[name = tensor<string, []>("op_10691_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_10691_end_0 = const()[name = tensor<string, []>("op_10691_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_10691_end_mask_0 = const()[name = tensor<string, []>("op_10691_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10691_cast = slice_by_index(begin = var_10691_begin_0, end = var_10691_end_0, end_mask = var_10691_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10691_cast")];
            tensor<int32, [4]> var_10695_begin_0 = const()[name = tensor<string, []>("op_10695_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_10695_end_0 = const()[name = tensor<string, []>("op_10695_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_10695_end_mask_0 = const()[name = tensor<string, []>("op_10695_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10695_cast = slice_by_index(begin = var_10695_begin_0, end = var_10695_end_0, end_mask = var_10695_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10695_cast")];
            tensor<int32, [4]> var_10699_begin_0 = const()[name = tensor<string, []>("op_10699_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_10699_end_0 = const()[name = tensor<string, []>("op_10699_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_10699_end_mask_0 = const()[name = tensor<string, []>("op_10699_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10699_cast = slice_by_index(begin = var_10699_begin_0, end = var_10699_end_0, end_mask = var_10699_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10699_cast")];
            tensor<int32, [4]> var_10703_begin_0 = const()[name = tensor<string, []>("op_10703_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_10703_end_0 = const()[name = tensor<string, []>("op_10703_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_10703_end_mask_0 = const()[name = tensor<string, []>("op_10703_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10703_cast = slice_by_index(begin = var_10703_begin_0, end = var_10703_end_0, end_mask = var_10703_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10703_cast")];
            tensor<int32, [4]> var_10707_begin_0 = const()[name = tensor<string, []>("op_10707_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_10707_end_0 = const()[name = tensor<string, []>("op_10707_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_10707_end_mask_0 = const()[name = tensor<string, []>("op_10707_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10707_cast = slice_by_index(begin = var_10707_begin_0, end = var_10707_end_0, end_mask = var_10707_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10707_cast")];
            tensor<int32, [4]> var_10711_begin_0 = const()[name = tensor<string, []>("op_10711_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_10711_end_0 = const()[name = tensor<string, []>("op_10711_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_10711_end_mask_0 = const()[name = tensor<string, []>("op_10711_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10711_cast = slice_by_index(begin = var_10711_begin_0, end = var_10711_end_0, end_mask = var_10711_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10711_cast")];
            tensor<int32, [4]> var_10715_begin_0 = const()[name = tensor<string, []>("op_10715_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_10715_end_0 = const()[name = tensor<string, []>("op_10715_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_10715_end_mask_0 = const()[name = tensor<string, []>("op_10715_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10715_cast = slice_by_index(begin = var_10715_begin_0, end = var_10715_end_0, end_mask = var_10715_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10715_cast")];
            tensor<int32, [4]> var_10719_begin_0 = const()[name = tensor<string, []>("op_10719_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_10719_end_0 = const()[name = tensor<string, []>("op_10719_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_10719_end_mask_0 = const()[name = tensor<string, []>("op_10719_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10719_cast = slice_by_index(begin = var_10719_begin_0, end = var_10719_end_0, end_mask = var_10719_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10719_cast")];
            tensor<int32, [4]> var_10723_begin_0 = const()[name = tensor<string, []>("op_10723_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_10723_end_0 = const()[name = tensor<string, []>("op_10723_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_10723_end_mask_0 = const()[name = tensor<string, []>("op_10723_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10723_cast = slice_by_index(begin = var_10723_begin_0, end = var_10723_end_0, end_mask = var_10723_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10723_cast")];
            tensor<int32, [4]> var_10727_begin_0 = const()[name = tensor<string, []>("op_10727_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_10727_end_0 = const()[name = tensor<string, []>("op_10727_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_10727_end_mask_0 = const()[name = tensor<string, []>("op_10727_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10727_cast = slice_by_index(begin = var_10727_begin_0, end = var_10727_end_0, end_mask = var_10727_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10727_cast")];
            tensor<int32, [4]> var_10731_begin_0 = const()[name = tensor<string, []>("op_10731_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_10731_end_0 = const()[name = tensor<string, []>("op_10731_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_10731_end_mask_0 = const()[name = tensor<string, []>("op_10731_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10731_cast = slice_by_index(begin = var_10731_begin_0, end = var_10731_end_0, end_mask = var_10731_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10731_cast")];
            tensor<int32, [4]> var_10735_begin_0 = const()[name = tensor<string, []>("op_10735_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_10735_end_0 = const()[name = tensor<string, []>("op_10735_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_10735_end_mask_0 = const()[name = tensor<string, []>("op_10735_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10735_cast = slice_by_index(begin = var_10735_begin_0, end = var_10735_end_0, end_mask = var_10735_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10735_cast")];
            tensor<string, []> var_10739_equation_0 = const()[name = tensor<string, []>("op_10739_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10739_cast = einsum(equation = var_10739_equation_0, values = (var_10581_cast, var_10498_cast))[name = tensor<string, []>("op_10739_cast")];
            tensor<fp16, []> var_10740_to_fp16 = const()[name = tensor<string, []>("op_10740_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_841_cast = mul(x = var_10739_cast, y = var_10740_to_fp16)[name = tensor<string, []>("aw_841_cast")];
            tensor<string, []> var_10743_equation_0 = const()[name = tensor<string, []>("op_10743_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10743_cast = einsum(equation = var_10743_equation_0, values = (var_10585_cast, var_10502_cast))[name = tensor<string, []>("op_10743_cast")];
            tensor<fp16, []> var_10744_to_fp16 = const()[name = tensor<string, []>("op_10744_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_843_cast = mul(x = var_10743_cast, y = var_10744_to_fp16)[name = tensor<string, []>("aw_843_cast")];
            tensor<string, []> var_10747_equation_0 = const()[name = tensor<string, []>("op_10747_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10747_cast = einsum(equation = var_10747_equation_0, values = (var_10589_cast, var_10506_cast))[name = tensor<string, []>("op_10747_cast")];
            tensor<fp16, []> var_10748_to_fp16 = const()[name = tensor<string, []>("op_10748_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_845_cast = mul(x = var_10747_cast, y = var_10748_to_fp16)[name = tensor<string, []>("aw_845_cast")];
            tensor<string, []> var_10751_equation_0 = const()[name = tensor<string, []>("op_10751_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10751_cast = einsum(equation = var_10751_equation_0, values = (var_10593_cast, var_10510_cast))[name = tensor<string, []>("op_10751_cast")];
            tensor<fp16, []> var_10752_to_fp16 = const()[name = tensor<string, []>("op_10752_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_847_cast = mul(x = var_10751_cast, y = var_10752_to_fp16)[name = tensor<string, []>("aw_847_cast")];
            tensor<string, []> var_10755_equation_0 = const()[name = tensor<string, []>("op_10755_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10755_cast = einsum(equation = var_10755_equation_0, values = (var_10597_cast, var_10514_cast))[name = tensor<string, []>("op_10755_cast")];
            tensor<fp16, []> var_10756_to_fp16 = const()[name = tensor<string, []>("op_10756_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_849_cast = mul(x = var_10755_cast, y = var_10756_to_fp16)[name = tensor<string, []>("aw_849_cast")];
            tensor<string, []> var_10759_equation_0 = const()[name = tensor<string, []>("op_10759_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10759_cast = einsum(equation = var_10759_equation_0, values = (var_10601_cast, var_10518_cast))[name = tensor<string, []>("op_10759_cast")];
            tensor<fp16, []> var_10760_to_fp16 = const()[name = tensor<string, []>("op_10760_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_851_cast = mul(x = var_10759_cast, y = var_10760_to_fp16)[name = tensor<string, []>("aw_851_cast")];
            tensor<string, []> var_10763_equation_0 = const()[name = tensor<string, []>("op_10763_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10763_cast = einsum(equation = var_10763_equation_0, values = (var_10605_cast, var_10522_cast))[name = tensor<string, []>("op_10763_cast")];
            tensor<fp16, []> var_10764_to_fp16 = const()[name = tensor<string, []>("op_10764_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_853_cast = mul(x = var_10763_cast, y = var_10764_to_fp16)[name = tensor<string, []>("aw_853_cast")];
            tensor<string, []> var_10767_equation_0 = const()[name = tensor<string, []>("op_10767_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10767_cast = einsum(equation = var_10767_equation_0, values = (var_10609_cast, var_10526_cast))[name = tensor<string, []>("op_10767_cast")];
            tensor<fp16, []> var_10768_to_fp16 = const()[name = tensor<string, []>("op_10768_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_855_cast = mul(x = var_10767_cast, y = var_10768_to_fp16)[name = tensor<string, []>("aw_855_cast")];
            tensor<string, []> var_10771_equation_0 = const()[name = tensor<string, []>("op_10771_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10771_cast = einsum(equation = var_10771_equation_0, values = (var_10613_cast, var_10530_cast))[name = tensor<string, []>("op_10771_cast")];
            tensor<fp16, []> var_10772_to_fp16 = const()[name = tensor<string, []>("op_10772_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_857_cast = mul(x = var_10771_cast, y = var_10772_to_fp16)[name = tensor<string, []>("aw_857_cast")];
            tensor<string, []> var_10775_equation_0 = const()[name = tensor<string, []>("op_10775_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10775_cast = einsum(equation = var_10775_equation_0, values = (var_10617_cast, var_10534_cast))[name = tensor<string, []>("op_10775_cast")];
            tensor<fp16, []> var_10776_to_fp16 = const()[name = tensor<string, []>("op_10776_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_859_cast = mul(x = var_10775_cast, y = var_10776_to_fp16)[name = tensor<string, []>("aw_859_cast")];
            tensor<string, []> var_10779_equation_0 = const()[name = tensor<string, []>("op_10779_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10779_cast = einsum(equation = var_10779_equation_0, values = (var_10621_cast, var_10538_cast))[name = tensor<string, []>("op_10779_cast")];
            tensor<fp16, []> var_10780_to_fp16 = const()[name = tensor<string, []>("op_10780_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_861_cast = mul(x = var_10779_cast, y = var_10780_to_fp16)[name = tensor<string, []>("aw_861_cast")];
            tensor<string, []> var_10783_equation_0 = const()[name = tensor<string, []>("op_10783_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10783_cast = einsum(equation = var_10783_equation_0, values = (var_10625_cast, var_10542_cast))[name = tensor<string, []>("op_10783_cast")];
            tensor<fp16, []> var_10784_to_fp16 = const()[name = tensor<string, []>("op_10784_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_863_cast = mul(x = var_10783_cast, y = var_10784_to_fp16)[name = tensor<string, []>("aw_863_cast")];
            tensor<string, []> var_10787_equation_0 = const()[name = tensor<string, []>("op_10787_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10787_cast = einsum(equation = var_10787_equation_0, values = (var_10629_cast, var_10546_cast))[name = tensor<string, []>("op_10787_cast")];
            tensor<fp16, []> var_10788_to_fp16 = const()[name = tensor<string, []>("op_10788_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_865_cast = mul(x = var_10787_cast, y = var_10788_to_fp16)[name = tensor<string, []>("aw_865_cast")];
            tensor<string, []> var_10791_equation_0 = const()[name = tensor<string, []>("op_10791_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10791_cast = einsum(equation = var_10791_equation_0, values = (var_10633_cast, var_10550_cast))[name = tensor<string, []>("op_10791_cast")];
            tensor<fp16, []> var_10792_to_fp16 = const()[name = tensor<string, []>("op_10792_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_867_cast = mul(x = var_10791_cast, y = var_10792_to_fp16)[name = tensor<string, []>("aw_867_cast")];
            tensor<string, []> var_10795_equation_0 = const()[name = tensor<string, []>("op_10795_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10795_cast = einsum(equation = var_10795_equation_0, values = (var_10637_cast, var_10554_cast))[name = tensor<string, []>("op_10795_cast")];
            tensor<fp16, []> var_10796_to_fp16 = const()[name = tensor<string, []>("op_10796_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_869_cast = mul(x = var_10795_cast, y = var_10796_to_fp16)[name = tensor<string, []>("aw_869_cast")];
            tensor<string, []> var_10799_equation_0 = const()[name = tensor<string, []>("op_10799_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10799_cast = einsum(equation = var_10799_equation_0, values = (var_10641_cast, var_10558_cast))[name = tensor<string, []>("op_10799_cast")];
            tensor<fp16, []> var_10800_to_fp16 = const()[name = tensor<string, []>("op_10800_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_871_cast = mul(x = var_10799_cast, y = var_10800_to_fp16)[name = tensor<string, []>("aw_871_cast")];
            tensor<string, []> var_10803_equation_0 = const()[name = tensor<string, []>("op_10803_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10803_cast = einsum(equation = var_10803_equation_0, values = (var_10645_cast, var_10562_cast))[name = tensor<string, []>("op_10803_cast")];
            tensor<fp16, []> var_10804_to_fp16 = const()[name = tensor<string, []>("op_10804_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_873_cast = mul(x = var_10803_cast, y = var_10804_to_fp16)[name = tensor<string, []>("aw_873_cast")];
            tensor<string, []> var_10807_equation_0 = const()[name = tensor<string, []>("op_10807_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10807_cast = einsum(equation = var_10807_equation_0, values = (var_10649_cast, var_10566_cast))[name = tensor<string, []>("op_10807_cast")];
            tensor<fp16, []> var_10808_to_fp16 = const()[name = tensor<string, []>("op_10808_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_875_cast = mul(x = var_10807_cast, y = var_10808_to_fp16)[name = tensor<string, []>("aw_875_cast")];
            tensor<string, []> var_10811_equation_0 = const()[name = tensor<string, []>("op_10811_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10811_cast = einsum(equation = var_10811_equation_0, values = (var_10653_cast, var_10570_cast))[name = tensor<string, []>("op_10811_cast")];
            tensor<fp16, []> var_10812_to_fp16 = const()[name = tensor<string, []>("op_10812_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_877_cast = mul(x = var_10811_cast, y = var_10812_to_fp16)[name = tensor<string, []>("aw_877_cast")];
            tensor<string, []> var_10815_equation_0 = const()[name = tensor<string, []>("op_10815_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10815_cast = einsum(equation = var_10815_equation_0, values = (var_10657_cast, var_10574_cast))[name = tensor<string, []>("op_10815_cast")];
            tensor<fp16, []> var_10816_to_fp16 = const()[name = tensor<string, []>("op_10816_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_879_cast = mul(x = var_10815_cast, y = var_10816_to_fp16)[name = tensor<string, []>("aw_879_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10818_cast = softmax(axis = var_2634, x = aw_841_cast)[name = tensor<string, []>("op_10818_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10819_cast = softmax(axis = var_2634, x = aw_843_cast)[name = tensor<string, []>("op_10819_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10820_cast = softmax(axis = var_2634, x = aw_845_cast)[name = tensor<string, []>("op_10820_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10821_cast = softmax(axis = var_2634, x = aw_847_cast)[name = tensor<string, []>("op_10821_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10822_cast = softmax(axis = var_2634, x = aw_849_cast)[name = tensor<string, []>("op_10822_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10823_cast = softmax(axis = var_2634, x = aw_851_cast)[name = tensor<string, []>("op_10823_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10824_cast = softmax(axis = var_2634, x = aw_853_cast)[name = tensor<string, []>("op_10824_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10825_cast = softmax(axis = var_2634, x = aw_855_cast)[name = tensor<string, []>("op_10825_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10826_cast = softmax(axis = var_2634, x = aw_857_cast)[name = tensor<string, []>("op_10826_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10827_cast = softmax(axis = var_2634, x = aw_859_cast)[name = tensor<string, []>("op_10827_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10828_cast = softmax(axis = var_2634, x = aw_861_cast)[name = tensor<string, []>("op_10828_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10829_cast = softmax(axis = var_2634, x = aw_863_cast)[name = tensor<string, []>("op_10829_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10830_cast = softmax(axis = var_2634, x = aw_865_cast)[name = tensor<string, []>("op_10830_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10831_cast = softmax(axis = var_2634, x = aw_867_cast)[name = tensor<string, []>("op_10831_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10832_cast = softmax(axis = var_2634, x = aw_869_cast)[name = tensor<string, []>("op_10832_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10833_cast = softmax(axis = var_2634, x = aw_871_cast)[name = tensor<string, []>("op_10833_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10834_cast = softmax(axis = var_2634, x = aw_873_cast)[name = tensor<string, []>("op_10834_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10835_cast = softmax(axis = var_2634, x = aw_875_cast)[name = tensor<string, []>("op_10835_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10836_cast = softmax(axis = var_2634, x = aw_877_cast)[name = tensor<string, []>("op_10836_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_10837_cast = softmax(axis = var_2634, x = aw_879_cast)[name = tensor<string, []>("op_10837_cast")];
            tensor<string, []> var_10839_equation_0 = const()[name = tensor<string, []>("op_10839_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10839_cast = einsum(equation = var_10839_equation_0, values = (var_10659_cast, var_10818_cast))[name = tensor<string, []>("op_10839_cast")];
            tensor<string, []> var_10841_equation_0 = const()[name = tensor<string, []>("op_10841_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10841_cast = einsum(equation = var_10841_equation_0, values = (var_10663_cast, var_10819_cast))[name = tensor<string, []>("op_10841_cast")];
            tensor<string, []> var_10843_equation_0 = const()[name = tensor<string, []>("op_10843_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10843_cast = einsum(equation = var_10843_equation_0, values = (var_10667_cast, var_10820_cast))[name = tensor<string, []>("op_10843_cast")];
            tensor<string, []> var_10845_equation_0 = const()[name = tensor<string, []>("op_10845_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10845_cast = einsum(equation = var_10845_equation_0, values = (var_10671_cast, var_10821_cast))[name = tensor<string, []>("op_10845_cast")];
            tensor<string, []> var_10847_equation_0 = const()[name = tensor<string, []>("op_10847_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10847_cast = einsum(equation = var_10847_equation_0, values = (var_10675_cast, var_10822_cast))[name = tensor<string, []>("op_10847_cast")];
            tensor<string, []> var_10849_equation_0 = const()[name = tensor<string, []>("op_10849_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10849_cast = einsum(equation = var_10849_equation_0, values = (var_10679_cast, var_10823_cast))[name = tensor<string, []>("op_10849_cast")];
            tensor<string, []> var_10851_equation_0 = const()[name = tensor<string, []>("op_10851_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10851_cast = einsum(equation = var_10851_equation_0, values = (var_10683_cast, var_10824_cast))[name = tensor<string, []>("op_10851_cast")];
            tensor<string, []> var_10853_equation_0 = const()[name = tensor<string, []>("op_10853_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10853_cast = einsum(equation = var_10853_equation_0, values = (var_10687_cast, var_10825_cast))[name = tensor<string, []>("op_10853_cast")];
            tensor<string, []> var_10855_equation_0 = const()[name = tensor<string, []>("op_10855_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10855_cast = einsum(equation = var_10855_equation_0, values = (var_10691_cast, var_10826_cast))[name = tensor<string, []>("op_10855_cast")];
            tensor<string, []> var_10857_equation_0 = const()[name = tensor<string, []>("op_10857_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10857_cast = einsum(equation = var_10857_equation_0, values = (var_10695_cast, var_10827_cast))[name = tensor<string, []>("op_10857_cast")];
            tensor<string, []> var_10859_equation_0 = const()[name = tensor<string, []>("op_10859_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10859_cast = einsum(equation = var_10859_equation_0, values = (var_10699_cast, var_10828_cast))[name = tensor<string, []>("op_10859_cast")];
            tensor<string, []> var_10861_equation_0 = const()[name = tensor<string, []>("op_10861_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10861_cast = einsum(equation = var_10861_equation_0, values = (var_10703_cast, var_10829_cast))[name = tensor<string, []>("op_10861_cast")];
            tensor<string, []> var_10863_equation_0 = const()[name = tensor<string, []>("op_10863_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10863_cast = einsum(equation = var_10863_equation_0, values = (var_10707_cast, var_10830_cast))[name = tensor<string, []>("op_10863_cast")];
            tensor<string, []> var_10865_equation_0 = const()[name = tensor<string, []>("op_10865_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10865_cast = einsum(equation = var_10865_equation_0, values = (var_10711_cast, var_10831_cast))[name = tensor<string, []>("op_10865_cast")];
            tensor<string, []> var_10867_equation_0 = const()[name = tensor<string, []>("op_10867_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10867_cast = einsum(equation = var_10867_equation_0, values = (var_10715_cast, var_10832_cast))[name = tensor<string, []>("op_10867_cast")];
            tensor<string, []> var_10869_equation_0 = const()[name = tensor<string, []>("op_10869_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10869_cast = einsum(equation = var_10869_equation_0, values = (var_10719_cast, var_10833_cast))[name = tensor<string, []>("op_10869_cast")];
            tensor<string, []> var_10871_equation_0 = const()[name = tensor<string, []>("op_10871_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10871_cast = einsum(equation = var_10871_equation_0, values = (var_10723_cast, var_10834_cast))[name = tensor<string, []>("op_10871_cast")];
            tensor<string, []> var_10873_equation_0 = const()[name = tensor<string, []>("op_10873_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10873_cast = einsum(equation = var_10873_equation_0, values = (var_10727_cast, var_10835_cast))[name = tensor<string, []>("op_10873_cast")];
            tensor<string, []> var_10875_equation_0 = const()[name = tensor<string, []>("op_10875_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10875_cast = einsum(equation = var_10875_equation_0, values = (var_10731_cast, var_10836_cast))[name = tensor<string, []>("op_10875_cast")];
            tensor<string, []> var_10877_equation_0 = const()[name = tensor<string, []>("op_10877_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10877_cast = einsum(equation = var_10877_equation_0, values = (var_10735_cast, var_10837_cast))[name = tensor<string, []>("op_10877_cast")];
            tensor<bool, []> input_201_interleave_0 = const()[name = tensor<string, []>("input_201_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_201_cast = concat(axis = var_2634, interleave = input_201_interleave_0, values = (var_10839_cast, var_10841_cast, var_10843_cast, var_10845_cast, var_10847_cast, var_10849_cast, var_10851_cast, var_10853_cast, var_10855_cast, var_10857_cast, var_10859_cast, var_10861_cast, var_10863_cast, var_10865_cast, var_10867_cast, var_10869_cast, var_10871_cast, var_10873_cast, var_10875_cast, var_10877_cast))[name = tensor<string, []>("input_201_cast")];
            tensor<int32, [2]> var_10883 = const()[name = tensor<string, []>("op_10883"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10885 = const()[name = tensor<string, []>("op_10885"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10887_pad_type_0 = const()[name = tensor<string, []>("op_10887_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10887_pad_0 = const()[name = tensor<string, []>("op_10887_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(781152896)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(784429760)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_10887_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16, dilations = var_10885, groups = var_2634, pad = var_10887_pad_0, pad_type = var_10887_pad_type_0, strides = var_10883, weight = down_blocks_2_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16, x = input_201_cast)[name = tensor<string, []>("op_10887_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_77_cast = add(x = var_10887_cast, y = inputs_75_cast)[name = tensor<string, []>("inputs_77_cast")];
            tensor<int32, [1]> var_10891 = const()[name = tensor<string, []>("op_10891"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_77_cast = reduce_mean(axes = var_10891, keep_dims = var_2629, x = inputs_77_cast)[name = tensor<string, []>("channels_mean_77_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_77_cast = sub(x = inputs_77_cast, y = channels_mean_77_cast)[name = tensor<string, []>("zero_mean_77_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_77_cast = mul(x = zero_mean_77_cast, y = zero_mean_77_cast)[name = tensor<string, []>("zero_mean_sq_77_cast")];
            tensor<int32, [1]> var_10895 = const()[name = tensor<string, []>("op_10895"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_10896_cast = reduce_mean(axes = var_10895, keep_dims = var_2629, x = zero_mean_sq_77_cast)[name = tensor<string, []>("op_10896_cast")];
            tensor<fp16, []> var_10897_to_fp16 = const()[name = tensor<string, []>("op_10897_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_10898_cast = add(x = var_10896_cast, y = var_10897_to_fp16)[name = tensor<string, []>("op_10898_cast")];
            tensor<fp16, []> denom_77_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_77_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_77_cast = rsqrt(epsilon = denom_77_epsilon_0_to_fp16, x = var_10898_cast)[name = tensor<string, []>("denom_77_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_77_cast = mul(x = zero_mean_77_cast, y = denom_77_cast)[name = tensor<string, []>("out_77_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_10902_to_fp16 = const()[name = tensor<string, []>("op_10902_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(784432384)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_10903_cast = add(x = out_77_cast, y = var_10902_to_fp16)[name = tensor<string, []>("op_10903_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_10905_to_fp16 = const()[name = tensor<string, []>("op_10905_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(784435008)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_203_cast = mul(x = var_10903_cast, y = var_10905_to_fp16)[name = tensor<string, []>("input_203_cast")];
            tensor<int32, [2]> var_10913 = const()[name = tensor<string, []>("op_10913"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10915 = const()[name = tensor<string, []>("op_10915"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10917_pad_type_0 = const()[name = tensor<string, []>("op_10917_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10917_pad_0 = const()[name = tensor<string, []>("op_10917_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(784437632)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(810652096)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_10917_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16, dilations = var_10915, groups = var_2634, pad = var_10917_pad_0, pad_type = var_10917_pad_type_0, strides = var_10913, weight = down_blocks_2_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16, x = input_203_cast)[name = tensor<string, []>("op_10917_cast")];
            tensor<int32, [2]> var_10918_split_sizes_0 = const()[name = tensor<string, []>("op_10918_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_10918_axis_0 = const()[name = tensor<string, []>("op_10918_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_10918_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_10918_cast_1 = split(axis = var_10918_axis_0, split_sizes = var_10918_split_sizes_0, x = var_10917_cast)[name = tensor<string, []>("op_10918_cast")];
            tensor<string, []> var_10920_mode_0 = const()[name = tensor<string, []>("op_10920_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_10920_cast = gelu(mode = var_10920_mode_0, x = var_10918_cast_1)[name = tensor<string, []>("op_10920_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_205_cast = mul(x = var_10918_cast_0, y = var_10920_cast)[name = tensor<string, []>("input_205_cast")];
            tensor<int32, [2]> var_10924 = const()[name = tensor<string, []>("op_10924"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10926 = const()[name = tensor<string, []>("op_10926"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10928_pad_type_0 = const()[name = tensor<string, []>("op_10928_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10928_pad_0 = const()[name = tensor<string, []>("op_10928_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(810672640)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(823779904)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_10928_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16, dilations = var_10926, groups = var_2634, pad = var_10928_pad_0, pad_type = var_10928_pad_type_0, strides = var_10924, weight = down_blocks_2_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16, x = input_205_cast)[name = tensor<string, []>("op_10928_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_79_cast = add(x = var_10928_cast, y = inputs_77_cast)[name = tensor<string, []>("inputs_79_cast")];
            tensor<int32, [1]> var_10938 = const()[name = tensor<string, []>("op_10938"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_79_cast = reduce_mean(axes = var_10938, keep_dims = var_2629, x = inputs_79_cast)[name = tensor<string, []>("channels_mean_79_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_79_cast = sub(x = inputs_79_cast, y = channels_mean_79_cast)[name = tensor<string, []>("zero_mean_79_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_79_cast = mul(x = zero_mean_79_cast, y = zero_mean_79_cast)[name = tensor<string, []>("zero_mean_sq_79_cast")];
            tensor<int32, [1]> var_10942 = const()[name = tensor<string, []>("op_10942"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_10943_cast = reduce_mean(axes = var_10942, keep_dims = var_2629, x = zero_mean_sq_79_cast)[name = tensor<string, []>("op_10943_cast")];
            tensor<fp16, []> var_10944_to_fp16 = const()[name = tensor<string, []>("op_10944_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_10945_cast = add(x = var_10943_cast, y = var_10944_to_fp16)[name = tensor<string, []>("op_10945_cast")];
            tensor<fp16, []> denom_79_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_79_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_79_cast = rsqrt(epsilon = denom_79_epsilon_0_to_fp16, x = var_10945_cast)[name = tensor<string, []>("denom_79_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_79_cast = mul(x = zero_mean_79_cast, y = denom_79_cast)[name = tensor<string, []>("out_79_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_10949_to_fp16 = const()[name = tensor<string, []>("op_10949_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(823782528)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_10950_cast = add(x = out_79_cast, y = var_10949_to_fp16)[name = tensor<string, []>("op_10950_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_10952_to_fp16 = const()[name = tensor<string, []>("op_10952_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(823785152)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_119_cast = mul(x = var_10950_cast, y = var_10952_to_fp16)[name = tensor<string, []>("hidden_states_119_cast")];
            tensor<int32, [2]> var_10959 = const()[name = tensor<string, []>("op_10959"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10961 = const()[name = tensor<string, []>("op_10961"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_53_pad_type_0 = const()[name = tensor<string, []>("q_53_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_53_pad_0 = const()[name = tensor<string, []>("q_53_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(823787776)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_53_cast = conv(dilations = var_10961, groups = var_2634, pad = q_53_pad_0, pad_type = q_53_pad_type_0, strides = var_10959, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16, x = hidden_states_119_cast)[name = tensor<string, []>("q_53_cast")];
            tensor<int32, [2]> var_10965 = const()[name = tensor<string, []>("op_10965"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10967 = const()[name = tensor<string, []>("op_10967"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_105_pad_type_0 = const()[name = tensor<string, []>("k_105_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_105_pad_0 = const()[name = tensor<string, []>("k_105_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(827064640)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_105_cast = conv(dilations = var_10967, groups = var_2634, pad = k_105_pad_0, pad_type = k_105_pad_type_0, strides = var_10965, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16, x = hidden_states_119_cast)[name = tensor<string, []>("k_105_cast")];
            tensor<int32, [2]> var_10971 = const()[name = tensor<string, []>("op_10971"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10973 = const()[name = tensor<string, []>("op_10973"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_53_pad_type_0 = const()[name = tensor<string, []>("v_53_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_53_pad_0 = const()[name = tensor<string, []>("v_53_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(830341504)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_53_cast = conv(dilations = var_10973, groups = var_2634, pad = v_53_pad_0, pad_type = v_53_pad_type_0, strides = var_10971, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16, x = hidden_states_119_cast)[name = tensor<string, []>("v_53_cast")];
            tensor<int32, [4]> var_10977_begin_0 = const()[name = tensor<string, []>("op_10977_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10977_end_0 = const()[name = tensor<string, []>("op_10977_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10977_end_mask_0 = const()[name = tensor<string, []>("op_10977_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10977_cast = slice_by_index(begin = var_10977_begin_0, end = var_10977_end_0, end_mask = var_10977_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_10977_cast")];
            tensor<int32, [4]> var_10981_begin_0 = const()[name = tensor<string, []>("op_10981_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_10981_end_0 = const()[name = tensor<string, []>("op_10981_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_10981_end_mask_0 = const()[name = tensor<string, []>("op_10981_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10981_cast = slice_by_index(begin = var_10981_begin_0, end = var_10981_end_0, end_mask = var_10981_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_10981_cast")];
            tensor<int32, [4]> var_10985_begin_0 = const()[name = tensor<string, []>("op_10985_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_10985_end_0 = const()[name = tensor<string, []>("op_10985_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_10985_end_mask_0 = const()[name = tensor<string, []>("op_10985_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10985_cast = slice_by_index(begin = var_10985_begin_0, end = var_10985_end_0, end_mask = var_10985_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_10985_cast")];
            tensor<int32, [4]> var_10989_begin_0 = const()[name = tensor<string, []>("op_10989_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_10989_end_0 = const()[name = tensor<string, []>("op_10989_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_10989_end_mask_0 = const()[name = tensor<string, []>("op_10989_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10989_cast = slice_by_index(begin = var_10989_begin_0, end = var_10989_end_0, end_mask = var_10989_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_10989_cast")];
            tensor<int32, [4]> var_10993_begin_0 = const()[name = tensor<string, []>("op_10993_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_10993_end_0 = const()[name = tensor<string, []>("op_10993_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_10993_end_mask_0 = const()[name = tensor<string, []>("op_10993_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10993_cast = slice_by_index(begin = var_10993_begin_0, end = var_10993_end_0, end_mask = var_10993_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_10993_cast")];
            tensor<int32, [4]> var_10997_begin_0 = const()[name = tensor<string, []>("op_10997_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_10997_end_0 = const()[name = tensor<string, []>("op_10997_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_10997_end_mask_0 = const()[name = tensor<string, []>("op_10997_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10997_cast = slice_by_index(begin = var_10997_begin_0, end = var_10997_end_0, end_mask = var_10997_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_10997_cast")];
            tensor<int32, [4]> var_11001_begin_0 = const()[name = tensor<string, []>("op_11001_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_11001_end_0 = const()[name = tensor<string, []>("op_11001_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_11001_end_mask_0 = const()[name = tensor<string, []>("op_11001_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11001_cast = slice_by_index(begin = var_11001_begin_0, end = var_11001_end_0, end_mask = var_11001_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_11001_cast")];
            tensor<int32, [4]> var_11005_begin_0 = const()[name = tensor<string, []>("op_11005_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_11005_end_0 = const()[name = tensor<string, []>("op_11005_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_11005_end_mask_0 = const()[name = tensor<string, []>("op_11005_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11005_cast = slice_by_index(begin = var_11005_begin_0, end = var_11005_end_0, end_mask = var_11005_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_11005_cast")];
            tensor<int32, [4]> var_11009_begin_0 = const()[name = tensor<string, []>("op_11009_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_11009_end_0 = const()[name = tensor<string, []>("op_11009_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_11009_end_mask_0 = const()[name = tensor<string, []>("op_11009_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11009_cast = slice_by_index(begin = var_11009_begin_0, end = var_11009_end_0, end_mask = var_11009_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_11009_cast")];
            tensor<int32, [4]> var_11013_begin_0 = const()[name = tensor<string, []>("op_11013_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_11013_end_0 = const()[name = tensor<string, []>("op_11013_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_11013_end_mask_0 = const()[name = tensor<string, []>("op_11013_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11013_cast = slice_by_index(begin = var_11013_begin_0, end = var_11013_end_0, end_mask = var_11013_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_11013_cast")];
            tensor<int32, [4]> var_11017_begin_0 = const()[name = tensor<string, []>("op_11017_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_11017_end_0 = const()[name = tensor<string, []>("op_11017_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_11017_end_mask_0 = const()[name = tensor<string, []>("op_11017_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11017_cast = slice_by_index(begin = var_11017_begin_0, end = var_11017_end_0, end_mask = var_11017_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_11017_cast")];
            tensor<int32, [4]> var_11021_begin_0 = const()[name = tensor<string, []>("op_11021_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_11021_end_0 = const()[name = tensor<string, []>("op_11021_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_11021_end_mask_0 = const()[name = tensor<string, []>("op_11021_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11021_cast = slice_by_index(begin = var_11021_begin_0, end = var_11021_end_0, end_mask = var_11021_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_11021_cast")];
            tensor<int32, [4]> var_11025_begin_0 = const()[name = tensor<string, []>("op_11025_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_11025_end_0 = const()[name = tensor<string, []>("op_11025_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_11025_end_mask_0 = const()[name = tensor<string, []>("op_11025_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11025_cast = slice_by_index(begin = var_11025_begin_0, end = var_11025_end_0, end_mask = var_11025_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_11025_cast")];
            tensor<int32, [4]> var_11029_begin_0 = const()[name = tensor<string, []>("op_11029_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_11029_end_0 = const()[name = tensor<string, []>("op_11029_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_11029_end_mask_0 = const()[name = tensor<string, []>("op_11029_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11029_cast = slice_by_index(begin = var_11029_begin_0, end = var_11029_end_0, end_mask = var_11029_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_11029_cast")];
            tensor<int32, [4]> var_11033_begin_0 = const()[name = tensor<string, []>("op_11033_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_11033_end_0 = const()[name = tensor<string, []>("op_11033_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_11033_end_mask_0 = const()[name = tensor<string, []>("op_11033_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11033_cast = slice_by_index(begin = var_11033_begin_0, end = var_11033_end_0, end_mask = var_11033_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_11033_cast")];
            tensor<int32, [4]> var_11037_begin_0 = const()[name = tensor<string, []>("op_11037_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_11037_end_0 = const()[name = tensor<string, []>("op_11037_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_11037_end_mask_0 = const()[name = tensor<string, []>("op_11037_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11037_cast = slice_by_index(begin = var_11037_begin_0, end = var_11037_end_0, end_mask = var_11037_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_11037_cast")];
            tensor<int32, [4]> var_11041_begin_0 = const()[name = tensor<string, []>("op_11041_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_11041_end_0 = const()[name = tensor<string, []>("op_11041_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_11041_end_mask_0 = const()[name = tensor<string, []>("op_11041_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11041_cast = slice_by_index(begin = var_11041_begin_0, end = var_11041_end_0, end_mask = var_11041_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_11041_cast")];
            tensor<int32, [4]> var_11045_begin_0 = const()[name = tensor<string, []>("op_11045_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_11045_end_0 = const()[name = tensor<string, []>("op_11045_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_11045_end_mask_0 = const()[name = tensor<string, []>("op_11045_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11045_cast = slice_by_index(begin = var_11045_begin_0, end = var_11045_end_0, end_mask = var_11045_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_11045_cast")];
            tensor<int32, [4]> var_11049_begin_0 = const()[name = tensor<string, []>("op_11049_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_11049_end_0 = const()[name = tensor<string, []>("op_11049_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_11049_end_mask_0 = const()[name = tensor<string, []>("op_11049_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11049_cast = slice_by_index(begin = var_11049_begin_0, end = var_11049_end_0, end_mask = var_11049_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_11049_cast")];
            tensor<int32, [4]> var_11053_begin_0 = const()[name = tensor<string, []>("op_11053_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_11053_end_0 = const()[name = tensor<string, []>("op_11053_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_11053_end_mask_0 = const()[name = tensor<string, []>("op_11053_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11053_cast = slice_by_index(begin = var_11053_begin_0, end = var_11053_end_0, end_mask = var_11053_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_11053_cast")];
            tensor<int32, [4]> k_107_perm_0 = const()[name = tensor<string, []>("k_107_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_11060_begin_0 = const()[name = tensor<string, []>("op_11060_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11060_end_0 = const()[name = tensor<string, []>("op_11060_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_11060_end_mask_0 = const()[name = tensor<string, []>("op_11060_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_113 = transpose(perm = k_107_perm_0, x = k_105_cast)[name = tensor<string, []>("transpose_113")];
            tensor<fp16, [2, 1024, 1, 64]> var_11060_cast = slice_by_index(begin = var_11060_begin_0, end = var_11060_end_0, end_mask = var_11060_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11060_cast")];
            tensor<int32, [4]> var_11064_begin_0 = const()[name = tensor<string, []>("op_11064_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_11064_end_0 = const()[name = tensor<string, []>("op_11064_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_11064_end_mask_0 = const()[name = tensor<string, []>("op_11064_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11064_cast = slice_by_index(begin = var_11064_begin_0, end = var_11064_end_0, end_mask = var_11064_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11064_cast")];
            tensor<int32, [4]> var_11068_begin_0 = const()[name = tensor<string, []>("op_11068_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_11068_end_0 = const()[name = tensor<string, []>("op_11068_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_11068_end_mask_0 = const()[name = tensor<string, []>("op_11068_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11068_cast = slice_by_index(begin = var_11068_begin_0, end = var_11068_end_0, end_mask = var_11068_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11068_cast")];
            tensor<int32, [4]> var_11072_begin_0 = const()[name = tensor<string, []>("op_11072_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_11072_end_0 = const()[name = tensor<string, []>("op_11072_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_11072_end_mask_0 = const()[name = tensor<string, []>("op_11072_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11072_cast = slice_by_index(begin = var_11072_begin_0, end = var_11072_end_0, end_mask = var_11072_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11072_cast")];
            tensor<int32, [4]> var_11076_begin_0 = const()[name = tensor<string, []>("op_11076_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_11076_end_0 = const()[name = tensor<string, []>("op_11076_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_11076_end_mask_0 = const()[name = tensor<string, []>("op_11076_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11076_cast = slice_by_index(begin = var_11076_begin_0, end = var_11076_end_0, end_mask = var_11076_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11076_cast")];
            tensor<int32, [4]> var_11080_begin_0 = const()[name = tensor<string, []>("op_11080_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_11080_end_0 = const()[name = tensor<string, []>("op_11080_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_11080_end_mask_0 = const()[name = tensor<string, []>("op_11080_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11080_cast = slice_by_index(begin = var_11080_begin_0, end = var_11080_end_0, end_mask = var_11080_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11080_cast")];
            tensor<int32, [4]> var_11084_begin_0 = const()[name = tensor<string, []>("op_11084_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_11084_end_0 = const()[name = tensor<string, []>("op_11084_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_11084_end_mask_0 = const()[name = tensor<string, []>("op_11084_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11084_cast = slice_by_index(begin = var_11084_begin_0, end = var_11084_end_0, end_mask = var_11084_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11084_cast")];
            tensor<int32, [4]> var_11088_begin_0 = const()[name = tensor<string, []>("op_11088_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_11088_end_0 = const()[name = tensor<string, []>("op_11088_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_11088_end_mask_0 = const()[name = tensor<string, []>("op_11088_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11088_cast = slice_by_index(begin = var_11088_begin_0, end = var_11088_end_0, end_mask = var_11088_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11088_cast")];
            tensor<int32, [4]> var_11092_begin_0 = const()[name = tensor<string, []>("op_11092_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11092_end_0 = const()[name = tensor<string, []>("op_11092_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_11092_end_mask_0 = const()[name = tensor<string, []>("op_11092_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11092_cast = slice_by_index(begin = var_11092_begin_0, end = var_11092_end_0, end_mask = var_11092_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11092_cast")];
            tensor<int32, [4]> var_11096_begin_0 = const()[name = tensor<string, []>("op_11096_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_11096_end_0 = const()[name = tensor<string, []>("op_11096_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_11096_end_mask_0 = const()[name = tensor<string, []>("op_11096_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11096_cast = slice_by_index(begin = var_11096_begin_0, end = var_11096_end_0, end_mask = var_11096_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11096_cast")];
            tensor<int32, [4]> var_11100_begin_0 = const()[name = tensor<string, []>("op_11100_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_11100_end_0 = const()[name = tensor<string, []>("op_11100_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_11100_end_mask_0 = const()[name = tensor<string, []>("op_11100_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11100_cast = slice_by_index(begin = var_11100_begin_0, end = var_11100_end_0, end_mask = var_11100_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11100_cast")];
            tensor<int32, [4]> var_11104_begin_0 = const()[name = tensor<string, []>("op_11104_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_11104_end_0 = const()[name = tensor<string, []>("op_11104_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_11104_end_mask_0 = const()[name = tensor<string, []>("op_11104_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11104_cast = slice_by_index(begin = var_11104_begin_0, end = var_11104_end_0, end_mask = var_11104_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11104_cast")];
            tensor<int32, [4]> var_11108_begin_0 = const()[name = tensor<string, []>("op_11108_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_11108_end_0 = const()[name = tensor<string, []>("op_11108_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_11108_end_mask_0 = const()[name = tensor<string, []>("op_11108_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11108_cast = slice_by_index(begin = var_11108_begin_0, end = var_11108_end_0, end_mask = var_11108_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11108_cast")];
            tensor<int32, [4]> var_11112_begin_0 = const()[name = tensor<string, []>("op_11112_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_11112_end_0 = const()[name = tensor<string, []>("op_11112_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_11112_end_mask_0 = const()[name = tensor<string, []>("op_11112_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11112_cast = slice_by_index(begin = var_11112_begin_0, end = var_11112_end_0, end_mask = var_11112_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11112_cast")];
            tensor<int32, [4]> var_11116_begin_0 = const()[name = tensor<string, []>("op_11116_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_11116_end_0 = const()[name = tensor<string, []>("op_11116_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_11116_end_mask_0 = const()[name = tensor<string, []>("op_11116_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11116_cast = slice_by_index(begin = var_11116_begin_0, end = var_11116_end_0, end_mask = var_11116_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11116_cast")];
            tensor<int32, [4]> var_11120_begin_0 = const()[name = tensor<string, []>("op_11120_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_11120_end_0 = const()[name = tensor<string, []>("op_11120_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_11120_end_mask_0 = const()[name = tensor<string, []>("op_11120_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11120_cast = slice_by_index(begin = var_11120_begin_0, end = var_11120_end_0, end_mask = var_11120_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11120_cast")];
            tensor<int32, [4]> var_11124_begin_0 = const()[name = tensor<string, []>("op_11124_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_11124_end_0 = const()[name = tensor<string, []>("op_11124_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_11124_end_mask_0 = const()[name = tensor<string, []>("op_11124_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11124_cast = slice_by_index(begin = var_11124_begin_0, end = var_11124_end_0, end_mask = var_11124_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11124_cast")];
            tensor<int32, [4]> var_11128_begin_0 = const()[name = tensor<string, []>("op_11128_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_11128_end_0 = const()[name = tensor<string, []>("op_11128_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_11128_end_mask_0 = const()[name = tensor<string, []>("op_11128_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11128_cast = slice_by_index(begin = var_11128_begin_0, end = var_11128_end_0, end_mask = var_11128_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11128_cast")];
            tensor<int32, [4]> var_11132_begin_0 = const()[name = tensor<string, []>("op_11132_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_11132_end_0 = const()[name = tensor<string, []>("op_11132_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_11132_end_mask_0 = const()[name = tensor<string, []>("op_11132_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11132_cast = slice_by_index(begin = var_11132_begin_0, end = var_11132_end_0, end_mask = var_11132_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11132_cast")];
            tensor<int32, [4]> var_11136_begin_0 = const()[name = tensor<string, []>("op_11136_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_11136_end_0 = const()[name = tensor<string, []>("op_11136_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_11136_end_mask_0 = const()[name = tensor<string, []>("op_11136_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11136_cast = slice_by_index(begin = var_11136_begin_0, end = var_11136_end_0, end_mask = var_11136_end_mask_0, x = transpose_113)[name = tensor<string, []>("op_11136_cast")];
            tensor<int32, [4]> var_11138_begin_0 = const()[name = tensor<string, []>("op_11138_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11138_end_0 = const()[name = tensor<string, []>("op_11138_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_11138_end_mask_0 = const()[name = tensor<string, []>("op_11138_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11138_cast = slice_by_index(begin = var_11138_begin_0, end = var_11138_end_0, end_mask = var_11138_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11138_cast")];
            tensor<int32, [4]> var_11142_begin_0 = const()[name = tensor<string, []>("op_11142_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_11142_end_0 = const()[name = tensor<string, []>("op_11142_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_11142_end_mask_0 = const()[name = tensor<string, []>("op_11142_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11142_cast = slice_by_index(begin = var_11142_begin_0, end = var_11142_end_0, end_mask = var_11142_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11142_cast")];
            tensor<int32, [4]> var_11146_begin_0 = const()[name = tensor<string, []>("op_11146_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_11146_end_0 = const()[name = tensor<string, []>("op_11146_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_11146_end_mask_0 = const()[name = tensor<string, []>("op_11146_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11146_cast = slice_by_index(begin = var_11146_begin_0, end = var_11146_end_0, end_mask = var_11146_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11146_cast")];
            tensor<int32, [4]> var_11150_begin_0 = const()[name = tensor<string, []>("op_11150_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_11150_end_0 = const()[name = tensor<string, []>("op_11150_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_11150_end_mask_0 = const()[name = tensor<string, []>("op_11150_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11150_cast = slice_by_index(begin = var_11150_begin_0, end = var_11150_end_0, end_mask = var_11150_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11150_cast")];
            tensor<int32, [4]> var_11154_begin_0 = const()[name = tensor<string, []>("op_11154_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_11154_end_0 = const()[name = tensor<string, []>("op_11154_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_11154_end_mask_0 = const()[name = tensor<string, []>("op_11154_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11154_cast = slice_by_index(begin = var_11154_begin_0, end = var_11154_end_0, end_mask = var_11154_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11154_cast")];
            tensor<int32, [4]> var_11158_begin_0 = const()[name = tensor<string, []>("op_11158_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_11158_end_0 = const()[name = tensor<string, []>("op_11158_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_11158_end_mask_0 = const()[name = tensor<string, []>("op_11158_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11158_cast = slice_by_index(begin = var_11158_begin_0, end = var_11158_end_0, end_mask = var_11158_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11158_cast")];
            tensor<int32, [4]> var_11162_begin_0 = const()[name = tensor<string, []>("op_11162_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_11162_end_0 = const()[name = tensor<string, []>("op_11162_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_11162_end_mask_0 = const()[name = tensor<string, []>("op_11162_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11162_cast = slice_by_index(begin = var_11162_begin_0, end = var_11162_end_0, end_mask = var_11162_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11162_cast")];
            tensor<int32, [4]> var_11166_begin_0 = const()[name = tensor<string, []>("op_11166_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_11166_end_0 = const()[name = tensor<string, []>("op_11166_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_11166_end_mask_0 = const()[name = tensor<string, []>("op_11166_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11166_cast = slice_by_index(begin = var_11166_begin_0, end = var_11166_end_0, end_mask = var_11166_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11166_cast")];
            tensor<int32, [4]> var_11170_begin_0 = const()[name = tensor<string, []>("op_11170_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_11170_end_0 = const()[name = tensor<string, []>("op_11170_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_11170_end_mask_0 = const()[name = tensor<string, []>("op_11170_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11170_cast = slice_by_index(begin = var_11170_begin_0, end = var_11170_end_0, end_mask = var_11170_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11170_cast")];
            tensor<int32, [4]> var_11174_begin_0 = const()[name = tensor<string, []>("op_11174_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_11174_end_0 = const()[name = tensor<string, []>("op_11174_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_11174_end_mask_0 = const()[name = tensor<string, []>("op_11174_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11174_cast = slice_by_index(begin = var_11174_begin_0, end = var_11174_end_0, end_mask = var_11174_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11174_cast")];
            tensor<int32, [4]> var_11178_begin_0 = const()[name = tensor<string, []>("op_11178_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_11178_end_0 = const()[name = tensor<string, []>("op_11178_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_11178_end_mask_0 = const()[name = tensor<string, []>("op_11178_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11178_cast = slice_by_index(begin = var_11178_begin_0, end = var_11178_end_0, end_mask = var_11178_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11178_cast")];
            tensor<int32, [4]> var_11182_begin_0 = const()[name = tensor<string, []>("op_11182_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_11182_end_0 = const()[name = tensor<string, []>("op_11182_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_11182_end_mask_0 = const()[name = tensor<string, []>("op_11182_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11182_cast = slice_by_index(begin = var_11182_begin_0, end = var_11182_end_0, end_mask = var_11182_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11182_cast")];
            tensor<int32, [4]> var_11186_begin_0 = const()[name = tensor<string, []>("op_11186_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_11186_end_0 = const()[name = tensor<string, []>("op_11186_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_11186_end_mask_0 = const()[name = tensor<string, []>("op_11186_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11186_cast = slice_by_index(begin = var_11186_begin_0, end = var_11186_end_0, end_mask = var_11186_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11186_cast")];
            tensor<int32, [4]> var_11190_begin_0 = const()[name = tensor<string, []>("op_11190_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_11190_end_0 = const()[name = tensor<string, []>("op_11190_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_11190_end_mask_0 = const()[name = tensor<string, []>("op_11190_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11190_cast = slice_by_index(begin = var_11190_begin_0, end = var_11190_end_0, end_mask = var_11190_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11190_cast")];
            tensor<int32, [4]> var_11194_begin_0 = const()[name = tensor<string, []>("op_11194_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_11194_end_0 = const()[name = tensor<string, []>("op_11194_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_11194_end_mask_0 = const()[name = tensor<string, []>("op_11194_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11194_cast = slice_by_index(begin = var_11194_begin_0, end = var_11194_end_0, end_mask = var_11194_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11194_cast")];
            tensor<int32, [4]> var_11198_begin_0 = const()[name = tensor<string, []>("op_11198_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_11198_end_0 = const()[name = tensor<string, []>("op_11198_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_11198_end_mask_0 = const()[name = tensor<string, []>("op_11198_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11198_cast = slice_by_index(begin = var_11198_begin_0, end = var_11198_end_0, end_mask = var_11198_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11198_cast")];
            tensor<int32, [4]> var_11202_begin_0 = const()[name = tensor<string, []>("op_11202_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_11202_end_0 = const()[name = tensor<string, []>("op_11202_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_11202_end_mask_0 = const()[name = tensor<string, []>("op_11202_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11202_cast = slice_by_index(begin = var_11202_begin_0, end = var_11202_end_0, end_mask = var_11202_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11202_cast")];
            tensor<int32, [4]> var_11206_begin_0 = const()[name = tensor<string, []>("op_11206_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_11206_end_0 = const()[name = tensor<string, []>("op_11206_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_11206_end_mask_0 = const()[name = tensor<string, []>("op_11206_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11206_cast = slice_by_index(begin = var_11206_begin_0, end = var_11206_end_0, end_mask = var_11206_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11206_cast")];
            tensor<int32, [4]> var_11210_begin_0 = const()[name = tensor<string, []>("op_11210_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_11210_end_0 = const()[name = tensor<string, []>("op_11210_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_11210_end_mask_0 = const()[name = tensor<string, []>("op_11210_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11210_cast = slice_by_index(begin = var_11210_begin_0, end = var_11210_end_0, end_mask = var_11210_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11210_cast")];
            tensor<int32, [4]> var_11214_begin_0 = const()[name = tensor<string, []>("op_11214_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_11214_end_0 = const()[name = tensor<string, []>("op_11214_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_11214_end_mask_0 = const()[name = tensor<string, []>("op_11214_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11214_cast = slice_by_index(begin = var_11214_begin_0, end = var_11214_end_0, end_mask = var_11214_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_11214_cast")];
            tensor<string, []> var_11218_equation_0 = const()[name = tensor<string, []>("op_11218_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11218_cast = einsum(equation = var_11218_equation_0, values = (var_11060_cast, var_10977_cast))[name = tensor<string, []>("op_11218_cast")];
            tensor<fp16, []> var_11219_to_fp16 = const()[name = tensor<string, []>("op_11219_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_881_cast = mul(x = var_11218_cast, y = var_11219_to_fp16)[name = tensor<string, []>("aw_881_cast")];
            tensor<string, []> var_11222_equation_0 = const()[name = tensor<string, []>("op_11222_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11222_cast = einsum(equation = var_11222_equation_0, values = (var_11064_cast, var_10981_cast))[name = tensor<string, []>("op_11222_cast")];
            tensor<fp16, []> var_11223_to_fp16 = const()[name = tensor<string, []>("op_11223_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_883_cast = mul(x = var_11222_cast, y = var_11223_to_fp16)[name = tensor<string, []>("aw_883_cast")];
            tensor<string, []> var_11226_equation_0 = const()[name = tensor<string, []>("op_11226_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11226_cast = einsum(equation = var_11226_equation_0, values = (var_11068_cast, var_10985_cast))[name = tensor<string, []>("op_11226_cast")];
            tensor<fp16, []> var_11227_to_fp16 = const()[name = tensor<string, []>("op_11227_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_885_cast = mul(x = var_11226_cast, y = var_11227_to_fp16)[name = tensor<string, []>("aw_885_cast")];
            tensor<string, []> var_11230_equation_0 = const()[name = tensor<string, []>("op_11230_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11230_cast = einsum(equation = var_11230_equation_0, values = (var_11072_cast, var_10989_cast))[name = tensor<string, []>("op_11230_cast")];
            tensor<fp16, []> var_11231_to_fp16 = const()[name = tensor<string, []>("op_11231_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_887_cast = mul(x = var_11230_cast, y = var_11231_to_fp16)[name = tensor<string, []>("aw_887_cast")];
            tensor<string, []> var_11234_equation_0 = const()[name = tensor<string, []>("op_11234_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11234_cast = einsum(equation = var_11234_equation_0, values = (var_11076_cast, var_10993_cast))[name = tensor<string, []>("op_11234_cast")];
            tensor<fp16, []> var_11235_to_fp16 = const()[name = tensor<string, []>("op_11235_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_889_cast = mul(x = var_11234_cast, y = var_11235_to_fp16)[name = tensor<string, []>("aw_889_cast")];
            tensor<string, []> var_11238_equation_0 = const()[name = tensor<string, []>("op_11238_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11238_cast = einsum(equation = var_11238_equation_0, values = (var_11080_cast, var_10997_cast))[name = tensor<string, []>("op_11238_cast")];
            tensor<fp16, []> var_11239_to_fp16 = const()[name = tensor<string, []>("op_11239_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_891_cast = mul(x = var_11238_cast, y = var_11239_to_fp16)[name = tensor<string, []>("aw_891_cast")];
            tensor<string, []> var_11242_equation_0 = const()[name = tensor<string, []>("op_11242_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11242_cast = einsum(equation = var_11242_equation_0, values = (var_11084_cast, var_11001_cast))[name = tensor<string, []>("op_11242_cast")];
            tensor<fp16, []> var_11243_to_fp16 = const()[name = tensor<string, []>("op_11243_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_893_cast = mul(x = var_11242_cast, y = var_11243_to_fp16)[name = tensor<string, []>("aw_893_cast")];
            tensor<string, []> var_11246_equation_0 = const()[name = tensor<string, []>("op_11246_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11246_cast = einsum(equation = var_11246_equation_0, values = (var_11088_cast, var_11005_cast))[name = tensor<string, []>("op_11246_cast")];
            tensor<fp16, []> var_11247_to_fp16 = const()[name = tensor<string, []>("op_11247_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_895_cast = mul(x = var_11246_cast, y = var_11247_to_fp16)[name = tensor<string, []>("aw_895_cast")];
            tensor<string, []> var_11250_equation_0 = const()[name = tensor<string, []>("op_11250_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11250_cast = einsum(equation = var_11250_equation_0, values = (var_11092_cast, var_11009_cast))[name = tensor<string, []>("op_11250_cast")];
            tensor<fp16, []> var_11251_to_fp16 = const()[name = tensor<string, []>("op_11251_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_897_cast = mul(x = var_11250_cast, y = var_11251_to_fp16)[name = tensor<string, []>("aw_897_cast")];
            tensor<string, []> var_11254_equation_0 = const()[name = tensor<string, []>("op_11254_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11254_cast = einsum(equation = var_11254_equation_0, values = (var_11096_cast, var_11013_cast))[name = tensor<string, []>("op_11254_cast")];
            tensor<fp16, []> var_11255_to_fp16 = const()[name = tensor<string, []>("op_11255_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_899_cast = mul(x = var_11254_cast, y = var_11255_to_fp16)[name = tensor<string, []>("aw_899_cast")];
            tensor<string, []> var_11258_equation_0 = const()[name = tensor<string, []>("op_11258_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11258_cast = einsum(equation = var_11258_equation_0, values = (var_11100_cast, var_11017_cast))[name = tensor<string, []>("op_11258_cast")];
            tensor<fp16, []> var_11259_to_fp16 = const()[name = tensor<string, []>("op_11259_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_901_cast = mul(x = var_11258_cast, y = var_11259_to_fp16)[name = tensor<string, []>("aw_901_cast")];
            tensor<string, []> var_11262_equation_0 = const()[name = tensor<string, []>("op_11262_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11262_cast = einsum(equation = var_11262_equation_0, values = (var_11104_cast, var_11021_cast))[name = tensor<string, []>("op_11262_cast")];
            tensor<fp16, []> var_11263_to_fp16 = const()[name = tensor<string, []>("op_11263_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_903_cast = mul(x = var_11262_cast, y = var_11263_to_fp16)[name = tensor<string, []>("aw_903_cast")];
            tensor<string, []> var_11266_equation_0 = const()[name = tensor<string, []>("op_11266_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11266_cast = einsum(equation = var_11266_equation_0, values = (var_11108_cast, var_11025_cast))[name = tensor<string, []>("op_11266_cast")];
            tensor<fp16, []> var_11267_to_fp16 = const()[name = tensor<string, []>("op_11267_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_905_cast = mul(x = var_11266_cast, y = var_11267_to_fp16)[name = tensor<string, []>("aw_905_cast")];
            tensor<string, []> var_11270_equation_0 = const()[name = tensor<string, []>("op_11270_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11270_cast = einsum(equation = var_11270_equation_0, values = (var_11112_cast, var_11029_cast))[name = tensor<string, []>("op_11270_cast")];
            tensor<fp16, []> var_11271_to_fp16 = const()[name = tensor<string, []>("op_11271_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_907_cast = mul(x = var_11270_cast, y = var_11271_to_fp16)[name = tensor<string, []>("aw_907_cast")];
            tensor<string, []> var_11274_equation_0 = const()[name = tensor<string, []>("op_11274_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11274_cast = einsum(equation = var_11274_equation_0, values = (var_11116_cast, var_11033_cast))[name = tensor<string, []>("op_11274_cast")];
            tensor<fp16, []> var_11275_to_fp16 = const()[name = tensor<string, []>("op_11275_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_909_cast = mul(x = var_11274_cast, y = var_11275_to_fp16)[name = tensor<string, []>("aw_909_cast")];
            tensor<string, []> var_11278_equation_0 = const()[name = tensor<string, []>("op_11278_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11278_cast = einsum(equation = var_11278_equation_0, values = (var_11120_cast, var_11037_cast))[name = tensor<string, []>("op_11278_cast")];
            tensor<fp16, []> var_11279_to_fp16 = const()[name = tensor<string, []>("op_11279_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_911_cast = mul(x = var_11278_cast, y = var_11279_to_fp16)[name = tensor<string, []>("aw_911_cast")];
            tensor<string, []> var_11282_equation_0 = const()[name = tensor<string, []>("op_11282_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11282_cast = einsum(equation = var_11282_equation_0, values = (var_11124_cast, var_11041_cast))[name = tensor<string, []>("op_11282_cast")];
            tensor<fp16, []> var_11283_to_fp16 = const()[name = tensor<string, []>("op_11283_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_913_cast = mul(x = var_11282_cast, y = var_11283_to_fp16)[name = tensor<string, []>("aw_913_cast")];
            tensor<string, []> var_11286_equation_0 = const()[name = tensor<string, []>("op_11286_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11286_cast = einsum(equation = var_11286_equation_0, values = (var_11128_cast, var_11045_cast))[name = tensor<string, []>("op_11286_cast")];
            tensor<fp16, []> var_11287_to_fp16 = const()[name = tensor<string, []>("op_11287_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_915_cast = mul(x = var_11286_cast, y = var_11287_to_fp16)[name = tensor<string, []>("aw_915_cast")];
            tensor<string, []> var_11290_equation_0 = const()[name = tensor<string, []>("op_11290_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11290_cast = einsum(equation = var_11290_equation_0, values = (var_11132_cast, var_11049_cast))[name = tensor<string, []>("op_11290_cast")];
            tensor<fp16, []> var_11291_to_fp16 = const()[name = tensor<string, []>("op_11291_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_917_cast = mul(x = var_11290_cast, y = var_11291_to_fp16)[name = tensor<string, []>("aw_917_cast")];
            tensor<string, []> var_11294_equation_0 = const()[name = tensor<string, []>("op_11294_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11294_cast = einsum(equation = var_11294_equation_0, values = (var_11136_cast, var_11053_cast))[name = tensor<string, []>("op_11294_cast")];
            tensor<fp16, []> var_11295_to_fp16 = const()[name = tensor<string, []>("op_11295_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_919_cast = mul(x = var_11294_cast, y = var_11295_to_fp16)[name = tensor<string, []>("aw_919_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11297_cast = softmax(axis = var_2634, x = aw_881_cast)[name = tensor<string, []>("op_11297_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11298_cast = softmax(axis = var_2634, x = aw_883_cast)[name = tensor<string, []>("op_11298_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11299_cast = softmax(axis = var_2634, x = aw_885_cast)[name = tensor<string, []>("op_11299_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11300_cast = softmax(axis = var_2634, x = aw_887_cast)[name = tensor<string, []>("op_11300_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11301_cast = softmax(axis = var_2634, x = aw_889_cast)[name = tensor<string, []>("op_11301_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11302_cast = softmax(axis = var_2634, x = aw_891_cast)[name = tensor<string, []>("op_11302_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11303_cast = softmax(axis = var_2634, x = aw_893_cast)[name = tensor<string, []>("op_11303_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11304_cast = softmax(axis = var_2634, x = aw_895_cast)[name = tensor<string, []>("op_11304_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11305_cast = softmax(axis = var_2634, x = aw_897_cast)[name = tensor<string, []>("op_11305_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11306_cast = softmax(axis = var_2634, x = aw_899_cast)[name = tensor<string, []>("op_11306_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11307_cast = softmax(axis = var_2634, x = aw_901_cast)[name = tensor<string, []>("op_11307_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11308_cast = softmax(axis = var_2634, x = aw_903_cast)[name = tensor<string, []>("op_11308_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11309_cast = softmax(axis = var_2634, x = aw_905_cast)[name = tensor<string, []>("op_11309_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11310_cast = softmax(axis = var_2634, x = aw_907_cast)[name = tensor<string, []>("op_11310_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11311_cast = softmax(axis = var_2634, x = aw_909_cast)[name = tensor<string, []>("op_11311_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11312_cast = softmax(axis = var_2634, x = aw_911_cast)[name = tensor<string, []>("op_11312_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11313_cast = softmax(axis = var_2634, x = aw_913_cast)[name = tensor<string, []>("op_11313_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11314_cast = softmax(axis = var_2634, x = aw_915_cast)[name = tensor<string, []>("op_11314_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11315_cast = softmax(axis = var_2634, x = aw_917_cast)[name = tensor<string, []>("op_11315_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_11316_cast = softmax(axis = var_2634, x = aw_919_cast)[name = tensor<string, []>("op_11316_cast")];
            tensor<string, []> var_11318_equation_0 = const()[name = tensor<string, []>("op_11318_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11318_cast = einsum(equation = var_11318_equation_0, values = (var_11138_cast, var_11297_cast))[name = tensor<string, []>("op_11318_cast")];
            tensor<string, []> var_11320_equation_0 = const()[name = tensor<string, []>("op_11320_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11320_cast = einsum(equation = var_11320_equation_0, values = (var_11142_cast, var_11298_cast))[name = tensor<string, []>("op_11320_cast")];
            tensor<string, []> var_11322_equation_0 = const()[name = tensor<string, []>("op_11322_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11322_cast = einsum(equation = var_11322_equation_0, values = (var_11146_cast, var_11299_cast))[name = tensor<string, []>("op_11322_cast")];
            tensor<string, []> var_11324_equation_0 = const()[name = tensor<string, []>("op_11324_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11324_cast = einsum(equation = var_11324_equation_0, values = (var_11150_cast, var_11300_cast))[name = tensor<string, []>("op_11324_cast")];
            tensor<string, []> var_11326_equation_0 = const()[name = tensor<string, []>("op_11326_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11326_cast = einsum(equation = var_11326_equation_0, values = (var_11154_cast, var_11301_cast))[name = tensor<string, []>("op_11326_cast")];
            tensor<string, []> var_11328_equation_0 = const()[name = tensor<string, []>("op_11328_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11328_cast = einsum(equation = var_11328_equation_0, values = (var_11158_cast, var_11302_cast))[name = tensor<string, []>("op_11328_cast")];
            tensor<string, []> var_11330_equation_0 = const()[name = tensor<string, []>("op_11330_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11330_cast = einsum(equation = var_11330_equation_0, values = (var_11162_cast, var_11303_cast))[name = tensor<string, []>("op_11330_cast")];
            tensor<string, []> var_11332_equation_0 = const()[name = tensor<string, []>("op_11332_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11332_cast = einsum(equation = var_11332_equation_0, values = (var_11166_cast, var_11304_cast))[name = tensor<string, []>("op_11332_cast")];
            tensor<string, []> var_11334_equation_0 = const()[name = tensor<string, []>("op_11334_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11334_cast = einsum(equation = var_11334_equation_0, values = (var_11170_cast, var_11305_cast))[name = tensor<string, []>("op_11334_cast")];
            tensor<string, []> var_11336_equation_0 = const()[name = tensor<string, []>("op_11336_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11336_cast = einsum(equation = var_11336_equation_0, values = (var_11174_cast, var_11306_cast))[name = tensor<string, []>("op_11336_cast")];
            tensor<string, []> var_11338_equation_0 = const()[name = tensor<string, []>("op_11338_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11338_cast = einsum(equation = var_11338_equation_0, values = (var_11178_cast, var_11307_cast))[name = tensor<string, []>("op_11338_cast")];
            tensor<string, []> var_11340_equation_0 = const()[name = tensor<string, []>("op_11340_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11340_cast = einsum(equation = var_11340_equation_0, values = (var_11182_cast, var_11308_cast))[name = tensor<string, []>("op_11340_cast")];
            tensor<string, []> var_11342_equation_0 = const()[name = tensor<string, []>("op_11342_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11342_cast = einsum(equation = var_11342_equation_0, values = (var_11186_cast, var_11309_cast))[name = tensor<string, []>("op_11342_cast")];
            tensor<string, []> var_11344_equation_0 = const()[name = tensor<string, []>("op_11344_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11344_cast = einsum(equation = var_11344_equation_0, values = (var_11190_cast, var_11310_cast))[name = tensor<string, []>("op_11344_cast")];
            tensor<string, []> var_11346_equation_0 = const()[name = tensor<string, []>("op_11346_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11346_cast = einsum(equation = var_11346_equation_0, values = (var_11194_cast, var_11311_cast))[name = tensor<string, []>("op_11346_cast")];
            tensor<string, []> var_11348_equation_0 = const()[name = tensor<string, []>("op_11348_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11348_cast = einsum(equation = var_11348_equation_0, values = (var_11198_cast, var_11312_cast))[name = tensor<string, []>("op_11348_cast")];
            tensor<string, []> var_11350_equation_0 = const()[name = tensor<string, []>("op_11350_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11350_cast = einsum(equation = var_11350_equation_0, values = (var_11202_cast, var_11313_cast))[name = tensor<string, []>("op_11350_cast")];
            tensor<string, []> var_11352_equation_0 = const()[name = tensor<string, []>("op_11352_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11352_cast = einsum(equation = var_11352_equation_0, values = (var_11206_cast, var_11314_cast))[name = tensor<string, []>("op_11352_cast")];
            tensor<string, []> var_11354_equation_0 = const()[name = tensor<string, []>("op_11354_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11354_cast = einsum(equation = var_11354_equation_0, values = (var_11210_cast, var_11315_cast))[name = tensor<string, []>("op_11354_cast")];
            tensor<string, []> var_11356_equation_0 = const()[name = tensor<string, []>("op_11356_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11356_cast = einsum(equation = var_11356_equation_0, values = (var_11214_cast, var_11316_cast))[name = tensor<string, []>("op_11356_cast")];
            tensor<bool, []> input_207_interleave_0 = const()[name = tensor<string, []>("input_207_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_207_cast = concat(axis = var_2634, interleave = input_207_interleave_0, values = (var_11318_cast, var_11320_cast, var_11322_cast, var_11324_cast, var_11326_cast, var_11328_cast, var_11330_cast, var_11332_cast, var_11334_cast, var_11336_cast, var_11338_cast, var_11340_cast, var_11342_cast, var_11344_cast, var_11346_cast, var_11348_cast, var_11350_cast, var_11352_cast, var_11354_cast, var_11356_cast))[name = tensor<string, []>("input_207_cast")];
            tensor<int32, [2]> var_11362 = const()[name = tensor<string, []>("op_11362"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11364 = const()[name = tensor<string, []>("op_11364"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11366_pad_type_0 = const()[name = tensor<string, []>("op_11366_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11366_pad_0 = const()[name = tensor<string, []>("op_11366_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(833618368)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(836895232)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_11366_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16, dilations = var_11364, groups = var_2634, pad = var_11366_pad_0, pad_type = var_11366_pad_type_0, strides = var_11362, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16, x = input_207_cast)[name = tensor<string, []>("op_11366_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_81_cast = add(x = var_11366_cast, y = inputs_79_cast)[name = tensor<string, []>("inputs_81_cast")];
            tensor<int32, [1]> var_11370 = const()[name = tensor<string, []>("op_11370"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_81_cast = reduce_mean(axes = var_11370, keep_dims = var_2629, x = inputs_81_cast)[name = tensor<string, []>("channels_mean_81_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_81_cast = sub(x = inputs_81_cast, y = channels_mean_81_cast)[name = tensor<string, []>("zero_mean_81_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_81_cast = mul(x = zero_mean_81_cast, y = zero_mean_81_cast)[name = tensor<string, []>("zero_mean_sq_81_cast")];
            tensor<int32, [1]> var_11374 = const()[name = tensor<string, []>("op_11374"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_11375_cast = reduce_mean(axes = var_11374, keep_dims = var_2629, x = zero_mean_sq_81_cast)[name = tensor<string, []>("op_11375_cast")];
            tensor<fp16, []> var_11376_to_fp16 = const()[name = tensor<string, []>("op_11376_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_11377_cast = add(x = var_11375_cast, y = var_11376_to_fp16)[name = tensor<string, []>("op_11377_cast")];
            tensor<fp16, []> denom_81_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_81_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_81_cast = rsqrt(epsilon = denom_81_epsilon_0_to_fp16, x = var_11377_cast)[name = tensor<string, []>("denom_81_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_81_cast = mul(x = zero_mean_81_cast, y = denom_81_cast)[name = tensor<string, []>("out_81_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_11381_to_fp16 = const()[name = tensor<string, []>("op_11381_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(836897856)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_11382_cast = add(x = out_81_cast, y = var_11381_to_fp16)[name = tensor<string, []>("op_11382_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_11384_to_fp16 = const()[name = tensor<string, []>("op_11384_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(836900480)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_121_cast = mul(x = var_11382_cast, y = var_11384_to_fp16)[name = tensor<string, []>("hidden_states_121_cast")];
            tensor<int32, [2]> var_11391 = const()[name = tensor<string, []>("op_11391"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11393 = const()[name = tensor<string, []>("op_11393"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_55_pad_type_0 = const()[name = tensor<string, []>("q_55_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_55_pad_0 = const()[name = tensor<string, []>("q_55_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(836903104)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_55_cast = conv(dilations = var_11393, groups = var_2634, pad = q_55_pad_0, pad_type = q_55_pad_type_0, strides = var_11391, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16, x = hidden_states_121_cast)[name = tensor<string, []>("q_55_cast")];
            tensor<int32, [2]> var_11397 = const()[name = tensor<string, []>("op_11397"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11399 = const()[name = tensor<string, []>("op_11399"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_109_pad_type_0 = const()[name = tensor<string, []>("k_109_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_109_pad_0 = const()[name = tensor<string, []>("k_109_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(840179968)))];
            tensor<fp16, [2, 1280, 1, 77]> k_109_cast = conv(dilations = var_11399, groups = var_2634, pad = k_109_pad_0, pad_type = k_109_pad_type_0, strides = var_11397, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_109_cast")];
            tensor<int32, [2]> var_11403 = const()[name = tensor<string, []>("op_11403"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11405 = const()[name = tensor<string, []>("op_11405"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_55_pad_type_0 = const()[name = tensor<string, []>("v_55_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_55_pad_0 = const()[name = tensor<string, []>("v_55_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(845422912)))];
            tensor<fp16, [2, 1280, 1, 77]> v_55_cast = conv(dilations = var_11405, groups = var_2634, pad = v_55_pad_0, pad_type = v_55_pad_type_0, strides = var_11403, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_55_cast")];
            tensor<int32, [4]> var_11409_begin_0 = const()[name = tensor<string, []>("op_11409_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11409_end_0 = const()[name = tensor<string, []>("op_11409_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_11409_end_mask_0 = const()[name = tensor<string, []>("op_11409_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11409_cast = slice_by_index(begin = var_11409_begin_0, end = var_11409_end_0, end_mask = var_11409_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11409_cast")];
            tensor<int32, [4]> var_11413_begin_0 = const()[name = tensor<string, []>("op_11413_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_11413_end_0 = const()[name = tensor<string, []>("op_11413_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_11413_end_mask_0 = const()[name = tensor<string, []>("op_11413_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11413_cast = slice_by_index(begin = var_11413_begin_0, end = var_11413_end_0, end_mask = var_11413_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11413_cast")];
            tensor<int32, [4]> var_11417_begin_0 = const()[name = tensor<string, []>("op_11417_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_11417_end_0 = const()[name = tensor<string, []>("op_11417_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_11417_end_mask_0 = const()[name = tensor<string, []>("op_11417_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11417_cast = slice_by_index(begin = var_11417_begin_0, end = var_11417_end_0, end_mask = var_11417_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11417_cast")];
            tensor<int32, [4]> var_11421_begin_0 = const()[name = tensor<string, []>("op_11421_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_11421_end_0 = const()[name = tensor<string, []>("op_11421_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_11421_end_mask_0 = const()[name = tensor<string, []>("op_11421_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11421_cast = slice_by_index(begin = var_11421_begin_0, end = var_11421_end_0, end_mask = var_11421_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11421_cast")];
            tensor<int32, [4]> var_11425_begin_0 = const()[name = tensor<string, []>("op_11425_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_11425_end_0 = const()[name = tensor<string, []>("op_11425_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_11425_end_mask_0 = const()[name = tensor<string, []>("op_11425_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11425_cast = slice_by_index(begin = var_11425_begin_0, end = var_11425_end_0, end_mask = var_11425_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11425_cast")];
            tensor<int32, [4]> var_11429_begin_0 = const()[name = tensor<string, []>("op_11429_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_11429_end_0 = const()[name = tensor<string, []>("op_11429_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_11429_end_mask_0 = const()[name = tensor<string, []>("op_11429_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11429_cast = slice_by_index(begin = var_11429_begin_0, end = var_11429_end_0, end_mask = var_11429_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11429_cast")];
            tensor<int32, [4]> var_11433_begin_0 = const()[name = tensor<string, []>("op_11433_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_11433_end_0 = const()[name = tensor<string, []>("op_11433_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_11433_end_mask_0 = const()[name = tensor<string, []>("op_11433_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11433_cast = slice_by_index(begin = var_11433_begin_0, end = var_11433_end_0, end_mask = var_11433_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11433_cast")];
            tensor<int32, [4]> var_11437_begin_0 = const()[name = tensor<string, []>("op_11437_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_11437_end_0 = const()[name = tensor<string, []>("op_11437_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_11437_end_mask_0 = const()[name = tensor<string, []>("op_11437_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11437_cast = slice_by_index(begin = var_11437_begin_0, end = var_11437_end_0, end_mask = var_11437_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11437_cast")];
            tensor<int32, [4]> var_11441_begin_0 = const()[name = tensor<string, []>("op_11441_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_11441_end_0 = const()[name = tensor<string, []>("op_11441_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_11441_end_mask_0 = const()[name = tensor<string, []>("op_11441_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11441_cast = slice_by_index(begin = var_11441_begin_0, end = var_11441_end_0, end_mask = var_11441_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11441_cast")];
            tensor<int32, [4]> var_11445_begin_0 = const()[name = tensor<string, []>("op_11445_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_11445_end_0 = const()[name = tensor<string, []>("op_11445_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_11445_end_mask_0 = const()[name = tensor<string, []>("op_11445_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11445_cast = slice_by_index(begin = var_11445_begin_0, end = var_11445_end_0, end_mask = var_11445_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11445_cast")];
            tensor<int32, [4]> var_11449_begin_0 = const()[name = tensor<string, []>("op_11449_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_11449_end_0 = const()[name = tensor<string, []>("op_11449_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_11449_end_mask_0 = const()[name = tensor<string, []>("op_11449_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11449_cast = slice_by_index(begin = var_11449_begin_0, end = var_11449_end_0, end_mask = var_11449_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11449_cast")];
            tensor<int32, [4]> var_11453_begin_0 = const()[name = tensor<string, []>("op_11453_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_11453_end_0 = const()[name = tensor<string, []>("op_11453_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_11453_end_mask_0 = const()[name = tensor<string, []>("op_11453_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11453_cast = slice_by_index(begin = var_11453_begin_0, end = var_11453_end_0, end_mask = var_11453_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11453_cast")];
            tensor<int32, [4]> var_11457_begin_0 = const()[name = tensor<string, []>("op_11457_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_11457_end_0 = const()[name = tensor<string, []>("op_11457_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_11457_end_mask_0 = const()[name = tensor<string, []>("op_11457_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11457_cast = slice_by_index(begin = var_11457_begin_0, end = var_11457_end_0, end_mask = var_11457_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11457_cast")];
            tensor<int32, [4]> var_11461_begin_0 = const()[name = tensor<string, []>("op_11461_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_11461_end_0 = const()[name = tensor<string, []>("op_11461_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_11461_end_mask_0 = const()[name = tensor<string, []>("op_11461_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11461_cast = slice_by_index(begin = var_11461_begin_0, end = var_11461_end_0, end_mask = var_11461_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11461_cast")];
            tensor<int32, [4]> var_11465_begin_0 = const()[name = tensor<string, []>("op_11465_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_11465_end_0 = const()[name = tensor<string, []>("op_11465_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_11465_end_mask_0 = const()[name = tensor<string, []>("op_11465_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11465_cast = slice_by_index(begin = var_11465_begin_0, end = var_11465_end_0, end_mask = var_11465_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11465_cast")];
            tensor<int32, [4]> var_11469_begin_0 = const()[name = tensor<string, []>("op_11469_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_11469_end_0 = const()[name = tensor<string, []>("op_11469_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_11469_end_mask_0 = const()[name = tensor<string, []>("op_11469_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11469_cast = slice_by_index(begin = var_11469_begin_0, end = var_11469_end_0, end_mask = var_11469_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11469_cast")];
            tensor<int32, [4]> var_11473_begin_0 = const()[name = tensor<string, []>("op_11473_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_11473_end_0 = const()[name = tensor<string, []>("op_11473_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_11473_end_mask_0 = const()[name = tensor<string, []>("op_11473_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11473_cast = slice_by_index(begin = var_11473_begin_0, end = var_11473_end_0, end_mask = var_11473_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11473_cast")];
            tensor<int32, [4]> var_11477_begin_0 = const()[name = tensor<string, []>("op_11477_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_11477_end_0 = const()[name = tensor<string, []>("op_11477_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_11477_end_mask_0 = const()[name = tensor<string, []>("op_11477_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11477_cast = slice_by_index(begin = var_11477_begin_0, end = var_11477_end_0, end_mask = var_11477_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11477_cast")];
            tensor<int32, [4]> var_11481_begin_0 = const()[name = tensor<string, []>("op_11481_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_11481_end_0 = const()[name = tensor<string, []>("op_11481_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_11481_end_mask_0 = const()[name = tensor<string, []>("op_11481_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11481_cast = slice_by_index(begin = var_11481_begin_0, end = var_11481_end_0, end_mask = var_11481_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11481_cast")];
            tensor<int32, [4]> var_11485_begin_0 = const()[name = tensor<string, []>("op_11485_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_11485_end_0 = const()[name = tensor<string, []>("op_11485_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_11485_end_mask_0 = const()[name = tensor<string, []>("op_11485_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11485_cast = slice_by_index(begin = var_11485_begin_0, end = var_11485_end_0, end_mask = var_11485_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11485_cast")];
            tensor<int32, [4]> k_111_perm_0 = const()[name = tensor<string, []>("k_111_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_11492_begin_0 = const()[name = tensor<string, []>("op_11492_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11492_end_0 = const()[name = tensor<string, []>("op_11492_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_11492_end_mask_0 = const()[name = tensor<string, []>("op_11492_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_112 = transpose(perm = k_111_perm_0, x = k_109_cast)[name = tensor<string, []>("transpose_112")];
            tensor<fp16, [2, 77, 1, 64]> var_11492_cast = slice_by_index(begin = var_11492_begin_0, end = var_11492_end_0, end_mask = var_11492_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11492_cast")];
            tensor<int32, [4]> var_11496_begin_0 = const()[name = tensor<string, []>("op_11496_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_11496_end_0 = const()[name = tensor<string, []>("op_11496_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_11496_end_mask_0 = const()[name = tensor<string, []>("op_11496_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11496_cast = slice_by_index(begin = var_11496_begin_0, end = var_11496_end_0, end_mask = var_11496_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11496_cast")];
            tensor<int32, [4]> var_11500_begin_0 = const()[name = tensor<string, []>("op_11500_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_11500_end_0 = const()[name = tensor<string, []>("op_11500_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_11500_end_mask_0 = const()[name = tensor<string, []>("op_11500_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11500_cast = slice_by_index(begin = var_11500_begin_0, end = var_11500_end_0, end_mask = var_11500_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11500_cast")];
            tensor<int32, [4]> var_11504_begin_0 = const()[name = tensor<string, []>("op_11504_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_11504_end_0 = const()[name = tensor<string, []>("op_11504_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_11504_end_mask_0 = const()[name = tensor<string, []>("op_11504_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11504_cast = slice_by_index(begin = var_11504_begin_0, end = var_11504_end_0, end_mask = var_11504_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11504_cast")];
            tensor<int32, [4]> var_11508_begin_0 = const()[name = tensor<string, []>("op_11508_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_11508_end_0 = const()[name = tensor<string, []>("op_11508_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_11508_end_mask_0 = const()[name = tensor<string, []>("op_11508_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11508_cast = slice_by_index(begin = var_11508_begin_0, end = var_11508_end_0, end_mask = var_11508_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11508_cast")];
            tensor<int32, [4]> var_11512_begin_0 = const()[name = tensor<string, []>("op_11512_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_11512_end_0 = const()[name = tensor<string, []>("op_11512_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_11512_end_mask_0 = const()[name = tensor<string, []>("op_11512_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11512_cast = slice_by_index(begin = var_11512_begin_0, end = var_11512_end_0, end_mask = var_11512_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11512_cast")];
            tensor<int32, [4]> var_11516_begin_0 = const()[name = tensor<string, []>("op_11516_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_11516_end_0 = const()[name = tensor<string, []>("op_11516_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_11516_end_mask_0 = const()[name = tensor<string, []>("op_11516_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11516_cast = slice_by_index(begin = var_11516_begin_0, end = var_11516_end_0, end_mask = var_11516_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11516_cast")];
            tensor<int32, [4]> var_11520_begin_0 = const()[name = tensor<string, []>("op_11520_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_11520_end_0 = const()[name = tensor<string, []>("op_11520_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_11520_end_mask_0 = const()[name = tensor<string, []>("op_11520_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11520_cast = slice_by_index(begin = var_11520_begin_0, end = var_11520_end_0, end_mask = var_11520_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11520_cast")];
            tensor<int32, [4]> var_11524_begin_0 = const()[name = tensor<string, []>("op_11524_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11524_end_0 = const()[name = tensor<string, []>("op_11524_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_11524_end_mask_0 = const()[name = tensor<string, []>("op_11524_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11524_cast = slice_by_index(begin = var_11524_begin_0, end = var_11524_end_0, end_mask = var_11524_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11524_cast")];
            tensor<int32, [4]> var_11528_begin_0 = const()[name = tensor<string, []>("op_11528_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_11528_end_0 = const()[name = tensor<string, []>("op_11528_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_11528_end_mask_0 = const()[name = tensor<string, []>("op_11528_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11528_cast = slice_by_index(begin = var_11528_begin_0, end = var_11528_end_0, end_mask = var_11528_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11528_cast")];
            tensor<int32, [4]> var_11532_begin_0 = const()[name = tensor<string, []>("op_11532_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_11532_end_0 = const()[name = tensor<string, []>("op_11532_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_11532_end_mask_0 = const()[name = tensor<string, []>("op_11532_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11532_cast = slice_by_index(begin = var_11532_begin_0, end = var_11532_end_0, end_mask = var_11532_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11532_cast")];
            tensor<int32, [4]> var_11536_begin_0 = const()[name = tensor<string, []>("op_11536_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_11536_end_0 = const()[name = tensor<string, []>("op_11536_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_11536_end_mask_0 = const()[name = tensor<string, []>("op_11536_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11536_cast = slice_by_index(begin = var_11536_begin_0, end = var_11536_end_0, end_mask = var_11536_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11536_cast")];
            tensor<int32, [4]> var_11540_begin_0 = const()[name = tensor<string, []>("op_11540_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_11540_end_0 = const()[name = tensor<string, []>("op_11540_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_11540_end_mask_0 = const()[name = tensor<string, []>("op_11540_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11540_cast = slice_by_index(begin = var_11540_begin_0, end = var_11540_end_0, end_mask = var_11540_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11540_cast")];
            tensor<int32, [4]> var_11544_begin_0 = const()[name = tensor<string, []>("op_11544_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_11544_end_0 = const()[name = tensor<string, []>("op_11544_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_11544_end_mask_0 = const()[name = tensor<string, []>("op_11544_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11544_cast = slice_by_index(begin = var_11544_begin_0, end = var_11544_end_0, end_mask = var_11544_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11544_cast")];
            tensor<int32, [4]> var_11548_begin_0 = const()[name = tensor<string, []>("op_11548_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_11548_end_0 = const()[name = tensor<string, []>("op_11548_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_11548_end_mask_0 = const()[name = tensor<string, []>("op_11548_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11548_cast = slice_by_index(begin = var_11548_begin_0, end = var_11548_end_0, end_mask = var_11548_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11548_cast")];
            tensor<int32, [4]> var_11552_begin_0 = const()[name = tensor<string, []>("op_11552_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_11552_end_0 = const()[name = tensor<string, []>("op_11552_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_11552_end_mask_0 = const()[name = tensor<string, []>("op_11552_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11552_cast = slice_by_index(begin = var_11552_begin_0, end = var_11552_end_0, end_mask = var_11552_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11552_cast")];
            tensor<int32, [4]> var_11556_begin_0 = const()[name = tensor<string, []>("op_11556_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_11556_end_0 = const()[name = tensor<string, []>("op_11556_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_11556_end_mask_0 = const()[name = tensor<string, []>("op_11556_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11556_cast = slice_by_index(begin = var_11556_begin_0, end = var_11556_end_0, end_mask = var_11556_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11556_cast")];
            tensor<int32, [4]> var_11560_begin_0 = const()[name = tensor<string, []>("op_11560_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_11560_end_0 = const()[name = tensor<string, []>("op_11560_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_11560_end_mask_0 = const()[name = tensor<string, []>("op_11560_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11560_cast = slice_by_index(begin = var_11560_begin_0, end = var_11560_end_0, end_mask = var_11560_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11560_cast")];
            tensor<int32, [4]> var_11564_begin_0 = const()[name = tensor<string, []>("op_11564_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_11564_end_0 = const()[name = tensor<string, []>("op_11564_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_11564_end_mask_0 = const()[name = tensor<string, []>("op_11564_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11564_cast = slice_by_index(begin = var_11564_begin_0, end = var_11564_end_0, end_mask = var_11564_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11564_cast")];
            tensor<int32, [4]> var_11568_begin_0 = const()[name = tensor<string, []>("op_11568_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_11568_end_0 = const()[name = tensor<string, []>("op_11568_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_11568_end_mask_0 = const()[name = tensor<string, []>("op_11568_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11568_cast = slice_by_index(begin = var_11568_begin_0, end = var_11568_end_0, end_mask = var_11568_end_mask_0, x = transpose_112)[name = tensor<string, []>("op_11568_cast")];
            tensor<int32, [4]> var_11570_begin_0 = const()[name = tensor<string, []>("op_11570_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11570_end_0 = const()[name = tensor<string, []>("op_11570_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_11570_end_mask_0 = const()[name = tensor<string, []>("op_11570_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11570_cast = slice_by_index(begin = var_11570_begin_0, end = var_11570_end_0, end_mask = var_11570_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11570_cast")];
            tensor<int32, [4]> var_11574_begin_0 = const()[name = tensor<string, []>("op_11574_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_11574_end_0 = const()[name = tensor<string, []>("op_11574_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_11574_end_mask_0 = const()[name = tensor<string, []>("op_11574_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11574_cast = slice_by_index(begin = var_11574_begin_0, end = var_11574_end_0, end_mask = var_11574_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11574_cast")];
            tensor<int32, [4]> var_11578_begin_0 = const()[name = tensor<string, []>("op_11578_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_11578_end_0 = const()[name = tensor<string, []>("op_11578_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_11578_end_mask_0 = const()[name = tensor<string, []>("op_11578_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11578_cast = slice_by_index(begin = var_11578_begin_0, end = var_11578_end_0, end_mask = var_11578_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11578_cast")];
            tensor<int32, [4]> var_11582_begin_0 = const()[name = tensor<string, []>("op_11582_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_11582_end_0 = const()[name = tensor<string, []>("op_11582_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_11582_end_mask_0 = const()[name = tensor<string, []>("op_11582_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11582_cast = slice_by_index(begin = var_11582_begin_0, end = var_11582_end_0, end_mask = var_11582_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11582_cast")];
            tensor<int32, [4]> var_11586_begin_0 = const()[name = tensor<string, []>("op_11586_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_11586_end_0 = const()[name = tensor<string, []>("op_11586_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_11586_end_mask_0 = const()[name = tensor<string, []>("op_11586_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11586_cast = slice_by_index(begin = var_11586_begin_0, end = var_11586_end_0, end_mask = var_11586_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11586_cast")];
            tensor<int32, [4]> var_11590_begin_0 = const()[name = tensor<string, []>("op_11590_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_11590_end_0 = const()[name = tensor<string, []>("op_11590_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_11590_end_mask_0 = const()[name = tensor<string, []>("op_11590_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11590_cast = slice_by_index(begin = var_11590_begin_0, end = var_11590_end_0, end_mask = var_11590_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11590_cast")];
            tensor<int32, [4]> var_11594_begin_0 = const()[name = tensor<string, []>("op_11594_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_11594_end_0 = const()[name = tensor<string, []>("op_11594_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_11594_end_mask_0 = const()[name = tensor<string, []>("op_11594_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11594_cast = slice_by_index(begin = var_11594_begin_0, end = var_11594_end_0, end_mask = var_11594_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11594_cast")];
            tensor<int32, [4]> var_11598_begin_0 = const()[name = tensor<string, []>("op_11598_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_11598_end_0 = const()[name = tensor<string, []>("op_11598_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_11598_end_mask_0 = const()[name = tensor<string, []>("op_11598_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11598_cast = slice_by_index(begin = var_11598_begin_0, end = var_11598_end_0, end_mask = var_11598_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11598_cast")];
            tensor<int32, [4]> var_11602_begin_0 = const()[name = tensor<string, []>("op_11602_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_11602_end_0 = const()[name = tensor<string, []>("op_11602_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_11602_end_mask_0 = const()[name = tensor<string, []>("op_11602_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11602_cast = slice_by_index(begin = var_11602_begin_0, end = var_11602_end_0, end_mask = var_11602_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11602_cast")];
            tensor<int32, [4]> var_11606_begin_0 = const()[name = tensor<string, []>("op_11606_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_11606_end_0 = const()[name = tensor<string, []>("op_11606_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_11606_end_mask_0 = const()[name = tensor<string, []>("op_11606_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11606_cast = slice_by_index(begin = var_11606_begin_0, end = var_11606_end_0, end_mask = var_11606_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11606_cast")];
            tensor<int32, [4]> var_11610_begin_0 = const()[name = tensor<string, []>("op_11610_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_11610_end_0 = const()[name = tensor<string, []>("op_11610_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_11610_end_mask_0 = const()[name = tensor<string, []>("op_11610_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11610_cast = slice_by_index(begin = var_11610_begin_0, end = var_11610_end_0, end_mask = var_11610_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11610_cast")];
            tensor<int32, [4]> var_11614_begin_0 = const()[name = tensor<string, []>("op_11614_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_11614_end_0 = const()[name = tensor<string, []>("op_11614_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_11614_end_mask_0 = const()[name = tensor<string, []>("op_11614_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11614_cast = slice_by_index(begin = var_11614_begin_0, end = var_11614_end_0, end_mask = var_11614_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11614_cast")];
            tensor<int32, [4]> var_11618_begin_0 = const()[name = tensor<string, []>("op_11618_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_11618_end_0 = const()[name = tensor<string, []>("op_11618_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_11618_end_mask_0 = const()[name = tensor<string, []>("op_11618_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11618_cast = slice_by_index(begin = var_11618_begin_0, end = var_11618_end_0, end_mask = var_11618_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11618_cast")];
            tensor<int32, [4]> var_11622_begin_0 = const()[name = tensor<string, []>("op_11622_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_11622_end_0 = const()[name = tensor<string, []>("op_11622_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_11622_end_mask_0 = const()[name = tensor<string, []>("op_11622_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11622_cast = slice_by_index(begin = var_11622_begin_0, end = var_11622_end_0, end_mask = var_11622_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11622_cast")];
            tensor<int32, [4]> var_11626_begin_0 = const()[name = tensor<string, []>("op_11626_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_11626_end_0 = const()[name = tensor<string, []>("op_11626_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_11626_end_mask_0 = const()[name = tensor<string, []>("op_11626_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11626_cast = slice_by_index(begin = var_11626_begin_0, end = var_11626_end_0, end_mask = var_11626_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11626_cast")];
            tensor<int32, [4]> var_11630_begin_0 = const()[name = tensor<string, []>("op_11630_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_11630_end_0 = const()[name = tensor<string, []>("op_11630_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_11630_end_mask_0 = const()[name = tensor<string, []>("op_11630_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11630_cast = slice_by_index(begin = var_11630_begin_0, end = var_11630_end_0, end_mask = var_11630_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11630_cast")];
            tensor<int32, [4]> var_11634_begin_0 = const()[name = tensor<string, []>("op_11634_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_11634_end_0 = const()[name = tensor<string, []>("op_11634_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_11634_end_mask_0 = const()[name = tensor<string, []>("op_11634_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11634_cast = slice_by_index(begin = var_11634_begin_0, end = var_11634_end_0, end_mask = var_11634_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11634_cast")];
            tensor<int32, [4]> var_11638_begin_0 = const()[name = tensor<string, []>("op_11638_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_11638_end_0 = const()[name = tensor<string, []>("op_11638_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_11638_end_mask_0 = const()[name = tensor<string, []>("op_11638_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11638_cast = slice_by_index(begin = var_11638_begin_0, end = var_11638_end_0, end_mask = var_11638_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11638_cast")];
            tensor<int32, [4]> var_11642_begin_0 = const()[name = tensor<string, []>("op_11642_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_11642_end_0 = const()[name = tensor<string, []>("op_11642_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_11642_end_mask_0 = const()[name = tensor<string, []>("op_11642_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11642_cast = slice_by_index(begin = var_11642_begin_0, end = var_11642_end_0, end_mask = var_11642_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11642_cast")];
            tensor<int32, [4]> var_11646_begin_0 = const()[name = tensor<string, []>("op_11646_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_11646_end_0 = const()[name = tensor<string, []>("op_11646_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_11646_end_mask_0 = const()[name = tensor<string, []>("op_11646_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11646_cast = slice_by_index(begin = var_11646_begin_0, end = var_11646_end_0, end_mask = var_11646_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11646_cast")];
            tensor<string, []> var_11650_equation_0 = const()[name = tensor<string, []>("op_11650_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11650_cast = einsum(equation = var_11650_equation_0, values = (var_11492_cast, var_11409_cast))[name = tensor<string, []>("op_11650_cast")];
            tensor<fp16, []> var_11651_to_fp16 = const()[name = tensor<string, []>("op_11651_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_921_cast = mul(x = var_11650_cast, y = var_11651_to_fp16)[name = tensor<string, []>("aw_921_cast")];
            tensor<string, []> var_11654_equation_0 = const()[name = tensor<string, []>("op_11654_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11654_cast = einsum(equation = var_11654_equation_0, values = (var_11496_cast, var_11413_cast))[name = tensor<string, []>("op_11654_cast")];
            tensor<fp16, []> var_11655_to_fp16 = const()[name = tensor<string, []>("op_11655_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_923_cast = mul(x = var_11654_cast, y = var_11655_to_fp16)[name = tensor<string, []>("aw_923_cast")];
            tensor<string, []> var_11658_equation_0 = const()[name = tensor<string, []>("op_11658_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11658_cast = einsum(equation = var_11658_equation_0, values = (var_11500_cast, var_11417_cast))[name = tensor<string, []>("op_11658_cast")];
            tensor<fp16, []> var_11659_to_fp16 = const()[name = tensor<string, []>("op_11659_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_925_cast = mul(x = var_11658_cast, y = var_11659_to_fp16)[name = tensor<string, []>("aw_925_cast")];
            tensor<string, []> var_11662_equation_0 = const()[name = tensor<string, []>("op_11662_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11662_cast = einsum(equation = var_11662_equation_0, values = (var_11504_cast, var_11421_cast))[name = tensor<string, []>("op_11662_cast")];
            tensor<fp16, []> var_11663_to_fp16 = const()[name = tensor<string, []>("op_11663_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_927_cast = mul(x = var_11662_cast, y = var_11663_to_fp16)[name = tensor<string, []>("aw_927_cast")];
            tensor<string, []> var_11666_equation_0 = const()[name = tensor<string, []>("op_11666_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11666_cast = einsum(equation = var_11666_equation_0, values = (var_11508_cast, var_11425_cast))[name = tensor<string, []>("op_11666_cast")];
            tensor<fp16, []> var_11667_to_fp16 = const()[name = tensor<string, []>("op_11667_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_929_cast = mul(x = var_11666_cast, y = var_11667_to_fp16)[name = tensor<string, []>("aw_929_cast")];
            tensor<string, []> var_11670_equation_0 = const()[name = tensor<string, []>("op_11670_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11670_cast = einsum(equation = var_11670_equation_0, values = (var_11512_cast, var_11429_cast))[name = tensor<string, []>("op_11670_cast")];
            tensor<fp16, []> var_11671_to_fp16 = const()[name = tensor<string, []>("op_11671_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_931_cast = mul(x = var_11670_cast, y = var_11671_to_fp16)[name = tensor<string, []>("aw_931_cast")];
            tensor<string, []> var_11674_equation_0 = const()[name = tensor<string, []>("op_11674_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11674_cast = einsum(equation = var_11674_equation_0, values = (var_11516_cast, var_11433_cast))[name = tensor<string, []>("op_11674_cast")];
            tensor<fp16, []> var_11675_to_fp16 = const()[name = tensor<string, []>("op_11675_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_933_cast = mul(x = var_11674_cast, y = var_11675_to_fp16)[name = tensor<string, []>("aw_933_cast")];
            tensor<string, []> var_11678_equation_0 = const()[name = tensor<string, []>("op_11678_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11678_cast = einsum(equation = var_11678_equation_0, values = (var_11520_cast, var_11437_cast))[name = tensor<string, []>("op_11678_cast")];
            tensor<fp16, []> var_11679_to_fp16 = const()[name = tensor<string, []>("op_11679_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_935_cast = mul(x = var_11678_cast, y = var_11679_to_fp16)[name = tensor<string, []>("aw_935_cast")];
            tensor<string, []> var_11682_equation_0 = const()[name = tensor<string, []>("op_11682_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11682_cast = einsum(equation = var_11682_equation_0, values = (var_11524_cast, var_11441_cast))[name = tensor<string, []>("op_11682_cast")];
            tensor<fp16, []> var_11683_to_fp16 = const()[name = tensor<string, []>("op_11683_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_937_cast = mul(x = var_11682_cast, y = var_11683_to_fp16)[name = tensor<string, []>("aw_937_cast")];
            tensor<string, []> var_11686_equation_0 = const()[name = tensor<string, []>("op_11686_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11686_cast = einsum(equation = var_11686_equation_0, values = (var_11528_cast, var_11445_cast))[name = tensor<string, []>("op_11686_cast")];
            tensor<fp16, []> var_11687_to_fp16 = const()[name = tensor<string, []>("op_11687_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_939_cast = mul(x = var_11686_cast, y = var_11687_to_fp16)[name = tensor<string, []>("aw_939_cast")];
            tensor<string, []> var_11690_equation_0 = const()[name = tensor<string, []>("op_11690_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11690_cast = einsum(equation = var_11690_equation_0, values = (var_11532_cast, var_11449_cast))[name = tensor<string, []>("op_11690_cast")];
            tensor<fp16, []> var_11691_to_fp16 = const()[name = tensor<string, []>("op_11691_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_941_cast = mul(x = var_11690_cast, y = var_11691_to_fp16)[name = tensor<string, []>("aw_941_cast")];
            tensor<string, []> var_11694_equation_0 = const()[name = tensor<string, []>("op_11694_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11694_cast = einsum(equation = var_11694_equation_0, values = (var_11536_cast, var_11453_cast))[name = tensor<string, []>("op_11694_cast")];
            tensor<fp16, []> var_11695_to_fp16 = const()[name = tensor<string, []>("op_11695_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_943_cast = mul(x = var_11694_cast, y = var_11695_to_fp16)[name = tensor<string, []>("aw_943_cast")];
            tensor<string, []> var_11698_equation_0 = const()[name = tensor<string, []>("op_11698_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11698_cast = einsum(equation = var_11698_equation_0, values = (var_11540_cast, var_11457_cast))[name = tensor<string, []>("op_11698_cast")];
            tensor<fp16, []> var_11699_to_fp16 = const()[name = tensor<string, []>("op_11699_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_945_cast = mul(x = var_11698_cast, y = var_11699_to_fp16)[name = tensor<string, []>("aw_945_cast")];
            tensor<string, []> var_11702_equation_0 = const()[name = tensor<string, []>("op_11702_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11702_cast = einsum(equation = var_11702_equation_0, values = (var_11544_cast, var_11461_cast))[name = tensor<string, []>("op_11702_cast")];
            tensor<fp16, []> var_11703_to_fp16 = const()[name = tensor<string, []>("op_11703_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_947_cast = mul(x = var_11702_cast, y = var_11703_to_fp16)[name = tensor<string, []>("aw_947_cast")];
            tensor<string, []> var_11706_equation_0 = const()[name = tensor<string, []>("op_11706_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11706_cast = einsum(equation = var_11706_equation_0, values = (var_11548_cast, var_11465_cast))[name = tensor<string, []>("op_11706_cast")];
            tensor<fp16, []> var_11707_to_fp16 = const()[name = tensor<string, []>("op_11707_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_949_cast = mul(x = var_11706_cast, y = var_11707_to_fp16)[name = tensor<string, []>("aw_949_cast")];
            tensor<string, []> var_11710_equation_0 = const()[name = tensor<string, []>("op_11710_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11710_cast = einsum(equation = var_11710_equation_0, values = (var_11552_cast, var_11469_cast))[name = tensor<string, []>("op_11710_cast")];
            tensor<fp16, []> var_11711_to_fp16 = const()[name = tensor<string, []>("op_11711_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_951_cast = mul(x = var_11710_cast, y = var_11711_to_fp16)[name = tensor<string, []>("aw_951_cast")];
            tensor<string, []> var_11714_equation_0 = const()[name = tensor<string, []>("op_11714_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11714_cast = einsum(equation = var_11714_equation_0, values = (var_11556_cast, var_11473_cast))[name = tensor<string, []>("op_11714_cast")];
            tensor<fp16, []> var_11715_to_fp16 = const()[name = tensor<string, []>("op_11715_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_953_cast = mul(x = var_11714_cast, y = var_11715_to_fp16)[name = tensor<string, []>("aw_953_cast")];
            tensor<string, []> var_11718_equation_0 = const()[name = tensor<string, []>("op_11718_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11718_cast = einsum(equation = var_11718_equation_0, values = (var_11560_cast, var_11477_cast))[name = tensor<string, []>("op_11718_cast")];
            tensor<fp16, []> var_11719_to_fp16 = const()[name = tensor<string, []>("op_11719_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_955_cast = mul(x = var_11718_cast, y = var_11719_to_fp16)[name = tensor<string, []>("aw_955_cast")];
            tensor<string, []> var_11722_equation_0 = const()[name = tensor<string, []>("op_11722_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11722_cast = einsum(equation = var_11722_equation_0, values = (var_11564_cast, var_11481_cast))[name = tensor<string, []>("op_11722_cast")];
            tensor<fp16, []> var_11723_to_fp16 = const()[name = tensor<string, []>("op_11723_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_957_cast = mul(x = var_11722_cast, y = var_11723_to_fp16)[name = tensor<string, []>("aw_957_cast")];
            tensor<string, []> var_11726_equation_0 = const()[name = tensor<string, []>("op_11726_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_11726_cast = einsum(equation = var_11726_equation_0, values = (var_11568_cast, var_11485_cast))[name = tensor<string, []>("op_11726_cast")];
            tensor<fp16, []> var_11727_to_fp16 = const()[name = tensor<string, []>("op_11727_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_959_cast = mul(x = var_11726_cast, y = var_11727_to_fp16)[name = tensor<string, []>("aw_959_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11729_cast = softmax(axis = var_2634, x = aw_921_cast)[name = tensor<string, []>("op_11729_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11730_cast = softmax(axis = var_2634, x = aw_923_cast)[name = tensor<string, []>("op_11730_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11731_cast = softmax(axis = var_2634, x = aw_925_cast)[name = tensor<string, []>("op_11731_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11732_cast = softmax(axis = var_2634, x = aw_927_cast)[name = tensor<string, []>("op_11732_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11733_cast = softmax(axis = var_2634, x = aw_929_cast)[name = tensor<string, []>("op_11733_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11734_cast = softmax(axis = var_2634, x = aw_931_cast)[name = tensor<string, []>("op_11734_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11735_cast = softmax(axis = var_2634, x = aw_933_cast)[name = tensor<string, []>("op_11735_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11736_cast = softmax(axis = var_2634, x = aw_935_cast)[name = tensor<string, []>("op_11736_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11737_cast = softmax(axis = var_2634, x = aw_937_cast)[name = tensor<string, []>("op_11737_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11738_cast = softmax(axis = var_2634, x = aw_939_cast)[name = tensor<string, []>("op_11738_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11739_cast = softmax(axis = var_2634, x = aw_941_cast)[name = tensor<string, []>("op_11739_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11740_cast = softmax(axis = var_2634, x = aw_943_cast)[name = tensor<string, []>("op_11740_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11741_cast = softmax(axis = var_2634, x = aw_945_cast)[name = tensor<string, []>("op_11741_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11742_cast = softmax(axis = var_2634, x = aw_947_cast)[name = tensor<string, []>("op_11742_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11743_cast = softmax(axis = var_2634, x = aw_949_cast)[name = tensor<string, []>("op_11743_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11744_cast = softmax(axis = var_2634, x = aw_951_cast)[name = tensor<string, []>("op_11744_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11745_cast = softmax(axis = var_2634, x = aw_953_cast)[name = tensor<string, []>("op_11745_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11746_cast = softmax(axis = var_2634, x = aw_955_cast)[name = tensor<string, []>("op_11746_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11747_cast = softmax(axis = var_2634, x = aw_957_cast)[name = tensor<string, []>("op_11747_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_11748_cast = softmax(axis = var_2634, x = aw_959_cast)[name = tensor<string, []>("op_11748_cast")];
            tensor<string, []> var_11750_equation_0 = const()[name = tensor<string, []>("op_11750_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11750_cast = einsum(equation = var_11750_equation_0, values = (var_11570_cast, var_11729_cast))[name = tensor<string, []>("op_11750_cast")];
            tensor<string, []> var_11752_equation_0 = const()[name = tensor<string, []>("op_11752_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11752_cast = einsum(equation = var_11752_equation_0, values = (var_11574_cast, var_11730_cast))[name = tensor<string, []>("op_11752_cast")];
            tensor<string, []> var_11754_equation_0 = const()[name = tensor<string, []>("op_11754_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11754_cast = einsum(equation = var_11754_equation_0, values = (var_11578_cast, var_11731_cast))[name = tensor<string, []>("op_11754_cast")];
            tensor<string, []> var_11756_equation_0 = const()[name = tensor<string, []>("op_11756_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11756_cast = einsum(equation = var_11756_equation_0, values = (var_11582_cast, var_11732_cast))[name = tensor<string, []>("op_11756_cast")];
            tensor<string, []> var_11758_equation_0 = const()[name = tensor<string, []>("op_11758_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11758_cast = einsum(equation = var_11758_equation_0, values = (var_11586_cast, var_11733_cast))[name = tensor<string, []>("op_11758_cast")];
            tensor<string, []> var_11760_equation_0 = const()[name = tensor<string, []>("op_11760_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11760_cast = einsum(equation = var_11760_equation_0, values = (var_11590_cast, var_11734_cast))[name = tensor<string, []>("op_11760_cast")];
            tensor<string, []> var_11762_equation_0 = const()[name = tensor<string, []>("op_11762_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11762_cast = einsum(equation = var_11762_equation_0, values = (var_11594_cast, var_11735_cast))[name = tensor<string, []>("op_11762_cast")];
            tensor<string, []> var_11764_equation_0 = const()[name = tensor<string, []>("op_11764_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11764_cast = einsum(equation = var_11764_equation_0, values = (var_11598_cast, var_11736_cast))[name = tensor<string, []>("op_11764_cast")];
            tensor<string, []> var_11766_equation_0 = const()[name = tensor<string, []>("op_11766_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11766_cast = einsum(equation = var_11766_equation_0, values = (var_11602_cast, var_11737_cast))[name = tensor<string, []>("op_11766_cast")];
            tensor<string, []> var_11768_equation_0 = const()[name = tensor<string, []>("op_11768_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11768_cast = einsum(equation = var_11768_equation_0, values = (var_11606_cast, var_11738_cast))[name = tensor<string, []>("op_11768_cast")];
            tensor<string, []> var_11770_equation_0 = const()[name = tensor<string, []>("op_11770_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11770_cast = einsum(equation = var_11770_equation_0, values = (var_11610_cast, var_11739_cast))[name = tensor<string, []>("op_11770_cast")];
            tensor<string, []> var_11772_equation_0 = const()[name = tensor<string, []>("op_11772_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11772_cast = einsum(equation = var_11772_equation_0, values = (var_11614_cast, var_11740_cast))[name = tensor<string, []>("op_11772_cast")];
            tensor<string, []> var_11774_equation_0 = const()[name = tensor<string, []>("op_11774_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11774_cast = einsum(equation = var_11774_equation_0, values = (var_11618_cast, var_11741_cast))[name = tensor<string, []>("op_11774_cast")];
            tensor<string, []> var_11776_equation_0 = const()[name = tensor<string, []>("op_11776_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11776_cast = einsum(equation = var_11776_equation_0, values = (var_11622_cast, var_11742_cast))[name = tensor<string, []>("op_11776_cast")];
            tensor<string, []> var_11778_equation_0 = const()[name = tensor<string, []>("op_11778_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11778_cast = einsum(equation = var_11778_equation_0, values = (var_11626_cast, var_11743_cast))[name = tensor<string, []>("op_11778_cast")];
            tensor<string, []> var_11780_equation_0 = const()[name = tensor<string, []>("op_11780_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11780_cast = einsum(equation = var_11780_equation_0, values = (var_11630_cast, var_11744_cast))[name = tensor<string, []>("op_11780_cast")];
            tensor<string, []> var_11782_equation_0 = const()[name = tensor<string, []>("op_11782_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11782_cast = einsum(equation = var_11782_equation_0, values = (var_11634_cast, var_11745_cast))[name = tensor<string, []>("op_11782_cast")];
            tensor<string, []> var_11784_equation_0 = const()[name = tensor<string, []>("op_11784_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11784_cast = einsum(equation = var_11784_equation_0, values = (var_11638_cast, var_11746_cast))[name = tensor<string, []>("op_11784_cast")];
            tensor<string, []> var_11786_equation_0 = const()[name = tensor<string, []>("op_11786_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11786_cast = einsum(equation = var_11786_equation_0, values = (var_11642_cast, var_11747_cast))[name = tensor<string, []>("op_11786_cast")];
            tensor<string, []> var_11788_equation_0 = const()[name = tensor<string, []>("op_11788_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_11788_cast = einsum(equation = var_11788_equation_0, values = (var_11646_cast, var_11748_cast))[name = tensor<string, []>("op_11788_cast")];
            tensor<bool, []> input_209_interleave_0 = const()[name = tensor<string, []>("input_209_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_209_cast = concat(axis = var_2634, interleave = input_209_interleave_0, values = (var_11750_cast, var_11752_cast, var_11754_cast, var_11756_cast, var_11758_cast, var_11760_cast, var_11762_cast, var_11764_cast, var_11766_cast, var_11768_cast, var_11770_cast, var_11772_cast, var_11774_cast, var_11776_cast, var_11778_cast, var_11780_cast, var_11782_cast, var_11784_cast, var_11786_cast, var_11788_cast))[name = tensor<string, []>("input_209_cast")];
            tensor<int32, [2]> var_11794 = const()[name = tensor<string, []>("op_11794"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11796 = const()[name = tensor<string, []>("op_11796"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11798_pad_type_0 = const()[name = tensor<string, []>("op_11798_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11798_pad_0 = const()[name = tensor<string, []>("op_11798_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(850665856)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(853942720)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_11798_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16, dilations = var_11796, groups = var_2634, pad = var_11798_pad_0, pad_type = var_11798_pad_type_0, strides = var_11794, weight = down_blocks_2_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16, x = input_209_cast)[name = tensor<string, []>("op_11798_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_83_cast = add(x = var_11798_cast, y = inputs_81_cast)[name = tensor<string, []>("inputs_83_cast")];
            tensor<int32, [1]> var_11802 = const()[name = tensor<string, []>("op_11802"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_83_cast = reduce_mean(axes = var_11802, keep_dims = var_2629, x = inputs_83_cast)[name = tensor<string, []>("channels_mean_83_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_83_cast = sub(x = inputs_83_cast, y = channels_mean_83_cast)[name = tensor<string, []>("zero_mean_83_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_83_cast = mul(x = zero_mean_83_cast, y = zero_mean_83_cast)[name = tensor<string, []>("zero_mean_sq_83_cast")];
            tensor<int32, [1]> var_11806 = const()[name = tensor<string, []>("op_11806"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_11807_cast = reduce_mean(axes = var_11806, keep_dims = var_2629, x = zero_mean_sq_83_cast)[name = tensor<string, []>("op_11807_cast")];
            tensor<fp16, []> var_11808_to_fp16 = const()[name = tensor<string, []>("op_11808_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_11809_cast = add(x = var_11807_cast, y = var_11808_to_fp16)[name = tensor<string, []>("op_11809_cast")];
            tensor<fp16, []> denom_83_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_83_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_83_cast = rsqrt(epsilon = denom_83_epsilon_0_to_fp16, x = var_11809_cast)[name = tensor<string, []>("denom_83_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_83_cast = mul(x = zero_mean_83_cast, y = denom_83_cast)[name = tensor<string, []>("out_83_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_11813_to_fp16 = const()[name = tensor<string, []>("op_11813_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(853945344)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_11814_cast = add(x = out_83_cast, y = var_11813_to_fp16)[name = tensor<string, []>("op_11814_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_11816_to_fp16 = const()[name = tensor<string, []>("op_11816_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(853947968)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_211_cast = mul(x = var_11814_cast, y = var_11816_to_fp16)[name = tensor<string, []>("input_211_cast")];
            tensor<int32, [2]> var_11824 = const()[name = tensor<string, []>("op_11824"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11826 = const()[name = tensor<string, []>("op_11826"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11828_pad_type_0 = const()[name = tensor<string, []>("op_11828_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11828_pad_0 = const()[name = tensor<string, []>("op_11828_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(853950592)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880165056)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_11828_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16, dilations = var_11826, groups = var_2634, pad = var_11828_pad_0, pad_type = var_11828_pad_type_0, strides = var_11824, weight = down_blocks_2_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16, x = input_211_cast)[name = tensor<string, []>("op_11828_cast")];
            tensor<int32, [2]> var_11829_split_sizes_0 = const()[name = tensor<string, []>("op_11829_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_11829_axis_0 = const()[name = tensor<string, []>("op_11829_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_11829_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_11829_cast_1 = split(axis = var_11829_axis_0, split_sizes = var_11829_split_sizes_0, x = var_11828_cast)[name = tensor<string, []>("op_11829_cast")];
            tensor<string, []> var_11831_mode_0 = const()[name = tensor<string, []>("op_11831_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_11831_cast = gelu(mode = var_11831_mode_0, x = var_11829_cast_1)[name = tensor<string, []>("op_11831_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_213_cast = mul(x = var_11829_cast_0, y = var_11831_cast)[name = tensor<string, []>("input_213_cast")];
            tensor<int32, [2]> var_11835 = const()[name = tensor<string, []>("op_11835"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11837 = const()[name = tensor<string, []>("op_11837"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11839_pad_type_0 = const()[name = tensor<string, []>("op_11839_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11839_pad_0 = const()[name = tensor<string, []>("op_11839_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880185600)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893292864)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_11839_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16, dilations = var_11837, groups = var_2634, pad = var_11839_pad_0, pad_type = var_11839_pad_type_0, strides = var_11835, weight = down_blocks_2_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16, x = input_213_cast)[name = tensor<string, []>("op_11839_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_125_cast = add(x = var_11839_cast, y = inputs_83_cast)[name = tensor<string, []>("hidden_states_125_cast")];
            tensor<int32, [4]> var_11841 = const()[name = tensor<string, []>("op_11841"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> input_215_cast = reshape(shape = var_11841, x = hidden_states_125_cast)[name = tensor<string, []>("input_215_cast")];
            tensor<int32, [2]> var_11845 = const()[name = tensor<string, []>("op_11845"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11847 = const()[name = tensor<string, []>("op_11847"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_127_pad_type_0 = const()[name = tensor<string, []>("hidden_states_127_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_127_pad_0 = const()[name = tensor<string, []>("hidden_states_127_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_proj_out_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893295488)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_proj_out_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(896572352)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_127_cast = conv(bias = down_blocks_2_attentions_0_proj_out_bias_to_fp16, dilations = var_11847, groups = var_2634, pad = hidden_states_127_pad_0, pad_type = hidden_states_127_pad_type_0, strides = var_11845, weight = down_blocks_2_attentions_0_proj_out_weight_to_fp16, x = input_215_cast)[name = tensor<string, []>("hidden_states_127_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_217_cast = add(x = hidden_states_127_cast, y = hidden_states_61_cast)[name = tensor<string, []>("input_217_cast")];
            tensor<int32, [5]> reshape_52_shape_0 = const()[name = tensor<string, []>("reshape_52_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_52_cast = reshape(shape = reshape_52_shape_0, x = input_217_cast)[name = tensor<string, []>("reshape_52_cast")];
            tensor<int32, [3]> reduce_mean_39_axes_0 = const()[name = tensor<string, []>("reduce_mean_39_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_39_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_39_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_39_cast = reduce_mean(axes = reduce_mean_39_axes_0, keep_dims = reduce_mean_39_keep_dims_0, x = reshape_52_cast)[name = tensor<string, []>("reduce_mean_39_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_26_cast = sub(x = reshape_52_cast, y = reduce_mean_39_cast)[name = tensor<string, []>("sub_26_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_13_cast = square(x = sub_26_cast)[name = tensor<string, []>("square_13_cast")];
            tensor<int32, [3]> reduce_mean_41_axes_0 = const()[name = tensor<string, []>("reduce_mean_41_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_41_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_41_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_41_cast = reduce_mean(axes = reduce_mean_41_axes_0, keep_dims = reduce_mean_41_keep_dims_0, x = square_13_cast)[name = tensor<string, []>("reduce_mean_41_cast")];
            tensor<fp16, []> add_26_y_0_to_fp16 = const()[name = tensor<string, []>("add_26_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_26_cast = add(x = reduce_mean_41_cast, y = add_26_y_0_to_fp16)[name = tensor<string, []>("add_26_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_13_cast = sqrt(x = add_26_cast)[name = tensor<string, []>("sqrt_13_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_13_cast = real_div(x = sub_26_cast, y = sqrt_13_cast)[name = tensor<string, []>("real_div_13_cast")];
            tensor<int32, [4]> reshape_53_shape_0 = const()[name = tensor<string, []>("reshape_53_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_53_cast = reshape(shape = reshape_53_shape_0, x = real_div_13_cast)[name = tensor<string, []>("reshape_53_cast")];
            tensor<fp16, [1280]> add_27_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_27_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(896574976)))];
            tensor<fp16, [1280]> add_27_beta_0_to_fp16 = const()[name = tensor<string, []>("add_27_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(896577600)))];
            tensor<fp16, []> add_27_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_27_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 32, 32]> add_27_cast = batch_norm(beta = add_27_beta_0_to_fp16, epsilon = add_27_epsilon_0_to_fp16, gamma = add_27_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_53_cast)[name = tensor<string, []>("add_27_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_221_cast = silu(x = add_27_cast)[name = tensor<string, []>("input_221_cast")];
            tensor<int32, [2]> var_11862 = const()[name = tensor<string, []>("op_11862"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11864 = const()[name = tensor<string, []>("op_11864"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_129_pad_type_0 = const()[name = tensor<string, []>("hidden_states_129_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_129_pad_0 = const()[name = tensor<string, []>("hidden_states_129_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> down_blocks_2_resnets_1_conv1_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_1_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(896580224)))];
            tensor<fp16, [1280]> down_blocks_2_resnets_1_conv1_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_1_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(926071488)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_129_cast = conv(bias = down_blocks_2_resnets_1_conv1_bias_to_fp16, dilations = var_11864, groups = var_2634, pad = hidden_states_129_pad_0, pad_type = hidden_states_129_pad_type_0, strides = var_11862, weight = down_blocks_2_resnets_1_conv1_weight_to_fp16, x = input_221_cast)[name = tensor<string, []>("hidden_states_129_cast")];
            tensor<int32, [2]> var_11870 = const()[name = tensor<string, []>("op_11870"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11872 = const()[name = tensor<string, []>("op_11872"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_11_pad_type_0 = const()[name = tensor<string, []>("temb_11_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_11_pad_0 = const()[name = tensor<string, []>("temb_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(926074112)))];
            tensor<fp16, [1280]> down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(929350976)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_11_cast = conv(bias = down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_11872, groups = var_2634, pad = temb_11_pad_0, pad_type = temb_11_pad_type_0, strides = var_11870, weight = down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16, x = input_23_cast)[name = tensor<string, []>("temb_11_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_225_cast = add(x = hidden_states_129_cast, y = temb_11_cast)[name = tensor<string, []>("input_225_cast")];
            tensor<int32, [5]> reshape_56_shape_0 = const()[name = tensor<string, []>("reshape_56_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_56_cast = reshape(shape = reshape_56_shape_0, x = input_225_cast)[name = tensor<string, []>("reshape_56_cast")];
            tensor<int32, [3]> reduce_mean_42_axes_0 = const()[name = tensor<string, []>("reduce_mean_42_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_42_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_42_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_42_cast = reduce_mean(axes = reduce_mean_42_axes_0, keep_dims = reduce_mean_42_keep_dims_0, x = reshape_56_cast)[name = tensor<string, []>("reduce_mean_42_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_28_cast = sub(x = reshape_56_cast, y = reduce_mean_42_cast)[name = tensor<string, []>("sub_28_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_14_cast = square(x = sub_28_cast)[name = tensor<string, []>("square_14_cast")];
            tensor<int32, [3]> reduce_mean_44_axes_0 = const()[name = tensor<string, []>("reduce_mean_44_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_44_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_44_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_44_cast = reduce_mean(axes = reduce_mean_44_axes_0, keep_dims = reduce_mean_44_keep_dims_0, x = square_14_cast)[name = tensor<string, []>("reduce_mean_44_cast")];
            tensor<fp16, []> add_28_y_0_to_fp16 = const()[name = tensor<string, []>("add_28_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_28_cast = add(x = reduce_mean_44_cast, y = add_28_y_0_to_fp16)[name = tensor<string, []>("add_28_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_14_cast = sqrt(x = add_28_cast)[name = tensor<string, []>("sqrt_14_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_14_cast = real_div(x = sub_28_cast, y = sqrt_14_cast)[name = tensor<string, []>("real_div_14_cast")];
            tensor<int32, [4]> reshape_57_shape_0 = const()[name = tensor<string, []>("reshape_57_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_57_cast = reshape(shape = reshape_57_shape_0, x = real_div_14_cast)[name = tensor<string, []>("reshape_57_cast")];
            tensor<fp16, [1280]> add_29_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_29_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(929353600)))];
            tensor<fp16, [1280]> add_29_beta_0_to_fp16 = const()[name = tensor<string, []>("add_29_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(929356224)))];
            tensor<fp16, []> add_29_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_29_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 32, 32]> add_29_cast = batch_norm(beta = add_29_beta_0_to_fp16, epsilon = add_29_epsilon_0_to_fp16, gamma = add_29_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_57_cast)[name = tensor<string, []>("add_29_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_229_cast = silu(x = add_29_cast)[name = tensor<string, []>("input_229_cast")];
            tensor<int32, [2]> var_11882 = const()[name = tensor<string, []>("op_11882"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11884 = const()[name = tensor<string, []>("op_11884"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_131_pad_type_0 = const()[name = tensor<string, []>("hidden_states_131_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_131_pad_0 = const()[name = tensor<string, []>("hidden_states_131_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> down_blocks_2_resnets_1_conv2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_1_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(929358848)))];
            tensor<fp16, [1280]> down_blocks_2_resnets_1_conv2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_1_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958850112)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_131_cast = conv(bias = down_blocks_2_resnets_1_conv2_bias_to_fp16, dilations = var_11884, groups = var_2634, pad = hidden_states_131_pad_0, pad_type = hidden_states_131_pad_type_0, strides = var_11882, weight = down_blocks_2_resnets_1_conv2_weight_to_fp16, x = input_229_cast)[name = tensor<string, []>("hidden_states_131_cast")];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_133_cast = add(x = input_217_cast, y = hidden_states_131_cast)[name = tensor<string, []>("hidden_states_133_cast")];
            tensor<int32, [5]> reshape_60_shape_0 = const()[name = tensor<string, []>("reshape_60_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_60_cast = reshape(shape = reshape_60_shape_0, x = hidden_states_133_cast)[name = tensor<string, []>("reshape_60_cast")];
            tensor<int32, [3]> reduce_mean_45_axes_0 = const()[name = tensor<string, []>("reduce_mean_45_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_45_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_45_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_45_cast = reduce_mean(axes = reduce_mean_45_axes_0, keep_dims = reduce_mean_45_keep_dims_0, x = reshape_60_cast)[name = tensor<string, []>("reduce_mean_45_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_30_cast = sub(x = reshape_60_cast, y = reduce_mean_45_cast)[name = tensor<string, []>("sub_30_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_15_cast = square(x = sub_30_cast)[name = tensor<string, []>("square_15_cast")];
            tensor<int32, [3]> reduce_mean_47_axes_0 = const()[name = tensor<string, []>("reduce_mean_47_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_47_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_47_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_47_cast = reduce_mean(axes = reduce_mean_47_axes_0, keep_dims = reduce_mean_47_keep_dims_0, x = square_15_cast)[name = tensor<string, []>("reduce_mean_47_cast")];
            tensor<fp16, []> add_30_y_0_to_fp16 = const()[name = tensor<string, []>("add_30_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_30_cast = add(x = reduce_mean_47_cast, y = add_30_y_0_to_fp16)[name = tensor<string, []>("add_30_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_15_cast = sqrt(x = add_30_cast)[name = tensor<string, []>("sqrt_15_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_15_cast = real_div(x = sub_30_cast, y = sqrt_15_cast)[name = tensor<string, []>("real_div_15_cast")];
            tensor<int32, [4]> reshape_61_shape_0 = const()[name = tensor<string, []>("reshape_61_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_61_cast = reshape(shape = reshape_61_shape_0, x = real_div_15_cast)[name = tensor<string, []>("reshape_61_cast")];
            tensor<fp16, [1280]> add_31_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_31_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958852736)))];
            tensor<fp16, [1280]> add_31_beta_0_to_fp16 = const()[name = tensor<string, []>("add_31_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958855360)))];
            tensor<fp16, []> add_31_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_31_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 32, 32]> add_31_cast = batch_norm(beta = add_31_beta_0_to_fp16, epsilon = add_31_epsilon_0_to_fp16, gamma = add_31_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_61_cast)[name = tensor<string, []>("add_31_cast")];
            tensor<int32, [2]> var_11922 = const()[name = tensor<string, []>("op_11922"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11924 = const()[name = tensor<string, []>("op_11924"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_135_pad_type_0 = const()[name = tensor<string, []>("hidden_states_135_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_135_pad_0 = const()[name = tensor<string, []>("hidden_states_135_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_proj_in_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958857984)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_proj_in_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(962134848)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_135_cast = conv(bias = down_blocks_2_attentions_1_proj_in_bias_to_fp16, dilations = var_11924, groups = var_2634, pad = hidden_states_135_pad_0, pad_type = hidden_states_135_pad_type_0, strides = var_11922, weight = down_blocks_2_attentions_1_proj_in_weight_to_fp16, x = add_31_cast)[name = tensor<string, []>("hidden_states_135_cast")];
            tensor<int32, [4]> var_11929 = const()[name = tensor<string, []>("op_11929"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_85_cast = reshape(shape = var_11929, x = hidden_states_135_cast)[name = tensor<string, []>("inputs_85_cast")];
            tensor<int32, [1]> var_11939 = const()[name = tensor<string, []>("op_11939"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_85_cast = reduce_mean(axes = var_11939, keep_dims = var_2629, x = inputs_85_cast)[name = tensor<string, []>("channels_mean_85_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_85_cast = sub(x = inputs_85_cast, y = channels_mean_85_cast)[name = tensor<string, []>("zero_mean_85_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_85_cast = mul(x = zero_mean_85_cast, y = zero_mean_85_cast)[name = tensor<string, []>("zero_mean_sq_85_cast")];
            tensor<int32, [1]> var_11943 = const()[name = tensor<string, []>("op_11943"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_11944_cast = reduce_mean(axes = var_11943, keep_dims = var_2629, x = zero_mean_sq_85_cast)[name = tensor<string, []>("op_11944_cast")];
            tensor<fp16, []> var_11945_to_fp16 = const()[name = tensor<string, []>("op_11945_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_11946_cast = add(x = var_11944_cast, y = var_11945_to_fp16)[name = tensor<string, []>("op_11946_cast")];
            tensor<fp16, []> denom_85_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_85_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_85_cast = rsqrt(epsilon = denom_85_epsilon_0_to_fp16, x = var_11946_cast)[name = tensor<string, []>("denom_85_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_85_cast = mul(x = zero_mean_85_cast, y = denom_85_cast)[name = tensor<string, []>("out_85_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_11950_to_fp16 = const()[name = tensor<string, []>("op_11950_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(962137472)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_11951_cast = add(x = out_85_cast, y = var_11950_to_fp16)[name = tensor<string, []>("op_11951_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_11953_to_fp16 = const()[name = tensor<string, []>("op_11953_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(962140096)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_137_cast = mul(x = var_11951_cast, y = var_11953_to_fp16)[name = tensor<string, []>("hidden_states_137_cast")];
            tensor<int32, [2]> var_11960 = const()[name = tensor<string, []>("op_11960"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11962 = const()[name = tensor<string, []>("op_11962"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_57_pad_type_0 = const()[name = tensor<string, []>("q_57_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_57_pad_0 = const()[name = tensor<string, []>("q_57_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(962142720)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_57_cast = conv(dilations = var_11962, groups = var_2634, pad = q_57_pad_0, pad_type = q_57_pad_type_0, strides = var_11960, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_137_cast)[name = tensor<string, []>("q_57_cast")];
            tensor<int32, [2]> var_11966 = const()[name = tensor<string, []>("op_11966"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11968 = const()[name = tensor<string, []>("op_11968"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_113_pad_type_0 = const()[name = tensor<string, []>("k_113_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_113_pad_0 = const()[name = tensor<string, []>("k_113_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(965419584)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_113_cast = conv(dilations = var_11968, groups = var_2634, pad = k_113_pad_0, pad_type = k_113_pad_type_0, strides = var_11966, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_137_cast)[name = tensor<string, []>("k_113_cast")];
            tensor<int32, [2]> var_11972 = const()[name = tensor<string, []>("op_11972"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11974 = const()[name = tensor<string, []>("op_11974"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_57_pad_type_0 = const()[name = tensor<string, []>("v_57_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_57_pad_0 = const()[name = tensor<string, []>("v_57_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(968696448)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_57_cast = conv(dilations = var_11974, groups = var_2634, pad = v_57_pad_0, pad_type = v_57_pad_type_0, strides = var_11972, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_137_cast)[name = tensor<string, []>("v_57_cast")];
            tensor<int32, [4]> var_11978_begin_0 = const()[name = tensor<string, []>("op_11978_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11978_end_0 = const()[name = tensor<string, []>("op_11978_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_11978_end_mask_0 = const()[name = tensor<string, []>("op_11978_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11978_cast = slice_by_index(begin = var_11978_begin_0, end = var_11978_end_0, end_mask = var_11978_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_11978_cast")];
            tensor<int32, [4]> var_11982_begin_0 = const()[name = tensor<string, []>("op_11982_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_11982_end_0 = const()[name = tensor<string, []>("op_11982_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_11982_end_mask_0 = const()[name = tensor<string, []>("op_11982_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11982_cast = slice_by_index(begin = var_11982_begin_0, end = var_11982_end_0, end_mask = var_11982_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_11982_cast")];
            tensor<int32, [4]> var_11986_begin_0 = const()[name = tensor<string, []>("op_11986_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_11986_end_0 = const()[name = tensor<string, []>("op_11986_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_11986_end_mask_0 = const()[name = tensor<string, []>("op_11986_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11986_cast = slice_by_index(begin = var_11986_begin_0, end = var_11986_end_0, end_mask = var_11986_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_11986_cast")];
            tensor<int32, [4]> var_11990_begin_0 = const()[name = tensor<string, []>("op_11990_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_11990_end_0 = const()[name = tensor<string, []>("op_11990_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_11990_end_mask_0 = const()[name = tensor<string, []>("op_11990_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11990_cast = slice_by_index(begin = var_11990_begin_0, end = var_11990_end_0, end_mask = var_11990_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_11990_cast")];
            tensor<int32, [4]> var_11994_begin_0 = const()[name = tensor<string, []>("op_11994_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_11994_end_0 = const()[name = tensor<string, []>("op_11994_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_11994_end_mask_0 = const()[name = tensor<string, []>("op_11994_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11994_cast = slice_by_index(begin = var_11994_begin_0, end = var_11994_end_0, end_mask = var_11994_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_11994_cast")];
            tensor<int32, [4]> var_11998_begin_0 = const()[name = tensor<string, []>("op_11998_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_11998_end_0 = const()[name = tensor<string, []>("op_11998_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_11998_end_mask_0 = const()[name = tensor<string, []>("op_11998_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11998_cast = slice_by_index(begin = var_11998_begin_0, end = var_11998_end_0, end_mask = var_11998_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_11998_cast")];
            tensor<int32, [4]> var_12002_begin_0 = const()[name = tensor<string, []>("op_12002_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_12002_end_0 = const()[name = tensor<string, []>("op_12002_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_12002_end_mask_0 = const()[name = tensor<string, []>("op_12002_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12002_cast = slice_by_index(begin = var_12002_begin_0, end = var_12002_end_0, end_mask = var_12002_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_12002_cast")];
            tensor<int32, [4]> var_12006_begin_0 = const()[name = tensor<string, []>("op_12006_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_12006_end_0 = const()[name = tensor<string, []>("op_12006_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_12006_end_mask_0 = const()[name = tensor<string, []>("op_12006_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12006_cast = slice_by_index(begin = var_12006_begin_0, end = var_12006_end_0, end_mask = var_12006_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_12006_cast")];
            tensor<int32, [4]> var_12010_begin_0 = const()[name = tensor<string, []>("op_12010_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_12010_end_0 = const()[name = tensor<string, []>("op_12010_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_12010_end_mask_0 = const()[name = tensor<string, []>("op_12010_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12010_cast = slice_by_index(begin = var_12010_begin_0, end = var_12010_end_0, end_mask = var_12010_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_12010_cast")];
            tensor<int32, [4]> var_12014_begin_0 = const()[name = tensor<string, []>("op_12014_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_12014_end_0 = const()[name = tensor<string, []>("op_12014_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_12014_end_mask_0 = const()[name = tensor<string, []>("op_12014_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12014_cast = slice_by_index(begin = var_12014_begin_0, end = var_12014_end_0, end_mask = var_12014_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_12014_cast")];
            tensor<int32, [4]> var_12018_begin_0 = const()[name = tensor<string, []>("op_12018_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_12018_end_0 = const()[name = tensor<string, []>("op_12018_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_12018_end_mask_0 = const()[name = tensor<string, []>("op_12018_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12018_cast = slice_by_index(begin = var_12018_begin_0, end = var_12018_end_0, end_mask = var_12018_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_12018_cast")];
            tensor<int32, [4]> var_12022_begin_0 = const()[name = tensor<string, []>("op_12022_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_12022_end_0 = const()[name = tensor<string, []>("op_12022_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_12022_end_mask_0 = const()[name = tensor<string, []>("op_12022_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12022_cast = slice_by_index(begin = var_12022_begin_0, end = var_12022_end_0, end_mask = var_12022_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_12022_cast")];
            tensor<int32, [4]> var_12026_begin_0 = const()[name = tensor<string, []>("op_12026_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_12026_end_0 = const()[name = tensor<string, []>("op_12026_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_12026_end_mask_0 = const()[name = tensor<string, []>("op_12026_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12026_cast = slice_by_index(begin = var_12026_begin_0, end = var_12026_end_0, end_mask = var_12026_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_12026_cast")];
            tensor<int32, [4]> var_12030_begin_0 = const()[name = tensor<string, []>("op_12030_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_12030_end_0 = const()[name = tensor<string, []>("op_12030_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_12030_end_mask_0 = const()[name = tensor<string, []>("op_12030_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12030_cast = slice_by_index(begin = var_12030_begin_0, end = var_12030_end_0, end_mask = var_12030_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_12030_cast")];
            tensor<int32, [4]> var_12034_begin_0 = const()[name = tensor<string, []>("op_12034_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_12034_end_0 = const()[name = tensor<string, []>("op_12034_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_12034_end_mask_0 = const()[name = tensor<string, []>("op_12034_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12034_cast = slice_by_index(begin = var_12034_begin_0, end = var_12034_end_0, end_mask = var_12034_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_12034_cast")];
            tensor<int32, [4]> var_12038_begin_0 = const()[name = tensor<string, []>("op_12038_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_12038_end_0 = const()[name = tensor<string, []>("op_12038_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_12038_end_mask_0 = const()[name = tensor<string, []>("op_12038_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12038_cast = slice_by_index(begin = var_12038_begin_0, end = var_12038_end_0, end_mask = var_12038_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_12038_cast")];
            tensor<int32, [4]> var_12042_begin_0 = const()[name = tensor<string, []>("op_12042_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_12042_end_0 = const()[name = tensor<string, []>("op_12042_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_12042_end_mask_0 = const()[name = tensor<string, []>("op_12042_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12042_cast = slice_by_index(begin = var_12042_begin_0, end = var_12042_end_0, end_mask = var_12042_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_12042_cast")];
            tensor<int32, [4]> var_12046_begin_0 = const()[name = tensor<string, []>("op_12046_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_12046_end_0 = const()[name = tensor<string, []>("op_12046_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_12046_end_mask_0 = const()[name = tensor<string, []>("op_12046_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12046_cast = slice_by_index(begin = var_12046_begin_0, end = var_12046_end_0, end_mask = var_12046_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_12046_cast")];
            tensor<int32, [4]> var_12050_begin_0 = const()[name = tensor<string, []>("op_12050_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_12050_end_0 = const()[name = tensor<string, []>("op_12050_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_12050_end_mask_0 = const()[name = tensor<string, []>("op_12050_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12050_cast = slice_by_index(begin = var_12050_begin_0, end = var_12050_end_0, end_mask = var_12050_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_12050_cast")];
            tensor<int32, [4]> var_12054_begin_0 = const()[name = tensor<string, []>("op_12054_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_12054_end_0 = const()[name = tensor<string, []>("op_12054_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_12054_end_mask_0 = const()[name = tensor<string, []>("op_12054_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12054_cast = slice_by_index(begin = var_12054_begin_0, end = var_12054_end_0, end_mask = var_12054_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_12054_cast")];
            tensor<int32, [4]> k_115_perm_0 = const()[name = tensor<string, []>("k_115_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_12061_begin_0 = const()[name = tensor<string, []>("op_12061_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12061_end_0 = const()[name = tensor<string, []>("op_12061_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_12061_end_mask_0 = const()[name = tensor<string, []>("op_12061_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_111 = transpose(perm = k_115_perm_0, x = k_113_cast)[name = tensor<string, []>("transpose_111")];
            tensor<fp16, [2, 1024, 1, 64]> var_12061_cast = slice_by_index(begin = var_12061_begin_0, end = var_12061_end_0, end_mask = var_12061_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12061_cast")];
            tensor<int32, [4]> var_12065_begin_0 = const()[name = tensor<string, []>("op_12065_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_12065_end_0 = const()[name = tensor<string, []>("op_12065_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_12065_end_mask_0 = const()[name = tensor<string, []>("op_12065_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12065_cast = slice_by_index(begin = var_12065_begin_0, end = var_12065_end_0, end_mask = var_12065_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12065_cast")];
            tensor<int32, [4]> var_12069_begin_0 = const()[name = tensor<string, []>("op_12069_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_12069_end_0 = const()[name = tensor<string, []>("op_12069_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_12069_end_mask_0 = const()[name = tensor<string, []>("op_12069_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12069_cast = slice_by_index(begin = var_12069_begin_0, end = var_12069_end_0, end_mask = var_12069_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12069_cast")];
            tensor<int32, [4]> var_12073_begin_0 = const()[name = tensor<string, []>("op_12073_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_12073_end_0 = const()[name = tensor<string, []>("op_12073_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_12073_end_mask_0 = const()[name = tensor<string, []>("op_12073_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12073_cast = slice_by_index(begin = var_12073_begin_0, end = var_12073_end_0, end_mask = var_12073_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12073_cast")];
            tensor<int32, [4]> var_12077_begin_0 = const()[name = tensor<string, []>("op_12077_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_12077_end_0 = const()[name = tensor<string, []>("op_12077_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_12077_end_mask_0 = const()[name = tensor<string, []>("op_12077_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12077_cast = slice_by_index(begin = var_12077_begin_0, end = var_12077_end_0, end_mask = var_12077_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12077_cast")];
            tensor<int32, [4]> var_12081_begin_0 = const()[name = tensor<string, []>("op_12081_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_12081_end_0 = const()[name = tensor<string, []>("op_12081_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_12081_end_mask_0 = const()[name = tensor<string, []>("op_12081_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12081_cast = slice_by_index(begin = var_12081_begin_0, end = var_12081_end_0, end_mask = var_12081_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12081_cast")];
            tensor<int32, [4]> var_12085_begin_0 = const()[name = tensor<string, []>("op_12085_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_12085_end_0 = const()[name = tensor<string, []>("op_12085_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_12085_end_mask_0 = const()[name = tensor<string, []>("op_12085_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12085_cast = slice_by_index(begin = var_12085_begin_0, end = var_12085_end_0, end_mask = var_12085_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12085_cast")];
            tensor<int32, [4]> var_12089_begin_0 = const()[name = tensor<string, []>("op_12089_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_12089_end_0 = const()[name = tensor<string, []>("op_12089_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_12089_end_mask_0 = const()[name = tensor<string, []>("op_12089_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12089_cast = slice_by_index(begin = var_12089_begin_0, end = var_12089_end_0, end_mask = var_12089_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12089_cast")];
            tensor<int32, [4]> var_12093_begin_0 = const()[name = tensor<string, []>("op_12093_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12093_end_0 = const()[name = tensor<string, []>("op_12093_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_12093_end_mask_0 = const()[name = tensor<string, []>("op_12093_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12093_cast = slice_by_index(begin = var_12093_begin_0, end = var_12093_end_0, end_mask = var_12093_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12093_cast")];
            tensor<int32, [4]> var_12097_begin_0 = const()[name = tensor<string, []>("op_12097_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_12097_end_0 = const()[name = tensor<string, []>("op_12097_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_12097_end_mask_0 = const()[name = tensor<string, []>("op_12097_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12097_cast = slice_by_index(begin = var_12097_begin_0, end = var_12097_end_0, end_mask = var_12097_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12097_cast")];
            tensor<int32, [4]> var_12101_begin_0 = const()[name = tensor<string, []>("op_12101_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_12101_end_0 = const()[name = tensor<string, []>("op_12101_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_12101_end_mask_0 = const()[name = tensor<string, []>("op_12101_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12101_cast = slice_by_index(begin = var_12101_begin_0, end = var_12101_end_0, end_mask = var_12101_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12101_cast")];
            tensor<int32, [4]> var_12105_begin_0 = const()[name = tensor<string, []>("op_12105_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_12105_end_0 = const()[name = tensor<string, []>("op_12105_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_12105_end_mask_0 = const()[name = tensor<string, []>("op_12105_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12105_cast = slice_by_index(begin = var_12105_begin_0, end = var_12105_end_0, end_mask = var_12105_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12105_cast")];
            tensor<int32, [4]> var_12109_begin_0 = const()[name = tensor<string, []>("op_12109_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_12109_end_0 = const()[name = tensor<string, []>("op_12109_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_12109_end_mask_0 = const()[name = tensor<string, []>("op_12109_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12109_cast = slice_by_index(begin = var_12109_begin_0, end = var_12109_end_0, end_mask = var_12109_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12109_cast")];
            tensor<int32, [4]> var_12113_begin_0 = const()[name = tensor<string, []>("op_12113_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_12113_end_0 = const()[name = tensor<string, []>("op_12113_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_12113_end_mask_0 = const()[name = tensor<string, []>("op_12113_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12113_cast = slice_by_index(begin = var_12113_begin_0, end = var_12113_end_0, end_mask = var_12113_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12113_cast")];
            tensor<int32, [4]> var_12117_begin_0 = const()[name = tensor<string, []>("op_12117_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_12117_end_0 = const()[name = tensor<string, []>("op_12117_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_12117_end_mask_0 = const()[name = tensor<string, []>("op_12117_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12117_cast = slice_by_index(begin = var_12117_begin_0, end = var_12117_end_0, end_mask = var_12117_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12117_cast")];
            tensor<int32, [4]> var_12121_begin_0 = const()[name = tensor<string, []>("op_12121_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_12121_end_0 = const()[name = tensor<string, []>("op_12121_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_12121_end_mask_0 = const()[name = tensor<string, []>("op_12121_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12121_cast = slice_by_index(begin = var_12121_begin_0, end = var_12121_end_0, end_mask = var_12121_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12121_cast")];
            tensor<int32, [4]> var_12125_begin_0 = const()[name = tensor<string, []>("op_12125_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12125_end_0 = const()[name = tensor<string, []>("op_12125_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_12125_end_mask_0 = const()[name = tensor<string, []>("op_12125_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12125_cast = slice_by_index(begin = var_12125_begin_0, end = var_12125_end_0, end_mask = var_12125_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12125_cast")];
            tensor<int32, [4]> var_12129_begin_0 = const()[name = tensor<string, []>("op_12129_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_12129_end_0 = const()[name = tensor<string, []>("op_12129_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_12129_end_mask_0 = const()[name = tensor<string, []>("op_12129_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12129_cast = slice_by_index(begin = var_12129_begin_0, end = var_12129_end_0, end_mask = var_12129_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12129_cast")];
            tensor<int32, [4]> var_12133_begin_0 = const()[name = tensor<string, []>("op_12133_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_12133_end_0 = const()[name = tensor<string, []>("op_12133_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_12133_end_mask_0 = const()[name = tensor<string, []>("op_12133_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12133_cast = slice_by_index(begin = var_12133_begin_0, end = var_12133_end_0, end_mask = var_12133_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12133_cast")];
            tensor<int32, [4]> var_12137_begin_0 = const()[name = tensor<string, []>("op_12137_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_12137_end_0 = const()[name = tensor<string, []>("op_12137_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_12137_end_mask_0 = const()[name = tensor<string, []>("op_12137_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12137_cast = slice_by_index(begin = var_12137_begin_0, end = var_12137_end_0, end_mask = var_12137_end_mask_0, x = transpose_111)[name = tensor<string, []>("op_12137_cast")];
            tensor<int32, [4]> var_12139_begin_0 = const()[name = tensor<string, []>("op_12139_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12139_end_0 = const()[name = tensor<string, []>("op_12139_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12139_end_mask_0 = const()[name = tensor<string, []>("op_12139_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12139_cast = slice_by_index(begin = var_12139_begin_0, end = var_12139_end_0, end_mask = var_12139_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12139_cast")];
            tensor<int32, [4]> var_12143_begin_0 = const()[name = tensor<string, []>("op_12143_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_12143_end_0 = const()[name = tensor<string, []>("op_12143_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_12143_end_mask_0 = const()[name = tensor<string, []>("op_12143_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12143_cast = slice_by_index(begin = var_12143_begin_0, end = var_12143_end_0, end_mask = var_12143_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12143_cast")];
            tensor<int32, [4]> var_12147_begin_0 = const()[name = tensor<string, []>("op_12147_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_12147_end_0 = const()[name = tensor<string, []>("op_12147_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_12147_end_mask_0 = const()[name = tensor<string, []>("op_12147_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12147_cast = slice_by_index(begin = var_12147_begin_0, end = var_12147_end_0, end_mask = var_12147_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12147_cast")];
            tensor<int32, [4]> var_12151_begin_0 = const()[name = tensor<string, []>("op_12151_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_12151_end_0 = const()[name = tensor<string, []>("op_12151_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_12151_end_mask_0 = const()[name = tensor<string, []>("op_12151_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12151_cast = slice_by_index(begin = var_12151_begin_0, end = var_12151_end_0, end_mask = var_12151_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12151_cast")];
            tensor<int32, [4]> var_12155_begin_0 = const()[name = tensor<string, []>("op_12155_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_12155_end_0 = const()[name = tensor<string, []>("op_12155_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_12155_end_mask_0 = const()[name = tensor<string, []>("op_12155_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12155_cast = slice_by_index(begin = var_12155_begin_0, end = var_12155_end_0, end_mask = var_12155_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12155_cast")];
            tensor<int32, [4]> var_12159_begin_0 = const()[name = tensor<string, []>("op_12159_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_12159_end_0 = const()[name = tensor<string, []>("op_12159_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_12159_end_mask_0 = const()[name = tensor<string, []>("op_12159_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12159_cast = slice_by_index(begin = var_12159_begin_0, end = var_12159_end_0, end_mask = var_12159_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12159_cast")];
            tensor<int32, [4]> var_12163_begin_0 = const()[name = tensor<string, []>("op_12163_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_12163_end_0 = const()[name = tensor<string, []>("op_12163_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_12163_end_mask_0 = const()[name = tensor<string, []>("op_12163_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12163_cast = slice_by_index(begin = var_12163_begin_0, end = var_12163_end_0, end_mask = var_12163_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12163_cast")];
            tensor<int32, [4]> var_12167_begin_0 = const()[name = tensor<string, []>("op_12167_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_12167_end_0 = const()[name = tensor<string, []>("op_12167_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_12167_end_mask_0 = const()[name = tensor<string, []>("op_12167_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12167_cast = slice_by_index(begin = var_12167_begin_0, end = var_12167_end_0, end_mask = var_12167_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12167_cast")];
            tensor<int32, [4]> var_12171_begin_0 = const()[name = tensor<string, []>("op_12171_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_12171_end_0 = const()[name = tensor<string, []>("op_12171_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_12171_end_mask_0 = const()[name = tensor<string, []>("op_12171_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12171_cast = slice_by_index(begin = var_12171_begin_0, end = var_12171_end_0, end_mask = var_12171_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12171_cast")];
            tensor<int32, [4]> var_12175_begin_0 = const()[name = tensor<string, []>("op_12175_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_12175_end_0 = const()[name = tensor<string, []>("op_12175_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_12175_end_mask_0 = const()[name = tensor<string, []>("op_12175_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12175_cast = slice_by_index(begin = var_12175_begin_0, end = var_12175_end_0, end_mask = var_12175_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12175_cast")];
            tensor<int32, [4]> var_12179_begin_0 = const()[name = tensor<string, []>("op_12179_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_12179_end_0 = const()[name = tensor<string, []>("op_12179_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_12179_end_mask_0 = const()[name = tensor<string, []>("op_12179_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12179_cast = slice_by_index(begin = var_12179_begin_0, end = var_12179_end_0, end_mask = var_12179_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12179_cast")];
            tensor<int32, [4]> var_12183_begin_0 = const()[name = tensor<string, []>("op_12183_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_12183_end_0 = const()[name = tensor<string, []>("op_12183_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_12183_end_mask_0 = const()[name = tensor<string, []>("op_12183_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12183_cast = slice_by_index(begin = var_12183_begin_0, end = var_12183_end_0, end_mask = var_12183_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12183_cast")];
            tensor<int32, [4]> var_12187_begin_0 = const()[name = tensor<string, []>("op_12187_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_12187_end_0 = const()[name = tensor<string, []>("op_12187_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_12187_end_mask_0 = const()[name = tensor<string, []>("op_12187_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12187_cast = slice_by_index(begin = var_12187_begin_0, end = var_12187_end_0, end_mask = var_12187_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12187_cast")];
            tensor<int32, [4]> var_12191_begin_0 = const()[name = tensor<string, []>("op_12191_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_12191_end_0 = const()[name = tensor<string, []>("op_12191_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_12191_end_mask_0 = const()[name = tensor<string, []>("op_12191_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12191_cast = slice_by_index(begin = var_12191_begin_0, end = var_12191_end_0, end_mask = var_12191_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12191_cast")];
            tensor<int32, [4]> var_12195_begin_0 = const()[name = tensor<string, []>("op_12195_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_12195_end_0 = const()[name = tensor<string, []>("op_12195_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_12195_end_mask_0 = const()[name = tensor<string, []>("op_12195_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12195_cast = slice_by_index(begin = var_12195_begin_0, end = var_12195_end_0, end_mask = var_12195_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12195_cast")];
            tensor<int32, [4]> var_12199_begin_0 = const()[name = tensor<string, []>("op_12199_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_12199_end_0 = const()[name = tensor<string, []>("op_12199_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_12199_end_mask_0 = const()[name = tensor<string, []>("op_12199_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12199_cast = slice_by_index(begin = var_12199_begin_0, end = var_12199_end_0, end_mask = var_12199_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12199_cast")];
            tensor<int32, [4]> var_12203_begin_0 = const()[name = tensor<string, []>("op_12203_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_12203_end_0 = const()[name = tensor<string, []>("op_12203_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_12203_end_mask_0 = const()[name = tensor<string, []>("op_12203_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12203_cast = slice_by_index(begin = var_12203_begin_0, end = var_12203_end_0, end_mask = var_12203_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12203_cast")];
            tensor<int32, [4]> var_12207_begin_0 = const()[name = tensor<string, []>("op_12207_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_12207_end_0 = const()[name = tensor<string, []>("op_12207_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_12207_end_mask_0 = const()[name = tensor<string, []>("op_12207_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12207_cast = slice_by_index(begin = var_12207_begin_0, end = var_12207_end_0, end_mask = var_12207_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12207_cast")];
            tensor<int32, [4]> var_12211_begin_0 = const()[name = tensor<string, []>("op_12211_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_12211_end_0 = const()[name = tensor<string, []>("op_12211_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_12211_end_mask_0 = const()[name = tensor<string, []>("op_12211_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12211_cast = slice_by_index(begin = var_12211_begin_0, end = var_12211_end_0, end_mask = var_12211_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12211_cast")];
            tensor<int32, [4]> var_12215_begin_0 = const()[name = tensor<string, []>("op_12215_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_12215_end_0 = const()[name = tensor<string, []>("op_12215_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_12215_end_mask_0 = const()[name = tensor<string, []>("op_12215_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12215_cast = slice_by_index(begin = var_12215_begin_0, end = var_12215_end_0, end_mask = var_12215_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12215_cast")];
            tensor<string, []> var_12219_equation_0 = const()[name = tensor<string, []>("op_12219_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12219_cast = einsum(equation = var_12219_equation_0, values = (var_12061_cast, var_11978_cast))[name = tensor<string, []>("op_12219_cast")];
            tensor<fp16, []> var_12220_to_fp16 = const()[name = tensor<string, []>("op_12220_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_961_cast = mul(x = var_12219_cast, y = var_12220_to_fp16)[name = tensor<string, []>("aw_961_cast")];
            tensor<string, []> var_12223_equation_0 = const()[name = tensor<string, []>("op_12223_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12223_cast = einsum(equation = var_12223_equation_0, values = (var_12065_cast, var_11982_cast))[name = tensor<string, []>("op_12223_cast")];
            tensor<fp16, []> var_12224_to_fp16 = const()[name = tensor<string, []>("op_12224_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_963_cast = mul(x = var_12223_cast, y = var_12224_to_fp16)[name = tensor<string, []>("aw_963_cast")];
            tensor<string, []> var_12227_equation_0 = const()[name = tensor<string, []>("op_12227_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12227_cast = einsum(equation = var_12227_equation_0, values = (var_12069_cast, var_11986_cast))[name = tensor<string, []>("op_12227_cast")];
            tensor<fp16, []> var_12228_to_fp16 = const()[name = tensor<string, []>("op_12228_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_965_cast = mul(x = var_12227_cast, y = var_12228_to_fp16)[name = tensor<string, []>("aw_965_cast")];
            tensor<string, []> var_12231_equation_0 = const()[name = tensor<string, []>("op_12231_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12231_cast = einsum(equation = var_12231_equation_0, values = (var_12073_cast, var_11990_cast))[name = tensor<string, []>("op_12231_cast")];
            tensor<fp16, []> var_12232_to_fp16 = const()[name = tensor<string, []>("op_12232_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_967_cast = mul(x = var_12231_cast, y = var_12232_to_fp16)[name = tensor<string, []>("aw_967_cast")];
            tensor<string, []> var_12235_equation_0 = const()[name = tensor<string, []>("op_12235_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12235_cast = einsum(equation = var_12235_equation_0, values = (var_12077_cast, var_11994_cast))[name = tensor<string, []>("op_12235_cast")];
            tensor<fp16, []> var_12236_to_fp16 = const()[name = tensor<string, []>("op_12236_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_969_cast = mul(x = var_12235_cast, y = var_12236_to_fp16)[name = tensor<string, []>("aw_969_cast")];
            tensor<string, []> var_12239_equation_0 = const()[name = tensor<string, []>("op_12239_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12239_cast = einsum(equation = var_12239_equation_0, values = (var_12081_cast, var_11998_cast))[name = tensor<string, []>("op_12239_cast")];
            tensor<fp16, []> var_12240_to_fp16 = const()[name = tensor<string, []>("op_12240_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_971_cast = mul(x = var_12239_cast, y = var_12240_to_fp16)[name = tensor<string, []>("aw_971_cast")];
            tensor<string, []> var_12243_equation_0 = const()[name = tensor<string, []>("op_12243_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12243_cast = einsum(equation = var_12243_equation_0, values = (var_12085_cast, var_12002_cast))[name = tensor<string, []>("op_12243_cast")];
            tensor<fp16, []> var_12244_to_fp16 = const()[name = tensor<string, []>("op_12244_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_973_cast = mul(x = var_12243_cast, y = var_12244_to_fp16)[name = tensor<string, []>("aw_973_cast")];
            tensor<string, []> var_12247_equation_0 = const()[name = tensor<string, []>("op_12247_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12247_cast = einsum(equation = var_12247_equation_0, values = (var_12089_cast, var_12006_cast))[name = tensor<string, []>("op_12247_cast")];
            tensor<fp16, []> var_12248_to_fp16 = const()[name = tensor<string, []>("op_12248_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_975_cast = mul(x = var_12247_cast, y = var_12248_to_fp16)[name = tensor<string, []>("aw_975_cast")];
            tensor<string, []> var_12251_equation_0 = const()[name = tensor<string, []>("op_12251_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12251_cast = einsum(equation = var_12251_equation_0, values = (var_12093_cast, var_12010_cast))[name = tensor<string, []>("op_12251_cast")];
            tensor<fp16, []> var_12252_to_fp16 = const()[name = tensor<string, []>("op_12252_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_977_cast = mul(x = var_12251_cast, y = var_12252_to_fp16)[name = tensor<string, []>("aw_977_cast")];
            tensor<string, []> var_12255_equation_0 = const()[name = tensor<string, []>("op_12255_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12255_cast = einsum(equation = var_12255_equation_0, values = (var_12097_cast, var_12014_cast))[name = tensor<string, []>("op_12255_cast")];
            tensor<fp16, []> var_12256_to_fp16 = const()[name = tensor<string, []>("op_12256_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_979_cast = mul(x = var_12255_cast, y = var_12256_to_fp16)[name = tensor<string, []>("aw_979_cast")];
            tensor<string, []> var_12259_equation_0 = const()[name = tensor<string, []>("op_12259_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12259_cast = einsum(equation = var_12259_equation_0, values = (var_12101_cast, var_12018_cast))[name = tensor<string, []>("op_12259_cast")];
            tensor<fp16, []> var_12260_to_fp16 = const()[name = tensor<string, []>("op_12260_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_981_cast = mul(x = var_12259_cast, y = var_12260_to_fp16)[name = tensor<string, []>("aw_981_cast")];
            tensor<string, []> var_12263_equation_0 = const()[name = tensor<string, []>("op_12263_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12263_cast = einsum(equation = var_12263_equation_0, values = (var_12105_cast, var_12022_cast))[name = tensor<string, []>("op_12263_cast")];
            tensor<fp16, []> var_12264_to_fp16 = const()[name = tensor<string, []>("op_12264_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_983_cast = mul(x = var_12263_cast, y = var_12264_to_fp16)[name = tensor<string, []>("aw_983_cast")];
            tensor<string, []> var_12267_equation_0 = const()[name = tensor<string, []>("op_12267_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12267_cast = einsum(equation = var_12267_equation_0, values = (var_12109_cast, var_12026_cast))[name = tensor<string, []>("op_12267_cast")];
            tensor<fp16, []> var_12268_to_fp16 = const()[name = tensor<string, []>("op_12268_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_985_cast = mul(x = var_12267_cast, y = var_12268_to_fp16)[name = tensor<string, []>("aw_985_cast")];
            tensor<string, []> var_12271_equation_0 = const()[name = tensor<string, []>("op_12271_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12271_cast = einsum(equation = var_12271_equation_0, values = (var_12113_cast, var_12030_cast))[name = tensor<string, []>("op_12271_cast")];
            tensor<fp16, []> var_12272_to_fp16 = const()[name = tensor<string, []>("op_12272_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_987_cast = mul(x = var_12271_cast, y = var_12272_to_fp16)[name = tensor<string, []>("aw_987_cast")];
            tensor<string, []> var_12275_equation_0 = const()[name = tensor<string, []>("op_12275_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12275_cast = einsum(equation = var_12275_equation_0, values = (var_12117_cast, var_12034_cast))[name = tensor<string, []>("op_12275_cast")];
            tensor<fp16, []> var_12276_to_fp16 = const()[name = tensor<string, []>("op_12276_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_989_cast = mul(x = var_12275_cast, y = var_12276_to_fp16)[name = tensor<string, []>("aw_989_cast")];
            tensor<string, []> var_12279_equation_0 = const()[name = tensor<string, []>("op_12279_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12279_cast = einsum(equation = var_12279_equation_0, values = (var_12121_cast, var_12038_cast))[name = tensor<string, []>("op_12279_cast")];
            tensor<fp16, []> var_12280_to_fp16 = const()[name = tensor<string, []>("op_12280_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_991_cast = mul(x = var_12279_cast, y = var_12280_to_fp16)[name = tensor<string, []>("aw_991_cast")];
            tensor<string, []> var_12283_equation_0 = const()[name = tensor<string, []>("op_12283_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12283_cast = einsum(equation = var_12283_equation_0, values = (var_12125_cast, var_12042_cast))[name = tensor<string, []>("op_12283_cast")];
            tensor<fp16, []> var_12284_to_fp16 = const()[name = tensor<string, []>("op_12284_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_993_cast = mul(x = var_12283_cast, y = var_12284_to_fp16)[name = tensor<string, []>("aw_993_cast")];
            tensor<string, []> var_12287_equation_0 = const()[name = tensor<string, []>("op_12287_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12287_cast = einsum(equation = var_12287_equation_0, values = (var_12129_cast, var_12046_cast))[name = tensor<string, []>("op_12287_cast")];
            tensor<fp16, []> var_12288_to_fp16 = const()[name = tensor<string, []>("op_12288_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_995_cast = mul(x = var_12287_cast, y = var_12288_to_fp16)[name = tensor<string, []>("aw_995_cast")];
            tensor<string, []> var_12291_equation_0 = const()[name = tensor<string, []>("op_12291_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12291_cast = einsum(equation = var_12291_equation_0, values = (var_12133_cast, var_12050_cast))[name = tensor<string, []>("op_12291_cast")];
            tensor<fp16, []> var_12292_to_fp16 = const()[name = tensor<string, []>("op_12292_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_997_cast = mul(x = var_12291_cast, y = var_12292_to_fp16)[name = tensor<string, []>("aw_997_cast")];
            tensor<string, []> var_12295_equation_0 = const()[name = tensor<string, []>("op_12295_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12295_cast = einsum(equation = var_12295_equation_0, values = (var_12137_cast, var_12054_cast))[name = tensor<string, []>("op_12295_cast")];
            tensor<fp16, []> var_12296_to_fp16 = const()[name = tensor<string, []>("op_12296_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_999_cast = mul(x = var_12295_cast, y = var_12296_to_fp16)[name = tensor<string, []>("aw_999_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12298_cast = softmax(axis = var_2634, x = aw_961_cast)[name = tensor<string, []>("op_12298_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12299_cast = softmax(axis = var_2634, x = aw_963_cast)[name = tensor<string, []>("op_12299_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12300_cast = softmax(axis = var_2634, x = aw_965_cast)[name = tensor<string, []>("op_12300_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12301_cast = softmax(axis = var_2634, x = aw_967_cast)[name = tensor<string, []>("op_12301_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12302_cast = softmax(axis = var_2634, x = aw_969_cast)[name = tensor<string, []>("op_12302_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12303_cast = softmax(axis = var_2634, x = aw_971_cast)[name = tensor<string, []>("op_12303_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12304_cast = softmax(axis = var_2634, x = aw_973_cast)[name = tensor<string, []>("op_12304_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12305_cast = softmax(axis = var_2634, x = aw_975_cast)[name = tensor<string, []>("op_12305_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12306_cast = softmax(axis = var_2634, x = aw_977_cast)[name = tensor<string, []>("op_12306_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12307_cast = softmax(axis = var_2634, x = aw_979_cast)[name = tensor<string, []>("op_12307_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12308_cast = softmax(axis = var_2634, x = aw_981_cast)[name = tensor<string, []>("op_12308_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12309_cast = softmax(axis = var_2634, x = aw_983_cast)[name = tensor<string, []>("op_12309_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12310_cast = softmax(axis = var_2634, x = aw_985_cast)[name = tensor<string, []>("op_12310_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12311_cast = softmax(axis = var_2634, x = aw_987_cast)[name = tensor<string, []>("op_12311_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12312_cast = softmax(axis = var_2634, x = aw_989_cast)[name = tensor<string, []>("op_12312_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12313_cast = softmax(axis = var_2634, x = aw_991_cast)[name = tensor<string, []>("op_12313_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12314_cast = softmax(axis = var_2634, x = aw_993_cast)[name = tensor<string, []>("op_12314_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12315_cast = softmax(axis = var_2634, x = aw_995_cast)[name = tensor<string, []>("op_12315_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12316_cast = softmax(axis = var_2634, x = aw_997_cast)[name = tensor<string, []>("op_12316_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_12317_cast = softmax(axis = var_2634, x = aw_999_cast)[name = tensor<string, []>("op_12317_cast")];
            tensor<string, []> var_12319_equation_0 = const()[name = tensor<string, []>("op_12319_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12319_cast = einsum(equation = var_12319_equation_0, values = (var_12139_cast, var_12298_cast))[name = tensor<string, []>("op_12319_cast")];
            tensor<string, []> var_12321_equation_0 = const()[name = tensor<string, []>("op_12321_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12321_cast = einsum(equation = var_12321_equation_0, values = (var_12143_cast, var_12299_cast))[name = tensor<string, []>("op_12321_cast")];
            tensor<string, []> var_12323_equation_0 = const()[name = tensor<string, []>("op_12323_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12323_cast = einsum(equation = var_12323_equation_0, values = (var_12147_cast, var_12300_cast))[name = tensor<string, []>("op_12323_cast")];
            tensor<string, []> var_12325_equation_0 = const()[name = tensor<string, []>("op_12325_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12325_cast = einsum(equation = var_12325_equation_0, values = (var_12151_cast, var_12301_cast))[name = tensor<string, []>("op_12325_cast")];
            tensor<string, []> var_12327_equation_0 = const()[name = tensor<string, []>("op_12327_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12327_cast = einsum(equation = var_12327_equation_0, values = (var_12155_cast, var_12302_cast))[name = tensor<string, []>("op_12327_cast")];
            tensor<string, []> var_12329_equation_0 = const()[name = tensor<string, []>("op_12329_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12329_cast = einsum(equation = var_12329_equation_0, values = (var_12159_cast, var_12303_cast))[name = tensor<string, []>("op_12329_cast")];
            tensor<string, []> var_12331_equation_0 = const()[name = tensor<string, []>("op_12331_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12331_cast = einsum(equation = var_12331_equation_0, values = (var_12163_cast, var_12304_cast))[name = tensor<string, []>("op_12331_cast")];
            tensor<string, []> var_12333_equation_0 = const()[name = tensor<string, []>("op_12333_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12333_cast = einsum(equation = var_12333_equation_0, values = (var_12167_cast, var_12305_cast))[name = tensor<string, []>("op_12333_cast")];
            tensor<string, []> var_12335_equation_0 = const()[name = tensor<string, []>("op_12335_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12335_cast = einsum(equation = var_12335_equation_0, values = (var_12171_cast, var_12306_cast))[name = tensor<string, []>("op_12335_cast")];
            tensor<string, []> var_12337_equation_0 = const()[name = tensor<string, []>("op_12337_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12337_cast = einsum(equation = var_12337_equation_0, values = (var_12175_cast, var_12307_cast))[name = tensor<string, []>("op_12337_cast")];
            tensor<string, []> var_12339_equation_0 = const()[name = tensor<string, []>("op_12339_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12339_cast = einsum(equation = var_12339_equation_0, values = (var_12179_cast, var_12308_cast))[name = tensor<string, []>("op_12339_cast")];
            tensor<string, []> var_12341_equation_0 = const()[name = tensor<string, []>("op_12341_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12341_cast = einsum(equation = var_12341_equation_0, values = (var_12183_cast, var_12309_cast))[name = tensor<string, []>("op_12341_cast")];
            tensor<string, []> var_12343_equation_0 = const()[name = tensor<string, []>("op_12343_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12343_cast = einsum(equation = var_12343_equation_0, values = (var_12187_cast, var_12310_cast))[name = tensor<string, []>("op_12343_cast")];
            tensor<string, []> var_12345_equation_0 = const()[name = tensor<string, []>("op_12345_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12345_cast = einsum(equation = var_12345_equation_0, values = (var_12191_cast, var_12311_cast))[name = tensor<string, []>("op_12345_cast")];
            tensor<string, []> var_12347_equation_0 = const()[name = tensor<string, []>("op_12347_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12347_cast = einsum(equation = var_12347_equation_0, values = (var_12195_cast, var_12312_cast))[name = tensor<string, []>("op_12347_cast")];
            tensor<string, []> var_12349_equation_0 = const()[name = tensor<string, []>("op_12349_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12349_cast = einsum(equation = var_12349_equation_0, values = (var_12199_cast, var_12313_cast))[name = tensor<string, []>("op_12349_cast")];
            tensor<string, []> var_12351_equation_0 = const()[name = tensor<string, []>("op_12351_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12351_cast = einsum(equation = var_12351_equation_0, values = (var_12203_cast, var_12314_cast))[name = tensor<string, []>("op_12351_cast")];
            tensor<string, []> var_12353_equation_0 = const()[name = tensor<string, []>("op_12353_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12353_cast = einsum(equation = var_12353_equation_0, values = (var_12207_cast, var_12315_cast))[name = tensor<string, []>("op_12353_cast")];
            tensor<string, []> var_12355_equation_0 = const()[name = tensor<string, []>("op_12355_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12355_cast = einsum(equation = var_12355_equation_0, values = (var_12211_cast, var_12316_cast))[name = tensor<string, []>("op_12355_cast")];
            tensor<string, []> var_12357_equation_0 = const()[name = tensor<string, []>("op_12357_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12357_cast = einsum(equation = var_12357_equation_0, values = (var_12215_cast, var_12317_cast))[name = tensor<string, []>("op_12357_cast")];
            tensor<bool, []> input_233_interleave_0 = const()[name = tensor<string, []>("input_233_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_233_cast = concat(axis = var_2634, interleave = input_233_interleave_0, values = (var_12319_cast, var_12321_cast, var_12323_cast, var_12325_cast, var_12327_cast, var_12329_cast, var_12331_cast, var_12333_cast, var_12335_cast, var_12337_cast, var_12339_cast, var_12341_cast, var_12343_cast, var_12345_cast, var_12347_cast, var_12349_cast, var_12351_cast, var_12353_cast, var_12355_cast, var_12357_cast))[name = tensor<string, []>("input_233_cast")];
            tensor<int32, [2]> var_12363 = const()[name = tensor<string, []>("op_12363"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12365 = const()[name = tensor<string, []>("op_12365"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_12367_pad_type_0 = const()[name = tensor<string, []>("op_12367_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_12367_pad_0 = const()[name = tensor<string, []>("op_12367_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(971973312)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(975250176)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_12367_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_12365, groups = var_2634, pad = var_12367_pad_0, pad_type = var_12367_pad_type_0, strides = var_12363, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_233_cast)[name = tensor<string, []>("op_12367_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_87_cast = add(x = var_12367_cast, y = inputs_85_cast)[name = tensor<string, []>("inputs_87_cast")];
            tensor<int32, [1]> var_12371 = const()[name = tensor<string, []>("op_12371"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_87_cast = reduce_mean(axes = var_12371, keep_dims = var_2629, x = inputs_87_cast)[name = tensor<string, []>("channels_mean_87_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_87_cast = sub(x = inputs_87_cast, y = channels_mean_87_cast)[name = tensor<string, []>("zero_mean_87_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_87_cast = mul(x = zero_mean_87_cast, y = zero_mean_87_cast)[name = tensor<string, []>("zero_mean_sq_87_cast")];
            tensor<int32, [1]> var_12375 = const()[name = tensor<string, []>("op_12375"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_12376_cast = reduce_mean(axes = var_12375, keep_dims = var_2629, x = zero_mean_sq_87_cast)[name = tensor<string, []>("op_12376_cast")];
            tensor<fp16, []> var_12377_to_fp16 = const()[name = tensor<string, []>("op_12377_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_12378_cast = add(x = var_12376_cast, y = var_12377_to_fp16)[name = tensor<string, []>("op_12378_cast")];
            tensor<fp16, []> denom_87_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_87_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_87_cast = rsqrt(epsilon = denom_87_epsilon_0_to_fp16, x = var_12378_cast)[name = tensor<string, []>("denom_87_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_87_cast = mul(x = zero_mean_87_cast, y = denom_87_cast)[name = tensor<string, []>("out_87_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_12382_to_fp16 = const()[name = tensor<string, []>("op_12382_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(975252800)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_12383_cast = add(x = out_87_cast, y = var_12382_to_fp16)[name = tensor<string, []>("op_12383_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_12385_to_fp16 = const()[name = tensor<string, []>("op_12385_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(975255424)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_139_cast = mul(x = var_12383_cast, y = var_12385_to_fp16)[name = tensor<string, []>("hidden_states_139_cast")];
            tensor<int32, [2]> var_12392 = const()[name = tensor<string, []>("op_12392"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12394 = const()[name = tensor<string, []>("op_12394"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_59_pad_type_0 = const()[name = tensor<string, []>("q_59_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_59_pad_0 = const()[name = tensor<string, []>("q_59_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(975258048)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_59_cast = conv(dilations = var_12394, groups = var_2634, pad = q_59_pad_0, pad_type = q_59_pad_type_0, strides = var_12392, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_139_cast)[name = tensor<string, []>("q_59_cast")];
            tensor<int32, [2]> var_12398 = const()[name = tensor<string, []>("op_12398"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12400 = const()[name = tensor<string, []>("op_12400"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_117_pad_type_0 = const()[name = tensor<string, []>("k_117_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_117_pad_0 = const()[name = tensor<string, []>("k_117_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(978534912)))];
            tensor<fp16, [2, 1280, 1, 77]> k_117_cast = conv(dilations = var_12400, groups = var_2634, pad = k_117_pad_0, pad_type = k_117_pad_type_0, strides = var_12398, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_117_cast")];
            tensor<int32, [2]> var_12404 = const()[name = tensor<string, []>("op_12404"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12406 = const()[name = tensor<string, []>("op_12406"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_59_pad_type_0 = const()[name = tensor<string, []>("v_59_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_59_pad_0 = const()[name = tensor<string, []>("v_59_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(983777856)))];
            tensor<fp16, [2, 1280, 1, 77]> v_59_cast = conv(dilations = var_12406, groups = var_2634, pad = v_59_pad_0, pad_type = v_59_pad_type_0, strides = var_12404, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_59_cast")];
            tensor<int32, [4]> var_12410_begin_0 = const()[name = tensor<string, []>("op_12410_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12410_end_0 = const()[name = tensor<string, []>("op_12410_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12410_end_mask_0 = const()[name = tensor<string, []>("op_12410_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12410_cast = slice_by_index(begin = var_12410_begin_0, end = var_12410_end_0, end_mask = var_12410_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12410_cast")];
            tensor<int32, [4]> var_12414_begin_0 = const()[name = tensor<string, []>("op_12414_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_12414_end_0 = const()[name = tensor<string, []>("op_12414_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_12414_end_mask_0 = const()[name = tensor<string, []>("op_12414_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12414_cast = slice_by_index(begin = var_12414_begin_0, end = var_12414_end_0, end_mask = var_12414_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12414_cast")];
            tensor<int32, [4]> var_12418_begin_0 = const()[name = tensor<string, []>("op_12418_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_12418_end_0 = const()[name = tensor<string, []>("op_12418_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_12418_end_mask_0 = const()[name = tensor<string, []>("op_12418_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12418_cast = slice_by_index(begin = var_12418_begin_0, end = var_12418_end_0, end_mask = var_12418_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12418_cast")];
            tensor<int32, [4]> var_12422_begin_0 = const()[name = tensor<string, []>("op_12422_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_12422_end_0 = const()[name = tensor<string, []>("op_12422_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_12422_end_mask_0 = const()[name = tensor<string, []>("op_12422_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12422_cast = slice_by_index(begin = var_12422_begin_0, end = var_12422_end_0, end_mask = var_12422_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12422_cast")];
            tensor<int32, [4]> var_12426_begin_0 = const()[name = tensor<string, []>("op_12426_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_12426_end_0 = const()[name = tensor<string, []>("op_12426_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_12426_end_mask_0 = const()[name = tensor<string, []>("op_12426_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12426_cast = slice_by_index(begin = var_12426_begin_0, end = var_12426_end_0, end_mask = var_12426_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12426_cast")];
            tensor<int32, [4]> var_12430_begin_0 = const()[name = tensor<string, []>("op_12430_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_12430_end_0 = const()[name = tensor<string, []>("op_12430_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_12430_end_mask_0 = const()[name = tensor<string, []>("op_12430_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12430_cast = slice_by_index(begin = var_12430_begin_0, end = var_12430_end_0, end_mask = var_12430_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12430_cast")];
            tensor<int32, [4]> var_12434_begin_0 = const()[name = tensor<string, []>("op_12434_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_12434_end_0 = const()[name = tensor<string, []>("op_12434_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_12434_end_mask_0 = const()[name = tensor<string, []>("op_12434_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12434_cast = slice_by_index(begin = var_12434_begin_0, end = var_12434_end_0, end_mask = var_12434_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12434_cast")];
            tensor<int32, [4]> var_12438_begin_0 = const()[name = tensor<string, []>("op_12438_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_12438_end_0 = const()[name = tensor<string, []>("op_12438_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_12438_end_mask_0 = const()[name = tensor<string, []>("op_12438_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12438_cast = slice_by_index(begin = var_12438_begin_0, end = var_12438_end_0, end_mask = var_12438_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12438_cast")];
            tensor<int32, [4]> var_12442_begin_0 = const()[name = tensor<string, []>("op_12442_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_12442_end_0 = const()[name = tensor<string, []>("op_12442_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_12442_end_mask_0 = const()[name = tensor<string, []>("op_12442_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12442_cast = slice_by_index(begin = var_12442_begin_0, end = var_12442_end_0, end_mask = var_12442_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12442_cast")];
            tensor<int32, [4]> var_12446_begin_0 = const()[name = tensor<string, []>("op_12446_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_12446_end_0 = const()[name = tensor<string, []>("op_12446_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_12446_end_mask_0 = const()[name = tensor<string, []>("op_12446_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12446_cast = slice_by_index(begin = var_12446_begin_0, end = var_12446_end_0, end_mask = var_12446_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12446_cast")];
            tensor<int32, [4]> var_12450_begin_0 = const()[name = tensor<string, []>("op_12450_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_12450_end_0 = const()[name = tensor<string, []>("op_12450_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_12450_end_mask_0 = const()[name = tensor<string, []>("op_12450_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12450_cast = slice_by_index(begin = var_12450_begin_0, end = var_12450_end_0, end_mask = var_12450_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12450_cast")];
            tensor<int32, [4]> var_12454_begin_0 = const()[name = tensor<string, []>("op_12454_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_12454_end_0 = const()[name = tensor<string, []>("op_12454_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_12454_end_mask_0 = const()[name = tensor<string, []>("op_12454_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12454_cast = slice_by_index(begin = var_12454_begin_0, end = var_12454_end_0, end_mask = var_12454_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12454_cast")];
            tensor<int32, [4]> var_12458_begin_0 = const()[name = tensor<string, []>("op_12458_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_12458_end_0 = const()[name = tensor<string, []>("op_12458_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_12458_end_mask_0 = const()[name = tensor<string, []>("op_12458_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12458_cast = slice_by_index(begin = var_12458_begin_0, end = var_12458_end_0, end_mask = var_12458_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12458_cast")];
            tensor<int32, [4]> var_12462_begin_0 = const()[name = tensor<string, []>("op_12462_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_12462_end_0 = const()[name = tensor<string, []>("op_12462_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_12462_end_mask_0 = const()[name = tensor<string, []>("op_12462_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12462_cast = slice_by_index(begin = var_12462_begin_0, end = var_12462_end_0, end_mask = var_12462_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12462_cast")];
            tensor<int32, [4]> var_12466_begin_0 = const()[name = tensor<string, []>("op_12466_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_12466_end_0 = const()[name = tensor<string, []>("op_12466_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_12466_end_mask_0 = const()[name = tensor<string, []>("op_12466_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12466_cast = slice_by_index(begin = var_12466_begin_0, end = var_12466_end_0, end_mask = var_12466_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12466_cast")];
            tensor<int32, [4]> var_12470_begin_0 = const()[name = tensor<string, []>("op_12470_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_12470_end_0 = const()[name = tensor<string, []>("op_12470_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_12470_end_mask_0 = const()[name = tensor<string, []>("op_12470_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12470_cast = slice_by_index(begin = var_12470_begin_0, end = var_12470_end_0, end_mask = var_12470_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12470_cast")];
            tensor<int32, [4]> var_12474_begin_0 = const()[name = tensor<string, []>("op_12474_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_12474_end_0 = const()[name = tensor<string, []>("op_12474_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_12474_end_mask_0 = const()[name = tensor<string, []>("op_12474_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12474_cast = slice_by_index(begin = var_12474_begin_0, end = var_12474_end_0, end_mask = var_12474_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12474_cast")];
            tensor<int32, [4]> var_12478_begin_0 = const()[name = tensor<string, []>("op_12478_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_12478_end_0 = const()[name = tensor<string, []>("op_12478_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_12478_end_mask_0 = const()[name = tensor<string, []>("op_12478_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12478_cast = slice_by_index(begin = var_12478_begin_0, end = var_12478_end_0, end_mask = var_12478_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12478_cast")];
            tensor<int32, [4]> var_12482_begin_0 = const()[name = tensor<string, []>("op_12482_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_12482_end_0 = const()[name = tensor<string, []>("op_12482_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_12482_end_mask_0 = const()[name = tensor<string, []>("op_12482_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12482_cast = slice_by_index(begin = var_12482_begin_0, end = var_12482_end_0, end_mask = var_12482_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12482_cast")];
            tensor<int32, [4]> var_12486_begin_0 = const()[name = tensor<string, []>("op_12486_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_12486_end_0 = const()[name = tensor<string, []>("op_12486_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_12486_end_mask_0 = const()[name = tensor<string, []>("op_12486_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12486_cast = slice_by_index(begin = var_12486_begin_0, end = var_12486_end_0, end_mask = var_12486_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12486_cast")];
            tensor<int32, [4]> k_119_perm_0 = const()[name = tensor<string, []>("k_119_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_12493_begin_0 = const()[name = tensor<string, []>("op_12493_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12493_end_0 = const()[name = tensor<string, []>("op_12493_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_12493_end_mask_0 = const()[name = tensor<string, []>("op_12493_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_110 = transpose(perm = k_119_perm_0, x = k_117_cast)[name = tensor<string, []>("transpose_110")];
            tensor<fp16, [2, 77, 1, 64]> var_12493_cast = slice_by_index(begin = var_12493_begin_0, end = var_12493_end_0, end_mask = var_12493_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12493_cast")];
            tensor<int32, [4]> var_12497_begin_0 = const()[name = tensor<string, []>("op_12497_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_12497_end_0 = const()[name = tensor<string, []>("op_12497_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_12497_end_mask_0 = const()[name = tensor<string, []>("op_12497_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12497_cast = slice_by_index(begin = var_12497_begin_0, end = var_12497_end_0, end_mask = var_12497_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12497_cast")];
            tensor<int32, [4]> var_12501_begin_0 = const()[name = tensor<string, []>("op_12501_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_12501_end_0 = const()[name = tensor<string, []>("op_12501_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_12501_end_mask_0 = const()[name = tensor<string, []>("op_12501_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12501_cast = slice_by_index(begin = var_12501_begin_0, end = var_12501_end_0, end_mask = var_12501_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12501_cast")];
            tensor<int32, [4]> var_12505_begin_0 = const()[name = tensor<string, []>("op_12505_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_12505_end_0 = const()[name = tensor<string, []>("op_12505_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_12505_end_mask_0 = const()[name = tensor<string, []>("op_12505_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12505_cast = slice_by_index(begin = var_12505_begin_0, end = var_12505_end_0, end_mask = var_12505_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12505_cast")];
            tensor<int32, [4]> var_12509_begin_0 = const()[name = tensor<string, []>("op_12509_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_12509_end_0 = const()[name = tensor<string, []>("op_12509_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_12509_end_mask_0 = const()[name = tensor<string, []>("op_12509_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12509_cast = slice_by_index(begin = var_12509_begin_0, end = var_12509_end_0, end_mask = var_12509_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12509_cast")];
            tensor<int32, [4]> var_12513_begin_0 = const()[name = tensor<string, []>("op_12513_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_12513_end_0 = const()[name = tensor<string, []>("op_12513_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_12513_end_mask_0 = const()[name = tensor<string, []>("op_12513_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12513_cast = slice_by_index(begin = var_12513_begin_0, end = var_12513_end_0, end_mask = var_12513_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12513_cast")];
            tensor<int32, [4]> var_12517_begin_0 = const()[name = tensor<string, []>("op_12517_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_12517_end_0 = const()[name = tensor<string, []>("op_12517_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_12517_end_mask_0 = const()[name = tensor<string, []>("op_12517_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12517_cast = slice_by_index(begin = var_12517_begin_0, end = var_12517_end_0, end_mask = var_12517_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12517_cast")];
            tensor<int32, [4]> var_12521_begin_0 = const()[name = tensor<string, []>("op_12521_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_12521_end_0 = const()[name = tensor<string, []>("op_12521_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_12521_end_mask_0 = const()[name = tensor<string, []>("op_12521_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12521_cast = slice_by_index(begin = var_12521_begin_0, end = var_12521_end_0, end_mask = var_12521_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12521_cast")];
            tensor<int32, [4]> var_12525_begin_0 = const()[name = tensor<string, []>("op_12525_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12525_end_0 = const()[name = tensor<string, []>("op_12525_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_12525_end_mask_0 = const()[name = tensor<string, []>("op_12525_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12525_cast = slice_by_index(begin = var_12525_begin_0, end = var_12525_end_0, end_mask = var_12525_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12525_cast")];
            tensor<int32, [4]> var_12529_begin_0 = const()[name = tensor<string, []>("op_12529_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_12529_end_0 = const()[name = tensor<string, []>("op_12529_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_12529_end_mask_0 = const()[name = tensor<string, []>("op_12529_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12529_cast = slice_by_index(begin = var_12529_begin_0, end = var_12529_end_0, end_mask = var_12529_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12529_cast")];
            tensor<int32, [4]> var_12533_begin_0 = const()[name = tensor<string, []>("op_12533_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_12533_end_0 = const()[name = tensor<string, []>("op_12533_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_12533_end_mask_0 = const()[name = tensor<string, []>("op_12533_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12533_cast = slice_by_index(begin = var_12533_begin_0, end = var_12533_end_0, end_mask = var_12533_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12533_cast")];
            tensor<int32, [4]> var_12537_begin_0 = const()[name = tensor<string, []>("op_12537_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_12537_end_0 = const()[name = tensor<string, []>("op_12537_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_12537_end_mask_0 = const()[name = tensor<string, []>("op_12537_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12537_cast = slice_by_index(begin = var_12537_begin_0, end = var_12537_end_0, end_mask = var_12537_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12537_cast")];
            tensor<int32, [4]> var_12541_begin_0 = const()[name = tensor<string, []>("op_12541_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_12541_end_0 = const()[name = tensor<string, []>("op_12541_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_12541_end_mask_0 = const()[name = tensor<string, []>("op_12541_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12541_cast = slice_by_index(begin = var_12541_begin_0, end = var_12541_end_0, end_mask = var_12541_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12541_cast")];
            tensor<int32, [4]> var_12545_begin_0 = const()[name = tensor<string, []>("op_12545_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_12545_end_0 = const()[name = tensor<string, []>("op_12545_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_12545_end_mask_0 = const()[name = tensor<string, []>("op_12545_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12545_cast = slice_by_index(begin = var_12545_begin_0, end = var_12545_end_0, end_mask = var_12545_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12545_cast")];
            tensor<int32, [4]> var_12549_begin_0 = const()[name = tensor<string, []>("op_12549_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_12549_end_0 = const()[name = tensor<string, []>("op_12549_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_12549_end_mask_0 = const()[name = tensor<string, []>("op_12549_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12549_cast = slice_by_index(begin = var_12549_begin_0, end = var_12549_end_0, end_mask = var_12549_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12549_cast")];
            tensor<int32, [4]> var_12553_begin_0 = const()[name = tensor<string, []>("op_12553_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_12553_end_0 = const()[name = tensor<string, []>("op_12553_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_12553_end_mask_0 = const()[name = tensor<string, []>("op_12553_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12553_cast = slice_by_index(begin = var_12553_begin_0, end = var_12553_end_0, end_mask = var_12553_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12553_cast")];
            tensor<int32, [4]> var_12557_begin_0 = const()[name = tensor<string, []>("op_12557_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12557_end_0 = const()[name = tensor<string, []>("op_12557_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_12557_end_mask_0 = const()[name = tensor<string, []>("op_12557_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12557_cast = slice_by_index(begin = var_12557_begin_0, end = var_12557_end_0, end_mask = var_12557_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12557_cast")];
            tensor<int32, [4]> var_12561_begin_0 = const()[name = tensor<string, []>("op_12561_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_12561_end_0 = const()[name = tensor<string, []>("op_12561_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_12561_end_mask_0 = const()[name = tensor<string, []>("op_12561_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12561_cast = slice_by_index(begin = var_12561_begin_0, end = var_12561_end_0, end_mask = var_12561_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12561_cast")];
            tensor<int32, [4]> var_12565_begin_0 = const()[name = tensor<string, []>("op_12565_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_12565_end_0 = const()[name = tensor<string, []>("op_12565_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_12565_end_mask_0 = const()[name = tensor<string, []>("op_12565_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12565_cast = slice_by_index(begin = var_12565_begin_0, end = var_12565_end_0, end_mask = var_12565_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12565_cast")];
            tensor<int32, [4]> var_12569_begin_0 = const()[name = tensor<string, []>("op_12569_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_12569_end_0 = const()[name = tensor<string, []>("op_12569_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_12569_end_mask_0 = const()[name = tensor<string, []>("op_12569_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12569_cast = slice_by_index(begin = var_12569_begin_0, end = var_12569_end_0, end_mask = var_12569_end_mask_0, x = transpose_110)[name = tensor<string, []>("op_12569_cast")];
            tensor<int32, [4]> var_12571_begin_0 = const()[name = tensor<string, []>("op_12571_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12571_end_0 = const()[name = tensor<string, []>("op_12571_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_12571_end_mask_0 = const()[name = tensor<string, []>("op_12571_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12571_cast = slice_by_index(begin = var_12571_begin_0, end = var_12571_end_0, end_mask = var_12571_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12571_cast")];
            tensor<int32, [4]> var_12575_begin_0 = const()[name = tensor<string, []>("op_12575_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_12575_end_0 = const()[name = tensor<string, []>("op_12575_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_12575_end_mask_0 = const()[name = tensor<string, []>("op_12575_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12575_cast = slice_by_index(begin = var_12575_begin_0, end = var_12575_end_0, end_mask = var_12575_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12575_cast")];
            tensor<int32, [4]> var_12579_begin_0 = const()[name = tensor<string, []>("op_12579_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_12579_end_0 = const()[name = tensor<string, []>("op_12579_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_12579_end_mask_0 = const()[name = tensor<string, []>("op_12579_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12579_cast = slice_by_index(begin = var_12579_begin_0, end = var_12579_end_0, end_mask = var_12579_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12579_cast")];
            tensor<int32, [4]> var_12583_begin_0 = const()[name = tensor<string, []>("op_12583_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_12583_end_0 = const()[name = tensor<string, []>("op_12583_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_12583_end_mask_0 = const()[name = tensor<string, []>("op_12583_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12583_cast = slice_by_index(begin = var_12583_begin_0, end = var_12583_end_0, end_mask = var_12583_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12583_cast")];
            tensor<int32, [4]> var_12587_begin_0 = const()[name = tensor<string, []>("op_12587_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_12587_end_0 = const()[name = tensor<string, []>("op_12587_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_12587_end_mask_0 = const()[name = tensor<string, []>("op_12587_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12587_cast = slice_by_index(begin = var_12587_begin_0, end = var_12587_end_0, end_mask = var_12587_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12587_cast")];
            tensor<int32, [4]> var_12591_begin_0 = const()[name = tensor<string, []>("op_12591_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_12591_end_0 = const()[name = tensor<string, []>("op_12591_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_12591_end_mask_0 = const()[name = tensor<string, []>("op_12591_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12591_cast = slice_by_index(begin = var_12591_begin_0, end = var_12591_end_0, end_mask = var_12591_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12591_cast")];
            tensor<int32, [4]> var_12595_begin_0 = const()[name = tensor<string, []>("op_12595_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_12595_end_0 = const()[name = tensor<string, []>("op_12595_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_12595_end_mask_0 = const()[name = tensor<string, []>("op_12595_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12595_cast = slice_by_index(begin = var_12595_begin_0, end = var_12595_end_0, end_mask = var_12595_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12595_cast")];
            tensor<int32, [4]> var_12599_begin_0 = const()[name = tensor<string, []>("op_12599_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_12599_end_0 = const()[name = tensor<string, []>("op_12599_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_12599_end_mask_0 = const()[name = tensor<string, []>("op_12599_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12599_cast = slice_by_index(begin = var_12599_begin_0, end = var_12599_end_0, end_mask = var_12599_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12599_cast")];
            tensor<int32, [4]> var_12603_begin_0 = const()[name = tensor<string, []>("op_12603_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_12603_end_0 = const()[name = tensor<string, []>("op_12603_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_12603_end_mask_0 = const()[name = tensor<string, []>("op_12603_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12603_cast = slice_by_index(begin = var_12603_begin_0, end = var_12603_end_0, end_mask = var_12603_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12603_cast")];
            tensor<int32, [4]> var_12607_begin_0 = const()[name = tensor<string, []>("op_12607_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_12607_end_0 = const()[name = tensor<string, []>("op_12607_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_12607_end_mask_0 = const()[name = tensor<string, []>("op_12607_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12607_cast = slice_by_index(begin = var_12607_begin_0, end = var_12607_end_0, end_mask = var_12607_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12607_cast")];
            tensor<int32, [4]> var_12611_begin_0 = const()[name = tensor<string, []>("op_12611_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_12611_end_0 = const()[name = tensor<string, []>("op_12611_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_12611_end_mask_0 = const()[name = tensor<string, []>("op_12611_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12611_cast = slice_by_index(begin = var_12611_begin_0, end = var_12611_end_0, end_mask = var_12611_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12611_cast")];
            tensor<int32, [4]> var_12615_begin_0 = const()[name = tensor<string, []>("op_12615_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_12615_end_0 = const()[name = tensor<string, []>("op_12615_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_12615_end_mask_0 = const()[name = tensor<string, []>("op_12615_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12615_cast = slice_by_index(begin = var_12615_begin_0, end = var_12615_end_0, end_mask = var_12615_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12615_cast")];
            tensor<int32, [4]> var_12619_begin_0 = const()[name = tensor<string, []>("op_12619_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_12619_end_0 = const()[name = tensor<string, []>("op_12619_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_12619_end_mask_0 = const()[name = tensor<string, []>("op_12619_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12619_cast = slice_by_index(begin = var_12619_begin_0, end = var_12619_end_0, end_mask = var_12619_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12619_cast")];
            tensor<int32, [4]> var_12623_begin_0 = const()[name = tensor<string, []>("op_12623_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_12623_end_0 = const()[name = tensor<string, []>("op_12623_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_12623_end_mask_0 = const()[name = tensor<string, []>("op_12623_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12623_cast = slice_by_index(begin = var_12623_begin_0, end = var_12623_end_0, end_mask = var_12623_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12623_cast")];
            tensor<int32, [4]> var_12627_begin_0 = const()[name = tensor<string, []>("op_12627_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_12627_end_0 = const()[name = tensor<string, []>("op_12627_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_12627_end_mask_0 = const()[name = tensor<string, []>("op_12627_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12627_cast = slice_by_index(begin = var_12627_begin_0, end = var_12627_end_0, end_mask = var_12627_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12627_cast")];
            tensor<int32, [4]> var_12631_begin_0 = const()[name = tensor<string, []>("op_12631_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_12631_end_0 = const()[name = tensor<string, []>("op_12631_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_12631_end_mask_0 = const()[name = tensor<string, []>("op_12631_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12631_cast = slice_by_index(begin = var_12631_begin_0, end = var_12631_end_0, end_mask = var_12631_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12631_cast")];
            tensor<int32, [4]> var_12635_begin_0 = const()[name = tensor<string, []>("op_12635_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_12635_end_0 = const()[name = tensor<string, []>("op_12635_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_12635_end_mask_0 = const()[name = tensor<string, []>("op_12635_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12635_cast = slice_by_index(begin = var_12635_begin_0, end = var_12635_end_0, end_mask = var_12635_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12635_cast")];
            tensor<int32, [4]> var_12639_begin_0 = const()[name = tensor<string, []>("op_12639_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_12639_end_0 = const()[name = tensor<string, []>("op_12639_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_12639_end_mask_0 = const()[name = tensor<string, []>("op_12639_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12639_cast = slice_by_index(begin = var_12639_begin_0, end = var_12639_end_0, end_mask = var_12639_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12639_cast")];
            tensor<int32, [4]> var_12643_begin_0 = const()[name = tensor<string, []>("op_12643_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_12643_end_0 = const()[name = tensor<string, []>("op_12643_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_12643_end_mask_0 = const()[name = tensor<string, []>("op_12643_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12643_cast = slice_by_index(begin = var_12643_begin_0, end = var_12643_end_0, end_mask = var_12643_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12643_cast")];
            tensor<int32, [4]> var_12647_begin_0 = const()[name = tensor<string, []>("op_12647_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_12647_end_0 = const()[name = tensor<string, []>("op_12647_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_12647_end_mask_0 = const()[name = tensor<string, []>("op_12647_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12647_cast = slice_by_index(begin = var_12647_begin_0, end = var_12647_end_0, end_mask = var_12647_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12647_cast")];
            tensor<string, []> var_12651_equation_0 = const()[name = tensor<string, []>("op_12651_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12651_cast = einsum(equation = var_12651_equation_0, values = (var_12493_cast, var_12410_cast))[name = tensor<string, []>("op_12651_cast")];
            tensor<fp16, []> var_12652_to_fp16 = const()[name = tensor<string, []>("op_12652_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1001_cast = mul(x = var_12651_cast, y = var_12652_to_fp16)[name = tensor<string, []>("aw_1001_cast")];
            tensor<string, []> var_12655_equation_0 = const()[name = tensor<string, []>("op_12655_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12655_cast = einsum(equation = var_12655_equation_0, values = (var_12497_cast, var_12414_cast))[name = tensor<string, []>("op_12655_cast")];
            tensor<fp16, []> var_12656_to_fp16 = const()[name = tensor<string, []>("op_12656_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1003_cast = mul(x = var_12655_cast, y = var_12656_to_fp16)[name = tensor<string, []>("aw_1003_cast")];
            tensor<string, []> var_12659_equation_0 = const()[name = tensor<string, []>("op_12659_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12659_cast = einsum(equation = var_12659_equation_0, values = (var_12501_cast, var_12418_cast))[name = tensor<string, []>("op_12659_cast")];
            tensor<fp16, []> var_12660_to_fp16 = const()[name = tensor<string, []>("op_12660_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1005_cast = mul(x = var_12659_cast, y = var_12660_to_fp16)[name = tensor<string, []>("aw_1005_cast")];
            tensor<string, []> var_12663_equation_0 = const()[name = tensor<string, []>("op_12663_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12663_cast = einsum(equation = var_12663_equation_0, values = (var_12505_cast, var_12422_cast))[name = tensor<string, []>("op_12663_cast")];
            tensor<fp16, []> var_12664_to_fp16 = const()[name = tensor<string, []>("op_12664_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1007_cast = mul(x = var_12663_cast, y = var_12664_to_fp16)[name = tensor<string, []>("aw_1007_cast")];
            tensor<string, []> var_12667_equation_0 = const()[name = tensor<string, []>("op_12667_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12667_cast = einsum(equation = var_12667_equation_0, values = (var_12509_cast, var_12426_cast))[name = tensor<string, []>("op_12667_cast")];
            tensor<fp16, []> var_12668_to_fp16 = const()[name = tensor<string, []>("op_12668_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1009_cast = mul(x = var_12667_cast, y = var_12668_to_fp16)[name = tensor<string, []>("aw_1009_cast")];
            tensor<string, []> var_12671_equation_0 = const()[name = tensor<string, []>("op_12671_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12671_cast = einsum(equation = var_12671_equation_0, values = (var_12513_cast, var_12430_cast))[name = tensor<string, []>("op_12671_cast")];
            tensor<fp16, []> var_12672_to_fp16 = const()[name = tensor<string, []>("op_12672_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1011_cast = mul(x = var_12671_cast, y = var_12672_to_fp16)[name = tensor<string, []>("aw_1011_cast")];
            tensor<string, []> var_12675_equation_0 = const()[name = tensor<string, []>("op_12675_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12675_cast = einsum(equation = var_12675_equation_0, values = (var_12517_cast, var_12434_cast))[name = tensor<string, []>("op_12675_cast")];
            tensor<fp16, []> var_12676_to_fp16 = const()[name = tensor<string, []>("op_12676_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1013_cast = mul(x = var_12675_cast, y = var_12676_to_fp16)[name = tensor<string, []>("aw_1013_cast")];
            tensor<string, []> var_12679_equation_0 = const()[name = tensor<string, []>("op_12679_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12679_cast = einsum(equation = var_12679_equation_0, values = (var_12521_cast, var_12438_cast))[name = tensor<string, []>("op_12679_cast")];
            tensor<fp16, []> var_12680_to_fp16 = const()[name = tensor<string, []>("op_12680_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1015_cast = mul(x = var_12679_cast, y = var_12680_to_fp16)[name = tensor<string, []>("aw_1015_cast")];
            tensor<string, []> var_12683_equation_0 = const()[name = tensor<string, []>("op_12683_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12683_cast = einsum(equation = var_12683_equation_0, values = (var_12525_cast, var_12442_cast))[name = tensor<string, []>("op_12683_cast")];
            tensor<fp16, []> var_12684_to_fp16 = const()[name = tensor<string, []>("op_12684_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1017_cast = mul(x = var_12683_cast, y = var_12684_to_fp16)[name = tensor<string, []>("aw_1017_cast")];
            tensor<string, []> var_12687_equation_0 = const()[name = tensor<string, []>("op_12687_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12687_cast = einsum(equation = var_12687_equation_0, values = (var_12529_cast, var_12446_cast))[name = tensor<string, []>("op_12687_cast")];
            tensor<fp16, []> var_12688_to_fp16 = const()[name = tensor<string, []>("op_12688_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1019_cast = mul(x = var_12687_cast, y = var_12688_to_fp16)[name = tensor<string, []>("aw_1019_cast")];
            tensor<string, []> var_12691_equation_0 = const()[name = tensor<string, []>("op_12691_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12691_cast = einsum(equation = var_12691_equation_0, values = (var_12533_cast, var_12450_cast))[name = tensor<string, []>("op_12691_cast")];
            tensor<fp16, []> var_12692_to_fp16 = const()[name = tensor<string, []>("op_12692_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1021_cast = mul(x = var_12691_cast, y = var_12692_to_fp16)[name = tensor<string, []>("aw_1021_cast")];
            tensor<string, []> var_12695_equation_0 = const()[name = tensor<string, []>("op_12695_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12695_cast = einsum(equation = var_12695_equation_0, values = (var_12537_cast, var_12454_cast))[name = tensor<string, []>("op_12695_cast")];
            tensor<fp16, []> var_12696_to_fp16 = const()[name = tensor<string, []>("op_12696_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1023_cast = mul(x = var_12695_cast, y = var_12696_to_fp16)[name = tensor<string, []>("aw_1023_cast")];
            tensor<string, []> var_12699_equation_0 = const()[name = tensor<string, []>("op_12699_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12699_cast = einsum(equation = var_12699_equation_0, values = (var_12541_cast, var_12458_cast))[name = tensor<string, []>("op_12699_cast")];
            tensor<fp16, []> var_12700_to_fp16 = const()[name = tensor<string, []>("op_12700_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1025_cast = mul(x = var_12699_cast, y = var_12700_to_fp16)[name = tensor<string, []>("aw_1025_cast")];
            tensor<string, []> var_12703_equation_0 = const()[name = tensor<string, []>("op_12703_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12703_cast = einsum(equation = var_12703_equation_0, values = (var_12545_cast, var_12462_cast))[name = tensor<string, []>("op_12703_cast")];
            tensor<fp16, []> var_12704_to_fp16 = const()[name = tensor<string, []>("op_12704_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1027_cast = mul(x = var_12703_cast, y = var_12704_to_fp16)[name = tensor<string, []>("aw_1027_cast")];
            tensor<string, []> var_12707_equation_0 = const()[name = tensor<string, []>("op_12707_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12707_cast = einsum(equation = var_12707_equation_0, values = (var_12549_cast, var_12466_cast))[name = tensor<string, []>("op_12707_cast")];
            tensor<fp16, []> var_12708_to_fp16 = const()[name = tensor<string, []>("op_12708_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1029_cast = mul(x = var_12707_cast, y = var_12708_to_fp16)[name = tensor<string, []>("aw_1029_cast")];
            tensor<string, []> var_12711_equation_0 = const()[name = tensor<string, []>("op_12711_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12711_cast = einsum(equation = var_12711_equation_0, values = (var_12553_cast, var_12470_cast))[name = tensor<string, []>("op_12711_cast")];
            tensor<fp16, []> var_12712_to_fp16 = const()[name = tensor<string, []>("op_12712_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1031_cast = mul(x = var_12711_cast, y = var_12712_to_fp16)[name = tensor<string, []>("aw_1031_cast")];
            tensor<string, []> var_12715_equation_0 = const()[name = tensor<string, []>("op_12715_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12715_cast = einsum(equation = var_12715_equation_0, values = (var_12557_cast, var_12474_cast))[name = tensor<string, []>("op_12715_cast")];
            tensor<fp16, []> var_12716_to_fp16 = const()[name = tensor<string, []>("op_12716_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1033_cast = mul(x = var_12715_cast, y = var_12716_to_fp16)[name = tensor<string, []>("aw_1033_cast")];
            tensor<string, []> var_12719_equation_0 = const()[name = tensor<string, []>("op_12719_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12719_cast = einsum(equation = var_12719_equation_0, values = (var_12561_cast, var_12478_cast))[name = tensor<string, []>("op_12719_cast")];
            tensor<fp16, []> var_12720_to_fp16 = const()[name = tensor<string, []>("op_12720_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1035_cast = mul(x = var_12719_cast, y = var_12720_to_fp16)[name = tensor<string, []>("aw_1035_cast")];
            tensor<string, []> var_12723_equation_0 = const()[name = tensor<string, []>("op_12723_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12723_cast = einsum(equation = var_12723_equation_0, values = (var_12565_cast, var_12482_cast))[name = tensor<string, []>("op_12723_cast")];
            tensor<fp16, []> var_12724_to_fp16 = const()[name = tensor<string, []>("op_12724_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1037_cast = mul(x = var_12723_cast, y = var_12724_to_fp16)[name = tensor<string, []>("aw_1037_cast")];
            tensor<string, []> var_12727_equation_0 = const()[name = tensor<string, []>("op_12727_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_12727_cast = einsum(equation = var_12727_equation_0, values = (var_12569_cast, var_12486_cast))[name = tensor<string, []>("op_12727_cast")];
            tensor<fp16, []> var_12728_to_fp16 = const()[name = tensor<string, []>("op_12728_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1039_cast = mul(x = var_12727_cast, y = var_12728_to_fp16)[name = tensor<string, []>("aw_1039_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12730_cast = softmax(axis = var_2634, x = aw_1001_cast)[name = tensor<string, []>("op_12730_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12731_cast = softmax(axis = var_2634, x = aw_1003_cast)[name = tensor<string, []>("op_12731_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12732_cast = softmax(axis = var_2634, x = aw_1005_cast)[name = tensor<string, []>("op_12732_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12733_cast = softmax(axis = var_2634, x = aw_1007_cast)[name = tensor<string, []>("op_12733_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12734_cast = softmax(axis = var_2634, x = aw_1009_cast)[name = tensor<string, []>("op_12734_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12735_cast = softmax(axis = var_2634, x = aw_1011_cast)[name = tensor<string, []>("op_12735_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12736_cast = softmax(axis = var_2634, x = aw_1013_cast)[name = tensor<string, []>("op_12736_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12737_cast = softmax(axis = var_2634, x = aw_1015_cast)[name = tensor<string, []>("op_12737_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12738_cast = softmax(axis = var_2634, x = aw_1017_cast)[name = tensor<string, []>("op_12738_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12739_cast = softmax(axis = var_2634, x = aw_1019_cast)[name = tensor<string, []>("op_12739_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12740_cast = softmax(axis = var_2634, x = aw_1021_cast)[name = tensor<string, []>("op_12740_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12741_cast = softmax(axis = var_2634, x = aw_1023_cast)[name = tensor<string, []>("op_12741_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12742_cast = softmax(axis = var_2634, x = aw_1025_cast)[name = tensor<string, []>("op_12742_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12743_cast = softmax(axis = var_2634, x = aw_1027_cast)[name = tensor<string, []>("op_12743_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12744_cast = softmax(axis = var_2634, x = aw_1029_cast)[name = tensor<string, []>("op_12744_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12745_cast = softmax(axis = var_2634, x = aw_1031_cast)[name = tensor<string, []>("op_12745_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12746_cast = softmax(axis = var_2634, x = aw_1033_cast)[name = tensor<string, []>("op_12746_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12747_cast = softmax(axis = var_2634, x = aw_1035_cast)[name = tensor<string, []>("op_12747_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12748_cast = softmax(axis = var_2634, x = aw_1037_cast)[name = tensor<string, []>("op_12748_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_12749_cast = softmax(axis = var_2634, x = aw_1039_cast)[name = tensor<string, []>("op_12749_cast")];
            tensor<string, []> var_12751_equation_0 = const()[name = tensor<string, []>("op_12751_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12751_cast = einsum(equation = var_12751_equation_0, values = (var_12571_cast, var_12730_cast))[name = tensor<string, []>("op_12751_cast")];
            tensor<string, []> var_12753_equation_0 = const()[name = tensor<string, []>("op_12753_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12753_cast = einsum(equation = var_12753_equation_0, values = (var_12575_cast, var_12731_cast))[name = tensor<string, []>("op_12753_cast")];
            tensor<string, []> var_12755_equation_0 = const()[name = tensor<string, []>("op_12755_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12755_cast = einsum(equation = var_12755_equation_0, values = (var_12579_cast, var_12732_cast))[name = tensor<string, []>("op_12755_cast")];
            tensor<string, []> var_12757_equation_0 = const()[name = tensor<string, []>("op_12757_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12757_cast = einsum(equation = var_12757_equation_0, values = (var_12583_cast, var_12733_cast))[name = tensor<string, []>("op_12757_cast")];
            tensor<string, []> var_12759_equation_0 = const()[name = tensor<string, []>("op_12759_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12759_cast = einsum(equation = var_12759_equation_0, values = (var_12587_cast, var_12734_cast))[name = tensor<string, []>("op_12759_cast")];
            tensor<string, []> var_12761_equation_0 = const()[name = tensor<string, []>("op_12761_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12761_cast = einsum(equation = var_12761_equation_0, values = (var_12591_cast, var_12735_cast))[name = tensor<string, []>("op_12761_cast")];
            tensor<string, []> var_12763_equation_0 = const()[name = tensor<string, []>("op_12763_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12763_cast = einsum(equation = var_12763_equation_0, values = (var_12595_cast, var_12736_cast))[name = tensor<string, []>("op_12763_cast")];
            tensor<string, []> var_12765_equation_0 = const()[name = tensor<string, []>("op_12765_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12765_cast = einsum(equation = var_12765_equation_0, values = (var_12599_cast, var_12737_cast))[name = tensor<string, []>("op_12765_cast")];
            tensor<string, []> var_12767_equation_0 = const()[name = tensor<string, []>("op_12767_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12767_cast = einsum(equation = var_12767_equation_0, values = (var_12603_cast, var_12738_cast))[name = tensor<string, []>("op_12767_cast")];
            tensor<string, []> var_12769_equation_0 = const()[name = tensor<string, []>("op_12769_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12769_cast = einsum(equation = var_12769_equation_0, values = (var_12607_cast, var_12739_cast))[name = tensor<string, []>("op_12769_cast")];
            tensor<string, []> var_12771_equation_0 = const()[name = tensor<string, []>("op_12771_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12771_cast = einsum(equation = var_12771_equation_0, values = (var_12611_cast, var_12740_cast))[name = tensor<string, []>("op_12771_cast")];
            tensor<string, []> var_12773_equation_0 = const()[name = tensor<string, []>("op_12773_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12773_cast = einsum(equation = var_12773_equation_0, values = (var_12615_cast, var_12741_cast))[name = tensor<string, []>("op_12773_cast")];
            tensor<string, []> var_12775_equation_0 = const()[name = tensor<string, []>("op_12775_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12775_cast = einsum(equation = var_12775_equation_0, values = (var_12619_cast, var_12742_cast))[name = tensor<string, []>("op_12775_cast")];
            tensor<string, []> var_12777_equation_0 = const()[name = tensor<string, []>("op_12777_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12777_cast = einsum(equation = var_12777_equation_0, values = (var_12623_cast, var_12743_cast))[name = tensor<string, []>("op_12777_cast")];
            tensor<string, []> var_12779_equation_0 = const()[name = tensor<string, []>("op_12779_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12779_cast = einsum(equation = var_12779_equation_0, values = (var_12627_cast, var_12744_cast))[name = tensor<string, []>("op_12779_cast")];
            tensor<string, []> var_12781_equation_0 = const()[name = tensor<string, []>("op_12781_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12781_cast = einsum(equation = var_12781_equation_0, values = (var_12631_cast, var_12745_cast))[name = tensor<string, []>("op_12781_cast")];
            tensor<string, []> var_12783_equation_0 = const()[name = tensor<string, []>("op_12783_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12783_cast = einsum(equation = var_12783_equation_0, values = (var_12635_cast, var_12746_cast))[name = tensor<string, []>("op_12783_cast")];
            tensor<string, []> var_12785_equation_0 = const()[name = tensor<string, []>("op_12785_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12785_cast = einsum(equation = var_12785_equation_0, values = (var_12639_cast, var_12747_cast))[name = tensor<string, []>("op_12785_cast")];
            tensor<string, []> var_12787_equation_0 = const()[name = tensor<string, []>("op_12787_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12787_cast = einsum(equation = var_12787_equation_0, values = (var_12643_cast, var_12748_cast))[name = tensor<string, []>("op_12787_cast")];
            tensor<string, []> var_12789_equation_0 = const()[name = tensor<string, []>("op_12789_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_12789_cast = einsum(equation = var_12789_equation_0, values = (var_12647_cast, var_12749_cast))[name = tensor<string, []>("op_12789_cast")];
            tensor<bool, []> input_235_interleave_0 = const()[name = tensor<string, []>("input_235_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_235_cast = concat(axis = var_2634, interleave = input_235_interleave_0, values = (var_12751_cast, var_12753_cast, var_12755_cast, var_12757_cast, var_12759_cast, var_12761_cast, var_12763_cast, var_12765_cast, var_12767_cast, var_12769_cast, var_12771_cast, var_12773_cast, var_12775_cast, var_12777_cast, var_12779_cast, var_12781_cast, var_12783_cast, var_12785_cast, var_12787_cast, var_12789_cast))[name = tensor<string, []>("input_235_cast")];
            tensor<int32, [2]> var_12795 = const()[name = tensor<string, []>("op_12795"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12797 = const()[name = tensor<string, []>("op_12797"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_12799_pad_type_0 = const()[name = tensor<string, []>("op_12799_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_12799_pad_0 = const()[name = tensor<string, []>("op_12799_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(989020800)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(992297664)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_12799_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_12797, groups = var_2634, pad = var_12799_pad_0, pad_type = var_12799_pad_type_0, strides = var_12795, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_235_cast)[name = tensor<string, []>("op_12799_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_89_cast = add(x = var_12799_cast, y = inputs_87_cast)[name = tensor<string, []>("inputs_89_cast")];
            tensor<int32, [1]> var_12803 = const()[name = tensor<string, []>("op_12803"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_89_cast = reduce_mean(axes = var_12803, keep_dims = var_2629, x = inputs_89_cast)[name = tensor<string, []>("channels_mean_89_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_89_cast = sub(x = inputs_89_cast, y = channels_mean_89_cast)[name = tensor<string, []>("zero_mean_89_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_89_cast = mul(x = zero_mean_89_cast, y = zero_mean_89_cast)[name = tensor<string, []>("zero_mean_sq_89_cast")];
            tensor<int32, [1]> var_12807 = const()[name = tensor<string, []>("op_12807"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_12808_cast = reduce_mean(axes = var_12807, keep_dims = var_2629, x = zero_mean_sq_89_cast)[name = tensor<string, []>("op_12808_cast")];
            tensor<fp16, []> var_12809_to_fp16 = const()[name = tensor<string, []>("op_12809_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_12810_cast = add(x = var_12808_cast, y = var_12809_to_fp16)[name = tensor<string, []>("op_12810_cast")];
            tensor<fp16, []> denom_89_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_89_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_89_cast = rsqrt(epsilon = denom_89_epsilon_0_to_fp16, x = var_12810_cast)[name = tensor<string, []>("denom_89_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_89_cast = mul(x = zero_mean_89_cast, y = denom_89_cast)[name = tensor<string, []>("out_89_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_12814_to_fp16 = const()[name = tensor<string, []>("op_12814_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(992300288)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_12815_cast = add(x = out_89_cast, y = var_12814_to_fp16)[name = tensor<string, []>("op_12815_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_12817_to_fp16 = const()[name = tensor<string, []>("op_12817_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(992302912)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_237_cast = mul(x = var_12815_cast, y = var_12817_to_fp16)[name = tensor<string, []>("input_237_cast")];
            tensor<int32, [2]> var_12825 = const()[name = tensor<string, []>("op_12825"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12827 = const()[name = tensor<string, []>("op_12827"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_12829_pad_type_0 = const()[name = tensor<string, []>("op_12829_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_12829_pad_0 = const()[name = tensor<string, []>("op_12829_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(992305536)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1018520000)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_12829_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_12827, groups = var_2634, pad = var_12829_pad_0, pad_type = var_12829_pad_type_0, strides = var_12825, weight = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_237_cast)[name = tensor<string, []>("op_12829_cast")];
            tensor<int32, [2]> var_12830_split_sizes_0 = const()[name = tensor<string, []>("op_12830_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_12830_axis_0 = const()[name = tensor<string, []>("op_12830_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_12830_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_12830_cast_1 = split(axis = var_12830_axis_0, split_sizes = var_12830_split_sizes_0, x = var_12829_cast)[name = tensor<string, []>("op_12830_cast")];
            tensor<string, []> var_12832_mode_0 = const()[name = tensor<string, []>("op_12832_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_12832_cast = gelu(mode = var_12832_mode_0, x = var_12830_cast_1)[name = tensor<string, []>("op_12832_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_239_cast = mul(x = var_12830_cast_0, y = var_12832_cast)[name = tensor<string, []>("input_239_cast")];
            tensor<int32, [2]> var_12836 = const()[name = tensor<string, []>("op_12836"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12838 = const()[name = tensor<string, []>("op_12838"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_12840_pad_type_0 = const()[name = tensor<string, []>("op_12840_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_12840_pad_0 = const()[name = tensor<string, []>("op_12840_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1018540544)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1031647808)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_12840_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_12838, groups = var_2634, pad = var_12840_pad_0, pad_type = var_12840_pad_type_0, strides = var_12836, weight = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_239_cast)[name = tensor<string, []>("op_12840_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_91_cast = add(x = var_12840_cast, y = inputs_89_cast)[name = tensor<string, []>("inputs_91_cast")];
            tensor<int32, [1]> var_12850 = const()[name = tensor<string, []>("op_12850"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_91_cast = reduce_mean(axes = var_12850, keep_dims = var_2629, x = inputs_91_cast)[name = tensor<string, []>("channels_mean_91_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_91_cast = sub(x = inputs_91_cast, y = channels_mean_91_cast)[name = tensor<string, []>("zero_mean_91_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_91_cast = mul(x = zero_mean_91_cast, y = zero_mean_91_cast)[name = tensor<string, []>("zero_mean_sq_91_cast")];
            tensor<int32, [1]> var_12854 = const()[name = tensor<string, []>("op_12854"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_12855_cast = reduce_mean(axes = var_12854, keep_dims = var_2629, x = zero_mean_sq_91_cast)[name = tensor<string, []>("op_12855_cast")];
            tensor<fp16, []> var_12856_to_fp16 = const()[name = tensor<string, []>("op_12856_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_12857_cast = add(x = var_12855_cast, y = var_12856_to_fp16)[name = tensor<string, []>("op_12857_cast")];
            tensor<fp16, []> denom_91_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_91_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_91_cast = rsqrt(epsilon = denom_91_epsilon_0_to_fp16, x = var_12857_cast)[name = tensor<string, []>("denom_91_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_91_cast = mul(x = zero_mean_91_cast, y = denom_91_cast)[name = tensor<string, []>("out_91_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_12861_to_fp16 = const()[name = tensor<string, []>("op_12861_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1031650432)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_12862_cast = add(x = out_91_cast, y = var_12861_to_fp16)[name = tensor<string, []>("op_12862_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_12864_to_fp16 = const()[name = tensor<string, []>("op_12864_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1031653056)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_143_cast = mul(x = var_12862_cast, y = var_12864_to_fp16)[name = tensor<string, []>("hidden_states_143_cast")];
            tensor<int32, [2]> var_12871 = const()[name = tensor<string, []>("op_12871"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12873 = const()[name = tensor<string, []>("op_12873"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_61_pad_type_0 = const()[name = tensor<string, []>("q_61_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_61_pad_0 = const()[name = tensor<string, []>("q_61_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1031655680)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_61_cast = conv(dilations = var_12873, groups = var_2634, pad = q_61_pad_0, pad_type = q_61_pad_type_0, strides = var_12871, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_143_cast)[name = tensor<string, []>("q_61_cast")];
            tensor<int32, [2]> var_12877 = const()[name = tensor<string, []>("op_12877"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12879 = const()[name = tensor<string, []>("op_12879"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_121_pad_type_0 = const()[name = tensor<string, []>("k_121_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_121_pad_0 = const()[name = tensor<string, []>("k_121_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1034932544)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_121_cast = conv(dilations = var_12879, groups = var_2634, pad = k_121_pad_0, pad_type = k_121_pad_type_0, strides = var_12877, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_143_cast)[name = tensor<string, []>("k_121_cast")];
            tensor<int32, [2]> var_12883 = const()[name = tensor<string, []>("op_12883"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12885 = const()[name = tensor<string, []>("op_12885"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_61_pad_type_0 = const()[name = tensor<string, []>("v_61_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_61_pad_0 = const()[name = tensor<string, []>("v_61_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1038209408)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_61_cast = conv(dilations = var_12885, groups = var_2634, pad = v_61_pad_0, pad_type = v_61_pad_type_0, strides = var_12883, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_143_cast)[name = tensor<string, []>("v_61_cast")];
            tensor<int32, [4]> var_12889_begin_0 = const()[name = tensor<string, []>("op_12889_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12889_end_0 = const()[name = tensor<string, []>("op_12889_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12889_end_mask_0 = const()[name = tensor<string, []>("op_12889_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12889_cast = slice_by_index(begin = var_12889_begin_0, end = var_12889_end_0, end_mask = var_12889_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12889_cast")];
            tensor<int32, [4]> var_12893_begin_0 = const()[name = tensor<string, []>("op_12893_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_12893_end_0 = const()[name = tensor<string, []>("op_12893_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_12893_end_mask_0 = const()[name = tensor<string, []>("op_12893_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12893_cast = slice_by_index(begin = var_12893_begin_0, end = var_12893_end_0, end_mask = var_12893_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12893_cast")];
            tensor<int32, [4]> var_12897_begin_0 = const()[name = tensor<string, []>("op_12897_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_12897_end_0 = const()[name = tensor<string, []>("op_12897_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_12897_end_mask_0 = const()[name = tensor<string, []>("op_12897_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12897_cast = slice_by_index(begin = var_12897_begin_0, end = var_12897_end_0, end_mask = var_12897_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12897_cast")];
            tensor<int32, [4]> var_12901_begin_0 = const()[name = tensor<string, []>("op_12901_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_12901_end_0 = const()[name = tensor<string, []>("op_12901_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_12901_end_mask_0 = const()[name = tensor<string, []>("op_12901_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12901_cast = slice_by_index(begin = var_12901_begin_0, end = var_12901_end_0, end_mask = var_12901_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12901_cast")];
            tensor<int32, [4]> var_12905_begin_0 = const()[name = tensor<string, []>("op_12905_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_12905_end_0 = const()[name = tensor<string, []>("op_12905_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_12905_end_mask_0 = const()[name = tensor<string, []>("op_12905_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12905_cast = slice_by_index(begin = var_12905_begin_0, end = var_12905_end_0, end_mask = var_12905_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12905_cast")];
            tensor<int32, [4]> var_12909_begin_0 = const()[name = tensor<string, []>("op_12909_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_12909_end_0 = const()[name = tensor<string, []>("op_12909_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_12909_end_mask_0 = const()[name = tensor<string, []>("op_12909_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12909_cast = slice_by_index(begin = var_12909_begin_0, end = var_12909_end_0, end_mask = var_12909_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12909_cast")];
            tensor<int32, [4]> var_12913_begin_0 = const()[name = tensor<string, []>("op_12913_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_12913_end_0 = const()[name = tensor<string, []>("op_12913_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_12913_end_mask_0 = const()[name = tensor<string, []>("op_12913_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12913_cast = slice_by_index(begin = var_12913_begin_0, end = var_12913_end_0, end_mask = var_12913_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12913_cast")];
            tensor<int32, [4]> var_12917_begin_0 = const()[name = tensor<string, []>("op_12917_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_12917_end_0 = const()[name = tensor<string, []>("op_12917_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_12917_end_mask_0 = const()[name = tensor<string, []>("op_12917_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12917_cast = slice_by_index(begin = var_12917_begin_0, end = var_12917_end_0, end_mask = var_12917_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12917_cast")];
            tensor<int32, [4]> var_12921_begin_0 = const()[name = tensor<string, []>("op_12921_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_12921_end_0 = const()[name = tensor<string, []>("op_12921_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_12921_end_mask_0 = const()[name = tensor<string, []>("op_12921_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12921_cast = slice_by_index(begin = var_12921_begin_0, end = var_12921_end_0, end_mask = var_12921_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12921_cast")];
            tensor<int32, [4]> var_12925_begin_0 = const()[name = tensor<string, []>("op_12925_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_12925_end_0 = const()[name = tensor<string, []>("op_12925_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_12925_end_mask_0 = const()[name = tensor<string, []>("op_12925_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12925_cast = slice_by_index(begin = var_12925_begin_0, end = var_12925_end_0, end_mask = var_12925_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12925_cast")];
            tensor<int32, [4]> var_12929_begin_0 = const()[name = tensor<string, []>("op_12929_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_12929_end_0 = const()[name = tensor<string, []>("op_12929_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_12929_end_mask_0 = const()[name = tensor<string, []>("op_12929_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12929_cast = slice_by_index(begin = var_12929_begin_0, end = var_12929_end_0, end_mask = var_12929_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12929_cast")];
            tensor<int32, [4]> var_12933_begin_0 = const()[name = tensor<string, []>("op_12933_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_12933_end_0 = const()[name = tensor<string, []>("op_12933_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_12933_end_mask_0 = const()[name = tensor<string, []>("op_12933_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12933_cast = slice_by_index(begin = var_12933_begin_0, end = var_12933_end_0, end_mask = var_12933_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12933_cast")];
            tensor<int32, [4]> var_12937_begin_0 = const()[name = tensor<string, []>("op_12937_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_12937_end_0 = const()[name = tensor<string, []>("op_12937_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_12937_end_mask_0 = const()[name = tensor<string, []>("op_12937_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12937_cast = slice_by_index(begin = var_12937_begin_0, end = var_12937_end_0, end_mask = var_12937_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12937_cast")];
            tensor<int32, [4]> var_12941_begin_0 = const()[name = tensor<string, []>("op_12941_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_12941_end_0 = const()[name = tensor<string, []>("op_12941_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_12941_end_mask_0 = const()[name = tensor<string, []>("op_12941_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12941_cast = slice_by_index(begin = var_12941_begin_0, end = var_12941_end_0, end_mask = var_12941_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12941_cast")];
            tensor<int32, [4]> var_12945_begin_0 = const()[name = tensor<string, []>("op_12945_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_12945_end_0 = const()[name = tensor<string, []>("op_12945_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_12945_end_mask_0 = const()[name = tensor<string, []>("op_12945_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12945_cast = slice_by_index(begin = var_12945_begin_0, end = var_12945_end_0, end_mask = var_12945_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12945_cast")];
            tensor<int32, [4]> var_12949_begin_0 = const()[name = tensor<string, []>("op_12949_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_12949_end_0 = const()[name = tensor<string, []>("op_12949_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_12949_end_mask_0 = const()[name = tensor<string, []>("op_12949_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12949_cast = slice_by_index(begin = var_12949_begin_0, end = var_12949_end_0, end_mask = var_12949_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12949_cast")];
            tensor<int32, [4]> var_12953_begin_0 = const()[name = tensor<string, []>("op_12953_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_12953_end_0 = const()[name = tensor<string, []>("op_12953_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_12953_end_mask_0 = const()[name = tensor<string, []>("op_12953_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12953_cast = slice_by_index(begin = var_12953_begin_0, end = var_12953_end_0, end_mask = var_12953_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12953_cast")];
            tensor<int32, [4]> var_12957_begin_0 = const()[name = tensor<string, []>("op_12957_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_12957_end_0 = const()[name = tensor<string, []>("op_12957_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_12957_end_mask_0 = const()[name = tensor<string, []>("op_12957_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12957_cast = slice_by_index(begin = var_12957_begin_0, end = var_12957_end_0, end_mask = var_12957_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12957_cast")];
            tensor<int32, [4]> var_12961_begin_0 = const()[name = tensor<string, []>("op_12961_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_12961_end_0 = const()[name = tensor<string, []>("op_12961_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_12961_end_mask_0 = const()[name = tensor<string, []>("op_12961_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12961_cast = slice_by_index(begin = var_12961_begin_0, end = var_12961_end_0, end_mask = var_12961_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12961_cast")];
            tensor<int32, [4]> var_12965_begin_0 = const()[name = tensor<string, []>("op_12965_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_12965_end_0 = const()[name = tensor<string, []>("op_12965_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_12965_end_mask_0 = const()[name = tensor<string, []>("op_12965_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12965_cast = slice_by_index(begin = var_12965_begin_0, end = var_12965_end_0, end_mask = var_12965_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_12965_cast")];
            tensor<int32, [4]> k_123_perm_0 = const()[name = tensor<string, []>("k_123_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_12972_begin_0 = const()[name = tensor<string, []>("op_12972_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12972_end_0 = const()[name = tensor<string, []>("op_12972_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_12972_end_mask_0 = const()[name = tensor<string, []>("op_12972_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_109 = transpose(perm = k_123_perm_0, x = k_121_cast)[name = tensor<string, []>("transpose_109")];
            tensor<fp16, [2, 1024, 1, 64]> var_12972_cast = slice_by_index(begin = var_12972_begin_0, end = var_12972_end_0, end_mask = var_12972_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_12972_cast")];
            tensor<int32, [4]> var_12976_begin_0 = const()[name = tensor<string, []>("op_12976_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_12976_end_0 = const()[name = tensor<string, []>("op_12976_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_12976_end_mask_0 = const()[name = tensor<string, []>("op_12976_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12976_cast = slice_by_index(begin = var_12976_begin_0, end = var_12976_end_0, end_mask = var_12976_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_12976_cast")];
            tensor<int32, [4]> var_12980_begin_0 = const()[name = tensor<string, []>("op_12980_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_12980_end_0 = const()[name = tensor<string, []>("op_12980_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_12980_end_mask_0 = const()[name = tensor<string, []>("op_12980_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12980_cast = slice_by_index(begin = var_12980_begin_0, end = var_12980_end_0, end_mask = var_12980_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_12980_cast")];
            tensor<int32, [4]> var_12984_begin_0 = const()[name = tensor<string, []>("op_12984_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_12984_end_0 = const()[name = tensor<string, []>("op_12984_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_12984_end_mask_0 = const()[name = tensor<string, []>("op_12984_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12984_cast = slice_by_index(begin = var_12984_begin_0, end = var_12984_end_0, end_mask = var_12984_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_12984_cast")];
            tensor<int32, [4]> var_12988_begin_0 = const()[name = tensor<string, []>("op_12988_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_12988_end_0 = const()[name = tensor<string, []>("op_12988_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_12988_end_mask_0 = const()[name = tensor<string, []>("op_12988_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12988_cast = slice_by_index(begin = var_12988_begin_0, end = var_12988_end_0, end_mask = var_12988_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_12988_cast")];
            tensor<int32, [4]> var_12992_begin_0 = const()[name = tensor<string, []>("op_12992_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_12992_end_0 = const()[name = tensor<string, []>("op_12992_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_12992_end_mask_0 = const()[name = tensor<string, []>("op_12992_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12992_cast = slice_by_index(begin = var_12992_begin_0, end = var_12992_end_0, end_mask = var_12992_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_12992_cast")];
            tensor<int32, [4]> var_12996_begin_0 = const()[name = tensor<string, []>("op_12996_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_12996_end_0 = const()[name = tensor<string, []>("op_12996_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_12996_end_mask_0 = const()[name = tensor<string, []>("op_12996_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12996_cast = slice_by_index(begin = var_12996_begin_0, end = var_12996_end_0, end_mask = var_12996_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_12996_cast")];
            tensor<int32, [4]> var_13000_begin_0 = const()[name = tensor<string, []>("op_13000_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_13000_end_0 = const()[name = tensor<string, []>("op_13000_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_13000_end_mask_0 = const()[name = tensor<string, []>("op_13000_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13000_cast = slice_by_index(begin = var_13000_begin_0, end = var_13000_end_0, end_mask = var_13000_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_13000_cast")];
            tensor<int32, [4]> var_13004_begin_0 = const()[name = tensor<string, []>("op_13004_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13004_end_0 = const()[name = tensor<string, []>("op_13004_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_13004_end_mask_0 = const()[name = tensor<string, []>("op_13004_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13004_cast = slice_by_index(begin = var_13004_begin_0, end = var_13004_end_0, end_mask = var_13004_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_13004_cast")];
            tensor<int32, [4]> var_13008_begin_0 = const()[name = tensor<string, []>("op_13008_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_13008_end_0 = const()[name = tensor<string, []>("op_13008_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_13008_end_mask_0 = const()[name = tensor<string, []>("op_13008_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13008_cast = slice_by_index(begin = var_13008_begin_0, end = var_13008_end_0, end_mask = var_13008_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_13008_cast")];
            tensor<int32, [4]> var_13012_begin_0 = const()[name = tensor<string, []>("op_13012_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_13012_end_0 = const()[name = tensor<string, []>("op_13012_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_13012_end_mask_0 = const()[name = tensor<string, []>("op_13012_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13012_cast = slice_by_index(begin = var_13012_begin_0, end = var_13012_end_0, end_mask = var_13012_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_13012_cast")];
            tensor<int32, [4]> var_13016_begin_0 = const()[name = tensor<string, []>("op_13016_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_13016_end_0 = const()[name = tensor<string, []>("op_13016_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_13016_end_mask_0 = const()[name = tensor<string, []>("op_13016_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13016_cast = slice_by_index(begin = var_13016_begin_0, end = var_13016_end_0, end_mask = var_13016_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_13016_cast")];
            tensor<int32, [4]> var_13020_begin_0 = const()[name = tensor<string, []>("op_13020_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_13020_end_0 = const()[name = tensor<string, []>("op_13020_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_13020_end_mask_0 = const()[name = tensor<string, []>("op_13020_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13020_cast = slice_by_index(begin = var_13020_begin_0, end = var_13020_end_0, end_mask = var_13020_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_13020_cast")];
            tensor<int32, [4]> var_13024_begin_0 = const()[name = tensor<string, []>("op_13024_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_13024_end_0 = const()[name = tensor<string, []>("op_13024_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_13024_end_mask_0 = const()[name = tensor<string, []>("op_13024_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13024_cast = slice_by_index(begin = var_13024_begin_0, end = var_13024_end_0, end_mask = var_13024_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_13024_cast")];
            tensor<int32, [4]> var_13028_begin_0 = const()[name = tensor<string, []>("op_13028_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_13028_end_0 = const()[name = tensor<string, []>("op_13028_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_13028_end_mask_0 = const()[name = tensor<string, []>("op_13028_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13028_cast = slice_by_index(begin = var_13028_begin_0, end = var_13028_end_0, end_mask = var_13028_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_13028_cast")];
            tensor<int32, [4]> var_13032_begin_0 = const()[name = tensor<string, []>("op_13032_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_13032_end_0 = const()[name = tensor<string, []>("op_13032_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_13032_end_mask_0 = const()[name = tensor<string, []>("op_13032_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13032_cast = slice_by_index(begin = var_13032_begin_0, end = var_13032_end_0, end_mask = var_13032_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_13032_cast")];
            tensor<int32, [4]> var_13036_begin_0 = const()[name = tensor<string, []>("op_13036_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13036_end_0 = const()[name = tensor<string, []>("op_13036_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_13036_end_mask_0 = const()[name = tensor<string, []>("op_13036_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13036_cast = slice_by_index(begin = var_13036_begin_0, end = var_13036_end_0, end_mask = var_13036_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_13036_cast")];
            tensor<int32, [4]> var_13040_begin_0 = const()[name = tensor<string, []>("op_13040_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_13040_end_0 = const()[name = tensor<string, []>("op_13040_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_13040_end_mask_0 = const()[name = tensor<string, []>("op_13040_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13040_cast = slice_by_index(begin = var_13040_begin_0, end = var_13040_end_0, end_mask = var_13040_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_13040_cast")];
            tensor<int32, [4]> var_13044_begin_0 = const()[name = tensor<string, []>("op_13044_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_13044_end_0 = const()[name = tensor<string, []>("op_13044_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_13044_end_mask_0 = const()[name = tensor<string, []>("op_13044_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13044_cast = slice_by_index(begin = var_13044_begin_0, end = var_13044_end_0, end_mask = var_13044_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_13044_cast")];
            tensor<int32, [4]> var_13048_begin_0 = const()[name = tensor<string, []>("op_13048_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_13048_end_0 = const()[name = tensor<string, []>("op_13048_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_13048_end_mask_0 = const()[name = tensor<string, []>("op_13048_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13048_cast = slice_by_index(begin = var_13048_begin_0, end = var_13048_end_0, end_mask = var_13048_end_mask_0, x = transpose_109)[name = tensor<string, []>("op_13048_cast")];
            tensor<int32, [4]> var_13050_begin_0 = const()[name = tensor<string, []>("op_13050_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13050_end_0 = const()[name = tensor<string, []>("op_13050_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_13050_end_mask_0 = const()[name = tensor<string, []>("op_13050_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13050_cast = slice_by_index(begin = var_13050_begin_0, end = var_13050_end_0, end_mask = var_13050_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13050_cast")];
            tensor<int32, [4]> var_13054_begin_0 = const()[name = tensor<string, []>("op_13054_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_13054_end_0 = const()[name = tensor<string, []>("op_13054_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_13054_end_mask_0 = const()[name = tensor<string, []>("op_13054_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13054_cast = slice_by_index(begin = var_13054_begin_0, end = var_13054_end_0, end_mask = var_13054_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13054_cast")];
            tensor<int32, [4]> var_13058_begin_0 = const()[name = tensor<string, []>("op_13058_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_13058_end_0 = const()[name = tensor<string, []>("op_13058_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_13058_end_mask_0 = const()[name = tensor<string, []>("op_13058_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13058_cast = slice_by_index(begin = var_13058_begin_0, end = var_13058_end_0, end_mask = var_13058_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13058_cast")];
            tensor<int32, [4]> var_13062_begin_0 = const()[name = tensor<string, []>("op_13062_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_13062_end_0 = const()[name = tensor<string, []>("op_13062_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_13062_end_mask_0 = const()[name = tensor<string, []>("op_13062_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13062_cast = slice_by_index(begin = var_13062_begin_0, end = var_13062_end_0, end_mask = var_13062_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13062_cast")];
            tensor<int32, [4]> var_13066_begin_0 = const()[name = tensor<string, []>("op_13066_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_13066_end_0 = const()[name = tensor<string, []>("op_13066_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_13066_end_mask_0 = const()[name = tensor<string, []>("op_13066_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13066_cast = slice_by_index(begin = var_13066_begin_0, end = var_13066_end_0, end_mask = var_13066_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13066_cast")];
            tensor<int32, [4]> var_13070_begin_0 = const()[name = tensor<string, []>("op_13070_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_13070_end_0 = const()[name = tensor<string, []>("op_13070_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_13070_end_mask_0 = const()[name = tensor<string, []>("op_13070_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13070_cast = slice_by_index(begin = var_13070_begin_0, end = var_13070_end_0, end_mask = var_13070_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13070_cast")];
            tensor<int32, [4]> var_13074_begin_0 = const()[name = tensor<string, []>("op_13074_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_13074_end_0 = const()[name = tensor<string, []>("op_13074_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_13074_end_mask_0 = const()[name = tensor<string, []>("op_13074_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13074_cast = slice_by_index(begin = var_13074_begin_0, end = var_13074_end_0, end_mask = var_13074_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13074_cast")];
            tensor<int32, [4]> var_13078_begin_0 = const()[name = tensor<string, []>("op_13078_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_13078_end_0 = const()[name = tensor<string, []>("op_13078_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_13078_end_mask_0 = const()[name = tensor<string, []>("op_13078_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13078_cast = slice_by_index(begin = var_13078_begin_0, end = var_13078_end_0, end_mask = var_13078_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13078_cast")];
            tensor<int32, [4]> var_13082_begin_0 = const()[name = tensor<string, []>("op_13082_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_13082_end_0 = const()[name = tensor<string, []>("op_13082_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_13082_end_mask_0 = const()[name = tensor<string, []>("op_13082_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13082_cast = slice_by_index(begin = var_13082_begin_0, end = var_13082_end_0, end_mask = var_13082_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13082_cast")];
            tensor<int32, [4]> var_13086_begin_0 = const()[name = tensor<string, []>("op_13086_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_13086_end_0 = const()[name = tensor<string, []>("op_13086_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_13086_end_mask_0 = const()[name = tensor<string, []>("op_13086_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13086_cast = slice_by_index(begin = var_13086_begin_0, end = var_13086_end_0, end_mask = var_13086_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13086_cast")];
            tensor<int32, [4]> var_13090_begin_0 = const()[name = tensor<string, []>("op_13090_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_13090_end_0 = const()[name = tensor<string, []>("op_13090_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_13090_end_mask_0 = const()[name = tensor<string, []>("op_13090_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13090_cast = slice_by_index(begin = var_13090_begin_0, end = var_13090_end_0, end_mask = var_13090_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13090_cast")];
            tensor<int32, [4]> var_13094_begin_0 = const()[name = tensor<string, []>("op_13094_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_13094_end_0 = const()[name = tensor<string, []>("op_13094_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_13094_end_mask_0 = const()[name = tensor<string, []>("op_13094_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13094_cast = slice_by_index(begin = var_13094_begin_0, end = var_13094_end_0, end_mask = var_13094_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13094_cast")];
            tensor<int32, [4]> var_13098_begin_0 = const()[name = tensor<string, []>("op_13098_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_13098_end_0 = const()[name = tensor<string, []>("op_13098_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_13098_end_mask_0 = const()[name = tensor<string, []>("op_13098_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13098_cast = slice_by_index(begin = var_13098_begin_0, end = var_13098_end_0, end_mask = var_13098_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13098_cast")];
            tensor<int32, [4]> var_13102_begin_0 = const()[name = tensor<string, []>("op_13102_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_13102_end_0 = const()[name = tensor<string, []>("op_13102_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_13102_end_mask_0 = const()[name = tensor<string, []>("op_13102_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13102_cast = slice_by_index(begin = var_13102_begin_0, end = var_13102_end_0, end_mask = var_13102_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13102_cast")];
            tensor<int32, [4]> var_13106_begin_0 = const()[name = tensor<string, []>("op_13106_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_13106_end_0 = const()[name = tensor<string, []>("op_13106_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_13106_end_mask_0 = const()[name = tensor<string, []>("op_13106_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13106_cast = slice_by_index(begin = var_13106_begin_0, end = var_13106_end_0, end_mask = var_13106_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13106_cast")];
            tensor<int32, [4]> var_13110_begin_0 = const()[name = tensor<string, []>("op_13110_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_13110_end_0 = const()[name = tensor<string, []>("op_13110_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_13110_end_mask_0 = const()[name = tensor<string, []>("op_13110_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13110_cast = slice_by_index(begin = var_13110_begin_0, end = var_13110_end_0, end_mask = var_13110_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13110_cast")];
            tensor<int32, [4]> var_13114_begin_0 = const()[name = tensor<string, []>("op_13114_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_13114_end_0 = const()[name = tensor<string, []>("op_13114_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_13114_end_mask_0 = const()[name = tensor<string, []>("op_13114_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13114_cast = slice_by_index(begin = var_13114_begin_0, end = var_13114_end_0, end_mask = var_13114_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13114_cast")];
            tensor<int32, [4]> var_13118_begin_0 = const()[name = tensor<string, []>("op_13118_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_13118_end_0 = const()[name = tensor<string, []>("op_13118_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_13118_end_mask_0 = const()[name = tensor<string, []>("op_13118_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13118_cast = slice_by_index(begin = var_13118_begin_0, end = var_13118_end_0, end_mask = var_13118_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13118_cast")];
            tensor<int32, [4]> var_13122_begin_0 = const()[name = tensor<string, []>("op_13122_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_13122_end_0 = const()[name = tensor<string, []>("op_13122_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_13122_end_mask_0 = const()[name = tensor<string, []>("op_13122_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13122_cast = slice_by_index(begin = var_13122_begin_0, end = var_13122_end_0, end_mask = var_13122_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13122_cast")];
            tensor<int32, [4]> var_13126_begin_0 = const()[name = tensor<string, []>("op_13126_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_13126_end_0 = const()[name = tensor<string, []>("op_13126_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_13126_end_mask_0 = const()[name = tensor<string, []>("op_13126_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13126_cast = slice_by_index(begin = var_13126_begin_0, end = var_13126_end_0, end_mask = var_13126_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13126_cast")];
            tensor<string, []> var_13130_equation_0 = const()[name = tensor<string, []>("op_13130_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13130_cast = einsum(equation = var_13130_equation_0, values = (var_12972_cast, var_12889_cast))[name = tensor<string, []>("op_13130_cast")];
            tensor<fp16, []> var_13131_to_fp16 = const()[name = tensor<string, []>("op_13131_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1041_cast = mul(x = var_13130_cast, y = var_13131_to_fp16)[name = tensor<string, []>("aw_1041_cast")];
            tensor<string, []> var_13134_equation_0 = const()[name = tensor<string, []>("op_13134_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13134_cast = einsum(equation = var_13134_equation_0, values = (var_12976_cast, var_12893_cast))[name = tensor<string, []>("op_13134_cast")];
            tensor<fp16, []> var_13135_to_fp16 = const()[name = tensor<string, []>("op_13135_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1043_cast = mul(x = var_13134_cast, y = var_13135_to_fp16)[name = tensor<string, []>("aw_1043_cast")];
            tensor<string, []> var_13138_equation_0 = const()[name = tensor<string, []>("op_13138_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13138_cast = einsum(equation = var_13138_equation_0, values = (var_12980_cast, var_12897_cast))[name = tensor<string, []>("op_13138_cast")];
            tensor<fp16, []> var_13139_to_fp16 = const()[name = tensor<string, []>("op_13139_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1045_cast = mul(x = var_13138_cast, y = var_13139_to_fp16)[name = tensor<string, []>("aw_1045_cast")];
            tensor<string, []> var_13142_equation_0 = const()[name = tensor<string, []>("op_13142_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13142_cast = einsum(equation = var_13142_equation_0, values = (var_12984_cast, var_12901_cast))[name = tensor<string, []>("op_13142_cast")];
            tensor<fp16, []> var_13143_to_fp16 = const()[name = tensor<string, []>("op_13143_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1047_cast = mul(x = var_13142_cast, y = var_13143_to_fp16)[name = tensor<string, []>("aw_1047_cast")];
            tensor<string, []> var_13146_equation_0 = const()[name = tensor<string, []>("op_13146_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13146_cast = einsum(equation = var_13146_equation_0, values = (var_12988_cast, var_12905_cast))[name = tensor<string, []>("op_13146_cast")];
            tensor<fp16, []> var_13147_to_fp16 = const()[name = tensor<string, []>("op_13147_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1049_cast = mul(x = var_13146_cast, y = var_13147_to_fp16)[name = tensor<string, []>("aw_1049_cast")];
            tensor<string, []> var_13150_equation_0 = const()[name = tensor<string, []>("op_13150_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13150_cast = einsum(equation = var_13150_equation_0, values = (var_12992_cast, var_12909_cast))[name = tensor<string, []>("op_13150_cast")];
            tensor<fp16, []> var_13151_to_fp16 = const()[name = tensor<string, []>("op_13151_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1051_cast = mul(x = var_13150_cast, y = var_13151_to_fp16)[name = tensor<string, []>("aw_1051_cast")];
            tensor<string, []> var_13154_equation_0 = const()[name = tensor<string, []>("op_13154_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13154_cast = einsum(equation = var_13154_equation_0, values = (var_12996_cast, var_12913_cast))[name = tensor<string, []>("op_13154_cast")];
            tensor<fp16, []> var_13155_to_fp16 = const()[name = tensor<string, []>("op_13155_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1053_cast = mul(x = var_13154_cast, y = var_13155_to_fp16)[name = tensor<string, []>("aw_1053_cast")];
            tensor<string, []> var_13158_equation_0 = const()[name = tensor<string, []>("op_13158_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13158_cast = einsum(equation = var_13158_equation_0, values = (var_13000_cast, var_12917_cast))[name = tensor<string, []>("op_13158_cast")];
            tensor<fp16, []> var_13159_to_fp16 = const()[name = tensor<string, []>("op_13159_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1055_cast = mul(x = var_13158_cast, y = var_13159_to_fp16)[name = tensor<string, []>("aw_1055_cast")];
            tensor<string, []> var_13162_equation_0 = const()[name = tensor<string, []>("op_13162_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13162_cast = einsum(equation = var_13162_equation_0, values = (var_13004_cast, var_12921_cast))[name = tensor<string, []>("op_13162_cast")];
            tensor<fp16, []> var_13163_to_fp16 = const()[name = tensor<string, []>("op_13163_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1057_cast = mul(x = var_13162_cast, y = var_13163_to_fp16)[name = tensor<string, []>("aw_1057_cast")];
            tensor<string, []> var_13166_equation_0 = const()[name = tensor<string, []>("op_13166_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13166_cast = einsum(equation = var_13166_equation_0, values = (var_13008_cast, var_12925_cast))[name = tensor<string, []>("op_13166_cast")];
            tensor<fp16, []> var_13167_to_fp16 = const()[name = tensor<string, []>("op_13167_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1059_cast = mul(x = var_13166_cast, y = var_13167_to_fp16)[name = tensor<string, []>("aw_1059_cast")];
            tensor<string, []> var_13170_equation_0 = const()[name = tensor<string, []>("op_13170_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13170_cast = einsum(equation = var_13170_equation_0, values = (var_13012_cast, var_12929_cast))[name = tensor<string, []>("op_13170_cast")];
            tensor<fp16, []> var_13171_to_fp16 = const()[name = tensor<string, []>("op_13171_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1061_cast = mul(x = var_13170_cast, y = var_13171_to_fp16)[name = tensor<string, []>("aw_1061_cast")];
            tensor<string, []> var_13174_equation_0 = const()[name = tensor<string, []>("op_13174_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13174_cast = einsum(equation = var_13174_equation_0, values = (var_13016_cast, var_12933_cast))[name = tensor<string, []>("op_13174_cast")];
            tensor<fp16, []> var_13175_to_fp16 = const()[name = tensor<string, []>("op_13175_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1063_cast = mul(x = var_13174_cast, y = var_13175_to_fp16)[name = tensor<string, []>("aw_1063_cast")];
            tensor<string, []> var_13178_equation_0 = const()[name = tensor<string, []>("op_13178_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13178_cast = einsum(equation = var_13178_equation_0, values = (var_13020_cast, var_12937_cast))[name = tensor<string, []>("op_13178_cast")];
            tensor<fp16, []> var_13179_to_fp16 = const()[name = tensor<string, []>("op_13179_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1065_cast = mul(x = var_13178_cast, y = var_13179_to_fp16)[name = tensor<string, []>("aw_1065_cast")];
            tensor<string, []> var_13182_equation_0 = const()[name = tensor<string, []>("op_13182_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13182_cast = einsum(equation = var_13182_equation_0, values = (var_13024_cast, var_12941_cast))[name = tensor<string, []>("op_13182_cast")];
            tensor<fp16, []> var_13183_to_fp16 = const()[name = tensor<string, []>("op_13183_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1067_cast = mul(x = var_13182_cast, y = var_13183_to_fp16)[name = tensor<string, []>("aw_1067_cast")];
            tensor<string, []> var_13186_equation_0 = const()[name = tensor<string, []>("op_13186_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13186_cast = einsum(equation = var_13186_equation_0, values = (var_13028_cast, var_12945_cast))[name = tensor<string, []>("op_13186_cast")];
            tensor<fp16, []> var_13187_to_fp16 = const()[name = tensor<string, []>("op_13187_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1069_cast = mul(x = var_13186_cast, y = var_13187_to_fp16)[name = tensor<string, []>("aw_1069_cast")];
            tensor<string, []> var_13190_equation_0 = const()[name = tensor<string, []>("op_13190_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13190_cast = einsum(equation = var_13190_equation_0, values = (var_13032_cast, var_12949_cast))[name = tensor<string, []>("op_13190_cast")];
            tensor<fp16, []> var_13191_to_fp16 = const()[name = tensor<string, []>("op_13191_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1071_cast = mul(x = var_13190_cast, y = var_13191_to_fp16)[name = tensor<string, []>("aw_1071_cast")];
            tensor<string, []> var_13194_equation_0 = const()[name = tensor<string, []>("op_13194_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13194_cast = einsum(equation = var_13194_equation_0, values = (var_13036_cast, var_12953_cast))[name = tensor<string, []>("op_13194_cast")];
            tensor<fp16, []> var_13195_to_fp16 = const()[name = tensor<string, []>("op_13195_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1073_cast = mul(x = var_13194_cast, y = var_13195_to_fp16)[name = tensor<string, []>("aw_1073_cast")];
            tensor<string, []> var_13198_equation_0 = const()[name = tensor<string, []>("op_13198_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13198_cast = einsum(equation = var_13198_equation_0, values = (var_13040_cast, var_12957_cast))[name = tensor<string, []>("op_13198_cast")];
            tensor<fp16, []> var_13199_to_fp16 = const()[name = tensor<string, []>("op_13199_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1075_cast = mul(x = var_13198_cast, y = var_13199_to_fp16)[name = tensor<string, []>("aw_1075_cast")];
            tensor<string, []> var_13202_equation_0 = const()[name = tensor<string, []>("op_13202_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13202_cast = einsum(equation = var_13202_equation_0, values = (var_13044_cast, var_12961_cast))[name = tensor<string, []>("op_13202_cast")];
            tensor<fp16, []> var_13203_to_fp16 = const()[name = tensor<string, []>("op_13203_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1077_cast = mul(x = var_13202_cast, y = var_13203_to_fp16)[name = tensor<string, []>("aw_1077_cast")];
            tensor<string, []> var_13206_equation_0 = const()[name = tensor<string, []>("op_13206_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13206_cast = einsum(equation = var_13206_equation_0, values = (var_13048_cast, var_12965_cast))[name = tensor<string, []>("op_13206_cast")];
            tensor<fp16, []> var_13207_to_fp16 = const()[name = tensor<string, []>("op_13207_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1079_cast = mul(x = var_13206_cast, y = var_13207_to_fp16)[name = tensor<string, []>("aw_1079_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13209_cast = softmax(axis = var_2634, x = aw_1041_cast)[name = tensor<string, []>("op_13209_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13210_cast = softmax(axis = var_2634, x = aw_1043_cast)[name = tensor<string, []>("op_13210_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13211_cast = softmax(axis = var_2634, x = aw_1045_cast)[name = tensor<string, []>("op_13211_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13212_cast = softmax(axis = var_2634, x = aw_1047_cast)[name = tensor<string, []>("op_13212_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13213_cast = softmax(axis = var_2634, x = aw_1049_cast)[name = tensor<string, []>("op_13213_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13214_cast = softmax(axis = var_2634, x = aw_1051_cast)[name = tensor<string, []>("op_13214_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13215_cast = softmax(axis = var_2634, x = aw_1053_cast)[name = tensor<string, []>("op_13215_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13216_cast = softmax(axis = var_2634, x = aw_1055_cast)[name = tensor<string, []>("op_13216_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13217_cast = softmax(axis = var_2634, x = aw_1057_cast)[name = tensor<string, []>("op_13217_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13218_cast = softmax(axis = var_2634, x = aw_1059_cast)[name = tensor<string, []>("op_13218_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13219_cast = softmax(axis = var_2634, x = aw_1061_cast)[name = tensor<string, []>("op_13219_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13220_cast = softmax(axis = var_2634, x = aw_1063_cast)[name = tensor<string, []>("op_13220_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13221_cast = softmax(axis = var_2634, x = aw_1065_cast)[name = tensor<string, []>("op_13221_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13222_cast = softmax(axis = var_2634, x = aw_1067_cast)[name = tensor<string, []>("op_13222_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13223_cast = softmax(axis = var_2634, x = aw_1069_cast)[name = tensor<string, []>("op_13223_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13224_cast = softmax(axis = var_2634, x = aw_1071_cast)[name = tensor<string, []>("op_13224_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13225_cast = softmax(axis = var_2634, x = aw_1073_cast)[name = tensor<string, []>("op_13225_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13226_cast = softmax(axis = var_2634, x = aw_1075_cast)[name = tensor<string, []>("op_13226_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13227_cast = softmax(axis = var_2634, x = aw_1077_cast)[name = tensor<string, []>("op_13227_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_13228_cast = softmax(axis = var_2634, x = aw_1079_cast)[name = tensor<string, []>("op_13228_cast")];
            tensor<string, []> var_13230_equation_0 = const()[name = tensor<string, []>("op_13230_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13230_cast = einsum(equation = var_13230_equation_0, values = (var_13050_cast, var_13209_cast))[name = tensor<string, []>("op_13230_cast")];
            tensor<string, []> var_13232_equation_0 = const()[name = tensor<string, []>("op_13232_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13232_cast = einsum(equation = var_13232_equation_0, values = (var_13054_cast, var_13210_cast))[name = tensor<string, []>("op_13232_cast")];
            tensor<string, []> var_13234_equation_0 = const()[name = tensor<string, []>("op_13234_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13234_cast = einsum(equation = var_13234_equation_0, values = (var_13058_cast, var_13211_cast))[name = tensor<string, []>("op_13234_cast")];
            tensor<string, []> var_13236_equation_0 = const()[name = tensor<string, []>("op_13236_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13236_cast = einsum(equation = var_13236_equation_0, values = (var_13062_cast, var_13212_cast))[name = tensor<string, []>("op_13236_cast")];
            tensor<string, []> var_13238_equation_0 = const()[name = tensor<string, []>("op_13238_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13238_cast = einsum(equation = var_13238_equation_0, values = (var_13066_cast, var_13213_cast))[name = tensor<string, []>("op_13238_cast")];
            tensor<string, []> var_13240_equation_0 = const()[name = tensor<string, []>("op_13240_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13240_cast = einsum(equation = var_13240_equation_0, values = (var_13070_cast, var_13214_cast))[name = tensor<string, []>("op_13240_cast")];
            tensor<string, []> var_13242_equation_0 = const()[name = tensor<string, []>("op_13242_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13242_cast = einsum(equation = var_13242_equation_0, values = (var_13074_cast, var_13215_cast))[name = tensor<string, []>("op_13242_cast")];
            tensor<string, []> var_13244_equation_0 = const()[name = tensor<string, []>("op_13244_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13244_cast = einsum(equation = var_13244_equation_0, values = (var_13078_cast, var_13216_cast))[name = tensor<string, []>("op_13244_cast")];
            tensor<string, []> var_13246_equation_0 = const()[name = tensor<string, []>("op_13246_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13246_cast = einsum(equation = var_13246_equation_0, values = (var_13082_cast, var_13217_cast))[name = tensor<string, []>("op_13246_cast")];
            tensor<string, []> var_13248_equation_0 = const()[name = tensor<string, []>("op_13248_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13248_cast = einsum(equation = var_13248_equation_0, values = (var_13086_cast, var_13218_cast))[name = tensor<string, []>("op_13248_cast")];
            tensor<string, []> var_13250_equation_0 = const()[name = tensor<string, []>("op_13250_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13250_cast = einsum(equation = var_13250_equation_0, values = (var_13090_cast, var_13219_cast))[name = tensor<string, []>("op_13250_cast")];
            tensor<string, []> var_13252_equation_0 = const()[name = tensor<string, []>("op_13252_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13252_cast = einsum(equation = var_13252_equation_0, values = (var_13094_cast, var_13220_cast))[name = tensor<string, []>("op_13252_cast")];
            tensor<string, []> var_13254_equation_0 = const()[name = tensor<string, []>("op_13254_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13254_cast = einsum(equation = var_13254_equation_0, values = (var_13098_cast, var_13221_cast))[name = tensor<string, []>("op_13254_cast")];
            tensor<string, []> var_13256_equation_0 = const()[name = tensor<string, []>("op_13256_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13256_cast = einsum(equation = var_13256_equation_0, values = (var_13102_cast, var_13222_cast))[name = tensor<string, []>("op_13256_cast")];
            tensor<string, []> var_13258_equation_0 = const()[name = tensor<string, []>("op_13258_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13258_cast = einsum(equation = var_13258_equation_0, values = (var_13106_cast, var_13223_cast))[name = tensor<string, []>("op_13258_cast")];
            tensor<string, []> var_13260_equation_0 = const()[name = tensor<string, []>("op_13260_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13260_cast = einsum(equation = var_13260_equation_0, values = (var_13110_cast, var_13224_cast))[name = tensor<string, []>("op_13260_cast")];
            tensor<string, []> var_13262_equation_0 = const()[name = tensor<string, []>("op_13262_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13262_cast = einsum(equation = var_13262_equation_0, values = (var_13114_cast, var_13225_cast))[name = tensor<string, []>("op_13262_cast")];
            tensor<string, []> var_13264_equation_0 = const()[name = tensor<string, []>("op_13264_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13264_cast = einsum(equation = var_13264_equation_0, values = (var_13118_cast, var_13226_cast))[name = tensor<string, []>("op_13264_cast")];
            tensor<string, []> var_13266_equation_0 = const()[name = tensor<string, []>("op_13266_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13266_cast = einsum(equation = var_13266_equation_0, values = (var_13122_cast, var_13227_cast))[name = tensor<string, []>("op_13266_cast")];
            tensor<string, []> var_13268_equation_0 = const()[name = tensor<string, []>("op_13268_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13268_cast = einsum(equation = var_13268_equation_0, values = (var_13126_cast, var_13228_cast))[name = tensor<string, []>("op_13268_cast")];
            tensor<bool, []> input_241_interleave_0 = const()[name = tensor<string, []>("input_241_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_241_cast = concat(axis = var_2634, interleave = input_241_interleave_0, values = (var_13230_cast, var_13232_cast, var_13234_cast, var_13236_cast, var_13238_cast, var_13240_cast, var_13242_cast, var_13244_cast, var_13246_cast, var_13248_cast, var_13250_cast, var_13252_cast, var_13254_cast, var_13256_cast, var_13258_cast, var_13260_cast, var_13262_cast, var_13264_cast, var_13266_cast, var_13268_cast))[name = tensor<string, []>("input_241_cast")];
            tensor<int32, [2]> var_13274 = const()[name = tensor<string, []>("op_13274"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13276 = const()[name = tensor<string, []>("op_13276"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_13278_pad_type_0 = const()[name = tensor<string, []>("op_13278_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_13278_pad_0 = const()[name = tensor<string, []>("op_13278_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1041486272)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1044763136)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_13278_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_13276, groups = var_2634, pad = var_13278_pad_0, pad_type = var_13278_pad_type_0, strides = var_13274, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_241_cast)[name = tensor<string, []>("op_13278_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_93_cast = add(x = var_13278_cast, y = inputs_91_cast)[name = tensor<string, []>("inputs_93_cast")];
            tensor<int32, [1]> var_13282 = const()[name = tensor<string, []>("op_13282"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_93_cast = reduce_mean(axes = var_13282, keep_dims = var_2629, x = inputs_93_cast)[name = tensor<string, []>("channels_mean_93_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_93_cast = sub(x = inputs_93_cast, y = channels_mean_93_cast)[name = tensor<string, []>("zero_mean_93_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_93_cast = mul(x = zero_mean_93_cast, y = zero_mean_93_cast)[name = tensor<string, []>("zero_mean_sq_93_cast")];
            tensor<int32, [1]> var_13286 = const()[name = tensor<string, []>("op_13286"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_13287_cast = reduce_mean(axes = var_13286, keep_dims = var_2629, x = zero_mean_sq_93_cast)[name = tensor<string, []>("op_13287_cast")];
            tensor<fp16, []> var_13288_to_fp16 = const()[name = tensor<string, []>("op_13288_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_13289_cast = add(x = var_13287_cast, y = var_13288_to_fp16)[name = tensor<string, []>("op_13289_cast")];
            tensor<fp16, []> denom_93_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_93_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_93_cast = rsqrt(epsilon = denom_93_epsilon_0_to_fp16, x = var_13289_cast)[name = tensor<string, []>("denom_93_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_93_cast = mul(x = zero_mean_93_cast, y = denom_93_cast)[name = tensor<string, []>("out_93_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_13293_to_fp16 = const()[name = tensor<string, []>("op_13293_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1044765760)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_13294_cast = add(x = out_93_cast, y = var_13293_to_fp16)[name = tensor<string, []>("op_13294_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_13296_to_fp16 = const()[name = tensor<string, []>("op_13296_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1044768384)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_145_cast = mul(x = var_13294_cast, y = var_13296_to_fp16)[name = tensor<string, []>("hidden_states_145_cast")];
            tensor<int32, [2]> var_13303 = const()[name = tensor<string, []>("op_13303"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13305 = const()[name = tensor<string, []>("op_13305"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_63_pad_type_0 = const()[name = tensor<string, []>("q_63_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_63_pad_0 = const()[name = tensor<string, []>("q_63_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1044771008)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_63_cast = conv(dilations = var_13305, groups = var_2634, pad = q_63_pad_0, pad_type = q_63_pad_type_0, strides = var_13303, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_145_cast)[name = tensor<string, []>("q_63_cast")];
            tensor<int32, [2]> var_13309 = const()[name = tensor<string, []>("op_13309"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13311 = const()[name = tensor<string, []>("op_13311"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_125_pad_type_0 = const()[name = tensor<string, []>("k_125_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_125_pad_0 = const()[name = tensor<string, []>("k_125_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1048047872)))];
            tensor<fp16, [2, 1280, 1, 77]> k_125_cast = conv(dilations = var_13311, groups = var_2634, pad = k_125_pad_0, pad_type = k_125_pad_type_0, strides = var_13309, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_125_cast")];
            tensor<int32, [2]> var_13315 = const()[name = tensor<string, []>("op_13315"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13317 = const()[name = tensor<string, []>("op_13317"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_63_pad_type_0 = const()[name = tensor<string, []>("v_63_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_63_pad_0 = const()[name = tensor<string, []>("v_63_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1053290816)))];
            tensor<fp16, [2, 1280, 1, 77]> v_63_cast = conv(dilations = var_13317, groups = var_2634, pad = v_63_pad_0, pad_type = v_63_pad_type_0, strides = var_13315, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_63_cast")];
            tensor<int32, [4]> var_13321_begin_0 = const()[name = tensor<string, []>("op_13321_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13321_end_0 = const()[name = tensor<string, []>("op_13321_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_13321_end_mask_0 = const()[name = tensor<string, []>("op_13321_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13321_cast = slice_by_index(begin = var_13321_begin_0, end = var_13321_end_0, end_mask = var_13321_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13321_cast")];
            tensor<int32, [4]> var_13325_begin_0 = const()[name = tensor<string, []>("op_13325_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_13325_end_0 = const()[name = tensor<string, []>("op_13325_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_13325_end_mask_0 = const()[name = tensor<string, []>("op_13325_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13325_cast = slice_by_index(begin = var_13325_begin_0, end = var_13325_end_0, end_mask = var_13325_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13325_cast")];
            tensor<int32, [4]> var_13329_begin_0 = const()[name = tensor<string, []>("op_13329_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_13329_end_0 = const()[name = tensor<string, []>("op_13329_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_13329_end_mask_0 = const()[name = tensor<string, []>("op_13329_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13329_cast = slice_by_index(begin = var_13329_begin_0, end = var_13329_end_0, end_mask = var_13329_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13329_cast")];
            tensor<int32, [4]> var_13333_begin_0 = const()[name = tensor<string, []>("op_13333_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_13333_end_0 = const()[name = tensor<string, []>("op_13333_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_13333_end_mask_0 = const()[name = tensor<string, []>("op_13333_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13333_cast = slice_by_index(begin = var_13333_begin_0, end = var_13333_end_0, end_mask = var_13333_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13333_cast")];
            tensor<int32, [4]> var_13337_begin_0 = const()[name = tensor<string, []>("op_13337_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_13337_end_0 = const()[name = tensor<string, []>("op_13337_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_13337_end_mask_0 = const()[name = tensor<string, []>("op_13337_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13337_cast = slice_by_index(begin = var_13337_begin_0, end = var_13337_end_0, end_mask = var_13337_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13337_cast")];
            tensor<int32, [4]> var_13341_begin_0 = const()[name = tensor<string, []>("op_13341_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_13341_end_0 = const()[name = tensor<string, []>("op_13341_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_13341_end_mask_0 = const()[name = tensor<string, []>("op_13341_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13341_cast = slice_by_index(begin = var_13341_begin_0, end = var_13341_end_0, end_mask = var_13341_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13341_cast")];
            tensor<int32, [4]> var_13345_begin_0 = const()[name = tensor<string, []>("op_13345_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_13345_end_0 = const()[name = tensor<string, []>("op_13345_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_13345_end_mask_0 = const()[name = tensor<string, []>("op_13345_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13345_cast = slice_by_index(begin = var_13345_begin_0, end = var_13345_end_0, end_mask = var_13345_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13345_cast")];
            tensor<int32, [4]> var_13349_begin_0 = const()[name = tensor<string, []>("op_13349_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_13349_end_0 = const()[name = tensor<string, []>("op_13349_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_13349_end_mask_0 = const()[name = tensor<string, []>("op_13349_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13349_cast = slice_by_index(begin = var_13349_begin_0, end = var_13349_end_0, end_mask = var_13349_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13349_cast")];
            tensor<int32, [4]> var_13353_begin_0 = const()[name = tensor<string, []>("op_13353_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_13353_end_0 = const()[name = tensor<string, []>("op_13353_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_13353_end_mask_0 = const()[name = tensor<string, []>("op_13353_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13353_cast = slice_by_index(begin = var_13353_begin_0, end = var_13353_end_0, end_mask = var_13353_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13353_cast")];
            tensor<int32, [4]> var_13357_begin_0 = const()[name = tensor<string, []>("op_13357_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_13357_end_0 = const()[name = tensor<string, []>("op_13357_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_13357_end_mask_0 = const()[name = tensor<string, []>("op_13357_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13357_cast = slice_by_index(begin = var_13357_begin_0, end = var_13357_end_0, end_mask = var_13357_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13357_cast")];
            tensor<int32, [4]> var_13361_begin_0 = const()[name = tensor<string, []>("op_13361_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_13361_end_0 = const()[name = tensor<string, []>("op_13361_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_13361_end_mask_0 = const()[name = tensor<string, []>("op_13361_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13361_cast = slice_by_index(begin = var_13361_begin_0, end = var_13361_end_0, end_mask = var_13361_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13361_cast")];
            tensor<int32, [4]> var_13365_begin_0 = const()[name = tensor<string, []>("op_13365_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_13365_end_0 = const()[name = tensor<string, []>("op_13365_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_13365_end_mask_0 = const()[name = tensor<string, []>("op_13365_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13365_cast = slice_by_index(begin = var_13365_begin_0, end = var_13365_end_0, end_mask = var_13365_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13365_cast")];
            tensor<int32, [4]> var_13369_begin_0 = const()[name = tensor<string, []>("op_13369_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_13369_end_0 = const()[name = tensor<string, []>("op_13369_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_13369_end_mask_0 = const()[name = tensor<string, []>("op_13369_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13369_cast = slice_by_index(begin = var_13369_begin_0, end = var_13369_end_0, end_mask = var_13369_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13369_cast")];
            tensor<int32, [4]> var_13373_begin_0 = const()[name = tensor<string, []>("op_13373_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_13373_end_0 = const()[name = tensor<string, []>("op_13373_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_13373_end_mask_0 = const()[name = tensor<string, []>("op_13373_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13373_cast = slice_by_index(begin = var_13373_begin_0, end = var_13373_end_0, end_mask = var_13373_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13373_cast")];
            tensor<int32, [4]> var_13377_begin_0 = const()[name = tensor<string, []>("op_13377_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_13377_end_0 = const()[name = tensor<string, []>("op_13377_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_13377_end_mask_0 = const()[name = tensor<string, []>("op_13377_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13377_cast = slice_by_index(begin = var_13377_begin_0, end = var_13377_end_0, end_mask = var_13377_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13377_cast")];
            tensor<int32, [4]> var_13381_begin_0 = const()[name = tensor<string, []>("op_13381_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_13381_end_0 = const()[name = tensor<string, []>("op_13381_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_13381_end_mask_0 = const()[name = tensor<string, []>("op_13381_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13381_cast = slice_by_index(begin = var_13381_begin_0, end = var_13381_end_0, end_mask = var_13381_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13381_cast")];
            tensor<int32, [4]> var_13385_begin_0 = const()[name = tensor<string, []>("op_13385_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_13385_end_0 = const()[name = tensor<string, []>("op_13385_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_13385_end_mask_0 = const()[name = tensor<string, []>("op_13385_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13385_cast = slice_by_index(begin = var_13385_begin_0, end = var_13385_end_0, end_mask = var_13385_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13385_cast")];
            tensor<int32, [4]> var_13389_begin_0 = const()[name = tensor<string, []>("op_13389_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_13389_end_0 = const()[name = tensor<string, []>("op_13389_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_13389_end_mask_0 = const()[name = tensor<string, []>("op_13389_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13389_cast = slice_by_index(begin = var_13389_begin_0, end = var_13389_end_0, end_mask = var_13389_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13389_cast")];
            tensor<int32, [4]> var_13393_begin_0 = const()[name = tensor<string, []>("op_13393_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_13393_end_0 = const()[name = tensor<string, []>("op_13393_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_13393_end_mask_0 = const()[name = tensor<string, []>("op_13393_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13393_cast = slice_by_index(begin = var_13393_begin_0, end = var_13393_end_0, end_mask = var_13393_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13393_cast")];
            tensor<int32, [4]> var_13397_begin_0 = const()[name = tensor<string, []>("op_13397_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_13397_end_0 = const()[name = tensor<string, []>("op_13397_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_13397_end_mask_0 = const()[name = tensor<string, []>("op_13397_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13397_cast = slice_by_index(begin = var_13397_begin_0, end = var_13397_end_0, end_mask = var_13397_end_mask_0, x = q_63_cast)[name = tensor<string, []>("op_13397_cast")];
            tensor<int32, [4]> k_127_perm_0 = const()[name = tensor<string, []>("k_127_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_13404_begin_0 = const()[name = tensor<string, []>("op_13404_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13404_end_0 = const()[name = tensor<string, []>("op_13404_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_13404_end_mask_0 = const()[name = tensor<string, []>("op_13404_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_108 = transpose(perm = k_127_perm_0, x = k_125_cast)[name = tensor<string, []>("transpose_108")];
            tensor<fp16, [2, 77, 1, 64]> var_13404_cast = slice_by_index(begin = var_13404_begin_0, end = var_13404_end_0, end_mask = var_13404_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13404_cast")];
            tensor<int32, [4]> var_13408_begin_0 = const()[name = tensor<string, []>("op_13408_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_13408_end_0 = const()[name = tensor<string, []>("op_13408_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_13408_end_mask_0 = const()[name = tensor<string, []>("op_13408_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13408_cast = slice_by_index(begin = var_13408_begin_0, end = var_13408_end_0, end_mask = var_13408_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13408_cast")];
            tensor<int32, [4]> var_13412_begin_0 = const()[name = tensor<string, []>("op_13412_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_13412_end_0 = const()[name = tensor<string, []>("op_13412_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_13412_end_mask_0 = const()[name = tensor<string, []>("op_13412_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13412_cast = slice_by_index(begin = var_13412_begin_0, end = var_13412_end_0, end_mask = var_13412_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13412_cast")];
            tensor<int32, [4]> var_13416_begin_0 = const()[name = tensor<string, []>("op_13416_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_13416_end_0 = const()[name = tensor<string, []>("op_13416_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_13416_end_mask_0 = const()[name = tensor<string, []>("op_13416_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13416_cast = slice_by_index(begin = var_13416_begin_0, end = var_13416_end_0, end_mask = var_13416_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13416_cast")];
            tensor<int32, [4]> var_13420_begin_0 = const()[name = tensor<string, []>("op_13420_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_13420_end_0 = const()[name = tensor<string, []>("op_13420_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_13420_end_mask_0 = const()[name = tensor<string, []>("op_13420_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13420_cast = slice_by_index(begin = var_13420_begin_0, end = var_13420_end_0, end_mask = var_13420_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13420_cast")];
            tensor<int32, [4]> var_13424_begin_0 = const()[name = tensor<string, []>("op_13424_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_13424_end_0 = const()[name = tensor<string, []>("op_13424_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_13424_end_mask_0 = const()[name = tensor<string, []>("op_13424_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13424_cast = slice_by_index(begin = var_13424_begin_0, end = var_13424_end_0, end_mask = var_13424_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13424_cast")];
            tensor<int32, [4]> var_13428_begin_0 = const()[name = tensor<string, []>("op_13428_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_13428_end_0 = const()[name = tensor<string, []>("op_13428_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_13428_end_mask_0 = const()[name = tensor<string, []>("op_13428_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13428_cast = slice_by_index(begin = var_13428_begin_0, end = var_13428_end_0, end_mask = var_13428_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13428_cast")];
            tensor<int32, [4]> var_13432_begin_0 = const()[name = tensor<string, []>("op_13432_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_13432_end_0 = const()[name = tensor<string, []>("op_13432_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_13432_end_mask_0 = const()[name = tensor<string, []>("op_13432_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13432_cast = slice_by_index(begin = var_13432_begin_0, end = var_13432_end_0, end_mask = var_13432_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13432_cast")];
            tensor<int32, [4]> var_13436_begin_0 = const()[name = tensor<string, []>("op_13436_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13436_end_0 = const()[name = tensor<string, []>("op_13436_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_13436_end_mask_0 = const()[name = tensor<string, []>("op_13436_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13436_cast = slice_by_index(begin = var_13436_begin_0, end = var_13436_end_0, end_mask = var_13436_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13436_cast")];
            tensor<int32, [4]> var_13440_begin_0 = const()[name = tensor<string, []>("op_13440_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_13440_end_0 = const()[name = tensor<string, []>("op_13440_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_13440_end_mask_0 = const()[name = tensor<string, []>("op_13440_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13440_cast = slice_by_index(begin = var_13440_begin_0, end = var_13440_end_0, end_mask = var_13440_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13440_cast")];
            tensor<int32, [4]> var_13444_begin_0 = const()[name = tensor<string, []>("op_13444_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_13444_end_0 = const()[name = tensor<string, []>("op_13444_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_13444_end_mask_0 = const()[name = tensor<string, []>("op_13444_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13444_cast = slice_by_index(begin = var_13444_begin_0, end = var_13444_end_0, end_mask = var_13444_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13444_cast")];
            tensor<int32, [4]> var_13448_begin_0 = const()[name = tensor<string, []>("op_13448_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_13448_end_0 = const()[name = tensor<string, []>("op_13448_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_13448_end_mask_0 = const()[name = tensor<string, []>("op_13448_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13448_cast = slice_by_index(begin = var_13448_begin_0, end = var_13448_end_0, end_mask = var_13448_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13448_cast")];
            tensor<int32, [4]> var_13452_begin_0 = const()[name = tensor<string, []>("op_13452_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_13452_end_0 = const()[name = tensor<string, []>("op_13452_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_13452_end_mask_0 = const()[name = tensor<string, []>("op_13452_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13452_cast = slice_by_index(begin = var_13452_begin_0, end = var_13452_end_0, end_mask = var_13452_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13452_cast")];
            tensor<int32, [4]> var_13456_begin_0 = const()[name = tensor<string, []>("op_13456_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_13456_end_0 = const()[name = tensor<string, []>("op_13456_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_13456_end_mask_0 = const()[name = tensor<string, []>("op_13456_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13456_cast = slice_by_index(begin = var_13456_begin_0, end = var_13456_end_0, end_mask = var_13456_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13456_cast")];
            tensor<int32, [4]> var_13460_begin_0 = const()[name = tensor<string, []>("op_13460_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_13460_end_0 = const()[name = tensor<string, []>("op_13460_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_13460_end_mask_0 = const()[name = tensor<string, []>("op_13460_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13460_cast = slice_by_index(begin = var_13460_begin_0, end = var_13460_end_0, end_mask = var_13460_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13460_cast")];
            tensor<int32, [4]> var_13464_begin_0 = const()[name = tensor<string, []>("op_13464_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_13464_end_0 = const()[name = tensor<string, []>("op_13464_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_13464_end_mask_0 = const()[name = tensor<string, []>("op_13464_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13464_cast = slice_by_index(begin = var_13464_begin_0, end = var_13464_end_0, end_mask = var_13464_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13464_cast")];
            tensor<int32, [4]> var_13468_begin_0 = const()[name = tensor<string, []>("op_13468_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13468_end_0 = const()[name = tensor<string, []>("op_13468_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_13468_end_mask_0 = const()[name = tensor<string, []>("op_13468_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13468_cast = slice_by_index(begin = var_13468_begin_0, end = var_13468_end_0, end_mask = var_13468_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13468_cast")];
            tensor<int32, [4]> var_13472_begin_0 = const()[name = tensor<string, []>("op_13472_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_13472_end_0 = const()[name = tensor<string, []>("op_13472_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_13472_end_mask_0 = const()[name = tensor<string, []>("op_13472_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13472_cast = slice_by_index(begin = var_13472_begin_0, end = var_13472_end_0, end_mask = var_13472_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13472_cast")];
            tensor<int32, [4]> var_13476_begin_0 = const()[name = tensor<string, []>("op_13476_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_13476_end_0 = const()[name = tensor<string, []>("op_13476_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_13476_end_mask_0 = const()[name = tensor<string, []>("op_13476_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13476_cast = slice_by_index(begin = var_13476_begin_0, end = var_13476_end_0, end_mask = var_13476_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13476_cast")];
            tensor<int32, [4]> var_13480_begin_0 = const()[name = tensor<string, []>("op_13480_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_13480_end_0 = const()[name = tensor<string, []>("op_13480_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_13480_end_mask_0 = const()[name = tensor<string, []>("op_13480_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13480_cast = slice_by_index(begin = var_13480_begin_0, end = var_13480_end_0, end_mask = var_13480_end_mask_0, x = transpose_108)[name = tensor<string, []>("op_13480_cast")];
            tensor<int32, [4]> var_13482_begin_0 = const()[name = tensor<string, []>("op_13482_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13482_end_0 = const()[name = tensor<string, []>("op_13482_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_13482_end_mask_0 = const()[name = tensor<string, []>("op_13482_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13482_cast = slice_by_index(begin = var_13482_begin_0, end = var_13482_end_0, end_mask = var_13482_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13482_cast")];
            tensor<int32, [4]> var_13486_begin_0 = const()[name = tensor<string, []>("op_13486_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_13486_end_0 = const()[name = tensor<string, []>("op_13486_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_13486_end_mask_0 = const()[name = tensor<string, []>("op_13486_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13486_cast = slice_by_index(begin = var_13486_begin_0, end = var_13486_end_0, end_mask = var_13486_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13486_cast")];
            tensor<int32, [4]> var_13490_begin_0 = const()[name = tensor<string, []>("op_13490_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_13490_end_0 = const()[name = tensor<string, []>("op_13490_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_13490_end_mask_0 = const()[name = tensor<string, []>("op_13490_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13490_cast = slice_by_index(begin = var_13490_begin_0, end = var_13490_end_0, end_mask = var_13490_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13490_cast")];
            tensor<int32, [4]> var_13494_begin_0 = const()[name = tensor<string, []>("op_13494_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_13494_end_0 = const()[name = tensor<string, []>("op_13494_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_13494_end_mask_0 = const()[name = tensor<string, []>("op_13494_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13494_cast = slice_by_index(begin = var_13494_begin_0, end = var_13494_end_0, end_mask = var_13494_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13494_cast")];
            tensor<int32, [4]> var_13498_begin_0 = const()[name = tensor<string, []>("op_13498_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_13498_end_0 = const()[name = tensor<string, []>("op_13498_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_13498_end_mask_0 = const()[name = tensor<string, []>("op_13498_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13498_cast = slice_by_index(begin = var_13498_begin_0, end = var_13498_end_0, end_mask = var_13498_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13498_cast")];
            tensor<int32, [4]> var_13502_begin_0 = const()[name = tensor<string, []>("op_13502_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_13502_end_0 = const()[name = tensor<string, []>("op_13502_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_13502_end_mask_0 = const()[name = tensor<string, []>("op_13502_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13502_cast = slice_by_index(begin = var_13502_begin_0, end = var_13502_end_0, end_mask = var_13502_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13502_cast")];
            tensor<int32, [4]> var_13506_begin_0 = const()[name = tensor<string, []>("op_13506_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_13506_end_0 = const()[name = tensor<string, []>("op_13506_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_13506_end_mask_0 = const()[name = tensor<string, []>("op_13506_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13506_cast = slice_by_index(begin = var_13506_begin_0, end = var_13506_end_0, end_mask = var_13506_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13506_cast")];
            tensor<int32, [4]> var_13510_begin_0 = const()[name = tensor<string, []>("op_13510_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_13510_end_0 = const()[name = tensor<string, []>("op_13510_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_13510_end_mask_0 = const()[name = tensor<string, []>("op_13510_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13510_cast = slice_by_index(begin = var_13510_begin_0, end = var_13510_end_0, end_mask = var_13510_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13510_cast")];
            tensor<int32, [4]> var_13514_begin_0 = const()[name = tensor<string, []>("op_13514_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_13514_end_0 = const()[name = tensor<string, []>("op_13514_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_13514_end_mask_0 = const()[name = tensor<string, []>("op_13514_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13514_cast = slice_by_index(begin = var_13514_begin_0, end = var_13514_end_0, end_mask = var_13514_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13514_cast")];
            tensor<int32, [4]> var_13518_begin_0 = const()[name = tensor<string, []>("op_13518_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_13518_end_0 = const()[name = tensor<string, []>("op_13518_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_13518_end_mask_0 = const()[name = tensor<string, []>("op_13518_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13518_cast = slice_by_index(begin = var_13518_begin_0, end = var_13518_end_0, end_mask = var_13518_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13518_cast")];
            tensor<int32, [4]> var_13522_begin_0 = const()[name = tensor<string, []>("op_13522_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_13522_end_0 = const()[name = tensor<string, []>("op_13522_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_13522_end_mask_0 = const()[name = tensor<string, []>("op_13522_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13522_cast = slice_by_index(begin = var_13522_begin_0, end = var_13522_end_0, end_mask = var_13522_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13522_cast")];
            tensor<int32, [4]> var_13526_begin_0 = const()[name = tensor<string, []>("op_13526_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_13526_end_0 = const()[name = tensor<string, []>("op_13526_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_13526_end_mask_0 = const()[name = tensor<string, []>("op_13526_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13526_cast = slice_by_index(begin = var_13526_begin_0, end = var_13526_end_0, end_mask = var_13526_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13526_cast")];
            tensor<int32, [4]> var_13530_begin_0 = const()[name = tensor<string, []>("op_13530_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_13530_end_0 = const()[name = tensor<string, []>("op_13530_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_13530_end_mask_0 = const()[name = tensor<string, []>("op_13530_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13530_cast = slice_by_index(begin = var_13530_begin_0, end = var_13530_end_0, end_mask = var_13530_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13530_cast")];
            tensor<int32, [4]> var_13534_begin_0 = const()[name = tensor<string, []>("op_13534_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_13534_end_0 = const()[name = tensor<string, []>("op_13534_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_13534_end_mask_0 = const()[name = tensor<string, []>("op_13534_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13534_cast = slice_by_index(begin = var_13534_begin_0, end = var_13534_end_0, end_mask = var_13534_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13534_cast")];
            tensor<int32, [4]> var_13538_begin_0 = const()[name = tensor<string, []>("op_13538_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_13538_end_0 = const()[name = tensor<string, []>("op_13538_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_13538_end_mask_0 = const()[name = tensor<string, []>("op_13538_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13538_cast = slice_by_index(begin = var_13538_begin_0, end = var_13538_end_0, end_mask = var_13538_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13538_cast")];
            tensor<int32, [4]> var_13542_begin_0 = const()[name = tensor<string, []>("op_13542_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_13542_end_0 = const()[name = tensor<string, []>("op_13542_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_13542_end_mask_0 = const()[name = tensor<string, []>("op_13542_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13542_cast = slice_by_index(begin = var_13542_begin_0, end = var_13542_end_0, end_mask = var_13542_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13542_cast")];
            tensor<int32, [4]> var_13546_begin_0 = const()[name = tensor<string, []>("op_13546_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_13546_end_0 = const()[name = tensor<string, []>("op_13546_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_13546_end_mask_0 = const()[name = tensor<string, []>("op_13546_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13546_cast = slice_by_index(begin = var_13546_begin_0, end = var_13546_end_0, end_mask = var_13546_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13546_cast")];
            tensor<int32, [4]> var_13550_begin_0 = const()[name = tensor<string, []>("op_13550_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_13550_end_0 = const()[name = tensor<string, []>("op_13550_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_13550_end_mask_0 = const()[name = tensor<string, []>("op_13550_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13550_cast = slice_by_index(begin = var_13550_begin_0, end = var_13550_end_0, end_mask = var_13550_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13550_cast")];
            tensor<int32, [4]> var_13554_begin_0 = const()[name = tensor<string, []>("op_13554_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_13554_end_0 = const()[name = tensor<string, []>("op_13554_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_13554_end_mask_0 = const()[name = tensor<string, []>("op_13554_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13554_cast = slice_by_index(begin = var_13554_begin_0, end = var_13554_end_0, end_mask = var_13554_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13554_cast")];
            tensor<int32, [4]> var_13558_begin_0 = const()[name = tensor<string, []>("op_13558_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_13558_end_0 = const()[name = tensor<string, []>("op_13558_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_13558_end_mask_0 = const()[name = tensor<string, []>("op_13558_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13558_cast = slice_by_index(begin = var_13558_begin_0, end = var_13558_end_0, end_mask = var_13558_end_mask_0, x = v_63_cast)[name = tensor<string, []>("op_13558_cast")];
            tensor<string, []> var_13562_equation_0 = const()[name = tensor<string, []>("op_13562_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13562_cast = einsum(equation = var_13562_equation_0, values = (var_13404_cast, var_13321_cast))[name = tensor<string, []>("op_13562_cast")];
            tensor<fp16, []> var_13563_to_fp16 = const()[name = tensor<string, []>("op_13563_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1081_cast = mul(x = var_13562_cast, y = var_13563_to_fp16)[name = tensor<string, []>("aw_1081_cast")];
            tensor<string, []> var_13566_equation_0 = const()[name = tensor<string, []>("op_13566_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13566_cast = einsum(equation = var_13566_equation_0, values = (var_13408_cast, var_13325_cast))[name = tensor<string, []>("op_13566_cast")];
            tensor<fp16, []> var_13567_to_fp16 = const()[name = tensor<string, []>("op_13567_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1083_cast = mul(x = var_13566_cast, y = var_13567_to_fp16)[name = tensor<string, []>("aw_1083_cast")];
            tensor<string, []> var_13570_equation_0 = const()[name = tensor<string, []>("op_13570_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13570_cast = einsum(equation = var_13570_equation_0, values = (var_13412_cast, var_13329_cast))[name = tensor<string, []>("op_13570_cast")];
            tensor<fp16, []> var_13571_to_fp16 = const()[name = tensor<string, []>("op_13571_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1085_cast = mul(x = var_13570_cast, y = var_13571_to_fp16)[name = tensor<string, []>("aw_1085_cast")];
            tensor<string, []> var_13574_equation_0 = const()[name = tensor<string, []>("op_13574_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13574_cast = einsum(equation = var_13574_equation_0, values = (var_13416_cast, var_13333_cast))[name = tensor<string, []>("op_13574_cast")];
            tensor<fp16, []> var_13575_to_fp16 = const()[name = tensor<string, []>("op_13575_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1087_cast = mul(x = var_13574_cast, y = var_13575_to_fp16)[name = tensor<string, []>("aw_1087_cast")];
            tensor<string, []> var_13578_equation_0 = const()[name = tensor<string, []>("op_13578_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13578_cast = einsum(equation = var_13578_equation_0, values = (var_13420_cast, var_13337_cast))[name = tensor<string, []>("op_13578_cast")];
            tensor<fp16, []> var_13579_to_fp16 = const()[name = tensor<string, []>("op_13579_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1089_cast = mul(x = var_13578_cast, y = var_13579_to_fp16)[name = tensor<string, []>("aw_1089_cast")];
            tensor<string, []> var_13582_equation_0 = const()[name = tensor<string, []>("op_13582_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13582_cast = einsum(equation = var_13582_equation_0, values = (var_13424_cast, var_13341_cast))[name = tensor<string, []>("op_13582_cast")];
            tensor<fp16, []> var_13583_to_fp16 = const()[name = tensor<string, []>("op_13583_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1091_cast = mul(x = var_13582_cast, y = var_13583_to_fp16)[name = tensor<string, []>("aw_1091_cast")];
            tensor<string, []> var_13586_equation_0 = const()[name = tensor<string, []>("op_13586_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13586_cast = einsum(equation = var_13586_equation_0, values = (var_13428_cast, var_13345_cast))[name = tensor<string, []>("op_13586_cast")];
            tensor<fp16, []> var_13587_to_fp16 = const()[name = tensor<string, []>("op_13587_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1093_cast = mul(x = var_13586_cast, y = var_13587_to_fp16)[name = tensor<string, []>("aw_1093_cast")];
            tensor<string, []> var_13590_equation_0 = const()[name = tensor<string, []>("op_13590_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13590_cast = einsum(equation = var_13590_equation_0, values = (var_13432_cast, var_13349_cast))[name = tensor<string, []>("op_13590_cast")];
            tensor<fp16, []> var_13591_to_fp16 = const()[name = tensor<string, []>("op_13591_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1095_cast = mul(x = var_13590_cast, y = var_13591_to_fp16)[name = tensor<string, []>("aw_1095_cast")];
            tensor<string, []> var_13594_equation_0 = const()[name = tensor<string, []>("op_13594_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13594_cast = einsum(equation = var_13594_equation_0, values = (var_13436_cast, var_13353_cast))[name = tensor<string, []>("op_13594_cast")];
            tensor<fp16, []> var_13595_to_fp16 = const()[name = tensor<string, []>("op_13595_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1097_cast = mul(x = var_13594_cast, y = var_13595_to_fp16)[name = tensor<string, []>("aw_1097_cast")];
            tensor<string, []> var_13598_equation_0 = const()[name = tensor<string, []>("op_13598_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13598_cast = einsum(equation = var_13598_equation_0, values = (var_13440_cast, var_13357_cast))[name = tensor<string, []>("op_13598_cast")];
            tensor<fp16, []> var_13599_to_fp16 = const()[name = tensor<string, []>("op_13599_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1099_cast = mul(x = var_13598_cast, y = var_13599_to_fp16)[name = tensor<string, []>("aw_1099_cast")];
            tensor<string, []> var_13602_equation_0 = const()[name = tensor<string, []>("op_13602_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13602_cast = einsum(equation = var_13602_equation_0, values = (var_13444_cast, var_13361_cast))[name = tensor<string, []>("op_13602_cast")];
            tensor<fp16, []> var_13603_to_fp16 = const()[name = tensor<string, []>("op_13603_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1101_cast = mul(x = var_13602_cast, y = var_13603_to_fp16)[name = tensor<string, []>("aw_1101_cast")];
            tensor<string, []> var_13606_equation_0 = const()[name = tensor<string, []>("op_13606_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13606_cast = einsum(equation = var_13606_equation_0, values = (var_13448_cast, var_13365_cast))[name = tensor<string, []>("op_13606_cast")];
            tensor<fp16, []> var_13607_to_fp16 = const()[name = tensor<string, []>("op_13607_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1103_cast = mul(x = var_13606_cast, y = var_13607_to_fp16)[name = tensor<string, []>("aw_1103_cast")];
            tensor<string, []> var_13610_equation_0 = const()[name = tensor<string, []>("op_13610_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13610_cast = einsum(equation = var_13610_equation_0, values = (var_13452_cast, var_13369_cast))[name = tensor<string, []>("op_13610_cast")];
            tensor<fp16, []> var_13611_to_fp16 = const()[name = tensor<string, []>("op_13611_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1105_cast = mul(x = var_13610_cast, y = var_13611_to_fp16)[name = tensor<string, []>("aw_1105_cast")];
            tensor<string, []> var_13614_equation_0 = const()[name = tensor<string, []>("op_13614_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13614_cast = einsum(equation = var_13614_equation_0, values = (var_13456_cast, var_13373_cast))[name = tensor<string, []>("op_13614_cast")];
            tensor<fp16, []> var_13615_to_fp16 = const()[name = tensor<string, []>("op_13615_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1107_cast = mul(x = var_13614_cast, y = var_13615_to_fp16)[name = tensor<string, []>("aw_1107_cast")];
            tensor<string, []> var_13618_equation_0 = const()[name = tensor<string, []>("op_13618_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13618_cast = einsum(equation = var_13618_equation_0, values = (var_13460_cast, var_13377_cast))[name = tensor<string, []>("op_13618_cast")];
            tensor<fp16, []> var_13619_to_fp16 = const()[name = tensor<string, []>("op_13619_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1109_cast = mul(x = var_13618_cast, y = var_13619_to_fp16)[name = tensor<string, []>("aw_1109_cast")];
            tensor<string, []> var_13622_equation_0 = const()[name = tensor<string, []>("op_13622_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13622_cast = einsum(equation = var_13622_equation_0, values = (var_13464_cast, var_13381_cast))[name = tensor<string, []>("op_13622_cast")];
            tensor<fp16, []> var_13623_to_fp16 = const()[name = tensor<string, []>("op_13623_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1111_cast = mul(x = var_13622_cast, y = var_13623_to_fp16)[name = tensor<string, []>("aw_1111_cast")];
            tensor<string, []> var_13626_equation_0 = const()[name = tensor<string, []>("op_13626_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13626_cast = einsum(equation = var_13626_equation_0, values = (var_13468_cast, var_13385_cast))[name = tensor<string, []>("op_13626_cast")];
            tensor<fp16, []> var_13627_to_fp16 = const()[name = tensor<string, []>("op_13627_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1113_cast = mul(x = var_13626_cast, y = var_13627_to_fp16)[name = tensor<string, []>("aw_1113_cast")];
            tensor<string, []> var_13630_equation_0 = const()[name = tensor<string, []>("op_13630_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13630_cast = einsum(equation = var_13630_equation_0, values = (var_13472_cast, var_13389_cast))[name = tensor<string, []>("op_13630_cast")];
            tensor<fp16, []> var_13631_to_fp16 = const()[name = tensor<string, []>("op_13631_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1115_cast = mul(x = var_13630_cast, y = var_13631_to_fp16)[name = tensor<string, []>("aw_1115_cast")];
            tensor<string, []> var_13634_equation_0 = const()[name = tensor<string, []>("op_13634_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13634_cast = einsum(equation = var_13634_equation_0, values = (var_13476_cast, var_13393_cast))[name = tensor<string, []>("op_13634_cast")];
            tensor<fp16, []> var_13635_to_fp16 = const()[name = tensor<string, []>("op_13635_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1117_cast = mul(x = var_13634_cast, y = var_13635_to_fp16)[name = tensor<string, []>("aw_1117_cast")];
            tensor<string, []> var_13638_equation_0 = const()[name = tensor<string, []>("op_13638_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_13638_cast = einsum(equation = var_13638_equation_0, values = (var_13480_cast, var_13397_cast))[name = tensor<string, []>("op_13638_cast")];
            tensor<fp16, []> var_13639_to_fp16 = const()[name = tensor<string, []>("op_13639_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1119_cast = mul(x = var_13638_cast, y = var_13639_to_fp16)[name = tensor<string, []>("aw_1119_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13641_cast = softmax(axis = var_2634, x = aw_1081_cast)[name = tensor<string, []>("op_13641_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13642_cast = softmax(axis = var_2634, x = aw_1083_cast)[name = tensor<string, []>("op_13642_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13643_cast = softmax(axis = var_2634, x = aw_1085_cast)[name = tensor<string, []>("op_13643_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13644_cast = softmax(axis = var_2634, x = aw_1087_cast)[name = tensor<string, []>("op_13644_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13645_cast = softmax(axis = var_2634, x = aw_1089_cast)[name = tensor<string, []>("op_13645_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13646_cast = softmax(axis = var_2634, x = aw_1091_cast)[name = tensor<string, []>("op_13646_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13647_cast = softmax(axis = var_2634, x = aw_1093_cast)[name = tensor<string, []>("op_13647_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13648_cast = softmax(axis = var_2634, x = aw_1095_cast)[name = tensor<string, []>("op_13648_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13649_cast = softmax(axis = var_2634, x = aw_1097_cast)[name = tensor<string, []>("op_13649_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13650_cast = softmax(axis = var_2634, x = aw_1099_cast)[name = tensor<string, []>("op_13650_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13651_cast = softmax(axis = var_2634, x = aw_1101_cast)[name = tensor<string, []>("op_13651_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13652_cast = softmax(axis = var_2634, x = aw_1103_cast)[name = tensor<string, []>("op_13652_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13653_cast = softmax(axis = var_2634, x = aw_1105_cast)[name = tensor<string, []>("op_13653_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13654_cast = softmax(axis = var_2634, x = aw_1107_cast)[name = tensor<string, []>("op_13654_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13655_cast = softmax(axis = var_2634, x = aw_1109_cast)[name = tensor<string, []>("op_13655_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13656_cast = softmax(axis = var_2634, x = aw_1111_cast)[name = tensor<string, []>("op_13656_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13657_cast = softmax(axis = var_2634, x = aw_1113_cast)[name = tensor<string, []>("op_13657_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13658_cast = softmax(axis = var_2634, x = aw_1115_cast)[name = tensor<string, []>("op_13658_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13659_cast = softmax(axis = var_2634, x = aw_1117_cast)[name = tensor<string, []>("op_13659_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_13660_cast = softmax(axis = var_2634, x = aw_1119_cast)[name = tensor<string, []>("op_13660_cast")];
            tensor<string, []> var_13662_equation_0 = const()[name = tensor<string, []>("op_13662_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13662_cast = einsum(equation = var_13662_equation_0, values = (var_13482_cast, var_13641_cast))[name = tensor<string, []>("op_13662_cast")];
            tensor<string, []> var_13664_equation_0 = const()[name = tensor<string, []>("op_13664_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13664_cast = einsum(equation = var_13664_equation_0, values = (var_13486_cast, var_13642_cast))[name = tensor<string, []>("op_13664_cast")];
            tensor<string, []> var_13666_equation_0 = const()[name = tensor<string, []>("op_13666_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13666_cast = einsum(equation = var_13666_equation_0, values = (var_13490_cast, var_13643_cast))[name = tensor<string, []>("op_13666_cast")];
            tensor<string, []> var_13668_equation_0 = const()[name = tensor<string, []>("op_13668_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13668_cast = einsum(equation = var_13668_equation_0, values = (var_13494_cast, var_13644_cast))[name = tensor<string, []>("op_13668_cast")];
            tensor<string, []> var_13670_equation_0 = const()[name = tensor<string, []>("op_13670_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13670_cast = einsum(equation = var_13670_equation_0, values = (var_13498_cast, var_13645_cast))[name = tensor<string, []>("op_13670_cast")];
            tensor<string, []> var_13672_equation_0 = const()[name = tensor<string, []>("op_13672_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13672_cast = einsum(equation = var_13672_equation_0, values = (var_13502_cast, var_13646_cast))[name = tensor<string, []>("op_13672_cast")];
            tensor<string, []> var_13674_equation_0 = const()[name = tensor<string, []>("op_13674_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13674_cast = einsum(equation = var_13674_equation_0, values = (var_13506_cast, var_13647_cast))[name = tensor<string, []>("op_13674_cast")];
            tensor<string, []> var_13676_equation_0 = const()[name = tensor<string, []>("op_13676_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13676_cast = einsum(equation = var_13676_equation_0, values = (var_13510_cast, var_13648_cast))[name = tensor<string, []>("op_13676_cast")];
            tensor<string, []> var_13678_equation_0 = const()[name = tensor<string, []>("op_13678_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13678_cast = einsum(equation = var_13678_equation_0, values = (var_13514_cast, var_13649_cast))[name = tensor<string, []>("op_13678_cast")];
            tensor<string, []> var_13680_equation_0 = const()[name = tensor<string, []>("op_13680_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13680_cast = einsum(equation = var_13680_equation_0, values = (var_13518_cast, var_13650_cast))[name = tensor<string, []>("op_13680_cast")];
            tensor<string, []> var_13682_equation_0 = const()[name = tensor<string, []>("op_13682_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13682_cast = einsum(equation = var_13682_equation_0, values = (var_13522_cast, var_13651_cast))[name = tensor<string, []>("op_13682_cast")];
            tensor<string, []> var_13684_equation_0 = const()[name = tensor<string, []>("op_13684_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13684_cast = einsum(equation = var_13684_equation_0, values = (var_13526_cast, var_13652_cast))[name = tensor<string, []>("op_13684_cast")];
            tensor<string, []> var_13686_equation_0 = const()[name = tensor<string, []>("op_13686_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13686_cast = einsum(equation = var_13686_equation_0, values = (var_13530_cast, var_13653_cast))[name = tensor<string, []>("op_13686_cast")];
            tensor<string, []> var_13688_equation_0 = const()[name = tensor<string, []>("op_13688_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13688_cast = einsum(equation = var_13688_equation_0, values = (var_13534_cast, var_13654_cast))[name = tensor<string, []>("op_13688_cast")];
            tensor<string, []> var_13690_equation_0 = const()[name = tensor<string, []>("op_13690_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13690_cast = einsum(equation = var_13690_equation_0, values = (var_13538_cast, var_13655_cast))[name = tensor<string, []>("op_13690_cast")];
            tensor<string, []> var_13692_equation_0 = const()[name = tensor<string, []>("op_13692_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13692_cast = einsum(equation = var_13692_equation_0, values = (var_13542_cast, var_13656_cast))[name = tensor<string, []>("op_13692_cast")];
            tensor<string, []> var_13694_equation_0 = const()[name = tensor<string, []>("op_13694_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13694_cast = einsum(equation = var_13694_equation_0, values = (var_13546_cast, var_13657_cast))[name = tensor<string, []>("op_13694_cast")];
            tensor<string, []> var_13696_equation_0 = const()[name = tensor<string, []>("op_13696_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13696_cast = einsum(equation = var_13696_equation_0, values = (var_13550_cast, var_13658_cast))[name = tensor<string, []>("op_13696_cast")];
            tensor<string, []> var_13698_equation_0 = const()[name = tensor<string, []>("op_13698_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13698_cast = einsum(equation = var_13698_equation_0, values = (var_13554_cast, var_13659_cast))[name = tensor<string, []>("op_13698_cast")];
            tensor<string, []> var_13700_equation_0 = const()[name = tensor<string, []>("op_13700_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_13700_cast = einsum(equation = var_13700_equation_0, values = (var_13558_cast, var_13660_cast))[name = tensor<string, []>("op_13700_cast")];
            tensor<bool, []> input_243_interleave_0 = const()[name = tensor<string, []>("input_243_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_243_cast = concat(axis = var_2634, interleave = input_243_interleave_0, values = (var_13662_cast, var_13664_cast, var_13666_cast, var_13668_cast, var_13670_cast, var_13672_cast, var_13674_cast, var_13676_cast, var_13678_cast, var_13680_cast, var_13682_cast, var_13684_cast, var_13686_cast, var_13688_cast, var_13690_cast, var_13692_cast, var_13694_cast, var_13696_cast, var_13698_cast, var_13700_cast))[name = tensor<string, []>("input_243_cast")];
            tensor<int32, [2]> var_13706 = const()[name = tensor<string, []>("op_13706"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13708 = const()[name = tensor<string, []>("op_13708"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_13710_pad_type_0 = const()[name = tensor<string, []>("op_13710_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_13710_pad_0 = const()[name = tensor<string, []>("op_13710_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1058533760)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1061810624)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_13710_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_13708, groups = var_2634, pad = var_13710_pad_0, pad_type = var_13710_pad_type_0, strides = var_13706, weight = down_blocks_2_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_243_cast)[name = tensor<string, []>("op_13710_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_95_cast = add(x = var_13710_cast, y = inputs_93_cast)[name = tensor<string, []>("inputs_95_cast")];
            tensor<int32, [1]> var_13714 = const()[name = tensor<string, []>("op_13714"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_95_cast = reduce_mean(axes = var_13714, keep_dims = var_2629, x = inputs_95_cast)[name = tensor<string, []>("channels_mean_95_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_95_cast = sub(x = inputs_95_cast, y = channels_mean_95_cast)[name = tensor<string, []>("zero_mean_95_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_95_cast = mul(x = zero_mean_95_cast, y = zero_mean_95_cast)[name = tensor<string, []>("zero_mean_sq_95_cast")];
            tensor<int32, [1]> var_13718 = const()[name = tensor<string, []>("op_13718"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_13719_cast = reduce_mean(axes = var_13718, keep_dims = var_2629, x = zero_mean_sq_95_cast)[name = tensor<string, []>("op_13719_cast")];
            tensor<fp16, []> var_13720_to_fp16 = const()[name = tensor<string, []>("op_13720_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_13721_cast = add(x = var_13719_cast, y = var_13720_to_fp16)[name = tensor<string, []>("op_13721_cast")];
            tensor<fp16, []> denom_95_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_95_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_95_cast = rsqrt(epsilon = denom_95_epsilon_0_to_fp16, x = var_13721_cast)[name = tensor<string, []>("denom_95_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_95_cast = mul(x = zero_mean_95_cast, y = denom_95_cast)[name = tensor<string, []>("out_95_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_13725_to_fp16 = const()[name = tensor<string, []>("op_13725_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1061813248)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_13726_cast = add(x = out_95_cast, y = var_13725_to_fp16)[name = tensor<string, []>("op_13726_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_13728_to_fp16 = const()[name = tensor<string, []>("op_13728_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1061815872)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_245_cast = mul(x = var_13726_cast, y = var_13728_to_fp16)[name = tensor<string, []>("input_245_cast")];
            tensor<int32, [2]> var_13736 = const()[name = tensor<string, []>("op_13736"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13738 = const()[name = tensor<string, []>("op_13738"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_13740_pad_type_0 = const()[name = tensor<string, []>("op_13740_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_13740_pad_0 = const()[name = tensor<string, []>("op_13740_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1061818496)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1088032960)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_13740_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_13738, groups = var_2634, pad = var_13740_pad_0, pad_type = var_13740_pad_type_0, strides = var_13736, weight = down_blocks_2_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_245_cast)[name = tensor<string, []>("op_13740_cast")];
            tensor<int32, [2]> var_13741_split_sizes_0 = const()[name = tensor<string, []>("op_13741_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_13741_axis_0 = const()[name = tensor<string, []>("op_13741_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_13741_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_13741_cast_1 = split(axis = var_13741_axis_0, split_sizes = var_13741_split_sizes_0, x = var_13740_cast)[name = tensor<string, []>("op_13741_cast")];
            tensor<string, []> var_13743_mode_0 = const()[name = tensor<string, []>("op_13743_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_13743_cast = gelu(mode = var_13743_mode_0, x = var_13741_cast_1)[name = tensor<string, []>("op_13743_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_247_cast = mul(x = var_13741_cast_0, y = var_13743_cast)[name = tensor<string, []>("input_247_cast")];
            tensor<int32, [2]> var_13747 = const()[name = tensor<string, []>("op_13747"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13749 = const()[name = tensor<string, []>("op_13749"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_13751_pad_type_0 = const()[name = tensor<string, []>("op_13751_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_13751_pad_0 = const()[name = tensor<string, []>("op_13751_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1088053504)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1101160768)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_13751_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_13749, groups = var_2634, pad = var_13751_pad_0, pad_type = var_13751_pad_type_0, strides = var_13747, weight = down_blocks_2_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_247_cast)[name = tensor<string, []>("op_13751_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_97_cast = add(x = var_13751_cast, y = inputs_95_cast)[name = tensor<string, []>("inputs_97_cast")];
            tensor<int32, [1]> var_13761 = const()[name = tensor<string, []>("op_13761"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_97_cast = reduce_mean(axes = var_13761, keep_dims = var_2629, x = inputs_97_cast)[name = tensor<string, []>("channels_mean_97_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_97_cast = sub(x = inputs_97_cast, y = channels_mean_97_cast)[name = tensor<string, []>("zero_mean_97_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_97_cast = mul(x = zero_mean_97_cast, y = zero_mean_97_cast)[name = tensor<string, []>("zero_mean_sq_97_cast")];
            tensor<int32, [1]> var_13765 = const()[name = tensor<string, []>("op_13765"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_13766_cast = reduce_mean(axes = var_13765, keep_dims = var_2629, x = zero_mean_sq_97_cast)[name = tensor<string, []>("op_13766_cast")];
            tensor<fp16, []> var_13767_to_fp16 = const()[name = tensor<string, []>("op_13767_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_13768_cast = add(x = var_13766_cast, y = var_13767_to_fp16)[name = tensor<string, []>("op_13768_cast")];
            tensor<fp16, []> denom_97_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_97_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_97_cast = rsqrt(epsilon = denom_97_epsilon_0_to_fp16, x = var_13768_cast)[name = tensor<string, []>("denom_97_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_97_cast = mul(x = zero_mean_97_cast, y = denom_97_cast)[name = tensor<string, []>("out_97_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_13772_to_fp16 = const()[name = tensor<string, []>("op_13772_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1101163392)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_13773_cast = add(x = out_97_cast, y = var_13772_to_fp16)[name = tensor<string, []>("op_13773_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_13775_to_fp16 = const()[name = tensor<string, []>("op_13775_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1101166016)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_149_cast = mul(x = var_13773_cast, y = var_13775_to_fp16)[name = tensor<string, []>("hidden_states_149_cast")];
            tensor<int32, [2]> var_13782 = const()[name = tensor<string, []>("op_13782"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13784 = const()[name = tensor<string, []>("op_13784"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_65_pad_type_0 = const()[name = tensor<string, []>("q_65_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_65_pad_0 = const()[name = tensor<string, []>("q_65_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1101168640)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_65_cast = conv(dilations = var_13784, groups = var_2634, pad = q_65_pad_0, pad_type = q_65_pad_type_0, strides = var_13782, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_q_weight_to_fp16, x = hidden_states_149_cast)[name = tensor<string, []>("q_65_cast")];
            tensor<int32, [2]> var_13788 = const()[name = tensor<string, []>("op_13788"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13790 = const()[name = tensor<string, []>("op_13790"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_129_pad_type_0 = const()[name = tensor<string, []>("k_129_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_129_pad_0 = const()[name = tensor<string, []>("k_129_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1104445504)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_129_cast = conv(dilations = var_13790, groups = var_2634, pad = k_129_pad_0, pad_type = k_129_pad_type_0, strides = var_13788, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_k_weight_to_fp16, x = hidden_states_149_cast)[name = tensor<string, []>("k_129_cast")];
            tensor<int32, [2]> var_13794 = const()[name = tensor<string, []>("op_13794"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13796 = const()[name = tensor<string, []>("op_13796"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_65_pad_type_0 = const()[name = tensor<string, []>("v_65_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_65_pad_0 = const()[name = tensor<string, []>("v_65_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1107722368)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_65_cast = conv(dilations = var_13796, groups = var_2634, pad = v_65_pad_0, pad_type = v_65_pad_type_0, strides = var_13794, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_v_weight_to_fp16, x = hidden_states_149_cast)[name = tensor<string, []>("v_65_cast")];
            tensor<int32, [4]> var_13800_begin_0 = const()[name = tensor<string, []>("op_13800_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13800_end_0 = const()[name = tensor<string, []>("op_13800_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_13800_end_mask_0 = const()[name = tensor<string, []>("op_13800_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13800_cast = slice_by_index(begin = var_13800_begin_0, end = var_13800_end_0, end_mask = var_13800_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13800_cast")];
            tensor<int32, [4]> var_13804_begin_0 = const()[name = tensor<string, []>("op_13804_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_13804_end_0 = const()[name = tensor<string, []>("op_13804_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_13804_end_mask_0 = const()[name = tensor<string, []>("op_13804_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13804_cast = slice_by_index(begin = var_13804_begin_0, end = var_13804_end_0, end_mask = var_13804_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13804_cast")];
            tensor<int32, [4]> var_13808_begin_0 = const()[name = tensor<string, []>("op_13808_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_13808_end_0 = const()[name = tensor<string, []>("op_13808_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_13808_end_mask_0 = const()[name = tensor<string, []>("op_13808_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13808_cast = slice_by_index(begin = var_13808_begin_0, end = var_13808_end_0, end_mask = var_13808_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13808_cast")];
            tensor<int32, [4]> var_13812_begin_0 = const()[name = tensor<string, []>("op_13812_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_13812_end_0 = const()[name = tensor<string, []>("op_13812_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_13812_end_mask_0 = const()[name = tensor<string, []>("op_13812_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13812_cast = slice_by_index(begin = var_13812_begin_0, end = var_13812_end_0, end_mask = var_13812_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13812_cast")];
            tensor<int32, [4]> var_13816_begin_0 = const()[name = tensor<string, []>("op_13816_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_13816_end_0 = const()[name = tensor<string, []>("op_13816_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_13816_end_mask_0 = const()[name = tensor<string, []>("op_13816_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13816_cast = slice_by_index(begin = var_13816_begin_0, end = var_13816_end_0, end_mask = var_13816_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13816_cast")];
            tensor<int32, [4]> var_13820_begin_0 = const()[name = tensor<string, []>("op_13820_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_13820_end_0 = const()[name = tensor<string, []>("op_13820_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_13820_end_mask_0 = const()[name = tensor<string, []>("op_13820_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13820_cast = slice_by_index(begin = var_13820_begin_0, end = var_13820_end_0, end_mask = var_13820_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13820_cast")];
            tensor<int32, [4]> var_13824_begin_0 = const()[name = tensor<string, []>("op_13824_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_13824_end_0 = const()[name = tensor<string, []>("op_13824_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_13824_end_mask_0 = const()[name = tensor<string, []>("op_13824_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13824_cast = slice_by_index(begin = var_13824_begin_0, end = var_13824_end_0, end_mask = var_13824_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13824_cast")];
            tensor<int32, [4]> var_13828_begin_0 = const()[name = tensor<string, []>("op_13828_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_13828_end_0 = const()[name = tensor<string, []>("op_13828_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_13828_end_mask_0 = const()[name = tensor<string, []>("op_13828_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13828_cast = slice_by_index(begin = var_13828_begin_0, end = var_13828_end_0, end_mask = var_13828_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13828_cast")];
            tensor<int32, [4]> var_13832_begin_0 = const()[name = tensor<string, []>("op_13832_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_13832_end_0 = const()[name = tensor<string, []>("op_13832_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_13832_end_mask_0 = const()[name = tensor<string, []>("op_13832_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13832_cast = slice_by_index(begin = var_13832_begin_0, end = var_13832_end_0, end_mask = var_13832_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13832_cast")];
            tensor<int32, [4]> var_13836_begin_0 = const()[name = tensor<string, []>("op_13836_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_13836_end_0 = const()[name = tensor<string, []>("op_13836_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_13836_end_mask_0 = const()[name = tensor<string, []>("op_13836_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13836_cast = slice_by_index(begin = var_13836_begin_0, end = var_13836_end_0, end_mask = var_13836_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13836_cast")];
            tensor<int32, [4]> var_13840_begin_0 = const()[name = tensor<string, []>("op_13840_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_13840_end_0 = const()[name = tensor<string, []>("op_13840_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_13840_end_mask_0 = const()[name = tensor<string, []>("op_13840_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13840_cast = slice_by_index(begin = var_13840_begin_0, end = var_13840_end_0, end_mask = var_13840_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13840_cast")];
            tensor<int32, [4]> var_13844_begin_0 = const()[name = tensor<string, []>("op_13844_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_13844_end_0 = const()[name = tensor<string, []>("op_13844_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_13844_end_mask_0 = const()[name = tensor<string, []>("op_13844_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13844_cast = slice_by_index(begin = var_13844_begin_0, end = var_13844_end_0, end_mask = var_13844_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13844_cast")];
            tensor<int32, [4]> var_13848_begin_0 = const()[name = tensor<string, []>("op_13848_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_13848_end_0 = const()[name = tensor<string, []>("op_13848_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_13848_end_mask_0 = const()[name = tensor<string, []>("op_13848_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13848_cast = slice_by_index(begin = var_13848_begin_0, end = var_13848_end_0, end_mask = var_13848_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13848_cast")];
            tensor<int32, [4]> var_13852_begin_0 = const()[name = tensor<string, []>("op_13852_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_13852_end_0 = const()[name = tensor<string, []>("op_13852_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_13852_end_mask_0 = const()[name = tensor<string, []>("op_13852_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13852_cast = slice_by_index(begin = var_13852_begin_0, end = var_13852_end_0, end_mask = var_13852_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13852_cast")];
            tensor<int32, [4]> var_13856_begin_0 = const()[name = tensor<string, []>("op_13856_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_13856_end_0 = const()[name = tensor<string, []>("op_13856_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_13856_end_mask_0 = const()[name = tensor<string, []>("op_13856_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13856_cast = slice_by_index(begin = var_13856_begin_0, end = var_13856_end_0, end_mask = var_13856_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13856_cast")];
            tensor<int32, [4]> var_13860_begin_0 = const()[name = tensor<string, []>("op_13860_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_13860_end_0 = const()[name = tensor<string, []>("op_13860_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_13860_end_mask_0 = const()[name = tensor<string, []>("op_13860_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13860_cast = slice_by_index(begin = var_13860_begin_0, end = var_13860_end_0, end_mask = var_13860_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13860_cast")];
            tensor<int32, [4]> var_13864_begin_0 = const()[name = tensor<string, []>("op_13864_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_13864_end_0 = const()[name = tensor<string, []>("op_13864_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_13864_end_mask_0 = const()[name = tensor<string, []>("op_13864_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13864_cast = slice_by_index(begin = var_13864_begin_0, end = var_13864_end_0, end_mask = var_13864_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13864_cast")];
            tensor<int32, [4]> var_13868_begin_0 = const()[name = tensor<string, []>("op_13868_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_13868_end_0 = const()[name = tensor<string, []>("op_13868_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_13868_end_mask_0 = const()[name = tensor<string, []>("op_13868_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13868_cast = slice_by_index(begin = var_13868_begin_0, end = var_13868_end_0, end_mask = var_13868_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13868_cast")];
            tensor<int32, [4]> var_13872_begin_0 = const()[name = tensor<string, []>("op_13872_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_13872_end_0 = const()[name = tensor<string, []>("op_13872_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_13872_end_mask_0 = const()[name = tensor<string, []>("op_13872_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13872_cast = slice_by_index(begin = var_13872_begin_0, end = var_13872_end_0, end_mask = var_13872_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13872_cast")];
            tensor<int32, [4]> var_13876_begin_0 = const()[name = tensor<string, []>("op_13876_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_13876_end_0 = const()[name = tensor<string, []>("op_13876_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_13876_end_mask_0 = const()[name = tensor<string, []>("op_13876_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13876_cast = slice_by_index(begin = var_13876_begin_0, end = var_13876_end_0, end_mask = var_13876_end_mask_0, x = q_65_cast)[name = tensor<string, []>("op_13876_cast")];
            tensor<int32, [4]> k_131_perm_0 = const()[name = tensor<string, []>("k_131_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_13883_begin_0 = const()[name = tensor<string, []>("op_13883_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13883_end_0 = const()[name = tensor<string, []>("op_13883_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_13883_end_mask_0 = const()[name = tensor<string, []>("op_13883_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_107 = transpose(perm = k_131_perm_0, x = k_129_cast)[name = tensor<string, []>("transpose_107")];
            tensor<fp16, [2, 1024, 1, 64]> var_13883_cast = slice_by_index(begin = var_13883_begin_0, end = var_13883_end_0, end_mask = var_13883_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13883_cast")];
            tensor<int32, [4]> var_13887_begin_0 = const()[name = tensor<string, []>("op_13887_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_13887_end_0 = const()[name = tensor<string, []>("op_13887_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_13887_end_mask_0 = const()[name = tensor<string, []>("op_13887_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13887_cast = slice_by_index(begin = var_13887_begin_0, end = var_13887_end_0, end_mask = var_13887_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13887_cast")];
            tensor<int32, [4]> var_13891_begin_0 = const()[name = tensor<string, []>("op_13891_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_13891_end_0 = const()[name = tensor<string, []>("op_13891_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_13891_end_mask_0 = const()[name = tensor<string, []>("op_13891_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13891_cast = slice_by_index(begin = var_13891_begin_0, end = var_13891_end_0, end_mask = var_13891_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13891_cast")];
            tensor<int32, [4]> var_13895_begin_0 = const()[name = tensor<string, []>("op_13895_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_13895_end_0 = const()[name = tensor<string, []>("op_13895_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_13895_end_mask_0 = const()[name = tensor<string, []>("op_13895_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13895_cast = slice_by_index(begin = var_13895_begin_0, end = var_13895_end_0, end_mask = var_13895_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13895_cast")];
            tensor<int32, [4]> var_13899_begin_0 = const()[name = tensor<string, []>("op_13899_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_13899_end_0 = const()[name = tensor<string, []>("op_13899_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_13899_end_mask_0 = const()[name = tensor<string, []>("op_13899_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13899_cast = slice_by_index(begin = var_13899_begin_0, end = var_13899_end_0, end_mask = var_13899_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13899_cast")];
            tensor<int32, [4]> var_13903_begin_0 = const()[name = tensor<string, []>("op_13903_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_13903_end_0 = const()[name = tensor<string, []>("op_13903_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_13903_end_mask_0 = const()[name = tensor<string, []>("op_13903_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13903_cast = slice_by_index(begin = var_13903_begin_0, end = var_13903_end_0, end_mask = var_13903_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13903_cast")];
            tensor<int32, [4]> var_13907_begin_0 = const()[name = tensor<string, []>("op_13907_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_13907_end_0 = const()[name = tensor<string, []>("op_13907_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_13907_end_mask_0 = const()[name = tensor<string, []>("op_13907_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13907_cast = slice_by_index(begin = var_13907_begin_0, end = var_13907_end_0, end_mask = var_13907_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13907_cast")];
            tensor<int32, [4]> var_13911_begin_0 = const()[name = tensor<string, []>("op_13911_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_13911_end_0 = const()[name = tensor<string, []>("op_13911_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_13911_end_mask_0 = const()[name = tensor<string, []>("op_13911_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13911_cast = slice_by_index(begin = var_13911_begin_0, end = var_13911_end_0, end_mask = var_13911_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13911_cast")];
            tensor<int32, [4]> var_13915_begin_0 = const()[name = tensor<string, []>("op_13915_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13915_end_0 = const()[name = tensor<string, []>("op_13915_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_13915_end_mask_0 = const()[name = tensor<string, []>("op_13915_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13915_cast = slice_by_index(begin = var_13915_begin_0, end = var_13915_end_0, end_mask = var_13915_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13915_cast")];
            tensor<int32, [4]> var_13919_begin_0 = const()[name = tensor<string, []>("op_13919_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_13919_end_0 = const()[name = tensor<string, []>("op_13919_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_13919_end_mask_0 = const()[name = tensor<string, []>("op_13919_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13919_cast = slice_by_index(begin = var_13919_begin_0, end = var_13919_end_0, end_mask = var_13919_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13919_cast")];
            tensor<int32, [4]> var_13923_begin_0 = const()[name = tensor<string, []>("op_13923_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_13923_end_0 = const()[name = tensor<string, []>("op_13923_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_13923_end_mask_0 = const()[name = tensor<string, []>("op_13923_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13923_cast = slice_by_index(begin = var_13923_begin_0, end = var_13923_end_0, end_mask = var_13923_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13923_cast")];
            tensor<int32, [4]> var_13927_begin_0 = const()[name = tensor<string, []>("op_13927_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_13927_end_0 = const()[name = tensor<string, []>("op_13927_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_13927_end_mask_0 = const()[name = tensor<string, []>("op_13927_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13927_cast = slice_by_index(begin = var_13927_begin_0, end = var_13927_end_0, end_mask = var_13927_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13927_cast")];
            tensor<int32, [4]> var_13931_begin_0 = const()[name = tensor<string, []>("op_13931_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_13931_end_0 = const()[name = tensor<string, []>("op_13931_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_13931_end_mask_0 = const()[name = tensor<string, []>("op_13931_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13931_cast = slice_by_index(begin = var_13931_begin_0, end = var_13931_end_0, end_mask = var_13931_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13931_cast")];
            tensor<int32, [4]> var_13935_begin_0 = const()[name = tensor<string, []>("op_13935_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_13935_end_0 = const()[name = tensor<string, []>("op_13935_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_13935_end_mask_0 = const()[name = tensor<string, []>("op_13935_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13935_cast = slice_by_index(begin = var_13935_begin_0, end = var_13935_end_0, end_mask = var_13935_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13935_cast")];
            tensor<int32, [4]> var_13939_begin_0 = const()[name = tensor<string, []>("op_13939_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_13939_end_0 = const()[name = tensor<string, []>("op_13939_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_13939_end_mask_0 = const()[name = tensor<string, []>("op_13939_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13939_cast = slice_by_index(begin = var_13939_begin_0, end = var_13939_end_0, end_mask = var_13939_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13939_cast")];
            tensor<int32, [4]> var_13943_begin_0 = const()[name = tensor<string, []>("op_13943_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_13943_end_0 = const()[name = tensor<string, []>("op_13943_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_13943_end_mask_0 = const()[name = tensor<string, []>("op_13943_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13943_cast = slice_by_index(begin = var_13943_begin_0, end = var_13943_end_0, end_mask = var_13943_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13943_cast")];
            tensor<int32, [4]> var_13947_begin_0 = const()[name = tensor<string, []>("op_13947_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13947_end_0 = const()[name = tensor<string, []>("op_13947_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_13947_end_mask_0 = const()[name = tensor<string, []>("op_13947_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13947_cast = slice_by_index(begin = var_13947_begin_0, end = var_13947_end_0, end_mask = var_13947_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13947_cast")];
            tensor<int32, [4]> var_13951_begin_0 = const()[name = tensor<string, []>("op_13951_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_13951_end_0 = const()[name = tensor<string, []>("op_13951_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_13951_end_mask_0 = const()[name = tensor<string, []>("op_13951_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13951_cast = slice_by_index(begin = var_13951_begin_0, end = var_13951_end_0, end_mask = var_13951_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13951_cast")];
            tensor<int32, [4]> var_13955_begin_0 = const()[name = tensor<string, []>("op_13955_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_13955_end_0 = const()[name = tensor<string, []>("op_13955_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_13955_end_mask_0 = const()[name = tensor<string, []>("op_13955_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13955_cast = slice_by_index(begin = var_13955_begin_0, end = var_13955_end_0, end_mask = var_13955_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13955_cast")];
            tensor<int32, [4]> var_13959_begin_0 = const()[name = tensor<string, []>("op_13959_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_13959_end_0 = const()[name = tensor<string, []>("op_13959_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_13959_end_mask_0 = const()[name = tensor<string, []>("op_13959_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_13959_cast = slice_by_index(begin = var_13959_begin_0, end = var_13959_end_0, end_mask = var_13959_end_mask_0, x = transpose_107)[name = tensor<string, []>("op_13959_cast")];
            tensor<int32, [4]> var_13961_begin_0 = const()[name = tensor<string, []>("op_13961_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13961_end_0 = const()[name = tensor<string, []>("op_13961_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_13961_end_mask_0 = const()[name = tensor<string, []>("op_13961_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13961_cast = slice_by_index(begin = var_13961_begin_0, end = var_13961_end_0, end_mask = var_13961_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_13961_cast")];
            tensor<int32, [4]> var_13965_begin_0 = const()[name = tensor<string, []>("op_13965_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_13965_end_0 = const()[name = tensor<string, []>("op_13965_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_13965_end_mask_0 = const()[name = tensor<string, []>("op_13965_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13965_cast = slice_by_index(begin = var_13965_begin_0, end = var_13965_end_0, end_mask = var_13965_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_13965_cast")];
            tensor<int32, [4]> var_13969_begin_0 = const()[name = tensor<string, []>("op_13969_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_13969_end_0 = const()[name = tensor<string, []>("op_13969_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_13969_end_mask_0 = const()[name = tensor<string, []>("op_13969_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13969_cast = slice_by_index(begin = var_13969_begin_0, end = var_13969_end_0, end_mask = var_13969_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_13969_cast")];
            tensor<int32, [4]> var_13973_begin_0 = const()[name = tensor<string, []>("op_13973_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_13973_end_0 = const()[name = tensor<string, []>("op_13973_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_13973_end_mask_0 = const()[name = tensor<string, []>("op_13973_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13973_cast = slice_by_index(begin = var_13973_begin_0, end = var_13973_end_0, end_mask = var_13973_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_13973_cast")];
            tensor<int32, [4]> var_13977_begin_0 = const()[name = tensor<string, []>("op_13977_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_13977_end_0 = const()[name = tensor<string, []>("op_13977_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_13977_end_mask_0 = const()[name = tensor<string, []>("op_13977_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13977_cast = slice_by_index(begin = var_13977_begin_0, end = var_13977_end_0, end_mask = var_13977_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_13977_cast")];
            tensor<int32, [4]> var_13981_begin_0 = const()[name = tensor<string, []>("op_13981_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_13981_end_0 = const()[name = tensor<string, []>("op_13981_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_13981_end_mask_0 = const()[name = tensor<string, []>("op_13981_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13981_cast = slice_by_index(begin = var_13981_begin_0, end = var_13981_end_0, end_mask = var_13981_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_13981_cast")];
            tensor<int32, [4]> var_13985_begin_0 = const()[name = tensor<string, []>("op_13985_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_13985_end_0 = const()[name = tensor<string, []>("op_13985_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_13985_end_mask_0 = const()[name = tensor<string, []>("op_13985_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13985_cast = slice_by_index(begin = var_13985_begin_0, end = var_13985_end_0, end_mask = var_13985_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_13985_cast")];
            tensor<int32, [4]> var_13989_begin_0 = const()[name = tensor<string, []>("op_13989_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_13989_end_0 = const()[name = tensor<string, []>("op_13989_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_13989_end_mask_0 = const()[name = tensor<string, []>("op_13989_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13989_cast = slice_by_index(begin = var_13989_begin_0, end = var_13989_end_0, end_mask = var_13989_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_13989_cast")];
            tensor<int32, [4]> var_13993_begin_0 = const()[name = tensor<string, []>("op_13993_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_13993_end_0 = const()[name = tensor<string, []>("op_13993_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_13993_end_mask_0 = const()[name = tensor<string, []>("op_13993_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13993_cast = slice_by_index(begin = var_13993_begin_0, end = var_13993_end_0, end_mask = var_13993_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_13993_cast")];
            tensor<int32, [4]> var_13997_begin_0 = const()[name = tensor<string, []>("op_13997_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_13997_end_0 = const()[name = tensor<string, []>("op_13997_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_13997_end_mask_0 = const()[name = tensor<string, []>("op_13997_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_13997_cast = slice_by_index(begin = var_13997_begin_0, end = var_13997_end_0, end_mask = var_13997_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_13997_cast")];
            tensor<int32, [4]> var_14001_begin_0 = const()[name = tensor<string, []>("op_14001_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_14001_end_0 = const()[name = tensor<string, []>("op_14001_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_14001_end_mask_0 = const()[name = tensor<string, []>("op_14001_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14001_cast = slice_by_index(begin = var_14001_begin_0, end = var_14001_end_0, end_mask = var_14001_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_14001_cast")];
            tensor<int32, [4]> var_14005_begin_0 = const()[name = tensor<string, []>("op_14005_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_14005_end_0 = const()[name = tensor<string, []>("op_14005_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_14005_end_mask_0 = const()[name = tensor<string, []>("op_14005_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14005_cast = slice_by_index(begin = var_14005_begin_0, end = var_14005_end_0, end_mask = var_14005_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_14005_cast")];
            tensor<int32, [4]> var_14009_begin_0 = const()[name = tensor<string, []>("op_14009_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_14009_end_0 = const()[name = tensor<string, []>("op_14009_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_14009_end_mask_0 = const()[name = tensor<string, []>("op_14009_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14009_cast = slice_by_index(begin = var_14009_begin_0, end = var_14009_end_0, end_mask = var_14009_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_14009_cast")];
            tensor<int32, [4]> var_14013_begin_0 = const()[name = tensor<string, []>("op_14013_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_14013_end_0 = const()[name = tensor<string, []>("op_14013_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_14013_end_mask_0 = const()[name = tensor<string, []>("op_14013_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14013_cast = slice_by_index(begin = var_14013_begin_0, end = var_14013_end_0, end_mask = var_14013_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_14013_cast")];
            tensor<int32, [4]> var_14017_begin_0 = const()[name = tensor<string, []>("op_14017_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_14017_end_0 = const()[name = tensor<string, []>("op_14017_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_14017_end_mask_0 = const()[name = tensor<string, []>("op_14017_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14017_cast = slice_by_index(begin = var_14017_begin_0, end = var_14017_end_0, end_mask = var_14017_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_14017_cast")];
            tensor<int32, [4]> var_14021_begin_0 = const()[name = tensor<string, []>("op_14021_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_14021_end_0 = const()[name = tensor<string, []>("op_14021_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_14021_end_mask_0 = const()[name = tensor<string, []>("op_14021_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14021_cast = slice_by_index(begin = var_14021_begin_0, end = var_14021_end_0, end_mask = var_14021_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_14021_cast")];
            tensor<int32, [4]> var_14025_begin_0 = const()[name = tensor<string, []>("op_14025_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_14025_end_0 = const()[name = tensor<string, []>("op_14025_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_14025_end_mask_0 = const()[name = tensor<string, []>("op_14025_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14025_cast = slice_by_index(begin = var_14025_begin_0, end = var_14025_end_0, end_mask = var_14025_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_14025_cast")];
            tensor<int32, [4]> var_14029_begin_0 = const()[name = tensor<string, []>("op_14029_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_14029_end_0 = const()[name = tensor<string, []>("op_14029_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_14029_end_mask_0 = const()[name = tensor<string, []>("op_14029_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14029_cast = slice_by_index(begin = var_14029_begin_0, end = var_14029_end_0, end_mask = var_14029_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_14029_cast")];
            tensor<int32, [4]> var_14033_begin_0 = const()[name = tensor<string, []>("op_14033_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_14033_end_0 = const()[name = tensor<string, []>("op_14033_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_14033_end_mask_0 = const()[name = tensor<string, []>("op_14033_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14033_cast = slice_by_index(begin = var_14033_begin_0, end = var_14033_end_0, end_mask = var_14033_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_14033_cast")];
            tensor<int32, [4]> var_14037_begin_0 = const()[name = tensor<string, []>("op_14037_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_14037_end_0 = const()[name = tensor<string, []>("op_14037_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_14037_end_mask_0 = const()[name = tensor<string, []>("op_14037_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14037_cast = slice_by_index(begin = var_14037_begin_0, end = var_14037_end_0, end_mask = var_14037_end_mask_0, x = v_65_cast)[name = tensor<string, []>("op_14037_cast")];
            tensor<string, []> var_14041_equation_0 = const()[name = tensor<string, []>("op_14041_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14041_cast = einsum(equation = var_14041_equation_0, values = (var_13883_cast, var_13800_cast))[name = tensor<string, []>("op_14041_cast")];
            tensor<fp16, []> var_14042_to_fp16 = const()[name = tensor<string, []>("op_14042_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1121_cast = mul(x = var_14041_cast, y = var_14042_to_fp16)[name = tensor<string, []>("aw_1121_cast")];
            tensor<string, []> var_14045_equation_0 = const()[name = tensor<string, []>("op_14045_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14045_cast = einsum(equation = var_14045_equation_0, values = (var_13887_cast, var_13804_cast))[name = tensor<string, []>("op_14045_cast")];
            tensor<fp16, []> var_14046_to_fp16 = const()[name = tensor<string, []>("op_14046_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1123_cast = mul(x = var_14045_cast, y = var_14046_to_fp16)[name = tensor<string, []>("aw_1123_cast")];
            tensor<string, []> var_14049_equation_0 = const()[name = tensor<string, []>("op_14049_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14049_cast = einsum(equation = var_14049_equation_0, values = (var_13891_cast, var_13808_cast))[name = tensor<string, []>("op_14049_cast")];
            tensor<fp16, []> var_14050_to_fp16 = const()[name = tensor<string, []>("op_14050_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1125_cast = mul(x = var_14049_cast, y = var_14050_to_fp16)[name = tensor<string, []>("aw_1125_cast")];
            tensor<string, []> var_14053_equation_0 = const()[name = tensor<string, []>("op_14053_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14053_cast = einsum(equation = var_14053_equation_0, values = (var_13895_cast, var_13812_cast))[name = tensor<string, []>("op_14053_cast")];
            tensor<fp16, []> var_14054_to_fp16 = const()[name = tensor<string, []>("op_14054_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1127_cast = mul(x = var_14053_cast, y = var_14054_to_fp16)[name = tensor<string, []>("aw_1127_cast")];
            tensor<string, []> var_14057_equation_0 = const()[name = tensor<string, []>("op_14057_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14057_cast = einsum(equation = var_14057_equation_0, values = (var_13899_cast, var_13816_cast))[name = tensor<string, []>("op_14057_cast")];
            tensor<fp16, []> var_14058_to_fp16 = const()[name = tensor<string, []>("op_14058_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1129_cast = mul(x = var_14057_cast, y = var_14058_to_fp16)[name = tensor<string, []>("aw_1129_cast")];
            tensor<string, []> var_14061_equation_0 = const()[name = tensor<string, []>("op_14061_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14061_cast = einsum(equation = var_14061_equation_0, values = (var_13903_cast, var_13820_cast))[name = tensor<string, []>("op_14061_cast")];
            tensor<fp16, []> var_14062_to_fp16 = const()[name = tensor<string, []>("op_14062_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1131_cast = mul(x = var_14061_cast, y = var_14062_to_fp16)[name = tensor<string, []>("aw_1131_cast")];
            tensor<string, []> var_14065_equation_0 = const()[name = tensor<string, []>("op_14065_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14065_cast = einsum(equation = var_14065_equation_0, values = (var_13907_cast, var_13824_cast))[name = tensor<string, []>("op_14065_cast")];
            tensor<fp16, []> var_14066_to_fp16 = const()[name = tensor<string, []>("op_14066_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1133_cast = mul(x = var_14065_cast, y = var_14066_to_fp16)[name = tensor<string, []>("aw_1133_cast")];
            tensor<string, []> var_14069_equation_0 = const()[name = tensor<string, []>("op_14069_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14069_cast = einsum(equation = var_14069_equation_0, values = (var_13911_cast, var_13828_cast))[name = tensor<string, []>("op_14069_cast")];
            tensor<fp16, []> var_14070_to_fp16 = const()[name = tensor<string, []>("op_14070_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1135_cast = mul(x = var_14069_cast, y = var_14070_to_fp16)[name = tensor<string, []>("aw_1135_cast")];
            tensor<string, []> var_14073_equation_0 = const()[name = tensor<string, []>("op_14073_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14073_cast = einsum(equation = var_14073_equation_0, values = (var_13915_cast, var_13832_cast))[name = tensor<string, []>("op_14073_cast")];
            tensor<fp16, []> var_14074_to_fp16 = const()[name = tensor<string, []>("op_14074_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1137_cast = mul(x = var_14073_cast, y = var_14074_to_fp16)[name = tensor<string, []>("aw_1137_cast")];
            tensor<string, []> var_14077_equation_0 = const()[name = tensor<string, []>("op_14077_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14077_cast = einsum(equation = var_14077_equation_0, values = (var_13919_cast, var_13836_cast))[name = tensor<string, []>("op_14077_cast")];
            tensor<fp16, []> var_14078_to_fp16 = const()[name = tensor<string, []>("op_14078_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1139_cast = mul(x = var_14077_cast, y = var_14078_to_fp16)[name = tensor<string, []>("aw_1139_cast")];
            tensor<string, []> var_14081_equation_0 = const()[name = tensor<string, []>("op_14081_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14081_cast = einsum(equation = var_14081_equation_0, values = (var_13923_cast, var_13840_cast))[name = tensor<string, []>("op_14081_cast")];
            tensor<fp16, []> var_14082_to_fp16 = const()[name = tensor<string, []>("op_14082_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1141_cast = mul(x = var_14081_cast, y = var_14082_to_fp16)[name = tensor<string, []>("aw_1141_cast")];
            tensor<string, []> var_14085_equation_0 = const()[name = tensor<string, []>("op_14085_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14085_cast = einsum(equation = var_14085_equation_0, values = (var_13927_cast, var_13844_cast))[name = tensor<string, []>("op_14085_cast")];
            tensor<fp16, []> var_14086_to_fp16 = const()[name = tensor<string, []>("op_14086_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1143_cast = mul(x = var_14085_cast, y = var_14086_to_fp16)[name = tensor<string, []>("aw_1143_cast")];
            tensor<string, []> var_14089_equation_0 = const()[name = tensor<string, []>("op_14089_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14089_cast = einsum(equation = var_14089_equation_0, values = (var_13931_cast, var_13848_cast))[name = tensor<string, []>("op_14089_cast")];
            tensor<fp16, []> var_14090_to_fp16 = const()[name = tensor<string, []>("op_14090_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1145_cast = mul(x = var_14089_cast, y = var_14090_to_fp16)[name = tensor<string, []>("aw_1145_cast")];
            tensor<string, []> var_14093_equation_0 = const()[name = tensor<string, []>("op_14093_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14093_cast = einsum(equation = var_14093_equation_0, values = (var_13935_cast, var_13852_cast))[name = tensor<string, []>("op_14093_cast")];
            tensor<fp16, []> var_14094_to_fp16 = const()[name = tensor<string, []>("op_14094_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1147_cast = mul(x = var_14093_cast, y = var_14094_to_fp16)[name = tensor<string, []>("aw_1147_cast")];
            tensor<string, []> var_14097_equation_0 = const()[name = tensor<string, []>("op_14097_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14097_cast = einsum(equation = var_14097_equation_0, values = (var_13939_cast, var_13856_cast))[name = tensor<string, []>("op_14097_cast")];
            tensor<fp16, []> var_14098_to_fp16 = const()[name = tensor<string, []>("op_14098_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1149_cast = mul(x = var_14097_cast, y = var_14098_to_fp16)[name = tensor<string, []>("aw_1149_cast")];
            tensor<string, []> var_14101_equation_0 = const()[name = tensor<string, []>("op_14101_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14101_cast = einsum(equation = var_14101_equation_0, values = (var_13943_cast, var_13860_cast))[name = tensor<string, []>("op_14101_cast")];
            tensor<fp16, []> var_14102_to_fp16 = const()[name = tensor<string, []>("op_14102_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1151_cast = mul(x = var_14101_cast, y = var_14102_to_fp16)[name = tensor<string, []>("aw_1151_cast")];
            tensor<string, []> var_14105_equation_0 = const()[name = tensor<string, []>("op_14105_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14105_cast = einsum(equation = var_14105_equation_0, values = (var_13947_cast, var_13864_cast))[name = tensor<string, []>("op_14105_cast")];
            tensor<fp16, []> var_14106_to_fp16 = const()[name = tensor<string, []>("op_14106_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1153_cast = mul(x = var_14105_cast, y = var_14106_to_fp16)[name = tensor<string, []>("aw_1153_cast")];
            tensor<string, []> var_14109_equation_0 = const()[name = tensor<string, []>("op_14109_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14109_cast = einsum(equation = var_14109_equation_0, values = (var_13951_cast, var_13868_cast))[name = tensor<string, []>("op_14109_cast")];
            tensor<fp16, []> var_14110_to_fp16 = const()[name = tensor<string, []>("op_14110_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1155_cast = mul(x = var_14109_cast, y = var_14110_to_fp16)[name = tensor<string, []>("aw_1155_cast")];
            tensor<string, []> var_14113_equation_0 = const()[name = tensor<string, []>("op_14113_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14113_cast = einsum(equation = var_14113_equation_0, values = (var_13955_cast, var_13872_cast))[name = tensor<string, []>("op_14113_cast")];
            tensor<fp16, []> var_14114_to_fp16 = const()[name = tensor<string, []>("op_14114_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1157_cast = mul(x = var_14113_cast, y = var_14114_to_fp16)[name = tensor<string, []>("aw_1157_cast")];
            tensor<string, []> var_14117_equation_0 = const()[name = tensor<string, []>("op_14117_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14117_cast = einsum(equation = var_14117_equation_0, values = (var_13959_cast, var_13876_cast))[name = tensor<string, []>("op_14117_cast")];
            tensor<fp16, []> var_14118_to_fp16 = const()[name = tensor<string, []>("op_14118_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1159_cast = mul(x = var_14117_cast, y = var_14118_to_fp16)[name = tensor<string, []>("aw_1159_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14120_cast = softmax(axis = var_2634, x = aw_1121_cast)[name = tensor<string, []>("op_14120_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14121_cast = softmax(axis = var_2634, x = aw_1123_cast)[name = tensor<string, []>("op_14121_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14122_cast = softmax(axis = var_2634, x = aw_1125_cast)[name = tensor<string, []>("op_14122_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14123_cast = softmax(axis = var_2634, x = aw_1127_cast)[name = tensor<string, []>("op_14123_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14124_cast = softmax(axis = var_2634, x = aw_1129_cast)[name = tensor<string, []>("op_14124_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14125_cast = softmax(axis = var_2634, x = aw_1131_cast)[name = tensor<string, []>("op_14125_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14126_cast = softmax(axis = var_2634, x = aw_1133_cast)[name = tensor<string, []>("op_14126_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14127_cast = softmax(axis = var_2634, x = aw_1135_cast)[name = tensor<string, []>("op_14127_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14128_cast = softmax(axis = var_2634, x = aw_1137_cast)[name = tensor<string, []>("op_14128_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14129_cast = softmax(axis = var_2634, x = aw_1139_cast)[name = tensor<string, []>("op_14129_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14130_cast = softmax(axis = var_2634, x = aw_1141_cast)[name = tensor<string, []>("op_14130_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14131_cast = softmax(axis = var_2634, x = aw_1143_cast)[name = tensor<string, []>("op_14131_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14132_cast = softmax(axis = var_2634, x = aw_1145_cast)[name = tensor<string, []>("op_14132_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14133_cast = softmax(axis = var_2634, x = aw_1147_cast)[name = tensor<string, []>("op_14133_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14134_cast = softmax(axis = var_2634, x = aw_1149_cast)[name = tensor<string, []>("op_14134_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14135_cast = softmax(axis = var_2634, x = aw_1151_cast)[name = tensor<string, []>("op_14135_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14136_cast = softmax(axis = var_2634, x = aw_1153_cast)[name = tensor<string, []>("op_14136_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14137_cast = softmax(axis = var_2634, x = aw_1155_cast)[name = tensor<string, []>("op_14137_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14138_cast = softmax(axis = var_2634, x = aw_1157_cast)[name = tensor<string, []>("op_14138_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14139_cast = softmax(axis = var_2634, x = aw_1159_cast)[name = tensor<string, []>("op_14139_cast")];
            tensor<string, []> var_14141_equation_0 = const()[name = tensor<string, []>("op_14141_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14141_cast = einsum(equation = var_14141_equation_0, values = (var_13961_cast, var_14120_cast))[name = tensor<string, []>("op_14141_cast")];
            tensor<string, []> var_14143_equation_0 = const()[name = tensor<string, []>("op_14143_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14143_cast = einsum(equation = var_14143_equation_0, values = (var_13965_cast, var_14121_cast))[name = tensor<string, []>("op_14143_cast")];
            tensor<string, []> var_14145_equation_0 = const()[name = tensor<string, []>("op_14145_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14145_cast = einsum(equation = var_14145_equation_0, values = (var_13969_cast, var_14122_cast))[name = tensor<string, []>("op_14145_cast")];
            tensor<string, []> var_14147_equation_0 = const()[name = tensor<string, []>("op_14147_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14147_cast = einsum(equation = var_14147_equation_0, values = (var_13973_cast, var_14123_cast))[name = tensor<string, []>("op_14147_cast")];
            tensor<string, []> var_14149_equation_0 = const()[name = tensor<string, []>("op_14149_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14149_cast = einsum(equation = var_14149_equation_0, values = (var_13977_cast, var_14124_cast))[name = tensor<string, []>("op_14149_cast")];
            tensor<string, []> var_14151_equation_0 = const()[name = tensor<string, []>("op_14151_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14151_cast = einsum(equation = var_14151_equation_0, values = (var_13981_cast, var_14125_cast))[name = tensor<string, []>("op_14151_cast")];
            tensor<string, []> var_14153_equation_0 = const()[name = tensor<string, []>("op_14153_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14153_cast = einsum(equation = var_14153_equation_0, values = (var_13985_cast, var_14126_cast))[name = tensor<string, []>("op_14153_cast")];
            tensor<string, []> var_14155_equation_0 = const()[name = tensor<string, []>("op_14155_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14155_cast = einsum(equation = var_14155_equation_0, values = (var_13989_cast, var_14127_cast))[name = tensor<string, []>("op_14155_cast")];
            tensor<string, []> var_14157_equation_0 = const()[name = tensor<string, []>("op_14157_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14157_cast = einsum(equation = var_14157_equation_0, values = (var_13993_cast, var_14128_cast))[name = tensor<string, []>("op_14157_cast")];
            tensor<string, []> var_14159_equation_0 = const()[name = tensor<string, []>("op_14159_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14159_cast = einsum(equation = var_14159_equation_0, values = (var_13997_cast, var_14129_cast))[name = tensor<string, []>("op_14159_cast")];
            tensor<string, []> var_14161_equation_0 = const()[name = tensor<string, []>("op_14161_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14161_cast = einsum(equation = var_14161_equation_0, values = (var_14001_cast, var_14130_cast))[name = tensor<string, []>("op_14161_cast")];
            tensor<string, []> var_14163_equation_0 = const()[name = tensor<string, []>("op_14163_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14163_cast = einsum(equation = var_14163_equation_0, values = (var_14005_cast, var_14131_cast))[name = tensor<string, []>("op_14163_cast")];
            tensor<string, []> var_14165_equation_0 = const()[name = tensor<string, []>("op_14165_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14165_cast = einsum(equation = var_14165_equation_0, values = (var_14009_cast, var_14132_cast))[name = tensor<string, []>("op_14165_cast")];
            tensor<string, []> var_14167_equation_0 = const()[name = tensor<string, []>("op_14167_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14167_cast = einsum(equation = var_14167_equation_0, values = (var_14013_cast, var_14133_cast))[name = tensor<string, []>("op_14167_cast")];
            tensor<string, []> var_14169_equation_0 = const()[name = tensor<string, []>("op_14169_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14169_cast = einsum(equation = var_14169_equation_0, values = (var_14017_cast, var_14134_cast))[name = tensor<string, []>("op_14169_cast")];
            tensor<string, []> var_14171_equation_0 = const()[name = tensor<string, []>("op_14171_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14171_cast = einsum(equation = var_14171_equation_0, values = (var_14021_cast, var_14135_cast))[name = tensor<string, []>("op_14171_cast")];
            tensor<string, []> var_14173_equation_0 = const()[name = tensor<string, []>("op_14173_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14173_cast = einsum(equation = var_14173_equation_0, values = (var_14025_cast, var_14136_cast))[name = tensor<string, []>("op_14173_cast")];
            tensor<string, []> var_14175_equation_0 = const()[name = tensor<string, []>("op_14175_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14175_cast = einsum(equation = var_14175_equation_0, values = (var_14029_cast, var_14137_cast))[name = tensor<string, []>("op_14175_cast")];
            tensor<string, []> var_14177_equation_0 = const()[name = tensor<string, []>("op_14177_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14177_cast = einsum(equation = var_14177_equation_0, values = (var_14033_cast, var_14138_cast))[name = tensor<string, []>("op_14177_cast")];
            tensor<string, []> var_14179_equation_0 = const()[name = tensor<string, []>("op_14179_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14179_cast = einsum(equation = var_14179_equation_0, values = (var_14037_cast, var_14139_cast))[name = tensor<string, []>("op_14179_cast")];
            tensor<bool, []> input_249_interleave_0 = const()[name = tensor<string, []>("input_249_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_249_cast = concat(axis = var_2634, interleave = input_249_interleave_0, values = (var_14141_cast, var_14143_cast, var_14145_cast, var_14147_cast, var_14149_cast, var_14151_cast, var_14153_cast, var_14155_cast, var_14157_cast, var_14159_cast, var_14161_cast, var_14163_cast, var_14165_cast, var_14167_cast, var_14169_cast, var_14171_cast, var_14173_cast, var_14175_cast, var_14177_cast, var_14179_cast))[name = tensor<string, []>("input_249_cast")];
            tensor<int32, [2]> var_14185 = const()[name = tensor<string, []>("op_14185"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14187 = const()[name = tensor<string, []>("op_14187"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_14189_pad_type_0 = const()[name = tensor<string, []>("op_14189_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_14189_pad_0 = const()[name = tensor<string, []>("op_14189_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1110999232)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1114276096)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_14189_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_out_0_bias_to_fp16, dilations = var_14187, groups = var_2634, pad = var_14189_pad_0, pad_type = var_14189_pad_type_0, strides = var_14185, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn1_to_out_0_weight_to_fp16, x = input_249_cast)[name = tensor<string, []>("op_14189_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_99_cast = add(x = var_14189_cast, y = inputs_97_cast)[name = tensor<string, []>("inputs_99_cast")];
            tensor<int32, [1]> var_14193 = const()[name = tensor<string, []>("op_14193"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_99_cast = reduce_mean(axes = var_14193, keep_dims = var_2629, x = inputs_99_cast)[name = tensor<string, []>("channels_mean_99_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_99_cast = sub(x = inputs_99_cast, y = channels_mean_99_cast)[name = tensor<string, []>("zero_mean_99_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_99_cast = mul(x = zero_mean_99_cast, y = zero_mean_99_cast)[name = tensor<string, []>("zero_mean_sq_99_cast")];
            tensor<int32, [1]> var_14197 = const()[name = tensor<string, []>("op_14197"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_14198_cast = reduce_mean(axes = var_14197, keep_dims = var_2629, x = zero_mean_sq_99_cast)[name = tensor<string, []>("op_14198_cast")];
            tensor<fp16, []> var_14199_to_fp16 = const()[name = tensor<string, []>("op_14199_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_14200_cast = add(x = var_14198_cast, y = var_14199_to_fp16)[name = tensor<string, []>("op_14200_cast")];
            tensor<fp16, []> denom_99_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_99_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_99_cast = rsqrt(epsilon = denom_99_epsilon_0_to_fp16, x = var_14200_cast)[name = tensor<string, []>("denom_99_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_99_cast = mul(x = zero_mean_99_cast, y = denom_99_cast)[name = tensor<string, []>("out_99_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_14204_to_fp16 = const()[name = tensor<string, []>("op_14204_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1114278720)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_14205_cast = add(x = out_99_cast, y = var_14204_to_fp16)[name = tensor<string, []>("op_14205_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_14207_to_fp16 = const()[name = tensor<string, []>("op_14207_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1114281344)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_151_cast = mul(x = var_14205_cast, y = var_14207_to_fp16)[name = tensor<string, []>("hidden_states_151_cast")];
            tensor<int32, [2]> var_14214 = const()[name = tensor<string, []>("op_14214"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14216 = const()[name = tensor<string, []>("op_14216"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_67_pad_type_0 = const()[name = tensor<string, []>("q_67_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_67_pad_0 = const()[name = tensor<string, []>("q_67_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1114283968)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_67_cast = conv(dilations = var_14216, groups = var_2634, pad = q_67_pad_0, pad_type = q_67_pad_type_0, strides = var_14214, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_q_weight_to_fp16, x = hidden_states_151_cast)[name = tensor<string, []>("q_67_cast")];
            tensor<int32, [2]> var_14220 = const()[name = tensor<string, []>("op_14220"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14222 = const()[name = tensor<string, []>("op_14222"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_133_pad_type_0 = const()[name = tensor<string, []>("k_133_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_133_pad_0 = const()[name = tensor<string, []>("k_133_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1117560832)))];
            tensor<fp16, [2, 1280, 1, 77]> k_133_cast = conv(dilations = var_14222, groups = var_2634, pad = k_133_pad_0, pad_type = k_133_pad_type_0, strides = var_14220, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_133_cast")];
            tensor<int32, [2]> var_14226 = const()[name = tensor<string, []>("op_14226"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14228 = const()[name = tensor<string, []>("op_14228"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_67_pad_type_0 = const()[name = tensor<string, []>("v_67_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_67_pad_0 = const()[name = tensor<string, []>("v_67_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1122803776)))];
            tensor<fp16, [2, 1280, 1, 77]> v_67_cast = conv(dilations = var_14228, groups = var_2634, pad = v_67_pad_0, pad_type = v_67_pad_type_0, strides = var_14226, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_67_cast")];
            tensor<int32, [4]> var_14232_begin_0 = const()[name = tensor<string, []>("op_14232_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14232_end_0 = const()[name = tensor<string, []>("op_14232_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_14232_end_mask_0 = const()[name = tensor<string, []>("op_14232_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14232_cast = slice_by_index(begin = var_14232_begin_0, end = var_14232_end_0, end_mask = var_14232_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14232_cast")];
            tensor<int32, [4]> var_14236_begin_0 = const()[name = tensor<string, []>("op_14236_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_14236_end_0 = const()[name = tensor<string, []>("op_14236_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_14236_end_mask_0 = const()[name = tensor<string, []>("op_14236_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14236_cast = slice_by_index(begin = var_14236_begin_0, end = var_14236_end_0, end_mask = var_14236_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14236_cast")];
            tensor<int32, [4]> var_14240_begin_0 = const()[name = tensor<string, []>("op_14240_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_14240_end_0 = const()[name = tensor<string, []>("op_14240_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_14240_end_mask_0 = const()[name = tensor<string, []>("op_14240_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14240_cast = slice_by_index(begin = var_14240_begin_0, end = var_14240_end_0, end_mask = var_14240_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14240_cast")];
            tensor<int32, [4]> var_14244_begin_0 = const()[name = tensor<string, []>("op_14244_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_14244_end_0 = const()[name = tensor<string, []>("op_14244_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_14244_end_mask_0 = const()[name = tensor<string, []>("op_14244_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14244_cast = slice_by_index(begin = var_14244_begin_0, end = var_14244_end_0, end_mask = var_14244_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14244_cast")];
            tensor<int32, [4]> var_14248_begin_0 = const()[name = tensor<string, []>("op_14248_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_14248_end_0 = const()[name = tensor<string, []>("op_14248_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_14248_end_mask_0 = const()[name = tensor<string, []>("op_14248_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14248_cast = slice_by_index(begin = var_14248_begin_0, end = var_14248_end_0, end_mask = var_14248_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14248_cast")];
            tensor<int32, [4]> var_14252_begin_0 = const()[name = tensor<string, []>("op_14252_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_14252_end_0 = const()[name = tensor<string, []>("op_14252_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_14252_end_mask_0 = const()[name = tensor<string, []>("op_14252_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14252_cast = slice_by_index(begin = var_14252_begin_0, end = var_14252_end_0, end_mask = var_14252_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14252_cast")];
            tensor<int32, [4]> var_14256_begin_0 = const()[name = tensor<string, []>("op_14256_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_14256_end_0 = const()[name = tensor<string, []>("op_14256_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_14256_end_mask_0 = const()[name = tensor<string, []>("op_14256_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14256_cast = slice_by_index(begin = var_14256_begin_0, end = var_14256_end_0, end_mask = var_14256_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14256_cast")];
            tensor<int32, [4]> var_14260_begin_0 = const()[name = tensor<string, []>("op_14260_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_14260_end_0 = const()[name = tensor<string, []>("op_14260_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_14260_end_mask_0 = const()[name = tensor<string, []>("op_14260_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14260_cast = slice_by_index(begin = var_14260_begin_0, end = var_14260_end_0, end_mask = var_14260_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14260_cast")];
            tensor<int32, [4]> var_14264_begin_0 = const()[name = tensor<string, []>("op_14264_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_14264_end_0 = const()[name = tensor<string, []>("op_14264_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_14264_end_mask_0 = const()[name = tensor<string, []>("op_14264_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14264_cast = slice_by_index(begin = var_14264_begin_0, end = var_14264_end_0, end_mask = var_14264_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14264_cast")];
            tensor<int32, [4]> var_14268_begin_0 = const()[name = tensor<string, []>("op_14268_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_14268_end_0 = const()[name = tensor<string, []>("op_14268_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_14268_end_mask_0 = const()[name = tensor<string, []>("op_14268_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14268_cast = slice_by_index(begin = var_14268_begin_0, end = var_14268_end_0, end_mask = var_14268_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14268_cast")];
            tensor<int32, [4]> var_14272_begin_0 = const()[name = tensor<string, []>("op_14272_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_14272_end_0 = const()[name = tensor<string, []>("op_14272_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_14272_end_mask_0 = const()[name = tensor<string, []>("op_14272_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14272_cast = slice_by_index(begin = var_14272_begin_0, end = var_14272_end_0, end_mask = var_14272_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14272_cast")];
            tensor<int32, [4]> var_14276_begin_0 = const()[name = tensor<string, []>("op_14276_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_14276_end_0 = const()[name = tensor<string, []>("op_14276_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_14276_end_mask_0 = const()[name = tensor<string, []>("op_14276_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14276_cast = slice_by_index(begin = var_14276_begin_0, end = var_14276_end_0, end_mask = var_14276_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14276_cast")];
            tensor<int32, [4]> var_14280_begin_0 = const()[name = tensor<string, []>("op_14280_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_14280_end_0 = const()[name = tensor<string, []>("op_14280_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_14280_end_mask_0 = const()[name = tensor<string, []>("op_14280_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14280_cast = slice_by_index(begin = var_14280_begin_0, end = var_14280_end_0, end_mask = var_14280_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14280_cast")];
            tensor<int32, [4]> var_14284_begin_0 = const()[name = tensor<string, []>("op_14284_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_14284_end_0 = const()[name = tensor<string, []>("op_14284_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_14284_end_mask_0 = const()[name = tensor<string, []>("op_14284_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14284_cast = slice_by_index(begin = var_14284_begin_0, end = var_14284_end_0, end_mask = var_14284_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14284_cast")];
            tensor<int32, [4]> var_14288_begin_0 = const()[name = tensor<string, []>("op_14288_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_14288_end_0 = const()[name = tensor<string, []>("op_14288_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_14288_end_mask_0 = const()[name = tensor<string, []>("op_14288_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14288_cast = slice_by_index(begin = var_14288_begin_0, end = var_14288_end_0, end_mask = var_14288_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14288_cast")];
            tensor<int32, [4]> var_14292_begin_0 = const()[name = tensor<string, []>("op_14292_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_14292_end_0 = const()[name = tensor<string, []>("op_14292_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_14292_end_mask_0 = const()[name = tensor<string, []>("op_14292_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14292_cast = slice_by_index(begin = var_14292_begin_0, end = var_14292_end_0, end_mask = var_14292_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14292_cast")];
            tensor<int32, [4]> var_14296_begin_0 = const()[name = tensor<string, []>("op_14296_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_14296_end_0 = const()[name = tensor<string, []>("op_14296_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_14296_end_mask_0 = const()[name = tensor<string, []>("op_14296_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14296_cast = slice_by_index(begin = var_14296_begin_0, end = var_14296_end_0, end_mask = var_14296_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14296_cast")];
            tensor<int32, [4]> var_14300_begin_0 = const()[name = tensor<string, []>("op_14300_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_14300_end_0 = const()[name = tensor<string, []>("op_14300_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_14300_end_mask_0 = const()[name = tensor<string, []>("op_14300_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14300_cast = slice_by_index(begin = var_14300_begin_0, end = var_14300_end_0, end_mask = var_14300_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14300_cast")];
            tensor<int32, [4]> var_14304_begin_0 = const()[name = tensor<string, []>("op_14304_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_14304_end_0 = const()[name = tensor<string, []>("op_14304_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_14304_end_mask_0 = const()[name = tensor<string, []>("op_14304_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14304_cast = slice_by_index(begin = var_14304_begin_0, end = var_14304_end_0, end_mask = var_14304_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14304_cast")];
            tensor<int32, [4]> var_14308_begin_0 = const()[name = tensor<string, []>("op_14308_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_14308_end_0 = const()[name = tensor<string, []>("op_14308_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_14308_end_mask_0 = const()[name = tensor<string, []>("op_14308_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14308_cast = slice_by_index(begin = var_14308_begin_0, end = var_14308_end_0, end_mask = var_14308_end_mask_0, x = q_67_cast)[name = tensor<string, []>("op_14308_cast")];
            tensor<int32, [4]> k_135_perm_0 = const()[name = tensor<string, []>("k_135_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_14315_begin_0 = const()[name = tensor<string, []>("op_14315_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14315_end_0 = const()[name = tensor<string, []>("op_14315_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_14315_end_mask_0 = const()[name = tensor<string, []>("op_14315_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_106 = transpose(perm = k_135_perm_0, x = k_133_cast)[name = tensor<string, []>("transpose_106")];
            tensor<fp16, [2, 77, 1, 64]> var_14315_cast = slice_by_index(begin = var_14315_begin_0, end = var_14315_end_0, end_mask = var_14315_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14315_cast")];
            tensor<int32, [4]> var_14319_begin_0 = const()[name = tensor<string, []>("op_14319_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_14319_end_0 = const()[name = tensor<string, []>("op_14319_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_14319_end_mask_0 = const()[name = tensor<string, []>("op_14319_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14319_cast = slice_by_index(begin = var_14319_begin_0, end = var_14319_end_0, end_mask = var_14319_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14319_cast")];
            tensor<int32, [4]> var_14323_begin_0 = const()[name = tensor<string, []>("op_14323_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_14323_end_0 = const()[name = tensor<string, []>("op_14323_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_14323_end_mask_0 = const()[name = tensor<string, []>("op_14323_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14323_cast = slice_by_index(begin = var_14323_begin_0, end = var_14323_end_0, end_mask = var_14323_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14323_cast")];
            tensor<int32, [4]> var_14327_begin_0 = const()[name = tensor<string, []>("op_14327_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_14327_end_0 = const()[name = tensor<string, []>("op_14327_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_14327_end_mask_0 = const()[name = tensor<string, []>("op_14327_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14327_cast = slice_by_index(begin = var_14327_begin_0, end = var_14327_end_0, end_mask = var_14327_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14327_cast")];
            tensor<int32, [4]> var_14331_begin_0 = const()[name = tensor<string, []>("op_14331_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_14331_end_0 = const()[name = tensor<string, []>("op_14331_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_14331_end_mask_0 = const()[name = tensor<string, []>("op_14331_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14331_cast = slice_by_index(begin = var_14331_begin_0, end = var_14331_end_0, end_mask = var_14331_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14331_cast")];
            tensor<int32, [4]> var_14335_begin_0 = const()[name = tensor<string, []>("op_14335_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_14335_end_0 = const()[name = tensor<string, []>("op_14335_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_14335_end_mask_0 = const()[name = tensor<string, []>("op_14335_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14335_cast = slice_by_index(begin = var_14335_begin_0, end = var_14335_end_0, end_mask = var_14335_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14335_cast")];
            tensor<int32, [4]> var_14339_begin_0 = const()[name = tensor<string, []>("op_14339_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_14339_end_0 = const()[name = tensor<string, []>("op_14339_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_14339_end_mask_0 = const()[name = tensor<string, []>("op_14339_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14339_cast = slice_by_index(begin = var_14339_begin_0, end = var_14339_end_0, end_mask = var_14339_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14339_cast")];
            tensor<int32, [4]> var_14343_begin_0 = const()[name = tensor<string, []>("op_14343_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_14343_end_0 = const()[name = tensor<string, []>("op_14343_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_14343_end_mask_0 = const()[name = tensor<string, []>("op_14343_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14343_cast = slice_by_index(begin = var_14343_begin_0, end = var_14343_end_0, end_mask = var_14343_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14343_cast")];
            tensor<int32, [4]> var_14347_begin_0 = const()[name = tensor<string, []>("op_14347_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_14347_end_0 = const()[name = tensor<string, []>("op_14347_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_14347_end_mask_0 = const()[name = tensor<string, []>("op_14347_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14347_cast = slice_by_index(begin = var_14347_begin_0, end = var_14347_end_0, end_mask = var_14347_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14347_cast")];
            tensor<int32, [4]> var_14351_begin_0 = const()[name = tensor<string, []>("op_14351_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_14351_end_0 = const()[name = tensor<string, []>("op_14351_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_14351_end_mask_0 = const()[name = tensor<string, []>("op_14351_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14351_cast = slice_by_index(begin = var_14351_begin_0, end = var_14351_end_0, end_mask = var_14351_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14351_cast")];
            tensor<int32, [4]> var_14355_begin_0 = const()[name = tensor<string, []>("op_14355_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_14355_end_0 = const()[name = tensor<string, []>("op_14355_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_14355_end_mask_0 = const()[name = tensor<string, []>("op_14355_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14355_cast = slice_by_index(begin = var_14355_begin_0, end = var_14355_end_0, end_mask = var_14355_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14355_cast")];
            tensor<int32, [4]> var_14359_begin_0 = const()[name = tensor<string, []>("op_14359_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_14359_end_0 = const()[name = tensor<string, []>("op_14359_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_14359_end_mask_0 = const()[name = tensor<string, []>("op_14359_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14359_cast = slice_by_index(begin = var_14359_begin_0, end = var_14359_end_0, end_mask = var_14359_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14359_cast")];
            tensor<int32, [4]> var_14363_begin_0 = const()[name = tensor<string, []>("op_14363_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_14363_end_0 = const()[name = tensor<string, []>("op_14363_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_14363_end_mask_0 = const()[name = tensor<string, []>("op_14363_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14363_cast = slice_by_index(begin = var_14363_begin_0, end = var_14363_end_0, end_mask = var_14363_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14363_cast")];
            tensor<int32, [4]> var_14367_begin_0 = const()[name = tensor<string, []>("op_14367_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_14367_end_0 = const()[name = tensor<string, []>("op_14367_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_14367_end_mask_0 = const()[name = tensor<string, []>("op_14367_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14367_cast = slice_by_index(begin = var_14367_begin_0, end = var_14367_end_0, end_mask = var_14367_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14367_cast")];
            tensor<int32, [4]> var_14371_begin_0 = const()[name = tensor<string, []>("op_14371_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_14371_end_0 = const()[name = tensor<string, []>("op_14371_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_14371_end_mask_0 = const()[name = tensor<string, []>("op_14371_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14371_cast = slice_by_index(begin = var_14371_begin_0, end = var_14371_end_0, end_mask = var_14371_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14371_cast")];
            tensor<int32, [4]> var_14375_begin_0 = const()[name = tensor<string, []>("op_14375_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_14375_end_0 = const()[name = tensor<string, []>("op_14375_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_14375_end_mask_0 = const()[name = tensor<string, []>("op_14375_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14375_cast = slice_by_index(begin = var_14375_begin_0, end = var_14375_end_0, end_mask = var_14375_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14375_cast")];
            tensor<int32, [4]> var_14379_begin_0 = const()[name = tensor<string, []>("op_14379_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_14379_end_0 = const()[name = tensor<string, []>("op_14379_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_14379_end_mask_0 = const()[name = tensor<string, []>("op_14379_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14379_cast = slice_by_index(begin = var_14379_begin_0, end = var_14379_end_0, end_mask = var_14379_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14379_cast")];
            tensor<int32, [4]> var_14383_begin_0 = const()[name = tensor<string, []>("op_14383_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_14383_end_0 = const()[name = tensor<string, []>("op_14383_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_14383_end_mask_0 = const()[name = tensor<string, []>("op_14383_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14383_cast = slice_by_index(begin = var_14383_begin_0, end = var_14383_end_0, end_mask = var_14383_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14383_cast")];
            tensor<int32, [4]> var_14387_begin_0 = const()[name = tensor<string, []>("op_14387_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_14387_end_0 = const()[name = tensor<string, []>("op_14387_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_14387_end_mask_0 = const()[name = tensor<string, []>("op_14387_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14387_cast = slice_by_index(begin = var_14387_begin_0, end = var_14387_end_0, end_mask = var_14387_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14387_cast")];
            tensor<int32, [4]> var_14391_begin_0 = const()[name = tensor<string, []>("op_14391_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_14391_end_0 = const()[name = tensor<string, []>("op_14391_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_14391_end_mask_0 = const()[name = tensor<string, []>("op_14391_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14391_cast = slice_by_index(begin = var_14391_begin_0, end = var_14391_end_0, end_mask = var_14391_end_mask_0, x = transpose_106)[name = tensor<string, []>("op_14391_cast")];
            tensor<int32, [4]> var_14393_begin_0 = const()[name = tensor<string, []>("op_14393_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14393_end_0 = const()[name = tensor<string, []>("op_14393_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_14393_end_mask_0 = const()[name = tensor<string, []>("op_14393_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14393_cast = slice_by_index(begin = var_14393_begin_0, end = var_14393_end_0, end_mask = var_14393_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14393_cast")];
            tensor<int32, [4]> var_14397_begin_0 = const()[name = tensor<string, []>("op_14397_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_14397_end_0 = const()[name = tensor<string, []>("op_14397_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_14397_end_mask_0 = const()[name = tensor<string, []>("op_14397_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14397_cast = slice_by_index(begin = var_14397_begin_0, end = var_14397_end_0, end_mask = var_14397_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14397_cast")];
            tensor<int32, [4]> var_14401_begin_0 = const()[name = tensor<string, []>("op_14401_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_14401_end_0 = const()[name = tensor<string, []>("op_14401_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_14401_end_mask_0 = const()[name = tensor<string, []>("op_14401_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14401_cast = slice_by_index(begin = var_14401_begin_0, end = var_14401_end_0, end_mask = var_14401_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14401_cast")];
            tensor<int32, [4]> var_14405_begin_0 = const()[name = tensor<string, []>("op_14405_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_14405_end_0 = const()[name = tensor<string, []>("op_14405_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_14405_end_mask_0 = const()[name = tensor<string, []>("op_14405_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14405_cast = slice_by_index(begin = var_14405_begin_0, end = var_14405_end_0, end_mask = var_14405_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14405_cast")];
            tensor<int32, [4]> var_14409_begin_0 = const()[name = tensor<string, []>("op_14409_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_14409_end_0 = const()[name = tensor<string, []>("op_14409_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_14409_end_mask_0 = const()[name = tensor<string, []>("op_14409_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14409_cast = slice_by_index(begin = var_14409_begin_0, end = var_14409_end_0, end_mask = var_14409_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14409_cast")];
            tensor<int32, [4]> var_14413_begin_0 = const()[name = tensor<string, []>("op_14413_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_14413_end_0 = const()[name = tensor<string, []>("op_14413_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_14413_end_mask_0 = const()[name = tensor<string, []>("op_14413_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14413_cast = slice_by_index(begin = var_14413_begin_0, end = var_14413_end_0, end_mask = var_14413_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14413_cast")];
            tensor<int32, [4]> var_14417_begin_0 = const()[name = tensor<string, []>("op_14417_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_14417_end_0 = const()[name = tensor<string, []>("op_14417_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_14417_end_mask_0 = const()[name = tensor<string, []>("op_14417_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14417_cast = slice_by_index(begin = var_14417_begin_0, end = var_14417_end_0, end_mask = var_14417_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14417_cast")];
            tensor<int32, [4]> var_14421_begin_0 = const()[name = tensor<string, []>("op_14421_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_14421_end_0 = const()[name = tensor<string, []>("op_14421_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_14421_end_mask_0 = const()[name = tensor<string, []>("op_14421_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14421_cast = slice_by_index(begin = var_14421_begin_0, end = var_14421_end_0, end_mask = var_14421_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14421_cast")];
            tensor<int32, [4]> var_14425_begin_0 = const()[name = tensor<string, []>("op_14425_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_14425_end_0 = const()[name = tensor<string, []>("op_14425_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_14425_end_mask_0 = const()[name = tensor<string, []>("op_14425_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14425_cast = slice_by_index(begin = var_14425_begin_0, end = var_14425_end_0, end_mask = var_14425_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14425_cast")];
            tensor<int32, [4]> var_14429_begin_0 = const()[name = tensor<string, []>("op_14429_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_14429_end_0 = const()[name = tensor<string, []>("op_14429_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_14429_end_mask_0 = const()[name = tensor<string, []>("op_14429_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14429_cast = slice_by_index(begin = var_14429_begin_0, end = var_14429_end_0, end_mask = var_14429_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14429_cast")];
            tensor<int32, [4]> var_14433_begin_0 = const()[name = tensor<string, []>("op_14433_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_14433_end_0 = const()[name = tensor<string, []>("op_14433_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_14433_end_mask_0 = const()[name = tensor<string, []>("op_14433_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14433_cast = slice_by_index(begin = var_14433_begin_0, end = var_14433_end_0, end_mask = var_14433_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14433_cast")];
            tensor<int32, [4]> var_14437_begin_0 = const()[name = tensor<string, []>("op_14437_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_14437_end_0 = const()[name = tensor<string, []>("op_14437_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_14437_end_mask_0 = const()[name = tensor<string, []>("op_14437_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14437_cast = slice_by_index(begin = var_14437_begin_0, end = var_14437_end_0, end_mask = var_14437_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14437_cast")];
            tensor<int32, [4]> var_14441_begin_0 = const()[name = tensor<string, []>("op_14441_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_14441_end_0 = const()[name = tensor<string, []>("op_14441_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_14441_end_mask_0 = const()[name = tensor<string, []>("op_14441_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14441_cast = slice_by_index(begin = var_14441_begin_0, end = var_14441_end_0, end_mask = var_14441_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14441_cast")];
            tensor<int32, [4]> var_14445_begin_0 = const()[name = tensor<string, []>("op_14445_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_14445_end_0 = const()[name = tensor<string, []>("op_14445_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_14445_end_mask_0 = const()[name = tensor<string, []>("op_14445_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14445_cast = slice_by_index(begin = var_14445_begin_0, end = var_14445_end_0, end_mask = var_14445_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14445_cast")];
            tensor<int32, [4]> var_14449_begin_0 = const()[name = tensor<string, []>("op_14449_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_14449_end_0 = const()[name = tensor<string, []>("op_14449_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_14449_end_mask_0 = const()[name = tensor<string, []>("op_14449_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14449_cast = slice_by_index(begin = var_14449_begin_0, end = var_14449_end_0, end_mask = var_14449_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14449_cast")];
            tensor<int32, [4]> var_14453_begin_0 = const()[name = tensor<string, []>("op_14453_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_14453_end_0 = const()[name = tensor<string, []>("op_14453_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_14453_end_mask_0 = const()[name = tensor<string, []>("op_14453_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14453_cast = slice_by_index(begin = var_14453_begin_0, end = var_14453_end_0, end_mask = var_14453_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14453_cast")];
            tensor<int32, [4]> var_14457_begin_0 = const()[name = tensor<string, []>("op_14457_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_14457_end_0 = const()[name = tensor<string, []>("op_14457_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_14457_end_mask_0 = const()[name = tensor<string, []>("op_14457_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14457_cast = slice_by_index(begin = var_14457_begin_0, end = var_14457_end_0, end_mask = var_14457_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14457_cast")];
            tensor<int32, [4]> var_14461_begin_0 = const()[name = tensor<string, []>("op_14461_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_14461_end_0 = const()[name = tensor<string, []>("op_14461_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_14461_end_mask_0 = const()[name = tensor<string, []>("op_14461_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14461_cast = slice_by_index(begin = var_14461_begin_0, end = var_14461_end_0, end_mask = var_14461_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14461_cast")];
            tensor<int32, [4]> var_14465_begin_0 = const()[name = tensor<string, []>("op_14465_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_14465_end_0 = const()[name = tensor<string, []>("op_14465_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_14465_end_mask_0 = const()[name = tensor<string, []>("op_14465_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14465_cast = slice_by_index(begin = var_14465_begin_0, end = var_14465_end_0, end_mask = var_14465_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14465_cast")];
            tensor<int32, [4]> var_14469_begin_0 = const()[name = tensor<string, []>("op_14469_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_14469_end_0 = const()[name = tensor<string, []>("op_14469_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_14469_end_mask_0 = const()[name = tensor<string, []>("op_14469_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14469_cast = slice_by_index(begin = var_14469_begin_0, end = var_14469_end_0, end_mask = var_14469_end_mask_0, x = v_67_cast)[name = tensor<string, []>("op_14469_cast")];
            tensor<string, []> var_14473_equation_0 = const()[name = tensor<string, []>("op_14473_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14473_cast = einsum(equation = var_14473_equation_0, values = (var_14315_cast, var_14232_cast))[name = tensor<string, []>("op_14473_cast")];
            tensor<fp16, []> var_14474_to_fp16 = const()[name = tensor<string, []>("op_14474_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1161_cast = mul(x = var_14473_cast, y = var_14474_to_fp16)[name = tensor<string, []>("aw_1161_cast")];
            tensor<string, []> var_14477_equation_0 = const()[name = tensor<string, []>("op_14477_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14477_cast = einsum(equation = var_14477_equation_0, values = (var_14319_cast, var_14236_cast))[name = tensor<string, []>("op_14477_cast")];
            tensor<fp16, []> var_14478_to_fp16 = const()[name = tensor<string, []>("op_14478_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1163_cast = mul(x = var_14477_cast, y = var_14478_to_fp16)[name = tensor<string, []>("aw_1163_cast")];
            tensor<string, []> var_14481_equation_0 = const()[name = tensor<string, []>("op_14481_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14481_cast = einsum(equation = var_14481_equation_0, values = (var_14323_cast, var_14240_cast))[name = tensor<string, []>("op_14481_cast")];
            tensor<fp16, []> var_14482_to_fp16 = const()[name = tensor<string, []>("op_14482_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1165_cast = mul(x = var_14481_cast, y = var_14482_to_fp16)[name = tensor<string, []>("aw_1165_cast")];
            tensor<string, []> var_14485_equation_0 = const()[name = tensor<string, []>("op_14485_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14485_cast = einsum(equation = var_14485_equation_0, values = (var_14327_cast, var_14244_cast))[name = tensor<string, []>("op_14485_cast")];
            tensor<fp16, []> var_14486_to_fp16 = const()[name = tensor<string, []>("op_14486_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1167_cast = mul(x = var_14485_cast, y = var_14486_to_fp16)[name = tensor<string, []>("aw_1167_cast")];
            tensor<string, []> var_14489_equation_0 = const()[name = tensor<string, []>("op_14489_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14489_cast = einsum(equation = var_14489_equation_0, values = (var_14331_cast, var_14248_cast))[name = tensor<string, []>("op_14489_cast")];
            tensor<fp16, []> var_14490_to_fp16 = const()[name = tensor<string, []>("op_14490_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1169_cast = mul(x = var_14489_cast, y = var_14490_to_fp16)[name = tensor<string, []>("aw_1169_cast")];
            tensor<string, []> var_14493_equation_0 = const()[name = tensor<string, []>("op_14493_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14493_cast = einsum(equation = var_14493_equation_0, values = (var_14335_cast, var_14252_cast))[name = tensor<string, []>("op_14493_cast")];
            tensor<fp16, []> var_14494_to_fp16 = const()[name = tensor<string, []>("op_14494_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1171_cast = mul(x = var_14493_cast, y = var_14494_to_fp16)[name = tensor<string, []>("aw_1171_cast")];
            tensor<string, []> var_14497_equation_0 = const()[name = tensor<string, []>("op_14497_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14497_cast = einsum(equation = var_14497_equation_0, values = (var_14339_cast, var_14256_cast))[name = tensor<string, []>("op_14497_cast")];
            tensor<fp16, []> var_14498_to_fp16 = const()[name = tensor<string, []>("op_14498_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1173_cast = mul(x = var_14497_cast, y = var_14498_to_fp16)[name = tensor<string, []>("aw_1173_cast")];
            tensor<string, []> var_14501_equation_0 = const()[name = tensor<string, []>("op_14501_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14501_cast = einsum(equation = var_14501_equation_0, values = (var_14343_cast, var_14260_cast))[name = tensor<string, []>("op_14501_cast")];
            tensor<fp16, []> var_14502_to_fp16 = const()[name = tensor<string, []>("op_14502_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1175_cast = mul(x = var_14501_cast, y = var_14502_to_fp16)[name = tensor<string, []>("aw_1175_cast")];
            tensor<string, []> var_14505_equation_0 = const()[name = tensor<string, []>("op_14505_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14505_cast = einsum(equation = var_14505_equation_0, values = (var_14347_cast, var_14264_cast))[name = tensor<string, []>("op_14505_cast")];
            tensor<fp16, []> var_14506_to_fp16 = const()[name = tensor<string, []>("op_14506_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1177_cast = mul(x = var_14505_cast, y = var_14506_to_fp16)[name = tensor<string, []>("aw_1177_cast")];
            tensor<string, []> var_14509_equation_0 = const()[name = tensor<string, []>("op_14509_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14509_cast = einsum(equation = var_14509_equation_0, values = (var_14351_cast, var_14268_cast))[name = tensor<string, []>("op_14509_cast")];
            tensor<fp16, []> var_14510_to_fp16 = const()[name = tensor<string, []>("op_14510_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1179_cast = mul(x = var_14509_cast, y = var_14510_to_fp16)[name = tensor<string, []>("aw_1179_cast")];
            tensor<string, []> var_14513_equation_0 = const()[name = tensor<string, []>("op_14513_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14513_cast = einsum(equation = var_14513_equation_0, values = (var_14355_cast, var_14272_cast))[name = tensor<string, []>("op_14513_cast")];
            tensor<fp16, []> var_14514_to_fp16 = const()[name = tensor<string, []>("op_14514_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1181_cast = mul(x = var_14513_cast, y = var_14514_to_fp16)[name = tensor<string, []>("aw_1181_cast")];
            tensor<string, []> var_14517_equation_0 = const()[name = tensor<string, []>("op_14517_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14517_cast = einsum(equation = var_14517_equation_0, values = (var_14359_cast, var_14276_cast))[name = tensor<string, []>("op_14517_cast")];
            tensor<fp16, []> var_14518_to_fp16 = const()[name = tensor<string, []>("op_14518_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1183_cast = mul(x = var_14517_cast, y = var_14518_to_fp16)[name = tensor<string, []>("aw_1183_cast")];
            tensor<string, []> var_14521_equation_0 = const()[name = tensor<string, []>("op_14521_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14521_cast = einsum(equation = var_14521_equation_0, values = (var_14363_cast, var_14280_cast))[name = tensor<string, []>("op_14521_cast")];
            tensor<fp16, []> var_14522_to_fp16 = const()[name = tensor<string, []>("op_14522_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1185_cast = mul(x = var_14521_cast, y = var_14522_to_fp16)[name = tensor<string, []>("aw_1185_cast")];
            tensor<string, []> var_14525_equation_0 = const()[name = tensor<string, []>("op_14525_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14525_cast = einsum(equation = var_14525_equation_0, values = (var_14367_cast, var_14284_cast))[name = tensor<string, []>("op_14525_cast")];
            tensor<fp16, []> var_14526_to_fp16 = const()[name = tensor<string, []>("op_14526_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1187_cast = mul(x = var_14525_cast, y = var_14526_to_fp16)[name = tensor<string, []>("aw_1187_cast")];
            tensor<string, []> var_14529_equation_0 = const()[name = tensor<string, []>("op_14529_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14529_cast = einsum(equation = var_14529_equation_0, values = (var_14371_cast, var_14288_cast))[name = tensor<string, []>("op_14529_cast")];
            tensor<fp16, []> var_14530_to_fp16 = const()[name = tensor<string, []>("op_14530_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1189_cast = mul(x = var_14529_cast, y = var_14530_to_fp16)[name = tensor<string, []>("aw_1189_cast")];
            tensor<string, []> var_14533_equation_0 = const()[name = tensor<string, []>("op_14533_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14533_cast = einsum(equation = var_14533_equation_0, values = (var_14375_cast, var_14292_cast))[name = tensor<string, []>("op_14533_cast")];
            tensor<fp16, []> var_14534_to_fp16 = const()[name = tensor<string, []>("op_14534_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1191_cast = mul(x = var_14533_cast, y = var_14534_to_fp16)[name = tensor<string, []>("aw_1191_cast")];
            tensor<string, []> var_14537_equation_0 = const()[name = tensor<string, []>("op_14537_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14537_cast = einsum(equation = var_14537_equation_0, values = (var_14379_cast, var_14296_cast))[name = tensor<string, []>("op_14537_cast")];
            tensor<fp16, []> var_14538_to_fp16 = const()[name = tensor<string, []>("op_14538_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1193_cast = mul(x = var_14537_cast, y = var_14538_to_fp16)[name = tensor<string, []>("aw_1193_cast")];
            tensor<string, []> var_14541_equation_0 = const()[name = tensor<string, []>("op_14541_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14541_cast = einsum(equation = var_14541_equation_0, values = (var_14383_cast, var_14300_cast))[name = tensor<string, []>("op_14541_cast")];
            tensor<fp16, []> var_14542_to_fp16 = const()[name = tensor<string, []>("op_14542_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1195_cast = mul(x = var_14541_cast, y = var_14542_to_fp16)[name = tensor<string, []>("aw_1195_cast")];
            tensor<string, []> var_14545_equation_0 = const()[name = tensor<string, []>("op_14545_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14545_cast = einsum(equation = var_14545_equation_0, values = (var_14387_cast, var_14304_cast))[name = tensor<string, []>("op_14545_cast")];
            tensor<fp16, []> var_14546_to_fp16 = const()[name = tensor<string, []>("op_14546_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1197_cast = mul(x = var_14545_cast, y = var_14546_to_fp16)[name = tensor<string, []>("aw_1197_cast")];
            tensor<string, []> var_14549_equation_0 = const()[name = tensor<string, []>("op_14549_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_14549_cast = einsum(equation = var_14549_equation_0, values = (var_14391_cast, var_14308_cast))[name = tensor<string, []>("op_14549_cast")];
            tensor<fp16, []> var_14550_to_fp16 = const()[name = tensor<string, []>("op_14550_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1199_cast = mul(x = var_14549_cast, y = var_14550_to_fp16)[name = tensor<string, []>("aw_1199_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14552_cast = softmax(axis = var_2634, x = aw_1161_cast)[name = tensor<string, []>("op_14552_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14553_cast = softmax(axis = var_2634, x = aw_1163_cast)[name = tensor<string, []>("op_14553_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14554_cast = softmax(axis = var_2634, x = aw_1165_cast)[name = tensor<string, []>("op_14554_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14555_cast = softmax(axis = var_2634, x = aw_1167_cast)[name = tensor<string, []>("op_14555_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14556_cast = softmax(axis = var_2634, x = aw_1169_cast)[name = tensor<string, []>("op_14556_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14557_cast = softmax(axis = var_2634, x = aw_1171_cast)[name = tensor<string, []>("op_14557_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14558_cast = softmax(axis = var_2634, x = aw_1173_cast)[name = tensor<string, []>("op_14558_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14559_cast = softmax(axis = var_2634, x = aw_1175_cast)[name = tensor<string, []>("op_14559_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14560_cast = softmax(axis = var_2634, x = aw_1177_cast)[name = tensor<string, []>("op_14560_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14561_cast = softmax(axis = var_2634, x = aw_1179_cast)[name = tensor<string, []>("op_14561_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14562_cast = softmax(axis = var_2634, x = aw_1181_cast)[name = tensor<string, []>("op_14562_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14563_cast = softmax(axis = var_2634, x = aw_1183_cast)[name = tensor<string, []>("op_14563_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14564_cast = softmax(axis = var_2634, x = aw_1185_cast)[name = tensor<string, []>("op_14564_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14565_cast = softmax(axis = var_2634, x = aw_1187_cast)[name = tensor<string, []>("op_14565_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14566_cast = softmax(axis = var_2634, x = aw_1189_cast)[name = tensor<string, []>("op_14566_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14567_cast = softmax(axis = var_2634, x = aw_1191_cast)[name = tensor<string, []>("op_14567_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14568_cast = softmax(axis = var_2634, x = aw_1193_cast)[name = tensor<string, []>("op_14568_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14569_cast = softmax(axis = var_2634, x = aw_1195_cast)[name = tensor<string, []>("op_14569_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14570_cast = softmax(axis = var_2634, x = aw_1197_cast)[name = tensor<string, []>("op_14570_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_14571_cast = softmax(axis = var_2634, x = aw_1199_cast)[name = tensor<string, []>("op_14571_cast")];
            tensor<string, []> var_14573_equation_0 = const()[name = tensor<string, []>("op_14573_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14573_cast = einsum(equation = var_14573_equation_0, values = (var_14393_cast, var_14552_cast))[name = tensor<string, []>("op_14573_cast")];
            tensor<string, []> var_14575_equation_0 = const()[name = tensor<string, []>("op_14575_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14575_cast = einsum(equation = var_14575_equation_0, values = (var_14397_cast, var_14553_cast))[name = tensor<string, []>("op_14575_cast")];
            tensor<string, []> var_14577_equation_0 = const()[name = tensor<string, []>("op_14577_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14577_cast = einsum(equation = var_14577_equation_0, values = (var_14401_cast, var_14554_cast))[name = tensor<string, []>("op_14577_cast")];
            tensor<string, []> var_14579_equation_0 = const()[name = tensor<string, []>("op_14579_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14579_cast = einsum(equation = var_14579_equation_0, values = (var_14405_cast, var_14555_cast))[name = tensor<string, []>("op_14579_cast")];
            tensor<string, []> var_14581_equation_0 = const()[name = tensor<string, []>("op_14581_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14581_cast = einsum(equation = var_14581_equation_0, values = (var_14409_cast, var_14556_cast))[name = tensor<string, []>("op_14581_cast")];
            tensor<string, []> var_14583_equation_0 = const()[name = tensor<string, []>("op_14583_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14583_cast = einsum(equation = var_14583_equation_0, values = (var_14413_cast, var_14557_cast))[name = tensor<string, []>("op_14583_cast")];
            tensor<string, []> var_14585_equation_0 = const()[name = tensor<string, []>("op_14585_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14585_cast = einsum(equation = var_14585_equation_0, values = (var_14417_cast, var_14558_cast))[name = tensor<string, []>("op_14585_cast")];
            tensor<string, []> var_14587_equation_0 = const()[name = tensor<string, []>("op_14587_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14587_cast = einsum(equation = var_14587_equation_0, values = (var_14421_cast, var_14559_cast))[name = tensor<string, []>("op_14587_cast")];
            tensor<string, []> var_14589_equation_0 = const()[name = tensor<string, []>("op_14589_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14589_cast = einsum(equation = var_14589_equation_0, values = (var_14425_cast, var_14560_cast))[name = tensor<string, []>("op_14589_cast")];
            tensor<string, []> var_14591_equation_0 = const()[name = tensor<string, []>("op_14591_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14591_cast = einsum(equation = var_14591_equation_0, values = (var_14429_cast, var_14561_cast))[name = tensor<string, []>("op_14591_cast")];
            tensor<string, []> var_14593_equation_0 = const()[name = tensor<string, []>("op_14593_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14593_cast = einsum(equation = var_14593_equation_0, values = (var_14433_cast, var_14562_cast))[name = tensor<string, []>("op_14593_cast")];
            tensor<string, []> var_14595_equation_0 = const()[name = tensor<string, []>("op_14595_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14595_cast = einsum(equation = var_14595_equation_0, values = (var_14437_cast, var_14563_cast))[name = tensor<string, []>("op_14595_cast")];
            tensor<string, []> var_14597_equation_0 = const()[name = tensor<string, []>("op_14597_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14597_cast = einsum(equation = var_14597_equation_0, values = (var_14441_cast, var_14564_cast))[name = tensor<string, []>("op_14597_cast")];
            tensor<string, []> var_14599_equation_0 = const()[name = tensor<string, []>("op_14599_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14599_cast = einsum(equation = var_14599_equation_0, values = (var_14445_cast, var_14565_cast))[name = tensor<string, []>("op_14599_cast")];
            tensor<string, []> var_14601_equation_0 = const()[name = tensor<string, []>("op_14601_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14601_cast = einsum(equation = var_14601_equation_0, values = (var_14449_cast, var_14566_cast))[name = tensor<string, []>("op_14601_cast")];
            tensor<string, []> var_14603_equation_0 = const()[name = tensor<string, []>("op_14603_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14603_cast = einsum(equation = var_14603_equation_0, values = (var_14453_cast, var_14567_cast))[name = tensor<string, []>("op_14603_cast")];
            tensor<string, []> var_14605_equation_0 = const()[name = tensor<string, []>("op_14605_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14605_cast = einsum(equation = var_14605_equation_0, values = (var_14457_cast, var_14568_cast))[name = tensor<string, []>("op_14605_cast")];
            tensor<string, []> var_14607_equation_0 = const()[name = tensor<string, []>("op_14607_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14607_cast = einsum(equation = var_14607_equation_0, values = (var_14461_cast, var_14569_cast))[name = tensor<string, []>("op_14607_cast")];
            tensor<string, []> var_14609_equation_0 = const()[name = tensor<string, []>("op_14609_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14609_cast = einsum(equation = var_14609_equation_0, values = (var_14465_cast, var_14570_cast))[name = tensor<string, []>("op_14609_cast")];
            tensor<string, []> var_14611_equation_0 = const()[name = tensor<string, []>("op_14611_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_14611_cast = einsum(equation = var_14611_equation_0, values = (var_14469_cast, var_14571_cast))[name = tensor<string, []>("op_14611_cast")];
            tensor<bool, []> input_251_interleave_0 = const()[name = tensor<string, []>("input_251_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_251_cast = concat(axis = var_2634, interleave = input_251_interleave_0, values = (var_14573_cast, var_14575_cast, var_14577_cast, var_14579_cast, var_14581_cast, var_14583_cast, var_14585_cast, var_14587_cast, var_14589_cast, var_14591_cast, var_14593_cast, var_14595_cast, var_14597_cast, var_14599_cast, var_14601_cast, var_14603_cast, var_14605_cast, var_14607_cast, var_14609_cast, var_14611_cast))[name = tensor<string, []>("input_251_cast")];
            tensor<int32, [2]> var_14617 = const()[name = tensor<string, []>("op_14617"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14619 = const()[name = tensor<string, []>("op_14619"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_14621_pad_type_0 = const()[name = tensor<string, []>("op_14621_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_14621_pad_0 = const()[name = tensor<string, []>("op_14621_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1128046720)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1131323584)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_14621_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_out_0_bias_to_fp16, dilations = var_14619, groups = var_2634, pad = var_14621_pad_0, pad_type = var_14621_pad_type_0, strides = var_14617, weight = down_blocks_2_attentions_1_transformer_blocks_2_attn2_to_out_0_weight_to_fp16, x = input_251_cast)[name = tensor<string, []>("op_14621_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_101_cast = add(x = var_14621_cast, y = inputs_99_cast)[name = tensor<string, []>("inputs_101_cast")];
            tensor<int32, [1]> var_14625 = const()[name = tensor<string, []>("op_14625"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_101_cast = reduce_mean(axes = var_14625, keep_dims = var_2629, x = inputs_101_cast)[name = tensor<string, []>("channels_mean_101_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_101_cast = sub(x = inputs_101_cast, y = channels_mean_101_cast)[name = tensor<string, []>("zero_mean_101_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_101_cast = mul(x = zero_mean_101_cast, y = zero_mean_101_cast)[name = tensor<string, []>("zero_mean_sq_101_cast")];
            tensor<int32, [1]> var_14629 = const()[name = tensor<string, []>("op_14629"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_14630_cast = reduce_mean(axes = var_14629, keep_dims = var_2629, x = zero_mean_sq_101_cast)[name = tensor<string, []>("op_14630_cast")];
            tensor<fp16, []> var_14631_to_fp16 = const()[name = tensor<string, []>("op_14631_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_14632_cast = add(x = var_14630_cast, y = var_14631_to_fp16)[name = tensor<string, []>("op_14632_cast")];
            tensor<fp16, []> denom_101_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_101_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_101_cast = rsqrt(epsilon = denom_101_epsilon_0_to_fp16, x = var_14632_cast)[name = tensor<string, []>("denom_101_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_101_cast = mul(x = zero_mean_101_cast, y = denom_101_cast)[name = tensor<string, []>("out_101_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_14636_to_fp16 = const()[name = tensor<string, []>("op_14636_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1131326208)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_14637_cast = add(x = out_101_cast, y = var_14636_to_fp16)[name = tensor<string, []>("op_14637_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_14639_to_fp16 = const()[name = tensor<string, []>("op_14639_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1131328832)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_253_cast = mul(x = var_14637_cast, y = var_14639_to_fp16)[name = tensor<string, []>("input_253_cast")];
            tensor<int32, [2]> var_14647 = const()[name = tensor<string, []>("op_14647"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14649 = const()[name = tensor<string, []>("op_14649"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_14651_pad_type_0 = const()[name = tensor<string, []>("op_14651_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_14651_pad_0 = const()[name = tensor<string, []>("op_14651_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_2_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_2_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1131331456)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_1_transformer_blocks_2_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_2_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1157545920)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_14651_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_2_ff_net_0_proj_bias_to_fp16, dilations = var_14649, groups = var_2634, pad = var_14651_pad_0, pad_type = var_14651_pad_type_0, strides = var_14647, weight = down_blocks_2_attentions_1_transformer_blocks_2_ff_net_0_proj_weight_to_fp16, x = input_253_cast)[name = tensor<string, []>("op_14651_cast")];
            tensor<int32, [2]> var_14652_split_sizes_0 = const()[name = tensor<string, []>("op_14652_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_14652_axis_0 = const()[name = tensor<string, []>("op_14652_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_14652_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_14652_cast_1 = split(axis = var_14652_axis_0, split_sizes = var_14652_split_sizes_0, x = var_14651_cast)[name = tensor<string, []>("op_14652_cast")];
            tensor<string, []> var_14654_mode_0 = const()[name = tensor<string, []>("op_14654_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_14654_cast = gelu(mode = var_14654_mode_0, x = var_14652_cast_1)[name = tensor<string, []>("op_14654_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_255_cast = mul(x = var_14652_cast_0, y = var_14654_cast)[name = tensor<string, []>("input_255_cast")];
            tensor<int32, [2]> var_14658 = const()[name = tensor<string, []>("op_14658"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14660 = const()[name = tensor<string, []>("op_14660"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_14662_pad_type_0 = const()[name = tensor<string, []>("op_14662_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_14662_pad_0 = const()[name = tensor<string, []>("op_14662_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_2_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_2_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1157566464)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_2_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_2_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1170673728)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_14662_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_2_ff_net_2_bias_to_fp16, dilations = var_14660, groups = var_2634, pad = var_14662_pad_0, pad_type = var_14662_pad_type_0, strides = var_14658, weight = down_blocks_2_attentions_1_transformer_blocks_2_ff_net_2_weight_to_fp16, x = input_255_cast)[name = tensor<string, []>("op_14662_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_103_cast = add(x = var_14662_cast, y = inputs_101_cast)[name = tensor<string, []>("inputs_103_cast")];
            tensor<int32, [1]> var_14672 = const()[name = tensor<string, []>("op_14672"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_103_cast = reduce_mean(axes = var_14672, keep_dims = var_2629, x = inputs_103_cast)[name = tensor<string, []>("channels_mean_103_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_103_cast = sub(x = inputs_103_cast, y = channels_mean_103_cast)[name = tensor<string, []>("zero_mean_103_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_103_cast = mul(x = zero_mean_103_cast, y = zero_mean_103_cast)[name = tensor<string, []>("zero_mean_sq_103_cast")];
            tensor<int32, [1]> var_14676 = const()[name = tensor<string, []>("op_14676"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_14677_cast = reduce_mean(axes = var_14676, keep_dims = var_2629, x = zero_mean_sq_103_cast)[name = tensor<string, []>("op_14677_cast")];
            tensor<fp16, []> var_14678_to_fp16 = const()[name = tensor<string, []>("op_14678_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_14679_cast = add(x = var_14677_cast, y = var_14678_to_fp16)[name = tensor<string, []>("op_14679_cast")];
            tensor<fp16, []> denom_103_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_103_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_103_cast = rsqrt(epsilon = denom_103_epsilon_0_to_fp16, x = var_14679_cast)[name = tensor<string, []>("denom_103_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_103_cast = mul(x = zero_mean_103_cast, y = denom_103_cast)[name = tensor<string, []>("out_103_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_14683_to_fp16 = const()[name = tensor<string, []>("op_14683_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1170676352)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_14684_cast = add(x = out_103_cast, y = var_14683_to_fp16)[name = tensor<string, []>("op_14684_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_14686_to_fp16 = const()[name = tensor<string, []>("op_14686_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1170678976)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_155_cast = mul(x = var_14684_cast, y = var_14686_to_fp16)[name = tensor<string, []>("hidden_states_155_cast")];
            tensor<int32, [2]> var_14693 = const()[name = tensor<string, []>("op_14693"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14695 = const()[name = tensor<string, []>("op_14695"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_69_pad_type_0 = const()[name = tensor<string, []>("q_69_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_69_pad_0 = const()[name = tensor<string, []>("q_69_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1170681600)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_69_cast = conv(dilations = var_14695, groups = var_2634, pad = q_69_pad_0, pad_type = q_69_pad_type_0, strides = var_14693, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_q_weight_to_fp16, x = hidden_states_155_cast)[name = tensor<string, []>("q_69_cast")];
            tensor<int32, [2]> var_14699 = const()[name = tensor<string, []>("op_14699"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14701 = const()[name = tensor<string, []>("op_14701"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_137_pad_type_0 = const()[name = tensor<string, []>("k_137_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_137_pad_0 = const()[name = tensor<string, []>("k_137_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1173958464)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_137_cast = conv(dilations = var_14701, groups = var_2634, pad = k_137_pad_0, pad_type = k_137_pad_type_0, strides = var_14699, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_k_weight_to_fp16, x = hidden_states_155_cast)[name = tensor<string, []>("k_137_cast")];
            tensor<int32, [2]> var_14705 = const()[name = tensor<string, []>("op_14705"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14707 = const()[name = tensor<string, []>("op_14707"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_69_pad_type_0 = const()[name = tensor<string, []>("v_69_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_69_pad_0 = const()[name = tensor<string, []>("v_69_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1177235328)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_69_cast = conv(dilations = var_14707, groups = var_2634, pad = v_69_pad_0, pad_type = v_69_pad_type_0, strides = var_14705, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_v_weight_to_fp16, x = hidden_states_155_cast)[name = tensor<string, []>("v_69_cast")];
            tensor<int32, [4]> var_14711_begin_0 = const()[name = tensor<string, []>("op_14711_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14711_end_0 = const()[name = tensor<string, []>("op_14711_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_14711_end_mask_0 = const()[name = tensor<string, []>("op_14711_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14711_cast = slice_by_index(begin = var_14711_begin_0, end = var_14711_end_0, end_mask = var_14711_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14711_cast")];
            tensor<int32, [4]> var_14715_begin_0 = const()[name = tensor<string, []>("op_14715_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_14715_end_0 = const()[name = tensor<string, []>("op_14715_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_14715_end_mask_0 = const()[name = tensor<string, []>("op_14715_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14715_cast = slice_by_index(begin = var_14715_begin_0, end = var_14715_end_0, end_mask = var_14715_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14715_cast")];
            tensor<int32, [4]> var_14719_begin_0 = const()[name = tensor<string, []>("op_14719_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_14719_end_0 = const()[name = tensor<string, []>("op_14719_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_14719_end_mask_0 = const()[name = tensor<string, []>("op_14719_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14719_cast = slice_by_index(begin = var_14719_begin_0, end = var_14719_end_0, end_mask = var_14719_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14719_cast")];
            tensor<int32, [4]> var_14723_begin_0 = const()[name = tensor<string, []>("op_14723_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_14723_end_0 = const()[name = tensor<string, []>("op_14723_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_14723_end_mask_0 = const()[name = tensor<string, []>("op_14723_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14723_cast = slice_by_index(begin = var_14723_begin_0, end = var_14723_end_0, end_mask = var_14723_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14723_cast")];
            tensor<int32, [4]> var_14727_begin_0 = const()[name = tensor<string, []>("op_14727_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_14727_end_0 = const()[name = tensor<string, []>("op_14727_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_14727_end_mask_0 = const()[name = tensor<string, []>("op_14727_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14727_cast = slice_by_index(begin = var_14727_begin_0, end = var_14727_end_0, end_mask = var_14727_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14727_cast")];
            tensor<int32, [4]> var_14731_begin_0 = const()[name = tensor<string, []>("op_14731_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_14731_end_0 = const()[name = tensor<string, []>("op_14731_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_14731_end_mask_0 = const()[name = tensor<string, []>("op_14731_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14731_cast = slice_by_index(begin = var_14731_begin_0, end = var_14731_end_0, end_mask = var_14731_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14731_cast")];
            tensor<int32, [4]> var_14735_begin_0 = const()[name = tensor<string, []>("op_14735_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_14735_end_0 = const()[name = tensor<string, []>("op_14735_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_14735_end_mask_0 = const()[name = tensor<string, []>("op_14735_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14735_cast = slice_by_index(begin = var_14735_begin_0, end = var_14735_end_0, end_mask = var_14735_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14735_cast")];
            tensor<int32, [4]> var_14739_begin_0 = const()[name = tensor<string, []>("op_14739_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_14739_end_0 = const()[name = tensor<string, []>("op_14739_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_14739_end_mask_0 = const()[name = tensor<string, []>("op_14739_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14739_cast = slice_by_index(begin = var_14739_begin_0, end = var_14739_end_0, end_mask = var_14739_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14739_cast")];
            tensor<int32, [4]> var_14743_begin_0 = const()[name = tensor<string, []>("op_14743_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_14743_end_0 = const()[name = tensor<string, []>("op_14743_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_14743_end_mask_0 = const()[name = tensor<string, []>("op_14743_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14743_cast = slice_by_index(begin = var_14743_begin_0, end = var_14743_end_0, end_mask = var_14743_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14743_cast")];
            tensor<int32, [4]> var_14747_begin_0 = const()[name = tensor<string, []>("op_14747_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_14747_end_0 = const()[name = tensor<string, []>("op_14747_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_14747_end_mask_0 = const()[name = tensor<string, []>("op_14747_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14747_cast = slice_by_index(begin = var_14747_begin_0, end = var_14747_end_0, end_mask = var_14747_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14747_cast")];
            tensor<int32, [4]> var_14751_begin_0 = const()[name = tensor<string, []>("op_14751_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_14751_end_0 = const()[name = tensor<string, []>("op_14751_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_14751_end_mask_0 = const()[name = tensor<string, []>("op_14751_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14751_cast = slice_by_index(begin = var_14751_begin_0, end = var_14751_end_0, end_mask = var_14751_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14751_cast")];
            tensor<int32, [4]> var_14755_begin_0 = const()[name = tensor<string, []>("op_14755_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_14755_end_0 = const()[name = tensor<string, []>("op_14755_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_14755_end_mask_0 = const()[name = tensor<string, []>("op_14755_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14755_cast = slice_by_index(begin = var_14755_begin_0, end = var_14755_end_0, end_mask = var_14755_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14755_cast")];
            tensor<int32, [4]> var_14759_begin_0 = const()[name = tensor<string, []>("op_14759_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_14759_end_0 = const()[name = tensor<string, []>("op_14759_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_14759_end_mask_0 = const()[name = tensor<string, []>("op_14759_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14759_cast = slice_by_index(begin = var_14759_begin_0, end = var_14759_end_0, end_mask = var_14759_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14759_cast")];
            tensor<int32, [4]> var_14763_begin_0 = const()[name = tensor<string, []>("op_14763_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_14763_end_0 = const()[name = tensor<string, []>("op_14763_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_14763_end_mask_0 = const()[name = tensor<string, []>("op_14763_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14763_cast = slice_by_index(begin = var_14763_begin_0, end = var_14763_end_0, end_mask = var_14763_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14763_cast")];
            tensor<int32, [4]> var_14767_begin_0 = const()[name = tensor<string, []>("op_14767_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_14767_end_0 = const()[name = tensor<string, []>("op_14767_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_14767_end_mask_0 = const()[name = tensor<string, []>("op_14767_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14767_cast = slice_by_index(begin = var_14767_begin_0, end = var_14767_end_0, end_mask = var_14767_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14767_cast")];
            tensor<int32, [4]> var_14771_begin_0 = const()[name = tensor<string, []>("op_14771_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_14771_end_0 = const()[name = tensor<string, []>("op_14771_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_14771_end_mask_0 = const()[name = tensor<string, []>("op_14771_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14771_cast = slice_by_index(begin = var_14771_begin_0, end = var_14771_end_0, end_mask = var_14771_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14771_cast")];
            tensor<int32, [4]> var_14775_begin_0 = const()[name = tensor<string, []>("op_14775_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_14775_end_0 = const()[name = tensor<string, []>("op_14775_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_14775_end_mask_0 = const()[name = tensor<string, []>("op_14775_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14775_cast = slice_by_index(begin = var_14775_begin_0, end = var_14775_end_0, end_mask = var_14775_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14775_cast")];
            tensor<int32, [4]> var_14779_begin_0 = const()[name = tensor<string, []>("op_14779_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_14779_end_0 = const()[name = tensor<string, []>("op_14779_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_14779_end_mask_0 = const()[name = tensor<string, []>("op_14779_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14779_cast = slice_by_index(begin = var_14779_begin_0, end = var_14779_end_0, end_mask = var_14779_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14779_cast")];
            tensor<int32, [4]> var_14783_begin_0 = const()[name = tensor<string, []>("op_14783_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_14783_end_0 = const()[name = tensor<string, []>("op_14783_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_14783_end_mask_0 = const()[name = tensor<string, []>("op_14783_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14783_cast = slice_by_index(begin = var_14783_begin_0, end = var_14783_end_0, end_mask = var_14783_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14783_cast")];
            tensor<int32, [4]> var_14787_begin_0 = const()[name = tensor<string, []>("op_14787_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_14787_end_0 = const()[name = tensor<string, []>("op_14787_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_14787_end_mask_0 = const()[name = tensor<string, []>("op_14787_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14787_cast = slice_by_index(begin = var_14787_begin_0, end = var_14787_end_0, end_mask = var_14787_end_mask_0, x = q_69_cast)[name = tensor<string, []>("op_14787_cast")];
            tensor<int32, [4]> k_139_perm_0 = const()[name = tensor<string, []>("k_139_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_14794_begin_0 = const()[name = tensor<string, []>("op_14794_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14794_end_0 = const()[name = tensor<string, []>("op_14794_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_14794_end_mask_0 = const()[name = tensor<string, []>("op_14794_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_105 = transpose(perm = k_139_perm_0, x = k_137_cast)[name = tensor<string, []>("transpose_105")];
            tensor<fp16, [2, 1024, 1, 64]> var_14794_cast = slice_by_index(begin = var_14794_begin_0, end = var_14794_end_0, end_mask = var_14794_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14794_cast")];
            tensor<int32, [4]> var_14798_begin_0 = const()[name = tensor<string, []>("op_14798_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_14798_end_0 = const()[name = tensor<string, []>("op_14798_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_14798_end_mask_0 = const()[name = tensor<string, []>("op_14798_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_14798_cast = slice_by_index(begin = var_14798_begin_0, end = var_14798_end_0, end_mask = var_14798_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14798_cast")];
            tensor<int32, [4]> var_14802_begin_0 = const()[name = tensor<string, []>("op_14802_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_14802_end_0 = const()[name = tensor<string, []>("op_14802_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_14802_end_mask_0 = const()[name = tensor<string, []>("op_14802_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_14802_cast = slice_by_index(begin = var_14802_begin_0, end = var_14802_end_0, end_mask = var_14802_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14802_cast")];
            tensor<int32, [4]> var_14806_begin_0 = const()[name = tensor<string, []>("op_14806_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_14806_end_0 = const()[name = tensor<string, []>("op_14806_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_14806_end_mask_0 = const()[name = tensor<string, []>("op_14806_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_14806_cast = slice_by_index(begin = var_14806_begin_0, end = var_14806_end_0, end_mask = var_14806_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14806_cast")];
            tensor<int32, [4]> var_14810_begin_0 = const()[name = tensor<string, []>("op_14810_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_14810_end_0 = const()[name = tensor<string, []>("op_14810_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_14810_end_mask_0 = const()[name = tensor<string, []>("op_14810_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_14810_cast = slice_by_index(begin = var_14810_begin_0, end = var_14810_end_0, end_mask = var_14810_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14810_cast")];
            tensor<int32, [4]> var_14814_begin_0 = const()[name = tensor<string, []>("op_14814_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_14814_end_0 = const()[name = tensor<string, []>("op_14814_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_14814_end_mask_0 = const()[name = tensor<string, []>("op_14814_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_14814_cast = slice_by_index(begin = var_14814_begin_0, end = var_14814_end_0, end_mask = var_14814_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14814_cast")];
            tensor<int32, [4]> var_14818_begin_0 = const()[name = tensor<string, []>("op_14818_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_14818_end_0 = const()[name = tensor<string, []>("op_14818_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_14818_end_mask_0 = const()[name = tensor<string, []>("op_14818_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_14818_cast = slice_by_index(begin = var_14818_begin_0, end = var_14818_end_0, end_mask = var_14818_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14818_cast")];
            tensor<int32, [4]> var_14822_begin_0 = const()[name = tensor<string, []>("op_14822_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_14822_end_0 = const()[name = tensor<string, []>("op_14822_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_14822_end_mask_0 = const()[name = tensor<string, []>("op_14822_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_14822_cast = slice_by_index(begin = var_14822_begin_0, end = var_14822_end_0, end_mask = var_14822_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14822_cast")];
            tensor<int32, [4]> var_14826_begin_0 = const()[name = tensor<string, []>("op_14826_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_14826_end_0 = const()[name = tensor<string, []>("op_14826_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_14826_end_mask_0 = const()[name = tensor<string, []>("op_14826_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_14826_cast = slice_by_index(begin = var_14826_begin_0, end = var_14826_end_0, end_mask = var_14826_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14826_cast")];
            tensor<int32, [4]> var_14830_begin_0 = const()[name = tensor<string, []>("op_14830_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_14830_end_0 = const()[name = tensor<string, []>("op_14830_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_14830_end_mask_0 = const()[name = tensor<string, []>("op_14830_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_14830_cast = slice_by_index(begin = var_14830_begin_0, end = var_14830_end_0, end_mask = var_14830_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14830_cast")];
            tensor<int32, [4]> var_14834_begin_0 = const()[name = tensor<string, []>("op_14834_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_14834_end_0 = const()[name = tensor<string, []>("op_14834_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_14834_end_mask_0 = const()[name = tensor<string, []>("op_14834_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_14834_cast = slice_by_index(begin = var_14834_begin_0, end = var_14834_end_0, end_mask = var_14834_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14834_cast")];
            tensor<int32, [4]> var_14838_begin_0 = const()[name = tensor<string, []>("op_14838_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_14838_end_0 = const()[name = tensor<string, []>("op_14838_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_14838_end_mask_0 = const()[name = tensor<string, []>("op_14838_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_14838_cast = slice_by_index(begin = var_14838_begin_0, end = var_14838_end_0, end_mask = var_14838_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14838_cast")];
            tensor<int32, [4]> var_14842_begin_0 = const()[name = tensor<string, []>("op_14842_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_14842_end_0 = const()[name = tensor<string, []>("op_14842_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_14842_end_mask_0 = const()[name = tensor<string, []>("op_14842_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_14842_cast = slice_by_index(begin = var_14842_begin_0, end = var_14842_end_0, end_mask = var_14842_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14842_cast")];
            tensor<int32, [4]> var_14846_begin_0 = const()[name = tensor<string, []>("op_14846_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_14846_end_0 = const()[name = tensor<string, []>("op_14846_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_14846_end_mask_0 = const()[name = tensor<string, []>("op_14846_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_14846_cast = slice_by_index(begin = var_14846_begin_0, end = var_14846_end_0, end_mask = var_14846_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14846_cast")];
            tensor<int32, [4]> var_14850_begin_0 = const()[name = tensor<string, []>("op_14850_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_14850_end_0 = const()[name = tensor<string, []>("op_14850_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_14850_end_mask_0 = const()[name = tensor<string, []>("op_14850_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_14850_cast = slice_by_index(begin = var_14850_begin_0, end = var_14850_end_0, end_mask = var_14850_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14850_cast")];
            tensor<int32, [4]> var_14854_begin_0 = const()[name = tensor<string, []>("op_14854_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_14854_end_0 = const()[name = tensor<string, []>("op_14854_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_14854_end_mask_0 = const()[name = tensor<string, []>("op_14854_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_14854_cast = slice_by_index(begin = var_14854_begin_0, end = var_14854_end_0, end_mask = var_14854_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14854_cast")];
            tensor<int32, [4]> var_14858_begin_0 = const()[name = tensor<string, []>("op_14858_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_14858_end_0 = const()[name = tensor<string, []>("op_14858_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_14858_end_mask_0 = const()[name = tensor<string, []>("op_14858_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_14858_cast = slice_by_index(begin = var_14858_begin_0, end = var_14858_end_0, end_mask = var_14858_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14858_cast")];
            tensor<int32, [4]> var_14862_begin_0 = const()[name = tensor<string, []>("op_14862_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_14862_end_0 = const()[name = tensor<string, []>("op_14862_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_14862_end_mask_0 = const()[name = tensor<string, []>("op_14862_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_14862_cast = slice_by_index(begin = var_14862_begin_0, end = var_14862_end_0, end_mask = var_14862_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14862_cast")];
            tensor<int32, [4]> var_14866_begin_0 = const()[name = tensor<string, []>("op_14866_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_14866_end_0 = const()[name = tensor<string, []>("op_14866_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_14866_end_mask_0 = const()[name = tensor<string, []>("op_14866_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_14866_cast = slice_by_index(begin = var_14866_begin_0, end = var_14866_end_0, end_mask = var_14866_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14866_cast")];
            tensor<int32, [4]> var_14870_begin_0 = const()[name = tensor<string, []>("op_14870_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_14870_end_0 = const()[name = tensor<string, []>("op_14870_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_14870_end_mask_0 = const()[name = tensor<string, []>("op_14870_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_14870_cast = slice_by_index(begin = var_14870_begin_0, end = var_14870_end_0, end_mask = var_14870_end_mask_0, x = transpose_105)[name = tensor<string, []>("op_14870_cast")];
            tensor<int32, [4]> var_14872_begin_0 = const()[name = tensor<string, []>("op_14872_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14872_end_0 = const()[name = tensor<string, []>("op_14872_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_14872_end_mask_0 = const()[name = tensor<string, []>("op_14872_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14872_cast = slice_by_index(begin = var_14872_begin_0, end = var_14872_end_0, end_mask = var_14872_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14872_cast")];
            tensor<int32, [4]> var_14876_begin_0 = const()[name = tensor<string, []>("op_14876_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_14876_end_0 = const()[name = tensor<string, []>("op_14876_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_14876_end_mask_0 = const()[name = tensor<string, []>("op_14876_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14876_cast = slice_by_index(begin = var_14876_begin_0, end = var_14876_end_0, end_mask = var_14876_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14876_cast")];
            tensor<int32, [4]> var_14880_begin_0 = const()[name = tensor<string, []>("op_14880_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_14880_end_0 = const()[name = tensor<string, []>("op_14880_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_14880_end_mask_0 = const()[name = tensor<string, []>("op_14880_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14880_cast = slice_by_index(begin = var_14880_begin_0, end = var_14880_end_0, end_mask = var_14880_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14880_cast")];
            tensor<int32, [4]> var_14884_begin_0 = const()[name = tensor<string, []>("op_14884_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_14884_end_0 = const()[name = tensor<string, []>("op_14884_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_14884_end_mask_0 = const()[name = tensor<string, []>("op_14884_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14884_cast = slice_by_index(begin = var_14884_begin_0, end = var_14884_end_0, end_mask = var_14884_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14884_cast")];
            tensor<int32, [4]> var_14888_begin_0 = const()[name = tensor<string, []>("op_14888_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_14888_end_0 = const()[name = tensor<string, []>("op_14888_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_14888_end_mask_0 = const()[name = tensor<string, []>("op_14888_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14888_cast = slice_by_index(begin = var_14888_begin_0, end = var_14888_end_0, end_mask = var_14888_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14888_cast")];
            tensor<int32, [4]> var_14892_begin_0 = const()[name = tensor<string, []>("op_14892_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_14892_end_0 = const()[name = tensor<string, []>("op_14892_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_14892_end_mask_0 = const()[name = tensor<string, []>("op_14892_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14892_cast = slice_by_index(begin = var_14892_begin_0, end = var_14892_end_0, end_mask = var_14892_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14892_cast")];
            tensor<int32, [4]> var_14896_begin_0 = const()[name = tensor<string, []>("op_14896_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_14896_end_0 = const()[name = tensor<string, []>("op_14896_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_14896_end_mask_0 = const()[name = tensor<string, []>("op_14896_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14896_cast = slice_by_index(begin = var_14896_begin_0, end = var_14896_end_0, end_mask = var_14896_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14896_cast")];
            tensor<int32, [4]> var_14900_begin_0 = const()[name = tensor<string, []>("op_14900_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_14900_end_0 = const()[name = tensor<string, []>("op_14900_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_14900_end_mask_0 = const()[name = tensor<string, []>("op_14900_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14900_cast = slice_by_index(begin = var_14900_begin_0, end = var_14900_end_0, end_mask = var_14900_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14900_cast")];
            tensor<int32, [4]> var_14904_begin_0 = const()[name = tensor<string, []>("op_14904_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_14904_end_0 = const()[name = tensor<string, []>("op_14904_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_14904_end_mask_0 = const()[name = tensor<string, []>("op_14904_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14904_cast = slice_by_index(begin = var_14904_begin_0, end = var_14904_end_0, end_mask = var_14904_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14904_cast")];
            tensor<int32, [4]> var_14908_begin_0 = const()[name = tensor<string, []>("op_14908_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_14908_end_0 = const()[name = tensor<string, []>("op_14908_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_14908_end_mask_0 = const()[name = tensor<string, []>("op_14908_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14908_cast = slice_by_index(begin = var_14908_begin_0, end = var_14908_end_0, end_mask = var_14908_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14908_cast")];
            tensor<int32, [4]> var_14912_begin_0 = const()[name = tensor<string, []>("op_14912_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_14912_end_0 = const()[name = tensor<string, []>("op_14912_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_14912_end_mask_0 = const()[name = tensor<string, []>("op_14912_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14912_cast = slice_by_index(begin = var_14912_begin_0, end = var_14912_end_0, end_mask = var_14912_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14912_cast")];
            tensor<int32, [4]> var_14916_begin_0 = const()[name = tensor<string, []>("op_14916_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_14916_end_0 = const()[name = tensor<string, []>("op_14916_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_14916_end_mask_0 = const()[name = tensor<string, []>("op_14916_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14916_cast = slice_by_index(begin = var_14916_begin_0, end = var_14916_end_0, end_mask = var_14916_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14916_cast")];
            tensor<int32, [4]> var_14920_begin_0 = const()[name = tensor<string, []>("op_14920_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_14920_end_0 = const()[name = tensor<string, []>("op_14920_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_14920_end_mask_0 = const()[name = tensor<string, []>("op_14920_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14920_cast = slice_by_index(begin = var_14920_begin_0, end = var_14920_end_0, end_mask = var_14920_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14920_cast")];
            tensor<int32, [4]> var_14924_begin_0 = const()[name = tensor<string, []>("op_14924_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_14924_end_0 = const()[name = tensor<string, []>("op_14924_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_14924_end_mask_0 = const()[name = tensor<string, []>("op_14924_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14924_cast = slice_by_index(begin = var_14924_begin_0, end = var_14924_end_0, end_mask = var_14924_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14924_cast")];
            tensor<int32, [4]> var_14928_begin_0 = const()[name = tensor<string, []>("op_14928_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_14928_end_0 = const()[name = tensor<string, []>("op_14928_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_14928_end_mask_0 = const()[name = tensor<string, []>("op_14928_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14928_cast = slice_by_index(begin = var_14928_begin_0, end = var_14928_end_0, end_mask = var_14928_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14928_cast")];
            tensor<int32, [4]> var_14932_begin_0 = const()[name = tensor<string, []>("op_14932_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_14932_end_0 = const()[name = tensor<string, []>("op_14932_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_14932_end_mask_0 = const()[name = tensor<string, []>("op_14932_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14932_cast = slice_by_index(begin = var_14932_begin_0, end = var_14932_end_0, end_mask = var_14932_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14932_cast")];
            tensor<int32, [4]> var_14936_begin_0 = const()[name = tensor<string, []>("op_14936_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_14936_end_0 = const()[name = tensor<string, []>("op_14936_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_14936_end_mask_0 = const()[name = tensor<string, []>("op_14936_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14936_cast = slice_by_index(begin = var_14936_begin_0, end = var_14936_end_0, end_mask = var_14936_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14936_cast")];
            tensor<int32, [4]> var_14940_begin_0 = const()[name = tensor<string, []>("op_14940_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_14940_end_0 = const()[name = tensor<string, []>("op_14940_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_14940_end_mask_0 = const()[name = tensor<string, []>("op_14940_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14940_cast = slice_by_index(begin = var_14940_begin_0, end = var_14940_end_0, end_mask = var_14940_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14940_cast")];
            tensor<int32, [4]> var_14944_begin_0 = const()[name = tensor<string, []>("op_14944_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_14944_end_0 = const()[name = tensor<string, []>("op_14944_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_14944_end_mask_0 = const()[name = tensor<string, []>("op_14944_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14944_cast = slice_by_index(begin = var_14944_begin_0, end = var_14944_end_0, end_mask = var_14944_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14944_cast")];
            tensor<int32, [4]> var_14948_begin_0 = const()[name = tensor<string, []>("op_14948_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_14948_end_0 = const()[name = tensor<string, []>("op_14948_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_14948_end_mask_0 = const()[name = tensor<string, []>("op_14948_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_14948_cast = slice_by_index(begin = var_14948_begin_0, end = var_14948_end_0, end_mask = var_14948_end_mask_0, x = v_69_cast)[name = tensor<string, []>("op_14948_cast")];
            tensor<string, []> var_14952_equation_0 = const()[name = tensor<string, []>("op_14952_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14952_cast = einsum(equation = var_14952_equation_0, values = (var_14794_cast, var_14711_cast))[name = tensor<string, []>("op_14952_cast")];
            tensor<fp16, []> var_14953_to_fp16 = const()[name = tensor<string, []>("op_14953_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1201_cast = mul(x = var_14952_cast, y = var_14953_to_fp16)[name = tensor<string, []>("aw_1201_cast")];
            tensor<string, []> var_14956_equation_0 = const()[name = tensor<string, []>("op_14956_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14956_cast = einsum(equation = var_14956_equation_0, values = (var_14798_cast, var_14715_cast))[name = tensor<string, []>("op_14956_cast")];
            tensor<fp16, []> var_14957_to_fp16 = const()[name = tensor<string, []>("op_14957_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1203_cast = mul(x = var_14956_cast, y = var_14957_to_fp16)[name = tensor<string, []>("aw_1203_cast")];
            tensor<string, []> var_14960_equation_0 = const()[name = tensor<string, []>("op_14960_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14960_cast = einsum(equation = var_14960_equation_0, values = (var_14802_cast, var_14719_cast))[name = tensor<string, []>("op_14960_cast")];
            tensor<fp16, []> var_14961_to_fp16 = const()[name = tensor<string, []>("op_14961_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1205_cast = mul(x = var_14960_cast, y = var_14961_to_fp16)[name = tensor<string, []>("aw_1205_cast")];
            tensor<string, []> var_14964_equation_0 = const()[name = tensor<string, []>("op_14964_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14964_cast = einsum(equation = var_14964_equation_0, values = (var_14806_cast, var_14723_cast))[name = tensor<string, []>("op_14964_cast")];
            tensor<fp16, []> var_14965_to_fp16 = const()[name = tensor<string, []>("op_14965_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1207_cast = mul(x = var_14964_cast, y = var_14965_to_fp16)[name = tensor<string, []>("aw_1207_cast")];
            tensor<string, []> var_14968_equation_0 = const()[name = tensor<string, []>("op_14968_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14968_cast = einsum(equation = var_14968_equation_0, values = (var_14810_cast, var_14727_cast))[name = tensor<string, []>("op_14968_cast")];
            tensor<fp16, []> var_14969_to_fp16 = const()[name = tensor<string, []>("op_14969_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1209_cast = mul(x = var_14968_cast, y = var_14969_to_fp16)[name = tensor<string, []>("aw_1209_cast")];
            tensor<string, []> var_14972_equation_0 = const()[name = tensor<string, []>("op_14972_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14972_cast = einsum(equation = var_14972_equation_0, values = (var_14814_cast, var_14731_cast))[name = tensor<string, []>("op_14972_cast")];
            tensor<fp16, []> var_14973_to_fp16 = const()[name = tensor<string, []>("op_14973_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1211_cast = mul(x = var_14972_cast, y = var_14973_to_fp16)[name = tensor<string, []>("aw_1211_cast")];
            tensor<string, []> var_14976_equation_0 = const()[name = tensor<string, []>("op_14976_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14976_cast = einsum(equation = var_14976_equation_0, values = (var_14818_cast, var_14735_cast))[name = tensor<string, []>("op_14976_cast")];
            tensor<fp16, []> var_14977_to_fp16 = const()[name = tensor<string, []>("op_14977_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1213_cast = mul(x = var_14976_cast, y = var_14977_to_fp16)[name = tensor<string, []>("aw_1213_cast")];
            tensor<string, []> var_14980_equation_0 = const()[name = tensor<string, []>("op_14980_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14980_cast = einsum(equation = var_14980_equation_0, values = (var_14822_cast, var_14739_cast))[name = tensor<string, []>("op_14980_cast")];
            tensor<fp16, []> var_14981_to_fp16 = const()[name = tensor<string, []>("op_14981_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1215_cast = mul(x = var_14980_cast, y = var_14981_to_fp16)[name = tensor<string, []>("aw_1215_cast")];
            tensor<string, []> var_14984_equation_0 = const()[name = tensor<string, []>("op_14984_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14984_cast = einsum(equation = var_14984_equation_0, values = (var_14826_cast, var_14743_cast))[name = tensor<string, []>("op_14984_cast")];
            tensor<fp16, []> var_14985_to_fp16 = const()[name = tensor<string, []>("op_14985_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1217_cast = mul(x = var_14984_cast, y = var_14985_to_fp16)[name = tensor<string, []>("aw_1217_cast")];
            tensor<string, []> var_14988_equation_0 = const()[name = tensor<string, []>("op_14988_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14988_cast = einsum(equation = var_14988_equation_0, values = (var_14830_cast, var_14747_cast))[name = tensor<string, []>("op_14988_cast")];
            tensor<fp16, []> var_14989_to_fp16 = const()[name = tensor<string, []>("op_14989_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1219_cast = mul(x = var_14988_cast, y = var_14989_to_fp16)[name = tensor<string, []>("aw_1219_cast")];
            tensor<string, []> var_14992_equation_0 = const()[name = tensor<string, []>("op_14992_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14992_cast = einsum(equation = var_14992_equation_0, values = (var_14834_cast, var_14751_cast))[name = tensor<string, []>("op_14992_cast")];
            tensor<fp16, []> var_14993_to_fp16 = const()[name = tensor<string, []>("op_14993_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1221_cast = mul(x = var_14992_cast, y = var_14993_to_fp16)[name = tensor<string, []>("aw_1221_cast")];
            tensor<string, []> var_14996_equation_0 = const()[name = tensor<string, []>("op_14996_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_14996_cast = einsum(equation = var_14996_equation_0, values = (var_14838_cast, var_14755_cast))[name = tensor<string, []>("op_14996_cast")];
            tensor<fp16, []> var_14997_to_fp16 = const()[name = tensor<string, []>("op_14997_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1223_cast = mul(x = var_14996_cast, y = var_14997_to_fp16)[name = tensor<string, []>("aw_1223_cast")];
            tensor<string, []> var_15000_equation_0 = const()[name = tensor<string, []>("op_15000_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15000_cast = einsum(equation = var_15000_equation_0, values = (var_14842_cast, var_14759_cast))[name = tensor<string, []>("op_15000_cast")];
            tensor<fp16, []> var_15001_to_fp16 = const()[name = tensor<string, []>("op_15001_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1225_cast = mul(x = var_15000_cast, y = var_15001_to_fp16)[name = tensor<string, []>("aw_1225_cast")];
            tensor<string, []> var_15004_equation_0 = const()[name = tensor<string, []>("op_15004_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15004_cast = einsum(equation = var_15004_equation_0, values = (var_14846_cast, var_14763_cast))[name = tensor<string, []>("op_15004_cast")];
            tensor<fp16, []> var_15005_to_fp16 = const()[name = tensor<string, []>("op_15005_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1227_cast = mul(x = var_15004_cast, y = var_15005_to_fp16)[name = tensor<string, []>("aw_1227_cast")];
            tensor<string, []> var_15008_equation_0 = const()[name = tensor<string, []>("op_15008_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15008_cast = einsum(equation = var_15008_equation_0, values = (var_14850_cast, var_14767_cast))[name = tensor<string, []>("op_15008_cast")];
            tensor<fp16, []> var_15009_to_fp16 = const()[name = tensor<string, []>("op_15009_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1229_cast = mul(x = var_15008_cast, y = var_15009_to_fp16)[name = tensor<string, []>("aw_1229_cast")];
            tensor<string, []> var_15012_equation_0 = const()[name = tensor<string, []>("op_15012_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15012_cast = einsum(equation = var_15012_equation_0, values = (var_14854_cast, var_14771_cast))[name = tensor<string, []>("op_15012_cast")];
            tensor<fp16, []> var_15013_to_fp16 = const()[name = tensor<string, []>("op_15013_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1231_cast = mul(x = var_15012_cast, y = var_15013_to_fp16)[name = tensor<string, []>("aw_1231_cast")];
            tensor<string, []> var_15016_equation_0 = const()[name = tensor<string, []>("op_15016_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15016_cast = einsum(equation = var_15016_equation_0, values = (var_14858_cast, var_14775_cast))[name = tensor<string, []>("op_15016_cast")];
            tensor<fp16, []> var_15017_to_fp16 = const()[name = tensor<string, []>("op_15017_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1233_cast = mul(x = var_15016_cast, y = var_15017_to_fp16)[name = tensor<string, []>("aw_1233_cast")];
            tensor<string, []> var_15020_equation_0 = const()[name = tensor<string, []>("op_15020_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15020_cast = einsum(equation = var_15020_equation_0, values = (var_14862_cast, var_14779_cast))[name = tensor<string, []>("op_15020_cast")];
            tensor<fp16, []> var_15021_to_fp16 = const()[name = tensor<string, []>("op_15021_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1235_cast = mul(x = var_15020_cast, y = var_15021_to_fp16)[name = tensor<string, []>("aw_1235_cast")];
            tensor<string, []> var_15024_equation_0 = const()[name = tensor<string, []>("op_15024_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15024_cast = einsum(equation = var_15024_equation_0, values = (var_14866_cast, var_14783_cast))[name = tensor<string, []>("op_15024_cast")];
            tensor<fp16, []> var_15025_to_fp16 = const()[name = tensor<string, []>("op_15025_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1237_cast = mul(x = var_15024_cast, y = var_15025_to_fp16)[name = tensor<string, []>("aw_1237_cast")];
            tensor<string, []> var_15028_equation_0 = const()[name = tensor<string, []>("op_15028_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15028_cast = einsum(equation = var_15028_equation_0, values = (var_14870_cast, var_14787_cast))[name = tensor<string, []>("op_15028_cast")];
            tensor<fp16, []> var_15029_to_fp16 = const()[name = tensor<string, []>("op_15029_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1239_cast = mul(x = var_15028_cast, y = var_15029_to_fp16)[name = tensor<string, []>("aw_1239_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15031_cast = softmax(axis = var_2634, x = aw_1201_cast)[name = tensor<string, []>("op_15031_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15032_cast = softmax(axis = var_2634, x = aw_1203_cast)[name = tensor<string, []>("op_15032_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15033_cast = softmax(axis = var_2634, x = aw_1205_cast)[name = tensor<string, []>("op_15033_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15034_cast = softmax(axis = var_2634, x = aw_1207_cast)[name = tensor<string, []>("op_15034_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15035_cast = softmax(axis = var_2634, x = aw_1209_cast)[name = tensor<string, []>("op_15035_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15036_cast = softmax(axis = var_2634, x = aw_1211_cast)[name = tensor<string, []>("op_15036_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15037_cast = softmax(axis = var_2634, x = aw_1213_cast)[name = tensor<string, []>("op_15037_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15038_cast = softmax(axis = var_2634, x = aw_1215_cast)[name = tensor<string, []>("op_15038_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15039_cast = softmax(axis = var_2634, x = aw_1217_cast)[name = tensor<string, []>("op_15039_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15040_cast = softmax(axis = var_2634, x = aw_1219_cast)[name = tensor<string, []>("op_15040_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15041_cast = softmax(axis = var_2634, x = aw_1221_cast)[name = tensor<string, []>("op_15041_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15042_cast = softmax(axis = var_2634, x = aw_1223_cast)[name = tensor<string, []>("op_15042_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15043_cast = softmax(axis = var_2634, x = aw_1225_cast)[name = tensor<string, []>("op_15043_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15044_cast = softmax(axis = var_2634, x = aw_1227_cast)[name = tensor<string, []>("op_15044_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15045_cast = softmax(axis = var_2634, x = aw_1229_cast)[name = tensor<string, []>("op_15045_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15046_cast = softmax(axis = var_2634, x = aw_1231_cast)[name = tensor<string, []>("op_15046_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15047_cast = softmax(axis = var_2634, x = aw_1233_cast)[name = tensor<string, []>("op_15047_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15048_cast = softmax(axis = var_2634, x = aw_1235_cast)[name = tensor<string, []>("op_15048_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15049_cast = softmax(axis = var_2634, x = aw_1237_cast)[name = tensor<string, []>("op_15049_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15050_cast = softmax(axis = var_2634, x = aw_1239_cast)[name = tensor<string, []>("op_15050_cast")];
            tensor<string, []> var_15052_equation_0 = const()[name = tensor<string, []>("op_15052_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15052_cast = einsum(equation = var_15052_equation_0, values = (var_14872_cast, var_15031_cast))[name = tensor<string, []>("op_15052_cast")];
            tensor<string, []> var_15054_equation_0 = const()[name = tensor<string, []>("op_15054_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15054_cast = einsum(equation = var_15054_equation_0, values = (var_14876_cast, var_15032_cast))[name = tensor<string, []>("op_15054_cast")];
            tensor<string, []> var_15056_equation_0 = const()[name = tensor<string, []>("op_15056_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15056_cast = einsum(equation = var_15056_equation_0, values = (var_14880_cast, var_15033_cast))[name = tensor<string, []>("op_15056_cast")];
            tensor<string, []> var_15058_equation_0 = const()[name = tensor<string, []>("op_15058_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15058_cast = einsum(equation = var_15058_equation_0, values = (var_14884_cast, var_15034_cast))[name = tensor<string, []>("op_15058_cast")];
            tensor<string, []> var_15060_equation_0 = const()[name = tensor<string, []>("op_15060_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15060_cast = einsum(equation = var_15060_equation_0, values = (var_14888_cast, var_15035_cast))[name = tensor<string, []>("op_15060_cast")];
            tensor<string, []> var_15062_equation_0 = const()[name = tensor<string, []>("op_15062_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15062_cast = einsum(equation = var_15062_equation_0, values = (var_14892_cast, var_15036_cast))[name = tensor<string, []>("op_15062_cast")];
            tensor<string, []> var_15064_equation_0 = const()[name = tensor<string, []>("op_15064_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15064_cast = einsum(equation = var_15064_equation_0, values = (var_14896_cast, var_15037_cast))[name = tensor<string, []>("op_15064_cast")];
            tensor<string, []> var_15066_equation_0 = const()[name = tensor<string, []>("op_15066_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15066_cast = einsum(equation = var_15066_equation_0, values = (var_14900_cast, var_15038_cast))[name = tensor<string, []>("op_15066_cast")];
            tensor<string, []> var_15068_equation_0 = const()[name = tensor<string, []>("op_15068_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15068_cast = einsum(equation = var_15068_equation_0, values = (var_14904_cast, var_15039_cast))[name = tensor<string, []>("op_15068_cast")];
            tensor<string, []> var_15070_equation_0 = const()[name = tensor<string, []>("op_15070_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15070_cast = einsum(equation = var_15070_equation_0, values = (var_14908_cast, var_15040_cast))[name = tensor<string, []>("op_15070_cast")];
            tensor<string, []> var_15072_equation_0 = const()[name = tensor<string, []>("op_15072_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15072_cast = einsum(equation = var_15072_equation_0, values = (var_14912_cast, var_15041_cast))[name = tensor<string, []>("op_15072_cast")];
            tensor<string, []> var_15074_equation_0 = const()[name = tensor<string, []>("op_15074_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15074_cast = einsum(equation = var_15074_equation_0, values = (var_14916_cast, var_15042_cast))[name = tensor<string, []>("op_15074_cast")];
            tensor<string, []> var_15076_equation_0 = const()[name = tensor<string, []>("op_15076_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15076_cast = einsum(equation = var_15076_equation_0, values = (var_14920_cast, var_15043_cast))[name = tensor<string, []>("op_15076_cast")];
            tensor<string, []> var_15078_equation_0 = const()[name = tensor<string, []>("op_15078_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15078_cast = einsum(equation = var_15078_equation_0, values = (var_14924_cast, var_15044_cast))[name = tensor<string, []>("op_15078_cast")];
            tensor<string, []> var_15080_equation_0 = const()[name = tensor<string, []>("op_15080_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15080_cast = einsum(equation = var_15080_equation_0, values = (var_14928_cast, var_15045_cast))[name = tensor<string, []>("op_15080_cast")];
            tensor<string, []> var_15082_equation_0 = const()[name = tensor<string, []>("op_15082_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15082_cast = einsum(equation = var_15082_equation_0, values = (var_14932_cast, var_15046_cast))[name = tensor<string, []>("op_15082_cast")];
            tensor<string, []> var_15084_equation_0 = const()[name = tensor<string, []>("op_15084_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15084_cast = einsum(equation = var_15084_equation_0, values = (var_14936_cast, var_15047_cast))[name = tensor<string, []>("op_15084_cast")];
            tensor<string, []> var_15086_equation_0 = const()[name = tensor<string, []>("op_15086_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15086_cast = einsum(equation = var_15086_equation_0, values = (var_14940_cast, var_15048_cast))[name = tensor<string, []>("op_15086_cast")];
            tensor<string, []> var_15088_equation_0 = const()[name = tensor<string, []>("op_15088_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15088_cast = einsum(equation = var_15088_equation_0, values = (var_14944_cast, var_15049_cast))[name = tensor<string, []>("op_15088_cast")];
            tensor<string, []> var_15090_equation_0 = const()[name = tensor<string, []>("op_15090_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15090_cast = einsum(equation = var_15090_equation_0, values = (var_14948_cast, var_15050_cast))[name = tensor<string, []>("op_15090_cast")];
            tensor<bool, []> input_257_interleave_0 = const()[name = tensor<string, []>("input_257_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_257_cast = concat(axis = var_2634, interleave = input_257_interleave_0, values = (var_15052_cast, var_15054_cast, var_15056_cast, var_15058_cast, var_15060_cast, var_15062_cast, var_15064_cast, var_15066_cast, var_15068_cast, var_15070_cast, var_15072_cast, var_15074_cast, var_15076_cast, var_15078_cast, var_15080_cast, var_15082_cast, var_15084_cast, var_15086_cast, var_15088_cast, var_15090_cast))[name = tensor<string, []>("input_257_cast")];
            tensor<int32, [2]> var_15096 = const()[name = tensor<string, []>("op_15096"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_15098 = const()[name = tensor<string, []>("op_15098"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_15100_pad_type_0 = const()[name = tensor<string, []>("op_15100_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_15100_pad_0 = const()[name = tensor<string, []>("op_15100_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1180512192)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1183789056)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_15100_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_out_0_bias_to_fp16, dilations = var_15098, groups = var_2634, pad = var_15100_pad_0, pad_type = var_15100_pad_type_0, strides = var_15096, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn1_to_out_0_weight_to_fp16, x = input_257_cast)[name = tensor<string, []>("op_15100_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_105_cast = add(x = var_15100_cast, y = inputs_103_cast)[name = tensor<string, []>("inputs_105_cast")];
            tensor<int32, [1]> var_15104 = const()[name = tensor<string, []>("op_15104"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_105_cast = reduce_mean(axes = var_15104, keep_dims = var_2629, x = inputs_105_cast)[name = tensor<string, []>("channels_mean_105_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_105_cast = sub(x = inputs_105_cast, y = channels_mean_105_cast)[name = tensor<string, []>("zero_mean_105_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_105_cast = mul(x = zero_mean_105_cast, y = zero_mean_105_cast)[name = tensor<string, []>("zero_mean_sq_105_cast")];
            tensor<int32, [1]> var_15108 = const()[name = tensor<string, []>("op_15108"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_15109_cast = reduce_mean(axes = var_15108, keep_dims = var_2629, x = zero_mean_sq_105_cast)[name = tensor<string, []>("op_15109_cast")];
            tensor<fp16, []> var_15110_to_fp16 = const()[name = tensor<string, []>("op_15110_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_15111_cast = add(x = var_15109_cast, y = var_15110_to_fp16)[name = tensor<string, []>("op_15111_cast")];
            tensor<fp16, []> denom_105_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_105_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_105_cast = rsqrt(epsilon = denom_105_epsilon_0_to_fp16, x = var_15111_cast)[name = tensor<string, []>("denom_105_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_105_cast = mul(x = zero_mean_105_cast, y = denom_105_cast)[name = tensor<string, []>("out_105_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_15115_to_fp16 = const()[name = tensor<string, []>("op_15115_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1183791680)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_15116_cast = add(x = out_105_cast, y = var_15115_to_fp16)[name = tensor<string, []>("op_15116_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_15118_to_fp16 = const()[name = tensor<string, []>("op_15118_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1183794304)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_157_cast = mul(x = var_15116_cast, y = var_15118_to_fp16)[name = tensor<string, []>("hidden_states_157_cast")];
            tensor<int32, [2]> var_15125 = const()[name = tensor<string, []>("op_15125"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_15127 = const()[name = tensor<string, []>("op_15127"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_71_pad_type_0 = const()[name = tensor<string, []>("q_71_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_71_pad_0 = const()[name = tensor<string, []>("q_71_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1183796928)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_71_cast = conv(dilations = var_15127, groups = var_2634, pad = q_71_pad_0, pad_type = q_71_pad_type_0, strides = var_15125, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_q_weight_to_fp16, x = hidden_states_157_cast)[name = tensor<string, []>("q_71_cast")];
            tensor<int32, [2]> var_15131 = const()[name = tensor<string, []>("op_15131"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_15133 = const()[name = tensor<string, []>("op_15133"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_141_pad_type_0 = const()[name = tensor<string, []>("k_141_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_141_pad_0 = const()[name = tensor<string, []>("k_141_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1187073792)))];
            tensor<fp16, [2, 1280, 1, 77]> k_141_cast = conv(dilations = var_15133, groups = var_2634, pad = k_141_pad_0, pad_type = k_141_pad_type_0, strides = var_15131, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_141_cast")];
            tensor<int32, [2]> var_15137 = const()[name = tensor<string, []>("op_15137"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_15139 = const()[name = tensor<string, []>("op_15139"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_71_pad_type_0 = const()[name = tensor<string, []>("v_71_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_71_pad_0 = const()[name = tensor<string, []>("v_71_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1192316736)))];
            tensor<fp16, [2, 1280, 1, 77]> v_71_cast = conv(dilations = var_15139, groups = var_2634, pad = v_71_pad_0, pad_type = v_71_pad_type_0, strides = var_15137, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_71_cast")];
            tensor<int32, [4]> var_15143_begin_0 = const()[name = tensor<string, []>("op_15143_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_15143_end_0 = const()[name = tensor<string, []>("op_15143_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_15143_end_mask_0 = const()[name = tensor<string, []>("op_15143_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15143_cast = slice_by_index(begin = var_15143_begin_0, end = var_15143_end_0, end_mask = var_15143_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15143_cast")];
            tensor<int32, [4]> var_15147_begin_0 = const()[name = tensor<string, []>("op_15147_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_15147_end_0 = const()[name = tensor<string, []>("op_15147_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_15147_end_mask_0 = const()[name = tensor<string, []>("op_15147_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15147_cast = slice_by_index(begin = var_15147_begin_0, end = var_15147_end_0, end_mask = var_15147_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15147_cast")];
            tensor<int32, [4]> var_15151_begin_0 = const()[name = tensor<string, []>("op_15151_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_15151_end_0 = const()[name = tensor<string, []>("op_15151_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_15151_end_mask_0 = const()[name = tensor<string, []>("op_15151_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15151_cast = slice_by_index(begin = var_15151_begin_0, end = var_15151_end_0, end_mask = var_15151_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15151_cast")];
            tensor<int32, [4]> var_15155_begin_0 = const()[name = tensor<string, []>("op_15155_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_15155_end_0 = const()[name = tensor<string, []>("op_15155_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_15155_end_mask_0 = const()[name = tensor<string, []>("op_15155_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15155_cast = slice_by_index(begin = var_15155_begin_0, end = var_15155_end_0, end_mask = var_15155_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15155_cast")];
            tensor<int32, [4]> var_15159_begin_0 = const()[name = tensor<string, []>("op_15159_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_15159_end_0 = const()[name = tensor<string, []>("op_15159_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_15159_end_mask_0 = const()[name = tensor<string, []>("op_15159_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15159_cast = slice_by_index(begin = var_15159_begin_0, end = var_15159_end_0, end_mask = var_15159_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15159_cast")];
            tensor<int32, [4]> var_15163_begin_0 = const()[name = tensor<string, []>("op_15163_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_15163_end_0 = const()[name = tensor<string, []>("op_15163_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_15163_end_mask_0 = const()[name = tensor<string, []>("op_15163_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15163_cast = slice_by_index(begin = var_15163_begin_0, end = var_15163_end_0, end_mask = var_15163_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15163_cast")];
            tensor<int32, [4]> var_15167_begin_0 = const()[name = tensor<string, []>("op_15167_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_15167_end_0 = const()[name = tensor<string, []>("op_15167_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_15167_end_mask_0 = const()[name = tensor<string, []>("op_15167_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15167_cast = slice_by_index(begin = var_15167_begin_0, end = var_15167_end_0, end_mask = var_15167_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15167_cast")];
            tensor<int32, [4]> var_15171_begin_0 = const()[name = tensor<string, []>("op_15171_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_15171_end_0 = const()[name = tensor<string, []>("op_15171_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_15171_end_mask_0 = const()[name = tensor<string, []>("op_15171_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15171_cast = slice_by_index(begin = var_15171_begin_0, end = var_15171_end_0, end_mask = var_15171_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15171_cast")];
            tensor<int32, [4]> var_15175_begin_0 = const()[name = tensor<string, []>("op_15175_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_15175_end_0 = const()[name = tensor<string, []>("op_15175_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_15175_end_mask_0 = const()[name = tensor<string, []>("op_15175_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15175_cast = slice_by_index(begin = var_15175_begin_0, end = var_15175_end_0, end_mask = var_15175_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15175_cast")];
            tensor<int32, [4]> var_15179_begin_0 = const()[name = tensor<string, []>("op_15179_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_15179_end_0 = const()[name = tensor<string, []>("op_15179_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_15179_end_mask_0 = const()[name = tensor<string, []>("op_15179_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15179_cast = slice_by_index(begin = var_15179_begin_0, end = var_15179_end_0, end_mask = var_15179_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15179_cast")];
            tensor<int32, [4]> var_15183_begin_0 = const()[name = tensor<string, []>("op_15183_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_15183_end_0 = const()[name = tensor<string, []>("op_15183_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_15183_end_mask_0 = const()[name = tensor<string, []>("op_15183_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15183_cast = slice_by_index(begin = var_15183_begin_0, end = var_15183_end_0, end_mask = var_15183_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15183_cast")];
            tensor<int32, [4]> var_15187_begin_0 = const()[name = tensor<string, []>("op_15187_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_15187_end_0 = const()[name = tensor<string, []>("op_15187_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_15187_end_mask_0 = const()[name = tensor<string, []>("op_15187_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15187_cast = slice_by_index(begin = var_15187_begin_0, end = var_15187_end_0, end_mask = var_15187_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15187_cast")];
            tensor<int32, [4]> var_15191_begin_0 = const()[name = tensor<string, []>("op_15191_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_15191_end_0 = const()[name = tensor<string, []>("op_15191_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_15191_end_mask_0 = const()[name = tensor<string, []>("op_15191_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15191_cast = slice_by_index(begin = var_15191_begin_0, end = var_15191_end_0, end_mask = var_15191_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15191_cast")];
            tensor<int32, [4]> var_15195_begin_0 = const()[name = tensor<string, []>("op_15195_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_15195_end_0 = const()[name = tensor<string, []>("op_15195_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_15195_end_mask_0 = const()[name = tensor<string, []>("op_15195_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15195_cast = slice_by_index(begin = var_15195_begin_0, end = var_15195_end_0, end_mask = var_15195_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15195_cast")];
            tensor<int32, [4]> var_15199_begin_0 = const()[name = tensor<string, []>("op_15199_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_15199_end_0 = const()[name = tensor<string, []>("op_15199_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_15199_end_mask_0 = const()[name = tensor<string, []>("op_15199_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15199_cast = slice_by_index(begin = var_15199_begin_0, end = var_15199_end_0, end_mask = var_15199_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15199_cast")];
            tensor<int32, [4]> var_15203_begin_0 = const()[name = tensor<string, []>("op_15203_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_15203_end_0 = const()[name = tensor<string, []>("op_15203_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_15203_end_mask_0 = const()[name = tensor<string, []>("op_15203_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15203_cast = slice_by_index(begin = var_15203_begin_0, end = var_15203_end_0, end_mask = var_15203_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15203_cast")];
            tensor<int32, [4]> var_15207_begin_0 = const()[name = tensor<string, []>("op_15207_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_15207_end_0 = const()[name = tensor<string, []>("op_15207_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_15207_end_mask_0 = const()[name = tensor<string, []>("op_15207_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15207_cast = slice_by_index(begin = var_15207_begin_0, end = var_15207_end_0, end_mask = var_15207_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15207_cast")];
            tensor<int32, [4]> var_15211_begin_0 = const()[name = tensor<string, []>("op_15211_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_15211_end_0 = const()[name = tensor<string, []>("op_15211_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_15211_end_mask_0 = const()[name = tensor<string, []>("op_15211_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15211_cast = slice_by_index(begin = var_15211_begin_0, end = var_15211_end_0, end_mask = var_15211_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15211_cast")];
            tensor<int32, [4]> var_15215_begin_0 = const()[name = tensor<string, []>("op_15215_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_15215_end_0 = const()[name = tensor<string, []>("op_15215_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_15215_end_mask_0 = const()[name = tensor<string, []>("op_15215_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15215_cast = slice_by_index(begin = var_15215_begin_0, end = var_15215_end_0, end_mask = var_15215_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15215_cast")];
            tensor<int32, [4]> var_15219_begin_0 = const()[name = tensor<string, []>("op_15219_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_15219_end_0 = const()[name = tensor<string, []>("op_15219_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_15219_end_mask_0 = const()[name = tensor<string, []>("op_15219_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15219_cast = slice_by_index(begin = var_15219_begin_0, end = var_15219_end_0, end_mask = var_15219_end_mask_0, x = q_71_cast)[name = tensor<string, []>("op_15219_cast")];
            tensor<int32, [4]> k_143_perm_0 = const()[name = tensor<string, []>("k_143_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_15226_begin_0 = const()[name = tensor<string, []>("op_15226_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_15226_end_0 = const()[name = tensor<string, []>("op_15226_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_15226_end_mask_0 = const()[name = tensor<string, []>("op_15226_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_104 = transpose(perm = k_143_perm_0, x = k_141_cast)[name = tensor<string, []>("transpose_104")];
            tensor<fp16, [2, 77, 1, 64]> var_15226_cast = slice_by_index(begin = var_15226_begin_0, end = var_15226_end_0, end_mask = var_15226_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15226_cast")];
            tensor<int32, [4]> var_15230_begin_0 = const()[name = tensor<string, []>("op_15230_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_15230_end_0 = const()[name = tensor<string, []>("op_15230_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_15230_end_mask_0 = const()[name = tensor<string, []>("op_15230_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15230_cast = slice_by_index(begin = var_15230_begin_0, end = var_15230_end_0, end_mask = var_15230_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15230_cast")];
            tensor<int32, [4]> var_15234_begin_0 = const()[name = tensor<string, []>("op_15234_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_15234_end_0 = const()[name = tensor<string, []>("op_15234_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_15234_end_mask_0 = const()[name = tensor<string, []>("op_15234_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15234_cast = slice_by_index(begin = var_15234_begin_0, end = var_15234_end_0, end_mask = var_15234_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15234_cast")];
            tensor<int32, [4]> var_15238_begin_0 = const()[name = tensor<string, []>("op_15238_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_15238_end_0 = const()[name = tensor<string, []>("op_15238_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_15238_end_mask_0 = const()[name = tensor<string, []>("op_15238_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15238_cast = slice_by_index(begin = var_15238_begin_0, end = var_15238_end_0, end_mask = var_15238_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15238_cast")];
            tensor<int32, [4]> var_15242_begin_0 = const()[name = tensor<string, []>("op_15242_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_15242_end_0 = const()[name = tensor<string, []>("op_15242_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_15242_end_mask_0 = const()[name = tensor<string, []>("op_15242_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15242_cast = slice_by_index(begin = var_15242_begin_0, end = var_15242_end_0, end_mask = var_15242_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15242_cast")];
            tensor<int32, [4]> var_15246_begin_0 = const()[name = tensor<string, []>("op_15246_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_15246_end_0 = const()[name = tensor<string, []>("op_15246_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_15246_end_mask_0 = const()[name = tensor<string, []>("op_15246_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15246_cast = slice_by_index(begin = var_15246_begin_0, end = var_15246_end_0, end_mask = var_15246_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15246_cast")];
            tensor<int32, [4]> var_15250_begin_0 = const()[name = tensor<string, []>("op_15250_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_15250_end_0 = const()[name = tensor<string, []>("op_15250_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_15250_end_mask_0 = const()[name = tensor<string, []>("op_15250_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15250_cast = slice_by_index(begin = var_15250_begin_0, end = var_15250_end_0, end_mask = var_15250_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15250_cast")];
            tensor<int32, [4]> var_15254_begin_0 = const()[name = tensor<string, []>("op_15254_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_15254_end_0 = const()[name = tensor<string, []>("op_15254_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_15254_end_mask_0 = const()[name = tensor<string, []>("op_15254_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15254_cast = slice_by_index(begin = var_15254_begin_0, end = var_15254_end_0, end_mask = var_15254_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15254_cast")];
            tensor<int32, [4]> var_15258_begin_0 = const()[name = tensor<string, []>("op_15258_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_15258_end_0 = const()[name = tensor<string, []>("op_15258_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_15258_end_mask_0 = const()[name = tensor<string, []>("op_15258_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15258_cast = slice_by_index(begin = var_15258_begin_0, end = var_15258_end_0, end_mask = var_15258_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15258_cast")];
            tensor<int32, [4]> var_15262_begin_0 = const()[name = tensor<string, []>("op_15262_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_15262_end_0 = const()[name = tensor<string, []>("op_15262_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_15262_end_mask_0 = const()[name = tensor<string, []>("op_15262_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15262_cast = slice_by_index(begin = var_15262_begin_0, end = var_15262_end_0, end_mask = var_15262_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15262_cast")];
            tensor<int32, [4]> var_15266_begin_0 = const()[name = tensor<string, []>("op_15266_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_15266_end_0 = const()[name = tensor<string, []>("op_15266_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_15266_end_mask_0 = const()[name = tensor<string, []>("op_15266_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15266_cast = slice_by_index(begin = var_15266_begin_0, end = var_15266_end_0, end_mask = var_15266_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15266_cast")];
            tensor<int32, [4]> var_15270_begin_0 = const()[name = tensor<string, []>("op_15270_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_15270_end_0 = const()[name = tensor<string, []>("op_15270_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_15270_end_mask_0 = const()[name = tensor<string, []>("op_15270_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15270_cast = slice_by_index(begin = var_15270_begin_0, end = var_15270_end_0, end_mask = var_15270_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15270_cast")];
            tensor<int32, [4]> var_15274_begin_0 = const()[name = tensor<string, []>("op_15274_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_15274_end_0 = const()[name = tensor<string, []>("op_15274_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_15274_end_mask_0 = const()[name = tensor<string, []>("op_15274_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15274_cast = slice_by_index(begin = var_15274_begin_0, end = var_15274_end_0, end_mask = var_15274_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15274_cast")];
            tensor<int32, [4]> var_15278_begin_0 = const()[name = tensor<string, []>("op_15278_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_15278_end_0 = const()[name = tensor<string, []>("op_15278_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_15278_end_mask_0 = const()[name = tensor<string, []>("op_15278_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15278_cast = slice_by_index(begin = var_15278_begin_0, end = var_15278_end_0, end_mask = var_15278_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15278_cast")];
            tensor<int32, [4]> var_15282_begin_0 = const()[name = tensor<string, []>("op_15282_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_15282_end_0 = const()[name = tensor<string, []>("op_15282_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_15282_end_mask_0 = const()[name = tensor<string, []>("op_15282_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15282_cast = slice_by_index(begin = var_15282_begin_0, end = var_15282_end_0, end_mask = var_15282_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15282_cast")];
            tensor<int32, [4]> var_15286_begin_0 = const()[name = tensor<string, []>("op_15286_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_15286_end_0 = const()[name = tensor<string, []>("op_15286_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_15286_end_mask_0 = const()[name = tensor<string, []>("op_15286_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15286_cast = slice_by_index(begin = var_15286_begin_0, end = var_15286_end_0, end_mask = var_15286_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15286_cast")];
            tensor<int32, [4]> var_15290_begin_0 = const()[name = tensor<string, []>("op_15290_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_15290_end_0 = const()[name = tensor<string, []>("op_15290_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_15290_end_mask_0 = const()[name = tensor<string, []>("op_15290_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15290_cast = slice_by_index(begin = var_15290_begin_0, end = var_15290_end_0, end_mask = var_15290_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15290_cast")];
            tensor<int32, [4]> var_15294_begin_0 = const()[name = tensor<string, []>("op_15294_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_15294_end_0 = const()[name = tensor<string, []>("op_15294_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_15294_end_mask_0 = const()[name = tensor<string, []>("op_15294_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15294_cast = slice_by_index(begin = var_15294_begin_0, end = var_15294_end_0, end_mask = var_15294_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15294_cast")];
            tensor<int32, [4]> var_15298_begin_0 = const()[name = tensor<string, []>("op_15298_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_15298_end_0 = const()[name = tensor<string, []>("op_15298_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_15298_end_mask_0 = const()[name = tensor<string, []>("op_15298_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15298_cast = slice_by_index(begin = var_15298_begin_0, end = var_15298_end_0, end_mask = var_15298_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15298_cast")];
            tensor<int32, [4]> var_15302_begin_0 = const()[name = tensor<string, []>("op_15302_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_15302_end_0 = const()[name = tensor<string, []>("op_15302_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_15302_end_mask_0 = const()[name = tensor<string, []>("op_15302_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15302_cast = slice_by_index(begin = var_15302_begin_0, end = var_15302_end_0, end_mask = var_15302_end_mask_0, x = transpose_104)[name = tensor<string, []>("op_15302_cast")];
            tensor<int32, [4]> var_15304_begin_0 = const()[name = tensor<string, []>("op_15304_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_15304_end_0 = const()[name = tensor<string, []>("op_15304_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_15304_end_mask_0 = const()[name = tensor<string, []>("op_15304_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15304_cast = slice_by_index(begin = var_15304_begin_0, end = var_15304_end_0, end_mask = var_15304_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15304_cast")];
            tensor<int32, [4]> var_15308_begin_0 = const()[name = tensor<string, []>("op_15308_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_15308_end_0 = const()[name = tensor<string, []>("op_15308_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_15308_end_mask_0 = const()[name = tensor<string, []>("op_15308_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15308_cast = slice_by_index(begin = var_15308_begin_0, end = var_15308_end_0, end_mask = var_15308_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15308_cast")];
            tensor<int32, [4]> var_15312_begin_0 = const()[name = tensor<string, []>("op_15312_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_15312_end_0 = const()[name = tensor<string, []>("op_15312_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_15312_end_mask_0 = const()[name = tensor<string, []>("op_15312_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15312_cast = slice_by_index(begin = var_15312_begin_0, end = var_15312_end_0, end_mask = var_15312_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15312_cast")];
            tensor<int32, [4]> var_15316_begin_0 = const()[name = tensor<string, []>("op_15316_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_15316_end_0 = const()[name = tensor<string, []>("op_15316_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_15316_end_mask_0 = const()[name = tensor<string, []>("op_15316_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15316_cast = slice_by_index(begin = var_15316_begin_0, end = var_15316_end_0, end_mask = var_15316_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15316_cast")];
            tensor<int32, [4]> var_15320_begin_0 = const()[name = tensor<string, []>("op_15320_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_15320_end_0 = const()[name = tensor<string, []>("op_15320_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_15320_end_mask_0 = const()[name = tensor<string, []>("op_15320_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15320_cast = slice_by_index(begin = var_15320_begin_0, end = var_15320_end_0, end_mask = var_15320_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15320_cast")];
            tensor<int32, [4]> var_15324_begin_0 = const()[name = tensor<string, []>("op_15324_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_15324_end_0 = const()[name = tensor<string, []>("op_15324_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_15324_end_mask_0 = const()[name = tensor<string, []>("op_15324_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15324_cast = slice_by_index(begin = var_15324_begin_0, end = var_15324_end_0, end_mask = var_15324_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15324_cast")];
            tensor<int32, [4]> var_15328_begin_0 = const()[name = tensor<string, []>("op_15328_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_15328_end_0 = const()[name = tensor<string, []>("op_15328_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_15328_end_mask_0 = const()[name = tensor<string, []>("op_15328_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15328_cast = slice_by_index(begin = var_15328_begin_0, end = var_15328_end_0, end_mask = var_15328_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15328_cast")];
            tensor<int32, [4]> var_15332_begin_0 = const()[name = tensor<string, []>("op_15332_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_15332_end_0 = const()[name = tensor<string, []>("op_15332_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_15332_end_mask_0 = const()[name = tensor<string, []>("op_15332_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15332_cast = slice_by_index(begin = var_15332_begin_0, end = var_15332_end_0, end_mask = var_15332_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15332_cast")];
            tensor<int32, [4]> var_15336_begin_0 = const()[name = tensor<string, []>("op_15336_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_15336_end_0 = const()[name = tensor<string, []>("op_15336_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_15336_end_mask_0 = const()[name = tensor<string, []>("op_15336_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15336_cast = slice_by_index(begin = var_15336_begin_0, end = var_15336_end_0, end_mask = var_15336_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15336_cast")];
            tensor<int32, [4]> var_15340_begin_0 = const()[name = tensor<string, []>("op_15340_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_15340_end_0 = const()[name = tensor<string, []>("op_15340_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_15340_end_mask_0 = const()[name = tensor<string, []>("op_15340_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15340_cast = slice_by_index(begin = var_15340_begin_0, end = var_15340_end_0, end_mask = var_15340_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15340_cast")];
            tensor<int32, [4]> var_15344_begin_0 = const()[name = tensor<string, []>("op_15344_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_15344_end_0 = const()[name = tensor<string, []>("op_15344_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_15344_end_mask_0 = const()[name = tensor<string, []>("op_15344_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15344_cast = slice_by_index(begin = var_15344_begin_0, end = var_15344_end_0, end_mask = var_15344_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15344_cast")];
            tensor<int32, [4]> var_15348_begin_0 = const()[name = tensor<string, []>("op_15348_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_15348_end_0 = const()[name = tensor<string, []>("op_15348_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_15348_end_mask_0 = const()[name = tensor<string, []>("op_15348_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15348_cast = slice_by_index(begin = var_15348_begin_0, end = var_15348_end_0, end_mask = var_15348_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15348_cast")];
            tensor<int32, [4]> var_15352_begin_0 = const()[name = tensor<string, []>("op_15352_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_15352_end_0 = const()[name = tensor<string, []>("op_15352_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_15352_end_mask_0 = const()[name = tensor<string, []>("op_15352_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15352_cast = slice_by_index(begin = var_15352_begin_0, end = var_15352_end_0, end_mask = var_15352_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15352_cast")];
            tensor<int32, [4]> var_15356_begin_0 = const()[name = tensor<string, []>("op_15356_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_15356_end_0 = const()[name = tensor<string, []>("op_15356_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_15356_end_mask_0 = const()[name = tensor<string, []>("op_15356_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15356_cast = slice_by_index(begin = var_15356_begin_0, end = var_15356_end_0, end_mask = var_15356_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15356_cast")];
            tensor<int32, [4]> var_15360_begin_0 = const()[name = tensor<string, []>("op_15360_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_15360_end_0 = const()[name = tensor<string, []>("op_15360_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_15360_end_mask_0 = const()[name = tensor<string, []>("op_15360_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15360_cast = slice_by_index(begin = var_15360_begin_0, end = var_15360_end_0, end_mask = var_15360_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15360_cast")];
            tensor<int32, [4]> var_15364_begin_0 = const()[name = tensor<string, []>("op_15364_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_15364_end_0 = const()[name = tensor<string, []>("op_15364_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_15364_end_mask_0 = const()[name = tensor<string, []>("op_15364_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15364_cast = slice_by_index(begin = var_15364_begin_0, end = var_15364_end_0, end_mask = var_15364_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15364_cast")];
            tensor<int32, [4]> var_15368_begin_0 = const()[name = tensor<string, []>("op_15368_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_15368_end_0 = const()[name = tensor<string, []>("op_15368_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_15368_end_mask_0 = const()[name = tensor<string, []>("op_15368_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15368_cast = slice_by_index(begin = var_15368_begin_0, end = var_15368_end_0, end_mask = var_15368_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15368_cast")];
            tensor<int32, [4]> var_15372_begin_0 = const()[name = tensor<string, []>("op_15372_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_15372_end_0 = const()[name = tensor<string, []>("op_15372_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_15372_end_mask_0 = const()[name = tensor<string, []>("op_15372_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15372_cast = slice_by_index(begin = var_15372_begin_0, end = var_15372_end_0, end_mask = var_15372_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15372_cast")];
            tensor<int32, [4]> var_15376_begin_0 = const()[name = tensor<string, []>("op_15376_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_15376_end_0 = const()[name = tensor<string, []>("op_15376_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_15376_end_mask_0 = const()[name = tensor<string, []>("op_15376_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15376_cast = slice_by_index(begin = var_15376_begin_0, end = var_15376_end_0, end_mask = var_15376_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15376_cast")];
            tensor<int32, [4]> var_15380_begin_0 = const()[name = tensor<string, []>("op_15380_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_15380_end_0 = const()[name = tensor<string, []>("op_15380_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_15380_end_mask_0 = const()[name = tensor<string, []>("op_15380_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15380_cast = slice_by_index(begin = var_15380_begin_0, end = var_15380_end_0, end_mask = var_15380_end_mask_0, x = v_71_cast)[name = tensor<string, []>("op_15380_cast")];
            tensor<string, []> var_15384_equation_0 = const()[name = tensor<string, []>("op_15384_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15384_cast = einsum(equation = var_15384_equation_0, values = (var_15226_cast, var_15143_cast))[name = tensor<string, []>("op_15384_cast")];
            tensor<fp16, []> var_15385_to_fp16 = const()[name = tensor<string, []>("op_15385_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1241_cast = mul(x = var_15384_cast, y = var_15385_to_fp16)[name = tensor<string, []>("aw_1241_cast")];
            tensor<string, []> var_15388_equation_0 = const()[name = tensor<string, []>("op_15388_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15388_cast = einsum(equation = var_15388_equation_0, values = (var_15230_cast, var_15147_cast))[name = tensor<string, []>("op_15388_cast")];
            tensor<fp16, []> var_15389_to_fp16 = const()[name = tensor<string, []>("op_15389_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1243_cast = mul(x = var_15388_cast, y = var_15389_to_fp16)[name = tensor<string, []>("aw_1243_cast")];
            tensor<string, []> var_15392_equation_0 = const()[name = tensor<string, []>("op_15392_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15392_cast = einsum(equation = var_15392_equation_0, values = (var_15234_cast, var_15151_cast))[name = tensor<string, []>("op_15392_cast")];
            tensor<fp16, []> var_15393_to_fp16 = const()[name = tensor<string, []>("op_15393_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1245_cast = mul(x = var_15392_cast, y = var_15393_to_fp16)[name = tensor<string, []>("aw_1245_cast")];
            tensor<string, []> var_15396_equation_0 = const()[name = tensor<string, []>("op_15396_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15396_cast = einsum(equation = var_15396_equation_0, values = (var_15238_cast, var_15155_cast))[name = tensor<string, []>("op_15396_cast")];
            tensor<fp16, []> var_15397_to_fp16 = const()[name = tensor<string, []>("op_15397_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1247_cast = mul(x = var_15396_cast, y = var_15397_to_fp16)[name = tensor<string, []>("aw_1247_cast")];
            tensor<string, []> var_15400_equation_0 = const()[name = tensor<string, []>("op_15400_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15400_cast = einsum(equation = var_15400_equation_0, values = (var_15242_cast, var_15159_cast))[name = tensor<string, []>("op_15400_cast")];
            tensor<fp16, []> var_15401_to_fp16 = const()[name = tensor<string, []>("op_15401_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1249_cast = mul(x = var_15400_cast, y = var_15401_to_fp16)[name = tensor<string, []>("aw_1249_cast")];
            tensor<string, []> var_15404_equation_0 = const()[name = tensor<string, []>("op_15404_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15404_cast = einsum(equation = var_15404_equation_0, values = (var_15246_cast, var_15163_cast))[name = tensor<string, []>("op_15404_cast")];
            tensor<fp16, []> var_15405_to_fp16 = const()[name = tensor<string, []>("op_15405_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1251_cast = mul(x = var_15404_cast, y = var_15405_to_fp16)[name = tensor<string, []>("aw_1251_cast")];
            tensor<string, []> var_15408_equation_0 = const()[name = tensor<string, []>("op_15408_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15408_cast = einsum(equation = var_15408_equation_0, values = (var_15250_cast, var_15167_cast))[name = tensor<string, []>("op_15408_cast")];
            tensor<fp16, []> var_15409_to_fp16 = const()[name = tensor<string, []>("op_15409_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1253_cast = mul(x = var_15408_cast, y = var_15409_to_fp16)[name = tensor<string, []>("aw_1253_cast")];
            tensor<string, []> var_15412_equation_0 = const()[name = tensor<string, []>("op_15412_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15412_cast = einsum(equation = var_15412_equation_0, values = (var_15254_cast, var_15171_cast))[name = tensor<string, []>("op_15412_cast")];
            tensor<fp16, []> var_15413_to_fp16 = const()[name = tensor<string, []>("op_15413_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1255_cast = mul(x = var_15412_cast, y = var_15413_to_fp16)[name = tensor<string, []>("aw_1255_cast")];
            tensor<string, []> var_15416_equation_0 = const()[name = tensor<string, []>("op_15416_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15416_cast = einsum(equation = var_15416_equation_0, values = (var_15258_cast, var_15175_cast))[name = tensor<string, []>("op_15416_cast")];
            tensor<fp16, []> var_15417_to_fp16 = const()[name = tensor<string, []>("op_15417_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1257_cast = mul(x = var_15416_cast, y = var_15417_to_fp16)[name = tensor<string, []>("aw_1257_cast")];
            tensor<string, []> var_15420_equation_0 = const()[name = tensor<string, []>("op_15420_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15420_cast = einsum(equation = var_15420_equation_0, values = (var_15262_cast, var_15179_cast))[name = tensor<string, []>("op_15420_cast")];
            tensor<fp16, []> var_15421_to_fp16 = const()[name = tensor<string, []>("op_15421_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1259_cast = mul(x = var_15420_cast, y = var_15421_to_fp16)[name = tensor<string, []>("aw_1259_cast")];
            tensor<string, []> var_15424_equation_0 = const()[name = tensor<string, []>("op_15424_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15424_cast = einsum(equation = var_15424_equation_0, values = (var_15266_cast, var_15183_cast))[name = tensor<string, []>("op_15424_cast")];
            tensor<fp16, []> var_15425_to_fp16 = const()[name = tensor<string, []>("op_15425_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1261_cast = mul(x = var_15424_cast, y = var_15425_to_fp16)[name = tensor<string, []>("aw_1261_cast")];
            tensor<string, []> var_15428_equation_0 = const()[name = tensor<string, []>("op_15428_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15428_cast = einsum(equation = var_15428_equation_0, values = (var_15270_cast, var_15187_cast))[name = tensor<string, []>("op_15428_cast")];
            tensor<fp16, []> var_15429_to_fp16 = const()[name = tensor<string, []>("op_15429_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1263_cast = mul(x = var_15428_cast, y = var_15429_to_fp16)[name = tensor<string, []>("aw_1263_cast")];
            tensor<string, []> var_15432_equation_0 = const()[name = tensor<string, []>("op_15432_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15432_cast = einsum(equation = var_15432_equation_0, values = (var_15274_cast, var_15191_cast))[name = tensor<string, []>("op_15432_cast")];
            tensor<fp16, []> var_15433_to_fp16 = const()[name = tensor<string, []>("op_15433_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1265_cast = mul(x = var_15432_cast, y = var_15433_to_fp16)[name = tensor<string, []>("aw_1265_cast")];
            tensor<string, []> var_15436_equation_0 = const()[name = tensor<string, []>("op_15436_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15436_cast = einsum(equation = var_15436_equation_0, values = (var_15278_cast, var_15195_cast))[name = tensor<string, []>("op_15436_cast")];
            tensor<fp16, []> var_15437_to_fp16 = const()[name = tensor<string, []>("op_15437_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1267_cast = mul(x = var_15436_cast, y = var_15437_to_fp16)[name = tensor<string, []>("aw_1267_cast")];
            tensor<string, []> var_15440_equation_0 = const()[name = tensor<string, []>("op_15440_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15440_cast = einsum(equation = var_15440_equation_0, values = (var_15282_cast, var_15199_cast))[name = tensor<string, []>("op_15440_cast")];
            tensor<fp16, []> var_15441_to_fp16 = const()[name = tensor<string, []>("op_15441_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1269_cast = mul(x = var_15440_cast, y = var_15441_to_fp16)[name = tensor<string, []>("aw_1269_cast")];
            tensor<string, []> var_15444_equation_0 = const()[name = tensor<string, []>("op_15444_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15444_cast = einsum(equation = var_15444_equation_0, values = (var_15286_cast, var_15203_cast))[name = tensor<string, []>("op_15444_cast")];
            tensor<fp16, []> var_15445_to_fp16 = const()[name = tensor<string, []>("op_15445_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1271_cast = mul(x = var_15444_cast, y = var_15445_to_fp16)[name = tensor<string, []>("aw_1271_cast")];
            tensor<string, []> var_15448_equation_0 = const()[name = tensor<string, []>("op_15448_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15448_cast = einsum(equation = var_15448_equation_0, values = (var_15290_cast, var_15207_cast))[name = tensor<string, []>("op_15448_cast")];
            tensor<fp16, []> var_15449_to_fp16 = const()[name = tensor<string, []>("op_15449_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1273_cast = mul(x = var_15448_cast, y = var_15449_to_fp16)[name = tensor<string, []>("aw_1273_cast")];
            tensor<string, []> var_15452_equation_0 = const()[name = tensor<string, []>("op_15452_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15452_cast = einsum(equation = var_15452_equation_0, values = (var_15294_cast, var_15211_cast))[name = tensor<string, []>("op_15452_cast")];
            tensor<fp16, []> var_15453_to_fp16 = const()[name = tensor<string, []>("op_15453_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1275_cast = mul(x = var_15452_cast, y = var_15453_to_fp16)[name = tensor<string, []>("aw_1275_cast")];
            tensor<string, []> var_15456_equation_0 = const()[name = tensor<string, []>("op_15456_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15456_cast = einsum(equation = var_15456_equation_0, values = (var_15298_cast, var_15215_cast))[name = tensor<string, []>("op_15456_cast")];
            tensor<fp16, []> var_15457_to_fp16 = const()[name = tensor<string, []>("op_15457_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1277_cast = mul(x = var_15456_cast, y = var_15457_to_fp16)[name = tensor<string, []>("aw_1277_cast")];
            tensor<string, []> var_15460_equation_0 = const()[name = tensor<string, []>("op_15460_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_15460_cast = einsum(equation = var_15460_equation_0, values = (var_15302_cast, var_15219_cast))[name = tensor<string, []>("op_15460_cast")];
            tensor<fp16, []> var_15461_to_fp16 = const()[name = tensor<string, []>("op_15461_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1279_cast = mul(x = var_15460_cast, y = var_15461_to_fp16)[name = tensor<string, []>("aw_1279_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15463_cast = softmax(axis = var_2634, x = aw_1241_cast)[name = tensor<string, []>("op_15463_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15464_cast = softmax(axis = var_2634, x = aw_1243_cast)[name = tensor<string, []>("op_15464_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15465_cast = softmax(axis = var_2634, x = aw_1245_cast)[name = tensor<string, []>("op_15465_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15466_cast = softmax(axis = var_2634, x = aw_1247_cast)[name = tensor<string, []>("op_15466_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15467_cast = softmax(axis = var_2634, x = aw_1249_cast)[name = tensor<string, []>("op_15467_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15468_cast = softmax(axis = var_2634, x = aw_1251_cast)[name = tensor<string, []>("op_15468_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15469_cast = softmax(axis = var_2634, x = aw_1253_cast)[name = tensor<string, []>("op_15469_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15470_cast = softmax(axis = var_2634, x = aw_1255_cast)[name = tensor<string, []>("op_15470_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15471_cast = softmax(axis = var_2634, x = aw_1257_cast)[name = tensor<string, []>("op_15471_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15472_cast = softmax(axis = var_2634, x = aw_1259_cast)[name = tensor<string, []>("op_15472_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15473_cast = softmax(axis = var_2634, x = aw_1261_cast)[name = tensor<string, []>("op_15473_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15474_cast = softmax(axis = var_2634, x = aw_1263_cast)[name = tensor<string, []>("op_15474_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15475_cast = softmax(axis = var_2634, x = aw_1265_cast)[name = tensor<string, []>("op_15475_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15476_cast = softmax(axis = var_2634, x = aw_1267_cast)[name = tensor<string, []>("op_15476_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15477_cast = softmax(axis = var_2634, x = aw_1269_cast)[name = tensor<string, []>("op_15477_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15478_cast = softmax(axis = var_2634, x = aw_1271_cast)[name = tensor<string, []>("op_15478_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15479_cast = softmax(axis = var_2634, x = aw_1273_cast)[name = tensor<string, []>("op_15479_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15480_cast = softmax(axis = var_2634, x = aw_1275_cast)[name = tensor<string, []>("op_15480_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15481_cast = softmax(axis = var_2634, x = aw_1277_cast)[name = tensor<string, []>("op_15481_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_15482_cast = softmax(axis = var_2634, x = aw_1279_cast)[name = tensor<string, []>("op_15482_cast")];
            tensor<string, []> var_15484_equation_0 = const()[name = tensor<string, []>("op_15484_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15484_cast = einsum(equation = var_15484_equation_0, values = (var_15304_cast, var_15463_cast))[name = tensor<string, []>("op_15484_cast")];
            tensor<string, []> var_15486_equation_0 = const()[name = tensor<string, []>("op_15486_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15486_cast = einsum(equation = var_15486_equation_0, values = (var_15308_cast, var_15464_cast))[name = tensor<string, []>("op_15486_cast")];
            tensor<string, []> var_15488_equation_0 = const()[name = tensor<string, []>("op_15488_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15488_cast = einsum(equation = var_15488_equation_0, values = (var_15312_cast, var_15465_cast))[name = tensor<string, []>("op_15488_cast")];
            tensor<string, []> var_15490_equation_0 = const()[name = tensor<string, []>("op_15490_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15490_cast = einsum(equation = var_15490_equation_0, values = (var_15316_cast, var_15466_cast))[name = tensor<string, []>("op_15490_cast")];
            tensor<string, []> var_15492_equation_0 = const()[name = tensor<string, []>("op_15492_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15492_cast = einsum(equation = var_15492_equation_0, values = (var_15320_cast, var_15467_cast))[name = tensor<string, []>("op_15492_cast")];
            tensor<string, []> var_15494_equation_0 = const()[name = tensor<string, []>("op_15494_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15494_cast = einsum(equation = var_15494_equation_0, values = (var_15324_cast, var_15468_cast))[name = tensor<string, []>("op_15494_cast")];
            tensor<string, []> var_15496_equation_0 = const()[name = tensor<string, []>("op_15496_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15496_cast = einsum(equation = var_15496_equation_0, values = (var_15328_cast, var_15469_cast))[name = tensor<string, []>("op_15496_cast")];
            tensor<string, []> var_15498_equation_0 = const()[name = tensor<string, []>("op_15498_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15498_cast = einsum(equation = var_15498_equation_0, values = (var_15332_cast, var_15470_cast))[name = tensor<string, []>("op_15498_cast")];
            tensor<string, []> var_15500_equation_0 = const()[name = tensor<string, []>("op_15500_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15500_cast = einsum(equation = var_15500_equation_0, values = (var_15336_cast, var_15471_cast))[name = tensor<string, []>("op_15500_cast")];
            tensor<string, []> var_15502_equation_0 = const()[name = tensor<string, []>("op_15502_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15502_cast = einsum(equation = var_15502_equation_0, values = (var_15340_cast, var_15472_cast))[name = tensor<string, []>("op_15502_cast")];
            tensor<string, []> var_15504_equation_0 = const()[name = tensor<string, []>("op_15504_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15504_cast = einsum(equation = var_15504_equation_0, values = (var_15344_cast, var_15473_cast))[name = tensor<string, []>("op_15504_cast")];
            tensor<string, []> var_15506_equation_0 = const()[name = tensor<string, []>("op_15506_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15506_cast = einsum(equation = var_15506_equation_0, values = (var_15348_cast, var_15474_cast))[name = tensor<string, []>("op_15506_cast")];
            tensor<string, []> var_15508_equation_0 = const()[name = tensor<string, []>("op_15508_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15508_cast = einsum(equation = var_15508_equation_0, values = (var_15352_cast, var_15475_cast))[name = tensor<string, []>("op_15508_cast")];
            tensor<string, []> var_15510_equation_0 = const()[name = tensor<string, []>("op_15510_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15510_cast = einsum(equation = var_15510_equation_0, values = (var_15356_cast, var_15476_cast))[name = tensor<string, []>("op_15510_cast")];
            tensor<string, []> var_15512_equation_0 = const()[name = tensor<string, []>("op_15512_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15512_cast = einsum(equation = var_15512_equation_0, values = (var_15360_cast, var_15477_cast))[name = tensor<string, []>("op_15512_cast")];
            tensor<string, []> var_15514_equation_0 = const()[name = tensor<string, []>("op_15514_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15514_cast = einsum(equation = var_15514_equation_0, values = (var_15364_cast, var_15478_cast))[name = tensor<string, []>("op_15514_cast")];
            tensor<string, []> var_15516_equation_0 = const()[name = tensor<string, []>("op_15516_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15516_cast = einsum(equation = var_15516_equation_0, values = (var_15368_cast, var_15479_cast))[name = tensor<string, []>("op_15516_cast")];
            tensor<string, []> var_15518_equation_0 = const()[name = tensor<string, []>("op_15518_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15518_cast = einsum(equation = var_15518_equation_0, values = (var_15372_cast, var_15480_cast))[name = tensor<string, []>("op_15518_cast")];
            tensor<string, []> var_15520_equation_0 = const()[name = tensor<string, []>("op_15520_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15520_cast = einsum(equation = var_15520_equation_0, values = (var_15376_cast, var_15481_cast))[name = tensor<string, []>("op_15520_cast")];
            tensor<string, []> var_15522_equation_0 = const()[name = tensor<string, []>("op_15522_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15522_cast = einsum(equation = var_15522_equation_0, values = (var_15380_cast, var_15482_cast))[name = tensor<string, []>("op_15522_cast")];
            tensor<bool, []> input_259_interleave_0 = const()[name = tensor<string, []>("input_259_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_259_cast = concat(axis = var_2634, interleave = input_259_interleave_0, values = (var_15484_cast, var_15486_cast, var_15488_cast, var_15490_cast, var_15492_cast, var_15494_cast, var_15496_cast, var_15498_cast, var_15500_cast, var_15502_cast, var_15504_cast, var_15506_cast, var_15508_cast, var_15510_cast, var_15512_cast, var_15514_cast, var_15516_cast, var_15518_cast, var_15520_cast, var_15522_cast))[name = tensor<string, []>("input_259_cast")];
            tensor<int32, [2]> var_15528 = const()[name = tensor<string, []>("op_15528"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_15530 = const()[name = tensor<string, []>("op_15530"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_15532_pad_type_0 = const()[name = tensor<string, []>("op_15532_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_15532_pad_0 = const()[name = tensor<string, []>("op_15532_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1197559680)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1200836544)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_15532_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_out_0_bias_to_fp16, dilations = var_15530, groups = var_2634, pad = var_15532_pad_0, pad_type = var_15532_pad_type_0, strides = var_15528, weight = down_blocks_2_attentions_1_transformer_blocks_3_attn2_to_out_0_weight_to_fp16, x = input_259_cast)[name = tensor<string, []>("op_15532_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_107_cast = add(x = var_15532_cast, y = inputs_105_cast)[name = tensor<string, []>("inputs_107_cast")];
            tensor<int32, [1]> var_15536 = const()[name = tensor<string, []>("op_15536"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_107_cast = reduce_mean(axes = var_15536, keep_dims = var_2629, x = inputs_107_cast)[name = tensor<string, []>("channels_mean_107_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_107_cast = sub(x = inputs_107_cast, y = channels_mean_107_cast)[name = tensor<string, []>("zero_mean_107_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_107_cast = mul(x = zero_mean_107_cast, y = zero_mean_107_cast)[name = tensor<string, []>("zero_mean_sq_107_cast")];
            tensor<int32, [1]> var_15540 = const()[name = tensor<string, []>("op_15540"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_15541_cast = reduce_mean(axes = var_15540, keep_dims = var_2629, x = zero_mean_sq_107_cast)[name = tensor<string, []>("op_15541_cast")];
            tensor<fp16, []> var_15542_to_fp16 = const()[name = tensor<string, []>("op_15542_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_15543_cast = add(x = var_15541_cast, y = var_15542_to_fp16)[name = tensor<string, []>("op_15543_cast")];
            tensor<fp16, []> denom_107_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_107_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_107_cast = rsqrt(epsilon = denom_107_epsilon_0_to_fp16, x = var_15543_cast)[name = tensor<string, []>("denom_107_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_107_cast = mul(x = zero_mean_107_cast, y = denom_107_cast)[name = tensor<string, []>("out_107_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_15547_to_fp16 = const()[name = tensor<string, []>("op_15547_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1200839168)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_15548_cast = add(x = out_107_cast, y = var_15547_to_fp16)[name = tensor<string, []>("op_15548_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_15550_to_fp16 = const()[name = tensor<string, []>("op_15550_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1200841792)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_261_cast = mul(x = var_15548_cast, y = var_15550_to_fp16)[name = tensor<string, []>("input_261_cast")];
            tensor<int32, [2]> var_15558 = const()[name = tensor<string, []>("op_15558"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_15560 = const()[name = tensor<string, []>("op_15560"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_15562_pad_type_0 = const()[name = tensor<string, []>("op_15562_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_15562_pad_0 = const()[name = tensor<string, []>("op_15562_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_3_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_3_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1200844416)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_1_transformer_blocks_3_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_3_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1227058880)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_15562_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_3_ff_net_0_proj_bias_to_fp16, dilations = var_15560, groups = var_2634, pad = var_15562_pad_0, pad_type = var_15562_pad_type_0, strides = var_15558, weight = down_blocks_2_attentions_1_transformer_blocks_3_ff_net_0_proj_weight_to_fp16, x = input_261_cast)[name = tensor<string, []>("op_15562_cast")];
            tensor<int32, [2]> var_15563_split_sizes_0 = const()[name = tensor<string, []>("op_15563_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_15563_axis_0 = const()[name = tensor<string, []>("op_15563_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_15563_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_15563_cast_1 = split(axis = var_15563_axis_0, split_sizes = var_15563_split_sizes_0, x = var_15562_cast)[name = tensor<string, []>("op_15563_cast")];
            tensor<string, []> var_15565_mode_0 = const()[name = tensor<string, []>("op_15565_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_15565_cast = gelu(mode = var_15565_mode_0, x = var_15563_cast_1)[name = tensor<string, []>("op_15565_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_263_cast = mul(x = var_15563_cast_0, y = var_15565_cast)[name = tensor<string, []>("input_263_cast")];
            tensor<int32, [2]> var_15569 = const()[name = tensor<string, []>("op_15569"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_15571 = const()[name = tensor<string, []>("op_15571"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_15573_pad_type_0 = const()[name = tensor<string, []>("op_15573_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_15573_pad_0 = const()[name = tensor<string, []>("op_15573_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_3_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_3_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1227079424)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_3_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_3_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1240186688)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_15573_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_3_ff_net_2_bias_to_fp16, dilations = var_15571, groups = var_2634, pad = var_15573_pad_0, pad_type = var_15573_pad_type_0, strides = var_15569, weight = down_blocks_2_attentions_1_transformer_blocks_3_ff_net_2_weight_to_fp16, x = input_263_cast)[name = tensor<string, []>("op_15573_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_109_cast = add(x = var_15573_cast, y = inputs_107_cast)[name = tensor<string, []>("inputs_109_cast")];
            tensor<int32, [1]> var_15583 = const()[name = tensor<string, []>("op_15583"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_109_cast = reduce_mean(axes = var_15583, keep_dims = var_2629, x = inputs_109_cast)[name = tensor<string, []>("channels_mean_109_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_109_cast = sub(x = inputs_109_cast, y = channels_mean_109_cast)[name = tensor<string, []>("zero_mean_109_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_109_cast = mul(x = zero_mean_109_cast, y = zero_mean_109_cast)[name = tensor<string, []>("zero_mean_sq_109_cast")];
            tensor<int32, [1]> var_15587 = const()[name = tensor<string, []>("op_15587"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_15588_cast = reduce_mean(axes = var_15587, keep_dims = var_2629, x = zero_mean_sq_109_cast)[name = tensor<string, []>("op_15588_cast")];
            tensor<fp16, []> var_15589_to_fp16 = const()[name = tensor<string, []>("op_15589_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_15590_cast = add(x = var_15588_cast, y = var_15589_to_fp16)[name = tensor<string, []>("op_15590_cast")];
            tensor<fp16, []> denom_109_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_109_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_109_cast = rsqrt(epsilon = denom_109_epsilon_0_to_fp16, x = var_15590_cast)[name = tensor<string, []>("denom_109_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_109_cast = mul(x = zero_mean_109_cast, y = denom_109_cast)[name = tensor<string, []>("out_109_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_15594_to_fp16 = const()[name = tensor<string, []>("op_15594_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1240189312)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_15595_cast = add(x = out_109_cast, y = var_15594_to_fp16)[name = tensor<string, []>("op_15595_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_15597_to_fp16 = const()[name = tensor<string, []>("op_15597_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1240191936)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_161_cast = mul(x = var_15595_cast, y = var_15597_to_fp16)[name = tensor<string, []>("hidden_states_161_cast")];
            tensor<int32, [2]> var_15604 = const()[name = tensor<string, []>("op_15604"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_15606 = const()[name = tensor<string, []>("op_15606"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_73_pad_type_0 = const()[name = tensor<string, []>("q_73_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_73_pad_0 = const()[name = tensor<string, []>("q_73_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1240194560)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_73_cast = conv(dilations = var_15606, groups = var_2634, pad = q_73_pad_0, pad_type = q_73_pad_type_0, strides = var_15604, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_q_weight_to_fp16, x = hidden_states_161_cast)[name = tensor<string, []>("q_73_cast")];
            tensor<int32, [2]> var_15610 = const()[name = tensor<string, []>("op_15610"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_15612 = const()[name = tensor<string, []>("op_15612"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_145_pad_type_0 = const()[name = tensor<string, []>("k_145_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_145_pad_0 = const()[name = tensor<string, []>("k_145_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1243471424)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_145_cast = conv(dilations = var_15612, groups = var_2634, pad = k_145_pad_0, pad_type = k_145_pad_type_0, strides = var_15610, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_k_weight_to_fp16, x = hidden_states_161_cast)[name = tensor<string, []>("k_145_cast")];
            tensor<int32, [2]> var_15616 = const()[name = tensor<string, []>("op_15616"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_15618 = const()[name = tensor<string, []>("op_15618"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_73_pad_type_0 = const()[name = tensor<string, []>("v_73_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_73_pad_0 = const()[name = tensor<string, []>("v_73_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1246748288)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_73_cast = conv(dilations = var_15618, groups = var_2634, pad = v_73_pad_0, pad_type = v_73_pad_type_0, strides = var_15616, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_v_weight_to_fp16, x = hidden_states_161_cast)[name = tensor<string, []>("v_73_cast")];
            tensor<int32, [4]> var_15622_begin_0 = const()[name = tensor<string, []>("op_15622_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_15622_end_0 = const()[name = tensor<string, []>("op_15622_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_15622_end_mask_0 = const()[name = tensor<string, []>("op_15622_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15622_cast = slice_by_index(begin = var_15622_begin_0, end = var_15622_end_0, end_mask = var_15622_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15622_cast")];
            tensor<int32, [4]> var_15626_begin_0 = const()[name = tensor<string, []>("op_15626_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_15626_end_0 = const()[name = tensor<string, []>("op_15626_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_15626_end_mask_0 = const()[name = tensor<string, []>("op_15626_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15626_cast = slice_by_index(begin = var_15626_begin_0, end = var_15626_end_0, end_mask = var_15626_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15626_cast")];
            tensor<int32, [4]> var_15630_begin_0 = const()[name = tensor<string, []>("op_15630_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_15630_end_0 = const()[name = tensor<string, []>("op_15630_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_15630_end_mask_0 = const()[name = tensor<string, []>("op_15630_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15630_cast = slice_by_index(begin = var_15630_begin_0, end = var_15630_end_0, end_mask = var_15630_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15630_cast")];
            tensor<int32, [4]> var_15634_begin_0 = const()[name = tensor<string, []>("op_15634_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_15634_end_0 = const()[name = tensor<string, []>("op_15634_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_15634_end_mask_0 = const()[name = tensor<string, []>("op_15634_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15634_cast = slice_by_index(begin = var_15634_begin_0, end = var_15634_end_0, end_mask = var_15634_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15634_cast")];
            tensor<int32, [4]> var_15638_begin_0 = const()[name = tensor<string, []>("op_15638_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_15638_end_0 = const()[name = tensor<string, []>("op_15638_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_15638_end_mask_0 = const()[name = tensor<string, []>("op_15638_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15638_cast = slice_by_index(begin = var_15638_begin_0, end = var_15638_end_0, end_mask = var_15638_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15638_cast")];
            tensor<int32, [4]> var_15642_begin_0 = const()[name = tensor<string, []>("op_15642_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_15642_end_0 = const()[name = tensor<string, []>("op_15642_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_15642_end_mask_0 = const()[name = tensor<string, []>("op_15642_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15642_cast = slice_by_index(begin = var_15642_begin_0, end = var_15642_end_0, end_mask = var_15642_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15642_cast")];
            tensor<int32, [4]> var_15646_begin_0 = const()[name = tensor<string, []>("op_15646_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_15646_end_0 = const()[name = tensor<string, []>("op_15646_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_15646_end_mask_0 = const()[name = tensor<string, []>("op_15646_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15646_cast = slice_by_index(begin = var_15646_begin_0, end = var_15646_end_0, end_mask = var_15646_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15646_cast")];
            tensor<int32, [4]> var_15650_begin_0 = const()[name = tensor<string, []>("op_15650_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_15650_end_0 = const()[name = tensor<string, []>("op_15650_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_15650_end_mask_0 = const()[name = tensor<string, []>("op_15650_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15650_cast = slice_by_index(begin = var_15650_begin_0, end = var_15650_end_0, end_mask = var_15650_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15650_cast")];
            tensor<int32, [4]> var_15654_begin_0 = const()[name = tensor<string, []>("op_15654_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_15654_end_0 = const()[name = tensor<string, []>("op_15654_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_15654_end_mask_0 = const()[name = tensor<string, []>("op_15654_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15654_cast = slice_by_index(begin = var_15654_begin_0, end = var_15654_end_0, end_mask = var_15654_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15654_cast")];
            tensor<int32, [4]> var_15658_begin_0 = const()[name = tensor<string, []>("op_15658_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_15658_end_0 = const()[name = tensor<string, []>("op_15658_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_15658_end_mask_0 = const()[name = tensor<string, []>("op_15658_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15658_cast = slice_by_index(begin = var_15658_begin_0, end = var_15658_end_0, end_mask = var_15658_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15658_cast")];
            tensor<int32, [4]> var_15662_begin_0 = const()[name = tensor<string, []>("op_15662_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_15662_end_0 = const()[name = tensor<string, []>("op_15662_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_15662_end_mask_0 = const()[name = tensor<string, []>("op_15662_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15662_cast = slice_by_index(begin = var_15662_begin_0, end = var_15662_end_0, end_mask = var_15662_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15662_cast")];
            tensor<int32, [4]> var_15666_begin_0 = const()[name = tensor<string, []>("op_15666_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_15666_end_0 = const()[name = tensor<string, []>("op_15666_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_15666_end_mask_0 = const()[name = tensor<string, []>("op_15666_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15666_cast = slice_by_index(begin = var_15666_begin_0, end = var_15666_end_0, end_mask = var_15666_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15666_cast")];
            tensor<int32, [4]> var_15670_begin_0 = const()[name = tensor<string, []>("op_15670_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_15670_end_0 = const()[name = tensor<string, []>("op_15670_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_15670_end_mask_0 = const()[name = tensor<string, []>("op_15670_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15670_cast = slice_by_index(begin = var_15670_begin_0, end = var_15670_end_0, end_mask = var_15670_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15670_cast")];
            tensor<int32, [4]> var_15674_begin_0 = const()[name = tensor<string, []>("op_15674_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_15674_end_0 = const()[name = tensor<string, []>("op_15674_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_15674_end_mask_0 = const()[name = tensor<string, []>("op_15674_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15674_cast = slice_by_index(begin = var_15674_begin_0, end = var_15674_end_0, end_mask = var_15674_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15674_cast")];
            tensor<int32, [4]> var_15678_begin_0 = const()[name = tensor<string, []>("op_15678_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_15678_end_0 = const()[name = tensor<string, []>("op_15678_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_15678_end_mask_0 = const()[name = tensor<string, []>("op_15678_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15678_cast = slice_by_index(begin = var_15678_begin_0, end = var_15678_end_0, end_mask = var_15678_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15678_cast")];
            tensor<int32, [4]> var_15682_begin_0 = const()[name = tensor<string, []>("op_15682_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_15682_end_0 = const()[name = tensor<string, []>("op_15682_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_15682_end_mask_0 = const()[name = tensor<string, []>("op_15682_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15682_cast = slice_by_index(begin = var_15682_begin_0, end = var_15682_end_0, end_mask = var_15682_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15682_cast")];
            tensor<int32, [4]> var_15686_begin_0 = const()[name = tensor<string, []>("op_15686_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_15686_end_0 = const()[name = tensor<string, []>("op_15686_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_15686_end_mask_0 = const()[name = tensor<string, []>("op_15686_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15686_cast = slice_by_index(begin = var_15686_begin_0, end = var_15686_end_0, end_mask = var_15686_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15686_cast")];
            tensor<int32, [4]> var_15690_begin_0 = const()[name = tensor<string, []>("op_15690_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_15690_end_0 = const()[name = tensor<string, []>("op_15690_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_15690_end_mask_0 = const()[name = tensor<string, []>("op_15690_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15690_cast = slice_by_index(begin = var_15690_begin_0, end = var_15690_end_0, end_mask = var_15690_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15690_cast")];
            tensor<int32, [4]> var_15694_begin_0 = const()[name = tensor<string, []>("op_15694_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_15694_end_0 = const()[name = tensor<string, []>("op_15694_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_15694_end_mask_0 = const()[name = tensor<string, []>("op_15694_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15694_cast = slice_by_index(begin = var_15694_begin_0, end = var_15694_end_0, end_mask = var_15694_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15694_cast")];
            tensor<int32, [4]> var_15698_begin_0 = const()[name = tensor<string, []>("op_15698_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_15698_end_0 = const()[name = tensor<string, []>("op_15698_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_15698_end_mask_0 = const()[name = tensor<string, []>("op_15698_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15698_cast = slice_by_index(begin = var_15698_begin_0, end = var_15698_end_0, end_mask = var_15698_end_mask_0, x = q_73_cast)[name = tensor<string, []>("op_15698_cast")];
            tensor<int32, [4]> k_147_perm_0 = const()[name = tensor<string, []>("k_147_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_15705_begin_0 = const()[name = tensor<string, []>("op_15705_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_15705_end_0 = const()[name = tensor<string, []>("op_15705_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_15705_end_mask_0 = const()[name = tensor<string, []>("op_15705_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_103 = transpose(perm = k_147_perm_0, x = k_145_cast)[name = tensor<string, []>("transpose_103")];
            tensor<fp16, [2, 1024, 1, 64]> var_15705_cast = slice_by_index(begin = var_15705_begin_0, end = var_15705_end_0, end_mask = var_15705_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15705_cast")];
            tensor<int32, [4]> var_15709_begin_0 = const()[name = tensor<string, []>("op_15709_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_15709_end_0 = const()[name = tensor<string, []>("op_15709_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_15709_end_mask_0 = const()[name = tensor<string, []>("op_15709_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_15709_cast = slice_by_index(begin = var_15709_begin_0, end = var_15709_end_0, end_mask = var_15709_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15709_cast")];
            tensor<int32, [4]> var_15713_begin_0 = const()[name = tensor<string, []>("op_15713_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_15713_end_0 = const()[name = tensor<string, []>("op_15713_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_15713_end_mask_0 = const()[name = tensor<string, []>("op_15713_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_15713_cast = slice_by_index(begin = var_15713_begin_0, end = var_15713_end_0, end_mask = var_15713_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15713_cast")];
            tensor<int32, [4]> var_15717_begin_0 = const()[name = tensor<string, []>("op_15717_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_15717_end_0 = const()[name = tensor<string, []>("op_15717_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_15717_end_mask_0 = const()[name = tensor<string, []>("op_15717_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_15717_cast = slice_by_index(begin = var_15717_begin_0, end = var_15717_end_0, end_mask = var_15717_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15717_cast")];
            tensor<int32, [4]> var_15721_begin_0 = const()[name = tensor<string, []>("op_15721_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_15721_end_0 = const()[name = tensor<string, []>("op_15721_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_15721_end_mask_0 = const()[name = tensor<string, []>("op_15721_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_15721_cast = slice_by_index(begin = var_15721_begin_0, end = var_15721_end_0, end_mask = var_15721_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15721_cast")];
            tensor<int32, [4]> var_15725_begin_0 = const()[name = tensor<string, []>("op_15725_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_15725_end_0 = const()[name = tensor<string, []>("op_15725_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_15725_end_mask_0 = const()[name = tensor<string, []>("op_15725_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_15725_cast = slice_by_index(begin = var_15725_begin_0, end = var_15725_end_0, end_mask = var_15725_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15725_cast")];
            tensor<int32, [4]> var_15729_begin_0 = const()[name = tensor<string, []>("op_15729_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_15729_end_0 = const()[name = tensor<string, []>("op_15729_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_15729_end_mask_0 = const()[name = tensor<string, []>("op_15729_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_15729_cast = slice_by_index(begin = var_15729_begin_0, end = var_15729_end_0, end_mask = var_15729_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15729_cast")];
            tensor<int32, [4]> var_15733_begin_0 = const()[name = tensor<string, []>("op_15733_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_15733_end_0 = const()[name = tensor<string, []>("op_15733_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_15733_end_mask_0 = const()[name = tensor<string, []>("op_15733_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_15733_cast = slice_by_index(begin = var_15733_begin_0, end = var_15733_end_0, end_mask = var_15733_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15733_cast")];
            tensor<int32, [4]> var_15737_begin_0 = const()[name = tensor<string, []>("op_15737_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_15737_end_0 = const()[name = tensor<string, []>("op_15737_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_15737_end_mask_0 = const()[name = tensor<string, []>("op_15737_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_15737_cast = slice_by_index(begin = var_15737_begin_0, end = var_15737_end_0, end_mask = var_15737_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15737_cast")];
            tensor<int32, [4]> var_15741_begin_0 = const()[name = tensor<string, []>("op_15741_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_15741_end_0 = const()[name = tensor<string, []>("op_15741_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_15741_end_mask_0 = const()[name = tensor<string, []>("op_15741_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_15741_cast = slice_by_index(begin = var_15741_begin_0, end = var_15741_end_0, end_mask = var_15741_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15741_cast")];
            tensor<int32, [4]> var_15745_begin_0 = const()[name = tensor<string, []>("op_15745_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_15745_end_0 = const()[name = tensor<string, []>("op_15745_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_15745_end_mask_0 = const()[name = tensor<string, []>("op_15745_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_15745_cast = slice_by_index(begin = var_15745_begin_0, end = var_15745_end_0, end_mask = var_15745_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15745_cast")];
            tensor<int32, [4]> var_15749_begin_0 = const()[name = tensor<string, []>("op_15749_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_15749_end_0 = const()[name = tensor<string, []>("op_15749_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_15749_end_mask_0 = const()[name = tensor<string, []>("op_15749_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_15749_cast = slice_by_index(begin = var_15749_begin_0, end = var_15749_end_0, end_mask = var_15749_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15749_cast")];
            tensor<int32, [4]> var_15753_begin_0 = const()[name = tensor<string, []>("op_15753_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_15753_end_0 = const()[name = tensor<string, []>("op_15753_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_15753_end_mask_0 = const()[name = tensor<string, []>("op_15753_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_15753_cast = slice_by_index(begin = var_15753_begin_0, end = var_15753_end_0, end_mask = var_15753_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15753_cast")];
            tensor<int32, [4]> var_15757_begin_0 = const()[name = tensor<string, []>("op_15757_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_15757_end_0 = const()[name = tensor<string, []>("op_15757_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_15757_end_mask_0 = const()[name = tensor<string, []>("op_15757_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_15757_cast = slice_by_index(begin = var_15757_begin_0, end = var_15757_end_0, end_mask = var_15757_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15757_cast")];
            tensor<int32, [4]> var_15761_begin_0 = const()[name = tensor<string, []>("op_15761_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_15761_end_0 = const()[name = tensor<string, []>("op_15761_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_15761_end_mask_0 = const()[name = tensor<string, []>("op_15761_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_15761_cast = slice_by_index(begin = var_15761_begin_0, end = var_15761_end_0, end_mask = var_15761_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15761_cast")];
            tensor<int32, [4]> var_15765_begin_0 = const()[name = tensor<string, []>("op_15765_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_15765_end_0 = const()[name = tensor<string, []>("op_15765_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_15765_end_mask_0 = const()[name = tensor<string, []>("op_15765_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_15765_cast = slice_by_index(begin = var_15765_begin_0, end = var_15765_end_0, end_mask = var_15765_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15765_cast")];
            tensor<int32, [4]> var_15769_begin_0 = const()[name = tensor<string, []>("op_15769_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_15769_end_0 = const()[name = tensor<string, []>("op_15769_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_15769_end_mask_0 = const()[name = tensor<string, []>("op_15769_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_15769_cast = slice_by_index(begin = var_15769_begin_0, end = var_15769_end_0, end_mask = var_15769_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15769_cast")];
            tensor<int32, [4]> var_15773_begin_0 = const()[name = tensor<string, []>("op_15773_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_15773_end_0 = const()[name = tensor<string, []>("op_15773_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_15773_end_mask_0 = const()[name = tensor<string, []>("op_15773_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_15773_cast = slice_by_index(begin = var_15773_begin_0, end = var_15773_end_0, end_mask = var_15773_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15773_cast")];
            tensor<int32, [4]> var_15777_begin_0 = const()[name = tensor<string, []>("op_15777_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_15777_end_0 = const()[name = tensor<string, []>("op_15777_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_15777_end_mask_0 = const()[name = tensor<string, []>("op_15777_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_15777_cast = slice_by_index(begin = var_15777_begin_0, end = var_15777_end_0, end_mask = var_15777_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15777_cast")];
            tensor<int32, [4]> var_15781_begin_0 = const()[name = tensor<string, []>("op_15781_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_15781_end_0 = const()[name = tensor<string, []>("op_15781_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_15781_end_mask_0 = const()[name = tensor<string, []>("op_15781_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_15781_cast = slice_by_index(begin = var_15781_begin_0, end = var_15781_end_0, end_mask = var_15781_end_mask_0, x = transpose_103)[name = tensor<string, []>("op_15781_cast")];
            tensor<int32, [4]> var_15783_begin_0 = const()[name = tensor<string, []>("op_15783_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_15783_end_0 = const()[name = tensor<string, []>("op_15783_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_15783_end_mask_0 = const()[name = tensor<string, []>("op_15783_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15783_cast = slice_by_index(begin = var_15783_begin_0, end = var_15783_end_0, end_mask = var_15783_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15783_cast")];
            tensor<int32, [4]> var_15787_begin_0 = const()[name = tensor<string, []>("op_15787_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_15787_end_0 = const()[name = tensor<string, []>("op_15787_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_15787_end_mask_0 = const()[name = tensor<string, []>("op_15787_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15787_cast = slice_by_index(begin = var_15787_begin_0, end = var_15787_end_0, end_mask = var_15787_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15787_cast")];
            tensor<int32, [4]> var_15791_begin_0 = const()[name = tensor<string, []>("op_15791_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_15791_end_0 = const()[name = tensor<string, []>("op_15791_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_15791_end_mask_0 = const()[name = tensor<string, []>("op_15791_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15791_cast = slice_by_index(begin = var_15791_begin_0, end = var_15791_end_0, end_mask = var_15791_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15791_cast")];
            tensor<int32, [4]> var_15795_begin_0 = const()[name = tensor<string, []>("op_15795_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_15795_end_0 = const()[name = tensor<string, []>("op_15795_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_15795_end_mask_0 = const()[name = tensor<string, []>("op_15795_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15795_cast = slice_by_index(begin = var_15795_begin_0, end = var_15795_end_0, end_mask = var_15795_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15795_cast")];
            tensor<int32, [4]> var_15799_begin_0 = const()[name = tensor<string, []>("op_15799_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_15799_end_0 = const()[name = tensor<string, []>("op_15799_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_15799_end_mask_0 = const()[name = tensor<string, []>("op_15799_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15799_cast = slice_by_index(begin = var_15799_begin_0, end = var_15799_end_0, end_mask = var_15799_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15799_cast")];
            tensor<int32, [4]> var_15803_begin_0 = const()[name = tensor<string, []>("op_15803_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_15803_end_0 = const()[name = tensor<string, []>("op_15803_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_15803_end_mask_0 = const()[name = tensor<string, []>("op_15803_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15803_cast = slice_by_index(begin = var_15803_begin_0, end = var_15803_end_0, end_mask = var_15803_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15803_cast")];
            tensor<int32, [4]> var_15807_begin_0 = const()[name = tensor<string, []>("op_15807_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_15807_end_0 = const()[name = tensor<string, []>("op_15807_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_15807_end_mask_0 = const()[name = tensor<string, []>("op_15807_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15807_cast = slice_by_index(begin = var_15807_begin_0, end = var_15807_end_0, end_mask = var_15807_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15807_cast")];
            tensor<int32, [4]> var_15811_begin_0 = const()[name = tensor<string, []>("op_15811_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_15811_end_0 = const()[name = tensor<string, []>("op_15811_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_15811_end_mask_0 = const()[name = tensor<string, []>("op_15811_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15811_cast = slice_by_index(begin = var_15811_begin_0, end = var_15811_end_0, end_mask = var_15811_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15811_cast")];
            tensor<int32, [4]> var_15815_begin_0 = const()[name = tensor<string, []>("op_15815_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_15815_end_0 = const()[name = tensor<string, []>("op_15815_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_15815_end_mask_0 = const()[name = tensor<string, []>("op_15815_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15815_cast = slice_by_index(begin = var_15815_begin_0, end = var_15815_end_0, end_mask = var_15815_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15815_cast")];
            tensor<int32, [4]> var_15819_begin_0 = const()[name = tensor<string, []>("op_15819_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_15819_end_0 = const()[name = tensor<string, []>("op_15819_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_15819_end_mask_0 = const()[name = tensor<string, []>("op_15819_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15819_cast = slice_by_index(begin = var_15819_begin_0, end = var_15819_end_0, end_mask = var_15819_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15819_cast")];
            tensor<int32, [4]> var_15823_begin_0 = const()[name = tensor<string, []>("op_15823_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_15823_end_0 = const()[name = tensor<string, []>("op_15823_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_15823_end_mask_0 = const()[name = tensor<string, []>("op_15823_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15823_cast = slice_by_index(begin = var_15823_begin_0, end = var_15823_end_0, end_mask = var_15823_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15823_cast")];
            tensor<int32, [4]> var_15827_begin_0 = const()[name = tensor<string, []>("op_15827_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_15827_end_0 = const()[name = tensor<string, []>("op_15827_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_15827_end_mask_0 = const()[name = tensor<string, []>("op_15827_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15827_cast = slice_by_index(begin = var_15827_begin_0, end = var_15827_end_0, end_mask = var_15827_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15827_cast")];
            tensor<int32, [4]> var_15831_begin_0 = const()[name = tensor<string, []>("op_15831_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_15831_end_0 = const()[name = tensor<string, []>("op_15831_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_15831_end_mask_0 = const()[name = tensor<string, []>("op_15831_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15831_cast = slice_by_index(begin = var_15831_begin_0, end = var_15831_end_0, end_mask = var_15831_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15831_cast")];
            tensor<int32, [4]> var_15835_begin_0 = const()[name = tensor<string, []>("op_15835_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_15835_end_0 = const()[name = tensor<string, []>("op_15835_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_15835_end_mask_0 = const()[name = tensor<string, []>("op_15835_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15835_cast = slice_by_index(begin = var_15835_begin_0, end = var_15835_end_0, end_mask = var_15835_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15835_cast")];
            tensor<int32, [4]> var_15839_begin_0 = const()[name = tensor<string, []>("op_15839_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_15839_end_0 = const()[name = tensor<string, []>("op_15839_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_15839_end_mask_0 = const()[name = tensor<string, []>("op_15839_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15839_cast = slice_by_index(begin = var_15839_begin_0, end = var_15839_end_0, end_mask = var_15839_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15839_cast")];
            tensor<int32, [4]> var_15843_begin_0 = const()[name = tensor<string, []>("op_15843_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_15843_end_0 = const()[name = tensor<string, []>("op_15843_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_15843_end_mask_0 = const()[name = tensor<string, []>("op_15843_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15843_cast = slice_by_index(begin = var_15843_begin_0, end = var_15843_end_0, end_mask = var_15843_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15843_cast")];
            tensor<int32, [4]> var_15847_begin_0 = const()[name = tensor<string, []>("op_15847_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_15847_end_0 = const()[name = tensor<string, []>("op_15847_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_15847_end_mask_0 = const()[name = tensor<string, []>("op_15847_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15847_cast = slice_by_index(begin = var_15847_begin_0, end = var_15847_end_0, end_mask = var_15847_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15847_cast")];
            tensor<int32, [4]> var_15851_begin_0 = const()[name = tensor<string, []>("op_15851_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_15851_end_0 = const()[name = tensor<string, []>("op_15851_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_15851_end_mask_0 = const()[name = tensor<string, []>("op_15851_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15851_cast = slice_by_index(begin = var_15851_begin_0, end = var_15851_end_0, end_mask = var_15851_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15851_cast")];
            tensor<int32, [4]> var_15855_begin_0 = const()[name = tensor<string, []>("op_15855_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_15855_end_0 = const()[name = tensor<string, []>("op_15855_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_15855_end_mask_0 = const()[name = tensor<string, []>("op_15855_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15855_cast = slice_by_index(begin = var_15855_begin_0, end = var_15855_end_0, end_mask = var_15855_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15855_cast")];
            tensor<int32, [4]> var_15859_begin_0 = const()[name = tensor<string, []>("op_15859_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_15859_end_0 = const()[name = tensor<string, []>("op_15859_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_15859_end_mask_0 = const()[name = tensor<string, []>("op_15859_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_15859_cast = slice_by_index(begin = var_15859_begin_0, end = var_15859_end_0, end_mask = var_15859_end_mask_0, x = v_73_cast)[name = tensor<string, []>("op_15859_cast")];
            tensor<string, []> var_15863_equation_0 = const()[name = tensor<string, []>("op_15863_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15863_cast = einsum(equation = var_15863_equation_0, values = (var_15705_cast, var_15622_cast))[name = tensor<string, []>("op_15863_cast")];
            tensor<fp16, []> var_15864_to_fp16 = const()[name = tensor<string, []>("op_15864_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1281_cast = mul(x = var_15863_cast, y = var_15864_to_fp16)[name = tensor<string, []>("aw_1281_cast")];
            tensor<string, []> var_15867_equation_0 = const()[name = tensor<string, []>("op_15867_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15867_cast = einsum(equation = var_15867_equation_0, values = (var_15709_cast, var_15626_cast))[name = tensor<string, []>("op_15867_cast")];
            tensor<fp16, []> var_15868_to_fp16 = const()[name = tensor<string, []>("op_15868_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1283_cast = mul(x = var_15867_cast, y = var_15868_to_fp16)[name = tensor<string, []>("aw_1283_cast")];
            tensor<string, []> var_15871_equation_0 = const()[name = tensor<string, []>("op_15871_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15871_cast = einsum(equation = var_15871_equation_0, values = (var_15713_cast, var_15630_cast))[name = tensor<string, []>("op_15871_cast")];
            tensor<fp16, []> var_15872_to_fp16 = const()[name = tensor<string, []>("op_15872_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1285_cast = mul(x = var_15871_cast, y = var_15872_to_fp16)[name = tensor<string, []>("aw_1285_cast")];
            tensor<string, []> var_15875_equation_0 = const()[name = tensor<string, []>("op_15875_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15875_cast = einsum(equation = var_15875_equation_0, values = (var_15717_cast, var_15634_cast))[name = tensor<string, []>("op_15875_cast")];
            tensor<fp16, []> var_15876_to_fp16 = const()[name = tensor<string, []>("op_15876_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1287_cast = mul(x = var_15875_cast, y = var_15876_to_fp16)[name = tensor<string, []>("aw_1287_cast")];
            tensor<string, []> var_15879_equation_0 = const()[name = tensor<string, []>("op_15879_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15879_cast = einsum(equation = var_15879_equation_0, values = (var_15721_cast, var_15638_cast))[name = tensor<string, []>("op_15879_cast")];
            tensor<fp16, []> var_15880_to_fp16 = const()[name = tensor<string, []>("op_15880_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1289_cast = mul(x = var_15879_cast, y = var_15880_to_fp16)[name = tensor<string, []>("aw_1289_cast")];
            tensor<string, []> var_15883_equation_0 = const()[name = tensor<string, []>("op_15883_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15883_cast = einsum(equation = var_15883_equation_0, values = (var_15725_cast, var_15642_cast))[name = tensor<string, []>("op_15883_cast")];
            tensor<fp16, []> var_15884_to_fp16 = const()[name = tensor<string, []>("op_15884_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1291_cast = mul(x = var_15883_cast, y = var_15884_to_fp16)[name = tensor<string, []>("aw_1291_cast")];
            tensor<string, []> var_15887_equation_0 = const()[name = tensor<string, []>("op_15887_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15887_cast = einsum(equation = var_15887_equation_0, values = (var_15729_cast, var_15646_cast))[name = tensor<string, []>("op_15887_cast")];
            tensor<fp16, []> var_15888_to_fp16 = const()[name = tensor<string, []>("op_15888_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1293_cast = mul(x = var_15887_cast, y = var_15888_to_fp16)[name = tensor<string, []>("aw_1293_cast")];
            tensor<string, []> var_15891_equation_0 = const()[name = tensor<string, []>("op_15891_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15891_cast = einsum(equation = var_15891_equation_0, values = (var_15733_cast, var_15650_cast))[name = tensor<string, []>("op_15891_cast")];
            tensor<fp16, []> var_15892_to_fp16 = const()[name = tensor<string, []>("op_15892_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1295_cast = mul(x = var_15891_cast, y = var_15892_to_fp16)[name = tensor<string, []>("aw_1295_cast")];
            tensor<string, []> var_15895_equation_0 = const()[name = tensor<string, []>("op_15895_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15895_cast = einsum(equation = var_15895_equation_0, values = (var_15737_cast, var_15654_cast))[name = tensor<string, []>("op_15895_cast")];
            tensor<fp16, []> var_15896_to_fp16 = const()[name = tensor<string, []>("op_15896_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1297_cast = mul(x = var_15895_cast, y = var_15896_to_fp16)[name = tensor<string, []>("aw_1297_cast")];
            tensor<string, []> var_15899_equation_0 = const()[name = tensor<string, []>("op_15899_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15899_cast = einsum(equation = var_15899_equation_0, values = (var_15741_cast, var_15658_cast))[name = tensor<string, []>("op_15899_cast")];
            tensor<fp16, []> var_15900_to_fp16 = const()[name = tensor<string, []>("op_15900_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1299_cast = mul(x = var_15899_cast, y = var_15900_to_fp16)[name = tensor<string, []>("aw_1299_cast")];
            tensor<string, []> var_15903_equation_0 = const()[name = tensor<string, []>("op_15903_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15903_cast = einsum(equation = var_15903_equation_0, values = (var_15745_cast, var_15662_cast))[name = tensor<string, []>("op_15903_cast")];
            tensor<fp16, []> var_15904_to_fp16 = const()[name = tensor<string, []>("op_15904_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1301_cast = mul(x = var_15903_cast, y = var_15904_to_fp16)[name = tensor<string, []>("aw_1301_cast")];
            tensor<string, []> var_15907_equation_0 = const()[name = tensor<string, []>("op_15907_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15907_cast = einsum(equation = var_15907_equation_0, values = (var_15749_cast, var_15666_cast))[name = tensor<string, []>("op_15907_cast")];
            tensor<fp16, []> var_15908_to_fp16 = const()[name = tensor<string, []>("op_15908_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1303_cast = mul(x = var_15907_cast, y = var_15908_to_fp16)[name = tensor<string, []>("aw_1303_cast")];
            tensor<string, []> var_15911_equation_0 = const()[name = tensor<string, []>("op_15911_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15911_cast = einsum(equation = var_15911_equation_0, values = (var_15753_cast, var_15670_cast))[name = tensor<string, []>("op_15911_cast")];
            tensor<fp16, []> var_15912_to_fp16 = const()[name = tensor<string, []>("op_15912_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1305_cast = mul(x = var_15911_cast, y = var_15912_to_fp16)[name = tensor<string, []>("aw_1305_cast")];
            tensor<string, []> var_15915_equation_0 = const()[name = tensor<string, []>("op_15915_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15915_cast = einsum(equation = var_15915_equation_0, values = (var_15757_cast, var_15674_cast))[name = tensor<string, []>("op_15915_cast")];
            tensor<fp16, []> var_15916_to_fp16 = const()[name = tensor<string, []>("op_15916_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1307_cast = mul(x = var_15915_cast, y = var_15916_to_fp16)[name = tensor<string, []>("aw_1307_cast")];
            tensor<string, []> var_15919_equation_0 = const()[name = tensor<string, []>("op_15919_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15919_cast = einsum(equation = var_15919_equation_0, values = (var_15761_cast, var_15678_cast))[name = tensor<string, []>("op_15919_cast")];
            tensor<fp16, []> var_15920_to_fp16 = const()[name = tensor<string, []>("op_15920_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1309_cast = mul(x = var_15919_cast, y = var_15920_to_fp16)[name = tensor<string, []>("aw_1309_cast")];
            tensor<string, []> var_15923_equation_0 = const()[name = tensor<string, []>("op_15923_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15923_cast = einsum(equation = var_15923_equation_0, values = (var_15765_cast, var_15682_cast))[name = tensor<string, []>("op_15923_cast")];
            tensor<fp16, []> var_15924_to_fp16 = const()[name = tensor<string, []>("op_15924_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1311_cast = mul(x = var_15923_cast, y = var_15924_to_fp16)[name = tensor<string, []>("aw_1311_cast")];
            tensor<string, []> var_15927_equation_0 = const()[name = tensor<string, []>("op_15927_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15927_cast = einsum(equation = var_15927_equation_0, values = (var_15769_cast, var_15686_cast))[name = tensor<string, []>("op_15927_cast")];
            tensor<fp16, []> var_15928_to_fp16 = const()[name = tensor<string, []>("op_15928_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1313_cast = mul(x = var_15927_cast, y = var_15928_to_fp16)[name = tensor<string, []>("aw_1313_cast")];
            tensor<string, []> var_15931_equation_0 = const()[name = tensor<string, []>("op_15931_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15931_cast = einsum(equation = var_15931_equation_0, values = (var_15773_cast, var_15690_cast))[name = tensor<string, []>("op_15931_cast")];
            tensor<fp16, []> var_15932_to_fp16 = const()[name = tensor<string, []>("op_15932_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1315_cast = mul(x = var_15931_cast, y = var_15932_to_fp16)[name = tensor<string, []>("aw_1315_cast")];
            tensor<string, []> var_15935_equation_0 = const()[name = tensor<string, []>("op_15935_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15935_cast = einsum(equation = var_15935_equation_0, values = (var_15777_cast, var_15694_cast))[name = tensor<string, []>("op_15935_cast")];
            tensor<fp16, []> var_15936_to_fp16 = const()[name = tensor<string, []>("op_15936_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1317_cast = mul(x = var_15935_cast, y = var_15936_to_fp16)[name = tensor<string, []>("aw_1317_cast")];
            tensor<string, []> var_15939_equation_0 = const()[name = tensor<string, []>("op_15939_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15939_cast = einsum(equation = var_15939_equation_0, values = (var_15781_cast, var_15698_cast))[name = tensor<string, []>("op_15939_cast")];
            tensor<fp16, []> var_15940_to_fp16 = const()[name = tensor<string, []>("op_15940_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1319_cast = mul(x = var_15939_cast, y = var_15940_to_fp16)[name = tensor<string, []>("aw_1319_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15942_cast = softmax(axis = var_2634, x = aw_1281_cast)[name = tensor<string, []>("op_15942_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15943_cast = softmax(axis = var_2634, x = aw_1283_cast)[name = tensor<string, []>("op_15943_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15944_cast = softmax(axis = var_2634, x = aw_1285_cast)[name = tensor<string, []>("op_15944_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15945_cast = softmax(axis = var_2634, x = aw_1287_cast)[name = tensor<string, []>("op_15945_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15946_cast = softmax(axis = var_2634, x = aw_1289_cast)[name = tensor<string, []>("op_15946_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15947_cast = softmax(axis = var_2634, x = aw_1291_cast)[name = tensor<string, []>("op_15947_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15948_cast = softmax(axis = var_2634, x = aw_1293_cast)[name = tensor<string, []>("op_15948_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15949_cast = softmax(axis = var_2634, x = aw_1295_cast)[name = tensor<string, []>("op_15949_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15950_cast = softmax(axis = var_2634, x = aw_1297_cast)[name = tensor<string, []>("op_15950_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15951_cast = softmax(axis = var_2634, x = aw_1299_cast)[name = tensor<string, []>("op_15951_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15952_cast = softmax(axis = var_2634, x = aw_1301_cast)[name = tensor<string, []>("op_15952_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15953_cast = softmax(axis = var_2634, x = aw_1303_cast)[name = tensor<string, []>("op_15953_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15954_cast = softmax(axis = var_2634, x = aw_1305_cast)[name = tensor<string, []>("op_15954_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15955_cast = softmax(axis = var_2634, x = aw_1307_cast)[name = tensor<string, []>("op_15955_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15956_cast = softmax(axis = var_2634, x = aw_1309_cast)[name = tensor<string, []>("op_15956_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15957_cast = softmax(axis = var_2634, x = aw_1311_cast)[name = tensor<string, []>("op_15957_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15958_cast = softmax(axis = var_2634, x = aw_1313_cast)[name = tensor<string, []>("op_15958_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15959_cast = softmax(axis = var_2634, x = aw_1315_cast)[name = tensor<string, []>("op_15959_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15960_cast = softmax(axis = var_2634, x = aw_1317_cast)[name = tensor<string, []>("op_15960_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_15961_cast = softmax(axis = var_2634, x = aw_1319_cast)[name = tensor<string, []>("op_15961_cast")];
            tensor<string, []> var_15963_equation_0 = const()[name = tensor<string, []>("op_15963_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15963_cast = einsum(equation = var_15963_equation_0, values = (var_15783_cast, var_15942_cast))[name = tensor<string, []>("op_15963_cast")];
            tensor<string, []> var_15965_equation_0 = const()[name = tensor<string, []>("op_15965_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15965_cast = einsum(equation = var_15965_equation_0, values = (var_15787_cast, var_15943_cast))[name = tensor<string, []>("op_15965_cast")];
            tensor<string, []> var_15967_equation_0 = const()[name = tensor<string, []>("op_15967_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15967_cast = einsum(equation = var_15967_equation_0, values = (var_15791_cast, var_15944_cast))[name = tensor<string, []>("op_15967_cast")];
            tensor<string, []> var_15969_equation_0 = const()[name = tensor<string, []>("op_15969_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15969_cast = einsum(equation = var_15969_equation_0, values = (var_15795_cast, var_15945_cast))[name = tensor<string, []>("op_15969_cast")];
            tensor<string, []> var_15971_equation_0 = const()[name = tensor<string, []>("op_15971_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15971_cast = einsum(equation = var_15971_equation_0, values = (var_15799_cast, var_15946_cast))[name = tensor<string, []>("op_15971_cast")];
            tensor<string, []> var_15973_equation_0 = const()[name = tensor<string, []>("op_15973_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15973_cast = einsum(equation = var_15973_equation_0, values = (var_15803_cast, var_15947_cast))[name = tensor<string, []>("op_15973_cast")];
            tensor<string, []> var_15975_equation_0 = const()[name = tensor<string, []>("op_15975_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15975_cast = einsum(equation = var_15975_equation_0, values = (var_15807_cast, var_15948_cast))[name = tensor<string, []>("op_15975_cast")];
            tensor<string, []> var_15977_equation_0 = const()[name = tensor<string, []>("op_15977_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15977_cast = einsum(equation = var_15977_equation_0, values = (var_15811_cast, var_15949_cast))[name = tensor<string, []>("op_15977_cast")];
            tensor<string, []> var_15979_equation_0 = const()[name = tensor<string, []>("op_15979_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15979_cast = einsum(equation = var_15979_equation_0, values = (var_15815_cast, var_15950_cast))[name = tensor<string, []>("op_15979_cast")];
            tensor<string, []> var_15981_equation_0 = const()[name = tensor<string, []>("op_15981_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15981_cast = einsum(equation = var_15981_equation_0, values = (var_15819_cast, var_15951_cast))[name = tensor<string, []>("op_15981_cast")];
            tensor<string, []> var_15983_equation_0 = const()[name = tensor<string, []>("op_15983_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15983_cast = einsum(equation = var_15983_equation_0, values = (var_15823_cast, var_15952_cast))[name = tensor<string, []>("op_15983_cast")];
            tensor<string, []> var_15985_equation_0 = const()[name = tensor<string, []>("op_15985_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15985_cast = einsum(equation = var_15985_equation_0, values = (var_15827_cast, var_15953_cast))[name = tensor<string, []>("op_15985_cast")];
            tensor<string, []> var_15987_equation_0 = const()[name = tensor<string, []>("op_15987_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15987_cast = einsum(equation = var_15987_equation_0, values = (var_15831_cast, var_15954_cast))[name = tensor<string, []>("op_15987_cast")];
            tensor<string, []> var_15989_equation_0 = const()[name = tensor<string, []>("op_15989_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15989_cast = einsum(equation = var_15989_equation_0, values = (var_15835_cast, var_15955_cast))[name = tensor<string, []>("op_15989_cast")];
            tensor<string, []> var_15991_equation_0 = const()[name = tensor<string, []>("op_15991_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15991_cast = einsum(equation = var_15991_equation_0, values = (var_15839_cast, var_15956_cast))[name = tensor<string, []>("op_15991_cast")];
            tensor<string, []> var_15993_equation_0 = const()[name = tensor<string, []>("op_15993_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15993_cast = einsum(equation = var_15993_equation_0, values = (var_15843_cast, var_15957_cast))[name = tensor<string, []>("op_15993_cast")];
            tensor<string, []> var_15995_equation_0 = const()[name = tensor<string, []>("op_15995_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15995_cast = einsum(equation = var_15995_equation_0, values = (var_15847_cast, var_15958_cast))[name = tensor<string, []>("op_15995_cast")];
            tensor<string, []> var_15997_equation_0 = const()[name = tensor<string, []>("op_15997_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15997_cast = einsum(equation = var_15997_equation_0, values = (var_15851_cast, var_15959_cast))[name = tensor<string, []>("op_15997_cast")];
            tensor<string, []> var_15999_equation_0 = const()[name = tensor<string, []>("op_15999_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_15999_cast = einsum(equation = var_15999_equation_0, values = (var_15855_cast, var_15960_cast))[name = tensor<string, []>("op_15999_cast")];
            tensor<string, []> var_16001_equation_0 = const()[name = tensor<string, []>("op_16001_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16001_cast = einsum(equation = var_16001_equation_0, values = (var_15859_cast, var_15961_cast))[name = tensor<string, []>("op_16001_cast")];
            tensor<bool, []> input_265_interleave_0 = const()[name = tensor<string, []>("input_265_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_265_cast = concat(axis = var_2634, interleave = input_265_interleave_0, values = (var_15963_cast, var_15965_cast, var_15967_cast, var_15969_cast, var_15971_cast, var_15973_cast, var_15975_cast, var_15977_cast, var_15979_cast, var_15981_cast, var_15983_cast, var_15985_cast, var_15987_cast, var_15989_cast, var_15991_cast, var_15993_cast, var_15995_cast, var_15997_cast, var_15999_cast, var_16001_cast))[name = tensor<string, []>("input_265_cast")];
            tensor<int32, [2]> var_16007 = const()[name = tensor<string, []>("op_16007"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_16009 = const()[name = tensor<string, []>("op_16009"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_16011_pad_type_0 = const()[name = tensor<string, []>("op_16011_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_16011_pad_0 = const()[name = tensor<string, []>("op_16011_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1250025152)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1253302016)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_16011_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_out_0_bias_to_fp16, dilations = var_16009, groups = var_2634, pad = var_16011_pad_0, pad_type = var_16011_pad_type_0, strides = var_16007, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn1_to_out_0_weight_to_fp16, x = input_265_cast)[name = tensor<string, []>("op_16011_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_111_cast = add(x = var_16011_cast, y = inputs_109_cast)[name = tensor<string, []>("inputs_111_cast")];
            tensor<int32, [1]> var_16015 = const()[name = tensor<string, []>("op_16015"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_111_cast = reduce_mean(axes = var_16015, keep_dims = var_2629, x = inputs_111_cast)[name = tensor<string, []>("channels_mean_111_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_111_cast = sub(x = inputs_111_cast, y = channels_mean_111_cast)[name = tensor<string, []>("zero_mean_111_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_111_cast = mul(x = zero_mean_111_cast, y = zero_mean_111_cast)[name = tensor<string, []>("zero_mean_sq_111_cast")];
            tensor<int32, [1]> var_16019 = const()[name = tensor<string, []>("op_16019"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_16020_cast = reduce_mean(axes = var_16019, keep_dims = var_2629, x = zero_mean_sq_111_cast)[name = tensor<string, []>("op_16020_cast")];
            tensor<fp16, []> var_16021_to_fp16 = const()[name = tensor<string, []>("op_16021_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_16022_cast = add(x = var_16020_cast, y = var_16021_to_fp16)[name = tensor<string, []>("op_16022_cast")];
            tensor<fp16, []> denom_111_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_111_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_111_cast = rsqrt(epsilon = denom_111_epsilon_0_to_fp16, x = var_16022_cast)[name = tensor<string, []>("denom_111_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_111_cast = mul(x = zero_mean_111_cast, y = denom_111_cast)[name = tensor<string, []>("out_111_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_16026_to_fp16 = const()[name = tensor<string, []>("op_16026_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1253304640)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_16027_cast = add(x = out_111_cast, y = var_16026_to_fp16)[name = tensor<string, []>("op_16027_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_16029_to_fp16 = const()[name = tensor<string, []>("op_16029_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1253307264)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_163_cast = mul(x = var_16027_cast, y = var_16029_to_fp16)[name = tensor<string, []>("hidden_states_163_cast")];
            tensor<int32, [2]> var_16036 = const()[name = tensor<string, []>("op_16036"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_16038 = const()[name = tensor<string, []>("op_16038"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_75_pad_type_0 = const()[name = tensor<string, []>("q_75_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_75_pad_0 = const()[name = tensor<string, []>("q_75_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1253309888)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_75_cast = conv(dilations = var_16038, groups = var_2634, pad = q_75_pad_0, pad_type = q_75_pad_type_0, strides = var_16036, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_q_weight_to_fp16, x = hidden_states_163_cast)[name = tensor<string, []>("q_75_cast")];
            tensor<int32, [2]> var_16042 = const()[name = tensor<string, []>("op_16042"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_16044 = const()[name = tensor<string, []>("op_16044"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_149_pad_type_0 = const()[name = tensor<string, []>("k_149_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_149_pad_0 = const()[name = tensor<string, []>("k_149_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1256586752)))];
            tensor<fp16, [2, 1280, 1, 77]> k_149_cast = conv(dilations = var_16044, groups = var_2634, pad = k_149_pad_0, pad_type = k_149_pad_type_0, strides = var_16042, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_149_cast")];
            tensor<int32, [2]> var_16048 = const()[name = tensor<string, []>("op_16048"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_16050 = const()[name = tensor<string, []>("op_16050"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_75_pad_type_0 = const()[name = tensor<string, []>("v_75_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_75_pad_0 = const()[name = tensor<string, []>("v_75_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1261829696)))];
            tensor<fp16, [2, 1280, 1, 77]> v_75_cast = conv(dilations = var_16050, groups = var_2634, pad = v_75_pad_0, pad_type = v_75_pad_type_0, strides = var_16048, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_75_cast")];
            tensor<int32, [4]> var_16054_begin_0 = const()[name = tensor<string, []>("op_16054_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_16054_end_0 = const()[name = tensor<string, []>("op_16054_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_16054_end_mask_0 = const()[name = tensor<string, []>("op_16054_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16054_cast = slice_by_index(begin = var_16054_begin_0, end = var_16054_end_0, end_mask = var_16054_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16054_cast")];
            tensor<int32, [4]> var_16058_begin_0 = const()[name = tensor<string, []>("op_16058_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_16058_end_0 = const()[name = tensor<string, []>("op_16058_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_16058_end_mask_0 = const()[name = tensor<string, []>("op_16058_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16058_cast = slice_by_index(begin = var_16058_begin_0, end = var_16058_end_0, end_mask = var_16058_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16058_cast")];
            tensor<int32, [4]> var_16062_begin_0 = const()[name = tensor<string, []>("op_16062_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_16062_end_0 = const()[name = tensor<string, []>("op_16062_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_16062_end_mask_0 = const()[name = tensor<string, []>("op_16062_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16062_cast = slice_by_index(begin = var_16062_begin_0, end = var_16062_end_0, end_mask = var_16062_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16062_cast")];
            tensor<int32, [4]> var_16066_begin_0 = const()[name = tensor<string, []>("op_16066_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_16066_end_0 = const()[name = tensor<string, []>("op_16066_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_16066_end_mask_0 = const()[name = tensor<string, []>("op_16066_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16066_cast = slice_by_index(begin = var_16066_begin_0, end = var_16066_end_0, end_mask = var_16066_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16066_cast")];
            tensor<int32, [4]> var_16070_begin_0 = const()[name = tensor<string, []>("op_16070_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_16070_end_0 = const()[name = tensor<string, []>("op_16070_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_16070_end_mask_0 = const()[name = tensor<string, []>("op_16070_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16070_cast = slice_by_index(begin = var_16070_begin_0, end = var_16070_end_0, end_mask = var_16070_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16070_cast")];
            tensor<int32, [4]> var_16074_begin_0 = const()[name = tensor<string, []>("op_16074_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_16074_end_0 = const()[name = tensor<string, []>("op_16074_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_16074_end_mask_0 = const()[name = tensor<string, []>("op_16074_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16074_cast = slice_by_index(begin = var_16074_begin_0, end = var_16074_end_0, end_mask = var_16074_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16074_cast")];
            tensor<int32, [4]> var_16078_begin_0 = const()[name = tensor<string, []>("op_16078_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_16078_end_0 = const()[name = tensor<string, []>("op_16078_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_16078_end_mask_0 = const()[name = tensor<string, []>("op_16078_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16078_cast = slice_by_index(begin = var_16078_begin_0, end = var_16078_end_0, end_mask = var_16078_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16078_cast")];
            tensor<int32, [4]> var_16082_begin_0 = const()[name = tensor<string, []>("op_16082_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_16082_end_0 = const()[name = tensor<string, []>("op_16082_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_16082_end_mask_0 = const()[name = tensor<string, []>("op_16082_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16082_cast = slice_by_index(begin = var_16082_begin_0, end = var_16082_end_0, end_mask = var_16082_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16082_cast")];
            tensor<int32, [4]> var_16086_begin_0 = const()[name = tensor<string, []>("op_16086_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_16086_end_0 = const()[name = tensor<string, []>("op_16086_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_16086_end_mask_0 = const()[name = tensor<string, []>("op_16086_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16086_cast = slice_by_index(begin = var_16086_begin_0, end = var_16086_end_0, end_mask = var_16086_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16086_cast")];
            tensor<int32, [4]> var_16090_begin_0 = const()[name = tensor<string, []>("op_16090_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_16090_end_0 = const()[name = tensor<string, []>("op_16090_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_16090_end_mask_0 = const()[name = tensor<string, []>("op_16090_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16090_cast = slice_by_index(begin = var_16090_begin_0, end = var_16090_end_0, end_mask = var_16090_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16090_cast")];
            tensor<int32, [4]> var_16094_begin_0 = const()[name = tensor<string, []>("op_16094_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_16094_end_0 = const()[name = tensor<string, []>("op_16094_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_16094_end_mask_0 = const()[name = tensor<string, []>("op_16094_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16094_cast = slice_by_index(begin = var_16094_begin_0, end = var_16094_end_0, end_mask = var_16094_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16094_cast")];
            tensor<int32, [4]> var_16098_begin_0 = const()[name = tensor<string, []>("op_16098_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_16098_end_0 = const()[name = tensor<string, []>("op_16098_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_16098_end_mask_0 = const()[name = tensor<string, []>("op_16098_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16098_cast = slice_by_index(begin = var_16098_begin_0, end = var_16098_end_0, end_mask = var_16098_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16098_cast")];
            tensor<int32, [4]> var_16102_begin_0 = const()[name = tensor<string, []>("op_16102_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_16102_end_0 = const()[name = tensor<string, []>("op_16102_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_16102_end_mask_0 = const()[name = tensor<string, []>("op_16102_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16102_cast = slice_by_index(begin = var_16102_begin_0, end = var_16102_end_0, end_mask = var_16102_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16102_cast")];
            tensor<int32, [4]> var_16106_begin_0 = const()[name = tensor<string, []>("op_16106_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_16106_end_0 = const()[name = tensor<string, []>("op_16106_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_16106_end_mask_0 = const()[name = tensor<string, []>("op_16106_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16106_cast = slice_by_index(begin = var_16106_begin_0, end = var_16106_end_0, end_mask = var_16106_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16106_cast")];
            tensor<int32, [4]> var_16110_begin_0 = const()[name = tensor<string, []>("op_16110_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_16110_end_0 = const()[name = tensor<string, []>("op_16110_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_16110_end_mask_0 = const()[name = tensor<string, []>("op_16110_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16110_cast = slice_by_index(begin = var_16110_begin_0, end = var_16110_end_0, end_mask = var_16110_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16110_cast")];
            tensor<int32, [4]> var_16114_begin_0 = const()[name = tensor<string, []>("op_16114_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_16114_end_0 = const()[name = tensor<string, []>("op_16114_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_16114_end_mask_0 = const()[name = tensor<string, []>("op_16114_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16114_cast = slice_by_index(begin = var_16114_begin_0, end = var_16114_end_0, end_mask = var_16114_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16114_cast")];
            tensor<int32, [4]> var_16118_begin_0 = const()[name = tensor<string, []>("op_16118_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_16118_end_0 = const()[name = tensor<string, []>("op_16118_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_16118_end_mask_0 = const()[name = tensor<string, []>("op_16118_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16118_cast = slice_by_index(begin = var_16118_begin_0, end = var_16118_end_0, end_mask = var_16118_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16118_cast")];
            tensor<int32, [4]> var_16122_begin_0 = const()[name = tensor<string, []>("op_16122_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_16122_end_0 = const()[name = tensor<string, []>("op_16122_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_16122_end_mask_0 = const()[name = tensor<string, []>("op_16122_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16122_cast = slice_by_index(begin = var_16122_begin_0, end = var_16122_end_0, end_mask = var_16122_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16122_cast")];
            tensor<int32, [4]> var_16126_begin_0 = const()[name = tensor<string, []>("op_16126_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_16126_end_0 = const()[name = tensor<string, []>("op_16126_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_16126_end_mask_0 = const()[name = tensor<string, []>("op_16126_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16126_cast = slice_by_index(begin = var_16126_begin_0, end = var_16126_end_0, end_mask = var_16126_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16126_cast")];
            tensor<int32, [4]> var_16130_begin_0 = const()[name = tensor<string, []>("op_16130_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_16130_end_0 = const()[name = tensor<string, []>("op_16130_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_16130_end_mask_0 = const()[name = tensor<string, []>("op_16130_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16130_cast = slice_by_index(begin = var_16130_begin_0, end = var_16130_end_0, end_mask = var_16130_end_mask_0, x = q_75_cast)[name = tensor<string, []>("op_16130_cast")];
            tensor<int32, [4]> k_151_perm_0 = const()[name = tensor<string, []>("k_151_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_16137_begin_0 = const()[name = tensor<string, []>("op_16137_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_16137_end_0 = const()[name = tensor<string, []>("op_16137_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_16137_end_mask_0 = const()[name = tensor<string, []>("op_16137_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_102 = transpose(perm = k_151_perm_0, x = k_149_cast)[name = tensor<string, []>("transpose_102")];
            tensor<fp16, [2, 77, 1, 64]> var_16137_cast = slice_by_index(begin = var_16137_begin_0, end = var_16137_end_0, end_mask = var_16137_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16137_cast")];
            tensor<int32, [4]> var_16141_begin_0 = const()[name = tensor<string, []>("op_16141_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_16141_end_0 = const()[name = tensor<string, []>("op_16141_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_16141_end_mask_0 = const()[name = tensor<string, []>("op_16141_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_16141_cast = slice_by_index(begin = var_16141_begin_0, end = var_16141_end_0, end_mask = var_16141_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16141_cast")];
            tensor<int32, [4]> var_16145_begin_0 = const()[name = tensor<string, []>("op_16145_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_16145_end_0 = const()[name = tensor<string, []>("op_16145_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_16145_end_mask_0 = const()[name = tensor<string, []>("op_16145_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_16145_cast = slice_by_index(begin = var_16145_begin_0, end = var_16145_end_0, end_mask = var_16145_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16145_cast")];
            tensor<int32, [4]> var_16149_begin_0 = const()[name = tensor<string, []>("op_16149_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_16149_end_0 = const()[name = tensor<string, []>("op_16149_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_16149_end_mask_0 = const()[name = tensor<string, []>("op_16149_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_16149_cast = slice_by_index(begin = var_16149_begin_0, end = var_16149_end_0, end_mask = var_16149_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16149_cast")];
            tensor<int32, [4]> var_16153_begin_0 = const()[name = tensor<string, []>("op_16153_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_16153_end_0 = const()[name = tensor<string, []>("op_16153_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_16153_end_mask_0 = const()[name = tensor<string, []>("op_16153_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_16153_cast = slice_by_index(begin = var_16153_begin_0, end = var_16153_end_0, end_mask = var_16153_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16153_cast")];
            tensor<int32, [4]> var_16157_begin_0 = const()[name = tensor<string, []>("op_16157_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_16157_end_0 = const()[name = tensor<string, []>("op_16157_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_16157_end_mask_0 = const()[name = tensor<string, []>("op_16157_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_16157_cast = slice_by_index(begin = var_16157_begin_0, end = var_16157_end_0, end_mask = var_16157_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16157_cast")];
            tensor<int32, [4]> var_16161_begin_0 = const()[name = tensor<string, []>("op_16161_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_16161_end_0 = const()[name = tensor<string, []>("op_16161_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_16161_end_mask_0 = const()[name = tensor<string, []>("op_16161_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_16161_cast = slice_by_index(begin = var_16161_begin_0, end = var_16161_end_0, end_mask = var_16161_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16161_cast")];
            tensor<int32, [4]> var_16165_begin_0 = const()[name = tensor<string, []>("op_16165_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_16165_end_0 = const()[name = tensor<string, []>("op_16165_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_16165_end_mask_0 = const()[name = tensor<string, []>("op_16165_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_16165_cast = slice_by_index(begin = var_16165_begin_0, end = var_16165_end_0, end_mask = var_16165_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16165_cast")];
            tensor<int32, [4]> var_16169_begin_0 = const()[name = tensor<string, []>("op_16169_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_16169_end_0 = const()[name = tensor<string, []>("op_16169_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_16169_end_mask_0 = const()[name = tensor<string, []>("op_16169_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_16169_cast = slice_by_index(begin = var_16169_begin_0, end = var_16169_end_0, end_mask = var_16169_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16169_cast")];
            tensor<int32, [4]> var_16173_begin_0 = const()[name = tensor<string, []>("op_16173_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_16173_end_0 = const()[name = tensor<string, []>("op_16173_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_16173_end_mask_0 = const()[name = tensor<string, []>("op_16173_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_16173_cast = slice_by_index(begin = var_16173_begin_0, end = var_16173_end_0, end_mask = var_16173_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16173_cast")];
            tensor<int32, [4]> var_16177_begin_0 = const()[name = tensor<string, []>("op_16177_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_16177_end_0 = const()[name = tensor<string, []>("op_16177_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_16177_end_mask_0 = const()[name = tensor<string, []>("op_16177_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_16177_cast = slice_by_index(begin = var_16177_begin_0, end = var_16177_end_0, end_mask = var_16177_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16177_cast")];
            tensor<int32, [4]> var_16181_begin_0 = const()[name = tensor<string, []>("op_16181_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_16181_end_0 = const()[name = tensor<string, []>("op_16181_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_16181_end_mask_0 = const()[name = tensor<string, []>("op_16181_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_16181_cast = slice_by_index(begin = var_16181_begin_0, end = var_16181_end_0, end_mask = var_16181_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16181_cast")];
            tensor<int32, [4]> var_16185_begin_0 = const()[name = tensor<string, []>("op_16185_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_16185_end_0 = const()[name = tensor<string, []>("op_16185_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_16185_end_mask_0 = const()[name = tensor<string, []>("op_16185_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_16185_cast = slice_by_index(begin = var_16185_begin_0, end = var_16185_end_0, end_mask = var_16185_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16185_cast")];
            tensor<int32, [4]> var_16189_begin_0 = const()[name = tensor<string, []>("op_16189_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_16189_end_0 = const()[name = tensor<string, []>("op_16189_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_16189_end_mask_0 = const()[name = tensor<string, []>("op_16189_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_16189_cast = slice_by_index(begin = var_16189_begin_0, end = var_16189_end_0, end_mask = var_16189_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16189_cast")];
            tensor<int32, [4]> var_16193_begin_0 = const()[name = tensor<string, []>("op_16193_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_16193_end_0 = const()[name = tensor<string, []>("op_16193_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_16193_end_mask_0 = const()[name = tensor<string, []>("op_16193_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_16193_cast = slice_by_index(begin = var_16193_begin_0, end = var_16193_end_0, end_mask = var_16193_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16193_cast")];
            tensor<int32, [4]> var_16197_begin_0 = const()[name = tensor<string, []>("op_16197_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_16197_end_0 = const()[name = tensor<string, []>("op_16197_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_16197_end_mask_0 = const()[name = tensor<string, []>("op_16197_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_16197_cast = slice_by_index(begin = var_16197_begin_0, end = var_16197_end_0, end_mask = var_16197_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16197_cast")];
            tensor<int32, [4]> var_16201_begin_0 = const()[name = tensor<string, []>("op_16201_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_16201_end_0 = const()[name = tensor<string, []>("op_16201_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_16201_end_mask_0 = const()[name = tensor<string, []>("op_16201_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_16201_cast = slice_by_index(begin = var_16201_begin_0, end = var_16201_end_0, end_mask = var_16201_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16201_cast")];
            tensor<int32, [4]> var_16205_begin_0 = const()[name = tensor<string, []>("op_16205_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_16205_end_0 = const()[name = tensor<string, []>("op_16205_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_16205_end_mask_0 = const()[name = tensor<string, []>("op_16205_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_16205_cast = slice_by_index(begin = var_16205_begin_0, end = var_16205_end_0, end_mask = var_16205_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16205_cast")];
            tensor<int32, [4]> var_16209_begin_0 = const()[name = tensor<string, []>("op_16209_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_16209_end_0 = const()[name = tensor<string, []>("op_16209_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_16209_end_mask_0 = const()[name = tensor<string, []>("op_16209_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_16209_cast = slice_by_index(begin = var_16209_begin_0, end = var_16209_end_0, end_mask = var_16209_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16209_cast")];
            tensor<int32, [4]> var_16213_begin_0 = const()[name = tensor<string, []>("op_16213_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_16213_end_0 = const()[name = tensor<string, []>("op_16213_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_16213_end_mask_0 = const()[name = tensor<string, []>("op_16213_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_16213_cast = slice_by_index(begin = var_16213_begin_0, end = var_16213_end_0, end_mask = var_16213_end_mask_0, x = transpose_102)[name = tensor<string, []>("op_16213_cast")];
            tensor<int32, [4]> var_16215_begin_0 = const()[name = tensor<string, []>("op_16215_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_16215_end_0 = const()[name = tensor<string, []>("op_16215_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_16215_end_mask_0 = const()[name = tensor<string, []>("op_16215_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16215_cast = slice_by_index(begin = var_16215_begin_0, end = var_16215_end_0, end_mask = var_16215_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16215_cast")];
            tensor<int32, [4]> var_16219_begin_0 = const()[name = tensor<string, []>("op_16219_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_16219_end_0 = const()[name = tensor<string, []>("op_16219_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_16219_end_mask_0 = const()[name = tensor<string, []>("op_16219_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16219_cast = slice_by_index(begin = var_16219_begin_0, end = var_16219_end_0, end_mask = var_16219_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16219_cast")];
            tensor<int32, [4]> var_16223_begin_0 = const()[name = tensor<string, []>("op_16223_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_16223_end_0 = const()[name = tensor<string, []>("op_16223_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_16223_end_mask_0 = const()[name = tensor<string, []>("op_16223_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16223_cast = slice_by_index(begin = var_16223_begin_0, end = var_16223_end_0, end_mask = var_16223_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16223_cast")];
            tensor<int32, [4]> var_16227_begin_0 = const()[name = tensor<string, []>("op_16227_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_16227_end_0 = const()[name = tensor<string, []>("op_16227_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_16227_end_mask_0 = const()[name = tensor<string, []>("op_16227_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16227_cast = slice_by_index(begin = var_16227_begin_0, end = var_16227_end_0, end_mask = var_16227_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16227_cast")];
            tensor<int32, [4]> var_16231_begin_0 = const()[name = tensor<string, []>("op_16231_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_16231_end_0 = const()[name = tensor<string, []>("op_16231_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_16231_end_mask_0 = const()[name = tensor<string, []>("op_16231_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16231_cast = slice_by_index(begin = var_16231_begin_0, end = var_16231_end_0, end_mask = var_16231_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16231_cast")];
            tensor<int32, [4]> var_16235_begin_0 = const()[name = tensor<string, []>("op_16235_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_16235_end_0 = const()[name = tensor<string, []>("op_16235_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_16235_end_mask_0 = const()[name = tensor<string, []>("op_16235_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16235_cast = slice_by_index(begin = var_16235_begin_0, end = var_16235_end_0, end_mask = var_16235_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16235_cast")];
            tensor<int32, [4]> var_16239_begin_0 = const()[name = tensor<string, []>("op_16239_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_16239_end_0 = const()[name = tensor<string, []>("op_16239_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_16239_end_mask_0 = const()[name = tensor<string, []>("op_16239_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16239_cast = slice_by_index(begin = var_16239_begin_0, end = var_16239_end_0, end_mask = var_16239_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16239_cast")];
            tensor<int32, [4]> var_16243_begin_0 = const()[name = tensor<string, []>("op_16243_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_16243_end_0 = const()[name = tensor<string, []>("op_16243_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_16243_end_mask_0 = const()[name = tensor<string, []>("op_16243_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16243_cast = slice_by_index(begin = var_16243_begin_0, end = var_16243_end_0, end_mask = var_16243_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16243_cast")];
            tensor<int32, [4]> var_16247_begin_0 = const()[name = tensor<string, []>("op_16247_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_16247_end_0 = const()[name = tensor<string, []>("op_16247_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_16247_end_mask_0 = const()[name = tensor<string, []>("op_16247_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16247_cast = slice_by_index(begin = var_16247_begin_0, end = var_16247_end_0, end_mask = var_16247_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16247_cast")];
            tensor<int32, [4]> var_16251_begin_0 = const()[name = tensor<string, []>("op_16251_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_16251_end_0 = const()[name = tensor<string, []>("op_16251_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_16251_end_mask_0 = const()[name = tensor<string, []>("op_16251_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16251_cast = slice_by_index(begin = var_16251_begin_0, end = var_16251_end_0, end_mask = var_16251_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16251_cast")];
            tensor<int32, [4]> var_16255_begin_0 = const()[name = tensor<string, []>("op_16255_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_16255_end_0 = const()[name = tensor<string, []>("op_16255_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_16255_end_mask_0 = const()[name = tensor<string, []>("op_16255_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16255_cast = slice_by_index(begin = var_16255_begin_0, end = var_16255_end_0, end_mask = var_16255_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16255_cast")];
            tensor<int32, [4]> var_16259_begin_0 = const()[name = tensor<string, []>("op_16259_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_16259_end_0 = const()[name = tensor<string, []>("op_16259_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_16259_end_mask_0 = const()[name = tensor<string, []>("op_16259_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16259_cast = slice_by_index(begin = var_16259_begin_0, end = var_16259_end_0, end_mask = var_16259_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16259_cast")];
            tensor<int32, [4]> var_16263_begin_0 = const()[name = tensor<string, []>("op_16263_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_16263_end_0 = const()[name = tensor<string, []>("op_16263_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_16263_end_mask_0 = const()[name = tensor<string, []>("op_16263_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16263_cast = slice_by_index(begin = var_16263_begin_0, end = var_16263_end_0, end_mask = var_16263_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16263_cast")];
            tensor<int32, [4]> var_16267_begin_0 = const()[name = tensor<string, []>("op_16267_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_16267_end_0 = const()[name = tensor<string, []>("op_16267_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_16267_end_mask_0 = const()[name = tensor<string, []>("op_16267_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16267_cast = slice_by_index(begin = var_16267_begin_0, end = var_16267_end_0, end_mask = var_16267_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16267_cast")];
            tensor<int32, [4]> var_16271_begin_0 = const()[name = tensor<string, []>("op_16271_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_16271_end_0 = const()[name = tensor<string, []>("op_16271_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_16271_end_mask_0 = const()[name = tensor<string, []>("op_16271_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16271_cast = slice_by_index(begin = var_16271_begin_0, end = var_16271_end_0, end_mask = var_16271_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16271_cast")];
            tensor<int32, [4]> var_16275_begin_0 = const()[name = tensor<string, []>("op_16275_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_16275_end_0 = const()[name = tensor<string, []>("op_16275_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_16275_end_mask_0 = const()[name = tensor<string, []>("op_16275_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16275_cast = slice_by_index(begin = var_16275_begin_0, end = var_16275_end_0, end_mask = var_16275_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16275_cast")];
            tensor<int32, [4]> var_16279_begin_0 = const()[name = tensor<string, []>("op_16279_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_16279_end_0 = const()[name = tensor<string, []>("op_16279_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_16279_end_mask_0 = const()[name = tensor<string, []>("op_16279_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16279_cast = slice_by_index(begin = var_16279_begin_0, end = var_16279_end_0, end_mask = var_16279_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16279_cast")];
            tensor<int32, [4]> var_16283_begin_0 = const()[name = tensor<string, []>("op_16283_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_16283_end_0 = const()[name = tensor<string, []>("op_16283_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_16283_end_mask_0 = const()[name = tensor<string, []>("op_16283_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16283_cast = slice_by_index(begin = var_16283_begin_0, end = var_16283_end_0, end_mask = var_16283_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16283_cast")];
            tensor<int32, [4]> var_16287_begin_0 = const()[name = tensor<string, []>("op_16287_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_16287_end_0 = const()[name = tensor<string, []>("op_16287_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_16287_end_mask_0 = const()[name = tensor<string, []>("op_16287_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16287_cast = slice_by_index(begin = var_16287_begin_0, end = var_16287_end_0, end_mask = var_16287_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16287_cast")];
            tensor<int32, [4]> var_16291_begin_0 = const()[name = tensor<string, []>("op_16291_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_16291_end_0 = const()[name = tensor<string, []>("op_16291_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_16291_end_mask_0 = const()[name = tensor<string, []>("op_16291_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_16291_cast = slice_by_index(begin = var_16291_begin_0, end = var_16291_end_0, end_mask = var_16291_end_mask_0, x = v_75_cast)[name = tensor<string, []>("op_16291_cast")];
            tensor<string, []> var_16295_equation_0 = const()[name = tensor<string, []>("op_16295_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16295_cast = einsum(equation = var_16295_equation_0, values = (var_16137_cast, var_16054_cast))[name = tensor<string, []>("op_16295_cast")];
            tensor<fp16, []> var_16296_to_fp16 = const()[name = tensor<string, []>("op_16296_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1321_cast = mul(x = var_16295_cast, y = var_16296_to_fp16)[name = tensor<string, []>("aw_1321_cast")];
            tensor<string, []> var_16299_equation_0 = const()[name = tensor<string, []>("op_16299_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16299_cast = einsum(equation = var_16299_equation_0, values = (var_16141_cast, var_16058_cast))[name = tensor<string, []>("op_16299_cast")];
            tensor<fp16, []> var_16300_to_fp16 = const()[name = tensor<string, []>("op_16300_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1323_cast = mul(x = var_16299_cast, y = var_16300_to_fp16)[name = tensor<string, []>("aw_1323_cast")];
            tensor<string, []> var_16303_equation_0 = const()[name = tensor<string, []>("op_16303_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16303_cast = einsum(equation = var_16303_equation_0, values = (var_16145_cast, var_16062_cast))[name = tensor<string, []>("op_16303_cast")];
            tensor<fp16, []> var_16304_to_fp16 = const()[name = tensor<string, []>("op_16304_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1325_cast = mul(x = var_16303_cast, y = var_16304_to_fp16)[name = tensor<string, []>("aw_1325_cast")];
            tensor<string, []> var_16307_equation_0 = const()[name = tensor<string, []>("op_16307_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16307_cast = einsum(equation = var_16307_equation_0, values = (var_16149_cast, var_16066_cast))[name = tensor<string, []>("op_16307_cast")];
            tensor<fp16, []> var_16308_to_fp16 = const()[name = tensor<string, []>("op_16308_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1327_cast = mul(x = var_16307_cast, y = var_16308_to_fp16)[name = tensor<string, []>("aw_1327_cast")];
            tensor<string, []> var_16311_equation_0 = const()[name = tensor<string, []>("op_16311_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16311_cast = einsum(equation = var_16311_equation_0, values = (var_16153_cast, var_16070_cast))[name = tensor<string, []>("op_16311_cast")];
            tensor<fp16, []> var_16312_to_fp16 = const()[name = tensor<string, []>("op_16312_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1329_cast = mul(x = var_16311_cast, y = var_16312_to_fp16)[name = tensor<string, []>("aw_1329_cast")];
            tensor<string, []> var_16315_equation_0 = const()[name = tensor<string, []>("op_16315_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16315_cast = einsum(equation = var_16315_equation_0, values = (var_16157_cast, var_16074_cast))[name = tensor<string, []>("op_16315_cast")];
            tensor<fp16, []> var_16316_to_fp16 = const()[name = tensor<string, []>("op_16316_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1331_cast = mul(x = var_16315_cast, y = var_16316_to_fp16)[name = tensor<string, []>("aw_1331_cast")];
            tensor<string, []> var_16319_equation_0 = const()[name = tensor<string, []>("op_16319_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16319_cast = einsum(equation = var_16319_equation_0, values = (var_16161_cast, var_16078_cast))[name = tensor<string, []>("op_16319_cast")];
            tensor<fp16, []> var_16320_to_fp16 = const()[name = tensor<string, []>("op_16320_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1333_cast = mul(x = var_16319_cast, y = var_16320_to_fp16)[name = tensor<string, []>("aw_1333_cast")];
            tensor<string, []> var_16323_equation_0 = const()[name = tensor<string, []>("op_16323_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16323_cast = einsum(equation = var_16323_equation_0, values = (var_16165_cast, var_16082_cast))[name = tensor<string, []>("op_16323_cast")];
            tensor<fp16, []> var_16324_to_fp16 = const()[name = tensor<string, []>("op_16324_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1335_cast = mul(x = var_16323_cast, y = var_16324_to_fp16)[name = tensor<string, []>("aw_1335_cast")];
            tensor<string, []> var_16327_equation_0 = const()[name = tensor<string, []>("op_16327_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16327_cast = einsum(equation = var_16327_equation_0, values = (var_16169_cast, var_16086_cast))[name = tensor<string, []>("op_16327_cast")];
            tensor<fp16, []> var_16328_to_fp16 = const()[name = tensor<string, []>("op_16328_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1337_cast = mul(x = var_16327_cast, y = var_16328_to_fp16)[name = tensor<string, []>("aw_1337_cast")];
            tensor<string, []> var_16331_equation_0 = const()[name = tensor<string, []>("op_16331_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16331_cast = einsum(equation = var_16331_equation_0, values = (var_16173_cast, var_16090_cast))[name = tensor<string, []>("op_16331_cast")];
            tensor<fp16, []> var_16332_to_fp16 = const()[name = tensor<string, []>("op_16332_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1339_cast = mul(x = var_16331_cast, y = var_16332_to_fp16)[name = tensor<string, []>("aw_1339_cast")];
            tensor<string, []> var_16335_equation_0 = const()[name = tensor<string, []>("op_16335_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16335_cast = einsum(equation = var_16335_equation_0, values = (var_16177_cast, var_16094_cast))[name = tensor<string, []>("op_16335_cast")];
            tensor<fp16, []> var_16336_to_fp16 = const()[name = tensor<string, []>("op_16336_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1341_cast = mul(x = var_16335_cast, y = var_16336_to_fp16)[name = tensor<string, []>("aw_1341_cast")];
            tensor<string, []> var_16339_equation_0 = const()[name = tensor<string, []>("op_16339_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16339_cast = einsum(equation = var_16339_equation_0, values = (var_16181_cast, var_16098_cast))[name = tensor<string, []>("op_16339_cast")];
            tensor<fp16, []> var_16340_to_fp16 = const()[name = tensor<string, []>("op_16340_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1343_cast = mul(x = var_16339_cast, y = var_16340_to_fp16)[name = tensor<string, []>("aw_1343_cast")];
            tensor<string, []> var_16343_equation_0 = const()[name = tensor<string, []>("op_16343_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16343_cast = einsum(equation = var_16343_equation_0, values = (var_16185_cast, var_16102_cast))[name = tensor<string, []>("op_16343_cast")];
            tensor<fp16, []> var_16344_to_fp16 = const()[name = tensor<string, []>("op_16344_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1345_cast = mul(x = var_16343_cast, y = var_16344_to_fp16)[name = tensor<string, []>("aw_1345_cast")];
            tensor<string, []> var_16347_equation_0 = const()[name = tensor<string, []>("op_16347_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16347_cast = einsum(equation = var_16347_equation_0, values = (var_16189_cast, var_16106_cast))[name = tensor<string, []>("op_16347_cast")];
            tensor<fp16, []> var_16348_to_fp16 = const()[name = tensor<string, []>("op_16348_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1347_cast = mul(x = var_16347_cast, y = var_16348_to_fp16)[name = tensor<string, []>("aw_1347_cast")];
            tensor<string, []> var_16351_equation_0 = const()[name = tensor<string, []>("op_16351_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16351_cast = einsum(equation = var_16351_equation_0, values = (var_16193_cast, var_16110_cast))[name = tensor<string, []>("op_16351_cast")];
            tensor<fp16, []> var_16352_to_fp16 = const()[name = tensor<string, []>("op_16352_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1349_cast = mul(x = var_16351_cast, y = var_16352_to_fp16)[name = tensor<string, []>("aw_1349_cast")];
            tensor<string, []> var_16355_equation_0 = const()[name = tensor<string, []>("op_16355_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16355_cast = einsum(equation = var_16355_equation_0, values = (var_16197_cast, var_16114_cast))[name = tensor<string, []>("op_16355_cast")];
            tensor<fp16, []> var_16356_to_fp16 = const()[name = tensor<string, []>("op_16356_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1351_cast = mul(x = var_16355_cast, y = var_16356_to_fp16)[name = tensor<string, []>("aw_1351_cast")];
            tensor<string, []> var_16359_equation_0 = const()[name = tensor<string, []>("op_16359_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16359_cast = einsum(equation = var_16359_equation_0, values = (var_16201_cast, var_16118_cast))[name = tensor<string, []>("op_16359_cast")];
            tensor<fp16, []> var_16360_to_fp16 = const()[name = tensor<string, []>("op_16360_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1353_cast = mul(x = var_16359_cast, y = var_16360_to_fp16)[name = tensor<string, []>("aw_1353_cast")];
            tensor<string, []> var_16363_equation_0 = const()[name = tensor<string, []>("op_16363_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16363_cast = einsum(equation = var_16363_equation_0, values = (var_16205_cast, var_16122_cast))[name = tensor<string, []>("op_16363_cast")];
            tensor<fp16, []> var_16364_to_fp16 = const()[name = tensor<string, []>("op_16364_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1355_cast = mul(x = var_16363_cast, y = var_16364_to_fp16)[name = tensor<string, []>("aw_1355_cast")];
            tensor<string, []> var_16367_equation_0 = const()[name = tensor<string, []>("op_16367_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16367_cast = einsum(equation = var_16367_equation_0, values = (var_16209_cast, var_16126_cast))[name = tensor<string, []>("op_16367_cast")];
            tensor<fp16, []> var_16368_to_fp16 = const()[name = tensor<string, []>("op_16368_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1357_cast = mul(x = var_16367_cast, y = var_16368_to_fp16)[name = tensor<string, []>("aw_1357_cast")];
            tensor<string, []> var_16371_equation_0 = const()[name = tensor<string, []>("op_16371_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_16371_cast = einsum(equation = var_16371_equation_0, values = (var_16213_cast, var_16130_cast))[name = tensor<string, []>("op_16371_cast")];
            tensor<fp16, []> var_16372_to_fp16 = const()[name = tensor<string, []>("op_16372_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1359_cast = mul(x = var_16371_cast, y = var_16372_to_fp16)[name = tensor<string, []>("aw_1359_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16374_cast = softmax(axis = var_2634, x = aw_1321_cast)[name = tensor<string, []>("op_16374_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16375_cast = softmax(axis = var_2634, x = aw_1323_cast)[name = tensor<string, []>("op_16375_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16376_cast = softmax(axis = var_2634, x = aw_1325_cast)[name = tensor<string, []>("op_16376_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16377_cast = softmax(axis = var_2634, x = aw_1327_cast)[name = tensor<string, []>("op_16377_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16378_cast = softmax(axis = var_2634, x = aw_1329_cast)[name = tensor<string, []>("op_16378_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16379_cast = softmax(axis = var_2634, x = aw_1331_cast)[name = tensor<string, []>("op_16379_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16380_cast = softmax(axis = var_2634, x = aw_1333_cast)[name = tensor<string, []>("op_16380_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16381_cast = softmax(axis = var_2634, x = aw_1335_cast)[name = tensor<string, []>("op_16381_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16382_cast = softmax(axis = var_2634, x = aw_1337_cast)[name = tensor<string, []>("op_16382_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16383_cast = softmax(axis = var_2634, x = aw_1339_cast)[name = tensor<string, []>("op_16383_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16384_cast = softmax(axis = var_2634, x = aw_1341_cast)[name = tensor<string, []>("op_16384_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16385_cast = softmax(axis = var_2634, x = aw_1343_cast)[name = tensor<string, []>("op_16385_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16386_cast = softmax(axis = var_2634, x = aw_1345_cast)[name = tensor<string, []>("op_16386_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16387_cast = softmax(axis = var_2634, x = aw_1347_cast)[name = tensor<string, []>("op_16387_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16388_cast = softmax(axis = var_2634, x = aw_1349_cast)[name = tensor<string, []>("op_16388_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16389_cast = softmax(axis = var_2634, x = aw_1351_cast)[name = tensor<string, []>("op_16389_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16390_cast = softmax(axis = var_2634, x = aw_1353_cast)[name = tensor<string, []>("op_16390_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16391_cast = softmax(axis = var_2634, x = aw_1355_cast)[name = tensor<string, []>("op_16391_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16392_cast = softmax(axis = var_2634, x = aw_1357_cast)[name = tensor<string, []>("op_16392_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_16393_cast = softmax(axis = var_2634, x = aw_1359_cast)[name = tensor<string, []>("op_16393_cast")];
            tensor<string, []> var_16395_equation_0 = const()[name = tensor<string, []>("op_16395_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16395_cast = einsum(equation = var_16395_equation_0, values = (var_16215_cast, var_16374_cast))[name = tensor<string, []>("op_16395_cast")];
            tensor<string, []> var_16397_equation_0 = const()[name = tensor<string, []>("op_16397_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16397_cast = einsum(equation = var_16397_equation_0, values = (var_16219_cast, var_16375_cast))[name = tensor<string, []>("op_16397_cast")];
            tensor<string, []> var_16399_equation_0 = const()[name = tensor<string, []>("op_16399_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16399_cast = einsum(equation = var_16399_equation_0, values = (var_16223_cast, var_16376_cast))[name = tensor<string, []>("op_16399_cast")];
            tensor<string, []> var_16401_equation_0 = const()[name = tensor<string, []>("op_16401_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16401_cast = einsum(equation = var_16401_equation_0, values = (var_16227_cast, var_16377_cast))[name = tensor<string, []>("op_16401_cast")];
            tensor<string, []> var_16403_equation_0 = const()[name = tensor<string, []>("op_16403_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16403_cast = einsum(equation = var_16403_equation_0, values = (var_16231_cast, var_16378_cast))[name = tensor<string, []>("op_16403_cast")];
            tensor<string, []> var_16405_equation_0 = const()[name = tensor<string, []>("op_16405_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16405_cast = einsum(equation = var_16405_equation_0, values = (var_16235_cast, var_16379_cast))[name = tensor<string, []>("op_16405_cast")];
            tensor<string, []> var_16407_equation_0 = const()[name = tensor<string, []>("op_16407_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16407_cast = einsum(equation = var_16407_equation_0, values = (var_16239_cast, var_16380_cast))[name = tensor<string, []>("op_16407_cast")];
            tensor<string, []> var_16409_equation_0 = const()[name = tensor<string, []>("op_16409_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16409_cast = einsum(equation = var_16409_equation_0, values = (var_16243_cast, var_16381_cast))[name = tensor<string, []>("op_16409_cast")];
            tensor<string, []> var_16411_equation_0 = const()[name = tensor<string, []>("op_16411_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16411_cast = einsum(equation = var_16411_equation_0, values = (var_16247_cast, var_16382_cast))[name = tensor<string, []>("op_16411_cast")];
            tensor<string, []> var_16413_equation_0 = const()[name = tensor<string, []>("op_16413_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16413_cast = einsum(equation = var_16413_equation_0, values = (var_16251_cast, var_16383_cast))[name = tensor<string, []>("op_16413_cast")];
            tensor<string, []> var_16415_equation_0 = const()[name = tensor<string, []>("op_16415_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16415_cast = einsum(equation = var_16415_equation_0, values = (var_16255_cast, var_16384_cast))[name = tensor<string, []>("op_16415_cast")];
            tensor<string, []> var_16417_equation_0 = const()[name = tensor<string, []>("op_16417_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16417_cast = einsum(equation = var_16417_equation_0, values = (var_16259_cast, var_16385_cast))[name = tensor<string, []>("op_16417_cast")];
            tensor<string, []> var_16419_equation_0 = const()[name = tensor<string, []>("op_16419_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16419_cast = einsum(equation = var_16419_equation_0, values = (var_16263_cast, var_16386_cast))[name = tensor<string, []>("op_16419_cast")];
            tensor<string, []> var_16421_equation_0 = const()[name = tensor<string, []>("op_16421_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16421_cast = einsum(equation = var_16421_equation_0, values = (var_16267_cast, var_16387_cast))[name = tensor<string, []>("op_16421_cast")];
            tensor<string, []> var_16423_equation_0 = const()[name = tensor<string, []>("op_16423_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16423_cast = einsum(equation = var_16423_equation_0, values = (var_16271_cast, var_16388_cast))[name = tensor<string, []>("op_16423_cast")];
            tensor<string, []> var_16425_equation_0 = const()[name = tensor<string, []>("op_16425_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16425_cast = einsum(equation = var_16425_equation_0, values = (var_16275_cast, var_16389_cast))[name = tensor<string, []>("op_16425_cast")];
            tensor<string, []> var_16427_equation_0 = const()[name = tensor<string, []>("op_16427_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16427_cast = einsum(equation = var_16427_equation_0, values = (var_16279_cast, var_16390_cast))[name = tensor<string, []>("op_16427_cast")];
            tensor<string, []> var_16429_equation_0 = const()[name = tensor<string, []>("op_16429_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16429_cast = einsum(equation = var_16429_equation_0, values = (var_16283_cast, var_16391_cast))[name = tensor<string, []>("op_16429_cast")];
            tensor<string, []> var_16431_equation_0 = const()[name = tensor<string, []>("op_16431_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16431_cast = einsum(equation = var_16431_equation_0, values = (var_16287_cast, var_16392_cast))[name = tensor<string, []>("op_16431_cast")];
            tensor<string, []> var_16433_equation_0 = const()[name = tensor<string, []>("op_16433_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16433_cast = einsum(equation = var_16433_equation_0, values = (var_16291_cast, var_16393_cast))[name = tensor<string, []>("op_16433_cast")];
            tensor<bool, []> input_267_interleave_0 = const()[name = tensor<string, []>("input_267_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_267_cast = concat(axis = var_2634, interleave = input_267_interleave_0, values = (var_16395_cast, var_16397_cast, var_16399_cast, var_16401_cast, var_16403_cast, var_16405_cast, var_16407_cast, var_16409_cast, var_16411_cast, var_16413_cast, var_16415_cast, var_16417_cast, var_16419_cast, var_16421_cast, var_16423_cast, var_16425_cast, var_16427_cast, var_16429_cast, var_16431_cast, var_16433_cast))[name = tensor<string, []>("input_267_cast")];
            tensor<int32, [2]> var_16439 = const()[name = tensor<string, []>("op_16439"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_16441 = const()[name = tensor<string, []>("op_16441"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_16443_pad_type_0 = const()[name = tensor<string, []>("op_16443_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_16443_pad_0 = const()[name = tensor<string, []>("op_16443_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1267072640)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1270349504)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_16443_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_out_0_bias_to_fp16, dilations = var_16441, groups = var_2634, pad = var_16443_pad_0, pad_type = var_16443_pad_type_0, strides = var_16439, weight = down_blocks_2_attentions_1_transformer_blocks_4_attn2_to_out_0_weight_to_fp16, x = input_267_cast)[name = tensor<string, []>("op_16443_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_113_cast = add(x = var_16443_cast, y = inputs_111_cast)[name = tensor<string, []>("inputs_113_cast")];
            tensor<int32, [1]> var_16447 = const()[name = tensor<string, []>("op_16447"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_113_cast = reduce_mean(axes = var_16447, keep_dims = var_2629, x = inputs_113_cast)[name = tensor<string, []>("channels_mean_113_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_113_cast = sub(x = inputs_113_cast, y = channels_mean_113_cast)[name = tensor<string, []>("zero_mean_113_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_113_cast = mul(x = zero_mean_113_cast, y = zero_mean_113_cast)[name = tensor<string, []>("zero_mean_sq_113_cast")];
            tensor<int32, [1]> var_16451 = const()[name = tensor<string, []>("op_16451"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_16452_cast = reduce_mean(axes = var_16451, keep_dims = var_2629, x = zero_mean_sq_113_cast)[name = tensor<string, []>("op_16452_cast")];
            tensor<fp16, []> var_16453_to_fp16 = const()[name = tensor<string, []>("op_16453_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_16454_cast = add(x = var_16452_cast, y = var_16453_to_fp16)[name = tensor<string, []>("op_16454_cast")];
            tensor<fp16, []> denom_113_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_113_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_113_cast = rsqrt(epsilon = denom_113_epsilon_0_to_fp16, x = var_16454_cast)[name = tensor<string, []>("denom_113_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_113_cast = mul(x = zero_mean_113_cast, y = denom_113_cast)[name = tensor<string, []>("out_113_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_16458_to_fp16 = const()[name = tensor<string, []>("op_16458_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1270352128)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_16459_cast = add(x = out_113_cast, y = var_16458_to_fp16)[name = tensor<string, []>("op_16459_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_16461_to_fp16 = const()[name = tensor<string, []>("op_16461_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1270354752)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_269_cast = mul(x = var_16459_cast, y = var_16461_to_fp16)[name = tensor<string, []>("input_269_cast")];
            tensor<int32, [2]> var_16469 = const()[name = tensor<string, []>("op_16469"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_16471 = const()[name = tensor<string, []>("op_16471"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_16473_pad_type_0 = const()[name = tensor<string, []>("op_16473_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_16473_pad_0 = const()[name = tensor<string, []>("op_16473_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_4_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_4_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1270357376)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_1_transformer_blocks_4_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_4_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1296571840)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_16473_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_4_ff_net_0_proj_bias_to_fp16, dilations = var_16471, groups = var_2634, pad = var_16473_pad_0, pad_type = var_16473_pad_type_0, strides = var_16469, weight = down_blocks_2_attentions_1_transformer_blocks_4_ff_net_0_proj_weight_to_fp16, x = input_269_cast)[name = tensor<string, []>("op_16473_cast")];
            tensor<int32, [2]> var_16474_split_sizes_0 = const()[name = tensor<string, []>("op_16474_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_16474_axis_0 = const()[name = tensor<string, []>("op_16474_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_16474_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_16474_cast_1 = split(axis = var_16474_axis_0, split_sizes = var_16474_split_sizes_0, x = var_16473_cast)[name = tensor<string, []>("op_16474_cast")];
            tensor<string, []> var_16476_mode_0 = const()[name = tensor<string, []>("op_16476_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_16476_cast = gelu(mode = var_16476_mode_0, x = var_16474_cast_1)[name = tensor<string, []>("op_16476_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_271_cast = mul(x = var_16474_cast_0, y = var_16476_cast)[name = tensor<string, []>("input_271_cast")];
            tensor<int32, [2]> var_16480 = const()[name = tensor<string, []>("op_16480"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_16482 = const()[name = tensor<string, []>("op_16482"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_16484_pad_type_0 = const()[name = tensor<string, []>("op_16484_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_16484_pad_0 = const()[name = tensor<string, []>("op_16484_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_4_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_4_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1296592384)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_4_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_4_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1309699648)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_16484_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_4_ff_net_2_bias_to_fp16, dilations = var_16482, groups = var_2634, pad = var_16484_pad_0, pad_type = var_16484_pad_type_0, strides = var_16480, weight = down_blocks_2_attentions_1_transformer_blocks_4_ff_net_2_weight_to_fp16, x = input_271_cast)[name = tensor<string, []>("op_16484_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_115_cast = add(x = var_16484_cast, y = inputs_113_cast)[name = tensor<string, []>("inputs_115_cast")];
            tensor<int32, [1]> var_16494 = const()[name = tensor<string, []>("op_16494"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_115_cast = reduce_mean(axes = var_16494, keep_dims = var_2629, x = inputs_115_cast)[name = tensor<string, []>("channels_mean_115_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_115_cast = sub(x = inputs_115_cast, y = channels_mean_115_cast)[name = tensor<string, []>("zero_mean_115_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_115_cast = mul(x = zero_mean_115_cast, y = zero_mean_115_cast)[name = tensor<string, []>("zero_mean_sq_115_cast")];
            tensor<int32, [1]> var_16498 = const()[name = tensor<string, []>("op_16498"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_16499_cast = reduce_mean(axes = var_16498, keep_dims = var_2629, x = zero_mean_sq_115_cast)[name = tensor<string, []>("op_16499_cast")];
            tensor<fp16, []> var_16500_to_fp16 = const()[name = tensor<string, []>("op_16500_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_16501_cast = add(x = var_16499_cast, y = var_16500_to_fp16)[name = tensor<string, []>("op_16501_cast")];
            tensor<fp16, []> denom_115_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_115_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_115_cast = rsqrt(epsilon = denom_115_epsilon_0_to_fp16, x = var_16501_cast)[name = tensor<string, []>("denom_115_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_115_cast = mul(x = zero_mean_115_cast, y = denom_115_cast)[name = tensor<string, []>("out_115_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_16505_to_fp16 = const()[name = tensor<string, []>("op_16505_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1309702272)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_16506_cast = add(x = out_115_cast, y = var_16505_to_fp16)[name = tensor<string, []>("op_16506_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_16508_to_fp16 = const()[name = tensor<string, []>("op_16508_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1309704896)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_167_cast = mul(x = var_16506_cast, y = var_16508_to_fp16)[name = tensor<string, []>("hidden_states_167_cast")];
            tensor<int32, [2]> var_16515 = const()[name = tensor<string, []>("op_16515"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_16517 = const()[name = tensor<string, []>("op_16517"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_77_pad_type_0 = const()[name = tensor<string, []>("q_77_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_77_pad_0 = const()[name = tensor<string, []>("q_77_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1309707520)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_77_cast = conv(dilations = var_16517, groups = var_2634, pad = q_77_pad_0, pad_type = q_77_pad_type_0, strides = var_16515, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_q_weight_to_fp16, x = hidden_states_167_cast)[name = tensor<string, []>("q_77_cast")];
            tensor<int32, [2]> var_16521 = const()[name = tensor<string, []>("op_16521"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_16523 = const()[name = tensor<string, []>("op_16523"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_153_pad_type_0 = const()[name = tensor<string, []>("k_153_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_153_pad_0 = const()[name = tensor<string, []>("k_153_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1312984384)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_153_cast = conv(dilations = var_16523, groups = var_2634, pad = k_153_pad_0, pad_type = k_153_pad_type_0, strides = var_16521, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_k_weight_to_fp16, x = hidden_states_167_cast)[name = tensor<string, []>("k_153_cast")];
            tensor<int32, [2]> var_16527 = const()[name = tensor<string, []>("op_16527"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_16529 = const()[name = tensor<string, []>("op_16529"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_77_pad_type_0 = const()[name = tensor<string, []>("v_77_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_77_pad_0 = const()[name = tensor<string, []>("v_77_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1316261248)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_77_cast = conv(dilations = var_16529, groups = var_2634, pad = v_77_pad_0, pad_type = v_77_pad_type_0, strides = var_16527, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_v_weight_to_fp16, x = hidden_states_167_cast)[name = tensor<string, []>("v_77_cast")];
            tensor<int32, [4]> var_16533_begin_0 = const()[name = tensor<string, []>("op_16533_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_16533_end_0 = const()[name = tensor<string, []>("op_16533_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_16533_end_mask_0 = const()[name = tensor<string, []>("op_16533_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16533_cast = slice_by_index(begin = var_16533_begin_0, end = var_16533_end_0, end_mask = var_16533_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16533_cast")];
            tensor<int32, [4]> var_16537_begin_0 = const()[name = tensor<string, []>("op_16537_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_16537_end_0 = const()[name = tensor<string, []>("op_16537_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_16537_end_mask_0 = const()[name = tensor<string, []>("op_16537_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16537_cast = slice_by_index(begin = var_16537_begin_0, end = var_16537_end_0, end_mask = var_16537_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16537_cast")];
            tensor<int32, [4]> var_16541_begin_0 = const()[name = tensor<string, []>("op_16541_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_16541_end_0 = const()[name = tensor<string, []>("op_16541_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_16541_end_mask_0 = const()[name = tensor<string, []>("op_16541_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16541_cast = slice_by_index(begin = var_16541_begin_0, end = var_16541_end_0, end_mask = var_16541_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16541_cast")];
            tensor<int32, [4]> var_16545_begin_0 = const()[name = tensor<string, []>("op_16545_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_16545_end_0 = const()[name = tensor<string, []>("op_16545_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_16545_end_mask_0 = const()[name = tensor<string, []>("op_16545_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16545_cast = slice_by_index(begin = var_16545_begin_0, end = var_16545_end_0, end_mask = var_16545_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16545_cast")];
            tensor<int32, [4]> var_16549_begin_0 = const()[name = tensor<string, []>("op_16549_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_16549_end_0 = const()[name = tensor<string, []>("op_16549_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_16549_end_mask_0 = const()[name = tensor<string, []>("op_16549_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16549_cast = slice_by_index(begin = var_16549_begin_0, end = var_16549_end_0, end_mask = var_16549_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16549_cast")];
            tensor<int32, [4]> var_16553_begin_0 = const()[name = tensor<string, []>("op_16553_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_16553_end_0 = const()[name = tensor<string, []>("op_16553_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_16553_end_mask_0 = const()[name = tensor<string, []>("op_16553_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16553_cast = slice_by_index(begin = var_16553_begin_0, end = var_16553_end_0, end_mask = var_16553_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16553_cast")];
            tensor<int32, [4]> var_16557_begin_0 = const()[name = tensor<string, []>("op_16557_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_16557_end_0 = const()[name = tensor<string, []>("op_16557_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_16557_end_mask_0 = const()[name = tensor<string, []>("op_16557_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16557_cast = slice_by_index(begin = var_16557_begin_0, end = var_16557_end_0, end_mask = var_16557_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16557_cast")];
            tensor<int32, [4]> var_16561_begin_0 = const()[name = tensor<string, []>("op_16561_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_16561_end_0 = const()[name = tensor<string, []>("op_16561_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_16561_end_mask_0 = const()[name = tensor<string, []>("op_16561_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16561_cast = slice_by_index(begin = var_16561_begin_0, end = var_16561_end_0, end_mask = var_16561_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16561_cast")];
            tensor<int32, [4]> var_16565_begin_0 = const()[name = tensor<string, []>("op_16565_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_16565_end_0 = const()[name = tensor<string, []>("op_16565_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_16565_end_mask_0 = const()[name = tensor<string, []>("op_16565_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16565_cast = slice_by_index(begin = var_16565_begin_0, end = var_16565_end_0, end_mask = var_16565_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16565_cast")];
            tensor<int32, [4]> var_16569_begin_0 = const()[name = tensor<string, []>("op_16569_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_16569_end_0 = const()[name = tensor<string, []>("op_16569_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_16569_end_mask_0 = const()[name = tensor<string, []>("op_16569_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16569_cast = slice_by_index(begin = var_16569_begin_0, end = var_16569_end_0, end_mask = var_16569_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16569_cast")];
            tensor<int32, [4]> var_16573_begin_0 = const()[name = tensor<string, []>("op_16573_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_16573_end_0 = const()[name = tensor<string, []>("op_16573_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_16573_end_mask_0 = const()[name = tensor<string, []>("op_16573_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16573_cast = slice_by_index(begin = var_16573_begin_0, end = var_16573_end_0, end_mask = var_16573_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16573_cast")];
            tensor<int32, [4]> var_16577_begin_0 = const()[name = tensor<string, []>("op_16577_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_16577_end_0 = const()[name = tensor<string, []>("op_16577_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_16577_end_mask_0 = const()[name = tensor<string, []>("op_16577_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16577_cast = slice_by_index(begin = var_16577_begin_0, end = var_16577_end_0, end_mask = var_16577_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16577_cast")];
            tensor<int32, [4]> var_16581_begin_0 = const()[name = tensor<string, []>("op_16581_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_16581_end_0 = const()[name = tensor<string, []>("op_16581_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_16581_end_mask_0 = const()[name = tensor<string, []>("op_16581_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16581_cast = slice_by_index(begin = var_16581_begin_0, end = var_16581_end_0, end_mask = var_16581_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16581_cast")];
            tensor<int32, [4]> var_16585_begin_0 = const()[name = tensor<string, []>("op_16585_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_16585_end_0 = const()[name = tensor<string, []>("op_16585_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_16585_end_mask_0 = const()[name = tensor<string, []>("op_16585_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16585_cast = slice_by_index(begin = var_16585_begin_0, end = var_16585_end_0, end_mask = var_16585_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16585_cast")];
            tensor<int32, [4]> var_16589_begin_0 = const()[name = tensor<string, []>("op_16589_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_16589_end_0 = const()[name = tensor<string, []>("op_16589_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_16589_end_mask_0 = const()[name = tensor<string, []>("op_16589_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16589_cast = slice_by_index(begin = var_16589_begin_0, end = var_16589_end_0, end_mask = var_16589_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16589_cast")];
            tensor<int32, [4]> var_16593_begin_0 = const()[name = tensor<string, []>("op_16593_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_16593_end_0 = const()[name = tensor<string, []>("op_16593_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_16593_end_mask_0 = const()[name = tensor<string, []>("op_16593_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16593_cast = slice_by_index(begin = var_16593_begin_0, end = var_16593_end_0, end_mask = var_16593_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16593_cast")];
            tensor<int32, [4]> var_16597_begin_0 = const()[name = tensor<string, []>("op_16597_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_16597_end_0 = const()[name = tensor<string, []>("op_16597_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_16597_end_mask_0 = const()[name = tensor<string, []>("op_16597_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16597_cast = slice_by_index(begin = var_16597_begin_0, end = var_16597_end_0, end_mask = var_16597_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16597_cast")];
            tensor<int32, [4]> var_16601_begin_0 = const()[name = tensor<string, []>("op_16601_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_16601_end_0 = const()[name = tensor<string, []>("op_16601_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_16601_end_mask_0 = const()[name = tensor<string, []>("op_16601_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16601_cast = slice_by_index(begin = var_16601_begin_0, end = var_16601_end_0, end_mask = var_16601_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16601_cast")];
            tensor<int32, [4]> var_16605_begin_0 = const()[name = tensor<string, []>("op_16605_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_16605_end_0 = const()[name = tensor<string, []>("op_16605_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_16605_end_mask_0 = const()[name = tensor<string, []>("op_16605_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16605_cast = slice_by_index(begin = var_16605_begin_0, end = var_16605_end_0, end_mask = var_16605_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16605_cast")];
            tensor<int32, [4]> var_16609_begin_0 = const()[name = tensor<string, []>("op_16609_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_16609_end_0 = const()[name = tensor<string, []>("op_16609_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_16609_end_mask_0 = const()[name = tensor<string, []>("op_16609_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16609_cast = slice_by_index(begin = var_16609_begin_0, end = var_16609_end_0, end_mask = var_16609_end_mask_0, x = q_77_cast)[name = tensor<string, []>("op_16609_cast")];
            tensor<int32, [4]> k_155_perm_0 = const()[name = tensor<string, []>("k_155_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_16616_begin_0 = const()[name = tensor<string, []>("op_16616_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_16616_end_0 = const()[name = tensor<string, []>("op_16616_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_16616_end_mask_0 = const()[name = tensor<string, []>("op_16616_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_101 = transpose(perm = k_155_perm_0, x = k_153_cast)[name = tensor<string, []>("transpose_101")];
            tensor<fp16, [2, 1024, 1, 64]> var_16616_cast = slice_by_index(begin = var_16616_begin_0, end = var_16616_end_0, end_mask = var_16616_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16616_cast")];
            tensor<int32, [4]> var_16620_begin_0 = const()[name = tensor<string, []>("op_16620_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_16620_end_0 = const()[name = tensor<string, []>("op_16620_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_16620_end_mask_0 = const()[name = tensor<string, []>("op_16620_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_16620_cast = slice_by_index(begin = var_16620_begin_0, end = var_16620_end_0, end_mask = var_16620_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16620_cast")];
            tensor<int32, [4]> var_16624_begin_0 = const()[name = tensor<string, []>("op_16624_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_16624_end_0 = const()[name = tensor<string, []>("op_16624_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_16624_end_mask_0 = const()[name = tensor<string, []>("op_16624_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_16624_cast = slice_by_index(begin = var_16624_begin_0, end = var_16624_end_0, end_mask = var_16624_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16624_cast")];
            tensor<int32, [4]> var_16628_begin_0 = const()[name = tensor<string, []>("op_16628_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_16628_end_0 = const()[name = tensor<string, []>("op_16628_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_16628_end_mask_0 = const()[name = tensor<string, []>("op_16628_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_16628_cast = slice_by_index(begin = var_16628_begin_0, end = var_16628_end_0, end_mask = var_16628_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16628_cast")];
            tensor<int32, [4]> var_16632_begin_0 = const()[name = tensor<string, []>("op_16632_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_16632_end_0 = const()[name = tensor<string, []>("op_16632_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_16632_end_mask_0 = const()[name = tensor<string, []>("op_16632_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_16632_cast = slice_by_index(begin = var_16632_begin_0, end = var_16632_end_0, end_mask = var_16632_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16632_cast")];
            tensor<int32, [4]> var_16636_begin_0 = const()[name = tensor<string, []>("op_16636_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_16636_end_0 = const()[name = tensor<string, []>("op_16636_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_16636_end_mask_0 = const()[name = tensor<string, []>("op_16636_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_16636_cast = slice_by_index(begin = var_16636_begin_0, end = var_16636_end_0, end_mask = var_16636_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16636_cast")];
            tensor<int32, [4]> var_16640_begin_0 = const()[name = tensor<string, []>("op_16640_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_16640_end_0 = const()[name = tensor<string, []>("op_16640_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_16640_end_mask_0 = const()[name = tensor<string, []>("op_16640_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_16640_cast = slice_by_index(begin = var_16640_begin_0, end = var_16640_end_0, end_mask = var_16640_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16640_cast")];
            tensor<int32, [4]> var_16644_begin_0 = const()[name = tensor<string, []>("op_16644_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_16644_end_0 = const()[name = tensor<string, []>("op_16644_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_16644_end_mask_0 = const()[name = tensor<string, []>("op_16644_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_16644_cast = slice_by_index(begin = var_16644_begin_0, end = var_16644_end_0, end_mask = var_16644_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16644_cast")];
            tensor<int32, [4]> var_16648_begin_0 = const()[name = tensor<string, []>("op_16648_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_16648_end_0 = const()[name = tensor<string, []>("op_16648_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_16648_end_mask_0 = const()[name = tensor<string, []>("op_16648_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_16648_cast = slice_by_index(begin = var_16648_begin_0, end = var_16648_end_0, end_mask = var_16648_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16648_cast")];
            tensor<int32, [4]> var_16652_begin_0 = const()[name = tensor<string, []>("op_16652_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_16652_end_0 = const()[name = tensor<string, []>("op_16652_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_16652_end_mask_0 = const()[name = tensor<string, []>("op_16652_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_16652_cast = slice_by_index(begin = var_16652_begin_0, end = var_16652_end_0, end_mask = var_16652_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16652_cast")];
            tensor<int32, [4]> var_16656_begin_0 = const()[name = tensor<string, []>("op_16656_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_16656_end_0 = const()[name = tensor<string, []>("op_16656_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_16656_end_mask_0 = const()[name = tensor<string, []>("op_16656_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_16656_cast = slice_by_index(begin = var_16656_begin_0, end = var_16656_end_0, end_mask = var_16656_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16656_cast")];
            tensor<int32, [4]> var_16660_begin_0 = const()[name = tensor<string, []>("op_16660_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_16660_end_0 = const()[name = tensor<string, []>("op_16660_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_16660_end_mask_0 = const()[name = tensor<string, []>("op_16660_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_16660_cast = slice_by_index(begin = var_16660_begin_0, end = var_16660_end_0, end_mask = var_16660_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16660_cast")];
            tensor<int32, [4]> var_16664_begin_0 = const()[name = tensor<string, []>("op_16664_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_16664_end_0 = const()[name = tensor<string, []>("op_16664_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_16664_end_mask_0 = const()[name = tensor<string, []>("op_16664_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_16664_cast = slice_by_index(begin = var_16664_begin_0, end = var_16664_end_0, end_mask = var_16664_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16664_cast")];
            tensor<int32, [4]> var_16668_begin_0 = const()[name = tensor<string, []>("op_16668_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_16668_end_0 = const()[name = tensor<string, []>("op_16668_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_16668_end_mask_0 = const()[name = tensor<string, []>("op_16668_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_16668_cast = slice_by_index(begin = var_16668_begin_0, end = var_16668_end_0, end_mask = var_16668_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16668_cast")];
            tensor<int32, [4]> var_16672_begin_0 = const()[name = tensor<string, []>("op_16672_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_16672_end_0 = const()[name = tensor<string, []>("op_16672_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_16672_end_mask_0 = const()[name = tensor<string, []>("op_16672_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_16672_cast = slice_by_index(begin = var_16672_begin_0, end = var_16672_end_0, end_mask = var_16672_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16672_cast")];
            tensor<int32, [4]> var_16676_begin_0 = const()[name = tensor<string, []>("op_16676_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_16676_end_0 = const()[name = tensor<string, []>("op_16676_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_16676_end_mask_0 = const()[name = tensor<string, []>("op_16676_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_16676_cast = slice_by_index(begin = var_16676_begin_0, end = var_16676_end_0, end_mask = var_16676_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16676_cast")];
            tensor<int32, [4]> var_16680_begin_0 = const()[name = tensor<string, []>("op_16680_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_16680_end_0 = const()[name = tensor<string, []>("op_16680_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_16680_end_mask_0 = const()[name = tensor<string, []>("op_16680_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_16680_cast = slice_by_index(begin = var_16680_begin_0, end = var_16680_end_0, end_mask = var_16680_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16680_cast")];
            tensor<int32, [4]> var_16684_begin_0 = const()[name = tensor<string, []>("op_16684_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_16684_end_0 = const()[name = tensor<string, []>("op_16684_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_16684_end_mask_0 = const()[name = tensor<string, []>("op_16684_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_16684_cast = slice_by_index(begin = var_16684_begin_0, end = var_16684_end_0, end_mask = var_16684_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16684_cast")];
            tensor<int32, [4]> var_16688_begin_0 = const()[name = tensor<string, []>("op_16688_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_16688_end_0 = const()[name = tensor<string, []>("op_16688_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_16688_end_mask_0 = const()[name = tensor<string, []>("op_16688_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_16688_cast = slice_by_index(begin = var_16688_begin_0, end = var_16688_end_0, end_mask = var_16688_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16688_cast")];
            tensor<int32, [4]> var_16692_begin_0 = const()[name = tensor<string, []>("op_16692_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_16692_end_0 = const()[name = tensor<string, []>("op_16692_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_16692_end_mask_0 = const()[name = tensor<string, []>("op_16692_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_16692_cast = slice_by_index(begin = var_16692_begin_0, end = var_16692_end_0, end_mask = var_16692_end_mask_0, x = transpose_101)[name = tensor<string, []>("op_16692_cast")];
            tensor<int32, [4]> var_16694_begin_0 = const()[name = tensor<string, []>("op_16694_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_16694_end_0 = const()[name = tensor<string, []>("op_16694_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_16694_end_mask_0 = const()[name = tensor<string, []>("op_16694_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16694_cast = slice_by_index(begin = var_16694_begin_0, end = var_16694_end_0, end_mask = var_16694_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16694_cast")];
            tensor<int32, [4]> var_16698_begin_0 = const()[name = tensor<string, []>("op_16698_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_16698_end_0 = const()[name = tensor<string, []>("op_16698_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_16698_end_mask_0 = const()[name = tensor<string, []>("op_16698_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16698_cast = slice_by_index(begin = var_16698_begin_0, end = var_16698_end_0, end_mask = var_16698_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16698_cast")];
            tensor<int32, [4]> var_16702_begin_0 = const()[name = tensor<string, []>("op_16702_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_16702_end_0 = const()[name = tensor<string, []>("op_16702_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_16702_end_mask_0 = const()[name = tensor<string, []>("op_16702_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16702_cast = slice_by_index(begin = var_16702_begin_0, end = var_16702_end_0, end_mask = var_16702_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16702_cast")];
            tensor<int32, [4]> var_16706_begin_0 = const()[name = tensor<string, []>("op_16706_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_16706_end_0 = const()[name = tensor<string, []>("op_16706_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_16706_end_mask_0 = const()[name = tensor<string, []>("op_16706_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16706_cast = slice_by_index(begin = var_16706_begin_0, end = var_16706_end_0, end_mask = var_16706_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16706_cast")];
            tensor<int32, [4]> var_16710_begin_0 = const()[name = tensor<string, []>("op_16710_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_16710_end_0 = const()[name = tensor<string, []>("op_16710_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_16710_end_mask_0 = const()[name = tensor<string, []>("op_16710_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16710_cast = slice_by_index(begin = var_16710_begin_0, end = var_16710_end_0, end_mask = var_16710_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16710_cast")];
            tensor<int32, [4]> var_16714_begin_0 = const()[name = tensor<string, []>("op_16714_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_16714_end_0 = const()[name = tensor<string, []>("op_16714_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_16714_end_mask_0 = const()[name = tensor<string, []>("op_16714_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16714_cast = slice_by_index(begin = var_16714_begin_0, end = var_16714_end_0, end_mask = var_16714_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16714_cast")];
            tensor<int32, [4]> var_16718_begin_0 = const()[name = tensor<string, []>("op_16718_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_16718_end_0 = const()[name = tensor<string, []>("op_16718_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_16718_end_mask_0 = const()[name = tensor<string, []>("op_16718_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16718_cast = slice_by_index(begin = var_16718_begin_0, end = var_16718_end_0, end_mask = var_16718_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16718_cast")];
            tensor<int32, [4]> var_16722_begin_0 = const()[name = tensor<string, []>("op_16722_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_16722_end_0 = const()[name = tensor<string, []>("op_16722_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_16722_end_mask_0 = const()[name = tensor<string, []>("op_16722_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16722_cast = slice_by_index(begin = var_16722_begin_0, end = var_16722_end_0, end_mask = var_16722_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16722_cast")];
            tensor<int32, [4]> var_16726_begin_0 = const()[name = tensor<string, []>("op_16726_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_16726_end_0 = const()[name = tensor<string, []>("op_16726_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_16726_end_mask_0 = const()[name = tensor<string, []>("op_16726_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16726_cast = slice_by_index(begin = var_16726_begin_0, end = var_16726_end_0, end_mask = var_16726_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16726_cast")];
            tensor<int32, [4]> var_16730_begin_0 = const()[name = tensor<string, []>("op_16730_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_16730_end_0 = const()[name = tensor<string, []>("op_16730_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_16730_end_mask_0 = const()[name = tensor<string, []>("op_16730_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16730_cast = slice_by_index(begin = var_16730_begin_0, end = var_16730_end_0, end_mask = var_16730_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16730_cast")];
            tensor<int32, [4]> var_16734_begin_0 = const()[name = tensor<string, []>("op_16734_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_16734_end_0 = const()[name = tensor<string, []>("op_16734_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_16734_end_mask_0 = const()[name = tensor<string, []>("op_16734_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16734_cast = slice_by_index(begin = var_16734_begin_0, end = var_16734_end_0, end_mask = var_16734_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16734_cast")];
            tensor<int32, [4]> var_16738_begin_0 = const()[name = tensor<string, []>("op_16738_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_16738_end_0 = const()[name = tensor<string, []>("op_16738_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_16738_end_mask_0 = const()[name = tensor<string, []>("op_16738_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16738_cast = slice_by_index(begin = var_16738_begin_0, end = var_16738_end_0, end_mask = var_16738_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16738_cast")];
            tensor<int32, [4]> var_16742_begin_0 = const()[name = tensor<string, []>("op_16742_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_16742_end_0 = const()[name = tensor<string, []>("op_16742_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_16742_end_mask_0 = const()[name = tensor<string, []>("op_16742_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16742_cast = slice_by_index(begin = var_16742_begin_0, end = var_16742_end_0, end_mask = var_16742_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16742_cast")];
            tensor<int32, [4]> var_16746_begin_0 = const()[name = tensor<string, []>("op_16746_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_16746_end_0 = const()[name = tensor<string, []>("op_16746_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_16746_end_mask_0 = const()[name = tensor<string, []>("op_16746_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16746_cast = slice_by_index(begin = var_16746_begin_0, end = var_16746_end_0, end_mask = var_16746_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16746_cast")];
            tensor<int32, [4]> var_16750_begin_0 = const()[name = tensor<string, []>("op_16750_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_16750_end_0 = const()[name = tensor<string, []>("op_16750_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_16750_end_mask_0 = const()[name = tensor<string, []>("op_16750_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16750_cast = slice_by_index(begin = var_16750_begin_0, end = var_16750_end_0, end_mask = var_16750_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16750_cast")];
            tensor<int32, [4]> var_16754_begin_0 = const()[name = tensor<string, []>("op_16754_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_16754_end_0 = const()[name = tensor<string, []>("op_16754_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_16754_end_mask_0 = const()[name = tensor<string, []>("op_16754_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16754_cast = slice_by_index(begin = var_16754_begin_0, end = var_16754_end_0, end_mask = var_16754_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16754_cast")];
            tensor<int32, [4]> var_16758_begin_0 = const()[name = tensor<string, []>("op_16758_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_16758_end_0 = const()[name = tensor<string, []>("op_16758_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_16758_end_mask_0 = const()[name = tensor<string, []>("op_16758_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16758_cast = slice_by_index(begin = var_16758_begin_0, end = var_16758_end_0, end_mask = var_16758_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16758_cast")];
            tensor<int32, [4]> var_16762_begin_0 = const()[name = tensor<string, []>("op_16762_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_16762_end_0 = const()[name = tensor<string, []>("op_16762_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_16762_end_mask_0 = const()[name = tensor<string, []>("op_16762_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16762_cast = slice_by_index(begin = var_16762_begin_0, end = var_16762_end_0, end_mask = var_16762_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16762_cast")];
            tensor<int32, [4]> var_16766_begin_0 = const()[name = tensor<string, []>("op_16766_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_16766_end_0 = const()[name = tensor<string, []>("op_16766_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_16766_end_mask_0 = const()[name = tensor<string, []>("op_16766_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16766_cast = slice_by_index(begin = var_16766_begin_0, end = var_16766_end_0, end_mask = var_16766_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16766_cast")];
            tensor<int32, [4]> var_16770_begin_0 = const()[name = tensor<string, []>("op_16770_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_16770_end_0 = const()[name = tensor<string, []>("op_16770_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_16770_end_mask_0 = const()[name = tensor<string, []>("op_16770_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16770_cast = slice_by_index(begin = var_16770_begin_0, end = var_16770_end_0, end_mask = var_16770_end_mask_0, x = v_77_cast)[name = tensor<string, []>("op_16770_cast")];
            tensor<string, []> var_16774_equation_0 = const()[name = tensor<string, []>("op_16774_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16774_cast = einsum(equation = var_16774_equation_0, values = (var_16616_cast, var_16533_cast))[name = tensor<string, []>("op_16774_cast")];
            tensor<fp16, []> var_16775_to_fp16 = const()[name = tensor<string, []>("op_16775_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1361_cast = mul(x = var_16774_cast, y = var_16775_to_fp16)[name = tensor<string, []>("aw_1361_cast")];
            tensor<string, []> var_16778_equation_0 = const()[name = tensor<string, []>("op_16778_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16778_cast = einsum(equation = var_16778_equation_0, values = (var_16620_cast, var_16537_cast))[name = tensor<string, []>("op_16778_cast")];
            tensor<fp16, []> var_16779_to_fp16 = const()[name = tensor<string, []>("op_16779_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1363_cast = mul(x = var_16778_cast, y = var_16779_to_fp16)[name = tensor<string, []>("aw_1363_cast")];
            tensor<string, []> var_16782_equation_0 = const()[name = tensor<string, []>("op_16782_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16782_cast = einsum(equation = var_16782_equation_0, values = (var_16624_cast, var_16541_cast))[name = tensor<string, []>("op_16782_cast")];
            tensor<fp16, []> var_16783_to_fp16 = const()[name = tensor<string, []>("op_16783_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1365_cast = mul(x = var_16782_cast, y = var_16783_to_fp16)[name = tensor<string, []>("aw_1365_cast")];
            tensor<string, []> var_16786_equation_0 = const()[name = tensor<string, []>("op_16786_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16786_cast = einsum(equation = var_16786_equation_0, values = (var_16628_cast, var_16545_cast))[name = tensor<string, []>("op_16786_cast")];
            tensor<fp16, []> var_16787_to_fp16 = const()[name = tensor<string, []>("op_16787_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1367_cast = mul(x = var_16786_cast, y = var_16787_to_fp16)[name = tensor<string, []>("aw_1367_cast")];
            tensor<string, []> var_16790_equation_0 = const()[name = tensor<string, []>("op_16790_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16790_cast = einsum(equation = var_16790_equation_0, values = (var_16632_cast, var_16549_cast))[name = tensor<string, []>("op_16790_cast")];
            tensor<fp16, []> var_16791_to_fp16 = const()[name = tensor<string, []>("op_16791_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1369_cast = mul(x = var_16790_cast, y = var_16791_to_fp16)[name = tensor<string, []>("aw_1369_cast")];
            tensor<string, []> var_16794_equation_0 = const()[name = tensor<string, []>("op_16794_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16794_cast = einsum(equation = var_16794_equation_0, values = (var_16636_cast, var_16553_cast))[name = tensor<string, []>("op_16794_cast")];
            tensor<fp16, []> var_16795_to_fp16 = const()[name = tensor<string, []>("op_16795_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1371_cast = mul(x = var_16794_cast, y = var_16795_to_fp16)[name = tensor<string, []>("aw_1371_cast")];
            tensor<string, []> var_16798_equation_0 = const()[name = tensor<string, []>("op_16798_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16798_cast = einsum(equation = var_16798_equation_0, values = (var_16640_cast, var_16557_cast))[name = tensor<string, []>("op_16798_cast")];
            tensor<fp16, []> var_16799_to_fp16 = const()[name = tensor<string, []>("op_16799_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1373_cast = mul(x = var_16798_cast, y = var_16799_to_fp16)[name = tensor<string, []>("aw_1373_cast")];
            tensor<string, []> var_16802_equation_0 = const()[name = tensor<string, []>("op_16802_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16802_cast = einsum(equation = var_16802_equation_0, values = (var_16644_cast, var_16561_cast))[name = tensor<string, []>("op_16802_cast")];
            tensor<fp16, []> var_16803_to_fp16 = const()[name = tensor<string, []>("op_16803_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1375_cast = mul(x = var_16802_cast, y = var_16803_to_fp16)[name = tensor<string, []>("aw_1375_cast")];
            tensor<string, []> var_16806_equation_0 = const()[name = tensor<string, []>("op_16806_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16806_cast = einsum(equation = var_16806_equation_0, values = (var_16648_cast, var_16565_cast))[name = tensor<string, []>("op_16806_cast")];
            tensor<fp16, []> var_16807_to_fp16 = const()[name = tensor<string, []>("op_16807_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1377_cast = mul(x = var_16806_cast, y = var_16807_to_fp16)[name = tensor<string, []>("aw_1377_cast")];
            tensor<string, []> var_16810_equation_0 = const()[name = tensor<string, []>("op_16810_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16810_cast = einsum(equation = var_16810_equation_0, values = (var_16652_cast, var_16569_cast))[name = tensor<string, []>("op_16810_cast")];
            tensor<fp16, []> var_16811_to_fp16 = const()[name = tensor<string, []>("op_16811_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1379_cast = mul(x = var_16810_cast, y = var_16811_to_fp16)[name = tensor<string, []>("aw_1379_cast")];
            tensor<string, []> var_16814_equation_0 = const()[name = tensor<string, []>("op_16814_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16814_cast = einsum(equation = var_16814_equation_0, values = (var_16656_cast, var_16573_cast))[name = tensor<string, []>("op_16814_cast")];
            tensor<fp16, []> var_16815_to_fp16 = const()[name = tensor<string, []>("op_16815_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1381_cast = mul(x = var_16814_cast, y = var_16815_to_fp16)[name = tensor<string, []>("aw_1381_cast")];
            tensor<string, []> var_16818_equation_0 = const()[name = tensor<string, []>("op_16818_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16818_cast = einsum(equation = var_16818_equation_0, values = (var_16660_cast, var_16577_cast))[name = tensor<string, []>("op_16818_cast")];
            tensor<fp16, []> var_16819_to_fp16 = const()[name = tensor<string, []>("op_16819_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1383_cast = mul(x = var_16818_cast, y = var_16819_to_fp16)[name = tensor<string, []>("aw_1383_cast")];
            tensor<string, []> var_16822_equation_0 = const()[name = tensor<string, []>("op_16822_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16822_cast = einsum(equation = var_16822_equation_0, values = (var_16664_cast, var_16581_cast))[name = tensor<string, []>("op_16822_cast")];
            tensor<fp16, []> var_16823_to_fp16 = const()[name = tensor<string, []>("op_16823_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1385_cast = mul(x = var_16822_cast, y = var_16823_to_fp16)[name = tensor<string, []>("aw_1385_cast")];
            tensor<string, []> var_16826_equation_0 = const()[name = tensor<string, []>("op_16826_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16826_cast = einsum(equation = var_16826_equation_0, values = (var_16668_cast, var_16585_cast))[name = tensor<string, []>("op_16826_cast")];
            tensor<fp16, []> var_16827_to_fp16 = const()[name = tensor<string, []>("op_16827_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1387_cast = mul(x = var_16826_cast, y = var_16827_to_fp16)[name = tensor<string, []>("aw_1387_cast")];
            tensor<string, []> var_16830_equation_0 = const()[name = tensor<string, []>("op_16830_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16830_cast = einsum(equation = var_16830_equation_0, values = (var_16672_cast, var_16589_cast))[name = tensor<string, []>("op_16830_cast")];
            tensor<fp16, []> var_16831_to_fp16 = const()[name = tensor<string, []>("op_16831_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1389_cast = mul(x = var_16830_cast, y = var_16831_to_fp16)[name = tensor<string, []>("aw_1389_cast")];
            tensor<string, []> var_16834_equation_0 = const()[name = tensor<string, []>("op_16834_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16834_cast = einsum(equation = var_16834_equation_0, values = (var_16676_cast, var_16593_cast))[name = tensor<string, []>("op_16834_cast")];
            tensor<fp16, []> var_16835_to_fp16 = const()[name = tensor<string, []>("op_16835_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1391_cast = mul(x = var_16834_cast, y = var_16835_to_fp16)[name = tensor<string, []>("aw_1391_cast")];
            tensor<string, []> var_16838_equation_0 = const()[name = tensor<string, []>("op_16838_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16838_cast = einsum(equation = var_16838_equation_0, values = (var_16680_cast, var_16597_cast))[name = tensor<string, []>("op_16838_cast")];
            tensor<fp16, []> var_16839_to_fp16 = const()[name = tensor<string, []>("op_16839_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1393_cast = mul(x = var_16838_cast, y = var_16839_to_fp16)[name = tensor<string, []>("aw_1393_cast")];
            tensor<string, []> var_16842_equation_0 = const()[name = tensor<string, []>("op_16842_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16842_cast = einsum(equation = var_16842_equation_0, values = (var_16684_cast, var_16601_cast))[name = tensor<string, []>("op_16842_cast")];
            tensor<fp16, []> var_16843_to_fp16 = const()[name = tensor<string, []>("op_16843_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1395_cast = mul(x = var_16842_cast, y = var_16843_to_fp16)[name = tensor<string, []>("aw_1395_cast")];
            tensor<string, []> var_16846_equation_0 = const()[name = tensor<string, []>("op_16846_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16846_cast = einsum(equation = var_16846_equation_0, values = (var_16688_cast, var_16605_cast))[name = tensor<string, []>("op_16846_cast")];
            tensor<fp16, []> var_16847_to_fp16 = const()[name = tensor<string, []>("op_16847_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1397_cast = mul(x = var_16846_cast, y = var_16847_to_fp16)[name = tensor<string, []>("aw_1397_cast")];
            tensor<string, []> var_16850_equation_0 = const()[name = tensor<string, []>("op_16850_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16850_cast = einsum(equation = var_16850_equation_0, values = (var_16692_cast, var_16609_cast))[name = tensor<string, []>("op_16850_cast")];
            tensor<fp16, []> var_16851_to_fp16 = const()[name = tensor<string, []>("op_16851_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1399_cast = mul(x = var_16850_cast, y = var_16851_to_fp16)[name = tensor<string, []>("aw_1399_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16853_cast = softmax(axis = var_2634, x = aw_1361_cast)[name = tensor<string, []>("op_16853_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16854_cast = softmax(axis = var_2634, x = aw_1363_cast)[name = tensor<string, []>("op_16854_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16855_cast = softmax(axis = var_2634, x = aw_1365_cast)[name = tensor<string, []>("op_16855_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16856_cast = softmax(axis = var_2634, x = aw_1367_cast)[name = tensor<string, []>("op_16856_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16857_cast = softmax(axis = var_2634, x = aw_1369_cast)[name = tensor<string, []>("op_16857_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16858_cast = softmax(axis = var_2634, x = aw_1371_cast)[name = tensor<string, []>("op_16858_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16859_cast = softmax(axis = var_2634, x = aw_1373_cast)[name = tensor<string, []>("op_16859_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16860_cast = softmax(axis = var_2634, x = aw_1375_cast)[name = tensor<string, []>("op_16860_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16861_cast = softmax(axis = var_2634, x = aw_1377_cast)[name = tensor<string, []>("op_16861_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16862_cast = softmax(axis = var_2634, x = aw_1379_cast)[name = tensor<string, []>("op_16862_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16863_cast = softmax(axis = var_2634, x = aw_1381_cast)[name = tensor<string, []>("op_16863_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16864_cast = softmax(axis = var_2634, x = aw_1383_cast)[name = tensor<string, []>("op_16864_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16865_cast = softmax(axis = var_2634, x = aw_1385_cast)[name = tensor<string, []>("op_16865_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16866_cast = softmax(axis = var_2634, x = aw_1387_cast)[name = tensor<string, []>("op_16866_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16867_cast = softmax(axis = var_2634, x = aw_1389_cast)[name = tensor<string, []>("op_16867_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16868_cast = softmax(axis = var_2634, x = aw_1391_cast)[name = tensor<string, []>("op_16868_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16869_cast = softmax(axis = var_2634, x = aw_1393_cast)[name = tensor<string, []>("op_16869_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16870_cast = softmax(axis = var_2634, x = aw_1395_cast)[name = tensor<string, []>("op_16870_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16871_cast = softmax(axis = var_2634, x = aw_1397_cast)[name = tensor<string, []>("op_16871_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_16872_cast = softmax(axis = var_2634, x = aw_1399_cast)[name = tensor<string, []>("op_16872_cast")];
            tensor<string, []> var_16874_equation_0 = const()[name = tensor<string, []>("op_16874_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16874_cast = einsum(equation = var_16874_equation_0, values = (var_16694_cast, var_16853_cast))[name = tensor<string, []>("op_16874_cast")];
            tensor<string, []> var_16876_equation_0 = const()[name = tensor<string, []>("op_16876_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16876_cast = einsum(equation = var_16876_equation_0, values = (var_16698_cast, var_16854_cast))[name = tensor<string, []>("op_16876_cast")];
            tensor<string, []> var_16878_equation_0 = const()[name = tensor<string, []>("op_16878_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16878_cast = einsum(equation = var_16878_equation_0, values = (var_16702_cast, var_16855_cast))[name = tensor<string, []>("op_16878_cast")];
            tensor<string, []> var_16880_equation_0 = const()[name = tensor<string, []>("op_16880_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16880_cast = einsum(equation = var_16880_equation_0, values = (var_16706_cast, var_16856_cast))[name = tensor<string, []>("op_16880_cast")];
            tensor<string, []> var_16882_equation_0 = const()[name = tensor<string, []>("op_16882_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16882_cast = einsum(equation = var_16882_equation_0, values = (var_16710_cast, var_16857_cast))[name = tensor<string, []>("op_16882_cast")];
            tensor<string, []> var_16884_equation_0 = const()[name = tensor<string, []>("op_16884_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16884_cast = einsum(equation = var_16884_equation_0, values = (var_16714_cast, var_16858_cast))[name = tensor<string, []>("op_16884_cast")];
            tensor<string, []> var_16886_equation_0 = const()[name = tensor<string, []>("op_16886_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16886_cast = einsum(equation = var_16886_equation_0, values = (var_16718_cast, var_16859_cast))[name = tensor<string, []>("op_16886_cast")];
            tensor<string, []> var_16888_equation_0 = const()[name = tensor<string, []>("op_16888_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16888_cast = einsum(equation = var_16888_equation_0, values = (var_16722_cast, var_16860_cast))[name = tensor<string, []>("op_16888_cast")];
            tensor<string, []> var_16890_equation_0 = const()[name = tensor<string, []>("op_16890_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16890_cast = einsum(equation = var_16890_equation_0, values = (var_16726_cast, var_16861_cast))[name = tensor<string, []>("op_16890_cast")];
            tensor<string, []> var_16892_equation_0 = const()[name = tensor<string, []>("op_16892_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16892_cast = einsum(equation = var_16892_equation_0, values = (var_16730_cast, var_16862_cast))[name = tensor<string, []>("op_16892_cast")];
            tensor<string, []> var_16894_equation_0 = const()[name = tensor<string, []>("op_16894_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16894_cast = einsum(equation = var_16894_equation_0, values = (var_16734_cast, var_16863_cast))[name = tensor<string, []>("op_16894_cast")];
            tensor<string, []> var_16896_equation_0 = const()[name = tensor<string, []>("op_16896_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16896_cast = einsum(equation = var_16896_equation_0, values = (var_16738_cast, var_16864_cast))[name = tensor<string, []>("op_16896_cast")];
            tensor<string, []> var_16898_equation_0 = const()[name = tensor<string, []>("op_16898_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16898_cast = einsum(equation = var_16898_equation_0, values = (var_16742_cast, var_16865_cast))[name = tensor<string, []>("op_16898_cast")];
            tensor<string, []> var_16900_equation_0 = const()[name = tensor<string, []>("op_16900_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16900_cast = einsum(equation = var_16900_equation_0, values = (var_16746_cast, var_16866_cast))[name = tensor<string, []>("op_16900_cast")];
            tensor<string, []> var_16902_equation_0 = const()[name = tensor<string, []>("op_16902_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16902_cast = einsum(equation = var_16902_equation_0, values = (var_16750_cast, var_16867_cast))[name = tensor<string, []>("op_16902_cast")];
            tensor<string, []> var_16904_equation_0 = const()[name = tensor<string, []>("op_16904_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16904_cast = einsum(equation = var_16904_equation_0, values = (var_16754_cast, var_16868_cast))[name = tensor<string, []>("op_16904_cast")];
            tensor<string, []> var_16906_equation_0 = const()[name = tensor<string, []>("op_16906_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16906_cast = einsum(equation = var_16906_equation_0, values = (var_16758_cast, var_16869_cast))[name = tensor<string, []>("op_16906_cast")];
            tensor<string, []> var_16908_equation_0 = const()[name = tensor<string, []>("op_16908_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16908_cast = einsum(equation = var_16908_equation_0, values = (var_16762_cast, var_16870_cast))[name = tensor<string, []>("op_16908_cast")];
            tensor<string, []> var_16910_equation_0 = const()[name = tensor<string, []>("op_16910_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16910_cast = einsum(equation = var_16910_equation_0, values = (var_16766_cast, var_16871_cast))[name = tensor<string, []>("op_16910_cast")];
            tensor<string, []> var_16912_equation_0 = const()[name = tensor<string, []>("op_16912_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_16912_cast = einsum(equation = var_16912_equation_0, values = (var_16770_cast, var_16872_cast))[name = tensor<string, []>("op_16912_cast")];
            tensor<bool, []> input_273_interleave_0 = const()[name = tensor<string, []>("input_273_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_273_cast = concat(axis = var_2634, interleave = input_273_interleave_0, values = (var_16874_cast, var_16876_cast, var_16878_cast, var_16880_cast, var_16882_cast, var_16884_cast, var_16886_cast, var_16888_cast, var_16890_cast, var_16892_cast, var_16894_cast, var_16896_cast, var_16898_cast, var_16900_cast, var_16902_cast, var_16904_cast, var_16906_cast, var_16908_cast, var_16910_cast, var_16912_cast))[name = tensor<string, []>("input_273_cast")];
            tensor<int32, [2]> var_16918 = const()[name = tensor<string, []>("op_16918"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_16920 = const()[name = tensor<string, []>("op_16920"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_16922_pad_type_0 = const()[name = tensor<string, []>("op_16922_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_16922_pad_0 = const()[name = tensor<string, []>("op_16922_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1319538112)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1322814976)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_16922_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_out_0_bias_to_fp16, dilations = var_16920, groups = var_2634, pad = var_16922_pad_0, pad_type = var_16922_pad_type_0, strides = var_16918, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn1_to_out_0_weight_to_fp16, x = input_273_cast)[name = tensor<string, []>("op_16922_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_117_cast = add(x = var_16922_cast, y = inputs_115_cast)[name = tensor<string, []>("inputs_117_cast")];
            tensor<int32, [1]> var_16926 = const()[name = tensor<string, []>("op_16926"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_117_cast = reduce_mean(axes = var_16926, keep_dims = var_2629, x = inputs_117_cast)[name = tensor<string, []>("channels_mean_117_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_117_cast = sub(x = inputs_117_cast, y = channels_mean_117_cast)[name = tensor<string, []>("zero_mean_117_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_117_cast = mul(x = zero_mean_117_cast, y = zero_mean_117_cast)[name = tensor<string, []>("zero_mean_sq_117_cast")];
            tensor<int32, [1]> var_16930 = const()[name = tensor<string, []>("op_16930"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_16931_cast = reduce_mean(axes = var_16930, keep_dims = var_2629, x = zero_mean_sq_117_cast)[name = tensor<string, []>("op_16931_cast")];
            tensor<fp16, []> var_16932_to_fp16 = const()[name = tensor<string, []>("op_16932_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_16933_cast = add(x = var_16931_cast, y = var_16932_to_fp16)[name = tensor<string, []>("op_16933_cast")];
            tensor<fp16, []> denom_117_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_117_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_117_cast = rsqrt(epsilon = denom_117_epsilon_0_to_fp16, x = var_16933_cast)[name = tensor<string, []>("denom_117_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_117_cast = mul(x = zero_mean_117_cast, y = denom_117_cast)[name = tensor<string, []>("out_117_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_16937_to_fp16 = const()[name = tensor<string, []>("op_16937_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1322817600)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_16938_cast = add(x = out_117_cast, y = var_16937_to_fp16)[name = tensor<string, []>("op_16938_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_16940_to_fp16 = const()[name = tensor<string, []>("op_16940_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1322820224)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_169_cast = mul(x = var_16938_cast, y = var_16940_to_fp16)[name = tensor<string, []>("hidden_states_169_cast")];
            tensor<int32, [2]> var_16947 = const()[name = tensor<string, []>("op_16947"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_16949 = const()[name = tensor<string, []>("op_16949"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_79_pad_type_0 = const()[name = tensor<string, []>("q_79_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_79_pad_0 = const()[name = tensor<string, []>("q_79_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1322822848)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_79_cast = conv(dilations = var_16949, groups = var_2634, pad = q_79_pad_0, pad_type = q_79_pad_type_0, strides = var_16947, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_q_weight_to_fp16, x = hidden_states_169_cast)[name = tensor<string, []>("q_79_cast")];
            tensor<int32, [2]> var_16953 = const()[name = tensor<string, []>("op_16953"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_16955 = const()[name = tensor<string, []>("op_16955"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_157_pad_type_0 = const()[name = tensor<string, []>("k_157_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_157_pad_0 = const()[name = tensor<string, []>("k_157_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1326099712)))];
            tensor<fp16, [2, 1280, 1, 77]> k_157_cast = conv(dilations = var_16955, groups = var_2634, pad = k_157_pad_0, pad_type = k_157_pad_type_0, strides = var_16953, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_157_cast")];
            tensor<int32, [2]> var_16959 = const()[name = tensor<string, []>("op_16959"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_16961 = const()[name = tensor<string, []>("op_16961"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_79_pad_type_0 = const()[name = tensor<string, []>("v_79_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_79_pad_0 = const()[name = tensor<string, []>("v_79_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1331342656)))];
            tensor<fp16, [2, 1280, 1, 77]> v_79_cast = conv(dilations = var_16961, groups = var_2634, pad = v_79_pad_0, pad_type = v_79_pad_type_0, strides = var_16959, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_79_cast")];
            tensor<int32, [4]> var_16965_begin_0 = const()[name = tensor<string, []>("op_16965_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_16965_end_0 = const()[name = tensor<string, []>("op_16965_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_16965_end_mask_0 = const()[name = tensor<string, []>("op_16965_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16965_cast = slice_by_index(begin = var_16965_begin_0, end = var_16965_end_0, end_mask = var_16965_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_16965_cast")];
            tensor<int32, [4]> var_16969_begin_0 = const()[name = tensor<string, []>("op_16969_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_16969_end_0 = const()[name = tensor<string, []>("op_16969_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_16969_end_mask_0 = const()[name = tensor<string, []>("op_16969_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16969_cast = slice_by_index(begin = var_16969_begin_0, end = var_16969_end_0, end_mask = var_16969_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_16969_cast")];
            tensor<int32, [4]> var_16973_begin_0 = const()[name = tensor<string, []>("op_16973_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_16973_end_0 = const()[name = tensor<string, []>("op_16973_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_16973_end_mask_0 = const()[name = tensor<string, []>("op_16973_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16973_cast = slice_by_index(begin = var_16973_begin_0, end = var_16973_end_0, end_mask = var_16973_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_16973_cast")];
            tensor<int32, [4]> var_16977_begin_0 = const()[name = tensor<string, []>("op_16977_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_16977_end_0 = const()[name = tensor<string, []>("op_16977_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_16977_end_mask_0 = const()[name = tensor<string, []>("op_16977_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16977_cast = slice_by_index(begin = var_16977_begin_0, end = var_16977_end_0, end_mask = var_16977_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_16977_cast")];
            tensor<int32, [4]> var_16981_begin_0 = const()[name = tensor<string, []>("op_16981_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_16981_end_0 = const()[name = tensor<string, []>("op_16981_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_16981_end_mask_0 = const()[name = tensor<string, []>("op_16981_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16981_cast = slice_by_index(begin = var_16981_begin_0, end = var_16981_end_0, end_mask = var_16981_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_16981_cast")];
            tensor<int32, [4]> var_16985_begin_0 = const()[name = tensor<string, []>("op_16985_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_16985_end_0 = const()[name = tensor<string, []>("op_16985_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_16985_end_mask_0 = const()[name = tensor<string, []>("op_16985_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16985_cast = slice_by_index(begin = var_16985_begin_0, end = var_16985_end_0, end_mask = var_16985_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_16985_cast")];
            tensor<int32, [4]> var_16989_begin_0 = const()[name = tensor<string, []>("op_16989_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_16989_end_0 = const()[name = tensor<string, []>("op_16989_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_16989_end_mask_0 = const()[name = tensor<string, []>("op_16989_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16989_cast = slice_by_index(begin = var_16989_begin_0, end = var_16989_end_0, end_mask = var_16989_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_16989_cast")];
            tensor<int32, [4]> var_16993_begin_0 = const()[name = tensor<string, []>("op_16993_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_16993_end_0 = const()[name = tensor<string, []>("op_16993_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_16993_end_mask_0 = const()[name = tensor<string, []>("op_16993_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16993_cast = slice_by_index(begin = var_16993_begin_0, end = var_16993_end_0, end_mask = var_16993_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_16993_cast")];
            tensor<int32, [4]> var_16997_begin_0 = const()[name = tensor<string, []>("op_16997_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_16997_end_0 = const()[name = tensor<string, []>("op_16997_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_16997_end_mask_0 = const()[name = tensor<string, []>("op_16997_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_16997_cast = slice_by_index(begin = var_16997_begin_0, end = var_16997_end_0, end_mask = var_16997_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_16997_cast")];
            tensor<int32, [4]> var_17001_begin_0 = const()[name = tensor<string, []>("op_17001_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_17001_end_0 = const()[name = tensor<string, []>("op_17001_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_17001_end_mask_0 = const()[name = tensor<string, []>("op_17001_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17001_cast = slice_by_index(begin = var_17001_begin_0, end = var_17001_end_0, end_mask = var_17001_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_17001_cast")];
            tensor<int32, [4]> var_17005_begin_0 = const()[name = tensor<string, []>("op_17005_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_17005_end_0 = const()[name = tensor<string, []>("op_17005_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_17005_end_mask_0 = const()[name = tensor<string, []>("op_17005_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17005_cast = slice_by_index(begin = var_17005_begin_0, end = var_17005_end_0, end_mask = var_17005_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_17005_cast")];
            tensor<int32, [4]> var_17009_begin_0 = const()[name = tensor<string, []>("op_17009_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_17009_end_0 = const()[name = tensor<string, []>("op_17009_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_17009_end_mask_0 = const()[name = tensor<string, []>("op_17009_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17009_cast = slice_by_index(begin = var_17009_begin_0, end = var_17009_end_0, end_mask = var_17009_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_17009_cast")];
            tensor<int32, [4]> var_17013_begin_0 = const()[name = tensor<string, []>("op_17013_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_17013_end_0 = const()[name = tensor<string, []>("op_17013_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_17013_end_mask_0 = const()[name = tensor<string, []>("op_17013_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17013_cast = slice_by_index(begin = var_17013_begin_0, end = var_17013_end_0, end_mask = var_17013_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_17013_cast")];
            tensor<int32, [4]> var_17017_begin_0 = const()[name = tensor<string, []>("op_17017_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_17017_end_0 = const()[name = tensor<string, []>("op_17017_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_17017_end_mask_0 = const()[name = tensor<string, []>("op_17017_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17017_cast = slice_by_index(begin = var_17017_begin_0, end = var_17017_end_0, end_mask = var_17017_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_17017_cast")];
            tensor<int32, [4]> var_17021_begin_0 = const()[name = tensor<string, []>("op_17021_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_17021_end_0 = const()[name = tensor<string, []>("op_17021_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_17021_end_mask_0 = const()[name = tensor<string, []>("op_17021_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17021_cast = slice_by_index(begin = var_17021_begin_0, end = var_17021_end_0, end_mask = var_17021_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_17021_cast")];
            tensor<int32, [4]> var_17025_begin_0 = const()[name = tensor<string, []>("op_17025_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_17025_end_0 = const()[name = tensor<string, []>("op_17025_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_17025_end_mask_0 = const()[name = tensor<string, []>("op_17025_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17025_cast = slice_by_index(begin = var_17025_begin_0, end = var_17025_end_0, end_mask = var_17025_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_17025_cast")];
            tensor<int32, [4]> var_17029_begin_0 = const()[name = tensor<string, []>("op_17029_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_17029_end_0 = const()[name = tensor<string, []>("op_17029_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_17029_end_mask_0 = const()[name = tensor<string, []>("op_17029_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17029_cast = slice_by_index(begin = var_17029_begin_0, end = var_17029_end_0, end_mask = var_17029_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_17029_cast")];
            tensor<int32, [4]> var_17033_begin_0 = const()[name = tensor<string, []>("op_17033_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_17033_end_0 = const()[name = tensor<string, []>("op_17033_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_17033_end_mask_0 = const()[name = tensor<string, []>("op_17033_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17033_cast = slice_by_index(begin = var_17033_begin_0, end = var_17033_end_0, end_mask = var_17033_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_17033_cast")];
            tensor<int32, [4]> var_17037_begin_0 = const()[name = tensor<string, []>("op_17037_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_17037_end_0 = const()[name = tensor<string, []>("op_17037_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_17037_end_mask_0 = const()[name = tensor<string, []>("op_17037_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17037_cast = slice_by_index(begin = var_17037_begin_0, end = var_17037_end_0, end_mask = var_17037_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_17037_cast")];
            tensor<int32, [4]> var_17041_begin_0 = const()[name = tensor<string, []>("op_17041_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_17041_end_0 = const()[name = tensor<string, []>("op_17041_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_17041_end_mask_0 = const()[name = tensor<string, []>("op_17041_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17041_cast = slice_by_index(begin = var_17041_begin_0, end = var_17041_end_0, end_mask = var_17041_end_mask_0, x = q_79_cast)[name = tensor<string, []>("op_17041_cast")];
            tensor<int32, [4]> k_159_perm_0 = const()[name = tensor<string, []>("k_159_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_17048_begin_0 = const()[name = tensor<string, []>("op_17048_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_17048_end_0 = const()[name = tensor<string, []>("op_17048_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_17048_end_mask_0 = const()[name = tensor<string, []>("op_17048_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_100 = transpose(perm = k_159_perm_0, x = k_157_cast)[name = tensor<string, []>("transpose_100")];
            tensor<fp16, [2, 77, 1, 64]> var_17048_cast = slice_by_index(begin = var_17048_begin_0, end = var_17048_end_0, end_mask = var_17048_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17048_cast")];
            tensor<int32, [4]> var_17052_begin_0 = const()[name = tensor<string, []>("op_17052_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_17052_end_0 = const()[name = tensor<string, []>("op_17052_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_17052_end_mask_0 = const()[name = tensor<string, []>("op_17052_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17052_cast = slice_by_index(begin = var_17052_begin_0, end = var_17052_end_0, end_mask = var_17052_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17052_cast")];
            tensor<int32, [4]> var_17056_begin_0 = const()[name = tensor<string, []>("op_17056_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_17056_end_0 = const()[name = tensor<string, []>("op_17056_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_17056_end_mask_0 = const()[name = tensor<string, []>("op_17056_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17056_cast = slice_by_index(begin = var_17056_begin_0, end = var_17056_end_0, end_mask = var_17056_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17056_cast")];
            tensor<int32, [4]> var_17060_begin_0 = const()[name = tensor<string, []>("op_17060_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_17060_end_0 = const()[name = tensor<string, []>("op_17060_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_17060_end_mask_0 = const()[name = tensor<string, []>("op_17060_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17060_cast = slice_by_index(begin = var_17060_begin_0, end = var_17060_end_0, end_mask = var_17060_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17060_cast")];
            tensor<int32, [4]> var_17064_begin_0 = const()[name = tensor<string, []>("op_17064_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_17064_end_0 = const()[name = tensor<string, []>("op_17064_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_17064_end_mask_0 = const()[name = tensor<string, []>("op_17064_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17064_cast = slice_by_index(begin = var_17064_begin_0, end = var_17064_end_0, end_mask = var_17064_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17064_cast")];
            tensor<int32, [4]> var_17068_begin_0 = const()[name = tensor<string, []>("op_17068_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_17068_end_0 = const()[name = tensor<string, []>("op_17068_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_17068_end_mask_0 = const()[name = tensor<string, []>("op_17068_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17068_cast = slice_by_index(begin = var_17068_begin_0, end = var_17068_end_0, end_mask = var_17068_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17068_cast")];
            tensor<int32, [4]> var_17072_begin_0 = const()[name = tensor<string, []>("op_17072_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_17072_end_0 = const()[name = tensor<string, []>("op_17072_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_17072_end_mask_0 = const()[name = tensor<string, []>("op_17072_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17072_cast = slice_by_index(begin = var_17072_begin_0, end = var_17072_end_0, end_mask = var_17072_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17072_cast")];
            tensor<int32, [4]> var_17076_begin_0 = const()[name = tensor<string, []>("op_17076_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_17076_end_0 = const()[name = tensor<string, []>("op_17076_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_17076_end_mask_0 = const()[name = tensor<string, []>("op_17076_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17076_cast = slice_by_index(begin = var_17076_begin_0, end = var_17076_end_0, end_mask = var_17076_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17076_cast")];
            tensor<int32, [4]> var_17080_begin_0 = const()[name = tensor<string, []>("op_17080_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_17080_end_0 = const()[name = tensor<string, []>("op_17080_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_17080_end_mask_0 = const()[name = tensor<string, []>("op_17080_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17080_cast = slice_by_index(begin = var_17080_begin_0, end = var_17080_end_0, end_mask = var_17080_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17080_cast")];
            tensor<int32, [4]> var_17084_begin_0 = const()[name = tensor<string, []>("op_17084_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_17084_end_0 = const()[name = tensor<string, []>("op_17084_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_17084_end_mask_0 = const()[name = tensor<string, []>("op_17084_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17084_cast = slice_by_index(begin = var_17084_begin_0, end = var_17084_end_0, end_mask = var_17084_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17084_cast")];
            tensor<int32, [4]> var_17088_begin_0 = const()[name = tensor<string, []>("op_17088_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_17088_end_0 = const()[name = tensor<string, []>("op_17088_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_17088_end_mask_0 = const()[name = tensor<string, []>("op_17088_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17088_cast = slice_by_index(begin = var_17088_begin_0, end = var_17088_end_0, end_mask = var_17088_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17088_cast")];
            tensor<int32, [4]> var_17092_begin_0 = const()[name = tensor<string, []>("op_17092_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_17092_end_0 = const()[name = tensor<string, []>("op_17092_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_17092_end_mask_0 = const()[name = tensor<string, []>("op_17092_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17092_cast = slice_by_index(begin = var_17092_begin_0, end = var_17092_end_0, end_mask = var_17092_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17092_cast")];
            tensor<int32, [4]> var_17096_begin_0 = const()[name = tensor<string, []>("op_17096_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_17096_end_0 = const()[name = tensor<string, []>("op_17096_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_17096_end_mask_0 = const()[name = tensor<string, []>("op_17096_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17096_cast = slice_by_index(begin = var_17096_begin_0, end = var_17096_end_0, end_mask = var_17096_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17096_cast")];
            tensor<int32, [4]> var_17100_begin_0 = const()[name = tensor<string, []>("op_17100_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_17100_end_0 = const()[name = tensor<string, []>("op_17100_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_17100_end_mask_0 = const()[name = tensor<string, []>("op_17100_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17100_cast = slice_by_index(begin = var_17100_begin_0, end = var_17100_end_0, end_mask = var_17100_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17100_cast")];
            tensor<int32, [4]> var_17104_begin_0 = const()[name = tensor<string, []>("op_17104_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_17104_end_0 = const()[name = tensor<string, []>("op_17104_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_17104_end_mask_0 = const()[name = tensor<string, []>("op_17104_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17104_cast = slice_by_index(begin = var_17104_begin_0, end = var_17104_end_0, end_mask = var_17104_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17104_cast")];
            tensor<int32, [4]> var_17108_begin_0 = const()[name = tensor<string, []>("op_17108_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_17108_end_0 = const()[name = tensor<string, []>("op_17108_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_17108_end_mask_0 = const()[name = tensor<string, []>("op_17108_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17108_cast = slice_by_index(begin = var_17108_begin_0, end = var_17108_end_0, end_mask = var_17108_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17108_cast")];
            tensor<int32, [4]> var_17112_begin_0 = const()[name = tensor<string, []>("op_17112_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_17112_end_0 = const()[name = tensor<string, []>("op_17112_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_17112_end_mask_0 = const()[name = tensor<string, []>("op_17112_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17112_cast = slice_by_index(begin = var_17112_begin_0, end = var_17112_end_0, end_mask = var_17112_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17112_cast")];
            tensor<int32, [4]> var_17116_begin_0 = const()[name = tensor<string, []>("op_17116_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_17116_end_0 = const()[name = tensor<string, []>("op_17116_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_17116_end_mask_0 = const()[name = tensor<string, []>("op_17116_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17116_cast = slice_by_index(begin = var_17116_begin_0, end = var_17116_end_0, end_mask = var_17116_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17116_cast")];
            tensor<int32, [4]> var_17120_begin_0 = const()[name = tensor<string, []>("op_17120_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_17120_end_0 = const()[name = tensor<string, []>("op_17120_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_17120_end_mask_0 = const()[name = tensor<string, []>("op_17120_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17120_cast = slice_by_index(begin = var_17120_begin_0, end = var_17120_end_0, end_mask = var_17120_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17120_cast")];
            tensor<int32, [4]> var_17124_begin_0 = const()[name = tensor<string, []>("op_17124_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_17124_end_0 = const()[name = tensor<string, []>("op_17124_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_17124_end_mask_0 = const()[name = tensor<string, []>("op_17124_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17124_cast = slice_by_index(begin = var_17124_begin_0, end = var_17124_end_0, end_mask = var_17124_end_mask_0, x = transpose_100)[name = tensor<string, []>("op_17124_cast")];
            tensor<int32, [4]> var_17126_begin_0 = const()[name = tensor<string, []>("op_17126_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_17126_end_0 = const()[name = tensor<string, []>("op_17126_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_17126_end_mask_0 = const()[name = tensor<string, []>("op_17126_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17126_cast = slice_by_index(begin = var_17126_begin_0, end = var_17126_end_0, end_mask = var_17126_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17126_cast")];
            tensor<int32, [4]> var_17130_begin_0 = const()[name = tensor<string, []>("op_17130_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_17130_end_0 = const()[name = tensor<string, []>("op_17130_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_17130_end_mask_0 = const()[name = tensor<string, []>("op_17130_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17130_cast = slice_by_index(begin = var_17130_begin_0, end = var_17130_end_0, end_mask = var_17130_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17130_cast")];
            tensor<int32, [4]> var_17134_begin_0 = const()[name = tensor<string, []>("op_17134_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_17134_end_0 = const()[name = tensor<string, []>("op_17134_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_17134_end_mask_0 = const()[name = tensor<string, []>("op_17134_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17134_cast = slice_by_index(begin = var_17134_begin_0, end = var_17134_end_0, end_mask = var_17134_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17134_cast")];
            tensor<int32, [4]> var_17138_begin_0 = const()[name = tensor<string, []>("op_17138_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_17138_end_0 = const()[name = tensor<string, []>("op_17138_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_17138_end_mask_0 = const()[name = tensor<string, []>("op_17138_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17138_cast = slice_by_index(begin = var_17138_begin_0, end = var_17138_end_0, end_mask = var_17138_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17138_cast")];
            tensor<int32, [4]> var_17142_begin_0 = const()[name = tensor<string, []>("op_17142_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_17142_end_0 = const()[name = tensor<string, []>("op_17142_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_17142_end_mask_0 = const()[name = tensor<string, []>("op_17142_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17142_cast = slice_by_index(begin = var_17142_begin_0, end = var_17142_end_0, end_mask = var_17142_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17142_cast")];
            tensor<int32, [4]> var_17146_begin_0 = const()[name = tensor<string, []>("op_17146_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_17146_end_0 = const()[name = tensor<string, []>("op_17146_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_17146_end_mask_0 = const()[name = tensor<string, []>("op_17146_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17146_cast = slice_by_index(begin = var_17146_begin_0, end = var_17146_end_0, end_mask = var_17146_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17146_cast")];
            tensor<int32, [4]> var_17150_begin_0 = const()[name = tensor<string, []>("op_17150_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_17150_end_0 = const()[name = tensor<string, []>("op_17150_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_17150_end_mask_0 = const()[name = tensor<string, []>("op_17150_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17150_cast = slice_by_index(begin = var_17150_begin_0, end = var_17150_end_0, end_mask = var_17150_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17150_cast")];
            tensor<int32, [4]> var_17154_begin_0 = const()[name = tensor<string, []>("op_17154_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_17154_end_0 = const()[name = tensor<string, []>("op_17154_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_17154_end_mask_0 = const()[name = tensor<string, []>("op_17154_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17154_cast = slice_by_index(begin = var_17154_begin_0, end = var_17154_end_0, end_mask = var_17154_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17154_cast")];
            tensor<int32, [4]> var_17158_begin_0 = const()[name = tensor<string, []>("op_17158_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_17158_end_0 = const()[name = tensor<string, []>("op_17158_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_17158_end_mask_0 = const()[name = tensor<string, []>("op_17158_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17158_cast = slice_by_index(begin = var_17158_begin_0, end = var_17158_end_0, end_mask = var_17158_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17158_cast")];
            tensor<int32, [4]> var_17162_begin_0 = const()[name = tensor<string, []>("op_17162_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_17162_end_0 = const()[name = tensor<string, []>("op_17162_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_17162_end_mask_0 = const()[name = tensor<string, []>("op_17162_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17162_cast = slice_by_index(begin = var_17162_begin_0, end = var_17162_end_0, end_mask = var_17162_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17162_cast")];
            tensor<int32, [4]> var_17166_begin_0 = const()[name = tensor<string, []>("op_17166_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_17166_end_0 = const()[name = tensor<string, []>("op_17166_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_17166_end_mask_0 = const()[name = tensor<string, []>("op_17166_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17166_cast = slice_by_index(begin = var_17166_begin_0, end = var_17166_end_0, end_mask = var_17166_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17166_cast")];
            tensor<int32, [4]> var_17170_begin_0 = const()[name = tensor<string, []>("op_17170_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_17170_end_0 = const()[name = tensor<string, []>("op_17170_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_17170_end_mask_0 = const()[name = tensor<string, []>("op_17170_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17170_cast = slice_by_index(begin = var_17170_begin_0, end = var_17170_end_0, end_mask = var_17170_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17170_cast")];
            tensor<int32, [4]> var_17174_begin_0 = const()[name = tensor<string, []>("op_17174_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_17174_end_0 = const()[name = tensor<string, []>("op_17174_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_17174_end_mask_0 = const()[name = tensor<string, []>("op_17174_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17174_cast = slice_by_index(begin = var_17174_begin_0, end = var_17174_end_0, end_mask = var_17174_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17174_cast")];
            tensor<int32, [4]> var_17178_begin_0 = const()[name = tensor<string, []>("op_17178_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_17178_end_0 = const()[name = tensor<string, []>("op_17178_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_17178_end_mask_0 = const()[name = tensor<string, []>("op_17178_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17178_cast = slice_by_index(begin = var_17178_begin_0, end = var_17178_end_0, end_mask = var_17178_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17178_cast")];
            tensor<int32, [4]> var_17182_begin_0 = const()[name = tensor<string, []>("op_17182_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_17182_end_0 = const()[name = tensor<string, []>("op_17182_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_17182_end_mask_0 = const()[name = tensor<string, []>("op_17182_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17182_cast = slice_by_index(begin = var_17182_begin_0, end = var_17182_end_0, end_mask = var_17182_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17182_cast")];
            tensor<int32, [4]> var_17186_begin_0 = const()[name = tensor<string, []>("op_17186_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_17186_end_0 = const()[name = tensor<string, []>("op_17186_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_17186_end_mask_0 = const()[name = tensor<string, []>("op_17186_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17186_cast = slice_by_index(begin = var_17186_begin_0, end = var_17186_end_0, end_mask = var_17186_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17186_cast")];
            tensor<int32, [4]> var_17190_begin_0 = const()[name = tensor<string, []>("op_17190_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_17190_end_0 = const()[name = tensor<string, []>("op_17190_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_17190_end_mask_0 = const()[name = tensor<string, []>("op_17190_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17190_cast = slice_by_index(begin = var_17190_begin_0, end = var_17190_end_0, end_mask = var_17190_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17190_cast")];
            tensor<int32, [4]> var_17194_begin_0 = const()[name = tensor<string, []>("op_17194_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_17194_end_0 = const()[name = tensor<string, []>("op_17194_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_17194_end_mask_0 = const()[name = tensor<string, []>("op_17194_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17194_cast = slice_by_index(begin = var_17194_begin_0, end = var_17194_end_0, end_mask = var_17194_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17194_cast")];
            tensor<int32, [4]> var_17198_begin_0 = const()[name = tensor<string, []>("op_17198_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_17198_end_0 = const()[name = tensor<string, []>("op_17198_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_17198_end_mask_0 = const()[name = tensor<string, []>("op_17198_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17198_cast = slice_by_index(begin = var_17198_begin_0, end = var_17198_end_0, end_mask = var_17198_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17198_cast")];
            tensor<int32, [4]> var_17202_begin_0 = const()[name = tensor<string, []>("op_17202_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_17202_end_0 = const()[name = tensor<string, []>("op_17202_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_17202_end_mask_0 = const()[name = tensor<string, []>("op_17202_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_17202_cast = slice_by_index(begin = var_17202_begin_0, end = var_17202_end_0, end_mask = var_17202_end_mask_0, x = v_79_cast)[name = tensor<string, []>("op_17202_cast")];
            tensor<string, []> var_17206_equation_0 = const()[name = tensor<string, []>("op_17206_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17206_cast = einsum(equation = var_17206_equation_0, values = (var_17048_cast, var_16965_cast))[name = tensor<string, []>("op_17206_cast")];
            tensor<fp16, []> var_17207_to_fp16 = const()[name = tensor<string, []>("op_17207_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1401_cast = mul(x = var_17206_cast, y = var_17207_to_fp16)[name = tensor<string, []>("aw_1401_cast")];
            tensor<string, []> var_17210_equation_0 = const()[name = tensor<string, []>("op_17210_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17210_cast = einsum(equation = var_17210_equation_0, values = (var_17052_cast, var_16969_cast))[name = tensor<string, []>("op_17210_cast")];
            tensor<fp16, []> var_17211_to_fp16 = const()[name = tensor<string, []>("op_17211_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1403_cast = mul(x = var_17210_cast, y = var_17211_to_fp16)[name = tensor<string, []>("aw_1403_cast")];
            tensor<string, []> var_17214_equation_0 = const()[name = tensor<string, []>("op_17214_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17214_cast = einsum(equation = var_17214_equation_0, values = (var_17056_cast, var_16973_cast))[name = tensor<string, []>("op_17214_cast")];
            tensor<fp16, []> var_17215_to_fp16 = const()[name = tensor<string, []>("op_17215_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1405_cast = mul(x = var_17214_cast, y = var_17215_to_fp16)[name = tensor<string, []>("aw_1405_cast")];
            tensor<string, []> var_17218_equation_0 = const()[name = tensor<string, []>("op_17218_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17218_cast = einsum(equation = var_17218_equation_0, values = (var_17060_cast, var_16977_cast))[name = tensor<string, []>("op_17218_cast")];
            tensor<fp16, []> var_17219_to_fp16 = const()[name = tensor<string, []>("op_17219_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1407_cast = mul(x = var_17218_cast, y = var_17219_to_fp16)[name = tensor<string, []>("aw_1407_cast")];
            tensor<string, []> var_17222_equation_0 = const()[name = tensor<string, []>("op_17222_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17222_cast = einsum(equation = var_17222_equation_0, values = (var_17064_cast, var_16981_cast))[name = tensor<string, []>("op_17222_cast")];
            tensor<fp16, []> var_17223_to_fp16 = const()[name = tensor<string, []>("op_17223_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1409_cast = mul(x = var_17222_cast, y = var_17223_to_fp16)[name = tensor<string, []>("aw_1409_cast")];
            tensor<string, []> var_17226_equation_0 = const()[name = tensor<string, []>("op_17226_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17226_cast = einsum(equation = var_17226_equation_0, values = (var_17068_cast, var_16985_cast))[name = tensor<string, []>("op_17226_cast")];
            tensor<fp16, []> var_17227_to_fp16 = const()[name = tensor<string, []>("op_17227_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1411_cast = mul(x = var_17226_cast, y = var_17227_to_fp16)[name = tensor<string, []>("aw_1411_cast")];
            tensor<string, []> var_17230_equation_0 = const()[name = tensor<string, []>("op_17230_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17230_cast = einsum(equation = var_17230_equation_0, values = (var_17072_cast, var_16989_cast))[name = tensor<string, []>("op_17230_cast")];
            tensor<fp16, []> var_17231_to_fp16 = const()[name = tensor<string, []>("op_17231_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1413_cast = mul(x = var_17230_cast, y = var_17231_to_fp16)[name = tensor<string, []>("aw_1413_cast")];
            tensor<string, []> var_17234_equation_0 = const()[name = tensor<string, []>("op_17234_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17234_cast = einsum(equation = var_17234_equation_0, values = (var_17076_cast, var_16993_cast))[name = tensor<string, []>("op_17234_cast")];
            tensor<fp16, []> var_17235_to_fp16 = const()[name = tensor<string, []>("op_17235_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1415_cast = mul(x = var_17234_cast, y = var_17235_to_fp16)[name = tensor<string, []>("aw_1415_cast")];
            tensor<string, []> var_17238_equation_0 = const()[name = tensor<string, []>("op_17238_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17238_cast = einsum(equation = var_17238_equation_0, values = (var_17080_cast, var_16997_cast))[name = tensor<string, []>("op_17238_cast")];
            tensor<fp16, []> var_17239_to_fp16 = const()[name = tensor<string, []>("op_17239_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1417_cast = mul(x = var_17238_cast, y = var_17239_to_fp16)[name = tensor<string, []>("aw_1417_cast")];
            tensor<string, []> var_17242_equation_0 = const()[name = tensor<string, []>("op_17242_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17242_cast = einsum(equation = var_17242_equation_0, values = (var_17084_cast, var_17001_cast))[name = tensor<string, []>("op_17242_cast")];
            tensor<fp16, []> var_17243_to_fp16 = const()[name = tensor<string, []>("op_17243_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1419_cast = mul(x = var_17242_cast, y = var_17243_to_fp16)[name = tensor<string, []>("aw_1419_cast")];
            tensor<string, []> var_17246_equation_0 = const()[name = tensor<string, []>("op_17246_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17246_cast = einsum(equation = var_17246_equation_0, values = (var_17088_cast, var_17005_cast))[name = tensor<string, []>("op_17246_cast")];
            tensor<fp16, []> var_17247_to_fp16 = const()[name = tensor<string, []>("op_17247_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1421_cast = mul(x = var_17246_cast, y = var_17247_to_fp16)[name = tensor<string, []>("aw_1421_cast")];
            tensor<string, []> var_17250_equation_0 = const()[name = tensor<string, []>("op_17250_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17250_cast = einsum(equation = var_17250_equation_0, values = (var_17092_cast, var_17009_cast))[name = tensor<string, []>("op_17250_cast")];
            tensor<fp16, []> var_17251_to_fp16 = const()[name = tensor<string, []>("op_17251_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1423_cast = mul(x = var_17250_cast, y = var_17251_to_fp16)[name = tensor<string, []>("aw_1423_cast")];
            tensor<string, []> var_17254_equation_0 = const()[name = tensor<string, []>("op_17254_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17254_cast = einsum(equation = var_17254_equation_0, values = (var_17096_cast, var_17013_cast))[name = tensor<string, []>("op_17254_cast")];
            tensor<fp16, []> var_17255_to_fp16 = const()[name = tensor<string, []>("op_17255_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1425_cast = mul(x = var_17254_cast, y = var_17255_to_fp16)[name = tensor<string, []>("aw_1425_cast")];
            tensor<string, []> var_17258_equation_0 = const()[name = tensor<string, []>("op_17258_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17258_cast = einsum(equation = var_17258_equation_0, values = (var_17100_cast, var_17017_cast))[name = tensor<string, []>("op_17258_cast")];
            tensor<fp16, []> var_17259_to_fp16 = const()[name = tensor<string, []>("op_17259_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1427_cast = mul(x = var_17258_cast, y = var_17259_to_fp16)[name = tensor<string, []>("aw_1427_cast")];
            tensor<string, []> var_17262_equation_0 = const()[name = tensor<string, []>("op_17262_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17262_cast = einsum(equation = var_17262_equation_0, values = (var_17104_cast, var_17021_cast))[name = tensor<string, []>("op_17262_cast")];
            tensor<fp16, []> var_17263_to_fp16 = const()[name = tensor<string, []>("op_17263_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1429_cast = mul(x = var_17262_cast, y = var_17263_to_fp16)[name = tensor<string, []>("aw_1429_cast")];
            tensor<string, []> var_17266_equation_0 = const()[name = tensor<string, []>("op_17266_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17266_cast = einsum(equation = var_17266_equation_0, values = (var_17108_cast, var_17025_cast))[name = tensor<string, []>("op_17266_cast")];
            tensor<fp16, []> var_17267_to_fp16 = const()[name = tensor<string, []>("op_17267_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1431_cast = mul(x = var_17266_cast, y = var_17267_to_fp16)[name = tensor<string, []>("aw_1431_cast")];
            tensor<string, []> var_17270_equation_0 = const()[name = tensor<string, []>("op_17270_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17270_cast = einsum(equation = var_17270_equation_0, values = (var_17112_cast, var_17029_cast))[name = tensor<string, []>("op_17270_cast")];
            tensor<fp16, []> var_17271_to_fp16 = const()[name = tensor<string, []>("op_17271_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1433_cast = mul(x = var_17270_cast, y = var_17271_to_fp16)[name = tensor<string, []>("aw_1433_cast")];
            tensor<string, []> var_17274_equation_0 = const()[name = tensor<string, []>("op_17274_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17274_cast = einsum(equation = var_17274_equation_0, values = (var_17116_cast, var_17033_cast))[name = tensor<string, []>("op_17274_cast")];
            tensor<fp16, []> var_17275_to_fp16 = const()[name = tensor<string, []>("op_17275_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1435_cast = mul(x = var_17274_cast, y = var_17275_to_fp16)[name = tensor<string, []>("aw_1435_cast")];
            tensor<string, []> var_17278_equation_0 = const()[name = tensor<string, []>("op_17278_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17278_cast = einsum(equation = var_17278_equation_0, values = (var_17120_cast, var_17037_cast))[name = tensor<string, []>("op_17278_cast")];
            tensor<fp16, []> var_17279_to_fp16 = const()[name = tensor<string, []>("op_17279_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1437_cast = mul(x = var_17278_cast, y = var_17279_to_fp16)[name = tensor<string, []>("aw_1437_cast")];
            tensor<string, []> var_17282_equation_0 = const()[name = tensor<string, []>("op_17282_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_17282_cast = einsum(equation = var_17282_equation_0, values = (var_17124_cast, var_17041_cast))[name = tensor<string, []>("op_17282_cast")];
            tensor<fp16, []> var_17283_to_fp16 = const()[name = tensor<string, []>("op_17283_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1439_cast = mul(x = var_17282_cast, y = var_17283_to_fp16)[name = tensor<string, []>("aw_1439_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17285_cast = softmax(axis = var_2634, x = aw_1401_cast)[name = tensor<string, []>("op_17285_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17286_cast = softmax(axis = var_2634, x = aw_1403_cast)[name = tensor<string, []>("op_17286_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17287_cast = softmax(axis = var_2634, x = aw_1405_cast)[name = tensor<string, []>("op_17287_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17288_cast = softmax(axis = var_2634, x = aw_1407_cast)[name = tensor<string, []>("op_17288_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17289_cast = softmax(axis = var_2634, x = aw_1409_cast)[name = tensor<string, []>("op_17289_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17290_cast = softmax(axis = var_2634, x = aw_1411_cast)[name = tensor<string, []>("op_17290_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17291_cast = softmax(axis = var_2634, x = aw_1413_cast)[name = tensor<string, []>("op_17291_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17292_cast = softmax(axis = var_2634, x = aw_1415_cast)[name = tensor<string, []>("op_17292_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17293_cast = softmax(axis = var_2634, x = aw_1417_cast)[name = tensor<string, []>("op_17293_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17294_cast = softmax(axis = var_2634, x = aw_1419_cast)[name = tensor<string, []>("op_17294_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17295_cast = softmax(axis = var_2634, x = aw_1421_cast)[name = tensor<string, []>("op_17295_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17296_cast = softmax(axis = var_2634, x = aw_1423_cast)[name = tensor<string, []>("op_17296_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17297_cast = softmax(axis = var_2634, x = aw_1425_cast)[name = tensor<string, []>("op_17297_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17298_cast = softmax(axis = var_2634, x = aw_1427_cast)[name = tensor<string, []>("op_17298_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17299_cast = softmax(axis = var_2634, x = aw_1429_cast)[name = tensor<string, []>("op_17299_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17300_cast = softmax(axis = var_2634, x = aw_1431_cast)[name = tensor<string, []>("op_17300_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17301_cast = softmax(axis = var_2634, x = aw_1433_cast)[name = tensor<string, []>("op_17301_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17302_cast = softmax(axis = var_2634, x = aw_1435_cast)[name = tensor<string, []>("op_17302_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17303_cast = softmax(axis = var_2634, x = aw_1437_cast)[name = tensor<string, []>("op_17303_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_17304_cast = softmax(axis = var_2634, x = aw_1439_cast)[name = tensor<string, []>("op_17304_cast")];
            tensor<string, []> var_17306_equation_0 = const()[name = tensor<string, []>("op_17306_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17306_cast = einsum(equation = var_17306_equation_0, values = (var_17126_cast, var_17285_cast))[name = tensor<string, []>("op_17306_cast")];
            tensor<string, []> var_17308_equation_0 = const()[name = tensor<string, []>("op_17308_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17308_cast = einsum(equation = var_17308_equation_0, values = (var_17130_cast, var_17286_cast))[name = tensor<string, []>("op_17308_cast")];
            tensor<string, []> var_17310_equation_0 = const()[name = tensor<string, []>("op_17310_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17310_cast = einsum(equation = var_17310_equation_0, values = (var_17134_cast, var_17287_cast))[name = tensor<string, []>("op_17310_cast")];
            tensor<string, []> var_17312_equation_0 = const()[name = tensor<string, []>("op_17312_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17312_cast = einsum(equation = var_17312_equation_0, values = (var_17138_cast, var_17288_cast))[name = tensor<string, []>("op_17312_cast")];
            tensor<string, []> var_17314_equation_0 = const()[name = tensor<string, []>("op_17314_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17314_cast = einsum(equation = var_17314_equation_0, values = (var_17142_cast, var_17289_cast))[name = tensor<string, []>("op_17314_cast")];
            tensor<string, []> var_17316_equation_0 = const()[name = tensor<string, []>("op_17316_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17316_cast = einsum(equation = var_17316_equation_0, values = (var_17146_cast, var_17290_cast))[name = tensor<string, []>("op_17316_cast")];
            tensor<string, []> var_17318_equation_0 = const()[name = tensor<string, []>("op_17318_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17318_cast = einsum(equation = var_17318_equation_0, values = (var_17150_cast, var_17291_cast))[name = tensor<string, []>("op_17318_cast")];
            tensor<string, []> var_17320_equation_0 = const()[name = tensor<string, []>("op_17320_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17320_cast = einsum(equation = var_17320_equation_0, values = (var_17154_cast, var_17292_cast))[name = tensor<string, []>("op_17320_cast")];
            tensor<string, []> var_17322_equation_0 = const()[name = tensor<string, []>("op_17322_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17322_cast = einsum(equation = var_17322_equation_0, values = (var_17158_cast, var_17293_cast))[name = tensor<string, []>("op_17322_cast")];
            tensor<string, []> var_17324_equation_0 = const()[name = tensor<string, []>("op_17324_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17324_cast = einsum(equation = var_17324_equation_0, values = (var_17162_cast, var_17294_cast))[name = tensor<string, []>("op_17324_cast")];
            tensor<string, []> var_17326_equation_0 = const()[name = tensor<string, []>("op_17326_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17326_cast = einsum(equation = var_17326_equation_0, values = (var_17166_cast, var_17295_cast))[name = tensor<string, []>("op_17326_cast")];
            tensor<string, []> var_17328_equation_0 = const()[name = tensor<string, []>("op_17328_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17328_cast = einsum(equation = var_17328_equation_0, values = (var_17170_cast, var_17296_cast))[name = tensor<string, []>("op_17328_cast")];
            tensor<string, []> var_17330_equation_0 = const()[name = tensor<string, []>("op_17330_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17330_cast = einsum(equation = var_17330_equation_0, values = (var_17174_cast, var_17297_cast))[name = tensor<string, []>("op_17330_cast")];
            tensor<string, []> var_17332_equation_0 = const()[name = tensor<string, []>("op_17332_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17332_cast = einsum(equation = var_17332_equation_0, values = (var_17178_cast, var_17298_cast))[name = tensor<string, []>("op_17332_cast")];
            tensor<string, []> var_17334_equation_0 = const()[name = tensor<string, []>("op_17334_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17334_cast = einsum(equation = var_17334_equation_0, values = (var_17182_cast, var_17299_cast))[name = tensor<string, []>("op_17334_cast")];
            tensor<string, []> var_17336_equation_0 = const()[name = tensor<string, []>("op_17336_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17336_cast = einsum(equation = var_17336_equation_0, values = (var_17186_cast, var_17300_cast))[name = tensor<string, []>("op_17336_cast")];
            tensor<string, []> var_17338_equation_0 = const()[name = tensor<string, []>("op_17338_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17338_cast = einsum(equation = var_17338_equation_0, values = (var_17190_cast, var_17301_cast))[name = tensor<string, []>("op_17338_cast")];
            tensor<string, []> var_17340_equation_0 = const()[name = tensor<string, []>("op_17340_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17340_cast = einsum(equation = var_17340_equation_0, values = (var_17194_cast, var_17302_cast))[name = tensor<string, []>("op_17340_cast")];
            tensor<string, []> var_17342_equation_0 = const()[name = tensor<string, []>("op_17342_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17342_cast = einsum(equation = var_17342_equation_0, values = (var_17198_cast, var_17303_cast))[name = tensor<string, []>("op_17342_cast")];
            tensor<string, []> var_17344_equation_0 = const()[name = tensor<string, []>("op_17344_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17344_cast = einsum(equation = var_17344_equation_0, values = (var_17202_cast, var_17304_cast))[name = tensor<string, []>("op_17344_cast")];
            tensor<bool, []> input_275_interleave_0 = const()[name = tensor<string, []>("input_275_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_275_cast = concat(axis = var_2634, interleave = input_275_interleave_0, values = (var_17306_cast, var_17308_cast, var_17310_cast, var_17312_cast, var_17314_cast, var_17316_cast, var_17318_cast, var_17320_cast, var_17322_cast, var_17324_cast, var_17326_cast, var_17328_cast, var_17330_cast, var_17332_cast, var_17334_cast, var_17336_cast, var_17338_cast, var_17340_cast, var_17342_cast, var_17344_cast))[name = tensor<string, []>("input_275_cast")];
            tensor<int32, [2]> var_17350 = const()[name = tensor<string, []>("op_17350"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_17352 = const()[name = tensor<string, []>("op_17352"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_17354_pad_type_0 = const()[name = tensor<string, []>("op_17354_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_17354_pad_0 = const()[name = tensor<string, []>("op_17354_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1336585600)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1339862464)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_17354_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_out_0_bias_to_fp16, dilations = var_17352, groups = var_2634, pad = var_17354_pad_0, pad_type = var_17354_pad_type_0, strides = var_17350, weight = down_blocks_2_attentions_1_transformer_blocks_5_attn2_to_out_0_weight_to_fp16, x = input_275_cast)[name = tensor<string, []>("op_17354_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_119_cast = add(x = var_17354_cast, y = inputs_117_cast)[name = tensor<string, []>("inputs_119_cast")];
            tensor<int32, [1]> var_17358 = const()[name = tensor<string, []>("op_17358"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_119_cast = reduce_mean(axes = var_17358, keep_dims = var_2629, x = inputs_119_cast)[name = tensor<string, []>("channels_mean_119_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_119_cast = sub(x = inputs_119_cast, y = channels_mean_119_cast)[name = tensor<string, []>("zero_mean_119_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_119_cast = mul(x = zero_mean_119_cast, y = zero_mean_119_cast)[name = tensor<string, []>("zero_mean_sq_119_cast")];
            tensor<int32, [1]> var_17362 = const()[name = tensor<string, []>("op_17362"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_17363_cast = reduce_mean(axes = var_17362, keep_dims = var_2629, x = zero_mean_sq_119_cast)[name = tensor<string, []>("op_17363_cast")];
            tensor<fp16, []> var_17364_to_fp16 = const()[name = tensor<string, []>("op_17364_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_17365_cast = add(x = var_17363_cast, y = var_17364_to_fp16)[name = tensor<string, []>("op_17365_cast")];
            tensor<fp16, []> denom_119_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_119_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_119_cast = rsqrt(epsilon = denom_119_epsilon_0_to_fp16, x = var_17365_cast)[name = tensor<string, []>("denom_119_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_119_cast = mul(x = zero_mean_119_cast, y = denom_119_cast)[name = tensor<string, []>("out_119_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_17369_to_fp16 = const()[name = tensor<string, []>("op_17369_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1339865088)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_17370_cast = add(x = out_119_cast, y = var_17369_to_fp16)[name = tensor<string, []>("op_17370_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_17372_to_fp16 = const()[name = tensor<string, []>("op_17372_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1339867712)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_277_cast = mul(x = var_17370_cast, y = var_17372_to_fp16)[name = tensor<string, []>("input_277_cast")];
            tensor<int32, [2]> var_17380 = const()[name = tensor<string, []>("op_17380"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_17382 = const()[name = tensor<string, []>("op_17382"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_17384_pad_type_0 = const()[name = tensor<string, []>("op_17384_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_17384_pad_0 = const()[name = tensor<string, []>("op_17384_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_5_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_5_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1339870336)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_1_transformer_blocks_5_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_5_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1366084800)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_17384_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_5_ff_net_0_proj_bias_to_fp16, dilations = var_17382, groups = var_2634, pad = var_17384_pad_0, pad_type = var_17384_pad_type_0, strides = var_17380, weight = down_blocks_2_attentions_1_transformer_blocks_5_ff_net_0_proj_weight_to_fp16, x = input_277_cast)[name = tensor<string, []>("op_17384_cast")];
            tensor<int32, [2]> var_17385_split_sizes_0 = const()[name = tensor<string, []>("op_17385_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_17385_axis_0 = const()[name = tensor<string, []>("op_17385_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_17385_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_17385_cast_1 = split(axis = var_17385_axis_0, split_sizes = var_17385_split_sizes_0, x = var_17384_cast)[name = tensor<string, []>("op_17385_cast")];
            tensor<string, []> var_17387_mode_0 = const()[name = tensor<string, []>("op_17387_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_17387_cast = gelu(mode = var_17387_mode_0, x = var_17385_cast_1)[name = tensor<string, []>("op_17387_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_279_cast = mul(x = var_17385_cast_0, y = var_17387_cast)[name = tensor<string, []>("input_279_cast")];
            tensor<int32, [2]> var_17391 = const()[name = tensor<string, []>("op_17391"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_17393 = const()[name = tensor<string, []>("op_17393"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_17395_pad_type_0 = const()[name = tensor<string, []>("op_17395_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_17395_pad_0 = const()[name = tensor<string, []>("op_17395_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_5_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_5_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1366105344)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_5_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_5_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1379212608)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_17395_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_5_ff_net_2_bias_to_fp16, dilations = var_17393, groups = var_2634, pad = var_17395_pad_0, pad_type = var_17395_pad_type_0, strides = var_17391, weight = down_blocks_2_attentions_1_transformer_blocks_5_ff_net_2_weight_to_fp16, x = input_279_cast)[name = tensor<string, []>("op_17395_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_121_cast = add(x = var_17395_cast, y = inputs_119_cast)[name = tensor<string, []>("inputs_121_cast")];
            tensor<int32, [1]> var_17405 = const()[name = tensor<string, []>("op_17405"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_121_cast = reduce_mean(axes = var_17405, keep_dims = var_2629, x = inputs_121_cast)[name = tensor<string, []>("channels_mean_121_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_121_cast = sub(x = inputs_121_cast, y = channels_mean_121_cast)[name = tensor<string, []>("zero_mean_121_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_121_cast = mul(x = zero_mean_121_cast, y = zero_mean_121_cast)[name = tensor<string, []>("zero_mean_sq_121_cast")];
            tensor<int32, [1]> var_17409 = const()[name = tensor<string, []>("op_17409"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_17410_cast = reduce_mean(axes = var_17409, keep_dims = var_2629, x = zero_mean_sq_121_cast)[name = tensor<string, []>("op_17410_cast")];
            tensor<fp16, []> var_17411_to_fp16 = const()[name = tensor<string, []>("op_17411_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_17412_cast = add(x = var_17410_cast, y = var_17411_to_fp16)[name = tensor<string, []>("op_17412_cast")];
            tensor<fp16, []> denom_121_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_121_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_121_cast = rsqrt(epsilon = denom_121_epsilon_0_to_fp16, x = var_17412_cast)[name = tensor<string, []>("denom_121_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_121_cast = mul(x = zero_mean_121_cast, y = denom_121_cast)[name = tensor<string, []>("out_121_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_17416_to_fp16 = const()[name = tensor<string, []>("op_17416_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1379215232)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_17417_cast = add(x = out_121_cast, y = var_17416_to_fp16)[name = tensor<string, []>("op_17417_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_17419_to_fp16 = const()[name = tensor<string, []>("op_17419_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1379217856)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_173_cast = mul(x = var_17417_cast, y = var_17419_to_fp16)[name = tensor<string, []>("hidden_states_173_cast")];
            tensor<int32, [2]> var_17426 = const()[name = tensor<string, []>("op_17426"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_17428 = const()[name = tensor<string, []>("op_17428"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_81_pad_type_0 = const()[name = tensor<string, []>("q_81_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_81_pad_0 = const()[name = tensor<string, []>("q_81_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1379220480)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_81_cast = conv(dilations = var_17428, groups = var_2634, pad = q_81_pad_0, pad_type = q_81_pad_type_0, strides = var_17426, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_q_weight_to_fp16, x = hidden_states_173_cast)[name = tensor<string, []>("q_81_cast")];
            tensor<int32, [2]> var_17432 = const()[name = tensor<string, []>("op_17432"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_17434 = const()[name = tensor<string, []>("op_17434"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_161_pad_type_0 = const()[name = tensor<string, []>("k_161_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_161_pad_0 = const()[name = tensor<string, []>("k_161_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1382497344)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_161_cast = conv(dilations = var_17434, groups = var_2634, pad = k_161_pad_0, pad_type = k_161_pad_type_0, strides = var_17432, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_k_weight_to_fp16, x = hidden_states_173_cast)[name = tensor<string, []>("k_161_cast")];
            tensor<int32, [2]> var_17438 = const()[name = tensor<string, []>("op_17438"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_17440 = const()[name = tensor<string, []>("op_17440"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_81_pad_type_0 = const()[name = tensor<string, []>("v_81_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_81_pad_0 = const()[name = tensor<string, []>("v_81_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1385774208)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_81_cast = conv(dilations = var_17440, groups = var_2634, pad = v_81_pad_0, pad_type = v_81_pad_type_0, strides = var_17438, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_v_weight_to_fp16, x = hidden_states_173_cast)[name = tensor<string, []>("v_81_cast")];
            tensor<int32, [4]> var_17444_begin_0 = const()[name = tensor<string, []>("op_17444_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_17444_end_0 = const()[name = tensor<string, []>("op_17444_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_17444_end_mask_0 = const()[name = tensor<string, []>("op_17444_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17444_cast = slice_by_index(begin = var_17444_begin_0, end = var_17444_end_0, end_mask = var_17444_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17444_cast")];
            tensor<int32, [4]> var_17448_begin_0 = const()[name = tensor<string, []>("op_17448_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_17448_end_0 = const()[name = tensor<string, []>("op_17448_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_17448_end_mask_0 = const()[name = tensor<string, []>("op_17448_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17448_cast = slice_by_index(begin = var_17448_begin_0, end = var_17448_end_0, end_mask = var_17448_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17448_cast")];
            tensor<int32, [4]> var_17452_begin_0 = const()[name = tensor<string, []>("op_17452_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_17452_end_0 = const()[name = tensor<string, []>("op_17452_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_17452_end_mask_0 = const()[name = tensor<string, []>("op_17452_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17452_cast = slice_by_index(begin = var_17452_begin_0, end = var_17452_end_0, end_mask = var_17452_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17452_cast")];
            tensor<int32, [4]> var_17456_begin_0 = const()[name = tensor<string, []>("op_17456_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_17456_end_0 = const()[name = tensor<string, []>("op_17456_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_17456_end_mask_0 = const()[name = tensor<string, []>("op_17456_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17456_cast = slice_by_index(begin = var_17456_begin_0, end = var_17456_end_0, end_mask = var_17456_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17456_cast")];
            tensor<int32, [4]> var_17460_begin_0 = const()[name = tensor<string, []>("op_17460_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_17460_end_0 = const()[name = tensor<string, []>("op_17460_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_17460_end_mask_0 = const()[name = tensor<string, []>("op_17460_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17460_cast = slice_by_index(begin = var_17460_begin_0, end = var_17460_end_0, end_mask = var_17460_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17460_cast")];
            tensor<int32, [4]> var_17464_begin_0 = const()[name = tensor<string, []>("op_17464_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_17464_end_0 = const()[name = tensor<string, []>("op_17464_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_17464_end_mask_0 = const()[name = tensor<string, []>("op_17464_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17464_cast = slice_by_index(begin = var_17464_begin_0, end = var_17464_end_0, end_mask = var_17464_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17464_cast")];
            tensor<int32, [4]> var_17468_begin_0 = const()[name = tensor<string, []>("op_17468_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_17468_end_0 = const()[name = tensor<string, []>("op_17468_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_17468_end_mask_0 = const()[name = tensor<string, []>("op_17468_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17468_cast = slice_by_index(begin = var_17468_begin_0, end = var_17468_end_0, end_mask = var_17468_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17468_cast")];
            tensor<int32, [4]> var_17472_begin_0 = const()[name = tensor<string, []>("op_17472_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_17472_end_0 = const()[name = tensor<string, []>("op_17472_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_17472_end_mask_0 = const()[name = tensor<string, []>("op_17472_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17472_cast = slice_by_index(begin = var_17472_begin_0, end = var_17472_end_0, end_mask = var_17472_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17472_cast")];
            tensor<int32, [4]> var_17476_begin_0 = const()[name = tensor<string, []>("op_17476_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_17476_end_0 = const()[name = tensor<string, []>("op_17476_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_17476_end_mask_0 = const()[name = tensor<string, []>("op_17476_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17476_cast = slice_by_index(begin = var_17476_begin_0, end = var_17476_end_0, end_mask = var_17476_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17476_cast")];
            tensor<int32, [4]> var_17480_begin_0 = const()[name = tensor<string, []>("op_17480_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_17480_end_0 = const()[name = tensor<string, []>("op_17480_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_17480_end_mask_0 = const()[name = tensor<string, []>("op_17480_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17480_cast = slice_by_index(begin = var_17480_begin_0, end = var_17480_end_0, end_mask = var_17480_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17480_cast")];
            tensor<int32, [4]> var_17484_begin_0 = const()[name = tensor<string, []>("op_17484_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_17484_end_0 = const()[name = tensor<string, []>("op_17484_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_17484_end_mask_0 = const()[name = tensor<string, []>("op_17484_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17484_cast = slice_by_index(begin = var_17484_begin_0, end = var_17484_end_0, end_mask = var_17484_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17484_cast")];
            tensor<int32, [4]> var_17488_begin_0 = const()[name = tensor<string, []>("op_17488_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_17488_end_0 = const()[name = tensor<string, []>("op_17488_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_17488_end_mask_0 = const()[name = tensor<string, []>("op_17488_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17488_cast = slice_by_index(begin = var_17488_begin_0, end = var_17488_end_0, end_mask = var_17488_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17488_cast")];
            tensor<int32, [4]> var_17492_begin_0 = const()[name = tensor<string, []>("op_17492_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_17492_end_0 = const()[name = tensor<string, []>("op_17492_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_17492_end_mask_0 = const()[name = tensor<string, []>("op_17492_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17492_cast = slice_by_index(begin = var_17492_begin_0, end = var_17492_end_0, end_mask = var_17492_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17492_cast")];
            tensor<int32, [4]> var_17496_begin_0 = const()[name = tensor<string, []>("op_17496_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_17496_end_0 = const()[name = tensor<string, []>("op_17496_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_17496_end_mask_0 = const()[name = tensor<string, []>("op_17496_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17496_cast = slice_by_index(begin = var_17496_begin_0, end = var_17496_end_0, end_mask = var_17496_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17496_cast")];
            tensor<int32, [4]> var_17500_begin_0 = const()[name = tensor<string, []>("op_17500_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_17500_end_0 = const()[name = tensor<string, []>("op_17500_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_17500_end_mask_0 = const()[name = tensor<string, []>("op_17500_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17500_cast = slice_by_index(begin = var_17500_begin_0, end = var_17500_end_0, end_mask = var_17500_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17500_cast")];
            tensor<int32, [4]> var_17504_begin_0 = const()[name = tensor<string, []>("op_17504_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_17504_end_0 = const()[name = tensor<string, []>("op_17504_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_17504_end_mask_0 = const()[name = tensor<string, []>("op_17504_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17504_cast = slice_by_index(begin = var_17504_begin_0, end = var_17504_end_0, end_mask = var_17504_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17504_cast")];
            tensor<int32, [4]> var_17508_begin_0 = const()[name = tensor<string, []>("op_17508_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_17508_end_0 = const()[name = tensor<string, []>("op_17508_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_17508_end_mask_0 = const()[name = tensor<string, []>("op_17508_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17508_cast = slice_by_index(begin = var_17508_begin_0, end = var_17508_end_0, end_mask = var_17508_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17508_cast")];
            tensor<int32, [4]> var_17512_begin_0 = const()[name = tensor<string, []>("op_17512_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_17512_end_0 = const()[name = tensor<string, []>("op_17512_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_17512_end_mask_0 = const()[name = tensor<string, []>("op_17512_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17512_cast = slice_by_index(begin = var_17512_begin_0, end = var_17512_end_0, end_mask = var_17512_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17512_cast")];
            tensor<int32, [4]> var_17516_begin_0 = const()[name = tensor<string, []>("op_17516_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_17516_end_0 = const()[name = tensor<string, []>("op_17516_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_17516_end_mask_0 = const()[name = tensor<string, []>("op_17516_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17516_cast = slice_by_index(begin = var_17516_begin_0, end = var_17516_end_0, end_mask = var_17516_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17516_cast")];
            tensor<int32, [4]> var_17520_begin_0 = const()[name = tensor<string, []>("op_17520_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_17520_end_0 = const()[name = tensor<string, []>("op_17520_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_17520_end_mask_0 = const()[name = tensor<string, []>("op_17520_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17520_cast = slice_by_index(begin = var_17520_begin_0, end = var_17520_end_0, end_mask = var_17520_end_mask_0, x = q_81_cast)[name = tensor<string, []>("op_17520_cast")];
            tensor<int32, [4]> k_163_perm_0 = const()[name = tensor<string, []>("k_163_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_17527_begin_0 = const()[name = tensor<string, []>("op_17527_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_17527_end_0 = const()[name = tensor<string, []>("op_17527_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_17527_end_mask_0 = const()[name = tensor<string, []>("op_17527_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_99 = transpose(perm = k_163_perm_0, x = k_161_cast)[name = tensor<string, []>("transpose_99")];
            tensor<fp16, [2, 1024, 1, 64]> var_17527_cast = slice_by_index(begin = var_17527_begin_0, end = var_17527_end_0, end_mask = var_17527_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17527_cast")];
            tensor<int32, [4]> var_17531_begin_0 = const()[name = tensor<string, []>("op_17531_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_17531_end_0 = const()[name = tensor<string, []>("op_17531_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_17531_end_mask_0 = const()[name = tensor<string, []>("op_17531_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_17531_cast = slice_by_index(begin = var_17531_begin_0, end = var_17531_end_0, end_mask = var_17531_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17531_cast")];
            tensor<int32, [4]> var_17535_begin_0 = const()[name = tensor<string, []>("op_17535_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_17535_end_0 = const()[name = tensor<string, []>("op_17535_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_17535_end_mask_0 = const()[name = tensor<string, []>("op_17535_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_17535_cast = slice_by_index(begin = var_17535_begin_0, end = var_17535_end_0, end_mask = var_17535_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17535_cast")];
            tensor<int32, [4]> var_17539_begin_0 = const()[name = tensor<string, []>("op_17539_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_17539_end_0 = const()[name = tensor<string, []>("op_17539_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_17539_end_mask_0 = const()[name = tensor<string, []>("op_17539_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_17539_cast = slice_by_index(begin = var_17539_begin_0, end = var_17539_end_0, end_mask = var_17539_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17539_cast")];
            tensor<int32, [4]> var_17543_begin_0 = const()[name = tensor<string, []>("op_17543_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_17543_end_0 = const()[name = tensor<string, []>("op_17543_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_17543_end_mask_0 = const()[name = tensor<string, []>("op_17543_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_17543_cast = slice_by_index(begin = var_17543_begin_0, end = var_17543_end_0, end_mask = var_17543_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17543_cast")];
            tensor<int32, [4]> var_17547_begin_0 = const()[name = tensor<string, []>("op_17547_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_17547_end_0 = const()[name = tensor<string, []>("op_17547_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_17547_end_mask_0 = const()[name = tensor<string, []>("op_17547_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_17547_cast = slice_by_index(begin = var_17547_begin_0, end = var_17547_end_0, end_mask = var_17547_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17547_cast")];
            tensor<int32, [4]> var_17551_begin_0 = const()[name = tensor<string, []>("op_17551_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_17551_end_0 = const()[name = tensor<string, []>("op_17551_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_17551_end_mask_0 = const()[name = tensor<string, []>("op_17551_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_17551_cast = slice_by_index(begin = var_17551_begin_0, end = var_17551_end_0, end_mask = var_17551_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17551_cast")];
            tensor<int32, [4]> var_17555_begin_0 = const()[name = tensor<string, []>("op_17555_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_17555_end_0 = const()[name = tensor<string, []>("op_17555_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_17555_end_mask_0 = const()[name = tensor<string, []>("op_17555_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_17555_cast = slice_by_index(begin = var_17555_begin_0, end = var_17555_end_0, end_mask = var_17555_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17555_cast")];
            tensor<int32, [4]> var_17559_begin_0 = const()[name = tensor<string, []>("op_17559_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_17559_end_0 = const()[name = tensor<string, []>("op_17559_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_17559_end_mask_0 = const()[name = tensor<string, []>("op_17559_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_17559_cast = slice_by_index(begin = var_17559_begin_0, end = var_17559_end_0, end_mask = var_17559_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17559_cast")];
            tensor<int32, [4]> var_17563_begin_0 = const()[name = tensor<string, []>("op_17563_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_17563_end_0 = const()[name = tensor<string, []>("op_17563_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_17563_end_mask_0 = const()[name = tensor<string, []>("op_17563_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_17563_cast = slice_by_index(begin = var_17563_begin_0, end = var_17563_end_0, end_mask = var_17563_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17563_cast")];
            tensor<int32, [4]> var_17567_begin_0 = const()[name = tensor<string, []>("op_17567_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_17567_end_0 = const()[name = tensor<string, []>("op_17567_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_17567_end_mask_0 = const()[name = tensor<string, []>("op_17567_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_17567_cast = slice_by_index(begin = var_17567_begin_0, end = var_17567_end_0, end_mask = var_17567_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17567_cast")];
            tensor<int32, [4]> var_17571_begin_0 = const()[name = tensor<string, []>("op_17571_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_17571_end_0 = const()[name = tensor<string, []>("op_17571_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_17571_end_mask_0 = const()[name = tensor<string, []>("op_17571_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_17571_cast = slice_by_index(begin = var_17571_begin_0, end = var_17571_end_0, end_mask = var_17571_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17571_cast")];
            tensor<int32, [4]> var_17575_begin_0 = const()[name = tensor<string, []>("op_17575_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_17575_end_0 = const()[name = tensor<string, []>("op_17575_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_17575_end_mask_0 = const()[name = tensor<string, []>("op_17575_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_17575_cast = slice_by_index(begin = var_17575_begin_0, end = var_17575_end_0, end_mask = var_17575_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17575_cast")];
            tensor<int32, [4]> var_17579_begin_0 = const()[name = tensor<string, []>("op_17579_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_17579_end_0 = const()[name = tensor<string, []>("op_17579_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_17579_end_mask_0 = const()[name = tensor<string, []>("op_17579_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_17579_cast = slice_by_index(begin = var_17579_begin_0, end = var_17579_end_0, end_mask = var_17579_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17579_cast")];
            tensor<int32, [4]> var_17583_begin_0 = const()[name = tensor<string, []>("op_17583_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_17583_end_0 = const()[name = tensor<string, []>("op_17583_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_17583_end_mask_0 = const()[name = tensor<string, []>("op_17583_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_17583_cast = slice_by_index(begin = var_17583_begin_0, end = var_17583_end_0, end_mask = var_17583_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17583_cast")];
            tensor<int32, [4]> var_17587_begin_0 = const()[name = tensor<string, []>("op_17587_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_17587_end_0 = const()[name = tensor<string, []>("op_17587_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_17587_end_mask_0 = const()[name = tensor<string, []>("op_17587_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_17587_cast = slice_by_index(begin = var_17587_begin_0, end = var_17587_end_0, end_mask = var_17587_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17587_cast")];
            tensor<int32, [4]> var_17591_begin_0 = const()[name = tensor<string, []>("op_17591_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_17591_end_0 = const()[name = tensor<string, []>("op_17591_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_17591_end_mask_0 = const()[name = tensor<string, []>("op_17591_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_17591_cast = slice_by_index(begin = var_17591_begin_0, end = var_17591_end_0, end_mask = var_17591_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17591_cast")];
            tensor<int32, [4]> var_17595_begin_0 = const()[name = tensor<string, []>("op_17595_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_17595_end_0 = const()[name = tensor<string, []>("op_17595_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_17595_end_mask_0 = const()[name = tensor<string, []>("op_17595_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_17595_cast = slice_by_index(begin = var_17595_begin_0, end = var_17595_end_0, end_mask = var_17595_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17595_cast")];
            tensor<int32, [4]> var_17599_begin_0 = const()[name = tensor<string, []>("op_17599_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_17599_end_0 = const()[name = tensor<string, []>("op_17599_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_17599_end_mask_0 = const()[name = tensor<string, []>("op_17599_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_17599_cast = slice_by_index(begin = var_17599_begin_0, end = var_17599_end_0, end_mask = var_17599_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17599_cast")];
            tensor<int32, [4]> var_17603_begin_0 = const()[name = tensor<string, []>("op_17603_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_17603_end_0 = const()[name = tensor<string, []>("op_17603_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_17603_end_mask_0 = const()[name = tensor<string, []>("op_17603_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_17603_cast = slice_by_index(begin = var_17603_begin_0, end = var_17603_end_0, end_mask = var_17603_end_mask_0, x = transpose_99)[name = tensor<string, []>("op_17603_cast")];
            tensor<int32, [4]> var_17605_begin_0 = const()[name = tensor<string, []>("op_17605_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_17605_end_0 = const()[name = tensor<string, []>("op_17605_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_17605_end_mask_0 = const()[name = tensor<string, []>("op_17605_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17605_cast = slice_by_index(begin = var_17605_begin_0, end = var_17605_end_0, end_mask = var_17605_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17605_cast")];
            tensor<int32, [4]> var_17609_begin_0 = const()[name = tensor<string, []>("op_17609_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_17609_end_0 = const()[name = tensor<string, []>("op_17609_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_17609_end_mask_0 = const()[name = tensor<string, []>("op_17609_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17609_cast = slice_by_index(begin = var_17609_begin_0, end = var_17609_end_0, end_mask = var_17609_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17609_cast")];
            tensor<int32, [4]> var_17613_begin_0 = const()[name = tensor<string, []>("op_17613_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_17613_end_0 = const()[name = tensor<string, []>("op_17613_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_17613_end_mask_0 = const()[name = tensor<string, []>("op_17613_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17613_cast = slice_by_index(begin = var_17613_begin_0, end = var_17613_end_0, end_mask = var_17613_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17613_cast")];
            tensor<int32, [4]> var_17617_begin_0 = const()[name = tensor<string, []>("op_17617_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_17617_end_0 = const()[name = tensor<string, []>("op_17617_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_17617_end_mask_0 = const()[name = tensor<string, []>("op_17617_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17617_cast = slice_by_index(begin = var_17617_begin_0, end = var_17617_end_0, end_mask = var_17617_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17617_cast")];
            tensor<int32, [4]> var_17621_begin_0 = const()[name = tensor<string, []>("op_17621_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_17621_end_0 = const()[name = tensor<string, []>("op_17621_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_17621_end_mask_0 = const()[name = tensor<string, []>("op_17621_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17621_cast = slice_by_index(begin = var_17621_begin_0, end = var_17621_end_0, end_mask = var_17621_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17621_cast")];
            tensor<int32, [4]> var_17625_begin_0 = const()[name = tensor<string, []>("op_17625_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_17625_end_0 = const()[name = tensor<string, []>("op_17625_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_17625_end_mask_0 = const()[name = tensor<string, []>("op_17625_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17625_cast = slice_by_index(begin = var_17625_begin_0, end = var_17625_end_0, end_mask = var_17625_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17625_cast")];
            tensor<int32, [4]> var_17629_begin_0 = const()[name = tensor<string, []>("op_17629_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_17629_end_0 = const()[name = tensor<string, []>("op_17629_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_17629_end_mask_0 = const()[name = tensor<string, []>("op_17629_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17629_cast = slice_by_index(begin = var_17629_begin_0, end = var_17629_end_0, end_mask = var_17629_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17629_cast")];
            tensor<int32, [4]> var_17633_begin_0 = const()[name = tensor<string, []>("op_17633_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_17633_end_0 = const()[name = tensor<string, []>("op_17633_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_17633_end_mask_0 = const()[name = tensor<string, []>("op_17633_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17633_cast = slice_by_index(begin = var_17633_begin_0, end = var_17633_end_0, end_mask = var_17633_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17633_cast")];
            tensor<int32, [4]> var_17637_begin_0 = const()[name = tensor<string, []>("op_17637_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_17637_end_0 = const()[name = tensor<string, []>("op_17637_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_17637_end_mask_0 = const()[name = tensor<string, []>("op_17637_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17637_cast = slice_by_index(begin = var_17637_begin_0, end = var_17637_end_0, end_mask = var_17637_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17637_cast")];
            tensor<int32, [4]> var_17641_begin_0 = const()[name = tensor<string, []>("op_17641_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_17641_end_0 = const()[name = tensor<string, []>("op_17641_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_17641_end_mask_0 = const()[name = tensor<string, []>("op_17641_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17641_cast = slice_by_index(begin = var_17641_begin_0, end = var_17641_end_0, end_mask = var_17641_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17641_cast")];
            tensor<int32, [4]> var_17645_begin_0 = const()[name = tensor<string, []>("op_17645_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_17645_end_0 = const()[name = tensor<string, []>("op_17645_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_17645_end_mask_0 = const()[name = tensor<string, []>("op_17645_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17645_cast = slice_by_index(begin = var_17645_begin_0, end = var_17645_end_0, end_mask = var_17645_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17645_cast")];
            tensor<int32, [4]> var_17649_begin_0 = const()[name = tensor<string, []>("op_17649_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_17649_end_0 = const()[name = tensor<string, []>("op_17649_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_17649_end_mask_0 = const()[name = tensor<string, []>("op_17649_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17649_cast = slice_by_index(begin = var_17649_begin_0, end = var_17649_end_0, end_mask = var_17649_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17649_cast")];
            tensor<int32, [4]> var_17653_begin_0 = const()[name = tensor<string, []>("op_17653_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_17653_end_0 = const()[name = tensor<string, []>("op_17653_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_17653_end_mask_0 = const()[name = tensor<string, []>("op_17653_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17653_cast = slice_by_index(begin = var_17653_begin_0, end = var_17653_end_0, end_mask = var_17653_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17653_cast")];
            tensor<int32, [4]> var_17657_begin_0 = const()[name = tensor<string, []>("op_17657_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_17657_end_0 = const()[name = tensor<string, []>("op_17657_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_17657_end_mask_0 = const()[name = tensor<string, []>("op_17657_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17657_cast = slice_by_index(begin = var_17657_begin_0, end = var_17657_end_0, end_mask = var_17657_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17657_cast")];
            tensor<int32, [4]> var_17661_begin_0 = const()[name = tensor<string, []>("op_17661_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_17661_end_0 = const()[name = tensor<string, []>("op_17661_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_17661_end_mask_0 = const()[name = tensor<string, []>("op_17661_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17661_cast = slice_by_index(begin = var_17661_begin_0, end = var_17661_end_0, end_mask = var_17661_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17661_cast")];
            tensor<int32, [4]> var_17665_begin_0 = const()[name = tensor<string, []>("op_17665_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_17665_end_0 = const()[name = tensor<string, []>("op_17665_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_17665_end_mask_0 = const()[name = tensor<string, []>("op_17665_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17665_cast = slice_by_index(begin = var_17665_begin_0, end = var_17665_end_0, end_mask = var_17665_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17665_cast")];
            tensor<int32, [4]> var_17669_begin_0 = const()[name = tensor<string, []>("op_17669_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_17669_end_0 = const()[name = tensor<string, []>("op_17669_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_17669_end_mask_0 = const()[name = tensor<string, []>("op_17669_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17669_cast = slice_by_index(begin = var_17669_begin_0, end = var_17669_end_0, end_mask = var_17669_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17669_cast")];
            tensor<int32, [4]> var_17673_begin_0 = const()[name = tensor<string, []>("op_17673_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_17673_end_0 = const()[name = tensor<string, []>("op_17673_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_17673_end_mask_0 = const()[name = tensor<string, []>("op_17673_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17673_cast = slice_by_index(begin = var_17673_begin_0, end = var_17673_end_0, end_mask = var_17673_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17673_cast")];
            tensor<int32, [4]> var_17677_begin_0 = const()[name = tensor<string, []>("op_17677_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_17677_end_0 = const()[name = tensor<string, []>("op_17677_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_17677_end_mask_0 = const()[name = tensor<string, []>("op_17677_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17677_cast = slice_by_index(begin = var_17677_begin_0, end = var_17677_end_0, end_mask = var_17677_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17677_cast")];
            tensor<int32, [4]> var_17681_begin_0 = const()[name = tensor<string, []>("op_17681_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_17681_end_0 = const()[name = tensor<string, []>("op_17681_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_17681_end_mask_0 = const()[name = tensor<string, []>("op_17681_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17681_cast = slice_by_index(begin = var_17681_begin_0, end = var_17681_end_0, end_mask = var_17681_end_mask_0, x = v_81_cast)[name = tensor<string, []>("op_17681_cast")];
            tensor<string, []> var_17685_equation_0 = const()[name = tensor<string, []>("op_17685_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17685_cast = einsum(equation = var_17685_equation_0, values = (var_17527_cast, var_17444_cast))[name = tensor<string, []>("op_17685_cast")];
            tensor<fp16, []> var_17686_to_fp16 = const()[name = tensor<string, []>("op_17686_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1441_cast = mul(x = var_17685_cast, y = var_17686_to_fp16)[name = tensor<string, []>("aw_1441_cast")];
            tensor<string, []> var_17689_equation_0 = const()[name = tensor<string, []>("op_17689_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17689_cast = einsum(equation = var_17689_equation_0, values = (var_17531_cast, var_17448_cast))[name = tensor<string, []>("op_17689_cast")];
            tensor<fp16, []> var_17690_to_fp16 = const()[name = tensor<string, []>("op_17690_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1443_cast = mul(x = var_17689_cast, y = var_17690_to_fp16)[name = tensor<string, []>("aw_1443_cast")];
            tensor<string, []> var_17693_equation_0 = const()[name = tensor<string, []>("op_17693_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17693_cast = einsum(equation = var_17693_equation_0, values = (var_17535_cast, var_17452_cast))[name = tensor<string, []>("op_17693_cast")];
            tensor<fp16, []> var_17694_to_fp16 = const()[name = tensor<string, []>("op_17694_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1445_cast = mul(x = var_17693_cast, y = var_17694_to_fp16)[name = tensor<string, []>("aw_1445_cast")];
            tensor<string, []> var_17697_equation_0 = const()[name = tensor<string, []>("op_17697_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17697_cast = einsum(equation = var_17697_equation_0, values = (var_17539_cast, var_17456_cast))[name = tensor<string, []>("op_17697_cast")];
            tensor<fp16, []> var_17698_to_fp16 = const()[name = tensor<string, []>("op_17698_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1447_cast = mul(x = var_17697_cast, y = var_17698_to_fp16)[name = tensor<string, []>("aw_1447_cast")];
            tensor<string, []> var_17701_equation_0 = const()[name = tensor<string, []>("op_17701_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17701_cast = einsum(equation = var_17701_equation_0, values = (var_17543_cast, var_17460_cast))[name = tensor<string, []>("op_17701_cast")];
            tensor<fp16, []> var_17702_to_fp16 = const()[name = tensor<string, []>("op_17702_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1449_cast = mul(x = var_17701_cast, y = var_17702_to_fp16)[name = tensor<string, []>("aw_1449_cast")];
            tensor<string, []> var_17705_equation_0 = const()[name = tensor<string, []>("op_17705_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17705_cast = einsum(equation = var_17705_equation_0, values = (var_17547_cast, var_17464_cast))[name = tensor<string, []>("op_17705_cast")];
            tensor<fp16, []> var_17706_to_fp16 = const()[name = tensor<string, []>("op_17706_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1451_cast = mul(x = var_17705_cast, y = var_17706_to_fp16)[name = tensor<string, []>("aw_1451_cast")];
            tensor<string, []> var_17709_equation_0 = const()[name = tensor<string, []>("op_17709_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17709_cast = einsum(equation = var_17709_equation_0, values = (var_17551_cast, var_17468_cast))[name = tensor<string, []>("op_17709_cast")];
            tensor<fp16, []> var_17710_to_fp16 = const()[name = tensor<string, []>("op_17710_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1453_cast = mul(x = var_17709_cast, y = var_17710_to_fp16)[name = tensor<string, []>("aw_1453_cast")];
            tensor<string, []> var_17713_equation_0 = const()[name = tensor<string, []>("op_17713_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17713_cast = einsum(equation = var_17713_equation_0, values = (var_17555_cast, var_17472_cast))[name = tensor<string, []>("op_17713_cast")];
            tensor<fp16, []> var_17714_to_fp16 = const()[name = tensor<string, []>("op_17714_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1455_cast = mul(x = var_17713_cast, y = var_17714_to_fp16)[name = tensor<string, []>("aw_1455_cast")];
            tensor<string, []> var_17717_equation_0 = const()[name = tensor<string, []>("op_17717_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17717_cast = einsum(equation = var_17717_equation_0, values = (var_17559_cast, var_17476_cast))[name = tensor<string, []>("op_17717_cast")];
            tensor<fp16, []> var_17718_to_fp16 = const()[name = tensor<string, []>("op_17718_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1457_cast = mul(x = var_17717_cast, y = var_17718_to_fp16)[name = tensor<string, []>("aw_1457_cast")];
            tensor<string, []> var_17721_equation_0 = const()[name = tensor<string, []>("op_17721_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17721_cast = einsum(equation = var_17721_equation_0, values = (var_17563_cast, var_17480_cast))[name = tensor<string, []>("op_17721_cast")];
            tensor<fp16, []> var_17722_to_fp16 = const()[name = tensor<string, []>("op_17722_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1459_cast = mul(x = var_17721_cast, y = var_17722_to_fp16)[name = tensor<string, []>("aw_1459_cast")];
            tensor<string, []> var_17725_equation_0 = const()[name = tensor<string, []>("op_17725_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17725_cast = einsum(equation = var_17725_equation_0, values = (var_17567_cast, var_17484_cast))[name = tensor<string, []>("op_17725_cast")];
            tensor<fp16, []> var_17726_to_fp16 = const()[name = tensor<string, []>("op_17726_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1461_cast = mul(x = var_17725_cast, y = var_17726_to_fp16)[name = tensor<string, []>("aw_1461_cast")];
            tensor<string, []> var_17729_equation_0 = const()[name = tensor<string, []>("op_17729_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17729_cast = einsum(equation = var_17729_equation_0, values = (var_17571_cast, var_17488_cast))[name = tensor<string, []>("op_17729_cast")];
            tensor<fp16, []> var_17730_to_fp16 = const()[name = tensor<string, []>("op_17730_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1463_cast = mul(x = var_17729_cast, y = var_17730_to_fp16)[name = tensor<string, []>("aw_1463_cast")];
            tensor<string, []> var_17733_equation_0 = const()[name = tensor<string, []>("op_17733_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17733_cast = einsum(equation = var_17733_equation_0, values = (var_17575_cast, var_17492_cast))[name = tensor<string, []>("op_17733_cast")];
            tensor<fp16, []> var_17734_to_fp16 = const()[name = tensor<string, []>("op_17734_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1465_cast = mul(x = var_17733_cast, y = var_17734_to_fp16)[name = tensor<string, []>("aw_1465_cast")];
            tensor<string, []> var_17737_equation_0 = const()[name = tensor<string, []>("op_17737_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17737_cast = einsum(equation = var_17737_equation_0, values = (var_17579_cast, var_17496_cast))[name = tensor<string, []>("op_17737_cast")];
            tensor<fp16, []> var_17738_to_fp16 = const()[name = tensor<string, []>("op_17738_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1467_cast = mul(x = var_17737_cast, y = var_17738_to_fp16)[name = tensor<string, []>("aw_1467_cast")];
            tensor<string, []> var_17741_equation_0 = const()[name = tensor<string, []>("op_17741_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17741_cast = einsum(equation = var_17741_equation_0, values = (var_17583_cast, var_17500_cast))[name = tensor<string, []>("op_17741_cast")];
            tensor<fp16, []> var_17742_to_fp16 = const()[name = tensor<string, []>("op_17742_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1469_cast = mul(x = var_17741_cast, y = var_17742_to_fp16)[name = tensor<string, []>("aw_1469_cast")];
            tensor<string, []> var_17745_equation_0 = const()[name = tensor<string, []>("op_17745_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17745_cast = einsum(equation = var_17745_equation_0, values = (var_17587_cast, var_17504_cast))[name = tensor<string, []>("op_17745_cast")];
            tensor<fp16, []> var_17746_to_fp16 = const()[name = tensor<string, []>("op_17746_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1471_cast = mul(x = var_17745_cast, y = var_17746_to_fp16)[name = tensor<string, []>("aw_1471_cast")];
            tensor<string, []> var_17749_equation_0 = const()[name = tensor<string, []>("op_17749_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17749_cast = einsum(equation = var_17749_equation_0, values = (var_17591_cast, var_17508_cast))[name = tensor<string, []>("op_17749_cast")];
            tensor<fp16, []> var_17750_to_fp16 = const()[name = tensor<string, []>("op_17750_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1473_cast = mul(x = var_17749_cast, y = var_17750_to_fp16)[name = tensor<string, []>("aw_1473_cast")];
            tensor<string, []> var_17753_equation_0 = const()[name = tensor<string, []>("op_17753_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17753_cast = einsum(equation = var_17753_equation_0, values = (var_17595_cast, var_17512_cast))[name = tensor<string, []>("op_17753_cast")];
            tensor<fp16, []> var_17754_to_fp16 = const()[name = tensor<string, []>("op_17754_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1475_cast = mul(x = var_17753_cast, y = var_17754_to_fp16)[name = tensor<string, []>("aw_1475_cast")];
            tensor<string, []> var_17757_equation_0 = const()[name = tensor<string, []>("op_17757_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17757_cast = einsum(equation = var_17757_equation_0, values = (var_17599_cast, var_17516_cast))[name = tensor<string, []>("op_17757_cast")];
            tensor<fp16, []> var_17758_to_fp16 = const()[name = tensor<string, []>("op_17758_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1477_cast = mul(x = var_17757_cast, y = var_17758_to_fp16)[name = tensor<string, []>("aw_1477_cast")];
            tensor<string, []> var_17761_equation_0 = const()[name = tensor<string, []>("op_17761_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17761_cast = einsum(equation = var_17761_equation_0, values = (var_17603_cast, var_17520_cast))[name = tensor<string, []>("op_17761_cast")];
            tensor<fp16, []> var_17762_to_fp16 = const()[name = tensor<string, []>("op_17762_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1479_cast = mul(x = var_17761_cast, y = var_17762_to_fp16)[name = tensor<string, []>("aw_1479_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17764_cast = softmax(axis = var_2634, x = aw_1441_cast)[name = tensor<string, []>("op_17764_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17765_cast = softmax(axis = var_2634, x = aw_1443_cast)[name = tensor<string, []>("op_17765_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17766_cast = softmax(axis = var_2634, x = aw_1445_cast)[name = tensor<string, []>("op_17766_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17767_cast = softmax(axis = var_2634, x = aw_1447_cast)[name = tensor<string, []>("op_17767_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17768_cast = softmax(axis = var_2634, x = aw_1449_cast)[name = tensor<string, []>("op_17768_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17769_cast = softmax(axis = var_2634, x = aw_1451_cast)[name = tensor<string, []>("op_17769_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17770_cast = softmax(axis = var_2634, x = aw_1453_cast)[name = tensor<string, []>("op_17770_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17771_cast = softmax(axis = var_2634, x = aw_1455_cast)[name = tensor<string, []>("op_17771_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17772_cast = softmax(axis = var_2634, x = aw_1457_cast)[name = tensor<string, []>("op_17772_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17773_cast = softmax(axis = var_2634, x = aw_1459_cast)[name = tensor<string, []>("op_17773_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17774_cast = softmax(axis = var_2634, x = aw_1461_cast)[name = tensor<string, []>("op_17774_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17775_cast = softmax(axis = var_2634, x = aw_1463_cast)[name = tensor<string, []>("op_17775_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17776_cast = softmax(axis = var_2634, x = aw_1465_cast)[name = tensor<string, []>("op_17776_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17777_cast = softmax(axis = var_2634, x = aw_1467_cast)[name = tensor<string, []>("op_17777_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17778_cast = softmax(axis = var_2634, x = aw_1469_cast)[name = tensor<string, []>("op_17778_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17779_cast = softmax(axis = var_2634, x = aw_1471_cast)[name = tensor<string, []>("op_17779_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17780_cast = softmax(axis = var_2634, x = aw_1473_cast)[name = tensor<string, []>("op_17780_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17781_cast = softmax(axis = var_2634, x = aw_1475_cast)[name = tensor<string, []>("op_17781_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17782_cast = softmax(axis = var_2634, x = aw_1477_cast)[name = tensor<string, []>("op_17782_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_17783_cast = softmax(axis = var_2634, x = aw_1479_cast)[name = tensor<string, []>("op_17783_cast")];
            tensor<string, []> var_17785_equation_0 = const()[name = tensor<string, []>("op_17785_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17785_cast = einsum(equation = var_17785_equation_0, values = (var_17605_cast, var_17764_cast))[name = tensor<string, []>("op_17785_cast")];
            tensor<string, []> var_17787_equation_0 = const()[name = tensor<string, []>("op_17787_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17787_cast = einsum(equation = var_17787_equation_0, values = (var_17609_cast, var_17765_cast))[name = tensor<string, []>("op_17787_cast")];
            tensor<string, []> var_17789_equation_0 = const()[name = tensor<string, []>("op_17789_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17789_cast = einsum(equation = var_17789_equation_0, values = (var_17613_cast, var_17766_cast))[name = tensor<string, []>("op_17789_cast")];
            tensor<string, []> var_17791_equation_0 = const()[name = tensor<string, []>("op_17791_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17791_cast = einsum(equation = var_17791_equation_0, values = (var_17617_cast, var_17767_cast))[name = tensor<string, []>("op_17791_cast")];
            tensor<string, []> var_17793_equation_0 = const()[name = tensor<string, []>("op_17793_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17793_cast = einsum(equation = var_17793_equation_0, values = (var_17621_cast, var_17768_cast))[name = tensor<string, []>("op_17793_cast")];
            tensor<string, []> var_17795_equation_0 = const()[name = tensor<string, []>("op_17795_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17795_cast = einsum(equation = var_17795_equation_0, values = (var_17625_cast, var_17769_cast))[name = tensor<string, []>("op_17795_cast")];
            tensor<string, []> var_17797_equation_0 = const()[name = tensor<string, []>("op_17797_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17797_cast = einsum(equation = var_17797_equation_0, values = (var_17629_cast, var_17770_cast))[name = tensor<string, []>("op_17797_cast")];
            tensor<string, []> var_17799_equation_0 = const()[name = tensor<string, []>("op_17799_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17799_cast = einsum(equation = var_17799_equation_0, values = (var_17633_cast, var_17771_cast))[name = tensor<string, []>("op_17799_cast")];
            tensor<string, []> var_17801_equation_0 = const()[name = tensor<string, []>("op_17801_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17801_cast = einsum(equation = var_17801_equation_0, values = (var_17637_cast, var_17772_cast))[name = tensor<string, []>("op_17801_cast")];
            tensor<string, []> var_17803_equation_0 = const()[name = tensor<string, []>("op_17803_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17803_cast = einsum(equation = var_17803_equation_0, values = (var_17641_cast, var_17773_cast))[name = tensor<string, []>("op_17803_cast")];
            tensor<string, []> var_17805_equation_0 = const()[name = tensor<string, []>("op_17805_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17805_cast = einsum(equation = var_17805_equation_0, values = (var_17645_cast, var_17774_cast))[name = tensor<string, []>("op_17805_cast")];
            tensor<string, []> var_17807_equation_0 = const()[name = tensor<string, []>("op_17807_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17807_cast = einsum(equation = var_17807_equation_0, values = (var_17649_cast, var_17775_cast))[name = tensor<string, []>("op_17807_cast")];
            tensor<string, []> var_17809_equation_0 = const()[name = tensor<string, []>("op_17809_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17809_cast = einsum(equation = var_17809_equation_0, values = (var_17653_cast, var_17776_cast))[name = tensor<string, []>("op_17809_cast")];
            tensor<string, []> var_17811_equation_0 = const()[name = tensor<string, []>("op_17811_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17811_cast = einsum(equation = var_17811_equation_0, values = (var_17657_cast, var_17777_cast))[name = tensor<string, []>("op_17811_cast")];
            tensor<string, []> var_17813_equation_0 = const()[name = tensor<string, []>("op_17813_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17813_cast = einsum(equation = var_17813_equation_0, values = (var_17661_cast, var_17778_cast))[name = tensor<string, []>("op_17813_cast")];
            tensor<string, []> var_17815_equation_0 = const()[name = tensor<string, []>("op_17815_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17815_cast = einsum(equation = var_17815_equation_0, values = (var_17665_cast, var_17779_cast))[name = tensor<string, []>("op_17815_cast")];
            tensor<string, []> var_17817_equation_0 = const()[name = tensor<string, []>("op_17817_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17817_cast = einsum(equation = var_17817_equation_0, values = (var_17669_cast, var_17780_cast))[name = tensor<string, []>("op_17817_cast")];
            tensor<string, []> var_17819_equation_0 = const()[name = tensor<string, []>("op_17819_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17819_cast = einsum(equation = var_17819_equation_0, values = (var_17673_cast, var_17781_cast))[name = tensor<string, []>("op_17819_cast")];
            tensor<string, []> var_17821_equation_0 = const()[name = tensor<string, []>("op_17821_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17821_cast = einsum(equation = var_17821_equation_0, values = (var_17677_cast, var_17782_cast))[name = tensor<string, []>("op_17821_cast")];
            tensor<string, []> var_17823_equation_0 = const()[name = tensor<string, []>("op_17823_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_17823_cast = einsum(equation = var_17823_equation_0, values = (var_17681_cast, var_17783_cast))[name = tensor<string, []>("op_17823_cast")];
            tensor<bool, []> input_281_interleave_0 = const()[name = tensor<string, []>("input_281_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_281_cast = concat(axis = var_2634, interleave = input_281_interleave_0, values = (var_17785_cast, var_17787_cast, var_17789_cast, var_17791_cast, var_17793_cast, var_17795_cast, var_17797_cast, var_17799_cast, var_17801_cast, var_17803_cast, var_17805_cast, var_17807_cast, var_17809_cast, var_17811_cast, var_17813_cast, var_17815_cast, var_17817_cast, var_17819_cast, var_17821_cast, var_17823_cast))[name = tensor<string, []>("input_281_cast")];
            tensor<int32, [2]> var_17829 = const()[name = tensor<string, []>("op_17829"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_17831 = const()[name = tensor<string, []>("op_17831"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_17833_pad_type_0 = const()[name = tensor<string, []>("op_17833_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_17833_pad_0 = const()[name = tensor<string, []>("op_17833_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1389051072)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1392327936)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_17833_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_out_0_bias_to_fp16, dilations = var_17831, groups = var_2634, pad = var_17833_pad_0, pad_type = var_17833_pad_type_0, strides = var_17829, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn1_to_out_0_weight_to_fp16, x = input_281_cast)[name = tensor<string, []>("op_17833_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_123_cast = add(x = var_17833_cast, y = inputs_121_cast)[name = tensor<string, []>("inputs_123_cast")];
            tensor<int32, [1]> var_17837 = const()[name = tensor<string, []>("op_17837"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_123_cast = reduce_mean(axes = var_17837, keep_dims = var_2629, x = inputs_123_cast)[name = tensor<string, []>("channels_mean_123_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_123_cast = sub(x = inputs_123_cast, y = channels_mean_123_cast)[name = tensor<string, []>("zero_mean_123_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_123_cast = mul(x = zero_mean_123_cast, y = zero_mean_123_cast)[name = tensor<string, []>("zero_mean_sq_123_cast")];
            tensor<int32, [1]> var_17841 = const()[name = tensor<string, []>("op_17841"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_17842_cast = reduce_mean(axes = var_17841, keep_dims = var_2629, x = zero_mean_sq_123_cast)[name = tensor<string, []>("op_17842_cast")];
            tensor<fp16, []> var_17843_to_fp16 = const()[name = tensor<string, []>("op_17843_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_17844_cast = add(x = var_17842_cast, y = var_17843_to_fp16)[name = tensor<string, []>("op_17844_cast")];
            tensor<fp16, []> denom_123_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_123_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_123_cast = rsqrt(epsilon = denom_123_epsilon_0_to_fp16, x = var_17844_cast)[name = tensor<string, []>("denom_123_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_123_cast = mul(x = zero_mean_123_cast, y = denom_123_cast)[name = tensor<string, []>("out_123_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_17848_to_fp16 = const()[name = tensor<string, []>("op_17848_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1392330560)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_17849_cast = add(x = out_123_cast, y = var_17848_to_fp16)[name = tensor<string, []>("op_17849_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_17851_to_fp16 = const()[name = tensor<string, []>("op_17851_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1392333184)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_175_cast = mul(x = var_17849_cast, y = var_17851_to_fp16)[name = tensor<string, []>("hidden_states_175_cast")];
            tensor<int32, [2]> var_17858 = const()[name = tensor<string, []>("op_17858"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_17860 = const()[name = tensor<string, []>("op_17860"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_83_pad_type_0 = const()[name = tensor<string, []>("q_83_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_83_pad_0 = const()[name = tensor<string, []>("q_83_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1392335808)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_83_cast = conv(dilations = var_17860, groups = var_2634, pad = q_83_pad_0, pad_type = q_83_pad_type_0, strides = var_17858, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_q_weight_to_fp16, x = hidden_states_175_cast)[name = tensor<string, []>("q_83_cast")];
            tensor<int32, [2]> var_17864 = const()[name = tensor<string, []>("op_17864"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_17866 = const()[name = tensor<string, []>("op_17866"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_165_pad_type_0 = const()[name = tensor<string, []>("k_165_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_165_pad_0 = const()[name = tensor<string, []>("k_165_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1395612672)))];
            tensor<fp16, [2, 1280, 1, 77]> k_165_cast = conv(dilations = var_17866, groups = var_2634, pad = k_165_pad_0, pad_type = k_165_pad_type_0, strides = var_17864, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_165_cast")];
            tensor<int32, [2]> var_17870 = const()[name = tensor<string, []>("op_17870"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_17872 = const()[name = tensor<string, []>("op_17872"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_83_pad_type_0 = const()[name = tensor<string, []>("v_83_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_83_pad_0 = const()[name = tensor<string, []>("v_83_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1400855616)))];
            tensor<fp16, [2, 1280, 1, 77]> v_83_cast = conv(dilations = var_17872, groups = var_2634, pad = v_83_pad_0, pad_type = v_83_pad_type_0, strides = var_17870, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_83_cast")];
            tensor<int32, [4]> var_17876_begin_0 = const()[name = tensor<string, []>("op_17876_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_17876_end_0 = const()[name = tensor<string, []>("op_17876_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_17876_end_mask_0 = const()[name = tensor<string, []>("op_17876_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17876_cast = slice_by_index(begin = var_17876_begin_0, end = var_17876_end_0, end_mask = var_17876_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17876_cast")];
            tensor<int32, [4]> var_17880_begin_0 = const()[name = tensor<string, []>("op_17880_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_17880_end_0 = const()[name = tensor<string, []>("op_17880_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_17880_end_mask_0 = const()[name = tensor<string, []>("op_17880_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17880_cast = slice_by_index(begin = var_17880_begin_0, end = var_17880_end_0, end_mask = var_17880_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17880_cast")];
            tensor<int32, [4]> var_17884_begin_0 = const()[name = tensor<string, []>("op_17884_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_17884_end_0 = const()[name = tensor<string, []>("op_17884_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_17884_end_mask_0 = const()[name = tensor<string, []>("op_17884_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17884_cast = slice_by_index(begin = var_17884_begin_0, end = var_17884_end_0, end_mask = var_17884_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17884_cast")];
            tensor<int32, [4]> var_17888_begin_0 = const()[name = tensor<string, []>("op_17888_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_17888_end_0 = const()[name = tensor<string, []>("op_17888_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_17888_end_mask_0 = const()[name = tensor<string, []>("op_17888_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17888_cast = slice_by_index(begin = var_17888_begin_0, end = var_17888_end_0, end_mask = var_17888_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17888_cast")];
            tensor<int32, [4]> var_17892_begin_0 = const()[name = tensor<string, []>("op_17892_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_17892_end_0 = const()[name = tensor<string, []>("op_17892_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_17892_end_mask_0 = const()[name = tensor<string, []>("op_17892_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17892_cast = slice_by_index(begin = var_17892_begin_0, end = var_17892_end_0, end_mask = var_17892_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17892_cast")];
            tensor<int32, [4]> var_17896_begin_0 = const()[name = tensor<string, []>("op_17896_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_17896_end_0 = const()[name = tensor<string, []>("op_17896_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_17896_end_mask_0 = const()[name = tensor<string, []>("op_17896_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17896_cast = slice_by_index(begin = var_17896_begin_0, end = var_17896_end_0, end_mask = var_17896_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17896_cast")];
            tensor<int32, [4]> var_17900_begin_0 = const()[name = tensor<string, []>("op_17900_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_17900_end_0 = const()[name = tensor<string, []>("op_17900_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_17900_end_mask_0 = const()[name = tensor<string, []>("op_17900_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17900_cast = slice_by_index(begin = var_17900_begin_0, end = var_17900_end_0, end_mask = var_17900_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17900_cast")];
            tensor<int32, [4]> var_17904_begin_0 = const()[name = tensor<string, []>("op_17904_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_17904_end_0 = const()[name = tensor<string, []>("op_17904_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_17904_end_mask_0 = const()[name = tensor<string, []>("op_17904_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17904_cast = slice_by_index(begin = var_17904_begin_0, end = var_17904_end_0, end_mask = var_17904_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17904_cast")];
            tensor<int32, [4]> var_17908_begin_0 = const()[name = tensor<string, []>("op_17908_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_17908_end_0 = const()[name = tensor<string, []>("op_17908_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_17908_end_mask_0 = const()[name = tensor<string, []>("op_17908_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17908_cast = slice_by_index(begin = var_17908_begin_0, end = var_17908_end_0, end_mask = var_17908_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17908_cast")];
            tensor<int32, [4]> var_17912_begin_0 = const()[name = tensor<string, []>("op_17912_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_17912_end_0 = const()[name = tensor<string, []>("op_17912_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_17912_end_mask_0 = const()[name = tensor<string, []>("op_17912_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17912_cast = slice_by_index(begin = var_17912_begin_0, end = var_17912_end_0, end_mask = var_17912_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17912_cast")];
            tensor<int32, [4]> var_17916_begin_0 = const()[name = tensor<string, []>("op_17916_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_17916_end_0 = const()[name = tensor<string, []>("op_17916_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_17916_end_mask_0 = const()[name = tensor<string, []>("op_17916_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17916_cast = slice_by_index(begin = var_17916_begin_0, end = var_17916_end_0, end_mask = var_17916_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17916_cast")];
            tensor<int32, [4]> var_17920_begin_0 = const()[name = tensor<string, []>("op_17920_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_17920_end_0 = const()[name = tensor<string, []>("op_17920_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_17920_end_mask_0 = const()[name = tensor<string, []>("op_17920_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17920_cast = slice_by_index(begin = var_17920_begin_0, end = var_17920_end_0, end_mask = var_17920_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17920_cast")];
            tensor<int32, [4]> var_17924_begin_0 = const()[name = tensor<string, []>("op_17924_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_17924_end_0 = const()[name = tensor<string, []>("op_17924_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_17924_end_mask_0 = const()[name = tensor<string, []>("op_17924_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17924_cast = slice_by_index(begin = var_17924_begin_0, end = var_17924_end_0, end_mask = var_17924_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17924_cast")];
            tensor<int32, [4]> var_17928_begin_0 = const()[name = tensor<string, []>("op_17928_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_17928_end_0 = const()[name = tensor<string, []>("op_17928_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_17928_end_mask_0 = const()[name = tensor<string, []>("op_17928_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17928_cast = slice_by_index(begin = var_17928_begin_0, end = var_17928_end_0, end_mask = var_17928_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17928_cast")];
            tensor<int32, [4]> var_17932_begin_0 = const()[name = tensor<string, []>("op_17932_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_17932_end_0 = const()[name = tensor<string, []>("op_17932_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_17932_end_mask_0 = const()[name = tensor<string, []>("op_17932_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17932_cast = slice_by_index(begin = var_17932_begin_0, end = var_17932_end_0, end_mask = var_17932_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17932_cast")];
            tensor<int32, [4]> var_17936_begin_0 = const()[name = tensor<string, []>("op_17936_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_17936_end_0 = const()[name = tensor<string, []>("op_17936_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_17936_end_mask_0 = const()[name = tensor<string, []>("op_17936_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17936_cast = slice_by_index(begin = var_17936_begin_0, end = var_17936_end_0, end_mask = var_17936_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17936_cast")];
            tensor<int32, [4]> var_17940_begin_0 = const()[name = tensor<string, []>("op_17940_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_17940_end_0 = const()[name = tensor<string, []>("op_17940_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_17940_end_mask_0 = const()[name = tensor<string, []>("op_17940_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17940_cast = slice_by_index(begin = var_17940_begin_0, end = var_17940_end_0, end_mask = var_17940_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17940_cast")];
            tensor<int32, [4]> var_17944_begin_0 = const()[name = tensor<string, []>("op_17944_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_17944_end_0 = const()[name = tensor<string, []>("op_17944_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_17944_end_mask_0 = const()[name = tensor<string, []>("op_17944_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17944_cast = slice_by_index(begin = var_17944_begin_0, end = var_17944_end_0, end_mask = var_17944_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17944_cast")];
            tensor<int32, [4]> var_17948_begin_0 = const()[name = tensor<string, []>("op_17948_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_17948_end_0 = const()[name = tensor<string, []>("op_17948_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_17948_end_mask_0 = const()[name = tensor<string, []>("op_17948_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17948_cast = slice_by_index(begin = var_17948_begin_0, end = var_17948_end_0, end_mask = var_17948_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17948_cast")];
            tensor<int32, [4]> var_17952_begin_0 = const()[name = tensor<string, []>("op_17952_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_17952_end_0 = const()[name = tensor<string, []>("op_17952_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_17952_end_mask_0 = const()[name = tensor<string, []>("op_17952_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_17952_cast = slice_by_index(begin = var_17952_begin_0, end = var_17952_end_0, end_mask = var_17952_end_mask_0, x = q_83_cast)[name = tensor<string, []>("op_17952_cast")];
            tensor<int32, [4]> k_167_perm_0 = const()[name = tensor<string, []>("k_167_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_17959_begin_0 = const()[name = tensor<string, []>("op_17959_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_17959_end_0 = const()[name = tensor<string, []>("op_17959_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_17959_end_mask_0 = const()[name = tensor<string, []>("op_17959_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_98 = transpose(perm = k_167_perm_0, x = k_165_cast)[name = tensor<string, []>("transpose_98")];
            tensor<fp16, [2, 77, 1, 64]> var_17959_cast = slice_by_index(begin = var_17959_begin_0, end = var_17959_end_0, end_mask = var_17959_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_17959_cast")];
            tensor<int32, [4]> var_17963_begin_0 = const()[name = tensor<string, []>("op_17963_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_17963_end_0 = const()[name = tensor<string, []>("op_17963_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_17963_end_mask_0 = const()[name = tensor<string, []>("op_17963_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17963_cast = slice_by_index(begin = var_17963_begin_0, end = var_17963_end_0, end_mask = var_17963_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_17963_cast")];
            tensor<int32, [4]> var_17967_begin_0 = const()[name = tensor<string, []>("op_17967_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_17967_end_0 = const()[name = tensor<string, []>("op_17967_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_17967_end_mask_0 = const()[name = tensor<string, []>("op_17967_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17967_cast = slice_by_index(begin = var_17967_begin_0, end = var_17967_end_0, end_mask = var_17967_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_17967_cast")];
            tensor<int32, [4]> var_17971_begin_0 = const()[name = tensor<string, []>("op_17971_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_17971_end_0 = const()[name = tensor<string, []>("op_17971_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_17971_end_mask_0 = const()[name = tensor<string, []>("op_17971_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17971_cast = slice_by_index(begin = var_17971_begin_0, end = var_17971_end_0, end_mask = var_17971_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_17971_cast")];
            tensor<int32, [4]> var_17975_begin_0 = const()[name = tensor<string, []>("op_17975_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_17975_end_0 = const()[name = tensor<string, []>("op_17975_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_17975_end_mask_0 = const()[name = tensor<string, []>("op_17975_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17975_cast = slice_by_index(begin = var_17975_begin_0, end = var_17975_end_0, end_mask = var_17975_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_17975_cast")];
            tensor<int32, [4]> var_17979_begin_0 = const()[name = tensor<string, []>("op_17979_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_17979_end_0 = const()[name = tensor<string, []>("op_17979_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_17979_end_mask_0 = const()[name = tensor<string, []>("op_17979_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17979_cast = slice_by_index(begin = var_17979_begin_0, end = var_17979_end_0, end_mask = var_17979_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_17979_cast")];
            tensor<int32, [4]> var_17983_begin_0 = const()[name = tensor<string, []>("op_17983_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_17983_end_0 = const()[name = tensor<string, []>("op_17983_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_17983_end_mask_0 = const()[name = tensor<string, []>("op_17983_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17983_cast = slice_by_index(begin = var_17983_begin_0, end = var_17983_end_0, end_mask = var_17983_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_17983_cast")];
            tensor<int32, [4]> var_17987_begin_0 = const()[name = tensor<string, []>("op_17987_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_17987_end_0 = const()[name = tensor<string, []>("op_17987_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_17987_end_mask_0 = const()[name = tensor<string, []>("op_17987_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17987_cast = slice_by_index(begin = var_17987_begin_0, end = var_17987_end_0, end_mask = var_17987_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_17987_cast")];
            tensor<int32, [4]> var_17991_begin_0 = const()[name = tensor<string, []>("op_17991_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_17991_end_0 = const()[name = tensor<string, []>("op_17991_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_17991_end_mask_0 = const()[name = tensor<string, []>("op_17991_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17991_cast = slice_by_index(begin = var_17991_begin_0, end = var_17991_end_0, end_mask = var_17991_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_17991_cast")];
            tensor<int32, [4]> var_17995_begin_0 = const()[name = tensor<string, []>("op_17995_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_17995_end_0 = const()[name = tensor<string, []>("op_17995_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_17995_end_mask_0 = const()[name = tensor<string, []>("op_17995_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17995_cast = slice_by_index(begin = var_17995_begin_0, end = var_17995_end_0, end_mask = var_17995_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_17995_cast")];
            tensor<int32, [4]> var_17999_begin_0 = const()[name = tensor<string, []>("op_17999_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_17999_end_0 = const()[name = tensor<string, []>("op_17999_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_17999_end_mask_0 = const()[name = tensor<string, []>("op_17999_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_17999_cast = slice_by_index(begin = var_17999_begin_0, end = var_17999_end_0, end_mask = var_17999_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_17999_cast")];
            tensor<int32, [4]> var_18003_begin_0 = const()[name = tensor<string, []>("op_18003_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_18003_end_0 = const()[name = tensor<string, []>("op_18003_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_18003_end_mask_0 = const()[name = tensor<string, []>("op_18003_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18003_cast = slice_by_index(begin = var_18003_begin_0, end = var_18003_end_0, end_mask = var_18003_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_18003_cast")];
            tensor<int32, [4]> var_18007_begin_0 = const()[name = tensor<string, []>("op_18007_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_18007_end_0 = const()[name = tensor<string, []>("op_18007_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_18007_end_mask_0 = const()[name = tensor<string, []>("op_18007_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18007_cast = slice_by_index(begin = var_18007_begin_0, end = var_18007_end_0, end_mask = var_18007_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_18007_cast")];
            tensor<int32, [4]> var_18011_begin_0 = const()[name = tensor<string, []>("op_18011_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_18011_end_0 = const()[name = tensor<string, []>("op_18011_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_18011_end_mask_0 = const()[name = tensor<string, []>("op_18011_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18011_cast = slice_by_index(begin = var_18011_begin_0, end = var_18011_end_0, end_mask = var_18011_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_18011_cast")];
            tensor<int32, [4]> var_18015_begin_0 = const()[name = tensor<string, []>("op_18015_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_18015_end_0 = const()[name = tensor<string, []>("op_18015_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_18015_end_mask_0 = const()[name = tensor<string, []>("op_18015_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18015_cast = slice_by_index(begin = var_18015_begin_0, end = var_18015_end_0, end_mask = var_18015_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_18015_cast")];
            tensor<int32, [4]> var_18019_begin_0 = const()[name = tensor<string, []>("op_18019_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_18019_end_0 = const()[name = tensor<string, []>("op_18019_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_18019_end_mask_0 = const()[name = tensor<string, []>("op_18019_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18019_cast = slice_by_index(begin = var_18019_begin_0, end = var_18019_end_0, end_mask = var_18019_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_18019_cast")];
            tensor<int32, [4]> var_18023_begin_0 = const()[name = tensor<string, []>("op_18023_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_18023_end_0 = const()[name = tensor<string, []>("op_18023_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_18023_end_mask_0 = const()[name = tensor<string, []>("op_18023_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18023_cast = slice_by_index(begin = var_18023_begin_0, end = var_18023_end_0, end_mask = var_18023_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_18023_cast")];
            tensor<int32, [4]> var_18027_begin_0 = const()[name = tensor<string, []>("op_18027_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_18027_end_0 = const()[name = tensor<string, []>("op_18027_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_18027_end_mask_0 = const()[name = tensor<string, []>("op_18027_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18027_cast = slice_by_index(begin = var_18027_begin_0, end = var_18027_end_0, end_mask = var_18027_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_18027_cast")];
            tensor<int32, [4]> var_18031_begin_0 = const()[name = tensor<string, []>("op_18031_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_18031_end_0 = const()[name = tensor<string, []>("op_18031_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_18031_end_mask_0 = const()[name = tensor<string, []>("op_18031_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18031_cast = slice_by_index(begin = var_18031_begin_0, end = var_18031_end_0, end_mask = var_18031_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_18031_cast")];
            tensor<int32, [4]> var_18035_begin_0 = const()[name = tensor<string, []>("op_18035_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_18035_end_0 = const()[name = tensor<string, []>("op_18035_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_18035_end_mask_0 = const()[name = tensor<string, []>("op_18035_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18035_cast = slice_by_index(begin = var_18035_begin_0, end = var_18035_end_0, end_mask = var_18035_end_mask_0, x = transpose_98)[name = tensor<string, []>("op_18035_cast")];
            tensor<int32, [4]> var_18037_begin_0 = const()[name = tensor<string, []>("op_18037_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_18037_end_0 = const()[name = tensor<string, []>("op_18037_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_18037_end_mask_0 = const()[name = tensor<string, []>("op_18037_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18037_cast = slice_by_index(begin = var_18037_begin_0, end = var_18037_end_0, end_mask = var_18037_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18037_cast")];
            tensor<int32, [4]> var_18041_begin_0 = const()[name = tensor<string, []>("op_18041_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_18041_end_0 = const()[name = tensor<string, []>("op_18041_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_18041_end_mask_0 = const()[name = tensor<string, []>("op_18041_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18041_cast = slice_by_index(begin = var_18041_begin_0, end = var_18041_end_0, end_mask = var_18041_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18041_cast")];
            tensor<int32, [4]> var_18045_begin_0 = const()[name = tensor<string, []>("op_18045_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_18045_end_0 = const()[name = tensor<string, []>("op_18045_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_18045_end_mask_0 = const()[name = tensor<string, []>("op_18045_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18045_cast = slice_by_index(begin = var_18045_begin_0, end = var_18045_end_0, end_mask = var_18045_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18045_cast")];
            tensor<int32, [4]> var_18049_begin_0 = const()[name = tensor<string, []>("op_18049_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_18049_end_0 = const()[name = tensor<string, []>("op_18049_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_18049_end_mask_0 = const()[name = tensor<string, []>("op_18049_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18049_cast = slice_by_index(begin = var_18049_begin_0, end = var_18049_end_0, end_mask = var_18049_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18049_cast")];
            tensor<int32, [4]> var_18053_begin_0 = const()[name = tensor<string, []>("op_18053_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_18053_end_0 = const()[name = tensor<string, []>("op_18053_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_18053_end_mask_0 = const()[name = tensor<string, []>("op_18053_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18053_cast = slice_by_index(begin = var_18053_begin_0, end = var_18053_end_0, end_mask = var_18053_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18053_cast")];
            tensor<int32, [4]> var_18057_begin_0 = const()[name = tensor<string, []>("op_18057_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_18057_end_0 = const()[name = tensor<string, []>("op_18057_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_18057_end_mask_0 = const()[name = tensor<string, []>("op_18057_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18057_cast = slice_by_index(begin = var_18057_begin_0, end = var_18057_end_0, end_mask = var_18057_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18057_cast")];
            tensor<int32, [4]> var_18061_begin_0 = const()[name = tensor<string, []>("op_18061_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_18061_end_0 = const()[name = tensor<string, []>("op_18061_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_18061_end_mask_0 = const()[name = tensor<string, []>("op_18061_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18061_cast = slice_by_index(begin = var_18061_begin_0, end = var_18061_end_0, end_mask = var_18061_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18061_cast")];
            tensor<int32, [4]> var_18065_begin_0 = const()[name = tensor<string, []>("op_18065_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_18065_end_0 = const()[name = tensor<string, []>("op_18065_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_18065_end_mask_0 = const()[name = tensor<string, []>("op_18065_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18065_cast = slice_by_index(begin = var_18065_begin_0, end = var_18065_end_0, end_mask = var_18065_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18065_cast")];
            tensor<int32, [4]> var_18069_begin_0 = const()[name = tensor<string, []>("op_18069_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_18069_end_0 = const()[name = tensor<string, []>("op_18069_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_18069_end_mask_0 = const()[name = tensor<string, []>("op_18069_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18069_cast = slice_by_index(begin = var_18069_begin_0, end = var_18069_end_0, end_mask = var_18069_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18069_cast")];
            tensor<int32, [4]> var_18073_begin_0 = const()[name = tensor<string, []>("op_18073_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_18073_end_0 = const()[name = tensor<string, []>("op_18073_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_18073_end_mask_0 = const()[name = tensor<string, []>("op_18073_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18073_cast = slice_by_index(begin = var_18073_begin_0, end = var_18073_end_0, end_mask = var_18073_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18073_cast")];
            tensor<int32, [4]> var_18077_begin_0 = const()[name = tensor<string, []>("op_18077_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_18077_end_0 = const()[name = tensor<string, []>("op_18077_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_18077_end_mask_0 = const()[name = tensor<string, []>("op_18077_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18077_cast = slice_by_index(begin = var_18077_begin_0, end = var_18077_end_0, end_mask = var_18077_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18077_cast")];
            tensor<int32, [4]> var_18081_begin_0 = const()[name = tensor<string, []>("op_18081_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_18081_end_0 = const()[name = tensor<string, []>("op_18081_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_18081_end_mask_0 = const()[name = tensor<string, []>("op_18081_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18081_cast = slice_by_index(begin = var_18081_begin_0, end = var_18081_end_0, end_mask = var_18081_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18081_cast")];
            tensor<int32, [4]> var_18085_begin_0 = const()[name = tensor<string, []>("op_18085_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_18085_end_0 = const()[name = tensor<string, []>("op_18085_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_18085_end_mask_0 = const()[name = tensor<string, []>("op_18085_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18085_cast = slice_by_index(begin = var_18085_begin_0, end = var_18085_end_0, end_mask = var_18085_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18085_cast")];
            tensor<int32, [4]> var_18089_begin_0 = const()[name = tensor<string, []>("op_18089_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_18089_end_0 = const()[name = tensor<string, []>("op_18089_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_18089_end_mask_0 = const()[name = tensor<string, []>("op_18089_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18089_cast = slice_by_index(begin = var_18089_begin_0, end = var_18089_end_0, end_mask = var_18089_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18089_cast")];
            tensor<int32, [4]> var_18093_begin_0 = const()[name = tensor<string, []>("op_18093_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_18093_end_0 = const()[name = tensor<string, []>("op_18093_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_18093_end_mask_0 = const()[name = tensor<string, []>("op_18093_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18093_cast = slice_by_index(begin = var_18093_begin_0, end = var_18093_end_0, end_mask = var_18093_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18093_cast")];
            tensor<int32, [4]> var_18097_begin_0 = const()[name = tensor<string, []>("op_18097_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_18097_end_0 = const()[name = tensor<string, []>("op_18097_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_18097_end_mask_0 = const()[name = tensor<string, []>("op_18097_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18097_cast = slice_by_index(begin = var_18097_begin_0, end = var_18097_end_0, end_mask = var_18097_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18097_cast")];
            tensor<int32, [4]> var_18101_begin_0 = const()[name = tensor<string, []>("op_18101_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_18101_end_0 = const()[name = tensor<string, []>("op_18101_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_18101_end_mask_0 = const()[name = tensor<string, []>("op_18101_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18101_cast = slice_by_index(begin = var_18101_begin_0, end = var_18101_end_0, end_mask = var_18101_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18101_cast")];
            tensor<int32, [4]> var_18105_begin_0 = const()[name = tensor<string, []>("op_18105_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_18105_end_0 = const()[name = tensor<string, []>("op_18105_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_18105_end_mask_0 = const()[name = tensor<string, []>("op_18105_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18105_cast = slice_by_index(begin = var_18105_begin_0, end = var_18105_end_0, end_mask = var_18105_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18105_cast")];
            tensor<int32, [4]> var_18109_begin_0 = const()[name = tensor<string, []>("op_18109_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_18109_end_0 = const()[name = tensor<string, []>("op_18109_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_18109_end_mask_0 = const()[name = tensor<string, []>("op_18109_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18109_cast = slice_by_index(begin = var_18109_begin_0, end = var_18109_end_0, end_mask = var_18109_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18109_cast")];
            tensor<int32, [4]> var_18113_begin_0 = const()[name = tensor<string, []>("op_18113_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_18113_end_0 = const()[name = tensor<string, []>("op_18113_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_18113_end_mask_0 = const()[name = tensor<string, []>("op_18113_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18113_cast = slice_by_index(begin = var_18113_begin_0, end = var_18113_end_0, end_mask = var_18113_end_mask_0, x = v_83_cast)[name = tensor<string, []>("op_18113_cast")];
            tensor<string, []> var_18117_equation_0 = const()[name = tensor<string, []>("op_18117_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18117_cast = einsum(equation = var_18117_equation_0, values = (var_17959_cast, var_17876_cast))[name = tensor<string, []>("op_18117_cast")];
            tensor<fp16, []> var_18118_to_fp16 = const()[name = tensor<string, []>("op_18118_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1481_cast = mul(x = var_18117_cast, y = var_18118_to_fp16)[name = tensor<string, []>("aw_1481_cast")];
            tensor<string, []> var_18121_equation_0 = const()[name = tensor<string, []>("op_18121_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18121_cast = einsum(equation = var_18121_equation_0, values = (var_17963_cast, var_17880_cast))[name = tensor<string, []>("op_18121_cast")];
            tensor<fp16, []> var_18122_to_fp16 = const()[name = tensor<string, []>("op_18122_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1483_cast = mul(x = var_18121_cast, y = var_18122_to_fp16)[name = tensor<string, []>("aw_1483_cast")];
            tensor<string, []> var_18125_equation_0 = const()[name = tensor<string, []>("op_18125_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18125_cast = einsum(equation = var_18125_equation_0, values = (var_17967_cast, var_17884_cast))[name = tensor<string, []>("op_18125_cast")];
            tensor<fp16, []> var_18126_to_fp16 = const()[name = tensor<string, []>("op_18126_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1485_cast = mul(x = var_18125_cast, y = var_18126_to_fp16)[name = tensor<string, []>("aw_1485_cast")];
            tensor<string, []> var_18129_equation_0 = const()[name = tensor<string, []>("op_18129_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18129_cast = einsum(equation = var_18129_equation_0, values = (var_17971_cast, var_17888_cast))[name = tensor<string, []>("op_18129_cast")];
            tensor<fp16, []> var_18130_to_fp16 = const()[name = tensor<string, []>("op_18130_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1487_cast = mul(x = var_18129_cast, y = var_18130_to_fp16)[name = tensor<string, []>("aw_1487_cast")];
            tensor<string, []> var_18133_equation_0 = const()[name = tensor<string, []>("op_18133_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18133_cast = einsum(equation = var_18133_equation_0, values = (var_17975_cast, var_17892_cast))[name = tensor<string, []>("op_18133_cast")];
            tensor<fp16, []> var_18134_to_fp16 = const()[name = tensor<string, []>("op_18134_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1489_cast = mul(x = var_18133_cast, y = var_18134_to_fp16)[name = tensor<string, []>("aw_1489_cast")];
            tensor<string, []> var_18137_equation_0 = const()[name = tensor<string, []>("op_18137_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18137_cast = einsum(equation = var_18137_equation_0, values = (var_17979_cast, var_17896_cast))[name = tensor<string, []>("op_18137_cast")];
            tensor<fp16, []> var_18138_to_fp16 = const()[name = tensor<string, []>("op_18138_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1491_cast = mul(x = var_18137_cast, y = var_18138_to_fp16)[name = tensor<string, []>("aw_1491_cast")];
            tensor<string, []> var_18141_equation_0 = const()[name = tensor<string, []>("op_18141_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18141_cast = einsum(equation = var_18141_equation_0, values = (var_17983_cast, var_17900_cast))[name = tensor<string, []>("op_18141_cast")];
            tensor<fp16, []> var_18142_to_fp16 = const()[name = tensor<string, []>("op_18142_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1493_cast = mul(x = var_18141_cast, y = var_18142_to_fp16)[name = tensor<string, []>("aw_1493_cast")];
            tensor<string, []> var_18145_equation_0 = const()[name = tensor<string, []>("op_18145_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18145_cast = einsum(equation = var_18145_equation_0, values = (var_17987_cast, var_17904_cast))[name = tensor<string, []>("op_18145_cast")];
            tensor<fp16, []> var_18146_to_fp16 = const()[name = tensor<string, []>("op_18146_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1495_cast = mul(x = var_18145_cast, y = var_18146_to_fp16)[name = tensor<string, []>("aw_1495_cast")];
            tensor<string, []> var_18149_equation_0 = const()[name = tensor<string, []>("op_18149_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18149_cast = einsum(equation = var_18149_equation_0, values = (var_17991_cast, var_17908_cast))[name = tensor<string, []>("op_18149_cast")];
            tensor<fp16, []> var_18150_to_fp16 = const()[name = tensor<string, []>("op_18150_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1497_cast = mul(x = var_18149_cast, y = var_18150_to_fp16)[name = tensor<string, []>("aw_1497_cast")];
            tensor<string, []> var_18153_equation_0 = const()[name = tensor<string, []>("op_18153_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18153_cast = einsum(equation = var_18153_equation_0, values = (var_17995_cast, var_17912_cast))[name = tensor<string, []>("op_18153_cast")];
            tensor<fp16, []> var_18154_to_fp16 = const()[name = tensor<string, []>("op_18154_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1499_cast = mul(x = var_18153_cast, y = var_18154_to_fp16)[name = tensor<string, []>("aw_1499_cast")];
            tensor<string, []> var_18157_equation_0 = const()[name = tensor<string, []>("op_18157_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18157_cast = einsum(equation = var_18157_equation_0, values = (var_17999_cast, var_17916_cast))[name = tensor<string, []>("op_18157_cast")];
            tensor<fp16, []> var_18158_to_fp16 = const()[name = tensor<string, []>("op_18158_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1501_cast = mul(x = var_18157_cast, y = var_18158_to_fp16)[name = tensor<string, []>("aw_1501_cast")];
            tensor<string, []> var_18161_equation_0 = const()[name = tensor<string, []>("op_18161_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18161_cast = einsum(equation = var_18161_equation_0, values = (var_18003_cast, var_17920_cast))[name = tensor<string, []>("op_18161_cast")];
            tensor<fp16, []> var_18162_to_fp16 = const()[name = tensor<string, []>("op_18162_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1503_cast = mul(x = var_18161_cast, y = var_18162_to_fp16)[name = tensor<string, []>("aw_1503_cast")];
            tensor<string, []> var_18165_equation_0 = const()[name = tensor<string, []>("op_18165_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18165_cast = einsum(equation = var_18165_equation_0, values = (var_18007_cast, var_17924_cast))[name = tensor<string, []>("op_18165_cast")];
            tensor<fp16, []> var_18166_to_fp16 = const()[name = tensor<string, []>("op_18166_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1505_cast = mul(x = var_18165_cast, y = var_18166_to_fp16)[name = tensor<string, []>("aw_1505_cast")];
            tensor<string, []> var_18169_equation_0 = const()[name = tensor<string, []>("op_18169_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18169_cast = einsum(equation = var_18169_equation_0, values = (var_18011_cast, var_17928_cast))[name = tensor<string, []>("op_18169_cast")];
            tensor<fp16, []> var_18170_to_fp16 = const()[name = tensor<string, []>("op_18170_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1507_cast = mul(x = var_18169_cast, y = var_18170_to_fp16)[name = tensor<string, []>("aw_1507_cast")];
            tensor<string, []> var_18173_equation_0 = const()[name = tensor<string, []>("op_18173_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18173_cast = einsum(equation = var_18173_equation_0, values = (var_18015_cast, var_17932_cast))[name = tensor<string, []>("op_18173_cast")];
            tensor<fp16, []> var_18174_to_fp16 = const()[name = tensor<string, []>("op_18174_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1509_cast = mul(x = var_18173_cast, y = var_18174_to_fp16)[name = tensor<string, []>("aw_1509_cast")];
            tensor<string, []> var_18177_equation_0 = const()[name = tensor<string, []>("op_18177_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18177_cast = einsum(equation = var_18177_equation_0, values = (var_18019_cast, var_17936_cast))[name = tensor<string, []>("op_18177_cast")];
            tensor<fp16, []> var_18178_to_fp16 = const()[name = tensor<string, []>("op_18178_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1511_cast = mul(x = var_18177_cast, y = var_18178_to_fp16)[name = tensor<string, []>("aw_1511_cast")];
            tensor<string, []> var_18181_equation_0 = const()[name = tensor<string, []>("op_18181_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18181_cast = einsum(equation = var_18181_equation_0, values = (var_18023_cast, var_17940_cast))[name = tensor<string, []>("op_18181_cast")];
            tensor<fp16, []> var_18182_to_fp16 = const()[name = tensor<string, []>("op_18182_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1513_cast = mul(x = var_18181_cast, y = var_18182_to_fp16)[name = tensor<string, []>("aw_1513_cast")];
            tensor<string, []> var_18185_equation_0 = const()[name = tensor<string, []>("op_18185_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18185_cast = einsum(equation = var_18185_equation_0, values = (var_18027_cast, var_17944_cast))[name = tensor<string, []>("op_18185_cast")];
            tensor<fp16, []> var_18186_to_fp16 = const()[name = tensor<string, []>("op_18186_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1515_cast = mul(x = var_18185_cast, y = var_18186_to_fp16)[name = tensor<string, []>("aw_1515_cast")];
            tensor<string, []> var_18189_equation_0 = const()[name = tensor<string, []>("op_18189_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18189_cast = einsum(equation = var_18189_equation_0, values = (var_18031_cast, var_17948_cast))[name = tensor<string, []>("op_18189_cast")];
            tensor<fp16, []> var_18190_to_fp16 = const()[name = tensor<string, []>("op_18190_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1517_cast = mul(x = var_18189_cast, y = var_18190_to_fp16)[name = tensor<string, []>("aw_1517_cast")];
            tensor<string, []> var_18193_equation_0 = const()[name = tensor<string, []>("op_18193_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_18193_cast = einsum(equation = var_18193_equation_0, values = (var_18035_cast, var_17952_cast))[name = tensor<string, []>("op_18193_cast")];
            tensor<fp16, []> var_18194_to_fp16 = const()[name = tensor<string, []>("op_18194_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1519_cast = mul(x = var_18193_cast, y = var_18194_to_fp16)[name = tensor<string, []>("aw_1519_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18196_cast = softmax(axis = var_2634, x = aw_1481_cast)[name = tensor<string, []>("op_18196_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18197_cast = softmax(axis = var_2634, x = aw_1483_cast)[name = tensor<string, []>("op_18197_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18198_cast = softmax(axis = var_2634, x = aw_1485_cast)[name = tensor<string, []>("op_18198_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18199_cast = softmax(axis = var_2634, x = aw_1487_cast)[name = tensor<string, []>("op_18199_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18200_cast = softmax(axis = var_2634, x = aw_1489_cast)[name = tensor<string, []>("op_18200_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18201_cast = softmax(axis = var_2634, x = aw_1491_cast)[name = tensor<string, []>("op_18201_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18202_cast = softmax(axis = var_2634, x = aw_1493_cast)[name = tensor<string, []>("op_18202_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18203_cast = softmax(axis = var_2634, x = aw_1495_cast)[name = tensor<string, []>("op_18203_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18204_cast = softmax(axis = var_2634, x = aw_1497_cast)[name = tensor<string, []>("op_18204_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18205_cast = softmax(axis = var_2634, x = aw_1499_cast)[name = tensor<string, []>("op_18205_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18206_cast = softmax(axis = var_2634, x = aw_1501_cast)[name = tensor<string, []>("op_18206_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18207_cast = softmax(axis = var_2634, x = aw_1503_cast)[name = tensor<string, []>("op_18207_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18208_cast = softmax(axis = var_2634, x = aw_1505_cast)[name = tensor<string, []>("op_18208_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18209_cast = softmax(axis = var_2634, x = aw_1507_cast)[name = tensor<string, []>("op_18209_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18210_cast = softmax(axis = var_2634, x = aw_1509_cast)[name = tensor<string, []>("op_18210_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18211_cast = softmax(axis = var_2634, x = aw_1511_cast)[name = tensor<string, []>("op_18211_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18212_cast = softmax(axis = var_2634, x = aw_1513_cast)[name = tensor<string, []>("op_18212_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18213_cast = softmax(axis = var_2634, x = aw_1515_cast)[name = tensor<string, []>("op_18213_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18214_cast = softmax(axis = var_2634, x = aw_1517_cast)[name = tensor<string, []>("op_18214_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_18215_cast = softmax(axis = var_2634, x = aw_1519_cast)[name = tensor<string, []>("op_18215_cast")];
            tensor<string, []> var_18217_equation_0 = const()[name = tensor<string, []>("op_18217_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18217_cast = einsum(equation = var_18217_equation_0, values = (var_18037_cast, var_18196_cast))[name = tensor<string, []>("op_18217_cast")];
            tensor<string, []> var_18219_equation_0 = const()[name = tensor<string, []>("op_18219_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18219_cast = einsum(equation = var_18219_equation_0, values = (var_18041_cast, var_18197_cast))[name = tensor<string, []>("op_18219_cast")];
            tensor<string, []> var_18221_equation_0 = const()[name = tensor<string, []>("op_18221_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18221_cast = einsum(equation = var_18221_equation_0, values = (var_18045_cast, var_18198_cast))[name = tensor<string, []>("op_18221_cast")];
            tensor<string, []> var_18223_equation_0 = const()[name = tensor<string, []>("op_18223_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18223_cast = einsum(equation = var_18223_equation_0, values = (var_18049_cast, var_18199_cast))[name = tensor<string, []>("op_18223_cast")];
            tensor<string, []> var_18225_equation_0 = const()[name = tensor<string, []>("op_18225_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18225_cast = einsum(equation = var_18225_equation_0, values = (var_18053_cast, var_18200_cast))[name = tensor<string, []>("op_18225_cast")];
            tensor<string, []> var_18227_equation_0 = const()[name = tensor<string, []>("op_18227_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18227_cast = einsum(equation = var_18227_equation_0, values = (var_18057_cast, var_18201_cast))[name = tensor<string, []>("op_18227_cast")];
            tensor<string, []> var_18229_equation_0 = const()[name = tensor<string, []>("op_18229_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18229_cast = einsum(equation = var_18229_equation_0, values = (var_18061_cast, var_18202_cast))[name = tensor<string, []>("op_18229_cast")];
            tensor<string, []> var_18231_equation_0 = const()[name = tensor<string, []>("op_18231_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18231_cast = einsum(equation = var_18231_equation_0, values = (var_18065_cast, var_18203_cast))[name = tensor<string, []>("op_18231_cast")];
            tensor<string, []> var_18233_equation_0 = const()[name = tensor<string, []>("op_18233_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18233_cast = einsum(equation = var_18233_equation_0, values = (var_18069_cast, var_18204_cast))[name = tensor<string, []>("op_18233_cast")];
            tensor<string, []> var_18235_equation_0 = const()[name = tensor<string, []>("op_18235_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18235_cast = einsum(equation = var_18235_equation_0, values = (var_18073_cast, var_18205_cast))[name = tensor<string, []>("op_18235_cast")];
            tensor<string, []> var_18237_equation_0 = const()[name = tensor<string, []>("op_18237_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18237_cast = einsum(equation = var_18237_equation_0, values = (var_18077_cast, var_18206_cast))[name = tensor<string, []>("op_18237_cast")];
            tensor<string, []> var_18239_equation_0 = const()[name = tensor<string, []>("op_18239_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18239_cast = einsum(equation = var_18239_equation_0, values = (var_18081_cast, var_18207_cast))[name = tensor<string, []>("op_18239_cast")];
            tensor<string, []> var_18241_equation_0 = const()[name = tensor<string, []>("op_18241_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18241_cast = einsum(equation = var_18241_equation_0, values = (var_18085_cast, var_18208_cast))[name = tensor<string, []>("op_18241_cast")];
            tensor<string, []> var_18243_equation_0 = const()[name = tensor<string, []>("op_18243_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18243_cast = einsum(equation = var_18243_equation_0, values = (var_18089_cast, var_18209_cast))[name = tensor<string, []>("op_18243_cast")];
            tensor<string, []> var_18245_equation_0 = const()[name = tensor<string, []>("op_18245_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18245_cast = einsum(equation = var_18245_equation_0, values = (var_18093_cast, var_18210_cast))[name = tensor<string, []>("op_18245_cast")];
            tensor<string, []> var_18247_equation_0 = const()[name = tensor<string, []>("op_18247_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18247_cast = einsum(equation = var_18247_equation_0, values = (var_18097_cast, var_18211_cast))[name = tensor<string, []>("op_18247_cast")];
            tensor<string, []> var_18249_equation_0 = const()[name = tensor<string, []>("op_18249_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18249_cast = einsum(equation = var_18249_equation_0, values = (var_18101_cast, var_18212_cast))[name = tensor<string, []>("op_18249_cast")];
            tensor<string, []> var_18251_equation_0 = const()[name = tensor<string, []>("op_18251_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18251_cast = einsum(equation = var_18251_equation_0, values = (var_18105_cast, var_18213_cast))[name = tensor<string, []>("op_18251_cast")];
            tensor<string, []> var_18253_equation_0 = const()[name = tensor<string, []>("op_18253_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18253_cast = einsum(equation = var_18253_equation_0, values = (var_18109_cast, var_18214_cast))[name = tensor<string, []>("op_18253_cast")];
            tensor<string, []> var_18255_equation_0 = const()[name = tensor<string, []>("op_18255_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18255_cast = einsum(equation = var_18255_equation_0, values = (var_18113_cast, var_18215_cast))[name = tensor<string, []>("op_18255_cast")];
            tensor<bool, []> input_283_interleave_0 = const()[name = tensor<string, []>("input_283_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_283_cast = concat(axis = var_2634, interleave = input_283_interleave_0, values = (var_18217_cast, var_18219_cast, var_18221_cast, var_18223_cast, var_18225_cast, var_18227_cast, var_18229_cast, var_18231_cast, var_18233_cast, var_18235_cast, var_18237_cast, var_18239_cast, var_18241_cast, var_18243_cast, var_18245_cast, var_18247_cast, var_18249_cast, var_18251_cast, var_18253_cast, var_18255_cast))[name = tensor<string, []>("input_283_cast")];
            tensor<int32, [2]> var_18261 = const()[name = tensor<string, []>("op_18261"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_18263 = const()[name = tensor<string, []>("op_18263"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_18265_pad_type_0 = const()[name = tensor<string, []>("op_18265_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_18265_pad_0 = const()[name = tensor<string, []>("op_18265_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1406098560)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1409375424)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_18265_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_out_0_bias_to_fp16, dilations = var_18263, groups = var_2634, pad = var_18265_pad_0, pad_type = var_18265_pad_type_0, strides = var_18261, weight = down_blocks_2_attentions_1_transformer_blocks_6_attn2_to_out_0_weight_to_fp16, x = input_283_cast)[name = tensor<string, []>("op_18265_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_125_cast = add(x = var_18265_cast, y = inputs_123_cast)[name = tensor<string, []>("inputs_125_cast")];
            tensor<int32, [1]> var_18269 = const()[name = tensor<string, []>("op_18269"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_125_cast = reduce_mean(axes = var_18269, keep_dims = var_2629, x = inputs_125_cast)[name = tensor<string, []>("channels_mean_125_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_125_cast = sub(x = inputs_125_cast, y = channels_mean_125_cast)[name = tensor<string, []>("zero_mean_125_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_125_cast = mul(x = zero_mean_125_cast, y = zero_mean_125_cast)[name = tensor<string, []>("zero_mean_sq_125_cast")];
            tensor<int32, [1]> var_18273 = const()[name = tensor<string, []>("op_18273"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_18274_cast = reduce_mean(axes = var_18273, keep_dims = var_2629, x = zero_mean_sq_125_cast)[name = tensor<string, []>("op_18274_cast")];
            tensor<fp16, []> var_18275_to_fp16 = const()[name = tensor<string, []>("op_18275_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_18276_cast = add(x = var_18274_cast, y = var_18275_to_fp16)[name = tensor<string, []>("op_18276_cast")];
            tensor<fp16, []> denom_125_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_125_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_125_cast = rsqrt(epsilon = denom_125_epsilon_0_to_fp16, x = var_18276_cast)[name = tensor<string, []>("denom_125_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_125_cast = mul(x = zero_mean_125_cast, y = denom_125_cast)[name = tensor<string, []>("out_125_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_18280_to_fp16 = const()[name = tensor<string, []>("op_18280_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1409378048)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_18281_cast = add(x = out_125_cast, y = var_18280_to_fp16)[name = tensor<string, []>("op_18281_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_18283_to_fp16 = const()[name = tensor<string, []>("op_18283_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1409380672)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_285_cast = mul(x = var_18281_cast, y = var_18283_to_fp16)[name = tensor<string, []>("input_285_cast")];
            tensor<int32, [2]> var_18291 = const()[name = tensor<string, []>("op_18291"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_18293 = const()[name = tensor<string, []>("op_18293"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_18295_pad_type_0 = const()[name = tensor<string, []>("op_18295_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_18295_pad_0 = const()[name = tensor<string, []>("op_18295_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_6_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_6_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1409383296)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_1_transformer_blocks_6_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_6_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1435597760)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_18295_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_6_ff_net_0_proj_bias_to_fp16, dilations = var_18293, groups = var_2634, pad = var_18295_pad_0, pad_type = var_18295_pad_type_0, strides = var_18291, weight = down_blocks_2_attentions_1_transformer_blocks_6_ff_net_0_proj_weight_to_fp16, x = input_285_cast)[name = tensor<string, []>("op_18295_cast")];
            tensor<int32, [2]> var_18296_split_sizes_0 = const()[name = tensor<string, []>("op_18296_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_18296_axis_0 = const()[name = tensor<string, []>("op_18296_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_18296_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_18296_cast_1 = split(axis = var_18296_axis_0, split_sizes = var_18296_split_sizes_0, x = var_18295_cast)[name = tensor<string, []>("op_18296_cast")];
            tensor<string, []> var_18298_mode_0 = const()[name = tensor<string, []>("op_18298_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_18298_cast = gelu(mode = var_18298_mode_0, x = var_18296_cast_1)[name = tensor<string, []>("op_18298_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_287_cast = mul(x = var_18296_cast_0, y = var_18298_cast)[name = tensor<string, []>("input_287_cast")];
            tensor<int32, [2]> var_18302 = const()[name = tensor<string, []>("op_18302"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_18304 = const()[name = tensor<string, []>("op_18304"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_18306_pad_type_0 = const()[name = tensor<string, []>("op_18306_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_18306_pad_0 = const()[name = tensor<string, []>("op_18306_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_6_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_6_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1435618304)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_6_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_6_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1448725568)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_18306_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_6_ff_net_2_bias_to_fp16, dilations = var_18304, groups = var_2634, pad = var_18306_pad_0, pad_type = var_18306_pad_type_0, strides = var_18302, weight = down_blocks_2_attentions_1_transformer_blocks_6_ff_net_2_weight_to_fp16, x = input_287_cast)[name = tensor<string, []>("op_18306_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_127_cast = add(x = var_18306_cast, y = inputs_125_cast)[name = tensor<string, []>("inputs_127_cast")];
            tensor<int32, [1]> var_18316 = const()[name = tensor<string, []>("op_18316"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_127_cast = reduce_mean(axes = var_18316, keep_dims = var_2629, x = inputs_127_cast)[name = tensor<string, []>("channels_mean_127_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_127_cast = sub(x = inputs_127_cast, y = channels_mean_127_cast)[name = tensor<string, []>("zero_mean_127_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_127_cast = mul(x = zero_mean_127_cast, y = zero_mean_127_cast)[name = tensor<string, []>("zero_mean_sq_127_cast")];
            tensor<int32, [1]> var_18320 = const()[name = tensor<string, []>("op_18320"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_18321_cast = reduce_mean(axes = var_18320, keep_dims = var_2629, x = zero_mean_sq_127_cast)[name = tensor<string, []>("op_18321_cast")];
            tensor<fp16, []> var_18322_to_fp16 = const()[name = tensor<string, []>("op_18322_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_18323_cast = add(x = var_18321_cast, y = var_18322_to_fp16)[name = tensor<string, []>("op_18323_cast")];
            tensor<fp16, []> denom_127_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_127_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_127_cast = rsqrt(epsilon = denom_127_epsilon_0_to_fp16, x = var_18323_cast)[name = tensor<string, []>("denom_127_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_127_cast = mul(x = zero_mean_127_cast, y = denom_127_cast)[name = tensor<string, []>("out_127_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_18327_to_fp16 = const()[name = tensor<string, []>("op_18327_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1448728192)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_18328_cast = add(x = out_127_cast, y = var_18327_to_fp16)[name = tensor<string, []>("op_18328_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_18330_to_fp16 = const()[name = tensor<string, []>("op_18330_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1448730816)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_179_cast = mul(x = var_18328_cast, y = var_18330_to_fp16)[name = tensor<string, []>("hidden_states_179_cast")];
            tensor<int32, [2]> var_18337 = const()[name = tensor<string, []>("op_18337"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_18339 = const()[name = tensor<string, []>("op_18339"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_85_pad_type_0 = const()[name = tensor<string, []>("q_85_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_85_pad_0 = const()[name = tensor<string, []>("q_85_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1448733440)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_85_cast = conv(dilations = var_18339, groups = var_2634, pad = q_85_pad_0, pad_type = q_85_pad_type_0, strides = var_18337, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_q_weight_to_fp16, x = hidden_states_179_cast)[name = tensor<string, []>("q_85_cast")];
            tensor<int32, [2]> var_18343 = const()[name = tensor<string, []>("op_18343"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_18345 = const()[name = tensor<string, []>("op_18345"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_169_pad_type_0 = const()[name = tensor<string, []>("k_169_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_169_pad_0 = const()[name = tensor<string, []>("k_169_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1452010304)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_169_cast = conv(dilations = var_18345, groups = var_2634, pad = k_169_pad_0, pad_type = k_169_pad_type_0, strides = var_18343, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_k_weight_to_fp16, x = hidden_states_179_cast)[name = tensor<string, []>("k_169_cast")];
            tensor<int32, [2]> var_18349 = const()[name = tensor<string, []>("op_18349"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_18351 = const()[name = tensor<string, []>("op_18351"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_85_pad_type_0 = const()[name = tensor<string, []>("v_85_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_85_pad_0 = const()[name = tensor<string, []>("v_85_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1455287168)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_85_cast = conv(dilations = var_18351, groups = var_2634, pad = v_85_pad_0, pad_type = v_85_pad_type_0, strides = var_18349, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_v_weight_to_fp16, x = hidden_states_179_cast)[name = tensor<string, []>("v_85_cast")];
            tensor<int32, [4]> var_18355_begin_0 = const()[name = tensor<string, []>("op_18355_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_18355_end_0 = const()[name = tensor<string, []>("op_18355_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_18355_end_mask_0 = const()[name = tensor<string, []>("op_18355_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18355_cast = slice_by_index(begin = var_18355_begin_0, end = var_18355_end_0, end_mask = var_18355_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18355_cast")];
            tensor<int32, [4]> var_18359_begin_0 = const()[name = tensor<string, []>("op_18359_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_18359_end_0 = const()[name = tensor<string, []>("op_18359_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_18359_end_mask_0 = const()[name = tensor<string, []>("op_18359_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18359_cast = slice_by_index(begin = var_18359_begin_0, end = var_18359_end_0, end_mask = var_18359_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18359_cast")];
            tensor<int32, [4]> var_18363_begin_0 = const()[name = tensor<string, []>("op_18363_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_18363_end_0 = const()[name = tensor<string, []>("op_18363_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_18363_end_mask_0 = const()[name = tensor<string, []>("op_18363_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18363_cast = slice_by_index(begin = var_18363_begin_0, end = var_18363_end_0, end_mask = var_18363_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18363_cast")];
            tensor<int32, [4]> var_18367_begin_0 = const()[name = tensor<string, []>("op_18367_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_18367_end_0 = const()[name = tensor<string, []>("op_18367_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_18367_end_mask_0 = const()[name = tensor<string, []>("op_18367_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18367_cast = slice_by_index(begin = var_18367_begin_0, end = var_18367_end_0, end_mask = var_18367_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18367_cast")];
            tensor<int32, [4]> var_18371_begin_0 = const()[name = tensor<string, []>("op_18371_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_18371_end_0 = const()[name = tensor<string, []>("op_18371_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_18371_end_mask_0 = const()[name = tensor<string, []>("op_18371_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18371_cast = slice_by_index(begin = var_18371_begin_0, end = var_18371_end_0, end_mask = var_18371_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18371_cast")];
            tensor<int32, [4]> var_18375_begin_0 = const()[name = tensor<string, []>("op_18375_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_18375_end_0 = const()[name = tensor<string, []>("op_18375_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_18375_end_mask_0 = const()[name = tensor<string, []>("op_18375_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18375_cast = slice_by_index(begin = var_18375_begin_0, end = var_18375_end_0, end_mask = var_18375_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18375_cast")];
            tensor<int32, [4]> var_18379_begin_0 = const()[name = tensor<string, []>("op_18379_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_18379_end_0 = const()[name = tensor<string, []>("op_18379_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_18379_end_mask_0 = const()[name = tensor<string, []>("op_18379_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18379_cast = slice_by_index(begin = var_18379_begin_0, end = var_18379_end_0, end_mask = var_18379_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18379_cast")];
            tensor<int32, [4]> var_18383_begin_0 = const()[name = tensor<string, []>("op_18383_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_18383_end_0 = const()[name = tensor<string, []>("op_18383_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_18383_end_mask_0 = const()[name = tensor<string, []>("op_18383_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18383_cast = slice_by_index(begin = var_18383_begin_0, end = var_18383_end_0, end_mask = var_18383_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18383_cast")];
            tensor<int32, [4]> var_18387_begin_0 = const()[name = tensor<string, []>("op_18387_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_18387_end_0 = const()[name = tensor<string, []>("op_18387_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_18387_end_mask_0 = const()[name = tensor<string, []>("op_18387_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18387_cast = slice_by_index(begin = var_18387_begin_0, end = var_18387_end_0, end_mask = var_18387_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18387_cast")];
            tensor<int32, [4]> var_18391_begin_0 = const()[name = tensor<string, []>("op_18391_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_18391_end_0 = const()[name = tensor<string, []>("op_18391_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_18391_end_mask_0 = const()[name = tensor<string, []>("op_18391_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18391_cast = slice_by_index(begin = var_18391_begin_0, end = var_18391_end_0, end_mask = var_18391_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18391_cast")];
            tensor<int32, [4]> var_18395_begin_0 = const()[name = tensor<string, []>("op_18395_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_18395_end_0 = const()[name = tensor<string, []>("op_18395_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_18395_end_mask_0 = const()[name = tensor<string, []>("op_18395_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18395_cast = slice_by_index(begin = var_18395_begin_0, end = var_18395_end_0, end_mask = var_18395_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18395_cast")];
            tensor<int32, [4]> var_18399_begin_0 = const()[name = tensor<string, []>("op_18399_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_18399_end_0 = const()[name = tensor<string, []>("op_18399_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_18399_end_mask_0 = const()[name = tensor<string, []>("op_18399_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18399_cast = slice_by_index(begin = var_18399_begin_0, end = var_18399_end_0, end_mask = var_18399_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18399_cast")];
            tensor<int32, [4]> var_18403_begin_0 = const()[name = tensor<string, []>("op_18403_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_18403_end_0 = const()[name = tensor<string, []>("op_18403_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_18403_end_mask_0 = const()[name = tensor<string, []>("op_18403_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18403_cast = slice_by_index(begin = var_18403_begin_0, end = var_18403_end_0, end_mask = var_18403_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18403_cast")];
            tensor<int32, [4]> var_18407_begin_0 = const()[name = tensor<string, []>("op_18407_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_18407_end_0 = const()[name = tensor<string, []>("op_18407_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_18407_end_mask_0 = const()[name = tensor<string, []>("op_18407_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18407_cast = slice_by_index(begin = var_18407_begin_0, end = var_18407_end_0, end_mask = var_18407_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18407_cast")];
            tensor<int32, [4]> var_18411_begin_0 = const()[name = tensor<string, []>("op_18411_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_18411_end_0 = const()[name = tensor<string, []>("op_18411_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_18411_end_mask_0 = const()[name = tensor<string, []>("op_18411_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18411_cast = slice_by_index(begin = var_18411_begin_0, end = var_18411_end_0, end_mask = var_18411_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18411_cast")];
            tensor<int32, [4]> var_18415_begin_0 = const()[name = tensor<string, []>("op_18415_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_18415_end_0 = const()[name = tensor<string, []>("op_18415_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_18415_end_mask_0 = const()[name = tensor<string, []>("op_18415_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18415_cast = slice_by_index(begin = var_18415_begin_0, end = var_18415_end_0, end_mask = var_18415_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18415_cast")];
            tensor<int32, [4]> var_18419_begin_0 = const()[name = tensor<string, []>("op_18419_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_18419_end_0 = const()[name = tensor<string, []>("op_18419_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_18419_end_mask_0 = const()[name = tensor<string, []>("op_18419_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18419_cast = slice_by_index(begin = var_18419_begin_0, end = var_18419_end_0, end_mask = var_18419_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18419_cast")];
            tensor<int32, [4]> var_18423_begin_0 = const()[name = tensor<string, []>("op_18423_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_18423_end_0 = const()[name = tensor<string, []>("op_18423_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_18423_end_mask_0 = const()[name = tensor<string, []>("op_18423_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18423_cast = slice_by_index(begin = var_18423_begin_0, end = var_18423_end_0, end_mask = var_18423_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18423_cast")];
            tensor<int32, [4]> var_18427_begin_0 = const()[name = tensor<string, []>("op_18427_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_18427_end_0 = const()[name = tensor<string, []>("op_18427_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_18427_end_mask_0 = const()[name = tensor<string, []>("op_18427_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18427_cast = slice_by_index(begin = var_18427_begin_0, end = var_18427_end_0, end_mask = var_18427_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18427_cast")];
            tensor<int32, [4]> var_18431_begin_0 = const()[name = tensor<string, []>("op_18431_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_18431_end_0 = const()[name = tensor<string, []>("op_18431_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_18431_end_mask_0 = const()[name = tensor<string, []>("op_18431_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18431_cast = slice_by_index(begin = var_18431_begin_0, end = var_18431_end_0, end_mask = var_18431_end_mask_0, x = q_85_cast)[name = tensor<string, []>("op_18431_cast")];
            tensor<int32, [4]> k_171_perm_0 = const()[name = tensor<string, []>("k_171_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_18438_begin_0 = const()[name = tensor<string, []>("op_18438_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_18438_end_0 = const()[name = tensor<string, []>("op_18438_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_18438_end_mask_0 = const()[name = tensor<string, []>("op_18438_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_97 = transpose(perm = k_171_perm_0, x = k_169_cast)[name = tensor<string, []>("transpose_97")];
            tensor<fp16, [2, 1024, 1, 64]> var_18438_cast = slice_by_index(begin = var_18438_begin_0, end = var_18438_end_0, end_mask = var_18438_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18438_cast")];
            tensor<int32, [4]> var_18442_begin_0 = const()[name = tensor<string, []>("op_18442_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_18442_end_0 = const()[name = tensor<string, []>("op_18442_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_18442_end_mask_0 = const()[name = tensor<string, []>("op_18442_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_18442_cast = slice_by_index(begin = var_18442_begin_0, end = var_18442_end_0, end_mask = var_18442_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18442_cast")];
            tensor<int32, [4]> var_18446_begin_0 = const()[name = tensor<string, []>("op_18446_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_18446_end_0 = const()[name = tensor<string, []>("op_18446_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_18446_end_mask_0 = const()[name = tensor<string, []>("op_18446_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_18446_cast = slice_by_index(begin = var_18446_begin_0, end = var_18446_end_0, end_mask = var_18446_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18446_cast")];
            tensor<int32, [4]> var_18450_begin_0 = const()[name = tensor<string, []>("op_18450_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_18450_end_0 = const()[name = tensor<string, []>("op_18450_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_18450_end_mask_0 = const()[name = tensor<string, []>("op_18450_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_18450_cast = slice_by_index(begin = var_18450_begin_0, end = var_18450_end_0, end_mask = var_18450_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18450_cast")];
            tensor<int32, [4]> var_18454_begin_0 = const()[name = tensor<string, []>("op_18454_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_18454_end_0 = const()[name = tensor<string, []>("op_18454_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_18454_end_mask_0 = const()[name = tensor<string, []>("op_18454_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_18454_cast = slice_by_index(begin = var_18454_begin_0, end = var_18454_end_0, end_mask = var_18454_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18454_cast")];
            tensor<int32, [4]> var_18458_begin_0 = const()[name = tensor<string, []>("op_18458_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_18458_end_0 = const()[name = tensor<string, []>("op_18458_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_18458_end_mask_0 = const()[name = tensor<string, []>("op_18458_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_18458_cast = slice_by_index(begin = var_18458_begin_0, end = var_18458_end_0, end_mask = var_18458_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18458_cast")];
            tensor<int32, [4]> var_18462_begin_0 = const()[name = tensor<string, []>("op_18462_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_18462_end_0 = const()[name = tensor<string, []>("op_18462_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_18462_end_mask_0 = const()[name = tensor<string, []>("op_18462_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_18462_cast = slice_by_index(begin = var_18462_begin_0, end = var_18462_end_0, end_mask = var_18462_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18462_cast")];
            tensor<int32, [4]> var_18466_begin_0 = const()[name = tensor<string, []>("op_18466_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_18466_end_0 = const()[name = tensor<string, []>("op_18466_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_18466_end_mask_0 = const()[name = tensor<string, []>("op_18466_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_18466_cast = slice_by_index(begin = var_18466_begin_0, end = var_18466_end_0, end_mask = var_18466_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18466_cast")];
            tensor<int32, [4]> var_18470_begin_0 = const()[name = tensor<string, []>("op_18470_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_18470_end_0 = const()[name = tensor<string, []>("op_18470_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_18470_end_mask_0 = const()[name = tensor<string, []>("op_18470_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_18470_cast = slice_by_index(begin = var_18470_begin_0, end = var_18470_end_0, end_mask = var_18470_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18470_cast")];
            tensor<int32, [4]> var_18474_begin_0 = const()[name = tensor<string, []>("op_18474_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_18474_end_0 = const()[name = tensor<string, []>("op_18474_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_18474_end_mask_0 = const()[name = tensor<string, []>("op_18474_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_18474_cast = slice_by_index(begin = var_18474_begin_0, end = var_18474_end_0, end_mask = var_18474_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18474_cast")];
            tensor<int32, [4]> var_18478_begin_0 = const()[name = tensor<string, []>("op_18478_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_18478_end_0 = const()[name = tensor<string, []>("op_18478_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_18478_end_mask_0 = const()[name = tensor<string, []>("op_18478_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_18478_cast = slice_by_index(begin = var_18478_begin_0, end = var_18478_end_0, end_mask = var_18478_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18478_cast")];
            tensor<int32, [4]> var_18482_begin_0 = const()[name = tensor<string, []>("op_18482_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_18482_end_0 = const()[name = tensor<string, []>("op_18482_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_18482_end_mask_0 = const()[name = tensor<string, []>("op_18482_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_18482_cast = slice_by_index(begin = var_18482_begin_0, end = var_18482_end_0, end_mask = var_18482_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18482_cast")];
            tensor<int32, [4]> var_18486_begin_0 = const()[name = tensor<string, []>("op_18486_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_18486_end_0 = const()[name = tensor<string, []>("op_18486_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_18486_end_mask_0 = const()[name = tensor<string, []>("op_18486_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_18486_cast = slice_by_index(begin = var_18486_begin_0, end = var_18486_end_0, end_mask = var_18486_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18486_cast")];
            tensor<int32, [4]> var_18490_begin_0 = const()[name = tensor<string, []>("op_18490_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_18490_end_0 = const()[name = tensor<string, []>("op_18490_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_18490_end_mask_0 = const()[name = tensor<string, []>("op_18490_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_18490_cast = slice_by_index(begin = var_18490_begin_0, end = var_18490_end_0, end_mask = var_18490_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18490_cast")];
            tensor<int32, [4]> var_18494_begin_0 = const()[name = tensor<string, []>("op_18494_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_18494_end_0 = const()[name = tensor<string, []>("op_18494_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_18494_end_mask_0 = const()[name = tensor<string, []>("op_18494_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_18494_cast = slice_by_index(begin = var_18494_begin_0, end = var_18494_end_0, end_mask = var_18494_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18494_cast")];
            tensor<int32, [4]> var_18498_begin_0 = const()[name = tensor<string, []>("op_18498_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_18498_end_0 = const()[name = tensor<string, []>("op_18498_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_18498_end_mask_0 = const()[name = tensor<string, []>("op_18498_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_18498_cast = slice_by_index(begin = var_18498_begin_0, end = var_18498_end_0, end_mask = var_18498_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18498_cast")];
            tensor<int32, [4]> var_18502_begin_0 = const()[name = tensor<string, []>("op_18502_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_18502_end_0 = const()[name = tensor<string, []>("op_18502_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_18502_end_mask_0 = const()[name = tensor<string, []>("op_18502_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_18502_cast = slice_by_index(begin = var_18502_begin_0, end = var_18502_end_0, end_mask = var_18502_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18502_cast")];
            tensor<int32, [4]> var_18506_begin_0 = const()[name = tensor<string, []>("op_18506_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_18506_end_0 = const()[name = tensor<string, []>("op_18506_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_18506_end_mask_0 = const()[name = tensor<string, []>("op_18506_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_18506_cast = slice_by_index(begin = var_18506_begin_0, end = var_18506_end_0, end_mask = var_18506_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18506_cast")];
            tensor<int32, [4]> var_18510_begin_0 = const()[name = tensor<string, []>("op_18510_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_18510_end_0 = const()[name = tensor<string, []>("op_18510_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_18510_end_mask_0 = const()[name = tensor<string, []>("op_18510_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_18510_cast = slice_by_index(begin = var_18510_begin_0, end = var_18510_end_0, end_mask = var_18510_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18510_cast")];
            tensor<int32, [4]> var_18514_begin_0 = const()[name = tensor<string, []>("op_18514_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_18514_end_0 = const()[name = tensor<string, []>("op_18514_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_18514_end_mask_0 = const()[name = tensor<string, []>("op_18514_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_18514_cast = slice_by_index(begin = var_18514_begin_0, end = var_18514_end_0, end_mask = var_18514_end_mask_0, x = transpose_97)[name = tensor<string, []>("op_18514_cast")];
            tensor<int32, [4]> var_18516_begin_0 = const()[name = tensor<string, []>("op_18516_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_18516_end_0 = const()[name = tensor<string, []>("op_18516_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_18516_end_mask_0 = const()[name = tensor<string, []>("op_18516_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18516_cast = slice_by_index(begin = var_18516_begin_0, end = var_18516_end_0, end_mask = var_18516_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18516_cast")];
            tensor<int32, [4]> var_18520_begin_0 = const()[name = tensor<string, []>("op_18520_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_18520_end_0 = const()[name = tensor<string, []>("op_18520_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_18520_end_mask_0 = const()[name = tensor<string, []>("op_18520_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18520_cast = slice_by_index(begin = var_18520_begin_0, end = var_18520_end_0, end_mask = var_18520_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18520_cast")];
            tensor<int32, [4]> var_18524_begin_0 = const()[name = tensor<string, []>("op_18524_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_18524_end_0 = const()[name = tensor<string, []>("op_18524_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_18524_end_mask_0 = const()[name = tensor<string, []>("op_18524_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18524_cast = slice_by_index(begin = var_18524_begin_0, end = var_18524_end_0, end_mask = var_18524_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18524_cast")];
            tensor<int32, [4]> var_18528_begin_0 = const()[name = tensor<string, []>("op_18528_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_18528_end_0 = const()[name = tensor<string, []>("op_18528_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_18528_end_mask_0 = const()[name = tensor<string, []>("op_18528_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18528_cast = slice_by_index(begin = var_18528_begin_0, end = var_18528_end_0, end_mask = var_18528_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18528_cast")];
            tensor<int32, [4]> var_18532_begin_0 = const()[name = tensor<string, []>("op_18532_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_18532_end_0 = const()[name = tensor<string, []>("op_18532_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_18532_end_mask_0 = const()[name = tensor<string, []>("op_18532_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18532_cast = slice_by_index(begin = var_18532_begin_0, end = var_18532_end_0, end_mask = var_18532_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18532_cast")];
            tensor<int32, [4]> var_18536_begin_0 = const()[name = tensor<string, []>("op_18536_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_18536_end_0 = const()[name = tensor<string, []>("op_18536_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_18536_end_mask_0 = const()[name = tensor<string, []>("op_18536_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18536_cast = slice_by_index(begin = var_18536_begin_0, end = var_18536_end_0, end_mask = var_18536_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18536_cast")];
            tensor<int32, [4]> var_18540_begin_0 = const()[name = tensor<string, []>("op_18540_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_18540_end_0 = const()[name = tensor<string, []>("op_18540_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_18540_end_mask_0 = const()[name = tensor<string, []>("op_18540_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18540_cast = slice_by_index(begin = var_18540_begin_0, end = var_18540_end_0, end_mask = var_18540_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18540_cast")];
            tensor<int32, [4]> var_18544_begin_0 = const()[name = tensor<string, []>("op_18544_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_18544_end_0 = const()[name = tensor<string, []>("op_18544_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_18544_end_mask_0 = const()[name = tensor<string, []>("op_18544_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18544_cast = slice_by_index(begin = var_18544_begin_0, end = var_18544_end_0, end_mask = var_18544_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18544_cast")];
            tensor<int32, [4]> var_18548_begin_0 = const()[name = tensor<string, []>("op_18548_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_18548_end_0 = const()[name = tensor<string, []>("op_18548_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_18548_end_mask_0 = const()[name = tensor<string, []>("op_18548_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18548_cast = slice_by_index(begin = var_18548_begin_0, end = var_18548_end_0, end_mask = var_18548_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18548_cast")];
            tensor<int32, [4]> var_18552_begin_0 = const()[name = tensor<string, []>("op_18552_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_18552_end_0 = const()[name = tensor<string, []>("op_18552_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_18552_end_mask_0 = const()[name = tensor<string, []>("op_18552_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18552_cast = slice_by_index(begin = var_18552_begin_0, end = var_18552_end_0, end_mask = var_18552_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18552_cast")];
            tensor<int32, [4]> var_18556_begin_0 = const()[name = tensor<string, []>("op_18556_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_18556_end_0 = const()[name = tensor<string, []>("op_18556_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_18556_end_mask_0 = const()[name = tensor<string, []>("op_18556_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18556_cast = slice_by_index(begin = var_18556_begin_0, end = var_18556_end_0, end_mask = var_18556_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18556_cast")];
            tensor<int32, [4]> var_18560_begin_0 = const()[name = tensor<string, []>("op_18560_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_18560_end_0 = const()[name = tensor<string, []>("op_18560_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_18560_end_mask_0 = const()[name = tensor<string, []>("op_18560_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18560_cast = slice_by_index(begin = var_18560_begin_0, end = var_18560_end_0, end_mask = var_18560_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18560_cast")];
            tensor<int32, [4]> var_18564_begin_0 = const()[name = tensor<string, []>("op_18564_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_18564_end_0 = const()[name = tensor<string, []>("op_18564_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_18564_end_mask_0 = const()[name = tensor<string, []>("op_18564_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18564_cast = slice_by_index(begin = var_18564_begin_0, end = var_18564_end_0, end_mask = var_18564_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18564_cast")];
            tensor<int32, [4]> var_18568_begin_0 = const()[name = tensor<string, []>("op_18568_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_18568_end_0 = const()[name = tensor<string, []>("op_18568_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_18568_end_mask_0 = const()[name = tensor<string, []>("op_18568_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18568_cast = slice_by_index(begin = var_18568_begin_0, end = var_18568_end_0, end_mask = var_18568_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18568_cast")];
            tensor<int32, [4]> var_18572_begin_0 = const()[name = tensor<string, []>("op_18572_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_18572_end_0 = const()[name = tensor<string, []>("op_18572_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_18572_end_mask_0 = const()[name = tensor<string, []>("op_18572_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18572_cast = slice_by_index(begin = var_18572_begin_0, end = var_18572_end_0, end_mask = var_18572_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18572_cast")];
            tensor<int32, [4]> var_18576_begin_0 = const()[name = tensor<string, []>("op_18576_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_18576_end_0 = const()[name = tensor<string, []>("op_18576_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_18576_end_mask_0 = const()[name = tensor<string, []>("op_18576_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18576_cast = slice_by_index(begin = var_18576_begin_0, end = var_18576_end_0, end_mask = var_18576_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18576_cast")];
            tensor<int32, [4]> var_18580_begin_0 = const()[name = tensor<string, []>("op_18580_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_18580_end_0 = const()[name = tensor<string, []>("op_18580_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_18580_end_mask_0 = const()[name = tensor<string, []>("op_18580_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18580_cast = slice_by_index(begin = var_18580_begin_0, end = var_18580_end_0, end_mask = var_18580_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18580_cast")];
            tensor<int32, [4]> var_18584_begin_0 = const()[name = tensor<string, []>("op_18584_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_18584_end_0 = const()[name = tensor<string, []>("op_18584_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_18584_end_mask_0 = const()[name = tensor<string, []>("op_18584_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18584_cast = slice_by_index(begin = var_18584_begin_0, end = var_18584_end_0, end_mask = var_18584_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18584_cast")];
            tensor<int32, [4]> var_18588_begin_0 = const()[name = tensor<string, []>("op_18588_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_18588_end_0 = const()[name = tensor<string, []>("op_18588_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_18588_end_mask_0 = const()[name = tensor<string, []>("op_18588_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18588_cast = slice_by_index(begin = var_18588_begin_0, end = var_18588_end_0, end_mask = var_18588_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18588_cast")];
            tensor<int32, [4]> var_18592_begin_0 = const()[name = tensor<string, []>("op_18592_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_18592_end_0 = const()[name = tensor<string, []>("op_18592_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_18592_end_mask_0 = const()[name = tensor<string, []>("op_18592_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18592_cast = slice_by_index(begin = var_18592_begin_0, end = var_18592_end_0, end_mask = var_18592_end_mask_0, x = v_85_cast)[name = tensor<string, []>("op_18592_cast")];
            tensor<string, []> var_18596_equation_0 = const()[name = tensor<string, []>("op_18596_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18596_cast = einsum(equation = var_18596_equation_0, values = (var_18438_cast, var_18355_cast))[name = tensor<string, []>("op_18596_cast")];
            tensor<fp16, []> var_18597_to_fp16 = const()[name = tensor<string, []>("op_18597_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1521_cast = mul(x = var_18596_cast, y = var_18597_to_fp16)[name = tensor<string, []>("aw_1521_cast")];
            tensor<string, []> var_18600_equation_0 = const()[name = tensor<string, []>("op_18600_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18600_cast = einsum(equation = var_18600_equation_0, values = (var_18442_cast, var_18359_cast))[name = tensor<string, []>("op_18600_cast")];
            tensor<fp16, []> var_18601_to_fp16 = const()[name = tensor<string, []>("op_18601_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1523_cast = mul(x = var_18600_cast, y = var_18601_to_fp16)[name = tensor<string, []>("aw_1523_cast")];
            tensor<string, []> var_18604_equation_0 = const()[name = tensor<string, []>("op_18604_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18604_cast = einsum(equation = var_18604_equation_0, values = (var_18446_cast, var_18363_cast))[name = tensor<string, []>("op_18604_cast")];
            tensor<fp16, []> var_18605_to_fp16 = const()[name = tensor<string, []>("op_18605_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1525_cast = mul(x = var_18604_cast, y = var_18605_to_fp16)[name = tensor<string, []>("aw_1525_cast")];
            tensor<string, []> var_18608_equation_0 = const()[name = tensor<string, []>("op_18608_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18608_cast = einsum(equation = var_18608_equation_0, values = (var_18450_cast, var_18367_cast))[name = tensor<string, []>("op_18608_cast")];
            tensor<fp16, []> var_18609_to_fp16 = const()[name = tensor<string, []>("op_18609_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1527_cast = mul(x = var_18608_cast, y = var_18609_to_fp16)[name = tensor<string, []>("aw_1527_cast")];
            tensor<string, []> var_18612_equation_0 = const()[name = tensor<string, []>("op_18612_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18612_cast = einsum(equation = var_18612_equation_0, values = (var_18454_cast, var_18371_cast))[name = tensor<string, []>("op_18612_cast")];
            tensor<fp16, []> var_18613_to_fp16 = const()[name = tensor<string, []>("op_18613_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1529_cast = mul(x = var_18612_cast, y = var_18613_to_fp16)[name = tensor<string, []>("aw_1529_cast")];
            tensor<string, []> var_18616_equation_0 = const()[name = tensor<string, []>("op_18616_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18616_cast = einsum(equation = var_18616_equation_0, values = (var_18458_cast, var_18375_cast))[name = tensor<string, []>("op_18616_cast")];
            tensor<fp16, []> var_18617_to_fp16 = const()[name = tensor<string, []>("op_18617_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1531_cast = mul(x = var_18616_cast, y = var_18617_to_fp16)[name = tensor<string, []>("aw_1531_cast")];
            tensor<string, []> var_18620_equation_0 = const()[name = tensor<string, []>("op_18620_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18620_cast = einsum(equation = var_18620_equation_0, values = (var_18462_cast, var_18379_cast))[name = tensor<string, []>("op_18620_cast")];
            tensor<fp16, []> var_18621_to_fp16 = const()[name = tensor<string, []>("op_18621_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1533_cast = mul(x = var_18620_cast, y = var_18621_to_fp16)[name = tensor<string, []>("aw_1533_cast")];
            tensor<string, []> var_18624_equation_0 = const()[name = tensor<string, []>("op_18624_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18624_cast = einsum(equation = var_18624_equation_0, values = (var_18466_cast, var_18383_cast))[name = tensor<string, []>("op_18624_cast")];
            tensor<fp16, []> var_18625_to_fp16 = const()[name = tensor<string, []>("op_18625_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1535_cast = mul(x = var_18624_cast, y = var_18625_to_fp16)[name = tensor<string, []>("aw_1535_cast")];
            tensor<string, []> var_18628_equation_0 = const()[name = tensor<string, []>("op_18628_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18628_cast = einsum(equation = var_18628_equation_0, values = (var_18470_cast, var_18387_cast))[name = tensor<string, []>("op_18628_cast")];
            tensor<fp16, []> var_18629_to_fp16 = const()[name = tensor<string, []>("op_18629_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1537_cast = mul(x = var_18628_cast, y = var_18629_to_fp16)[name = tensor<string, []>("aw_1537_cast")];
            tensor<string, []> var_18632_equation_0 = const()[name = tensor<string, []>("op_18632_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18632_cast = einsum(equation = var_18632_equation_0, values = (var_18474_cast, var_18391_cast))[name = tensor<string, []>("op_18632_cast")];
            tensor<fp16, []> var_18633_to_fp16 = const()[name = tensor<string, []>("op_18633_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1539_cast = mul(x = var_18632_cast, y = var_18633_to_fp16)[name = tensor<string, []>("aw_1539_cast")];
            tensor<string, []> var_18636_equation_0 = const()[name = tensor<string, []>("op_18636_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18636_cast = einsum(equation = var_18636_equation_0, values = (var_18478_cast, var_18395_cast))[name = tensor<string, []>("op_18636_cast")];
            tensor<fp16, []> var_18637_to_fp16 = const()[name = tensor<string, []>("op_18637_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1541_cast = mul(x = var_18636_cast, y = var_18637_to_fp16)[name = tensor<string, []>("aw_1541_cast")];
            tensor<string, []> var_18640_equation_0 = const()[name = tensor<string, []>("op_18640_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18640_cast = einsum(equation = var_18640_equation_0, values = (var_18482_cast, var_18399_cast))[name = tensor<string, []>("op_18640_cast")];
            tensor<fp16, []> var_18641_to_fp16 = const()[name = tensor<string, []>("op_18641_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1543_cast = mul(x = var_18640_cast, y = var_18641_to_fp16)[name = tensor<string, []>("aw_1543_cast")];
            tensor<string, []> var_18644_equation_0 = const()[name = tensor<string, []>("op_18644_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18644_cast = einsum(equation = var_18644_equation_0, values = (var_18486_cast, var_18403_cast))[name = tensor<string, []>("op_18644_cast")];
            tensor<fp16, []> var_18645_to_fp16 = const()[name = tensor<string, []>("op_18645_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1545_cast = mul(x = var_18644_cast, y = var_18645_to_fp16)[name = tensor<string, []>("aw_1545_cast")];
            tensor<string, []> var_18648_equation_0 = const()[name = tensor<string, []>("op_18648_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18648_cast = einsum(equation = var_18648_equation_0, values = (var_18490_cast, var_18407_cast))[name = tensor<string, []>("op_18648_cast")];
            tensor<fp16, []> var_18649_to_fp16 = const()[name = tensor<string, []>("op_18649_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1547_cast = mul(x = var_18648_cast, y = var_18649_to_fp16)[name = tensor<string, []>("aw_1547_cast")];
            tensor<string, []> var_18652_equation_0 = const()[name = tensor<string, []>("op_18652_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18652_cast = einsum(equation = var_18652_equation_0, values = (var_18494_cast, var_18411_cast))[name = tensor<string, []>("op_18652_cast")];
            tensor<fp16, []> var_18653_to_fp16 = const()[name = tensor<string, []>("op_18653_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1549_cast = mul(x = var_18652_cast, y = var_18653_to_fp16)[name = tensor<string, []>("aw_1549_cast")];
            tensor<string, []> var_18656_equation_0 = const()[name = tensor<string, []>("op_18656_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18656_cast = einsum(equation = var_18656_equation_0, values = (var_18498_cast, var_18415_cast))[name = tensor<string, []>("op_18656_cast")];
            tensor<fp16, []> var_18657_to_fp16 = const()[name = tensor<string, []>("op_18657_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1551_cast = mul(x = var_18656_cast, y = var_18657_to_fp16)[name = tensor<string, []>("aw_1551_cast")];
            tensor<string, []> var_18660_equation_0 = const()[name = tensor<string, []>("op_18660_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18660_cast = einsum(equation = var_18660_equation_0, values = (var_18502_cast, var_18419_cast))[name = tensor<string, []>("op_18660_cast")];
            tensor<fp16, []> var_18661_to_fp16 = const()[name = tensor<string, []>("op_18661_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1553_cast = mul(x = var_18660_cast, y = var_18661_to_fp16)[name = tensor<string, []>("aw_1553_cast")];
            tensor<string, []> var_18664_equation_0 = const()[name = tensor<string, []>("op_18664_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18664_cast = einsum(equation = var_18664_equation_0, values = (var_18506_cast, var_18423_cast))[name = tensor<string, []>("op_18664_cast")];
            tensor<fp16, []> var_18665_to_fp16 = const()[name = tensor<string, []>("op_18665_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1555_cast = mul(x = var_18664_cast, y = var_18665_to_fp16)[name = tensor<string, []>("aw_1555_cast")];
            tensor<string, []> var_18668_equation_0 = const()[name = tensor<string, []>("op_18668_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18668_cast = einsum(equation = var_18668_equation_0, values = (var_18510_cast, var_18427_cast))[name = tensor<string, []>("op_18668_cast")];
            tensor<fp16, []> var_18669_to_fp16 = const()[name = tensor<string, []>("op_18669_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1557_cast = mul(x = var_18668_cast, y = var_18669_to_fp16)[name = tensor<string, []>("aw_1557_cast")];
            tensor<string, []> var_18672_equation_0 = const()[name = tensor<string, []>("op_18672_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18672_cast = einsum(equation = var_18672_equation_0, values = (var_18514_cast, var_18431_cast))[name = tensor<string, []>("op_18672_cast")];
            tensor<fp16, []> var_18673_to_fp16 = const()[name = tensor<string, []>("op_18673_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1559_cast = mul(x = var_18672_cast, y = var_18673_to_fp16)[name = tensor<string, []>("aw_1559_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18675_cast = softmax(axis = var_2634, x = aw_1521_cast)[name = tensor<string, []>("op_18675_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18676_cast = softmax(axis = var_2634, x = aw_1523_cast)[name = tensor<string, []>("op_18676_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18677_cast = softmax(axis = var_2634, x = aw_1525_cast)[name = tensor<string, []>("op_18677_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18678_cast = softmax(axis = var_2634, x = aw_1527_cast)[name = tensor<string, []>("op_18678_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18679_cast = softmax(axis = var_2634, x = aw_1529_cast)[name = tensor<string, []>("op_18679_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18680_cast = softmax(axis = var_2634, x = aw_1531_cast)[name = tensor<string, []>("op_18680_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18681_cast = softmax(axis = var_2634, x = aw_1533_cast)[name = tensor<string, []>("op_18681_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18682_cast = softmax(axis = var_2634, x = aw_1535_cast)[name = tensor<string, []>("op_18682_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18683_cast = softmax(axis = var_2634, x = aw_1537_cast)[name = tensor<string, []>("op_18683_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18684_cast = softmax(axis = var_2634, x = aw_1539_cast)[name = tensor<string, []>("op_18684_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18685_cast = softmax(axis = var_2634, x = aw_1541_cast)[name = tensor<string, []>("op_18685_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18686_cast = softmax(axis = var_2634, x = aw_1543_cast)[name = tensor<string, []>("op_18686_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18687_cast = softmax(axis = var_2634, x = aw_1545_cast)[name = tensor<string, []>("op_18687_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18688_cast = softmax(axis = var_2634, x = aw_1547_cast)[name = tensor<string, []>("op_18688_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18689_cast = softmax(axis = var_2634, x = aw_1549_cast)[name = tensor<string, []>("op_18689_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18690_cast = softmax(axis = var_2634, x = aw_1551_cast)[name = tensor<string, []>("op_18690_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18691_cast = softmax(axis = var_2634, x = aw_1553_cast)[name = tensor<string, []>("op_18691_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18692_cast = softmax(axis = var_2634, x = aw_1555_cast)[name = tensor<string, []>("op_18692_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18693_cast = softmax(axis = var_2634, x = aw_1557_cast)[name = tensor<string, []>("op_18693_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_18694_cast = softmax(axis = var_2634, x = aw_1559_cast)[name = tensor<string, []>("op_18694_cast")];
            tensor<string, []> var_18696_equation_0 = const()[name = tensor<string, []>("op_18696_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18696_cast = einsum(equation = var_18696_equation_0, values = (var_18516_cast, var_18675_cast))[name = tensor<string, []>("op_18696_cast")];
            tensor<string, []> var_18698_equation_0 = const()[name = tensor<string, []>("op_18698_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18698_cast = einsum(equation = var_18698_equation_0, values = (var_18520_cast, var_18676_cast))[name = tensor<string, []>("op_18698_cast")];
            tensor<string, []> var_18700_equation_0 = const()[name = tensor<string, []>("op_18700_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18700_cast = einsum(equation = var_18700_equation_0, values = (var_18524_cast, var_18677_cast))[name = tensor<string, []>("op_18700_cast")];
            tensor<string, []> var_18702_equation_0 = const()[name = tensor<string, []>("op_18702_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18702_cast = einsum(equation = var_18702_equation_0, values = (var_18528_cast, var_18678_cast))[name = tensor<string, []>("op_18702_cast")];
            tensor<string, []> var_18704_equation_0 = const()[name = tensor<string, []>("op_18704_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18704_cast = einsum(equation = var_18704_equation_0, values = (var_18532_cast, var_18679_cast))[name = tensor<string, []>("op_18704_cast")];
            tensor<string, []> var_18706_equation_0 = const()[name = tensor<string, []>("op_18706_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18706_cast = einsum(equation = var_18706_equation_0, values = (var_18536_cast, var_18680_cast))[name = tensor<string, []>("op_18706_cast")];
            tensor<string, []> var_18708_equation_0 = const()[name = tensor<string, []>("op_18708_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18708_cast = einsum(equation = var_18708_equation_0, values = (var_18540_cast, var_18681_cast))[name = tensor<string, []>("op_18708_cast")];
            tensor<string, []> var_18710_equation_0 = const()[name = tensor<string, []>("op_18710_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18710_cast = einsum(equation = var_18710_equation_0, values = (var_18544_cast, var_18682_cast))[name = tensor<string, []>("op_18710_cast")];
            tensor<string, []> var_18712_equation_0 = const()[name = tensor<string, []>("op_18712_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18712_cast = einsum(equation = var_18712_equation_0, values = (var_18548_cast, var_18683_cast))[name = tensor<string, []>("op_18712_cast")];
            tensor<string, []> var_18714_equation_0 = const()[name = tensor<string, []>("op_18714_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18714_cast = einsum(equation = var_18714_equation_0, values = (var_18552_cast, var_18684_cast))[name = tensor<string, []>("op_18714_cast")];
            tensor<string, []> var_18716_equation_0 = const()[name = tensor<string, []>("op_18716_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18716_cast = einsum(equation = var_18716_equation_0, values = (var_18556_cast, var_18685_cast))[name = tensor<string, []>("op_18716_cast")];
            tensor<string, []> var_18718_equation_0 = const()[name = tensor<string, []>("op_18718_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18718_cast = einsum(equation = var_18718_equation_0, values = (var_18560_cast, var_18686_cast))[name = tensor<string, []>("op_18718_cast")];
            tensor<string, []> var_18720_equation_0 = const()[name = tensor<string, []>("op_18720_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18720_cast = einsum(equation = var_18720_equation_0, values = (var_18564_cast, var_18687_cast))[name = tensor<string, []>("op_18720_cast")];
            tensor<string, []> var_18722_equation_0 = const()[name = tensor<string, []>("op_18722_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18722_cast = einsum(equation = var_18722_equation_0, values = (var_18568_cast, var_18688_cast))[name = tensor<string, []>("op_18722_cast")];
            tensor<string, []> var_18724_equation_0 = const()[name = tensor<string, []>("op_18724_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18724_cast = einsum(equation = var_18724_equation_0, values = (var_18572_cast, var_18689_cast))[name = tensor<string, []>("op_18724_cast")];
            tensor<string, []> var_18726_equation_0 = const()[name = tensor<string, []>("op_18726_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18726_cast = einsum(equation = var_18726_equation_0, values = (var_18576_cast, var_18690_cast))[name = tensor<string, []>("op_18726_cast")];
            tensor<string, []> var_18728_equation_0 = const()[name = tensor<string, []>("op_18728_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18728_cast = einsum(equation = var_18728_equation_0, values = (var_18580_cast, var_18691_cast))[name = tensor<string, []>("op_18728_cast")];
            tensor<string, []> var_18730_equation_0 = const()[name = tensor<string, []>("op_18730_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18730_cast = einsum(equation = var_18730_equation_0, values = (var_18584_cast, var_18692_cast))[name = tensor<string, []>("op_18730_cast")];
            tensor<string, []> var_18732_equation_0 = const()[name = tensor<string, []>("op_18732_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18732_cast = einsum(equation = var_18732_equation_0, values = (var_18588_cast, var_18693_cast))[name = tensor<string, []>("op_18732_cast")];
            tensor<string, []> var_18734_equation_0 = const()[name = tensor<string, []>("op_18734_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_18734_cast = einsum(equation = var_18734_equation_0, values = (var_18592_cast, var_18694_cast))[name = tensor<string, []>("op_18734_cast")];
            tensor<bool, []> input_289_interleave_0 = const()[name = tensor<string, []>("input_289_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_289_cast = concat(axis = var_2634, interleave = input_289_interleave_0, values = (var_18696_cast, var_18698_cast, var_18700_cast, var_18702_cast, var_18704_cast, var_18706_cast, var_18708_cast, var_18710_cast, var_18712_cast, var_18714_cast, var_18716_cast, var_18718_cast, var_18720_cast, var_18722_cast, var_18724_cast, var_18726_cast, var_18728_cast, var_18730_cast, var_18732_cast, var_18734_cast))[name = tensor<string, []>("input_289_cast")];
            tensor<int32, [2]> var_18740 = const()[name = tensor<string, []>("op_18740"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_18742 = const()[name = tensor<string, []>("op_18742"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_18744_pad_type_0 = const()[name = tensor<string, []>("op_18744_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_18744_pad_0 = const()[name = tensor<string, []>("op_18744_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1458564032)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1461840896)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_18744_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_out_0_bias_to_fp16, dilations = var_18742, groups = var_2634, pad = var_18744_pad_0, pad_type = var_18744_pad_type_0, strides = var_18740, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn1_to_out_0_weight_to_fp16, x = input_289_cast)[name = tensor<string, []>("op_18744_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_129_cast = add(x = var_18744_cast, y = inputs_127_cast)[name = tensor<string, []>("inputs_129_cast")];
            tensor<int32, [1]> var_18748 = const()[name = tensor<string, []>("op_18748"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_129_cast = reduce_mean(axes = var_18748, keep_dims = var_2629, x = inputs_129_cast)[name = tensor<string, []>("channels_mean_129_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_129_cast = sub(x = inputs_129_cast, y = channels_mean_129_cast)[name = tensor<string, []>("zero_mean_129_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_129_cast = mul(x = zero_mean_129_cast, y = zero_mean_129_cast)[name = tensor<string, []>("zero_mean_sq_129_cast")];
            tensor<int32, [1]> var_18752 = const()[name = tensor<string, []>("op_18752"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_18753_cast = reduce_mean(axes = var_18752, keep_dims = var_2629, x = zero_mean_sq_129_cast)[name = tensor<string, []>("op_18753_cast")];
            tensor<fp16, []> var_18754_to_fp16 = const()[name = tensor<string, []>("op_18754_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_18755_cast = add(x = var_18753_cast, y = var_18754_to_fp16)[name = tensor<string, []>("op_18755_cast")];
            tensor<fp16, []> denom_129_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_129_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_129_cast = rsqrt(epsilon = denom_129_epsilon_0_to_fp16, x = var_18755_cast)[name = tensor<string, []>("denom_129_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_129_cast = mul(x = zero_mean_129_cast, y = denom_129_cast)[name = tensor<string, []>("out_129_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_18759_to_fp16 = const()[name = tensor<string, []>("op_18759_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1461843520)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_18760_cast = add(x = out_129_cast, y = var_18759_to_fp16)[name = tensor<string, []>("op_18760_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_18762_to_fp16 = const()[name = tensor<string, []>("op_18762_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1461846144)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_181_cast = mul(x = var_18760_cast, y = var_18762_to_fp16)[name = tensor<string, []>("hidden_states_181_cast")];
            tensor<int32, [2]> var_18769 = const()[name = tensor<string, []>("op_18769"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_18771 = const()[name = tensor<string, []>("op_18771"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_87_pad_type_0 = const()[name = tensor<string, []>("q_87_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_87_pad_0 = const()[name = tensor<string, []>("q_87_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1461848768)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_87_cast = conv(dilations = var_18771, groups = var_2634, pad = q_87_pad_0, pad_type = q_87_pad_type_0, strides = var_18769, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_q_weight_to_fp16, x = hidden_states_181_cast)[name = tensor<string, []>("q_87_cast")];
            tensor<int32, [2]> var_18775 = const()[name = tensor<string, []>("op_18775"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_18777 = const()[name = tensor<string, []>("op_18777"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_173_pad_type_0 = const()[name = tensor<string, []>("k_173_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_173_pad_0 = const()[name = tensor<string, []>("k_173_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1465125632)))];
            tensor<fp16, [2, 1280, 1, 77]> k_173_cast = conv(dilations = var_18777, groups = var_2634, pad = k_173_pad_0, pad_type = k_173_pad_type_0, strides = var_18775, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_173_cast")];
            tensor<int32, [2]> var_18781 = const()[name = tensor<string, []>("op_18781"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_18783 = const()[name = tensor<string, []>("op_18783"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_87_pad_type_0 = const()[name = tensor<string, []>("v_87_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_87_pad_0 = const()[name = tensor<string, []>("v_87_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1470368576)))];
            tensor<fp16, [2, 1280, 1, 77]> v_87_cast = conv(dilations = var_18783, groups = var_2634, pad = v_87_pad_0, pad_type = v_87_pad_type_0, strides = var_18781, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_87_cast")];
            tensor<int32, [4]> var_18787_begin_0 = const()[name = tensor<string, []>("op_18787_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_18787_end_0 = const()[name = tensor<string, []>("op_18787_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_18787_end_mask_0 = const()[name = tensor<string, []>("op_18787_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18787_cast = slice_by_index(begin = var_18787_begin_0, end = var_18787_end_0, end_mask = var_18787_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18787_cast")];
            tensor<int32, [4]> var_18791_begin_0 = const()[name = tensor<string, []>("op_18791_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_18791_end_0 = const()[name = tensor<string, []>("op_18791_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_18791_end_mask_0 = const()[name = tensor<string, []>("op_18791_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18791_cast = slice_by_index(begin = var_18791_begin_0, end = var_18791_end_0, end_mask = var_18791_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18791_cast")];
            tensor<int32, [4]> var_18795_begin_0 = const()[name = tensor<string, []>("op_18795_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_18795_end_0 = const()[name = tensor<string, []>("op_18795_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_18795_end_mask_0 = const()[name = tensor<string, []>("op_18795_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18795_cast = slice_by_index(begin = var_18795_begin_0, end = var_18795_end_0, end_mask = var_18795_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18795_cast")];
            tensor<int32, [4]> var_18799_begin_0 = const()[name = tensor<string, []>("op_18799_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_18799_end_0 = const()[name = tensor<string, []>("op_18799_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_18799_end_mask_0 = const()[name = tensor<string, []>("op_18799_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18799_cast = slice_by_index(begin = var_18799_begin_0, end = var_18799_end_0, end_mask = var_18799_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18799_cast")];
            tensor<int32, [4]> var_18803_begin_0 = const()[name = tensor<string, []>("op_18803_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_18803_end_0 = const()[name = tensor<string, []>("op_18803_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_18803_end_mask_0 = const()[name = tensor<string, []>("op_18803_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18803_cast = slice_by_index(begin = var_18803_begin_0, end = var_18803_end_0, end_mask = var_18803_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18803_cast")];
            tensor<int32, [4]> var_18807_begin_0 = const()[name = tensor<string, []>("op_18807_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_18807_end_0 = const()[name = tensor<string, []>("op_18807_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_18807_end_mask_0 = const()[name = tensor<string, []>("op_18807_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18807_cast = slice_by_index(begin = var_18807_begin_0, end = var_18807_end_0, end_mask = var_18807_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18807_cast")];
            tensor<int32, [4]> var_18811_begin_0 = const()[name = tensor<string, []>("op_18811_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_18811_end_0 = const()[name = tensor<string, []>("op_18811_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_18811_end_mask_0 = const()[name = tensor<string, []>("op_18811_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18811_cast = slice_by_index(begin = var_18811_begin_0, end = var_18811_end_0, end_mask = var_18811_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18811_cast")];
            tensor<int32, [4]> var_18815_begin_0 = const()[name = tensor<string, []>("op_18815_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_18815_end_0 = const()[name = tensor<string, []>("op_18815_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_18815_end_mask_0 = const()[name = tensor<string, []>("op_18815_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18815_cast = slice_by_index(begin = var_18815_begin_0, end = var_18815_end_0, end_mask = var_18815_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18815_cast")];
            tensor<int32, [4]> var_18819_begin_0 = const()[name = tensor<string, []>("op_18819_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_18819_end_0 = const()[name = tensor<string, []>("op_18819_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_18819_end_mask_0 = const()[name = tensor<string, []>("op_18819_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18819_cast = slice_by_index(begin = var_18819_begin_0, end = var_18819_end_0, end_mask = var_18819_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18819_cast")];
            tensor<int32, [4]> var_18823_begin_0 = const()[name = tensor<string, []>("op_18823_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_18823_end_0 = const()[name = tensor<string, []>("op_18823_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_18823_end_mask_0 = const()[name = tensor<string, []>("op_18823_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18823_cast = slice_by_index(begin = var_18823_begin_0, end = var_18823_end_0, end_mask = var_18823_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18823_cast")];
            tensor<int32, [4]> var_18827_begin_0 = const()[name = tensor<string, []>("op_18827_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_18827_end_0 = const()[name = tensor<string, []>("op_18827_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_18827_end_mask_0 = const()[name = tensor<string, []>("op_18827_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18827_cast = slice_by_index(begin = var_18827_begin_0, end = var_18827_end_0, end_mask = var_18827_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18827_cast")];
            tensor<int32, [4]> var_18831_begin_0 = const()[name = tensor<string, []>("op_18831_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_18831_end_0 = const()[name = tensor<string, []>("op_18831_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_18831_end_mask_0 = const()[name = tensor<string, []>("op_18831_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18831_cast = slice_by_index(begin = var_18831_begin_0, end = var_18831_end_0, end_mask = var_18831_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18831_cast")];
            tensor<int32, [4]> var_18835_begin_0 = const()[name = tensor<string, []>("op_18835_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_18835_end_0 = const()[name = tensor<string, []>("op_18835_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_18835_end_mask_0 = const()[name = tensor<string, []>("op_18835_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18835_cast = slice_by_index(begin = var_18835_begin_0, end = var_18835_end_0, end_mask = var_18835_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18835_cast")];
            tensor<int32, [4]> var_18839_begin_0 = const()[name = tensor<string, []>("op_18839_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_18839_end_0 = const()[name = tensor<string, []>("op_18839_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_18839_end_mask_0 = const()[name = tensor<string, []>("op_18839_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18839_cast = slice_by_index(begin = var_18839_begin_0, end = var_18839_end_0, end_mask = var_18839_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18839_cast")];
            tensor<int32, [4]> var_18843_begin_0 = const()[name = tensor<string, []>("op_18843_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_18843_end_0 = const()[name = tensor<string, []>("op_18843_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_18843_end_mask_0 = const()[name = tensor<string, []>("op_18843_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18843_cast = slice_by_index(begin = var_18843_begin_0, end = var_18843_end_0, end_mask = var_18843_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18843_cast")];
            tensor<int32, [4]> var_18847_begin_0 = const()[name = tensor<string, []>("op_18847_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_18847_end_0 = const()[name = tensor<string, []>("op_18847_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_18847_end_mask_0 = const()[name = tensor<string, []>("op_18847_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18847_cast = slice_by_index(begin = var_18847_begin_0, end = var_18847_end_0, end_mask = var_18847_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18847_cast")];
            tensor<int32, [4]> var_18851_begin_0 = const()[name = tensor<string, []>("op_18851_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_18851_end_0 = const()[name = tensor<string, []>("op_18851_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_18851_end_mask_0 = const()[name = tensor<string, []>("op_18851_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18851_cast = slice_by_index(begin = var_18851_begin_0, end = var_18851_end_0, end_mask = var_18851_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18851_cast")];
            tensor<int32, [4]> var_18855_begin_0 = const()[name = tensor<string, []>("op_18855_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_18855_end_0 = const()[name = tensor<string, []>("op_18855_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_18855_end_mask_0 = const()[name = tensor<string, []>("op_18855_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18855_cast = slice_by_index(begin = var_18855_begin_0, end = var_18855_end_0, end_mask = var_18855_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18855_cast")];
            tensor<int32, [4]> var_18859_begin_0 = const()[name = tensor<string, []>("op_18859_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_18859_end_0 = const()[name = tensor<string, []>("op_18859_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_18859_end_mask_0 = const()[name = tensor<string, []>("op_18859_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18859_cast = slice_by_index(begin = var_18859_begin_0, end = var_18859_end_0, end_mask = var_18859_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18859_cast")];
            tensor<int32, [4]> var_18863_begin_0 = const()[name = tensor<string, []>("op_18863_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_18863_end_0 = const()[name = tensor<string, []>("op_18863_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_18863_end_mask_0 = const()[name = tensor<string, []>("op_18863_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_18863_cast = slice_by_index(begin = var_18863_begin_0, end = var_18863_end_0, end_mask = var_18863_end_mask_0, x = q_87_cast)[name = tensor<string, []>("op_18863_cast")];
            tensor<int32, [4]> k_175_perm_0 = const()[name = tensor<string, []>("k_175_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_18870_begin_0 = const()[name = tensor<string, []>("op_18870_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_18870_end_0 = const()[name = tensor<string, []>("op_18870_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_18870_end_mask_0 = const()[name = tensor<string, []>("op_18870_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_96 = transpose(perm = k_175_perm_0, x = k_173_cast)[name = tensor<string, []>("transpose_96")];
            tensor<fp16, [2, 77, 1, 64]> var_18870_cast = slice_by_index(begin = var_18870_begin_0, end = var_18870_end_0, end_mask = var_18870_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18870_cast")];
            tensor<int32, [4]> var_18874_begin_0 = const()[name = tensor<string, []>("op_18874_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_18874_end_0 = const()[name = tensor<string, []>("op_18874_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_18874_end_mask_0 = const()[name = tensor<string, []>("op_18874_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18874_cast = slice_by_index(begin = var_18874_begin_0, end = var_18874_end_0, end_mask = var_18874_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18874_cast")];
            tensor<int32, [4]> var_18878_begin_0 = const()[name = tensor<string, []>("op_18878_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_18878_end_0 = const()[name = tensor<string, []>("op_18878_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_18878_end_mask_0 = const()[name = tensor<string, []>("op_18878_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18878_cast = slice_by_index(begin = var_18878_begin_0, end = var_18878_end_0, end_mask = var_18878_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18878_cast")];
            tensor<int32, [4]> var_18882_begin_0 = const()[name = tensor<string, []>("op_18882_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_18882_end_0 = const()[name = tensor<string, []>("op_18882_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_18882_end_mask_0 = const()[name = tensor<string, []>("op_18882_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18882_cast = slice_by_index(begin = var_18882_begin_0, end = var_18882_end_0, end_mask = var_18882_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18882_cast")];
            tensor<int32, [4]> var_18886_begin_0 = const()[name = tensor<string, []>("op_18886_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_18886_end_0 = const()[name = tensor<string, []>("op_18886_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_18886_end_mask_0 = const()[name = tensor<string, []>("op_18886_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18886_cast = slice_by_index(begin = var_18886_begin_0, end = var_18886_end_0, end_mask = var_18886_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18886_cast")];
            tensor<int32, [4]> var_18890_begin_0 = const()[name = tensor<string, []>("op_18890_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_18890_end_0 = const()[name = tensor<string, []>("op_18890_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_18890_end_mask_0 = const()[name = tensor<string, []>("op_18890_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18890_cast = slice_by_index(begin = var_18890_begin_0, end = var_18890_end_0, end_mask = var_18890_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18890_cast")];
            tensor<int32, [4]> var_18894_begin_0 = const()[name = tensor<string, []>("op_18894_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_18894_end_0 = const()[name = tensor<string, []>("op_18894_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_18894_end_mask_0 = const()[name = tensor<string, []>("op_18894_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18894_cast = slice_by_index(begin = var_18894_begin_0, end = var_18894_end_0, end_mask = var_18894_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18894_cast")];
            tensor<int32, [4]> var_18898_begin_0 = const()[name = tensor<string, []>("op_18898_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_18898_end_0 = const()[name = tensor<string, []>("op_18898_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_18898_end_mask_0 = const()[name = tensor<string, []>("op_18898_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18898_cast = slice_by_index(begin = var_18898_begin_0, end = var_18898_end_0, end_mask = var_18898_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18898_cast")];
            tensor<int32, [4]> var_18902_begin_0 = const()[name = tensor<string, []>("op_18902_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_18902_end_0 = const()[name = tensor<string, []>("op_18902_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_18902_end_mask_0 = const()[name = tensor<string, []>("op_18902_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18902_cast = slice_by_index(begin = var_18902_begin_0, end = var_18902_end_0, end_mask = var_18902_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18902_cast")];
            tensor<int32, [4]> var_18906_begin_0 = const()[name = tensor<string, []>("op_18906_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_18906_end_0 = const()[name = tensor<string, []>("op_18906_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_18906_end_mask_0 = const()[name = tensor<string, []>("op_18906_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18906_cast = slice_by_index(begin = var_18906_begin_0, end = var_18906_end_0, end_mask = var_18906_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18906_cast")];
            tensor<int32, [4]> var_18910_begin_0 = const()[name = tensor<string, []>("op_18910_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_18910_end_0 = const()[name = tensor<string, []>("op_18910_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_18910_end_mask_0 = const()[name = tensor<string, []>("op_18910_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18910_cast = slice_by_index(begin = var_18910_begin_0, end = var_18910_end_0, end_mask = var_18910_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18910_cast")];
            tensor<int32, [4]> var_18914_begin_0 = const()[name = tensor<string, []>("op_18914_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_18914_end_0 = const()[name = tensor<string, []>("op_18914_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_18914_end_mask_0 = const()[name = tensor<string, []>("op_18914_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18914_cast = slice_by_index(begin = var_18914_begin_0, end = var_18914_end_0, end_mask = var_18914_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18914_cast")];
            tensor<int32, [4]> var_18918_begin_0 = const()[name = tensor<string, []>("op_18918_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_18918_end_0 = const()[name = tensor<string, []>("op_18918_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_18918_end_mask_0 = const()[name = tensor<string, []>("op_18918_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18918_cast = slice_by_index(begin = var_18918_begin_0, end = var_18918_end_0, end_mask = var_18918_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18918_cast")];
            tensor<int32, [4]> var_18922_begin_0 = const()[name = tensor<string, []>("op_18922_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_18922_end_0 = const()[name = tensor<string, []>("op_18922_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_18922_end_mask_0 = const()[name = tensor<string, []>("op_18922_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18922_cast = slice_by_index(begin = var_18922_begin_0, end = var_18922_end_0, end_mask = var_18922_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18922_cast")];
            tensor<int32, [4]> var_18926_begin_0 = const()[name = tensor<string, []>("op_18926_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_18926_end_0 = const()[name = tensor<string, []>("op_18926_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_18926_end_mask_0 = const()[name = tensor<string, []>("op_18926_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18926_cast = slice_by_index(begin = var_18926_begin_0, end = var_18926_end_0, end_mask = var_18926_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18926_cast")];
            tensor<int32, [4]> var_18930_begin_0 = const()[name = tensor<string, []>("op_18930_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_18930_end_0 = const()[name = tensor<string, []>("op_18930_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_18930_end_mask_0 = const()[name = tensor<string, []>("op_18930_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18930_cast = slice_by_index(begin = var_18930_begin_0, end = var_18930_end_0, end_mask = var_18930_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18930_cast")];
            tensor<int32, [4]> var_18934_begin_0 = const()[name = tensor<string, []>("op_18934_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_18934_end_0 = const()[name = tensor<string, []>("op_18934_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_18934_end_mask_0 = const()[name = tensor<string, []>("op_18934_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18934_cast = slice_by_index(begin = var_18934_begin_0, end = var_18934_end_0, end_mask = var_18934_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18934_cast")];
            tensor<int32, [4]> var_18938_begin_0 = const()[name = tensor<string, []>("op_18938_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_18938_end_0 = const()[name = tensor<string, []>("op_18938_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_18938_end_mask_0 = const()[name = tensor<string, []>("op_18938_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18938_cast = slice_by_index(begin = var_18938_begin_0, end = var_18938_end_0, end_mask = var_18938_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18938_cast")];
            tensor<int32, [4]> var_18942_begin_0 = const()[name = tensor<string, []>("op_18942_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_18942_end_0 = const()[name = tensor<string, []>("op_18942_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_18942_end_mask_0 = const()[name = tensor<string, []>("op_18942_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18942_cast = slice_by_index(begin = var_18942_begin_0, end = var_18942_end_0, end_mask = var_18942_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18942_cast")];
            tensor<int32, [4]> var_18946_begin_0 = const()[name = tensor<string, []>("op_18946_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_18946_end_0 = const()[name = tensor<string, []>("op_18946_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_18946_end_mask_0 = const()[name = tensor<string, []>("op_18946_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_18946_cast = slice_by_index(begin = var_18946_begin_0, end = var_18946_end_0, end_mask = var_18946_end_mask_0, x = transpose_96)[name = tensor<string, []>("op_18946_cast")];
            tensor<int32, [4]> var_18948_begin_0 = const()[name = tensor<string, []>("op_18948_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_18948_end_0 = const()[name = tensor<string, []>("op_18948_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_18948_end_mask_0 = const()[name = tensor<string, []>("op_18948_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18948_cast = slice_by_index(begin = var_18948_begin_0, end = var_18948_end_0, end_mask = var_18948_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_18948_cast")];
            tensor<int32, [4]> var_18952_begin_0 = const()[name = tensor<string, []>("op_18952_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_18952_end_0 = const()[name = tensor<string, []>("op_18952_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_18952_end_mask_0 = const()[name = tensor<string, []>("op_18952_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18952_cast = slice_by_index(begin = var_18952_begin_0, end = var_18952_end_0, end_mask = var_18952_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_18952_cast")];
            tensor<int32, [4]> var_18956_begin_0 = const()[name = tensor<string, []>("op_18956_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_18956_end_0 = const()[name = tensor<string, []>("op_18956_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_18956_end_mask_0 = const()[name = tensor<string, []>("op_18956_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18956_cast = slice_by_index(begin = var_18956_begin_0, end = var_18956_end_0, end_mask = var_18956_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_18956_cast")];
            tensor<int32, [4]> var_18960_begin_0 = const()[name = tensor<string, []>("op_18960_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_18960_end_0 = const()[name = tensor<string, []>("op_18960_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_18960_end_mask_0 = const()[name = tensor<string, []>("op_18960_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18960_cast = slice_by_index(begin = var_18960_begin_0, end = var_18960_end_0, end_mask = var_18960_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_18960_cast")];
            tensor<int32, [4]> var_18964_begin_0 = const()[name = tensor<string, []>("op_18964_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_18964_end_0 = const()[name = tensor<string, []>("op_18964_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_18964_end_mask_0 = const()[name = tensor<string, []>("op_18964_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18964_cast = slice_by_index(begin = var_18964_begin_0, end = var_18964_end_0, end_mask = var_18964_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_18964_cast")];
            tensor<int32, [4]> var_18968_begin_0 = const()[name = tensor<string, []>("op_18968_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_18968_end_0 = const()[name = tensor<string, []>("op_18968_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_18968_end_mask_0 = const()[name = tensor<string, []>("op_18968_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18968_cast = slice_by_index(begin = var_18968_begin_0, end = var_18968_end_0, end_mask = var_18968_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_18968_cast")];
            tensor<int32, [4]> var_18972_begin_0 = const()[name = tensor<string, []>("op_18972_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_18972_end_0 = const()[name = tensor<string, []>("op_18972_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_18972_end_mask_0 = const()[name = tensor<string, []>("op_18972_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18972_cast = slice_by_index(begin = var_18972_begin_0, end = var_18972_end_0, end_mask = var_18972_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_18972_cast")];
            tensor<int32, [4]> var_18976_begin_0 = const()[name = tensor<string, []>("op_18976_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_18976_end_0 = const()[name = tensor<string, []>("op_18976_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_18976_end_mask_0 = const()[name = tensor<string, []>("op_18976_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18976_cast = slice_by_index(begin = var_18976_begin_0, end = var_18976_end_0, end_mask = var_18976_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_18976_cast")];
            tensor<int32, [4]> var_18980_begin_0 = const()[name = tensor<string, []>("op_18980_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_18980_end_0 = const()[name = tensor<string, []>("op_18980_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_18980_end_mask_0 = const()[name = tensor<string, []>("op_18980_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18980_cast = slice_by_index(begin = var_18980_begin_0, end = var_18980_end_0, end_mask = var_18980_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_18980_cast")];
            tensor<int32, [4]> var_18984_begin_0 = const()[name = tensor<string, []>("op_18984_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_18984_end_0 = const()[name = tensor<string, []>("op_18984_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_18984_end_mask_0 = const()[name = tensor<string, []>("op_18984_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18984_cast = slice_by_index(begin = var_18984_begin_0, end = var_18984_end_0, end_mask = var_18984_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_18984_cast")];
            tensor<int32, [4]> var_18988_begin_0 = const()[name = tensor<string, []>("op_18988_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_18988_end_0 = const()[name = tensor<string, []>("op_18988_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_18988_end_mask_0 = const()[name = tensor<string, []>("op_18988_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18988_cast = slice_by_index(begin = var_18988_begin_0, end = var_18988_end_0, end_mask = var_18988_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_18988_cast")];
            tensor<int32, [4]> var_18992_begin_0 = const()[name = tensor<string, []>("op_18992_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_18992_end_0 = const()[name = tensor<string, []>("op_18992_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_18992_end_mask_0 = const()[name = tensor<string, []>("op_18992_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18992_cast = slice_by_index(begin = var_18992_begin_0, end = var_18992_end_0, end_mask = var_18992_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_18992_cast")];
            tensor<int32, [4]> var_18996_begin_0 = const()[name = tensor<string, []>("op_18996_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_18996_end_0 = const()[name = tensor<string, []>("op_18996_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_18996_end_mask_0 = const()[name = tensor<string, []>("op_18996_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_18996_cast = slice_by_index(begin = var_18996_begin_0, end = var_18996_end_0, end_mask = var_18996_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_18996_cast")];
            tensor<int32, [4]> var_19000_begin_0 = const()[name = tensor<string, []>("op_19000_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_19000_end_0 = const()[name = tensor<string, []>("op_19000_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_19000_end_mask_0 = const()[name = tensor<string, []>("op_19000_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19000_cast = slice_by_index(begin = var_19000_begin_0, end = var_19000_end_0, end_mask = var_19000_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_19000_cast")];
            tensor<int32, [4]> var_19004_begin_0 = const()[name = tensor<string, []>("op_19004_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_19004_end_0 = const()[name = tensor<string, []>("op_19004_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_19004_end_mask_0 = const()[name = tensor<string, []>("op_19004_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19004_cast = slice_by_index(begin = var_19004_begin_0, end = var_19004_end_0, end_mask = var_19004_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_19004_cast")];
            tensor<int32, [4]> var_19008_begin_0 = const()[name = tensor<string, []>("op_19008_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_19008_end_0 = const()[name = tensor<string, []>("op_19008_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_19008_end_mask_0 = const()[name = tensor<string, []>("op_19008_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19008_cast = slice_by_index(begin = var_19008_begin_0, end = var_19008_end_0, end_mask = var_19008_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_19008_cast")];
            tensor<int32, [4]> var_19012_begin_0 = const()[name = tensor<string, []>("op_19012_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_19012_end_0 = const()[name = tensor<string, []>("op_19012_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_19012_end_mask_0 = const()[name = tensor<string, []>("op_19012_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19012_cast = slice_by_index(begin = var_19012_begin_0, end = var_19012_end_0, end_mask = var_19012_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_19012_cast")];
            tensor<int32, [4]> var_19016_begin_0 = const()[name = tensor<string, []>("op_19016_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_19016_end_0 = const()[name = tensor<string, []>("op_19016_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_19016_end_mask_0 = const()[name = tensor<string, []>("op_19016_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19016_cast = slice_by_index(begin = var_19016_begin_0, end = var_19016_end_0, end_mask = var_19016_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_19016_cast")];
            tensor<int32, [4]> var_19020_begin_0 = const()[name = tensor<string, []>("op_19020_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_19020_end_0 = const()[name = tensor<string, []>("op_19020_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_19020_end_mask_0 = const()[name = tensor<string, []>("op_19020_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19020_cast = slice_by_index(begin = var_19020_begin_0, end = var_19020_end_0, end_mask = var_19020_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_19020_cast")];
            tensor<int32, [4]> var_19024_begin_0 = const()[name = tensor<string, []>("op_19024_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_19024_end_0 = const()[name = tensor<string, []>("op_19024_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_19024_end_mask_0 = const()[name = tensor<string, []>("op_19024_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19024_cast = slice_by_index(begin = var_19024_begin_0, end = var_19024_end_0, end_mask = var_19024_end_mask_0, x = v_87_cast)[name = tensor<string, []>("op_19024_cast")];
            tensor<string, []> var_19028_equation_0 = const()[name = tensor<string, []>("op_19028_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19028_cast = einsum(equation = var_19028_equation_0, values = (var_18870_cast, var_18787_cast))[name = tensor<string, []>("op_19028_cast")];
            tensor<fp16, []> var_19029_to_fp16 = const()[name = tensor<string, []>("op_19029_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1561_cast = mul(x = var_19028_cast, y = var_19029_to_fp16)[name = tensor<string, []>("aw_1561_cast")];
            tensor<string, []> var_19032_equation_0 = const()[name = tensor<string, []>("op_19032_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19032_cast = einsum(equation = var_19032_equation_0, values = (var_18874_cast, var_18791_cast))[name = tensor<string, []>("op_19032_cast")];
            tensor<fp16, []> var_19033_to_fp16 = const()[name = tensor<string, []>("op_19033_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1563_cast = mul(x = var_19032_cast, y = var_19033_to_fp16)[name = tensor<string, []>("aw_1563_cast")];
            tensor<string, []> var_19036_equation_0 = const()[name = tensor<string, []>("op_19036_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19036_cast = einsum(equation = var_19036_equation_0, values = (var_18878_cast, var_18795_cast))[name = tensor<string, []>("op_19036_cast")];
            tensor<fp16, []> var_19037_to_fp16 = const()[name = tensor<string, []>("op_19037_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1565_cast = mul(x = var_19036_cast, y = var_19037_to_fp16)[name = tensor<string, []>("aw_1565_cast")];
            tensor<string, []> var_19040_equation_0 = const()[name = tensor<string, []>("op_19040_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19040_cast = einsum(equation = var_19040_equation_0, values = (var_18882_cast, var_18799_cast))[name = tensor<string, []>("op_19040_cast")];
            tensor<fp16, []> var_19041_to_fp16 = const()[name = tensor<string, []>("op_19041_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1567_cast = mul(x = var_19040_cast, y = var_19041_to_fp16)[name = tensor<string, []>("aw_1567_cast")];
            tensor<string, []> var_19044_equation_0 = const()[name = tensor<string, []>("op_19044_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19044_cast = einsum(equation = var_19044_equation_0, values = (var_18886_cast, var_18803_cast))[name = tensor<string, []>("op_19044_cast")];
            tensor<fp16, []> var_19045_to_fp16 = const()[name = tensor<string, []>("op_19045_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1569_cast = mul(x = var_19044_cast, y = var_19045_to_fp16)[name = tensor<string, []>("aw_1569_cast")];
            tensor<string, []> var_19048_equation_0 = const()[name = tensor<string, []>("op_19048_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19048_cast = einsum(equation = var_19048_equation_0, values = (var_18890_cast, var_18807_cast))[name = tensor<string, []>("op_19048_cast")];
            tensor<fp16, []> var_19049_to_fp16 = const()[name = tensor<string, []>("op_19049_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1571_cast = mul(x = var_19048_cast, y = var_19049_to_fp16)[name = tensor<string, []>("aw_1571_cast")];
            tensor<string, []> var_19052_equation_0 = const()[name = tensor<string, []>("op_19052_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19052_cast = einsum(equation = var_19052_equation_0, values = (var_18894_cast, var_18811_cast))[name = tensor<string, []>("op_19052_cast")];
            tensor<fp16, []> var_19053_to_fp16 = const()[name = tensor<string, []>("op_19053_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1573_cast = mul(x = var_19052_cast, y = var_19053_to_fp16)[name = tensor<string, []>("aw_1573_cast")];
            tensor<string, []> var_19056_equation_0 = const()[name = tensor<string, []>("op_19056_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19056_cast = einsum(equation = var_19056_equation_0, values = (var_18898_cast, var_18815_cast))[name = tensor<string, []>("op_19056_cast")];
            tensor<fp16, []> var_19057_to_fp16 = const()[name = tensor<string, []>("op_19057_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1575_cast = mul(x = var_19056_cast, y = var_19057_to_fp16)[name = tensor<string, []>("aw_1575_cast")];
            tensor<string, []> var_19060_equation_0 = const()[name = tensor<string, []>("op_19060_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19060_cast = einsum(equation = var_19060_equation_0, values = (var_18902_cast, var_18819_cast))[name = tensor<string, []>("op_19060_cast")];
            tensor<fp16, []> var_19061_to_fp16 = const()[name = tensor<string, []>("op_19061_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1577_cast = mul(x = var_19060_cast, y = var_19061_to_fp16)[name = tensor<string, []>("aw_1577_cast")];
            tensor<string, []> var_19064_equation_0 = const()[name = tensor<string, []>("op_19064_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19064_cast = einsum(equation = var_19064_equation_0, values = (var_18906_cast, var_18823_cast))[name = tensor<string, []>("op_19064_cast")];
            tensor<fp16, []> var_19065_to_fp16 = const()[name = tensor<string, []>("op_19065_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1579_cast = mul(x = var_19064_cast, y = var_19065_to_fp16)[name = tensor<string, []>("aw_1579_cast")];
            tensor<string, []> var_19068_equation_0 = const()[name = tensor<string, []>("op_19068_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19068_cast = einsum(equation = var_19068_equation_0, values = (var_18910_cast, var_18827_cast))[name = tensor<string, []>("op_19068_cast")];
            tensor<fp16, []> var_19069_to_fp16 = const()[name = tensor<string, []>("op_19069_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1581_cast = mul(x = var_19068_cast, y = var_19069_to_fp16)[name = tensor<string, []>("aw_1581_cast")];
            tensor<string, []> var_19072_equation_0 = const()[name = tensor<string, []>("op_19072_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19072_cast = einsum(equation = var_19072_equation_0, values = (var_18914_cast, var_18831_cast))[name = tensor<string, []>("op_19072_cast")];
            tensor<fp16, []> var_19073_to_fp16 = const()[name = tensor<string, []>("op_19073_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1583_cast = mul(x = var_19072_cast, y = var_19073_to_fp16)[name = tensor<string, []>("aw_1583_cast")];
            tensor<string, []> var_19076_equation_0 = const()[name = tensor<string, []>("op_19076_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19076_cast = einsum(equation = var_19076_equation_0, values = (var_18918_cast, var_18835_cast))[name = tensor<string, []>("op_19076_cast")];
            tensor<fp16, []> var_19077_to_fp16 = const()[name = tensor<string, []>("op_19077_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1585_cast = mul(x = var_19076_cast, y = var_19077_to_fp16)[name = tensor<string, []>("aw_1585_cast")];
            tensor<string, []> var_19080_equation_0 = const()[name = tensor<string, []>("op_19080_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19080_cast = einsum(equation = var_19080_equation_0, values = (var_18922_cast, var_18839_cast))[name = tensor<string, []>("op_19080_cast")];
            tensor<fp16, []> var_19081_to_fp16 = const()[name = tensor<string, []>("op_19081_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1587_cast = mul(x = var_19080_cast, y = var_19081_to_fp16)[name = tensor<string, []>("aw_1587_cast")];
            tensor<string, []> var_19084_equation_0 = const()[name = tensor<string, []>("op_19084_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19084_cast = einsum(equation = var_19084_equation_0, values = (var_18926_cast, var_18843_cast))[name = tensor<string, []>("op_19084_cast")];
            tensor<fp16, []> var_19085_to_fp16 = const()[name = tensor<string, []>("op_19085_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1589_cast = mul(x = var_19084_cast, y = var_19085_to_fp16)[name = tensor<string, []>("aw_1589_cast")];
            tensor<string, []> var_19088_equation_0 = const()[name = tensor<string, []>("op_19088_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19088_cast = einsum(equation = var_19088_equation_0, values = (var_18930_cast, var_18847_cast))[name = tensor<string, []>("op_19088_cast")];
            tensor<fp16, []> var_19089_to_fp16 = const()[name = tensor<string, []>("op_19089_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1591_cast = mul(x = var_19088_cast, y = var_19089_to_fp16)[name = tensor<string, []>("aw_1591_cast")];
            tensor<string, []> var_19092_equation_0 = const()[name = tensor<string, []>("op_19092_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19092_cast = einsum(equation = var_19092_equation_0, values = (var_18934_cast, var_18851_cast))[name = tensor<string, []>("op_19092_cast")];
            tensor<fp16, []> var_19093_to_fp16 = const()[name = tensor<string, []>("op_19093_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1593_cast = mul(x = var_19092_cast, y = var_19093_to_fp16)[name = tensor<string, []>("aw_1593_cast")];
            tensor<string, []> var_19096_equation_0 = const()[name = tensor<string, []>("op_19096_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19096_cast = einsum(equation = var_19096_equation_0, values = (var_18938_cast, var_18855_cast))[name = tensor<string, []>("op_19096_cast")];
            tensor<fp16, []> var_19097_to_fp16 = const()[name = tensor<string, []>("op_19097_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1595_cast = mul(x = var_19096_cast, y = var_19097_to_fp16)[name = tensor<string, []>("aw_1595_cast")];
            tensor<string, []> var_19100_equation_0 = const()[name = tensor<string, []>("op_19100_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19100_cast = einsum(equation = var_19100_equation_0, values = (var_18942_cast, var_18859_cast))[name = tensor<string, []>("op_19100_cast")];
            tensor<fp16, []> var_19101_to_fp16 = const()[name = tensor<string, []>("op_19101_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1597_cast = mul(x = var_19100_cast, y = var_19101_to_fp16)[name = tensor<string, []>("aw_1597_cast")];
            tensor<string, []> var_19104_equation_0 = const()[name = tensor<string, []>("op_19104_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19104_cast = einsum(equation = var_19104_equation_0, values = (var_18946_cast, var_18863_cast))[name = tensor<string, []>("op_19104_cast")];
            tensor<fp16, []> var_19105_to_fp16 = const()[name = tensor<string, []>("op_19105_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1599_cast = mul(x = var_19104_cast, y = var_19105_to_fp16)[name = tensor<string, []>("aw_1599_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19107_cast = softmax(axis = var_2634, x = aw_1561_cast)[name = tensor<string, []>("op_19107_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19108_cast = softmax(axis = var_2634, x = aw_1563_cast)[name = tensor<string, []>("op_19108_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19109_cast = softmax(axis = var_2634, x = aw_1565_cast)[name = tensor<string, []>("op_19109_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19110_cast = softmax(axis = var_2634, x = aw_1567_cast)[name = tensor<string, []>("op_19110_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19111_cast = softmax(axis = var_2634, x = aw_1569_cast)[name = tensor<string, []>("op_19111_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19112_cast = softmax(axis = var_2634, x = aw_1571_cast)[name = tensor<string, []>("op_19112_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19113_cast = softmax(axis = var_2634, x = aw_1573_cast)[name = tensor<string, []>("op_19113_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19114_cast = softmax(axis = var_2634, x = aw_1575_cast)[name = tensor<string, []>("op_19114_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19115_cast = softmax(axis = var_2634, x = aw_1577_cast)[name = tensor<string, []>("op_19115_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19116_cast = softmax(axis = var_2634, x = aw_1579_cast)[name = tensor<string, []>("op_19116_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19117_cast = softmax(axis = var_2634, x = aw_1581_cast)[name = tensor<string, []>("op_19117_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19118_cast = softmax(axis = var_2634, x = aw_1583_cast)[name = tensor<string, []>("op_19118_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19119_cast = softmax(axis = var_2634, x = aw_1585_cast)[name = tensor<string, []>("op_19119_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19120_cast = softmax(axis = var_2634, x = aw_1587_cast)[name = tensor<string, []>("op_19120_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19121_cast = softmax(axis = var_2634, x = aw_1589_cast)[name = tensor<string, []>("op_19121_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19122_cast = softmax(axis = var_2634, x = aw_1591_cast)[name = tensor<string, []>("op_19122_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19123_cast = softmax(axis = var_2634, x = aw_1593_cast)[name = tensor<string, []>("op_19123_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19124_cast = softmax(axis = var_2634, x = aw_1595_cast)[name = tensor<string, []>("op_19124_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19125_cast = softmax(axis = var_2634, x = aw_1597_cast)[name = tensor<string, []>("op_19125_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_19126_cast = softmax(axis = var_2634, x = aw_1599_cast)[name = tensor<string, []>("op_19126_cast")];
            tensor<string, []> var_19128_equation_0 = const()[name = tensor<string, []>("op_19128_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19128_cast = einsum(equation = var_19128_equation_0, values = (var_18948_cast, var_19107_cast))[name = tensor<string, []>("op_19128_cast")];
            tensor<string, []> var_19130_equation_0 = const()[name = tensor<string, []>("op_19130_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19130_cast = einsum(equation = var_19130_equation_0, values = (var_18952_cast, var_19108_cast))[name = tensor<string, []>("op_19130_cast")];
            tensor<string, []> var_19132_equation_0 = const()[name = tensor<string, []>("op_19132_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19132_cast = einsum(equation = var_19132_equation_0, values = (var_18956_cast, var_19109_cast))[name = tensor<string, []>("op_19132_cast")];
            tensor<string, []> var_19134_equation_0 = const()[name = tensor<string, []>("op_19134_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19134_cast = einsum(equation = var_19134_equation_0, values = (var_18960_cast, var_19110_cast))[name = tensor<string, []>("op_19134_cast")];
            tensor<string, []> var_19136_equation_0 = const()[name = tensor<string, []>("op_19136_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19136_cast = einsum(equation = var_19136_equation_0, values = (var_18964_cast, var_19111_cast))[name = tensor<string, []>("op_19136_cast")];
            tensor<string, []> var_19138_equation_0 = const()[name = tensor<string, []>("op_19138_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19138_cast = einsum(equation = var_19138_equation_0, values = (var_18968_cast, var_19112_cast))[name = tensor<string, []>("op_19138_cast")];
            tensor<string, []> var_19140_equation_0 = const()[name = tensor<string, []>("op_19140_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19140_cast = einsum(equation = var_19140_equation_0, values = (var_18972_cast, var_19113_cast))[name = tensor<string, []>("op_19140_cast")];
            tensor<string, []> var_19142_equation_0 = const()[name = tensor<string, []>("op_19142_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19142_cast = einsum(equation = var_19142_equation_0, values = (var_18976_cast, var_19114_cast))[name = tensor<string, []>("op_19142_cast")];
            tensor<string, []> var_19144_equation_0 = const()[name = tensor<string, []>("op_19144_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19144_cast = einsum(equation = var_19144_equation_0, values = (var_18980_cast, var_19115_cast))[name = tensor<string, []>("op_19144_cast")];
            tensor<string, []> var_19146_equation_0 = const()[name = tensor<string, []>("op_19146_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19146_cast = einsum(equation = var_19146_equation_0, values = (var_18984_cast, var_19116_cast))[name = tensor<string, []>("op_19146_cast")];
            tensor<string, []> var_19148_equation_0 = const()[name = tensor<string, []>("op_19148_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19148_cast = einsum(equation = var_19148_equation_0, values = (var_18988_cast, var_19117_cast))[name = tensor<string, []>("op_19148_cast")];
            tensor<string, []> var_19150_equation_0 = const()[name = tensor<string, []>("op_19150_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19150_cast = einsum(equation = var_19150_equation_0, values = (var_18992_cast, var_19118_cast))[name = tensor<string, []>("op_19150_cast")];
            tensor<string, []> var_19152_equation_0 = const()[name = tensor<string, []>("op_19152_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19152_cast = einsum(equation = var_19152_equation_0, values = (var_18996_cast, var_19119_cast))[name = tensor<string, []>("op_19152_cast")];
            tensor<string, []> var_19154_equation_0 = const()[name = tensor<string, []>("op_19154_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19154_cast = einsum(equation = var_19154_equation_0, values = (var_19000_cast, var_19120_cast))[name = tensor<string, []>("op_19154_cast")];
            tensor<string, []> var_19156_equation_0 = const()[name = tensor<string, []>("op_19156_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19156_cast = einsum(equation = var_19156_equation_0, values = (var_19004_cast, var_19121_cast))[name = tensor<string, []>("op_19156_cast")];
            tensor<string, []> var_19158_equation_0 = const()[name = tensor<string, []>("op_19158_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19158_cast = einsum(equation = var_19158_equation_0, values = (var_19008_cast, var_19122_cast))[name = tensor<string, []>("op_19158_cast")];
            tensor<string, []> var_19160_equation_0 = const()[name = tensor<string, []>("op_19160_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19160_cast = einsum(equation = var_19160_equation_0, values = (var_19012_cast, var_19123_cast))[name = tensor<string, []>("op_19160_cast")];
            tensor<string, []> var_19162_equation_0 = const()[name = tensor<string, []>("op_19162_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19162_cast = einsum(equation = var_19162_equation_0, values = (var_19016_cast, var_19124_cast))[name = tensor<string, []>("op_19162_cast")];
            tensor<string, []> var_19164_equation_0 = const()[name = tensor<string, []>("op_19164_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19164_cast = einsum(equation = var_19164_equation_0, values = (var_19020_cast, var_19125_cast))[name = tensor<string, []>("op_19164_cast")];
            tensor<string, []> var_19166_equation_0 = const()[name = tensor<string, []>("op_19166_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19166_cast = einsum(equation = var_19166_equation_0, values = (var_19024_cast, var_19126_cast))[name = tensor<string, []>("op_19166_cast")];
            tensor<bool, []> input_291_interleave_0 = const()[name = tensor<string, []>("input_291_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_291_cast = concat(axis = var_2634, interleave = input_291_interleave_0, values = (var_19128_cast, var_19130_cast, var_19132_cast, var_19134_cast, var_19136_cast, var_19138_cast, var_19140_cast, var_19142_cast, var_19144_cast, var_19146_cast, var_19148_cast, var_19150_cast, var_19152_cast, var_19154_cast, var_19156_cast, var_19158_cast, var_19160_cast, var_19162_cast, var_19164_cast, var_19166_cast))[name = tensor<string, []>("input_291_cast")];
            tensor<int32, [2]> var_19172 = const()[name = tensor<string, []>("op_19172"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_19174 = const()[name = tensor<string, []>("op_19174"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_19176_pad_type_0 = const()[name = tensor<string, []>("op_19176_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_19176_pad_0 = const()[name = tensor<string, []>("op_19176_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1475611520)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1478888384)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_19176_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_out_0_bias_to_fp16, dilations = var_19174, groups = var_2634, pad = var_19176_pad_0, pad_type = var_19176_pad_type_0, strides = var_19172, weight = down_blocks_2_attentions_1_transformer_blocks_7_attn2_to_out_0_weight_to_fp16, x = input_291_cast)[name = tensor<string, []>("op_19176_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_131_cast = add(x = var_19176_cast, y = inputs_129_cast)[name = tensor<string, []>("inputs_131_cast")];
            tensor<int32, [1]> var_19180 = const()[name = tensor<string, []>("op_19180"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_131_cast = reduce_mean(axes = var_19180, keep_dims = var_2629, x = inputs_131_cast)[name = tensor<string, []>("channels_mean_131_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_131_cast = sub(x = inputs_131_cast, y = channels_mean_131_cast)[name = tensor<string, []>("zero_mean_131_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_131_cast = mul(x = zero_mean_131_cast, y = zero_mean_131_cast)[name = tensor<string, []>("zero_mean_sq_131_cast")];
            tensor<int32, [1]> var_19184 = const()[name = tensor<string, []>("op_19184"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_19185_cast = reduce_mean(axes = var_19184, keep_dims = var_2629, x = zero_mean_sq_131_cast)[name = tensor<string, []>("op_19185_cast")];
            tensor<fp16, []> var_19186_to_fp16 = const()[name = tensor<string, []>("op_19186_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_19187_cast = add(x = var_19185_cast, y = var_19186_to_fp16)[name = tensor<string, []>("op_19187_cast")];
            tensor<fp16, []> denom_131_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_131_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_131_cast = rsqrt(epsilon = denom_131_epsilon_0_to_fp16, x = var_19187_cast)[name = tensor<string, []>("denom_131_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_131_cast = mul(x = zero_mean_131_cast, y = denom_131_cast)[name = tensor<string, []>("out_131_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_19191_to_fp16 = const()[name = tensor<string, []>("op_19191_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1478891008)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_19192_cast = add(x = out_131_cast, y = var_19191_to_fp16)[name = tensor<string, []>("op_19192_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_19194_to_fp16 = const()[name = tensor<string, []>("op_19194_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1478893632)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_293_cast = mul(x = var_19192_cast, y = var_19194_to_fp16)[name = tensor<string, []>("input_293_cast")];
            tensor<int32, [2]> var_19202 = const()[name = tensor<string, []>("op_19202"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_19204 = const()[name = tensor<string, []>("op_19204"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_19206_pad_type_0 = const()[name = tensor<string, []>("op_19206_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_19206_pad_0 = const()[name = tensor<string, []>("op_19206_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_7_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_7_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1478896256)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_1_transformer_blocks_7_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_7_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1505110720)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_19206_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_7_ff_net_0_proj_bias_to_fp16, dilations = var_19204, groups = var_2634, pad = var_19206_pad_0, pad_type = var_19206_pad_type_0, strides = var_19202, weight = down_blocks_2_attentions_1_transformer_blocks_7_ff_net_0_proj_weight_to_fp16, x = input_293_cast)[name = tensor<string, []>("op_19206_cast")];
            tensor<int32, [2]> var_19207_split_sizes_0 = const()[name = tensor<string, []>("op_19207_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_19207_axis_0 = const()[name = tensor<string, []>("op_19207_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_19207_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_19207_cast_1 = split(axis = var_19207_axis_0, split_sizes = var_19207_split_sizes_0, x = var_19206_cast)[name = tensor<string, []>("op_19207_cast")];
            tensor<string, []> var_19209_mode_0 = const()[name = tensor<string, []>("op_19209_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_19209_cast = gelu(mode = var_19209_mode_0, x = var_19207_cast_1)[name = tensor<string, []>("op_19209_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_295_cast = mul(x = var_19207_cast_0, y = var_19209_cast)[name = tensor<string, []>("input_295_cast")];
            tensor<int32, [2]> var_19213 = const()[name = tensor<string, []>("op_19213"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_19215 = const()[name = tensor<string, []>("op_19215"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_19217_pad_type_0 = const()[name = tensor<string, []>("op_19217_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_19217_pad_0 = const()[name = tensor<string, []>("op_19217_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_7_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_7_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1505131264)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_7_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_7_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1518238528)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_19217_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_7_ff_net_2_bias_to_fp16, dilations = var_19215, groups = var_2634, pad = var_19217_pad_0, pad_type = var_19217_pad_type_0, strides = var_19213, weight = down_blocks_2_attentions_1_transformer_blocks_7_ff_net_2_weight_to_fp16, x = input_295_cast)[name = tensor<string, []>("op_19217_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_133_cast = add(x = var_19217_cast, y = inputs_131_cast)[name = tensor<string, []>("inputs_133_cast")];
            tensor<int32, [1]> var_19227 = const()[name = tensor<string, []>("op_19227"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_133_cast = reduce_mean(axes = var_19227, keep_dims = var_2629, x = inputs_133_cast)[name = tensor<string, []>("channels_mean_133_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_133_cast = sub(x = inputs_133_cast, y = channels_mean_133_cast)[name = tensor<string, []>("zero_mean_133_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_133_cast = mul(x = zero_mean_133_cast, y = zero_mean_133_cast)[name = tensor<string, []>("zero_mean_sq_133_cast")];
            tensor<int32, [1]> var_19231 = const()[name = tensor<string, []>("op_19231"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_19232_cast = reduce_mean(axes = var_19231, keep_dims = var_2629, x = zero_mean_sq_133_cast)[name = tensor<string, []>("op_19232_cast")];
            tensor<fp16, []> var_19233_to_fp16 = const()[name = tensor<string, []>("op_19233_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_19234_cast = add(x = var_19232_cast, y = var_19233_to_fp16)[name = tensor<string, []>("op_19234_cast")];
            tensor<fp16, []> denom_133_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_133_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_133_cast = rsqrt(epsilon = denom_133_epsilon_0_to_fp16, x = var_19234_cast)[name = tensor<string, []>("denom_133_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_133_cast = mul(x = zero_mean_133_cast, y = denom_133_cast)[name = tensor<string, []>("out_133_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_19238_to_fp16 = const()[name = tensor<string, []>("op_19238_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1518241152)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_19239_cast = add(x = out_133_cast, y = var_19238_to_fp16)[name = tensor<string, []>("op_19239_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_19241_to_fp16 = const()[name = tensor<string, []>("op_19241_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1518243776)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_185_cast = mul(x = var_19239_cast, y = var_19241_to_fp16)[name = tensor<string, []>("hidden_states_185_cast")];
            tensor<int32, [2]> var_19248 = const()[name = tensor<string, []>("op_19248"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_19250 = const()[name = tensor<string, []>("op_19250"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_89_pad_type_0 = const()[name = tensor<string, []>("q_89_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_89_pad_0 = const()[name = tensor<string, []>("q_89_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1518246400)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_89_cast = conv(dilations = var_19250, groups = var_2634, pad = q_89_pad_0, pad_type = q_89_pad_type_0, strides = var_19248, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_q_weight_to_fp16, x = hidden_states_185_cast)[name = tensor<string, []>("q_89_cast")];
            tensor<int32, [2]> var_19254 = const()[name = tensor<string, []>("op_19254"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_19256 = const()[name = tensor<string, []>("op_19256"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_177_pad_type_0 = const()[name = tensor<string, []>("k_177_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_177_pad_0 = const()[name = tensor<string, []>("k_177_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1521523264)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_177_cast = conv(dilations = var_19256, groups = var_2634, pad = k_177_pad_0, pad_type = k_177_pad_type_0, strides = var_19254, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_k_weight_to_fp16, x = hidden_states_185_cast)[name = tensor<string, []>("k_177_cast")];
            tensor<int32, [2]> var_19260 = const()[name = tensor<string, []>("op_19260"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_19262 = const()[name = tensor<string, []>("op_19262"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_89_pad_type_0 = const()[name = tensor<string, []>("v_89_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_89_pad_0 = const()[name = tensor<string, []>("v_89_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1524800128)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_89_cast = conv(dilations = var_19262, groups = var_2634, pad = v_89_pad_0, pad_type = v_89_pad_type_0, strides = var_19260, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_v_weight_to_fp16, x = hidden_states_185_cast)[name = tensor<string, []>("v_89_cast")];
            tensor<int32, [4]> var_19266_begin_0 = const()[name = tensor<string, []>("op_19266_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_19266_end_0 = const()[name = tensor<string, []>("op_19266_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_19266_end_mask_0 = const()[name = tensor<string, []>("op_19266_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19266_cast = slice_by_index(begin = var_19266_begin_0, end = var_19266_end_0, end_mask = var_19266_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19266_cast")];
            tensor<int32, [4]> var_19270_begin_0 = const()[name = tensor<string, []>("op_19270_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_19270_end_0 = const()[name = tensor<string, []>("op_19270_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_19270_end_mask_0 = const()[name = tensor<string, []>("op_19270_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19270_cast = slice_by_index(begin = var_19270_begin_0, end = var_19270_end_0, end_mask = var_19270_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19270_cast")];
            tensor<int32, [4]> var_19274_begin_0 = const()[name = tensor<string, []>("op_19274_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_19274_end_0 = const()[name = tensor<string, []>("op_19274_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_19274_end_mask_0 = const()[name = tensor<string, []>("op_19274_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19274_cast = slice_by_index(begin = var_19274_begin_0, end = var_19274_end_0, end_mask = var_19274_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19274_cast")];
            tensor<int32, [4]> var_19278_begin_0 = const()[name = tensor<string, []>("op_19278_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_19278_end_0 = const()[name = tensor<string, []>("op_19278_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_19278_end_mask_0 = const()[name = tensor<string, []>("op_19278_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19278_cast = slice_by_index(begin = var_19278_begin_0, end = var_19278_end_0, end_mask = var_19278_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19278_cast")];
            tensor<int32, [4]> var_19282_begin_0 = const()[name = tensor<string, []>("op_19282_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_19282_end_0 = const()[name = tensor<string, []>("op_19282_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_19282_end_mask_0 = const()[name = tensor<string, []>("op_19282_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19282_cast = slice_by_index(begin = var_19282_begin_0, end = var_19282_end_0, end_mask = var_19282_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19282_cast")];
            tensor<int32, [4]> var_19286_begin_0 = const()[name = tensor<string, []>("op_19286_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_19286_end_0 = const()[name = tensor<string, []>("op_19286_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_19286_end_mask_0 = const()[name = tensor<string, []>("op_19286_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19286_cast = slice_by_index(begin = var_19286_begin_0, end = var_19286_end_0, end_mask = var_19286_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19286_cast")];
            tensor<int32, [4]> var_19290_begin_0 = const()[name = tensor<string, []>("op_19290_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_19290_end_0 = const()[name = tensor<string, []>("op_19290_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_19290_end_mask_0 = const()[name = tensor<string, []>("op_19290_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19290_cast = slice_by_index(begin = var_19290_begin_0, end = var_19290_end_0, end_mask = var_19290_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19290_cast")];
            tensor<int32, [4]> var_19294_begin_0 = const()[name = tensor<string, []>("op_19294_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_19294_end_0 = const()[name = tensor<string, []>("op_19294_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_19294_end_mask_0 = const()[name = tensor<string, []>("op_19294_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19294_cast = slice_by_index(begin = var_19294_begin_0, end = var_19294_end_0, end_mask = var_19294_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19294_cast")];
            tensor<int32, [4]> var_19298_begin_0 = const()[name = tensor<string, []>("op_19298_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_19298_end_0 = const()[name = tensor<string, []>("op_19298_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_19298_end_mask_0 = const()[name = tensor<string, []>("op_19298_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19298_cast = slice_by_index(begin = var_19298_begin_0, end = var_19298_end_0, end_mask = var_19298_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19298_cast")];
            tensor<int32, [4]> var_19302_begin_0 = const()[name = tensor<string, []>("op_19302_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_19302_end_0 = const()[name = tensor<string, []>("op_19302_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_19302_end_mask_0 = const()[name = tensor<string, []>("op_19302_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19302_cast = slice_by_index(begin = var_19302_begin_0, end = var_19302_end_0, end_mask = var_19302_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19302_cast")];
            tensor<int32, [4]> var_19306_begin_0 = const()[name = tensor<string, []>("op_19306_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_19306_end_0 = const()[name = tensor<string, []>("op_19306_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_19306_end_mask_0 = const()[name = tensor<string, []>("op_19306_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19306_cast = slice_by_index(begin = var_19306_begin_0, end = var_19306_end_0, end_mask = var_19306_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19306_cast")];
            tensor<int32, [4]> var_19310_begin_0 = const()[name = tensor<string, []>("op_19310_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_19310_end_0 = const()[name = tensor<string, []>("op_19310_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_19310_end_mask_0 = const()[name = tensor<string, []>("op_19310_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19310_cast = slice_by_index(begin = var_19310_begin_0, end = var_19310_end_0, end_mask = var_19310_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19310_cast")];
            tensor<int32, [4]> var_19314_begin_0 = const()[name = tensor<string, []>("op_19314_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_19314_end_0 = const()[name = tensor<string, []>("op_19314_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_19314_end_mask_0 = const()[name = tensor<string, []>("op_19314_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19314_cast = slice_by_index(begin = var_19314_begin_0, end = var_19314_end_0, end_mask = var_19314_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19314_cast")];
            tensor<int32, [4]> var_19318_begin_0 = const()[name = tensor<string, []>("op_19318_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_19318_end_0 = const()[name = tensor<string, []>("op_19318_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_19318_end_mask_0 = const()[name = tensor<string, []>("op_19318_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19318_cast = slice_by_index(begin = var_19318_begin_0, end = var_19318_end_0, end_mask = var_19318_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19318_cast")];
            tensor<int32, [4]> var_19322_begin_0 = const()[name = tensor<string, []>("op_19322_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_19322_end_0 = const()[name = tensor<string, []>("op_19322_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_19322_end_mask_0 = const()[name = tensor<string, []>("op_19322_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19322_cast = slice_by_index(begin = var_19322_begin_0, end = var_19322_end_0, end_mask = var_19322_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19322_cast")];
            tensor<int32, [4]> var_19326_begin_0 = const()[name = tensor<string, []>("op_19326_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_19326_end_0 = const()[name = tensor<string, []>("op_19326_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_19326_end_mask_0 = const()[name = tensor<string, []>("op_19326_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19326_cast = slice_by_index(begin = var_19326_begin_0, end = var_19326_end_0, end_mask = var_19326_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19326_cast")];
            tensor<int32, [4]> var_19330_begin_0 = const()[name = tensor<string, []>("op_19330_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_19330_end_0 = const()[name = tensor<string, []>("op_19330_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_19330_end_mask_0 = const()[name = tensor<string, []>("op_19330_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19330_cast = slice_by_index(begin = var_19330_begin_0, end = var_19330_end_0, end_mask = var_19330_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19330_cast")];
            tensor<int32, [4]> var_19334_begin_0 = const()[name = tensor<string, []>("op_19334_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_19334_end_0 = const()[name = tensor<string, []>("op_19334_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_19334_end_mask_0 = const()[name = tensor<string, []>("op_19334_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19334_cast = slice_by_index(begin = var_19334_begin_0, end = var_19334_end_0, end_mask = var_19334_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19334_cast")];
            tensor<int32, [4]> var_19338_begin_0 = const()[name = tensor<string, []>("op_19338_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_19338_end_0 = const()[name = tensor<string, []>("op_19338_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_19338_end_mask_0 = const()[name = tensor<string, []>("op_19338_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19338_cast = slice_by_index(begin = var_19338_begin_0, end = var_19338_end_0, end_mask = var_19338_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19338_cast")];
            tensor<int32, [4]> var_19342_begin_0 = const()[name = tensor<string, []>("op_19342_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_19342_end_0 = const()[name = tensor<string, []>("op_19342_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_19342_end_mask_0 = const()[name = tensor<string, []>("op_19342_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19342_cast = slice_by_index(begin = var_19342_begin_0, end = var_19342_end_0, end_mask = var_19342_end_mask_0, x = q_89_cast)[name = tensor<string, []>("op_19342_cast")];
            tensor<int32, [4]> k_179_perm_0 = const()[name = tensor<string, []>("k_179_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_19349_begin_0 = const()[name = tensor<string, []>("op_19349_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_19349_end_0 = const()[name = tensor<string, []>("op_19349_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_19349_end_mask_0 = const()[name = tensor<string, []>("op_19349_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_95 = transpose(perm = k_179_perm_0, x = k_177_cast)[name = tensor<string, []>("transpose_95")];
            tensor<fp16, [2, 1024, 1, 64]> var_19349_cast = slice_by_index(begin = var_19349_begin_0, end = var_19349_end_0, end_mask = var_19349_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19349_cast")];
            tensor<int32, [4]> var_19353_begin_0 = const()[name = tensor<string, []>("op_19353_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_19353_end_0 = const()[name = tensor<string, []>("op_19353_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_19353_end_mask_0 = const()[name = tensor<string, []>("op_19353_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_19353_cast = slice_by_index(begin = var_19353_begin_0, end = var_19353_end_0, end_mask = var_19353_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19353_cast")];
            tensor<int32, [4]> var_19357_begin_0 = const()[name = tensor<string, []>("op_19357_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_19357_end_0 = const()[name = tensor<string, []>("op_19357_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_19357_end_mask_0 = const()[name = tensor<string, []>("op_19357_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_19357_cast = slice_by_index(begin = var_19357_begin_0, end = var_19357_end_0, end_mask = var_19357_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19357_cast")];
            tensor<int32, [4]> var_19361_begin_0 = const()[name = tensor<string, []>("op_19361_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_19361_end_0 = const()[name = tensor<string, []>("op_19361_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_19361_end_mask_0 = const()[name = tensor<string, []>("op_19361_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_19361_cast = slice_by_index(begin = var_19361_begin_0, end = var_19361_end_0, end_mask = var_19361_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19361_cast")];
            tensor<int32, [4]> var_19365_begin_0 = const()[name = tensor<string, []>("op_19365_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_19365_end_0 = const()[name = tensor<string, []>("op_19365_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_19365_end_mask_0 = const()[name = tensor<string, []>("op_19365_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_19365_cast = slice_by_index(begin = var_19365_begin_0, end = var_19365_end_0, end_mask = var_19365_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19365_cast")];
            tensor<int32, [4]> var_19369_begin_0 = const()[name = tensor<string, []>("op_19369_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_19369_end_0 = const()[name = tensor<string, []>("op_19369_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_19369_end_mask_0 = const()[name = tensor<string, []>("op_19369_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_19369_cast = slice_by_index(begin = var_19369_begin_0, end = var_19369_end_0, end_mask = var_19369_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19369_cast")];
            tensor<int32, [4]> var_19373_begin_0 = const()[name = tensor<string, []>("op_19373_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_19373_end_0 = const()[name = tensor<string, []>("op_19373_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_19373_end_mask_0 = const()[name = tensor<string, []>("op_19373_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_19373_cast = slice_by_index(begin = var_19373_begin_0, end = var_19373_end_0, end_mask = var_19373_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19373_cast")];
            tensor<int32, [4]> var_19377_begin_0 = const()[name = tensor<string, []>("op_19377_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_19377_end_0 = const()[name = tensor<string, []>("op_19377_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_19377_end_mask_0 = const()[name = tensor<string, []>("op_19377_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_19377_cast = slice_by_index(begin = var_19377_begin_0, end = var_19377_end_0, end_mask = var_19377_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19377_cast")];
            tensor<int32, [4]> var_19381_begin_0 = const()[name = tensor<string, []>("op_19381_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_19381_end_0 = const()[name = tensor<string, []>("op_19381_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_19381_end_mask_0 = const()[name = tensor<string, []>("op_19381_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_19381_cast = slice_by_index(begin = var_19381_begin_0, end = var_19381_end_0, end_mask = var_19381_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19381_cast")];
            tensor<int32, [4]> var_19385_begin_0 = const()[name = tensor<string, []>("op_19385_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_19385_end_0 = const()[name = tensor<string, []>("op_19385_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_19385_end_mask_0 = const()[name = tensor<string, []>("op_19385_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_19385_cast = slice_by_index(begin = var_19385_begin_0, end = var_19385_end_0, end_mask = var_19385_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19385_cast")];
            tensor<int32, [4]> var_19389_begin_0 = const()[name = tensor<string, []>("op_19389_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_19389_end_0 = const()[name = tensor<string, []>("op_19389_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_19389_end_mask_0 = const()[name = tensor<string, []>("op_19389_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_19389_cast = slice_by_index(begin = var_19389_begin_0, end = var_19389_end_0, end_mask = var_19389_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19389_cast")];
            tensor<int32, [4]> var_19393_begin_0 = const()[name = tensor<string, []>("op_19393_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_19393_end_0 = const()[name = tensor<string, []>("op_19393_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_19393_end_mask_0 = const()[name = tensor<string, []>("op_19393_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_19393_cast = slice_by_index(begin = var_19393_begin_0, end = var_19393_end_0, end_mask = var_19393_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19393_cast")];
            tensor<int32, [4]> var_19397_begin_0 = const()[name = tensor<string, []>("op_19397_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_19397_end_0 = const()[name = tensor<string, []>("op_19397_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_19397_end_mask_0 = const()[name = tensor<string, []>("op_19397_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_19397_cast = slice_by_index(begin = var_19397_begin_0, end = var_19397_end_0, end_mask = var_19397_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19397_cast")];
            tensor<int32, [4]> var_19401_begin_0 = const()[name = tensor<string, []>("op_19401_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_19401_end_0 = const()[name = tensor<string, []>("op_19401_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_19401_end_mask_0 = const()[name = tensor<string, []>("op_19401_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_19401_cast = slice_by_index(begin = var_19401_begin_0, end = var_19401_end_0, end_mask = var_19401_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19401_cast")];
            tensor<int32, [4]> var_19405_begin_0 = const()[name = tensor<string, []>("op_19405_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_19405_end_0 = const()[name = tensor<string, []>("op_19405_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_19405_end_mask_0 = const()[name = tensor<string, []>("op_19405_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_19405_cast = slice_by_index(begin = var_19405_begin_0, end = var_19405_end_0, end_mask = var_19405_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19405_cast")];
            tensor<int32, [4]> var_19409_begin_0 = const()[name = tensor<string, []>("op_19409_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_19409_end_0 = const()[name = tensor<string, []>("op_19409_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_19409_end_mask_0 = const()[name = tensor<string, []>("op_19409_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_19409_cast = slice_by_index(begin = var_19409_begin_0, end = var_19409_end_0, end_mask = var_19409_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19409_cast")];
            tensor<int32, [4]> var_19413_begin_0 = const()[name = tensor<string, []>("op_19413_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_19413_end_0 = const()[name = tensor<string, []>("op_19413_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_19413_end_mask_0 = const()[name = tensor<string, []>("op_19413_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_19413_cast = slice_by_index(begin = var_19413_begin_0, end = var_19413_end_0, end_mask = var_19413_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19413_cast")];
            tensor<int32, [4]> var_19417_begin_0 = const()[name = tensor<string, []>("op_19417_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_19417_end_0 = const()[name = tensor<string, []>("op_19417_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_19417_end_mask_0 = const()[name = tensor<string, []>("op_19417_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_19417_cast = slice_by_index(begin = var_19417_begin_0, end = var_19417_end_0, end_mask = var_19417_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19417_cast")];
            tensor<int32, [4]> var_19421_begin_0 = const()[name = tensor<string, []>("op_19421_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_19421_end_0 = const()[name = tensor<string, []>("op_19421_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_19421_end_mask_0 = const()[name = tensor<string, []>("op_19421_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_19421_cast = slice_by_index(begin = var_19421_begin_0, end = var_19421_end_0, end_mask = var_19421_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19421_cast")];
            tensor<int32, [4]> var_19425_begin_0 = const()[name = tensor<string, []>("op_19425_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_19425_end_0 = const()[name = tensor<string, []>("op_19425_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_19425_end_mask_0 = const()[name = tensor<string, []>("op_19425_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_19425_cast = slice_by_index(begin = var_19425_begin_0, end = var_19425_end_0, end_mask = var_19425_end_mask_0, x = transpose_95)[name = tensor<string, []>("op_19425_cast")];
            tensor<int32, [4]> var_19427_begin_0 = const()[name = tensor<string, []>("op_19427_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_19427_end_0 = const()[name = tensor<string, []>("op_19427_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_19427_end_mask_0 = const()[name = tensor<string, []>("op_19427_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19427_cast = slice_by_index(begin = var_19427_begin_0, end = var_19427_end_0, end_mask = var_19427_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19427_cast")];
            tensor<int32, [4]> var_19431_begin_0 = const()[name = tensor<string, []>("op_19431_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_19431_end_0 = const()[name = tensor<string, []>("op_19431_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_19431_end_mask_0 = const()[name = tensor<string, []>("op_19431_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19431_cast = slice_by_index(begin = var_19431_begin_0, end = var_19431_end_0, end_mask = var_19431_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19431_cast")];
            tensor<int32, [4]> var_19435_begin_0 = const()[name = tensor<string, []>("op_19435_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_19435_end_0 = const()[name = tensor<string, []>("op_19435_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_19435_end_mask_0 = const()[name = tensor<string, []>("op_19435_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19435_cast = slice_by_index(begin = var_19435_begin_0, end = var_19435_end_0, end_mask = var_19435_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19435_cast")];
            tensor<int32, [4]> var_19439_begin_0 = const()[name = tensor<string, []>("op_19439_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_19439_end_0 = const()[name = tensor<string, []>("op_19439_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_19439_end_mask_0 = const()[name = tensor<string, []>("op_19439_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19439_cast = slice_by_index(begin = var_19439_begin_0, end = var_19439_end_0, end_mask = var_19439_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19439_cast")];
            tensor<int32, [4]> var_19443_begin_0 = const()[name = tensor<string, []>("op_19443_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_19443_end_0 = const()[name = tensor<string, []>("op_19443_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_19443_end_mask_0 = const()[name = tensor<string, []>("op_19443_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19443_cast = slice_by_index(begin = var_19443_begin_0, end = var_19443_end_0, end_mask = var_19443_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19443_cast")];
            tensor<int32, [4]> var_19447_begin_0 = const()[name = tensor<string, []>("op_19447_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_19447_end_0 = const()[name = tensor<string, []>("op_19447_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_19447_end_mask_0 = const()[name = tensor<string, []>("op_19447_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19447_cast = slice_by_index(begin = var_19447_begin_0, end = var_19447_end_0, end_mask = var_19447_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19447_cast")];
            tensor<int32, [4]> var_19451_begin_0 = const()[name = tensor<string, []>("op_19451_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_19451_end_0 = const()[name = tensor<string, []>("op_19451_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_19451_end_mask_0 = const()[name = tensor<string, []>("op_19451_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19451_cast = slice_by_index(begin = var_19451_begin_0, end = var_19451_end_0, end_mask = var_19451_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19451_cast")];
            tensor<int32, [4]> var_19455_begin_0 = const()[name = tensor<string, []>("op_19455_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_19455_end_0 = const()[name = tensor<string, []>("op_19455_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_19455_end_mask_0 = const()[name = tensor<string, []>("op_19455_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19455_cast = slice_by_index(begin = var_19455_begin_0, end = var_19455_end_0, end_mask = var_19455_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19455_cast")];
            tensor<int32, [4]> var_19459_begin_0 = const()[name = tensor<string, []>("op_19459_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_19459_end_0 = const()[name = tensor<string, []>("op_19459_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_19459_end_mask_0 = const()[name = tensor<string, []>("op_19459_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19459_cast = slice_by_index(begin = var_19459_begin_0, end = var_19459_end_0, end_mask = var_19459_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19459_cast")];
            tensor<int32, [4]> var_19463_begin_0 = const()[name = tensor<string, []>("op_19463_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_19463_end_0 = const()[name = tensor<string, []>("op_19463_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_19463_end_mask_0 = const()[name = tensor<string, []>("op_19463_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19463_cast = slice_by_index(begin = var_19463_begin_0, end = var_19463_end_0, end_mask = var_19463_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19463_cast")];
            tensor<int32, [4]> var_19467_begin_0 = const()[name = tensor<string, []>("op_19467_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_19467_end_0 = const()[name = tensor<string, []>("op_19467_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_19467_end_mask_0 = const()[name = tensor<string, []>("op_19467_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19467_cast = slice_by_index(begin = var_19467_begin_0, end = var_19467_end_0, end_mask = var_19467_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19467_cast")];
            tensor<int32, [4]> var_19471_begin_0 = const()[name = tensor<string, []>("op_19471_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_19471_end_0 = const()[name = tensor<string, []>("op_19471_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_19471_end_mask_0 = const()[name = tensor<string, []>("op_19471_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19471_cast = slice_by_index(begin = var_19471_begin_0, end = var_19471_end_0, end_mask = var_19471_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19471_cast")];
            tensor<int32, [4]> var_19475_begin_0 = const()[name = tensor<string, []>("op_19475_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_19475_end_0 = const()[name = tensor<string, []>("op_19475_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_19475_end_mask_0 = const()[name = tensor<string, []>("op_19475_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19475_cast = slice_by_index(begin = var_19475_begin_0, end = var_19475_end_0, end_mask = var_19475_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19475_cast")];
            tensor<int32, [4]> var_19479_begin_0 = const()[name = tensor<string, []>("op_19479_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_19479_end_0 = const()[name = tensor<string, []>("op_19479_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_19479_end_mask_0 = const()[name = tensor<string, []>("op_19479_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19479_cast = slice_by_index(begin = var_19479_begin_0, end = var_19479_end_0, end_mask = var_19479_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19479_cast")];
            tensor<int32, [4]> var_19483_begin_0 = const()[name = tensor<string, []>("op_19483_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_19483_end_0 = const()[name = tensor<string, []>("op_19483_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_19483_end_mask_0 = const()[name = tensor<string, []>("op_19483_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19483_cast = slice_by_index(begin = var_19483_begin_0, end = var_19483_end_0, end_mask = var_19483_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19483_cast")];
            tensor<int32, [4]> var_19487_begin_0 = const()[name = tensor<string, []>("op_19487_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_19487_end_0 = const()[name = tensor<string, []>("op_19487_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_19487_end_mask_0 = const()[name = tensor<string, []>("op_19487_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19487_cast = slice_by_index(begin = var_19487_begin_0, end = var_19487_end_0, end_mask = var_19487_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19487_cast")];
            tensor<int32, [4]> var_19491_begin_0 = const()[name = tensor<string, []>("op_19491_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_19491_end_0 = const()[name = tensor<string, []>("op_19491_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_19491_end_mask_0 = const()[name = tensor<string, []>("op_19491_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19491_cast = slice_by_index(begin = var_19491_begin_0, end = var_19491_end_0, end_mask = var_19491_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19491_cast")];
            tensor<int32, [4]> var_19495_begin_0 = const()[name = tensor<string, []>("op_19495_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_19495_end_0 = const()[name = tensor<string, []>("op_19495_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_19495_end_mask_0 = const()[name = tensor<string, []>("op_19495_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19495_cast = slice_by_index(begin = var_19495_begin_0, end = var_19495_end_0, end_mask = var_19495_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19495_cast")];
            tensor<int32, [4]> var_19499_begin_0 = const()[name = tensor<string, []>("op_19499_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_19499_end_0 = const()[name = tensor<string, []>("op_19499_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_19499_end_mask_0 = const()[name = tensor<string, []>("op_19499_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19499_cast = slice_by_index(begin = var_19499_begin_0, end = var_19499_end_0, end_mask = var_19499_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19499_cast")];
            tensor<int32, [4]> var_19503_begin_0 = const()[name = tensor<string, []>("op_19503_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_19503_end_0 = const()[name = tensor<string, []>("op_19503_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_19503_end_mask_0 = const()[name = tensor<string, []>("op_19503_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19503_cast = slice_by_index(begin = var_19503_begin_0, end = var_19503_end_0, end_mask = var_19503_end_mask_0, x = v_89_cast)[name = tensor<string, []>("op_19503_cast")];
            tensor<string, []> var_19507_equation_0 = const()[name = tensor<string, []>("op_19507_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19507_cast = einsum(equation = var_19507_equation_0, values = (var_19349_cast, var_19266_cast))[name = tensor<string, []>("op_19507_cast")];
            tensor<fp16, []> var_19508_to_fp16 = const()[name = tensor<string, []>("op_19508_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1601_cast = mul(x = var_19507_cast, y = var_19508_to_fp16)[name = tensor<string, []>("aw_1601_cast")];
            tensor<string, []> var_19511_equation_0 = const()[name = tensor<string, []>("op_19511_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19511_cast = einsum(equation = var_19511_equation_0, values = (var_19353_cast, var_19270_cast))[name = tensor<string, []>("op_19511_cast")];
            tensor<fp16, []> var_19512_to_fp16 = const()[name = tensor<string, []>("op_19512_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1603_cast = mul(x = var_19511_cast, y = var_19512_to_fp16)[name = tensor<string, []>("aw_1603_cast")];
            tensor<string, []> var_19515_equation_0 = const()[name = tensor<string, []>("op_19515_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19515_cast = einsum(equation = var_19515_equation_0, values = (var_19357_cast, var_19274_cast))[name = tensor<string, []>("op_19515_cast")];
            tensor<fp16, []> var_19516_to_fp16 = const()[name = tensor<string, []>("op_19516_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1605_cast = mul(x = var_19515_cast, y = var_19516_to_fp16)[name = tensor<string, []>("aw_1605_cast")];
            tensor<string, []> var_19519_equation_0 = const()[name = tensor<string, []>("op_19519_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19519_cast = einsum(equation = var_19519_equation_0, values = (var_19361_cast, var_19278_cast))[name = tensor<string, []>("op_19519_cast")];
            tensor<fp16, []> var_19520_to_fp16 = const()[name = tensor<string, []>("op_19520_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1607_cast = mul(x = var_19519_cast, y = var_19520_to_fp16)[name = tensor<string, []>("aw_1607_cast")];
            tensor<string, []> var_19523_equation_0 = const()[name = tensor<string, []>("op_19523_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19523_cast = einsum(equation = var_19523_equation_0, values = (var_19365_cast, var_19282_cast))[name = tensor<string, []>("op_19523_cast")];
            tensor<fp16, []> var_19524_to_fp16 = const()[name = tensor<string, []>("op_19524_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1609_cast = mul(x = var_19523_cast, y = var_19524_to_fp16)[name = tensor<string, []>("aw_1609_cast")];
            tensor<string, []> var_19527_equation_0 = const()[name = tensor<string, []>("op_19527_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19527_cast = einsum(equation = var_19527_equation_0, values = (var_19369_cast, var_19286_cast))[name = tensor<string, []>("op_19527_cast")];
            tensor<fp16, []> var_19528_to_fp16 = const()[name = tensor<string, []>("op_19528_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1611_cast = mul(x = var_19527_cast, y = var_19528_to_fp16)[name = tensor<string, []>("aw_1611_cast")];
            tensor<string, []> var_19531_equation_0 = const()[name = tensor<string, []>("op_19531_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19531_cast = einsum(equation = var_19531_equation_0, values = (var_19373_cast, var_19290_cast))[name = tensor<string, []>("op_19531_cast")];
            tensor<fp16, []> var_19532_to_fp16 = const()[name = tensor<string, []>("op_19532_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1613_cast = mul(x = var_19531_cast, y = var_19532_to_fp16)[name = tensor<string, []>("aw_1613_cast")];
            tensor<string, []> var_19535_equation_0 = const()[name = tensor<string, []>("op_19535_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19535_cast = einsum(equation = var_19535_equation_0, values = (var_19377_cast, var_19294_cast))[name = tensor<string, []>("op_19535_cast")];
            tensor<fp16, []> var_19536_to_fp16 = const()[name = tensor<string, []>("op_19536_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1615_cast = mul(x = var_19535_cast, y = var_19536_to_fp16)[name = tensor<string, []>("aw_1615_cast")];
            tensor<string, []> var_19539_equation_0 = const()[name = tensor<string, []>("op_19539_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19539_cast = einsum(equation = var_19539_equation_0, values = (var_19381_cast, var_19298_cast))[name = tensor<string, []>("op_19539_cast")];
            tensor<fp16, []> var_19540_to_fp16 = const()[name = tensor<string, []>("op_19540_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1617_cast = mul(x = var_19539_cast, y = var_19540_to_fp16)[name = tensor<string, []>("aw_1617_cast")];
            tensor<string, []> var_19543_equation_0 = const()[name = tensor<string, []>("op_19543_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19543_cast = einsum(equation = var_19543_equation_0, values = (var_19385_cast, var_19302_cast))[name = tensor<string, []>("op_19543_cast")];
            tensor<fp16, []> var_19544_to_fp16 = const()[name = tensor<string, []>("op_19544_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1619_cast = mul(x = var_19543_cast, y = var_19544_to_fp16)[name = tensor<string, []>("aw_1619_cast")];
            tensor<string, []> var_19547_equation_0 = const()[name = tensor<string, []>("op_19547_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19547_cast = einsum(equation = var_19547_equation_0, values = (var_19389_cast, var_19306_cast))[name = tensor<string, []>("op_19547_cast")];
            tensor<fp16, []> var_19548_to_fp16 = const()[name = tensor<string, []>("op_19548_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1621_cast = mul(x = var_19547_cast, y = var_19548_to_fp16)[name = tensor<string, []>("aw_1621_cast")];
            tensor<string, []> var_19551_equation_0 = const()[name = tensor<string, []>("op_19551_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19551_cast = einsum(equation = var_19551_equation_0, values = (var_19393_cast, var_19310_cast))[name = tensor<string, []>("op_19551_cast")];
            tensor<fp16, []> var_19552_to_fp16 = const()[name = tensor<string, []>("op_19552_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1623_cast = mul(x = var_19551_cast, y = var_19552_to_fp16)[name = tensor<string, []>("aw_1623_cast")];
            tensor<string, []> var_19555_equation_0 = const()[name = tensor<string, []>("op_19555_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19555_cast = einsum(equation = var_19555_equation_0, values = (var_19397_cast, var_19314_cast))[name = tensor<string, []>("op_19555_cast")];
            tensor<fp16, []> var_19556_to_fp16 = const()[name = tensor<string, []>("op_19556_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1625_cast = mul(x = var_19555_cast, y = var_19556_to_fp16)[name = tensor<string, []>("aw_1625_cast")];
            tensor<string, []> var_19559_equation_0 = const()[name = tensor<string, []>("op_19559_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19559_cast = einsum(equation = var_19559_equation_0, values = (var_19401_cast, var_19318_cast))[name = tensor<string, []>("op_19559_cast")];
            tensor<fp16, []> var_19560_to_fp16 = const()[name = tensor<string, []>("op_19560_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1627_cast = mul(x = var_19559_cast, y = var_19560_to_fp16)[name = tensor<string, []>("aw_1627_cast")];
            tensor<string, []> var_19563_equation_0 = const()[name = tensor<string, []>("op_19563_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19563_cast = einsum(equation = var_19563_equation_0, values = (var_19405_cast, var_19322_cast))[name = tensor<string, []>("op_19563_cast")];
            tensor<fp16, []> var_19564_to_fp16 = const()[name = tensor<string, []>("op_19564_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1629_cast = mul(x = var_19563_cast, y = var_19564_to_fp16)[name = tensor<string, []>("aw_1629_cast")];
            tensor<string, []> var_19567_equation_0 = const()[name = tensor<string, []>("op_19567_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19567_cast = einsum(equation = var_19567_equation_0, values = (var_19409_cast, var_19326_cast))[name = tensor<string, []>("op_19567_cast")];
            tensor<fp16, []> var_19568_to_fp16 = const()[name = tensor<string, []>("op_19568_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1631_cast = mul(x = var_19567_cast, y = var_19568_to_fp16)[name = tensor<string, []>("aw_1631_cast")];
            tensor<string, []> var_19571_equation_0 = const()[name = tensor<string, []>("op_19571_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19571_cast = einsum(equation = var_19571_equation_0, values = (var_19413_cast, var_19330_cast))[name = tensor<string, []>("op_19571_cast")];
            tensor<fp16, []> var_19572_to_fp16 = const()[name = tensor<string, []>("op_19572_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1633_cast = mul(x = var_19571_cast, y = var_19572_to_fp16)[name = tensor<string, []>("aw_1633_cast")];
            tensor<string, []> var_19575_equation_0 = const()[name = tensor<string, []>("op_19575_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19575_cast = einsum(equation = var_19575_equation_0, values = (var_19417_cast, var_19334_cast))[name = tensor<string, []>("op_19575_cast")];
            tensor<fp16, []> var_19576_to_fp16 = const()[name = tensor<string, []>("op_19576_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1635_cast = mul(x = var_19575_cast, y = var_19576_to_fp16)[name = tensor<string, []>("aw_1635_cast")];
            tensor<string, []> var_19579_equation_0 = const()[name = tensor<string, []>("op_19579_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19579_cast = einsum(equation = var_19579_equation_0, values = (var_19421_cast, var_19338_cast))[name = tensor<string, []>("op_19579_cast")];
            tensor<fp16, []> var_19580_to_fp16 = const()[name = tensor<string, []>("op_19580_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1637_cast = mul(x = var_19579_cast, y = var_19580_to_fp16)[name = tensor<string, []>("aw_1637_cast")];
            tensor<string, []> var_19583_equation_0 = const()[name = tensor<string, []>("op_19583_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19583_cast = einsum(equation = var_19583_equation_0, values = (var_19425_cast, var_19342_cast))[name = tensor<string, []>("op_19583_cast")];
            tensor<fp16, []> var_19584_to_fp16 = const()[name = tensor<string, []>("op_19584_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1639_cast = mul(x = var_19583_cast, y = var_19584_to_fp16)[name = tensor<string, []>("aw_1639_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19586_cast = softmax(axis = var_2634, x = aw_1601_cast)[name = tensor<string, []>("op_19586_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19587_cast = softmax(axis = var_2634, x = aw_1603_cast)[name = tensor<string, []>("op_19587_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19588_cast = softmax(axis = var_2634, x = aw_1605_cast)[name = tensor<string, []>("op_19588_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19589_cast = softmax(axis = var_2634, x = aw_1607_cast)[name = tensor<string, []>("op_19589_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19590_cast = softmax(axis = var_2634, x = aw_1609_cast)[name = tensor<string, []>("op_19590_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19591_cast = softmax(axis = var_2634, x = aw_1611_cast)[name = tensor<string, []>("op_19591_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19592_cast = softmax(axis = var_2634, x = aw_1613_cast)[name = tensor<string, []>("op_19592_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19593_cast = softmax(axis = var_2634, x = aw_1615_cast)[name = tensor<string, []>("op_19593_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19594_cast = softmax(axis = var_2634, x = aw_1617_cast)[name = tensor<string, []>("op_19594_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19595_cast = softmax(axis = var_2634, x = aw_1619_cast)[name = tensor<string, []>("op_19595_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19596_cast = softmax(axis = var_2634, x = aw_1621_cast)[name = tensor<string, []>("op_19596_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19597_cast = softmax(axis = var_2634, x = aw_1623_cast)[name = tensor<string, []>("op_19597_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19598_cast = softmax(axis = var_2634, x = aw_1625_cast)[name = tensor<string, []>("op_19598_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19599_cast = softmax(axis = var_2634, x = aw_1627_cast)[name = tensor<string, []>("op_19599_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19600_cast = softmax(axis = var_2634, x = aw_1629_cast)[name = tensor<string, []>("op_19600_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19601_cast = softmax(axis = var_2634, x = aw_1631_cast)[name = tensor<string, []>("op_19601_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19602_cast = softmax(axis = var_2634, x = aw_1633_cast)[name = tensor<string, []>("op_19602_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19603_cast = softmax(axis = var_2634, x = aw_1635_cast)[name = tensor<string, []>("op_19603_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19604_cast = softmax(axis = var_2634, x = aw_1637_cast)[name = tensor<string, []>("op_19604_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_19605_cast = softmax(axis = var_2634, x = aw_1639_cast)[name = tensor<string, []>("op_19605_cast")];
            tensor<string, []> var_19607_equation_0 = const()[name = tensor<string, []>("op_19607_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19607_cast = einsum(equation = var_19607_equation_0, values = (var_19427_cast, var_19586_cast))[name = tensor<string, []>("op_19607_cast")];
            tensor<string, []> var_19609_equation_0 = const()[name = tensor<string, []>("op_19609_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19609_cast = einsum(equation = var_19609_equation_0, values = (var_19431_cast, var_19587_cast))[name = tensor<string, []>("op_19609_cast")];
            tensor<string, []> var_19611_equation_0 = const()[name = tensor<string, []>("op_19611_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19611_cast = einsum(equation = var_19611_equation_0, values = (var_19435_cast, var_19588_cast))[name = tensor<string, []>("op_19611_cast")];
            tensor<string, []> var_19613_equation_0 = const()[name = tensor<string, []>("op_19613_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19613_cast = einsum(equation = var_19613_equation_0, values = (var_19439_cast, var_19589_cast))[name = tensor<string, []>("op_19613_cast")];
            tensor<string, []> var_19615_equation_0 = const()[name = tensor<string, []>("op_19615_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19615_cast = einsum(equation = var_19615_equation_0, values = (var_19443_cast, var_19590_cast))[name = tensor<string, []>("op_19615_cast")];
            tensor<string, []> var_19617_equation_0 = const()[name = tensor<string, []>("op_19617_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19617_cast = einsum(equation = var_19617_equation_0, values = (var_19447_cast, var_19591_cast))[name = tensor<string, []>("op_19617_cast")];
            tensor<string, []> var_19619_equation_0 = const()[name = tensor<string, []>("op_19619_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19619_cast = einsum(equation = var_19619_equation_0, values = (var_19451_cast, var_19592_cast))[name = tensor<string, []>("op_19619_cast")];
            tensor<string, []> var_19621_equation_0 = const()[name = tensor<string, []>("op_19621_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19621_cast = einsum(equation = var_19621_equation_0, values = (var_19455_cast, var_19593_cast))[name = tensor<string, []>("op_19621_cast")];
            tensor<string, []> var_19623_equation_0 = const()[name = tensor<string, []>("op_19623_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19623_cast = einsum(equation = var_19623_equation_0, values = (var_19459_cast, var_19594_cast))[name = tensor<string, []>("op_19623_cast")];
            tensor<string, []> var_19625_equation_0 = const()[name = tensor<string, []>("op_19625_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19625_cast = einsum(equation = var_19625_equation_0, values = (var_19463_cast, var_19595_cast))[name = tensor<string, []>("op_19625_cast")];
            tensor<string, []> var_19627_equation_0 = const()[name = tensor<string, []>("op_19627_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19627_cast = einsum(equation = var_19627_equation_0, values = (var_19467_cast, var_19596_cast))[name = tensor<string, []>("op_19627_cast")];
            tensor<string, []> var_19629_equation_0 = const()[name = tensor<string, []>("op_19629_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19629_cast = einsum(equation = var_19629_equation_0, values = (var_19471_cast, var_19597_cast))[name = tensor<string, []>("op_19629_cast")];
            tensor<string, []> var_19631_equation_0 = const()[name = tensor<string, []>("op_19631_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19631_cast = einsum(equation = var_19631_equation_0, values = (var_19475_cast, var_19598_cast))[name = tensor<string, []>("op_19631_cast")];
            tensor<string, []> var_19633_equation_0 = const()[name = tensor<string, []>("op_19633_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19633_cast = einsum(equation = var_19633_equation_0, values = (var_19479_cast, var_19599_cast))[name = tensor<string, []>("op_19633_cast")];
            tensor<string, []> var_19635_equation_0 = const()[name = tensor<string, []>("op_19635_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19635_cast = einsum(equation = var_19635_equation_0, values = (var_19483_cast, var_19600_cast))[name = tensor<string, []>("op_19635_cast")];
            tensor<string, []> var_19637_equation_0 = const()[name = tensor<string, []>("op_19637_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19637_cast = einsum(equation = var_19637_equation_0, values = (var_19487_cast, var_19601_cast))[name = tensor<string, []>("op_19637_cast")];
            tensor<string, []> var_19639_equation_0 = const()[name = tensor<string, []>("op_19639_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19639_cast = einsum(equation = var_19639_equation_0, values = (var_19491_cast, var_19602_cast))[name = tensor<string, []>("op_19639_cast")];
            tensor<string, []> var_19641_equation_0 = const()[name = tensor<string, []>("op_19641_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19641_cast = einsum(equation = var_19641_equation_0, values = (var_19495_cast, var_19603_cast))[name = tensor<string, []>("op_19641_cast")];
            tensor<string, []> var_19643_equation_0 = const()[name = tensor<string, []>("op_19643_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19643_cast = einsum(equation = var_19643_equation_0, values = (var_19499_cast, var_19604_cast))[name = tensor<string, []>("op_19643_cast")];
            tensor<string, []> var_19645_equation_0 = const()[name = tensor<string, []>("op_19645_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_19645_cast = einsum(equation = var_19645_equation_0, values = (var_19503_cast, var_19605_cast))[name = tensor<string, []>("op_19645_cast")];
            tensor<bool, []> input_297_interleave_0 = const()[name = tensor<string, []>("input_297_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_297_cast = concat(axis = var_2634, interleave = input_297_interleave_0, values = (var_19607_cast, var_19609_cast, var_19611_cast, var_19613_cast, var_19615_cast, var_19617_cast, var_19619_cast, var_19621_cast, var_19623_cast, var_19625_cast, var_19627_cast, var_19629_cast, var_19631_cast, var_19633_cast, var_19635_cast, var_19637_cast, var_19639_cast, var_19641_cast, var_19643_cast, var_19645_cast))[name = tensor<string, []>("input_297_cast")];
            tensor<int32, [2]> var_19651 = const()[name = tensor<string, []>("op_19651"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_19653 = const()[name = tensor<string, []>("op_19653"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_19655_pad_type_0 = const()[name = tensor<string, []>("op_19655_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_19655_pad_0 = const()[name = tensor<string, []>("op_19655_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1528076992)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1531353856)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_19655_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_out_0_bias_to_fp16, dilations = var_19653, groups = var_2634, pad = var_19655_pad_0, pad_type = var_19655_pad_type_0, strides = var_19651, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn1_to_out_0_weight_to_fp16, x = input_297_cast)[name = tensor<string, []>("op_19655_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_135_cast = add(x = var_19655_cast, y = inputs_133_cast)[name = tensor<string, []>("inputs_135_cast")];
            tensor<int32, [1]> var_19659 = const()[name = tensor<string, []>("op_19659"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_135_cast = reduce_mean(axes = var_19659, keep_dims = var_2629, x = inputs_135_cast)[name = tensor<string, []>("channels_mean_135_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_135_cast = sub(x = inputs_135_cast, y = channels_mean_135_cast)[name = tensor<string, []>("zero_mean_135_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_135_cast = mul(x = zero_mean_135_cast, y = zero_mean_135_cast)[name = tensor<string, []>("zero_mean_sq_135_cast")];
            tensor<int32, [1]> var_19663 = const()[name = tensor<string, []>("op_19663"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_19664_cast = reduce_mean(axes = var_19663, keep_dims = var_2629, x = zero_mean_sq_135_cast)[name = tensor<string, []>("op_19664_cast")];
            tensor<fp16, []> var_19665_to_fp16 = const()[name = tensor<string, []>("op_19665_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_19666_cast = add(x = var_19664_cast, y = var_19665_to_fp16)[name = tensor<string, []>("op_19666_cast")];
            tensor<fp16, []> denom_135_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_135_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_135_cast = rsqrt(epsilon = denom_135_epsilon_0_to_fp16, x = var_19666_cast)[name = tensor<string, []>("denom_135_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_135_cast = mul(x = zero_mean_135_cast, y = denom_135_cast)[name = tensor<string, []>("out_135_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_19670_to_fp16 = const()[name = tensor<string, []>("op_19670_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1531356480)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_19671_cast = add(x = out_135_cast, y = var_19670_to_fp16)[name = tensor<string, []>("op_19671_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_19673_to_fp16 = const()[name = tensor<string, []>("op_19673_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1531359104)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_187_cast = mul(x = var_19671_cast, y = var_19673_to_fp16)[name = tensor<string, []>("hidden_states_187_cast")];
            tensor<int32, [2]> var_19680 = const()[name = tensor<string, []>("op_19680"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_19682 = const()[name = tensor<string, []>("op_19682"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_91_pad_type_0 = const()[name = tensor<string, []>("q_91_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_91_pad_0 = const()[name = tensor<string, []>("q_91_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1531361728)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_91_cast = conv(dilations = var_19682, groups = var_2634, pad = q_91_pad_0, pad_type = q_91_pad_type_0, strides = var_19680, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_q_weight_to_fp16, x = hidden_states_187_cast)[name = tensor<string, []>("q_91_cast")];
            tensor<int32, [2]> var_19686 = const()[name = tensor<string, []>("op_19686"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_19688 = const()[name = tensor<string, []>("op_19688"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_181_pad_type_0 = const()[name = tensor<string, []>("k_181_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_181_pad_0 = const()[name = tensor<string, []>("k_181_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1534638592)))];
            tensor<fp16, [2, 1280, 1, 77]> k_181_cast = conv(dilations = var_19688, groups = var_2634, pad = k_181_pad_0, pad_type = k_181_pad_type_0, strides = var_19686, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_181_cast")];
            tensor<int32, [2]> var_19692 = const()[name = tensor<string, []>("op_19692"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_19694 = const()[name = tensor<string, []>("op_19694"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_91_pad_type_0 = const()[name = tensor<string, []>("v_91_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_91_pad_0 = const()[name = tensor<string, []>("v_91_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1539881536)))];
            tensor<fp16, [2, 1280, 1, 77]> v_91_cast = conv(dilations = var_19694, groups = var_2634, pad = v_91_pad_0, pad_type = v_91_pad_type_0, strides = var_19692, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_91_cast")];
            tensor<int32, [4]> var_19698_begin_0 = const()[name = tensor<string, []>("op_19698_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_19698_end_0 = const()[name = tensor<string, []>("op_19698_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_19698_end_mask_0 = const()[name = tensor<string, []>("op_19698_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19698_cast = slice_by_index(begin = var_19698_begin_0, end = var_19698_end_0, end_mask = var_19698_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19698_cast")];
            tensor<int32, [4]> var_19702_begin_0 = const()[name = tensor<string, []>("op_19702_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_19702_end_0 = const()[name = tensor<string, []>("op_19702_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_19702_end_mask_0 = const()[name = tensor<string, []>("op_19702_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19702_cast = slice_by_index(begin = var_19702_begin_0, end = var_19702_end_0, end_mask = var_19702_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19702_cast")];
            tensor<int32, [4]> var_19706_begin_0 = const()[name = tensor<string, []>("op_19706_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_19706_end_0 = const()[name = tensor<string, []>("op_19706_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_19706_end_mask_0 = const()[name = tensor<string, []>("op_19706_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19706_cast = slice_by_index(begin = var_19706_begin_0, end = var_19706_end_0, end_mask = var_19706_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19706_cast")];
            tensor<int32, [4]> var_19710_begin_0 = const()[name = tensor<string, []>("op_19710_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_19710_end_0 = const()[name = tensor<string, []>("op_19710_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_19710_end_mask_0 = const()[name = tensor<string, []>("op_19710_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19710_cast = slice_by_index(begin = var_19710_begin_0, end = var_19710_end_0, end_mask = var_19710_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19710_cast")];
            tensor<int32, [4]> var_19714_begin_0 = const()[name = tensor<string, []>("op_19714_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_19714_end_0 = const()[name = tensor<string, []>("op_19714_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_19714_end_mask_0 = const()[name = tensor<string, []>("op_19714_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19714_cast = slice_by_index(begin = var_19714_begin_0, end = var_19714_end_0, end_mask = var_19714_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19714_cast")];
            tensor<int32, [4]> var_19718_begin_0 = const()[name = tensor<string, []>("op_19718_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_19718_end_0 = const()[name = tensor<string, []>("op_19718_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_19718_end_mask_0 = const()[name = tensor<string, []>("op_19718_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19718_cast = slice_by_index(begin = var_19718_begin_0, end = var_19718_end_0, end_mask = var_19718_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19718_cast")];
            tensor<int32, [4]> var_19722_begin_0 = const()[name = tensor<string, []>("op_19722_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_19722_end_0 = const()[name = tensor<string, []>("op_19722_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_19722_end_mask_0 = const()[name = tensor<string, []>("op_19722_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19722_cast = slice_by_index(begin = var_19722_begin_0, end = var_19722_end_0, end_mask = var_19722_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19722_cast")];
            tensor<int32, [4]> var_19726_begin_0 = const()[name = tensor<string, []>("op_19726_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_19726_end_0 = const()[name = tensor<string, []>("op_19726_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_19726_end_mask_0 = const()[name = tensor<string, []>("op_19726_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19726_cast = slice_by_index(begin = var_19726_begin_0, end = var_19726_end_0, end_mask = var_19726_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19726_cast")];
            tensor<int32, [4]> var_19730_begin_0 = const()[name = tensor<string, []>("op_19730_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_19730_end_0 = const()[name = tensor<string, []>("op_19730_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_19730_end_mask_0 = const()[name = tensor<string, []>("op_19730_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19730_cast = slice_by_index(begin = var_19730_begin_0, end = var_19730_end_0, end_mask = var_19730_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19730_cast")];
            tensor<int32, [4]> var_19734_begin_0 = const()[name = tensor<string, []>("op_19734_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_19734_end_0 = const()[name = tensor<string, []>("op_19734_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_19734_end_mask_0 = const()[name = tensor<string, []>("op_19734_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19734_cast = slice_by_index(begin = var_19734_begin_0, end = var_19734_end_0, end_mask = var_19734_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19734_cast")];
            tensor<int32, [4]> var_19738_begin_0 = const()[name = tensor<string, []>("op_19738_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_19738_end_0 = const()[name = tensor<string, []>("op_19738_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_19738_end_mask_0 = const()[name = tensor<string, []>("op_19738_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19738_cast = slice_by_index(begin = var_19738_begin_0, end = var_19738_end_0, end_mask = var_19738_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19738_cast")];
            tensor<int32, [4]> var_19742_begin_0 = const()[name = tensor<string, []>("op_19742_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_19742_end_0 = const()[name = tensor<string, []>("op_19742_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_19742_end_mask_0 = const()[name = tensor<string, []>("op_19742_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19742_cast = slice_by_index(begin = var_19742_begin_0, end = var_19742_end_0, end_mask = var_19742_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19742_cast")];
            tensor<int32, [4]> var_19746_begin_0 = const()[name = tensor<string, []>("op_19746_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_19746_end_0 = const()[name = tensor<string, []>("op_19746_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_19746_end_mask_0 = const()[name = tensor<string, []>("op_19746_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19746_cast = slice_by_index(begin = var_19746_begin_0, end = var_19746_end_0, end_mask = var_19746_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19746_cast")];
            tensor<int32, [4]> var_19750_begin_0 = const()[name = tensor<string, []>("op_19750_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_19750_end_0 = const()[name = tensor<string, []>("op_19750_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_19750_end_mask_0 = const()[name = tensor<string, []>("op_19750_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19750_cast = slice_by_index(begin = var_19750_begin_0, end = var_19750_end_0, end_mask = var_19750_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19750_cast")];
            tensor<int32, [4]> var_19754_begin_0 = const()[name = tensor<string, []>("op_19754_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_19754_end_0 = const()[name = tensor<string, []>("op_19754_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_19754_end_mask_0 = const()[name = tensor<string, []>("op_19754_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19754_cast = slice_by_index(begin = var_19754_begin_0, end = var_19754_end_0, end_mask = var_19754_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19754_cast")];
            tensor<int32, [4]> var_19758_begin_0 = const()[name = tensor<string, []>("op_19758_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_19758_end_0 = const()[name = tensor<string, []>("op_19758_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_19758_end_mask_0 = const()[name = tensor<string, []>("op_19758_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19758_cast = slice_by_index(begin = var_19758_begin_0, end = var_19758_end_0, end_mask = var_19758_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19758_cast")];
            tensor<int32, [4]> var_19762_begin_0 = const()[name = tensor<string, []>("op_19762_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_19762_end_0 = const()[name = tensor<string, []>("op_19762_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_19762_end_mask_0 = const()[name = tensor<string, []>("op_19762_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19762_cast = slice_by_index(begin = var_19762_begin_0, end = var_19762_end_0, end_mask = var_19762_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19762_cast")];
            tensor<int32, [4]> var_19766_begin_0 = const()[name = tensor<string, []>("op_19766_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_19766_end_0 = const()[name = tensor<string, []>("op_19766_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_19766_end_mask_0 = const()[name = tensor<string, []>("op_19766_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19766_cast = slice_by_index(begin = var_19766_begin_0, end = var_19766_end_0, end_mask = var_19766_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19766_cast")];
            tensor<int32, [4]> var_19770_begin_0 = const()[name = tensor<string, []>("op_19770_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_19770_end_0 = const()[name = tensor<string, []>("op_19770_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_19770_end_mask_0 = const()[name = tensor<string, []>("op_19770_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19770_cast = slice_by_index(begin = var_19770_begin_0, end = var_19770_end_0, end_mask = var_19770_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19770_cast")];
            tensor<int32, [4]> var_19774_begin_0 = const()[name = tensor<string, []>("op_19774_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_19774_end_0 = const()[name = tensor<string, []>("op_19774_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_19774_end_mask_0 = const()[name = tensor<string, []>("op_19774_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_19774_cast = slice_by_index(begin = var_19774_begin_0, end = var_19774_end_0, end_mask = var_19774_end_mask_0, x = q_91_cast)[name = tensor<string, []>("op_19774_cast")];
            tensor<int32, [4]> k_183_perm_0 = const()[name = tensor<string, []>("k_183_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_19781_begin_0 = const()[name = tensor<string, []>("op_19781_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_19781_end_0 = const()[name = tensor<string, []>("op_19781_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_19781_end_mask_0 = const()[name = tensor<string, []>("op_19781_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_94 = transpose(perm = k_183_perm_0, x = k_181_cast)[name = tensor<string, []>("transpose_94")];
            tensor<fp16, [2, 77, 1, 64]> var_19781_cast = slice_by_index(begin = var_19781_begin_0, end = var_19781_end_0, end_mask = var_19781_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19781_cast")];
            tensor<int32, [4]> var_19785_begin_0 = const()[name = tensor<string, []>("op_19785_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_19785_end_0 = const()[name = tensor<string, []>("op_19785_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_19785_end_mask_0 = const()[name = tensor<string, []>("op_19785_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_19785_cast = slice_by_index(begin = var_19785_begin_0, end = var_19785_end_0, end_mask = var_19785_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19785_cast")];
            tensor<int32, [4]> var_19789_begin_0 = const()[name = tensor<string, []>("op_19789_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_19789_end_0 = const()[name = tensor<string, []>("op_19789_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_19789_end_mask_0 = const()[name = tensor<string, []>("op_19789_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_19789_cast = slice_by_index(begin = var_19789_begin_0, end = var_19789_end_0, end_mask = var_19789_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19789_cast")];
            tensor<int32, [4]> var_19793_begin_0 = const()[name = tensor<string, []>("op_19793_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_19793_end_0 = const()[name = tensor<string, []>("op_19793_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_19793_end_mask_0 = const()[name = tensor<string, []>("op_19793_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_19793_cast = slice_by_index(begin = var_19793_begin_0, end = var_19793_end_0, end_mask = var_19793_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19793_cast")];
            tensor<int32, [4]> var_19797_begin_0 = const()[name = tensor<string, []>("op_19797_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_19797_end_0 = const()[name = tensor<string, []>("op_19797_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_19797_end_mask_0 = const()[name = tensor<string, []>("op_19797_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_19797_cast = slice_by_index(begin = var_19797_begin_0, end = var_19797_end_0, end_mask = var_19797_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19797_cast")];
            tensor<int32, [4]> var_19801_begin_0 = const()[name = tensor<string, []>("op_19801_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_19801_end_0 = const()[name = tensor<string, []>("op_19801_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_19801_end_mask_0 = const()[name = tensor<string, []>("op_19801_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_19801_cast = slice_by_index(begin = var_19801_begin_0, end = var_19801_end_0, end_mask = var_19801_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19801_cast")];
            tensor<int32, [4]> var_19805_begin_0 = const()[name = tensor<string, []>("op_19805_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_19805_end_0 = const()[name = tensor<string, []>("op_19805_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_19805_end_mask_0 = const()[name = tensor<string, []>("op_19805_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_19805_cast = slice_by_index(begin = var_19805_begin_0, end = var_19805_end_0, end_mask = var_19805_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19805_cast")];
            tensor<int32, [4]> var_19809_begin_0 = const()[name = tensor<string, []>("op_19809_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_19809_end_0 = const()[name = tensor<string, []>("op_19809_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_19809_end_mask_0 = const()[name = tensor<string, []>("op_19809_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_19809_cast = slice_by_index(begin = var_19809_begin_0, end = var_19809_end_0, end_mask = var_19809_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19809_cast")];
            tensor<int32, [4]> var_19813_begin_0 = const()[name = tensor<string, []>("op_19813_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_19813_end_0 = const()[name = tensor<string, []>("op_19813_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_19813_end_mask_0 = const()[name = tensor<string, []>("op_19813_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_19813_cast = slice_by_index(begin = var_19813_begin_0, end = var_19813_end_0, end_mask = var_19813_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19813_cast")];
            tensor<int32, [4]> var_19817_begin_0 = const()[name = tensor<string, []>("op_19817_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_19817_end_0 = const()[name = tensor<string, []>("op_19817_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_19817_end_mask_0 = const()[name = tensor<string, []>("op_19817_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_19817_cast = slice_by_index(begin = var_19817_begin_0, end = var_19817_end_0, end_mask = var_19817_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19817_cast")];
            tensor<int32, [4]> var_19821_begin_0 = const()[name = tensor<string, []>("op_19821_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_19821_end_0 = const()[name = tensor<string, []>("op_19821_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_19821_end_mask_0 = const()[name = tensor<string, []>("op_19821_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_19821_cast = slice_by_index(begin = var_19821_begin_0, end = var_19821_end_0, end_mask = var_19821_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19821_cast")];
            tensor<int32, [4]> var_19825_begin_0 = const()[name = tensor<string, []>("op_19825_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_19825_end_0 = const()[name = tensor<string, []>("op_19825_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_19825_end_mask_0 = const()[name = tensor<string, []>("op_19825_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_19825_cast = slice_by_index(begin = var_19825_begin_0, end = var_19825_end_0, end_mask = var_19825_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19825_cast")];
            tensor<int32, [4]> var_19829_begin_0 = const()[name = tensor<string, []>("op_19829_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_19829_end_0 = const()[name = tensor<string, []>("op_19829_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_19829_end_mask_0 = const()[name = tensor<string, []>("op_19829_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_19829_cast = slice_by_index(begin = var_19829_begin_0, end = var_19829_end_0, end_mask = var_19829_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19829_cast")];
            tensor<int32, [4]> var_19833_begin_0 = const()[name = tensor<string, []>("op_19833_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_19833_end_0 = const()[name = tensor<string, []>("op_19833_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_19833_end_mask_0 = const()[name = tensor<string, []>("op_19833_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_19833_cast = slice_by_index(begin = var_19833_begin_0, end = var_19833_end_0, end_mask = var_19833_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19833_cast")];
            tensor<int32, [4]> var_19837_begin_0 = const()[name = tensor<string, []>("op_19837_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_19837_end_0 = const()[name = tensor<string, []>("op_19837_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_19837_end_mask_0 = const()[name = tensor<string, []>("op_19837_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_19837_cast = slice_by_index(begin = var_19837_begin_0, end = var_19837_end_0, end_mask = var_19837_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19837_cast")];
            tensor<int32, [4]> var_19841_begin_0 = const()[name = tensor<string, []>("op_19841_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_19841_end_0 = const()[name = tensor<string, []>("op_19841_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_19841_end_mask_0 = const()[name = tensor<string, []>("op_19841_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_19841_cast = slice_by_index(begin = var_19841_begin_0, end = var_19841_end_0, end_mask = var_19841_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19841_cast")];
            tensor<int32, [4]> var_19845_begin_0 = const()[name = tensor<string, []>("op_19845_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_19845_end_0 = const()[name = tensor<string, []>("op_19845_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_19845_end_mask_0 = const()[name = tensor<string, []>("op_19845_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_19845_cast = slice_by_index(begin = var_19845_begin_0, end = var_19845_end_0, end_mask = var_19845_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19845_cast")];
            tensor<int32, [4]> var_19849_begin_0 = const()[name = tensor<string, []>("op_19849_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_19849_end_0 = const()[name = tensor<string, []>("op_19849_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_19849_end_mask_0 = const()[name = tensor<string, []>("op_19849_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_19849_cast = slice_by_index(begin = var_19849_begin_0, end = var_19849_end_0, end_mask = var_19849_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19849_cast")];
            tensor<int32, [4]> var_19853_begin_0 = const()[name = tensor<string, []>("op_19853_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_19853_end_0 = const()[name = tensor<string, []>("op_19853_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_19853_end_mask_0 = const()[name = tensor<string, []>("op_19853_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_19853_cast = slice_by_index(begin = var_19853_begin_0, end = var_19853_end_0, end_mask = var_19853_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19853_cast")];
            tensor<int32, [4]> var_19857_begin_0 = const()[name = tensor<string, []>("op_19857_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_19857_end_0 = const()[name = tensor<string, []>("op_19857_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_19857_end_mask_0 = const()[name = tensor<string, []>("op_19857_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_19857_cast = slice_by_index(begin = var_19857_begin_0, end = var_19857_end_0, end_mask = var_19857_end_mask_0, x = transpose_94)[name = tensor<string, []>("op_19857_cast")];
            tensor<int32, [4]> var_19859_begin_0 = const()[name = tensor<string, []>("op_19859_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_19859_end_0 = const()[name = tensor<string, []>("op_19859_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_19859_end_mask_0 = const()[name = tensor<string, []>("op_19859_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19859_cast = slice_by_index(begin = var_19859_begin_0, end = var_19859_end_0, end_mask = var_19859_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19859_cast")];
            tensor<int32, [4]> var_19863_begin_0 = const()[name = tensor<string, []>("op_19863_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_19863_end_0 = const()[name = tensor<string, []>("op_19863_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_19863_end_mask_0 = const()[name = tensor<string, []>("op_19863_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19863_cast = slice_by_index(begin = var_19863_begin_0, end = var_19863_end_0, end_mask = var_19863_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19863_cast")];
            tensor<int32, [4]> var_19867_begin_0 = const()[name = tensor<string, []>("op_19867_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_19867_end_0 = const()[name = tensor<string, []>("op_19867_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_19867_end_mask_0 = const()[name = tensor<string, []>("op_19867_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19867_cast = slice_by_index(begin = var_19867_begin_0, end = var_19867_end_0, end_mask = var_19867_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19867_cast")];
            tensor<int32, [4]> var_19871_begin_0 = const()[name = tensor<string, []>("op_19871_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_19871_end_0 = const()[name = tensor<string, []>("op_19871_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_19871_end_mask_0 = const()[name = tensor<string, []>("op_19871_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19871_cast = slice_by_index(begin = var_19871_begin_0, end = var_19871_end_0, end_mask = var_19871_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19871_cast")];
            tensor<int32, [4]> var_19875_begin_0 = const()[name = tensor<string, []>("op_19875_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_19875_end_0 = const()[name = tensor<string, []>("op_19875_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_19875_end_mask_0 = const()[name = tensor<string, []>("op_19875_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19875_cast = slice_by_index(begin = var_19875_begin_0, end = var_19875_end_0, end_mask = var_19875_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19875_cast")];
            tensor<int32, [4]> var_19879_begin_0 = const()[name = tensor<string, []>("op_19879_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_19879_end_0 = const()[name = tensor<string, []>("op_19879_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_19879_end_mask_0 = const()[name = tensor<string, []>("op_19879_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19879_cast = slice_by_index(begin = var_19879_begin_0, end = var_19879_end_0, end_mask = var_19879_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19879_cast")];
            tensor<int32, [4]> var_19883_begin_0 = const()[name = tensor<string, []>("op_19883_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_19883_end_0 = const()[name = tensor<string, []>("op_19883_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_19883_end_mask_0 = const()[name = tensor<string, []>("op_19883_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19883_cast = slice_by_index(begin = var_19883_begin_0, end = var_19883_end_0, end_mask = var_19883_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19883_cast")];
            tensor<int32, [4]> var_19887_begin_0 = const()[name = tensor<string, []>("op_19887_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_19887_end_0 = const()[name = tensor<string, []>("op_19887_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_19887_end_mask_0 = const()[name = tensor<string, []>("op_19887_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19887_cast = slice_by_index(begin = var_19887_begin_0, end = var_19887_end_0, end_mask = var_19887_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19887_cast")];
            tensor<int32, [4]> var_19891_begin_0 = const()[name = tensor<string, []>("op_19891_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_19891_end_0 = const()[name = tensor<string, []>("op_19891_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_19891_end_mask_0 = const()[name = tensor<string, []>("op_19891_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19891_cast = slice_by_index(begin = var_19891_begin_0, end = var_19891_end_0, end_mask = var_19891_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19891_cast")];
            tensor<int32, [4]> var_19895_begin_0 = const()[name = tensor<string, []>("op_19895_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_19895_end_0 = const()[name = tensor<string, []>("op_19895_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_19895_end_mask_0 = const()[name = tensor<string, []>("op_19895_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19895_cast = slice_by_index(begin = var_19895_begin_0, end = var_19895_end_0, end_mask = var_19895_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19895_cast")];
            tensor<int32, [4]> var_19899_begin_0 = const()[name = tensor<string, []>("op_19899_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_19899_end_0 = const()[name = tensor<string, []>("op_19899_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_19899_end_mask_0 = const()[name = tensor<string, []>("op_19899_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19899_cast = slice_by_index(begin = var_19899_begin_0, end = var_19899_end_0, end_mask = var_19899_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19899_cast")];
            tensor<int32, [4]> var_19903_begin_0 = const()[name = tensor<string, []>("op_19903_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_19903_end_0 = const()[name = tensor<string, []>("op_19903_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_19903_end_mask_0 = const()[name = tensor<string, []>("op_19903_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19903_cast = slice_by_index(begin = var_19903_begin_0, end = var_19903_end_0, end_mask = var_19903_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19903_cast")];
            tensor<int32, [4]> var_19907_begin_0 = const()[name = tensor<string, []>("op_19907_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_19907_end_0 = const()[name = tensor<string, []>("op_19907_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_19907_end_mask_0 = const()[name = tensor<string, []>("op_19907_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19907_cast = slice_by_index(begin = var_19907_begin_0, end = var_19907_end_0, end_mask = var_19907_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19907_cast")];
            tensor<int32, [4]> var_19911_begin_0 = const()[name = tensor<string, []>("op_19911_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_19911_end_0 = const()[name = tensor<string, []>("op_19911_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_19911_end_mask_0 = const()[name = tensor<string, []>("op_19911_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19911_cast = slice_by_index(begin = var_19911_begin_0, end = var_19911_end_0, end_mask = var_19911_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19911_cast")];
            tensor<int32, [4]> var_19915_begin_0 = const()[name = tensor<string, []>("op_19915_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_19915_end_0 = const()[name = tensor<string, []>("op_19915_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_19915_end_mask_0 = const()[name = tensor<string, []>("op_19915_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19915_cast = slice_by_index(begin = var_19915_begin_0, end = var_19915_end_0, end_mask = var_19915_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19915_cast")];
            tensor<int32, [4]> var_19919_begin_0 = const()[name = tensor<string, []>("op_19919_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_19919_end_0 = const()[name = tensor<string, []>("op_19919_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_19919_end_mask_0 = const()[name = tensor<string, []>("op_19919_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19919_cast = slice_by_index(begin = var_19919_begin_0, end = var_19919_end_0, end_mask = var_19919_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19919_cast")];
            tensor<int32, [4]> var_19923_begin_0 = const()[name = tensor<string, []>("op_19923_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_19923_end_0 = const()[name = tensor<string, []>("op_19923_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_19923_end_mask_0 = const()[name = tensor<string, []>("op_19923_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19923_cast = slice_by_index(begin = var_19923_begin_0, end = var_19923_end_0, end_mask = var_19923_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19923_cast")];
            tensor<int32, [4]> var_19927_begin_0 = const()[name = tensor<string, []>("op_19927_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_19927_end_0 = const()[name = tensor<string, []>("op_19927_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_19927_end_mask_0 = const()[name = tensor<string, []>("op_19927_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19927_cast = slice_by_index(begin = var_19927_begin_0, end = var_19927_end_0, end_mask = var_19927_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19927_cast")];
            tensor<int32, [4]> var_19931_begin_0 = const()[name = tensor<string, []>("op_19931_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_19931_end_0 = const()[name = tensor<string, []>("op_19931_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_19931_end_mask_0 = const()[name = tensor<string, []>("op_19931_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19931_cast = slice_by_index(begin = var_19931_begin_0, end = var_19931_end_0, end_mask = var_19931_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19931_cast")];
            tensor<int32, [4]> var_19935_begin_0 = const()[name = tensor<string, []>("op_19935_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_19935_end_0 = const()[name = tensor<string, []>("op_19935_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_19935_end_mask_0 = const()[name = tensor<string, []>("op_19935_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_19935_cast = slice_by_index(begin = var_19935_begin_0, end = var_19935_end_0, end_mask = var_19935_end_mask_0, x = v_91_cast)[name = tensor<string, []>("op_19935_cast")];
            tensor<string, []> var_19939_equation_0 = const()[name = tensor<string, []>("op_19939_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19939_cast = einsum(equation = var_19939_equation_0, values = (var_19781_cast, var_19698_cast))[name = tensor<string, []>("op_19939_cast")];
            tensor<fp16, []> var_19940_to_fp16 = const()[name = tensor<string, []>("op_19940_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1641_cast = mul(x = var_19939_cast, y = var_19940_to_fp16)[name = tensor<string, []>("aw_1641_cast")];
            tensor<string, []> var_19943_equation_0 = const()[name = tensor<string, []>("op_19943_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19943_cast = einsum(equation = var_19943_equation_0, values = (var_19785_cast, var_19702_cast))[name = tensor<string, []>("op_19943_cast")];
            tensor<fp16, []> var_19944_to_fp16 = const()[name = tensor<string, []>("op_19944_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1643_cast = mul(x = var_19943_cast, y = var_19944_to_fp16)[name = tensor<string, []>("aw_1643_cast")];
            tensor<string, []> var_19947_equation_0 = const()[name = tensor<string, []>("op_19947_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19947_cast = einsum(equation = var_19947_equation_0, values = (var_19789_cast, var_19706_cast))[name = tensor<string, []>("op_19947_cast")];
            tensor<fp16, []> var_19948_to_fp16 = const()[name = tensor<string, []>("op_19948_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1645_cast = mul(x = var_19947_cast, y = var_19948_to_fp16)[name = tensor<string, []>("aw_1645_cast")];
            tensor<string, []> var_19951_equation_0 = const()[name = tensor<string, []>("op_19951_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19951_cast = einsum(equation = var_19951_equation_0, values = (var_19793_cast, var_19710_cast))[name = tensor<string, []>("op_19951_cast")];
            tensor<fp16, []> var_19952_to_fp16 = const()[name = tensor<string, []>("op_19952_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1647_cast = mul(x = var_19951_cast, y = var_19952_to_fp16)[name = tensor<string, []>("aw_1647_cast")];
            tensor<string, []> var_19955_equation_0 = const()[name = tensor<string, []>("op_19955_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19955_cast = einsum(equation = var_19955_equation_0, values = (var_19797_cast, var_19714_cast))[name = tensor<string, []>("op_19955_cast")];
            tensor<fp16, []> var_19956_to_fp16 = const()[name = tensor<string, []>("op_19956_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1649_cast = mul(x = var_19955_cast, y = var_19956_to_fp16)[name = tensor<string, []>("aw_1649_cast")];
            tensor<string, []> var_19959_equation_0 = const()[name = tensor<string, []>("op_19959_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19959_cast = einsum(equation = var_19959_equation_0, values = (var_19801_cast, var_19718_cast))[name = tensor<string, []>("op_19959_cast")];
            tensor<fp16, []> var_19960_to_fp16 = const()[name = tensor<string, []>("op_19960_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1651_cast = mul(x = var_19959_cast, y = var_19960_to_fp16)[name = tensor<string, []>("aw_1651_cast")];
            tensor<string, []> var_19963_equation_0 = const()[name = tensor<string, []>("op_19963_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19963_cast = einsum(equation = var_19963_equation_0, values = (var_19805_cast, var_19722_cast))[name = tensor<string, []>("op_19963_cast")];
            tensor<fp16, []> var_19964_to_fp16 = const()[name = tensor<string, []>("op_19964_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1653_cast = mul(x = var_19963_cast, y = var_19964_to_fp16)[name = tensor<string, []>("aw_1653_cast")];
            tensor<string, []> var_19967_equation_0 = const()[name = tensor<string, []>("op_19967_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19967_cast = einsum(equation = var_19967_equation_0, values = (var_19809_cast, var_19726_cast))[name = tensor<string, []>("op_19967_cast")];
            tensor<fp16, []> var_19968_to_fp16 = const()[name = tensor<string, []>("op_19968_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1655_cast = mul(x = var_19967_cast, y = var_19968_to_fp16)[name = tensor<string, []>("aw_1655_cast")];
            tensor<string, []> var_19971_equation_0 = const()[name = tensor<string, []>("op_19971_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19971_cast = einsum(equation = var_19971_equation_0, values = (var_19813_cast, var_19730_cast))[name = tensor<string, []>("op_19971_cast")];
            tensor<fp16, []> var_19972_to_fp16 = const()[name = tensor<string, []>("op_19972_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1657_cast = mul(x = var_19971_cast, y = var_19972_to_fp16)[name = tensor<string, []>("aw_1657_cast")];
            tensor<string, []> var_19975_equation_0 = const()[name = tensor<string, []>("op_19975_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19975_cast = einsum(equation = var_19975_equation_0, values = (var_19817_cast, var_19734_cast))[name = tensor<string, []>("op_19975_cast")];
            tensor<fp16, []> var_19976_to_fp16 = const()[name = tensor<string, []>("op_19976_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1659_cast = mul(x = var_19975_cast, y = var_19976_to_fp16)[name = tensor<string, []>("aw_1659_cast")];
            tensor<string, []> var_19979_equation_0 = const()[name = tensor<string, []>("op_19979_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19979_cast = einsum(equation = var_19979_equation_0, values = (var_19821_cast, var_19738_cast))[name = tensor<string, []>("op_19979_cast")];
            tensor<fp16, []> var_19980_to_fp16 = const()[name = tensor<string, []>("op_19980_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1661_cast = mul(x = var_19979_cast, y = var_19980_to_fp16)[name = tensor<string, []>("aw_1661_cast")];
            tensor<string, []> var_19983_equation_0 = const()[name = tensor<string, []>("op_19983_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19983_cast = einsum(equation = var_19983_equation_0, values = (var_19825_cast, var_19742_cast))[name = tensor<string, []>("op_19983_cast")];
            tensor<fp16, []> var_19984_to_fp16 = const()[name = tensor<string, []>("op_19984_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1663_cast = mul(x = var_19983_cast, y = var_19984_to_fp16)[name = tensor<string, []>("aw_1663_cast")];
            tensor<string, []> var_19987_equation_0 = const()[name = tensor<string, []>("op_19987_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19987_cast = einsum(equation = var_19987_equation_0, values = (var_19829_cast, var_19746_cast))[name = tensor<string, []>("op_19987_cast")];
            tensor<fp16, []> var_19988_to_fp16 = const()[name = tensor<string, []>("op_19988_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1665_cast = mul(x = var_19987_cast, y = var_19988_to_fp16)[name = tensor<string, []>("aw_1665_cast")];
            tensor<string, []> var_19991_equation_0 = const()[name = tensor<string, []>("op_19991_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19991_cast = einsum(equation = var_19991_equation_0, values = (var_19833_cast, var_19750_cast))[name = tensor<string, []>("op_19991_cast")];
            tensor<fp16, []> var_19992_to_fp16 = const()[name = tensor<string, []>("op_19992_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1667_cast = mul(x = var_19991_cast, y = var_19992_to_fp16)[name = tensor<string, []>("aw_1667_cast")];
            tensor<string, []> var_19995_equation_0 = const()[name = tensor<string, []>("op_19995_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19995_cast = einsum(equation = var_19995_equation_0, values = (var_19837_cast, var_19754_cast))[name = tensor<string, []>("op_19995_cast")];
            tensor<fp16, []> var_19996_to_fp16 = const()[name = tensor<string, []>("op_19996_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1669_cast = mul(x = var_19995_cast, y = var_19996_to_fp16)[name = tensor<string, []>("aw_1669_cast")];
            tensor<string, []> var_19999_equation_0 = const()[name = tensor<string, []>("op_19999_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_19999_cast = einsum(equation = var_19999_equation_0, values = (var_19841_cast, var_19758_cast))[name = tensor<string, []>("op_19999_cast")];
            tensor<fp16, []> var_20000_to_fp16 = const()[name = tensor<string, []>("op_20000_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1671_cast = mul(x = var_19999_cast, y = var_20000_to_fp16)[name = tensor<string, []>("aw_1671_cast")];
            tensor<string, []> var_20003_equation_0 = const()[name = tensor<string, []>("op_20003_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20003_cast = einsum(equation = var_20003_equation_0, values = (var_19845_cast, var_19762_cast))[name = tensor<string, []>("op_20003_cast")];
            tensor<fp16, []> var_20004_to_fp16 = const()[name = tensor<string, []>("op_20004_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1673_cast = mul(x = var_20003_cast, y = var_20004_to_fp16)[name = tensor<string, []>("aw_1673_cast")];
            tensor<string, []> var_20007_equation_0 = const()[name = tensor<string, []>("op_20007_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20007_cast = einsum(equation = var_20007_equation_0, values = (var_19849_cast, var_19766_cast))[name = tensor<string, []>("op_20007_cast")];
            tensor<fp16, []> var_20008_to_fp16 = const()[name = tensor<string, []>("op_20008_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1675_cast = mul(x = var_20007_cast, y = var_20008_to_fp16)[name = tensor<string, []>("aw_1675_cast")];
            tensor<string, []> var_20011_equation_0 = const()[name = tensor<string, []>("op_20011_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20011_cast = einsum(equation = var_20011_equation_0, values = (var_19853_cast, var_19770_cast))[name = tensor<string, []>("op_20011_cast")];
            tensor<fp16, []> var_20012_to_fp16 = const()[name = tensor<string, []>("op_20012_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1677_cast = mul(x = var_20011_cast, y = var_20012_to_fp16)[name = tensor<string, []>("aw_1677_cast")];
            tensor<string, []> var_20015_equation_0 = const()[name = tensor<string, []>("op_20015_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20015_cast = einsum(equation = var_20015_equation_0, values = (var_19857_cast, var_19774_cast))[name = tensor<string, []>("op_20015_cast")];
            tensor<fp16, []> var_20016_to_fp16 = const()[name = tensor<string, []>("op_20016_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1679_cast = mul(x = var_20015_cast, y = var_20016_to_fp16)[name = tensor<string, []>("aw_1679_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20018_cast = softmax(axis = var_2634, x = aw_1641_cast)[name = tensor<string, []>("op_20018_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20019_cast = softmax(axis = var_2634, x = aw_1643_cast)[name = tensor<string, []>("op_20019_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20020_cast = softmax(axis = var_2634, x = aw_1645_cast)[name = tensor<string, []>("op_20020_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20021_cast = softmax(axis = var_2634, x = aw_1647_cast)[name = tensor<string, []>("op_20021_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20022_cast = softmax(axis = var_2634, x = aw_1649_cast)[name = tensor<string, []>("op_20022_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20023_cast = softmax(axis = var_2634, x = aw_1651_cast)[name = tensor<string, []>("op_20023_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20024_cast = softmax(axis = var_2634, x = aw_1653_cast)[name = tensor<string, []>("op_20024_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20025_cast = softmax(axis = var_2634, x = aw_1655_cast)[name = tensor<string, []>("op_20025_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20026_cast = softmax(axis = var_2634, x = aw_1657_cast)[name = tensor<string, []>("op_20026_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20027_cast = softmax(axis = var_2634, x = aw_1659_cast)[name = tensor<string, []>("op_20027_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20028_cast = softmax(axis = var_2634, x = aw_1661_cast)[name = tensor<string, []>("op_20028_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20029_cast = softmax(axis = var_2634, x = aw_1663_cast)[name = tensor<string, []>("op_20029_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20030_cast = softmax(axis = var_2634, x = aw_1665_cast)[name = tensor<string, []>("op_20030_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20031_cast = softmax(axis = var_2634, x = aw_1667_cast)[name = tensor<string, []>("op_20031_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20032_cast = softmax(axis = var_2634, x = aw_1669_cast)[name = tensor<string, []>("op_20032_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20033_cast = softmax(axis = var_2634, x = aw_1671_cast)[name = tensor<string, []>("op_20033_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20034_cast = softmax(axis = var_2634, x = aw_1673_cast)[name = tensor<string, []>("op_20034_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20035_cast = softmax(axis = var_2634, x = aw_1675_cast)[name = tensor<string, []>("op_20035_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20036_cast = softmax(axis = var_2634, x = aw_1677_cast)[name = tensor<string, []>("op_20036_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20037_cast = softmax(axis = var_2634, x = aw_1679_cast)[name = tensor<string, []>("op_20037_cast")];
            tensor<string, []> var_20039_equation_0 = const()[name = tensor<string, []>("op_20039_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20039_cast = einsum(equation = var_20039_equation_0, values = (var_19859_cast, var_20018_cast))[name = tensor<string, []>("op_20039_cast")];
            tensor<string, []> var_20041_equation_0 = const()[name = tensor<string, []>("op_20041_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20041_cast = einsum(equation = var_20041_equation_0, values = (var_19863_cast, var_20019_cast))[name = tensor<string, []>("op_20041_cast")];
            tensor<string, []> var_20043_equation_0 = const()[name = tensor<string, []>("op_20043_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20043_cast = einsum(equation = var_20043_equation_0, values = (var_19867_cast, var_20020_cast))[name = tensor<string, []>("op_20043_cast")];
            tensor<string, []> var_20045_equation_0 = const()[name = tensor<string, []>("op_20045_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20045_cast = einsum(equation = var_20045_equation_0, values = (var_19871_cast, var_20021_cast))[name = tensor<string, []>("op_20045_cast")];
            tensor<string, []> var_20047_equation_0 = const()[name = tensor<string, []>("op_20047_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20047_cast = einsum(equation = var_20047_equation_0, values = (var_19875_cast, var_20022_cast))[name = tensor<string, []>("op_20047_cast")];
            tensor<string, []> var_20049_equation_0 = const()[name = tensor<string, []>("op_20049_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20049_cast = einsum(equation = var_20049_equation_0, values = (var_19879_cast, var_20023_cast))[name = tensor<string, []>("op_20049_cast")];
            tensor<string, []> var_20051_equation_0 = const()[name = tensor<string, []>("op_20051_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20051_cast = einsum(equation = var_20051_equation_0, values = (var_19883_cast, var_20024_cast))[name = tensor<string, []>("op_20051_cast")];
            tensor<string, []> var_20053_equation_0 = const()[name = tensor<string, []>("op_20053_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20053_cast = einsum(equation = var_20053_equation_0, values = (var_19887_cast, var_20025_cast))[name = tensor<string, []>("op_20053_cast")];
            tensor<string, []> var_20055_equation_0 = const()[name = tensor<string, []>("op_20055_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20055_cast = einsum(equation = var_20055_equation_0, values = (var_19891_cast, var_20026_cast))[name = tensor<string, []>("op_20055_cast")];
            tensor<string, []> var_20057_equation_0 = const()[name = tensor<string, []>("op_20057_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20057_cast = einsum(equation = var_20057_equation_0, values = (var_19895_cast, var_20027_cast))[name = tensor<string, []>("op_20057_cast")];
            tensor<string, []> var_20059_equation_0 = const()[name = tensor<string, []>("op_20059_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20059_cast = einsum(equation = var_20059_equation_0, values = (var_19899_cast, var_20028_cast))[name = tensor<string, []>("op_20059_cast")];
            tensor<string, []> var_20061_equation_0 = const()[name = tensor<string, []>("op_20061_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20061_cast = einsum(equation = var_20061_equation_0, values = (var_19903_cast, var_20029_cast))[name = tensor<string, []>("op_20061_cast")];
            tensor<string, []> var_20063_equation_0 = const()[name = tensor<string, []>("op_20063_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20063_cast = einsum(equation = var_20063_equation_0, values = (var_19907_cast, var_20030_cast))[name = tensor<string, []>("op_20063_cast")];
            tensor<string, []> var_20065_equation_0 = const()[name = tensor<string, []>("op_20065_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20065_cast = einsum(equation = var_20065_equation_0, values = (var_19911_cast, var_20031_cast))[name = tensor<string, []>("op_20065_cast")];
            tensor<string, []> var_20067_equation_0 = const()[name = tensor<string, []>("op_20067_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20067_cast = einsum(equation = var_20067_equation_0, values = (var_19915_cast, var_20032_cast))[name = tensor<string, []>("op_20067_cast")];
            tensor<string, []> var_20069_equation_0 = const()[name = tensor<string, []>("op_20069_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20069_cast = einsum(equation = var_20069_equation_0, values = (var_19919_cast, var_20033_cast))[name = tensor<string, []>("op_20069_cast")];
            tensor<string, []> var_20071_equation_0 = const()[name = tensor<string, []>("op_20071_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20071_cast = einsum(equation = var_20071_equation_0, values = (var_19923_cast, var_20034_cast))[name = tensor<string, []>("op_20071_cast")];
            tensor<string, []> var_20073_equation_0 = const()[name = tensor<string, []>("op_20073_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20073_cast = einsum(equation = var_20073_equation_0, values = (var_19927_cast, var_20035_cast))[name = tensor<string, []>("op_20073_cast")];
            tensor<string, []> var_20075_equation_0 = const()[name = tensor<string, []>("op_20075_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20075_cast = einsum(equation = var_20075_equation_0, values = (var_19931_cast, var_20036_cast))[name = tensor<string, []>("op_20075_cast")];
            tensor<string, []> var_20077_equation_0 = const()[name = tensor<string, []>("op_20077_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20077_cast = einsum(equation = var_20077_equation_0, values = (var_19935_cast, var_20037_cast))[name = tensor<string, []>("op_20077_cast")];
            tensor<bool, []> input_299_interleave_0 = const()[name = tensor<string, []>("input_299_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_299_cast = concat(axis = var_2634, interleave = input_299_interleave_0, values = (var_20039_cast, var_20041_cast, var_20043_cast, var_20045_cast, var_20047_cast, var_20049_cast, var_20051_cast, var_20053_cast, var_20055_cast, var_20057_cast, var_20059_cast, var_20061_cast, var_20063_cast, var_20065_cast, var_20067_cast, var_20069_cast, var_20071_cast, var_20073_cast, var_20075_cast, var_20077_cast))[name = tensor<string, []>("input_299_cast")];
            tensor<int32, [2]> var_20083 = const()[name = tensor<string, []>("op_20083"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_20085 = const()[name = tensor<string, []>("op_20085"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_20087_pad_type_0 = const()[name = tensor<string, []>("op_20087_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_20087_pad_0 = const()[name = tensor<string, []>("op_20087_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1545124480)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1548401344)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_20087_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_out_0_bias_to_fp16, dilations = var_20085, groups = var_2634, pad = var_20087_pad_0, pad_type = var_20087_pad_type_0, strides = var_20083, weight = down_blocks_2_attentions_1_transformer_blocks_8_attn2_to_out_0_weight_to_fp16, x = input_299_cast)[name = tensor<string, []>("op_20087_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_137_cast = add(x = var_20087_cast, y = inputs_135_cast)[name = tensor<string, []>("inputs_137_cast")];
            tensor<int32, [1]> var_20091 = const()[name = tensor<string, []>("op_20091"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_137_cast = reduce_mean(axes = var_20091, keep_dims = var_2629, x = inputs_137_cast)[name = tensor<string, []>("channels_mean_137_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_137_cast = sub(x = inputs_137_cast, y = channels_mean_137_cast)[name = tensor<string, []>("zero_mean_137_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_137_cast = mul(x = zero_mean_137_cast, y = zero_mean_137_cast)[name = tensor<string, []>("zero_mean_sq_137_cast")];
            tensor<int32, [1]> var_20095 = const()[name = tensor<string, []>("op_20095"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_20096_cast = reduce_mean(axes = var_20095, keep_dims = var_2629, x = zero_mean_sq_137_cast)[name = tensor<string, []>("op_20096_cast")];
            tensor<fp16, []> var_20097_to_fp16 = const()[name = tensor<string, []>("op_20097_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_20098_cast = add(x = var_20096_cast, y = var_20097_to_fp16)[name = tensor<string, []>("op_20098_cast")];
            tensor<fp16, []> denom_137_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_137_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_137_cast = rsqrt(epsilon = denom_137_epsilon_0_to_fp16, x = var_20098_cast)[name = tensor<string, []>("denom_137_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_137_cast = mul(x = zero_mean_137_cast, y = denom_137_cast)[name = tensor<string, []>("out_137_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_20102_to_fp16 = const()[name = tensor<string, []>("op_20102_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1548403968)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_20103_cast = add(x = out_137_cast, y = var_20102_to_fp16)[name = tensor<string, []>("op_20103_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_20105_to_fp16 = const()[name = tensor<string, []>("op_20105_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1548406592)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_301_cast = mul(x = var_20103_cast, y = var_20105_to_fp16)[name = tensor<string, []>("input_301_cast")];
            tensor<int32, [2]> var_20113 = const()[name = tensor<string, []>("op_20113"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_20115 = const()[name = tensor<string, []>("op_20115"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_20117_pad_type_0 = const()[name = tensor<string, []>("op_20117_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_20117_pad_0 = const()[name = tensor<string, []>("op_20117_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_8_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_8_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1548409216)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_1_transformer_blocks_8_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_8_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1574623680)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_20117_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_8_ff_net_0_proj_bias_to_fp16, dilations = var_20115, groups = var_2634, pad = var_20117_pad_0, pad_type = var_20117_pad_type_0, strides = var_20113, weight = down_blocks_2_attentions_1_transformer_blocks_8_ff_net_0_proj_weight_to_fp16, x = input_301_cast)[name = tensor<string, []>("op_20117_cast")];
            tensor<int32, [2]> var_20118_split_sizes_0 = const()[name = tensor<string, []>("op_20118_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_20118_axis_0 = const()[name = tensor<string, []>("op_20118_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_20118_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_20118_cast_1 = split(axis = var_20118_axis_0, split_sizes = var_20118_split_sizes_0, x = var_20117_cast)[name = tensor<string, []>("op_20118_cast")];
            tensor<string, []> var_20120_mode_0 = const()[name = tensor<string, []>("op_20120_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_20120_cast = gelu(mode = var_20120_mode_0, x = var_20118_cast_1)[name = tensor<string, []>("op_20120_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_303_cast = mul(x = var_20118_cast_0, y = var_20120_cast)[name = tensor<string, []>("input_303_cast")];
            tensor<int32, [2]> var_20124 = const()[name = tensor<string, []>("op_20124"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_20126 = const()[name = tensor<string, []>("op_20126"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_20128_pad_type_0 = const()[name = tensor<string, []>("op_20128_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_20128_pad_0 = const()[name = tensor<string, []>("op_20128_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_8_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_8_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1574644224)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_8_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_8_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1587751488)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_20128_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_8_ff_net_2_bias_to_fp16, dilations = var_20126, groups = var_2634, pad = var_20128_pad_0, pad_type = var_20128_pad_type_0, strides = var_20124, weight = down_blocks_2_attentions_1_transformer_blocks_8_ff_net_2_weight_to_fp16, x = input_303_cast)[name = tensor<string, []>("op_20128_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_139_cast = add(x = var_20128_cast, y = inputs_137_cast)[name = tensor<string, []>("inputs_139_cast")];
            tensor<int32, [1]> var_20138 = const()[name = tensor<string, []>("op_20138"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_139_cast = reduce_mean(axes = var_20138, keep_dims = var_2629, x = inputs_139_cast)[name = tensor<string, []>("channels_mean_139_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_139_cast = sub(x = inputs_139_cast, y = channels_mean_139_cast)[name = tensor<string, []>("zero_mean_139_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_139_cast = mul(x = zero_mean_139_cast, y = zero_mean_139_cast)[name = tensor<string, []>("zero_mean_sq_139_cast")];
            tensor<int32, [1]> var_20142 = const()[name = tensor<string, []>("op_20142"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_20143_cast = reduce_mean(axes = var_20142, keep_dims = var_2629, x = zero_mean_sq_139_cast)[name = tensor<string, []>("op_20143_cast")];
            tensor<fp16, []> var_20144_to_fp16 = const()[name = tensor<string, []>("op_20144_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_20145_cast = add(x = var_20143_cast, y = var_20144_to_fp16)[name = tensor<string, []>("op_20145_cast")];
            tensor<fp16, []> denom_139_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_139_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_139_cast = rsqrt(epsilon = denom_139_epsilon_0_to_fp16, x = var_20145_cast)[name = tensor<string, []>("denom_139_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_139_cast = mul(x = zero_mean_139_cast, y = denom_139_cast)[name = tensor<string, []>("out_139_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_20149_to_fp16 = const()[name = tensor<string, []>("op_20149_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1587754112)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_20150_cast = add(x = out_139_cast, y = var_20149_to_fp16)[name = tensor<string, []>("op_20150_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_20152_to_fp16 = const()[name = tensor<string, []>("op_20152_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1587756736)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_191_cast = mul(x = var_20150_cast, y = var_20152_to_fp16)[name = tensor<string, []>("hidden_states_191_cast")];
            tensor<int32, [2]> var_20159 = const()[name = tensor<string, []>("op_20159"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_20161 = const()[name = tensor<string, []>("op_20161"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_93_pad_type_0 = const()[name = tensor<string, []>("q_93_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_93_pad_0 = const()[name = tensor<string, []>("q_93_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1587759360)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_93_cast = conv(dilations = var_20161, groups = var_2634, pad = q_93_pad_0, pad_type = q_93_pad_type_0, strides = var_20159, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_q_weight_to_fp16, x = hidden_states_191_cast)[name = tensor<string, []>("q_93_cast")];
            tensor<int32, [2]> var_20165 = const()[name = tensor<string, []>("op_20165"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_20167 = const()[name = tensor<string, []>("op_20167"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_185_pad_type_0 = const()[name = tensor<string, []>("k_185_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_185_pad_0 = const()[name = tensor<string, []>("k_185_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1591036224)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_185_cast = conv(dilations = var_20167, groups = var_2634, pad = k_185_pad_0, pad_type = k_185_pad_type_0, strides = var_20165, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_k_weight_to_fp16, x = hidden_states_191_cast)[name = tensor<string, []>("k_185_cast")];
            tensor<int32, [2]> var_20171 = const()[name = tensor<string, []>("op_20171"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_20173 = const()[name = tensor<string, []>("op_20173"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_93_pad_type_0 = const()[name = tensor<string, []>("v_93_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_93_pad_0 = const()[name = tensor<string, []>("v_93_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1594313088)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_93_cast = conv(dilations = var_20173, groups = var_2634, pad = v_93_pad_0, pad_type = v_93_pad_type_0, strides = var_20171, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_v_weight_to_fp16, x = hidden_states_191_cast)[name = tensor<string, []>("v_93_cast")];
            tensor<int32, [4]> var_20177_begin_0 = const()[name = tensor<string, []>("op_20177_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_20177_end_0 = const()[name = tensor<string, []>("op_20177_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_20177_end_mask_0 = const()[name = tensor<string, []>("op_20177_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20177_cast = slice_by_index(begin = var_20177_begin_0, end = var_20177_end_0, end_mask = var_20177_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20177_cast")];
            tensor<int32, [4]> var_20181_begin_0 = const()[name = tensor<string, []>("op_20181_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_20181_end_0 = const()[name = tensor<string, []>("op_20181_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_20181_end_mask_0 = const()[name = tensor<string, []>("op_20181_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20181_cast = slice_by_index(begin = var_20181_begin_0, end = var_20181_end_0, end_mask = var_20181_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20181_cast")];
            tensor<int32, [4]> var_20185_begin_0 = const()[name = tensor<string, []>("op_20185_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_20185_end_0 = const()[name = tensor<string, []>("op_20185_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_20185_end_mask_0 = const()[name = tensor<string, []>("op_20185_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20185_cast = slice_by_index(begin = var_20185_begin_0, end = var_20185_end_0, end_mask = var_20185_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20185_cast")];
            tensor<int32, [4]> var_20189_begin_0 = const()[name = tensor<string, []>("op_20189_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_20189_end_0 = const()[name = tensor<string, []>("op_20189_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_20189_end_mask_0 = const()[name = tensor<string, []>("op_20189_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20189_cast = slice_by_index(begin = var_20189_begin_0, end = var_20189_end_0, end_mask = var_20189_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20189_cast")];
            tensor<int32, [4]> var_20193_begin_0 = const()[name = tensor<string, []>("op_20193_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_20193_end_0 = const()[name = tensor<string, []>("op_20193_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_20193_end_mask_0 = const()[name = tensor<string, []>("op_20193_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20193_cast = slice_by_index(begin = var_20193_begin_0, end = var_20193_end_0, end_mask = var_20193_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20193_cast")];
            tensor<int32, [4]> var_20197_begin_0 = const()[name = tensor<string, []>("op_20197_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_20197_end_0 = const()[name = tensor<string, []>("op_20197_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_20197_end_mask_0 = const()[name = tensor<string, []>("op_20197_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20197_cast = slice_by_index(begin = var_20197_begin_0, end = var_20197_end_0, end_mask = var_20197_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20197_cast")];
            tensor<int32, [4]> var_20201_begin_0 = const()[name = tensor<string, []>("op_20201_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_20201_end_0 = const()[name = tensor<string, []>("op_20201_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_20201_end_mask_0 = const()[name = tensor<string, []>("op_20201_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20201_cast = slice_by_index(begin = var_20201_begin_0, end = var_20201_end_0, end_mask = var_20201_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20201_cast")];
            tensor<int32, [4]> var_20205_begin_0 = const()[name = tensor<string, []>("op_20205_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_20205_end_0 = const()[name = tensor<string, []>("op_20205_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_20205_end_mask_0 = const()[name = tensor<string, []>("op_20205_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20205_cast = slice_by_index(begin = var_20205_begin_0, end = var_20205_end_0, end_mask = var_20205_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20205_cast")];
            tensor<int32, [4]> var_20209_begin_0 = const()[name = tensor<string, []>("op_20209_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_20209_end_0 = const()[name = tensor<string, []>("op_20209_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_20209_end_mask_0 = const()[name = tensor<string, []>("op_20209_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20209_cast = slice_by_index(begin = var_20209_begin_0, end = var_20209_end_0, end_mask = var_20209_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20209_cast")];
            tensor<int32, [4]> var_20213_begin_0 = const()[name = tensor<string, []>("op_20213_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_20213_end_0 = const()[name = tensor<string, []>("op_20213_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_20213_end_mask_0 = const()[name = tensor<string, []>("op_20213_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20213_cast = slice_by_index(begin = var_20213_begin_0, end = var_20213_end_0, end_mask = var_20213_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20213_cast")];
            tensor<int32, [4]> var_20217_begin_0 = const()[name = tensor<string, []>("op_20217_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_20217_end_0 = const()[name = tensor<string, []>("op_20217_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_20217_end_mask_0 = const()[name = tensor<string, []>("op_20217_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20217_cast = slice_by_index(begin = var_20217_begin_0, end = var_20217_end_0, end_mask = var_20217_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20217_cast")];
            tensor<int32, [4]> var_20221_begin_0 = const()[name = tensor<string, []>("op_20221_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_20221_end_0 = const()[name = tensor<string, []>("op_20221_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_20221_end_mask_0 = const()[name = tensor<string, []>("op_20221_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20221_cast = slice_by_index(begin = var_20221_begin_0, end = var_20221_end_0, end_mask = var_20221_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20221_cast")];
            tensor<int32, [4]> var_20225_begin_0 = const()[name = tensor<string, []>("op_20225_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_20225_end_0 = const()[name = tensor<string, []>("op_20225_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_20225_end_mask_0 = const()[name = tensor<string, []>("op_20225_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20225_cast = slice_by_index(begin = var_20225_begin_0, end = var_20225_end_0, end_mask = var_20225_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20225_cast")];
            tensor<int32, [4]> var_20229_begin_0 = const()[name = tensor<string, []>("op_20229_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_20229_end_0 = const()[name = tensor<string, []>("op_20229_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_20229_end_mask_0 = const()[name = tensor<string, []>("op_20229_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20229_cast = slice_by_index(begin = var_20229_begin_0, end = var_20229_end_0, end_mask = var_20229_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20229_cast")];
            tensor<int32, [4]> var_20233_begin_0 = const()[name = tensor<string, []>("op_20233_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_20233_end_0 = const()[name = tensor<string, []>("op_20233_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_20233_end_mask_0 = const()[name = tensor<string, []>("op_20233_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20233_cast = slice_by_index(begin = var_20233_begin_0, end = var_20233_end_0, end_mask = var_20233_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20233_cast")];
            tensor<int32, [4]> var_20237_begin_0 = const()[name = tensor<string, []>("op_20237_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_20237_end_0 = const()[name = tensor<string, []>("op_20237_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_20237_end_mask_0 = const()[name = tensor<string, []>("op_20237_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20237_cast = slice_by_index(begin = var_20237_begin_0, end = var_20237_end_0, end_mask = var_20237_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20237_cast")];
            tensor<int32, [4]> var_20241_begin_0 = const()[name = tensor<string, []>("op_20241_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_20241_end_0 = const()[name = tensor<string, []>("op_20241_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_20241_end_mask_0 = const()[name = tensor<string, []>("op_20241_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20241_cast = slice_by_index(begin = var_20241_begin_0, end = var_20241_end_0, end_mask = var_20241_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20241_cast")];
            tensor<int32, [4]> var_20245_begin_0 = const()[name = tensor<string, []>("op_20245_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_20245_end_0 = const()[name = tensor<string, []>("op_20245_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_20245_end_mask_0 = const()[name = tensor<string, []>("op_20245_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20245_cast = slice_by_index(begin = var_20245_begin_0, end = var_20245_end_0, end_mask = var_20245_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20245_cast")];
            tensor<int32, [4]> var_20249_begin_0 = const()[name = tensor<string, []>("op_20249_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_20249_end_0 = const()[name = tensor<string, []>("op_20249_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_20249_end_mask_0 = const()[name = tensor<string, []>("op_20249_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20249_cast = slice_by_index(begin = var_20249_begin_0, end = var_20249_end_0, end_mask = var_20249_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20249_cast")];
            tensor<int32, [4]> var_20253_begin_0 = const()[name = tensor<string, []>("op_20253_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_20253_end_0 = const()[name = tensor<string, []>("op_20253_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_20253_end_mask_0 = const()[name = tensor<string, []>("op_20253_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20253_cast = slice_by_index(begin = var_20253_begin_0, end = var_20253_end_0, end_mask = var_20253_end_mask_0, x = q_93_cast)[name = tensor<string, []>("op_20253_cast")];
            tensor<int32, [4]> k_187_perm_0 = const()[name = tensor<string, []>("k_187_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_20260_begin_0 = const()[name = tensor<string, []>("op_20260_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_20260_end_0 = const()[name = tensor<string, []>("op_20260_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_20260_end_mask_0 = const()[name = tensor<string, []>("op_20260_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_93 = transpose(perm = k_187_perm_0, x = k_185_cast)[name = tensor<string, []>("transpose_93")];
            tensor<fp16, [2, 1024, 1, 64]> var_20260_cast = slice_by_index(begin = var_20260_begin_0, end = var_20260_end_0, end_mask = var_20260_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20260_cast")];
            tensor<int32, [4]> var_20264_begin_0 = const()[name = tensor<string, []>("op_20264_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_20264_end_0 = const()[name = tensor<string, []>("op_20264_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_20264_end_mask_0 = const()[name = tensor<string, []>("op_20264_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_20264_cast = slice_by_index(begin = var_20264_begin_0, end = var_20264_end_0, end_mask = var_20264_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20264_cast")];
            tensor<int32, [4]> var_20268_begin_0 = const()[name = tensor<string, []>("op_20268_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_20268_end_0 = const()[name = tensor<string, []>("op_20268_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_20268_end_mask_0 = const()[name = tensor<string, []>("op_20268_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_20268_cast = slice_by_index(begin = var_20268_begin_0, end = var_20268_end_0, end_mask = var_20268_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20268_cast")];
            tensor<int32, [4]> var_20272_begin_0 = const()[name = tensor<string, []>("op_20272_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_20272_end_0 = const()[name = tensor<string, []>("op_20272_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_20272_end_mask_0 = const()[name = tensor<string, []>("op_20272_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_20272_cast = slice_by_index(begin = var_20272_begin_0, end = var_20272_end_0, end_mask = var_20272_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20272_cast")];
            tensor<int32, [4]> var_20276_begin_0 = const()[name = tensor<string, []>("op_20276_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_20276_end_0 = const()[name = tensor<string, []>("op_20276_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_20276_end_mask_0 = const()[name = tensor<string, []>("op_20276_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_20276_cast = slice_by_index(begin = var_20276_begin_0, end = var_20276_end_0, end_mask = var_20276_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20276_cast")];
            tensor<int32, [4]> var_20280_begin_0 = const()[name = tensor<string, []>("op_20280_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_20280_end_0 = const()[name = tensor<string, []>("op_20280_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_20280_end_mask_0 = const()[name = tensor<string, []>("op_20280_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_20280_cast = slice_by_index(begin = var_20280_begin_0, end = var_20280_end_0, end_mask = var_20280_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20280_cast")];
            tensor<int32, [4]> var_20284_begin_0 = const()[name = tensor<string, []>("op_20284_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_20284_end_0 = const()[name = tensor<string, []>("op_20284_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_20284_end_mask_0 = const()[name = tensor<string, []>("op_20284_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_20284_cast = slice_by_index(begin = var_20284_begin_0, end = var_20284_end_0, end_mask = var_20284_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20284_cast")];
            tensor<int32, [4]> var_20288_begin_0 = const()[name = tensor<string, []>("op_20288_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_20288_end_0 = const()[name = tensor<string, []>("op_20288_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_20288_end_mask_0 = const()[name = tensor<string, []>("op_20288_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_20288_cast = slice_by_index(begin = var_20288_begin_0, end = var_20288_end_0, end_mask = var_20288_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20288_cast")];
            tensor<int32, [4]> var_20292_begin_0 = const()[name = tensor<string, []>("op_20292_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_20292_end_0 = const()[name = tensor<string, []>("op_20292_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_20292_end_mask_0 = const()[name = tensor<string, []>("op_20292_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_20292_cast = slice_by_index(begin = var_20292_begin_0, end = var_20292_end_0, end_mask = var_20292_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20292_cast")];
            tensor<int32, [4]> var_20296_begin_0 = const()[name = tensor<string, []>("op_20296_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_20296_end_0 = const()[name = tensor<string, []>("op_20296_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_20296_end_mask_0 = const()[name = tensor<string, []>("op_20296_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_20296_cast = slice_by_index(begin = var_20296_begin_0, end = var_20296_end_0, end_mask = var_20296_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20296_cast")];
            tensor<int32, [4]> var_20300_begin_0 = const()[name = tensor<string, []>("op_20300_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_20300_end_0 = const()[name = tensor<string, []>("op_20300_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_20300_end_mask_0 = const()[name = tensor<string, []>("op_20300_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_20300_cast = slice_by_index(begin = var_20300_begin_0, end = var_20300_end_0, end_mask = var_20300_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20300_cast")];
            tensor<int32, [4]> var_20304_begin_0 = const()[name = tensor<string, []>("op_20304_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_20304_end_0 = const()[name = tensor<string, []>("op_20304_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_20304_end_mask_0 = const()[name = tensor<string, []>("op_20304_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_20304_cast = slice_by_index(begin = var_20304_begin_0, end = var_20304_end_0, end_mask = var_20304_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20304_cast")];
            tensor<int32, [4]> var_20308_begin_0 = const()[name = tensor<string, []>("op_20308_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_20308_end_0 = const()[name = tensor<string, []>("op_20308_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_20308_end_mask_0 = const()[name = tensor<string, []>("op_20308_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_20308_cast = slice_by_index(begin = var_20308_begin_0, end = var_20308_end_0, end_mask = var_20308_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20308_cast")];
            tensor<int32, [4]> var_20312_begin_0 = const()[name = tensor<string, []>("op_20312_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_20312_end_0 = const()[name = tensor<string, []>("op_20312_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_20312_end_mask_0 = const()[name = tensor<string, []>("op_20312_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_20312_cast = slice_by_index(begin = var_20312_begin_0, end = var_20312_end_0, end_mask = var_20312_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20312_cast")];
            tensor<int32, [4]> var_20316_begin_0 = const()[name = tensor<string, []>("op_20316_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_20316_end_0 = const()[name = tensor<string, []>("op_20316_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_20316_end_mask_0 = const()[name = tensor<string, []>("op_20316_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_20316_cast = slice_by_index(begin = var_20316_begin_0, end = var_20316_end_0, end_mask = var_20316_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20316_cast")];
            tensor<int32, [4]> var_20320_begin_0 = const()[name = tensor<string, []>("op_20320_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_20320_end_0 = const()[name = tensor<string, []>("op_20320_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_20320_end_mask_0 = const()[name = tensor<string, []>("op_20320_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_20320_cast = slice_by_index(begin = var_20320_begin_0, end = var_20320_end_0, end_mask = var_20320_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20320_cast")];
            tensor<int32, [4]> var_20324_begin_0 = const()[name = tensor<string, []>("op_20324_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_20324_end_0 = const()[name = tensor<string, []>("op_20324_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_20324_end_mask_0 = const()[name = tensor<string, []>("op_20324_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_20324_cast = slice_by_index(begin = var_20324_begin_0, end = var_20324_end_0, end_mask = var_20324_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20324_cast")];
            tensor<int32, [4]> var_20328_begin_0 = const()[name = tensor<string, []>("op_20328_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_20328_end_0 = const()[name = tensor<string, []>("op_20328_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_20328_end_mask_0 = const()[name = tensor<string, []>("op_20328_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_20328_cast = slice_by_index(begin = var_20328_begin_0, end = var_20328_end_0, end_mask = var_20328_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20328_cast")];
            tensor<int32, [4]> var_20332_begin_0 = const()[name = tensor<string, []>("op_20332_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_20332_end_0 = const()[name = tensor<string, []>("op_20332_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_20332_end_mask_0 = const()[name = tensor<string, []>("op_20332_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_20332_cast = slice_by_index(begin = var_20332_begin_0, end = var_20332_end_0, end_mask = var_20332_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20332_cast")];
            tensor<int32, [4]> var_20336_begin_0 = const()[name = tensor<string, []>("op_20336_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_20336_end_0 = const()[name = tensor<string, []>("op_20336_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_20336_end_mask_0 = const()[name = tensor<string, []>("op_20336_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_20336_cast = slice_by_index(begin = var_20336_begin_0, end = var_20336_end_0, end_mask = var_20336_end_mask_0, x = transpose_93)[name = tensor<string, []>("op_20336_cast")];
            tensor<int32, [4]> var_20338_begin_0 = const()[name = tensor<string, []>("op_20338_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_20338_end_0 = const()[name = tensor<string, []>("op_20338_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_20338_end_mask_0 = const()[name = tensor<string, []>("op_20338_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20338_cast = slice_by_index(begin = var_20338_begin_0, end = var_20338_end_0, end_mask = var_20338_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20338_cast")];
            tensor<int32, [4]> var_20342_begin_0 = const()[name = tensor<string, []>("op_20342_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_20342_end_0 = const()[name = tensor<string, []>("op_20342_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_20342_end_mask_0 = const()[name = tensor<string, []>("op_20342_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20342_cast = slice_by_index(begin = var_20342_begin_0, end = var_20342_end_0, end_mask = var_20342_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20342_cast")];
            tensor<int32, [4]> var_20346_begin_0 = const()[name = tensor<string, []>("op_20346_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_20346_end_0 = const()[name = tensor<string, []>("op_20346_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_20346_end_mask_0 = const()[name = tensor<string, []>("op_20346_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20346_cast = slice_by_index(begin = var_20346_begin_0, end = var_20346_end_0, end_mask = var_20346_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20346_cast")];
            tensor<int32, [4]> var_20350_begin_0 = const()[name = tensor<string, []>("op_20350_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_20350_end_0 = const()[name = tensor<string, []>("op_20350_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_20350_end_mask_0 = const()[name = tensor<string, []>("op_20350_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20350_cast = slice_by_index(begin = var_20350_begin_0, end = var_20350_end_0, end_mask = var_20350_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20350_cast")];
            tensor<int32, [4]> var_20354_begin_0 = const()[name = tensor<string, []>("op_20354_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_20354_end_0 = const()[name = tensor<string, []>("op_20354_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_20354_end_mask_0 = const()[name = tensor<string, []>("op_20354_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20354_cast = slice_by_index(begin = var_20354_begin_0, end = var_20354_end_0, end_mask = var_20354_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20354_cast")];
            tensor<int32, [4]> var_20358_begin_0 = const()[name = tensor<string, []>("op_20358_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_20358_end_0 = const()[name = tensor<string, []>("op_20358_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_20358_end_mask_0 = const()[name = tensor<string, []>("op_20358_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20358_cast = slice_by_index(begin = var_20358_begin_0, end = var_20358_end_0, end_mask = var_20358_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20358_cast")];
            tensor<int32, [4]> var_20362_begin_0 = const()[name = tensor<string, []>("op_20362_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_20362_end_0 = const()[name = tensor<string, []>("op_20362_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_20362_end_mask_0 = const()[name = tensor<string, []>("op_20362_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20362_cast = slice_by_index(begin = var_20362_begin_0, end = var_20362_end_0, end_mask = var_20362_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20362_cast")];
            tensor<int32, [4]> var_20366_begin_0 = const()[name = tensor<string, []>("op_20366_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_20366_end_0 = const()[name = tensor<string, []>("op_20366_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_20366_end_mask_0 = const()[name = tensor<string, []>("op_20366_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20366_cast = slice_by_index(begin = var_20366_begin_0, end = var_20366_end_0, end_mask = var_20366_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20366_cast")];
            tensor<int32, [4]> var_20370_begin_0 = const()[name = tensor<string, []>("op_20370_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_20370_end_0 = const()[name = tensor<string, []>("op_20370_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_20370_end_mask_0 = const()[name = tensor<string, []>("op_20370_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20370_cast = slice_by_index(begin = var_20370_begin_0, end = var_20370_end_0, end_mask = var_20370_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20370_cast")];
            tensor<int32, [4]> var_20374_begin_0 = const()[name = tensor<string, []>("op_20374_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_20374_end_0 = const()[name = tensor<string, []>("op_20374_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_20374_end_mask_0 = const()[name = tensor<string, []>("op_20374_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20374_cast = slice_by_index(begin = var_20374_begin_0, end = var_20374_end_0, end_mask = var_20374_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20374_cast")];
            tensor<int32, [4]> var_20378_begin_0 = const()[name = tensor<string, []>("op_20378_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_20378_end_0 = const()[name = tensor<string, []>("op_20378_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_20378_end_mask_0 = const()[name = tensor<string, []>("op_20378_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20378_cast = slice_by_index(begin = var_20378_begin_0, end = var_20378_end_0, end_mask = var_20378_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20378_cast")];
            tensor<int32, [4]> var_20382_begin_0 = const()[name = tensor<string, []>("op_20382_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_20382_end_0 = const()[name = tensor<string, []>("op_20382_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_20382_end_mask_0 = const()[name = tensor<string, []>("op_20382_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20382_cast = slice_by_index(begin = var_20382_begin_0, end = var_20382_end_0, end_mask = var_20382_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20382_cast")];
            tensor<int32, [4]> var_20386_begin_0 = const()[name = tensor<string, []>("op_20386_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_20386_end_0 = const()[name = tensor<string, []>("op_20386_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_20386_end_mask_0 = const()[name = tensor<string, []>("op_20386_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20386_cast = slice_by_index(begin = var_20386_begin_0, end = var_20386_end_0, end_mask = var_20386_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20386_cast")];
            tensor<int32, [4]> var_20390_begin_0 = const()[name = tensor<string, []>("op_20390_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_20390_end_0 = const()[name = tensor<string, []>("op_20390_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_20390_end_mask_0 = const()[name = tensor<string, []>("op_20390_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20390_cast = slice_by_index(begin = var_20390_begin_0, end = var_20390_end_0, end_mask = var_20390_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20390_cast")];
            tensor<int32, [4]> var_20394_begin_0 = const()[name = tensor<string, []>("op_20394_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_20394_end_0 = const()[name = tensor<string, []>("op_20394_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_20394_end_mask_0 = const()[name = tensor<string, []>("op_20394_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20394_cast = slice_by_index(begin = var_20394_begin_0, end = var_20394_end_0, end_mask = var_20394_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20394_cast")];
            tensor<int32, [4]> var_20398_begin_0 = const()[name = tensor<string, []>("op_20398_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_20398_end_0 = const()[name = tensor<string, []>("op_20398_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_20398_end_mask_0 = const()[name = tensor<string, []>("op_20398_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20398_cast = slice_by_index(begin = var_20398_begin_0, end = var_20398_end_0, end_mask = var_20398_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20398_cast")];
            tensor<int32, [4]> var_20402_begin_0 = const()[name = tensor<string, []>("op_20402_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_20402_end_0 = const()[name = tensor<string, []>("op_20402_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_20402_end_mask_0 = const()[name = tensor<string, []>("op_20402_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20402_cast = slice_by_index(begin = var_20402_begin_0, end = var_20402_end_0, end_mask = var_20402_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20402_cast")];
            tensor<int32, [4]> var_20406_begin_0 = const()[name = tensor<string, []>("op_20406_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_20406_end_0 = const()[name = tensor<string, []>("op_20406_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_20406_end_mask_0 = const()[name = tensor<string, []>("op_20406_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20406_cast = slice_by_index(begin = var_20406_begin_0, end = var_20406_end_0, end_mask = var_20406_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20406_cast")];
            tensor<int32, [4]> var_20410_begin_0 = const()[name = tensor<string, []>("op_20410_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_20410_end_0 = const()[name = tensor<string, []>("op_20410_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_20410_end_mask_0 = const()[name = tensor<string, []>("op_20410_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20410_cast = slice_by_index(begin = var_20410_begin_0, end = var_20410_end_0, end_mask = var_20410_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20410_cast")];
            tensor<int32, [4]> var_20414_begin_0 = const()[name = tensor<string, []>("op_20414_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_20414_end_0 = const()[name = tensor<string, []>("op_20414_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_20414_end_mask_0 = const()[name = tensor<string, []>("op_20414_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20414_cast = slice_by_index(begin = var_20414_begin_0, end = var_20414_end_0, end_mask = var_20414_end_mask_0, x = v_93_cast)[name = tensor<string, []>("op_20414_cast")];
            tensor<string, []> var_20418_equation_0 = const()[name = tensor<string, []>("op_20418_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20418_cast = einsum(equation = var_20418_equation_0, values = (var_20260_cast, var_20177_cast))[name = tensor<string, []>("op_20418_cast")];
            tensor<fp16, []> var_20419_to_fp16 = const()[name = tensor<string, []>("op_20419_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1681_cast = mul(x = var_20418_cast, y = var_20419_to_fp16)[name = tensor<string, []>("aw_1681_cast")];
            tensor<string, []> var_20422_equation_0 = const()[name = tensor<string, []>("op_20422_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20422_cast = einsum(equation = var_20422_equation_0, values = (var_20264_cast, var_20181_cast))[name = tensor<string, []>("op_20422_cast")];
            tensor<fp16, []> var_20423_to_fp16 = const()[name = tensor<string, []>("op_20423_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1683_cast = mul(x = var_20422_cast, y = var_20423_to_fp16)[name = tensor<string, []>("aw_1683_cast")];
            tensor<string, []> var_20426_equation_0 = const()[name = tensor<string, []>("op_20426_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20426_cast = einsum(equation = var_20426_equation_0, values = (var_20268_cast, var_20185_cast))[name = tensor<string, []>("op_20426_cast")];
            tensor<fp16, []> var_20427_to_fp16 = const()[name = tensor<string, []>("op_20427_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1685_cast = mul(x = var_20426_cast, y = var_20427_to_fp16)[name = tensor<string, []>("aw_1685_cast")];
            tensor<string, []> var_20430_equation_0 = const()[name = tensor<string, []>("op_20430_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20430_cast = einsum(equation = var_20430_equation_0, values = (var_20272_cast, var_20189_cast))[name = tensor<string, []>("op_20430_cast")];
            tensor<fp16, []> var_20431_to_fp16 = const()[name = tensor<string, []>("op_20431_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1687_cast = mul(x = var_20430_cast, y = var_20431_to_fp16)[name = tensor<string, []>("aw_1687_cast")];
            tensor<string, []> var_20434_equation_0 = const()[name = tensor<string, []>("op_20434_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20434_cast = einsum(equation = var_20434_equation_0, values = (var_20276_cast, var_20193_cast))[name = tensor<string, []>("op_20434_cast")];
            tensor<fp16, []> var_20435_to_fp16 = const()[name = tensor<string, []>("op_20435_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1689_cast = mul(x = var_20434_cast, y = var_20435_to_fp16)[name = tensor<string, []>("aw_1689_cast")];
            tensor<string, []> var_20438_equation_0 = const()[name = tensor<string, []>("op_20438_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20438_cast = einsum(equation = var_20438_equation_0, values = (var_20280_cast, var_20197_cast))[name = tensor<string, []>("op_20438_cast")];
            tensor<fp16, []> var_20439_to_fp16 = const()[name = tensor<string, []>("op_20439_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1691_cast = mul(x = var_20438_cast, y = var_20439_to_fp16)[name = tensor<string, []>("aw_1691_cast")];
            tensor<string, []> var_20442_equation_0 = const()[name = tensor<string, []>("op_20442_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20442_cast = einsum(equation = var_20442_equation_0, values = (var_20284_cast, var_20201_cast))[name = tensor<string, []>("op_20442_cast")];
            tensor<fp16, []> var_20443_to_fp16 = const()[name = tensor<string, []>("op_20443_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1693_cast = mul(x = var_20442_cast, y = var_20443_to_fp16)[name = tensor<string, []>("aw_1693_cast")];
            tensor<string, []> var_20446_equation_0 = const()[name = tensor<string, []>("op_20446_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20446_cast = einsum(equation = var_20446_equation_0, values = (var_20288_cast, var_20205_cast))[name = tensor<string, []>("op_20446_cast")];
            tensor<fp16, []> var_20447_to_fp16 = const()[name = tensor<string, []>("op_20447_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1695_cast = mul(x = var_20446_cast, y = var_20447_to_fp16)[name = tensor<string, []>("aw_1695_cast")];
            tensor<string, []> var_20450_equation_0 = const()[name = tensor<string, []>("op_20450_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20450_cast = einsum(equation = var_20450_equation_0, values = (var_20292_cast, var_20209_cast))[name = tensor<string, []>("op_20450_cast")];
            tensor<fp16, []> var_20451_to_fp16 = const()[name = tensor<string, []>("op_20451_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1697_cast = mul(x = var_20450_cast, y = var_20451_to_fp16)[name = tensor<string, []>("aw_1697_cast")];
            tensor<string, []> var_20454_equation_0 = const()[name = tensor<string, []>("op_20454_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20454_cast = einsum(equation = var_20454_equation_0, values = (var_20296_cast, var_20213_cast))[name = tensor<string, []>("op_20454_cast")];
            tensor<fp16, []> var_20455_to_fp16 = const()[name = tensor<string, []>("op_20455_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1699_cast = mul(x = var_20454_cast, y = var_20455_to_fp16)[name = tensor<string, []>("aw_1699_cast")];
            tensor<string, []> var_20458_equation_0 = const()[name = tensor<string, []>("op_20458_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20458_cast = einsum(equation = var_20458_equation_0, values = (var_20300_cast, var_20217_cast))[name = tensor<string, []>("op_20458_cast")];
            tensor<fp16, []> var_20459_to_fp16 = const()[name = tensor<string, []>("op_20459_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1701_cast = mul(x = var_20458_cast, y = var_20459_to_fp16)[name = tensor<string, []>("aw_1701_cast")];
            tensor<string, []> var_20462_equation_0 = const()[name = tensor<string, []>("op_20462_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20462_cast = einsum(equation = var_20462_equation_0, values = (var_20304_cast, var_20221_cast))[name = tensor<string, []>("op_20462_cast")];
            tensor<fp16, []> var_20463_to_fp16 = const()[name = tensor<string, []>("op_20463_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1703_cast = mul(x = var_20462_cast, y = var_20463_to_fp16)[name = tensor<string, []>("aw_1703_cast")];
            tensor<string, []> var_20466_equation_0 = const()[name = tensor<string, []>("op_20466_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20466_cast = einsum(equation = var_20466_equation_0, values = (var_20308_cast, var_20225_cast))[name = tensor<string, []>("op_20466_cast")];
            tensor<fp16, []> var_20467_to_fp16 = const()[name = tensor<string, []>("op_20467_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1705_cast = mul(x = var_20466_cast, y = var_20467_to_fp16)[name = tensor<string, []>("aw_1705_cast")];
            tensor<string, []> var_20470_equation_0 = const()[name = tensor<string, []>("op_20470_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20470_cast = einsum(equation = var_20470_equation_0, values = (var_20312_cast, var_20229_cast))[name = tensor<string, []>("op_20470_cast")];
            tensor<fp16, []> var_20471_to_fp16 = const()[name = tensor<string, []>("op_20471_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1707_cast = mul(x = var_20470_cast, y = var_20471_to_fp16)[name = tensor<string, []>("aw_1707_cast")];
            tensor<string, []> var_20474_equation_0 = const()[name = tensor<string, []>("op_20474_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20474_cast = einsum(equation = var_20474_equation_0, values = (var_20316_cast, var_20233_cast))[name = tensor<string, []>("op_20474_cast")];
            tensor<fp16, []> var_20475_to_fp16 = const()[name = tensor<string, []>("op_20475_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1709_cast = mul(x = var_20474_cast, y = var_20475_to_fp16)[name = tensor<string, []>("aw_1709_cast")];
            tensor<string, []> var_20478_equation_0 = const()[name = tensor<string, []>("op_20478_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20478_cast = einsum(equation = var_20478_equation_0, values = (var_20320_cast, var_20237_cast))[name = tensor<string, []>("op_20478_cast")];
            tensor<fp16, []> var_20479_to_fp16 = const()[name = tensor<string, []>("op_20479_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1711_cast = mul(x = var_20478_cast, y = var_20479_to_fp16)[name = tensor<string, []>("aw_1711_cast")];
            tensor<string, []> var_20482_equation_0 = const()[name = tensor<string, []>("op_20482_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20482_cast = einsum(equation = var_20482_equation_0, values = (var_20324_cast, var_20241_cast))[name = tensor<string, []>("op_20482_cast")];
            tensor<fp16, []> var_20483_to_fp16 = const()[name = tensor<string, []>("op_20483_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1713_cast = mul(x = var_20482_cast, y = var_20483_to_fp16)[name = tensor<string, []>("aw_1713_cast")];
            tensor<string, []> var_20486_equation_0 = const()[name = tensor<string, []>("op_20486_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20486_cast = einsum(equation = var_20486_equation_0, values = (var_20328_cast, var_20245_cast))[name = tensor<string, []>("op_20486_cast")];
            tensor<fp16, []> var_20487_to_fp16 = const()[name = tensor<string, []>("op_20487_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1715_cast = mul(x = var_20486_cast, y = var_20487_to_fp16)[name = tensor<string, []>("aw_1715_cast")];
            tensor<string, []> var_20490_equation_0 = const()[name = tensor<string, []>("op_20490_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20490_cast = einsum(equation = var_20490_equation_0, values = (var_20332_cast, var_20249_cast))[name = tensor<string, []>("op_20490_cast")];
            tensor<fp16, []> var_20491_to_fp16 = const()[name = tensor<string, []>("op_20491_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1717_cast = mul(x = var_20490_cast, y = var_20491_to_fp16)[name = tensor<string, []>("aw_1717_cast")];
            tensor<string, []> var_20494_equation_0 = const()[name = tensor<string, []>("op_20494_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20494_cast = einsum(equation = var_20494_equation_0, values = (var_20336_cast, var_20253_cast))[name = tensor<string, []>("op_20494_cast")];
            tensor<fp16, []> var_20495_to_fp16 = const()[name = tensor<string, []>("op_20495_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1719_cast = mul(x = var_20494_cast, y = var_20495_to_fp16)[name = tensor<string, []>("aw_1719_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20497_cast = softmax(axis = var_2634, x = aw_1681_cast)[name = tensor<string, []>("op_20497_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20498_cast = softmax(axis = var_2634, x = aw_1683_cast)[name = tensor<string, []>("op_20498_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20499_cast = softmax(axis = var_2634, x = aw_1685_cast)[name = tensor<string, []>("op_20499_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20500_cast = softmax(axis = var_2634, x = aw_1687_cast)[name = tensor<string, []>("op_20500_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20501_cast = softmax(axis = var_2634, x = aw_1689_cast)[name = tensor<string, []>("op_20501_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20502_cast = softmax(axis = var_2634, x = aw_1691_cast)[name = tensor<string, []>("op_20502_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20503_cast = softmax(axis = var_2634, x = aw_1693_cast)[name = tensor<string, []>("op_20503_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20504_cast = softmax(axis = var_2634, x = aw_1695_cast)[name = tensor<string, []>("op_20504_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20505_cast = softmax(axis = var_2634, x = aw_1697_cast)[name = tensor<string, []>("op_20505_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20506_cast = softmax(axis = var_2634, x = aw_1699_cast)[name = tensor<string, []>("op_20506_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20507_cast = softmax(axis = var_2634, x = aw_1701_cast)[name = tensor<string, []>("op_20507_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20508_cast = softmax(axis = var_2634, x = aw_1703_cast)[name = tensor<string, []>("op_20508_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20509_cast = softmax(axis = var_2634, x = aw_1705_cast)[name = tensor<string, []>("op_20509_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20510_cast = softmax(axis = var_2634, x = aw_1707_cast)[name = tensor<string, []>("op_20510_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20511_cast = softmax(axis = var_2634, x = aw_1709_cast)[name = tensor<string, []>("op_20511_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20512_cast = softmax(axis = var_2634, x = aw_1711_cast)[name = tensor<string, []>("op_20512_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20513_cast = softmax(axis = var_2634, x = aw_1713_cast)[name = tensor<string, []>("op_20513_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20514_cast = softmax(axis = var_2634, x = aw_1715_cast)[name = tensor<string, []>("op_20514_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20515_cast = softmax(axis = var_2634, x = aw_1717_cast)[name = tensor<string, []>("op_20515_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_20516_cast = softmax(axis = var_2634, x = aw_1719_cast)[name = tensor<string, []>("op_20516_cast")];
            tensor<string, []> var_20518_equation_0 = const()[name = tensor<string, []>("op_20518_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20518_cast = einsum(equation = var_20518_equation_0, values = (var_20338_cast, var_20497_cast))[name = tensor<string, []>("op_20518_cast")];
            tensor<string, []> var_20520_equation_0 = const()[name = tensor<string, []>("op_20520_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20520_cast = einsum(equation = var_20520_equation_0, values = (var_20342_cast, var_20498_cast))[name = tensor<string, []>("op_20520_cast")];
            tensor<string, []> var_20522_equation_0 = const()[name = tensor<string, []>("op_20522_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20522_cast = einsum(equation = var_20522_equation_0, values = (var_20346_cast, var_20499_cast))[name = tensor<string, []>("op_20522_cast")];
            tensor<string, []> var_20524_equation_0 = const()[name = tensor<string, []>("op_20524_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20524_cast = einsum(equation = var_20524_equation_0, values = (var_20350_cast, var_20500_cast))[name = tensor<string, []>("op_20524_cast")];
            tensor<string, []> var_20526_equation_0 = const()[name = tensor<string, []>("op_20526_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20526_cast = einsum(equation = var_20526_equation_0, values = (var_20354_cast, var_20501_cast))[name = tensor<string, []>("op_20526_cast")];
            tensor<string, []> var_20528_equation_0 = const()[name = tensor<string, []>("op_20528_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20528_cast = einsum(equation = var_20528_equation_0, values = (var_20358_cast, var_20502_cast))[name = tensor<string, []>("op_20528_cast")];
            tensor<string, []> var_20530_equation_0 = const()[name = tensor<string, []>("op_20530_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20530_cast = einsum(equation = var_20530_equation_0, values = (var_20362_cast, var_20503_cast))[name = tensor<string, []>("op_20530_cast")];
            tensor<string, []> var_20532_equation_0 = const()[name = tensor<string, []>("op_20532_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20532_cast = einsum(equation = var_20532_equation_0, values = (var_20366_cast, var_20504_cast))[name = tensor<string, []>("op_20532_cast")];
            tensor<string, []> var_20534_equation_0 = const()[name = tensor<string, []>("op_20534_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20534_cast = einsum(equation = var_20534_equation_0, values = (var_20370_cast, var_20505_cast))[name = tensor<string, []>("op_20534_cast")];
            tensor<string, []> var_20536_equation_0 = const()[name = tensor<string, []>("op_20536_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20536_cast = einsum(equation = var_20536_equation_0, values = (var_20374_cast, var_20506_cast))[name = tensor<string, []>("op_20536_cast")];
            tensor<string, []> var_20538_equation_0 = const()[name = tensor<string, []>("op_20538_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20538_cast = einsum(equation = var_20538_equation_0, values = (var_20378_cast, var_20507_cast))[name = tensor<string, []>("op_20538_cast")];
            tensor<string, []> var_20540_equation_0 = const()[name = tensor<string, []>("op_20540_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20540_cast = einsum(equation = var_20540_equation_0, values = (var_20382_cast, var_20508_cast))[name = tensor<string, []>("op_20540_cast")];
            tensor<string, []> var_20542_equation_0 = const()[name = tensor<string, []>("op_20542_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20542_cast = einsum(equation = var_20542_equation_0, values = (var_20386_cast, var_20509_cast))[name = tensor<string, []>("op_20542_cast")];
            tensor<string, []> var_20544_equation_0 = const()[name = tensor<string, []>("op_20544_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20544_cast = einsum(equation = var_20544_equation_0, values = (var_20390_cast, var_20510_cast))[name = tensor<string, []>("op_20544_cast")];
            tensor<string, []> var_20546_equation_0 = const()[name = tensor<string, []>("op_20546_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20546_cast = einsum(equation = var_20546_equation_0, values = (var_20394_cast, var_20511_cast))[name = tensor<string, []>("op_20546_cast")];
            tensor<string, []> var_20548_equation_0 = const()[name = tensor<string, []>("op_20548_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20548_cast = einsum(equation = var_20548_equation_0, values = (var_20398_cast, var_20512_cast))[name = tensor<string, []>("op_20548_cast")];
            tensor<string, []> var_20550_equation_0 = const()[name = tensor<string, []>("op_20550_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20550_cast = einsum(equation = var_20550_equation_0, values = (var_20402_cast, var_20513_cast))[name = tensor<string, []>("op_20550_cast")];
            tensor<string, []> var_20552_equation_0 = const()[name = tensor<string, []>("op_20552_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20552_cast = einsum(equation = var_20552_equation_0, values = (var_20406_cast, var_20514_cast))[name = tensor<string, []>("op_20552_cast")];
            tensor<string, []> var_20554_equation_0 = const()[name = tensor<string, []>("op_20554_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20554_cast = einsum(equation = var_20554_equation_0, values = (var_20410_cast, var_20515_cast))[name = tensor<string, []>("op_20554_cast")];
            tensor<string, []> var_20556_equation_0 = const()[name = tensor<string, []>("op_20556_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20556_cast = einsum(equation = var_20556_equation_0, values = (var_20414_cast, var_20516_cast))[name = tensor<string, []>("op_20556_cast")];
            tensor<bool, []> input_305_interleave_0 = const()[name = tensor<string, []>("input_305_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_305_cast = concat(axis = var_2634, interleave = input_305_interleave_0, values = (var_20518_cast, var_20520_cast, var_20522_cast, var_20524_cast, var_20526_cast, var_20528_cast, var_20530_cast, var_20532_cast, var_20534_cast, var_20536_cast, var_20538_cast, var_20540_cast, var_20542_cast, var_20544_cast, var_20546_cast, var_20548_cast, var_20550_cast, var_20552_cast, var_20554_cast, var_20556_cast))[name = tensor<string, []>("input_305_cast")];
            tensor<int32, [2]> var_20562 = const()[name = tensor<string, []>("op_20562"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_20564 = const()[name = tensor<string, []>("op_20564"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_20566_pad_type_0 = const()[name = tensor<string, []>("op_20566_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_20566_pad_0 = const()[name = tensor<string, []>("op_20566_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1597589952)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1600866816)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_20566_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_out_0_bias_to_fp16, dilations = var_20564, groups = var_2634, pad = var_20566_pad_0, pad_type = var_20566_pad_type_0, strides = var_20562, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn1_to_out_0_weight_to_fp16, x = input_305_cast)[name = tensor<string, []>("op_20566_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_141_cast = add(x = var_20566_cast, y = inputs_139_cast)[name = tensor<string, []>("inputs_141_cast")];
            tensor<int32, [1]> var_20570 = const()[name = tensor<string, []>("op_20570"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_141_cast = reduce_mean(axes = var_20570, keep_dims = var_2629, x = inputs_141_cast)[name = tensor<string, []>("channels_mean_141_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_141_cast = sub(x = inputs_141_cast, y = channels_mean_141_cast)[name = tensor<string, []>("zero_mean_141_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_141_cast = mul(x = zero_mean_141_cast, y = zero_mean_141_cast)[name = tensor<string, []>("zero_mean_sq_141_cast")];
            tensor<int32, [1]> var_20574 = const()[name = tensor<string, []>("op_20574"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_20575_cast = reduce_mean(axes = var_20574, keep_dims = var_2629, x = zero_mean_sq_141_cast)[name = tensor<string, []>("op_20575_cast")];
            tensor<fp16, []> var_20576_to_fp16 = const()[name = tensor<string, []>("op_20576_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_20577_cast = add(x = var_20575_cast, y = var_20576_to_fp16)[name = tensor<string, []>("op_20577_cast")];
            tensor<fp16, []> denom_141_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_141_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_141_cast = rsqrt(epsilon = denom_141_epsilon_0_to_fp16, x = var_20577_cast)[name = tensor<string, []>("denom_141_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_141_cast = mul(x = zero_mean_141_cast, y = denom_141_cast)[name = tensor<string, []>("out_141_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_20581_to_fp16 = const()[name = tensor<string, []>("op_20581_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1600869440)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_20582_cast = add(x = out_141_cast, y = var_20581_to_fp16)[name = tensor<string, []>("op_20582_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_20584_to_fp16 = const()[name = tensor<string, []>("op_20584_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1600872064)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_193_cast = mul(x = var_20582_cast, y = var_20584_to_fp16)[name = tensor<string, []>("hidden_states_193_cast")];
            tensor<int32, [2]> var_20591 = const()[name = tensor<string, []>("op_20591"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_20593 = const()[name = tensor<string, []>("op_20593"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_95_pad_type_0 = const()[name = tensor<string, []>("q_95_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_95_pad_0 = const()[name = tensor<string, []>("q_95_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1600874688)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_95_cast = conv(dilations = var_20593, groups = var_2634, pad = q_95_pad_0, pad_type = q_95_pad_type_0, strides = var_20591, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_q_weight_to_fp16, x = hidden_states_193_cast)[name = tensor<string, []>("q_95_cast")];
            tensor<int32, [2]> var_20597 = const()[name = tensor<string, []>("op_20597"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_20599 = const()[name = tensor<string, []>("op_20599"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_189_pad_type_0 = const()[name = tensor<string, []>("k_189_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_189_pad_0 = const()[name = tensor<string, []>("k_189_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1604151552)))];
            tensor<fp16, [2, 1280, 1, 77]> k_189_cast = conv(dilations = var_20599, groups = var_2634, pad = k_189_pad_0, pad_type = k_189_pad_type_0, strides = var_20597, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_189_cast")];
            tensor<int32, [2]> var_20603 = const()[name = tensor<string, []>("op_20603"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_20605 = const()[name = tensor<string, []>("op_20605"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_95_pad_type_0 = const()[name = tensor<string, []>("v_95_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_95_pad_0 = const()[name = tensor<string, []>("v_95_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1609394496)))];
            tensor<fp16, [2, 1280, 1, 77]> v_95_cast = conv(dilations = var_20605, groups = var_2634, pad = v_95_pad_0, pad_type = v_95_pad_type_0, strides = var_20603, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_95_cast")];
            tensor<int32, [4]> var_20609_begin_0 = const()[name = tensor<string, []>("op_20609_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_20609_end_0 = const()[name = tensor<string, []>("op_20609_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_20609_end_mask_0 = const()[name = tensor<string, []>("op_20609_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20609_cast = slice_by_index(begin = var_20609_begin_0, end = var_20609_end_0, end_mask = var_20609_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20609_cast")];
            tensor<int32, [4]> var_20613_begin_0 = const()[name = tensor<string, []>("op_20613_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_20613_end_0 = const()[name = tensor<string, []>("op_20613_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_20613_end_mask_0 = const()[name = tensor<string, []>("op_20613_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20613_cast = slice_by_index(begin = var_20613_begin_0, end = var_20613_end_0, end_mask = var_20613_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20613_cast")];
            tensor<int32, [4]> var_20617_begin_0 = const()[name = tensor<string, []>("op_20617_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_20617_end_0 = const()[name = tensor<string, []>("op_20617_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_20617_end_mask_0 = const()[name = tensor<string, []>("op_20617_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20617_cast = slice_by_index(begin = var_20617_begin_0, end = var_20617_end_0, end_mask = var_20617_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20617_cast")];
            tensor<int32, [4]> var_20621_begin_0 = const()[name = tensor<string, []>("op_20621_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_20621_end_0 = const()[name = tensor<string, []>("op_20621_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_20621_end_mask_0 = const()[name = tensor<string, []>("op_20621_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20621_cast = slice_by_index(begin = var_20621_begin_0, end = var_20621_end_0, end_mask = var_20621_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20621_cast")];
            tensor<int32, [4]> var_20625_begin_0 = const()[name = tensor<string, []>("op_20625_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_20625_end_0 = const()[name = tensor<string, []>("op_20625_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_20625_end_mask_0 = const()[name = tensor<string, []>("op_20625_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20625_cast = slice_by_index(begin = var_20625_begin_0, end = var_20625_end_0, end_mask = var_20625_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20625_cast")];
            tensor<int32, [4]> var_20629_begin_0 = const()[name = tensor<string, []>("op_20629_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_20629_end_0 = const()[name = tensor<string, []>("op_20629_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_20629_end_mask_0 = const()[name = tensor<string, []>("op_20629_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20629_cast = slice_by_index(begin = var_20629_begin_0, end = var_20629_end_0, end_mask = var_20629_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20629_cast")];
            tensor<int32, [4]> var_20633_begin_0 = const()[name = tensor<string, []>("op_20633_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_20633_end_0 = const()[name = tensor<string, []>("op_20633_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_20633_end_mask_0 = const()[name = tensor<string, []>("op_20633_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20633_cast = slice_by_index(begin = var_20633_begin_0, end = var_20633_end_0, end_mask = var_20633_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20633_cast")];
            tensor<int32, [4]> var_20637_begin_0 = const()[name = tensor<string, []>("op_20637_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_20637_end_0 = const()[name = tensor<string, []>("op_20637_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_20637_end_mask_0 = const()[name = tensor<string, []>("op_20637_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20637_cast = slice_by_index(begin = var_20637_begin_0, end = var_20637_end_0, end_mask = var_20637_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20637_cast")];
            tensor<int32, [4]> var_20641_begin_0 = const()[name = tensor<string, []>("op_20641_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_20641_end_0 = const()[name = tensor<string, []>("op_20641_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_20641_end_mask_0 = const()[name = tensor<string, []>("op_20641_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20641_cast = slice_by_index(begin = var_20641_begin_0, end = var_20641_end_0, end_mask = var_20641_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20641_cast")];
            tensor<int32, [4]> var_20645_begin_0 = const()[name = tensor<string, []>("op_20645_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_20645_end_0 = const()[name = tensor<string, []>("op_20645_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_20645_end_mask_0 = const()[name = tensor<string, []>("op_20645_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20645_cast = slice_by_index(begin = var_20645_begin_0, end = var_20645_end_0, end_mask = var_20645_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20645_cast")];
            tensor<int32, [4]> var_20649_begin_0 = const()[name = tensor<string, []>("op_20649_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_20649_end_0 = const()[name = tensor<string, []>("op_20649_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_20649_end_mask_0 = const()[name = tensor<string, []>("op_20649_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20649_cast = slice_by_index(begin = var_20649_begin_0, end = var_20649_end_0, end_mask = var_20649_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20649_cast")];
            tensor<int32, [4]> var_20653_begin_0 = const()[name = tensor<string, []>("op_20653_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_20653_end_0 = const()[name = tensor<string, []>("op_20653_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_20653_end_mask_0 = const()[name = tensor<string, []>("op_20653_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20653_cast = slice_by_index(begin = var_20653_begin_0, end = var_20653_end_0, end_mask = var_20653_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20653_cast")];
            tensor<int32, [4]> var_20657_begin_0 = const()[name = tensor<string, []>("op_20657_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_20657_end_0 = const()[name = tensor<string, []>("op_20657_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_20657_end_mask_0 = const()[name = tensor<string, []>("op_20657_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20657_cast = slice_by_index(begin = var_20657_begin_0, end = var_20657_end_0, end_mask = var_20657_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20657_cast")];
            tensor<int32, [4]> var_20661_begin_0 = const()[name = tensor<string, []>("op_20661_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_20661_end_0 = const()[name = tensor<string, []>("op_20661_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_20661_end_mask_0 = const()[name = tensor<string, []>("op_20661_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20661_cast = slice_by_index(begin = var_20661_begin_0, end = var_20661_end_0, end_mask = var_20661_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20661_cast")];
            tensor<int32, [4]> var_20665_begin_0 = const()[name = tensor<string, []>("op_20665_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_20665_end_0 = const()[name = tensor<string, []>("op_20665_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_20665_end_mask_0 = const()[name = tensor<string, []>("op_20665_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20665_cast = slice_by_index(begin = var_20665_begin_0, end = var_20665_end_0, end_mask = var_20665_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20665_cast")];
            tensor<int32, [4]> var_20669_begin_0 = const()[name = tensor<string, []>("op_20669_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_20669_end_0 = const()[name = tensor<string, []>("op_20669_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_20669_end_mask_0 = const()[name = tensor<string, []>("op_20669_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20669_cast = slice_by_index(begin = var_20669_begin_0, end = var_20669_end_0, end_mask = var_20669_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20669_cast")];
            tensor<int32, [4]> var_20673_begin_0 = const()[name = tensor<string, []>("op_20673_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_20673_end_0 = const()[name = tensor<string, []>("op_20673_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_20673_end_mask_0 = const()[name = tensor<string, []>("op_20673_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20673_cast = slice_by_index(begin = var_20673_begin_0, end = var_20673_end_0, end_mask = var_20673_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20673_cast")];
            tensor<int32, [4]> var_20677_begin_0 = const()[name = tensor<string, []>("op_20677_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_20677_end_0 = const()[name = tensor<string, []>("op_20677_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_20677_end_mask_0 = const()[name = tensor<string, []>("op_20677_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20677_cast = slice_by_index(begin = var_20677_begin_0, end = var_20677_end_0, end_mask = var_20677_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20677_cast")];
            tensor<int32, [4]> var_20681_begin_0 = const()[name = tensor<string, []>("op_20681_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_20681_end_0 = const()[name = tensor<string, []>("op_20681_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_20681_end_mask_0 = const()[name = tensor<string, []>("op_20681_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20681_cast = slice_by_index(begin = var_20681_begin_0, end = var_20681_end_0, end_mask = var_20681_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20681_cast")];
            tensor<int32, [4]> var_20685_begin_0 = const()[name = tensor<string, []>("op_20685_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_20685_end_0 = const()[name = tensor<string, []>("op_20685_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_20685_end_mask_0 = const()[name = tensor<string, []>("op_20685_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_20685_cast = slice_by_index(begin = var_20685_begin_0, end = var_20685_end_0, end_mask = var_20685_end_mask_0, x = q_95_cast)[name = tensor<string, []>("op_20685_cast")];
            tensor<int32, [4]> k_191_perm_0 = const()[name = tensor<string, []>("k_191_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_20692_begin_0 = const()[name = tensor<string, []>("op_20692_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_20692_end_0 = const()[name = tensor<string, []>("op_20692_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_20692_end_mask_0 = const()[name = tensor<string, []>("op_20692_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_92 = transpose(perm = k_191_perm_0, x = k_189_cast)[name = tensor<string, []>("transpose_92")];
            tensor<fp16, [2, 77, 1, 64]> var_20692_cast = slice_by_index(begin = var_20692_begin_0, end = var_20692_end_0, end_mask = var_20692_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20692_cast")];
            tensor<int32, [4]> var_20696_begin_0 = const()[name = tensor<string, []>("op_20696_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_20696_end_0 = const()[name = tensor<string, []>("op_20696_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_20696_end_mask_0 = const()[name = tensor<string, []>("op_20696_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_20696_cast = slice_by_index(begin = var_20696_begin_0, end = var_20696_end_0, end_mask = var_20696_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20696_cast")];
            tensor<int32, [4]> var_20700_begin_0 = const()[name = tensor<string, []>("op_20700_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_20700_end_0 = const()[name = tensor<string, []>("op_20700_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_20700_end_mask_0 = const()[name = tensor<string, []>("op_20700_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_20700_cast = slice_by_index(begin = var_20700_begin_0, end = var_20700_end_0, end_mask = var_20700_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20700_cast")];
            tensor<int32, [4]> var_20704_begin_0 = const()[name = tensor<string, []>("op_20704_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_20704_end_0 = const()[name = tensor<string, []>("op_20704_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_20704_end_mask_0 = const()[name = tensor<string, []>("op_20704_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_20704_cast = slice_by_index(begin = var_20704_begin_0, end = var_20704_end_0, end_mask = var_20704_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20704_cast")];
            tensor<int32, [4]> var_20708_begin_0 = const()[name = tensor<string, []>("op_20708_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_20708_end_0 = const()[name = tensor<string, []>("op_20708_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_20708_end_mask_0 = const()[name = tensor<string, []>("op_20708_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_20708_cast = slice_by_index(begin = var_20708_begin_0, end = var_20708_end_0, end_mask = var_20708_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20708_cast")];
            tensor<int32, [4]> var_20712_begin_0 = const()[name = tensor<string, []>("op_20712_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_20712_end_0 = const()[name = tensor<string, []>("op_20712_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_20712_end_mask_0 = const()[name = tensor<string, []>("op_20712_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_20712_cast = slice_by_index(begin = var_20712_begin_0, end = var_20712_end_0, end_mask = var_20712_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20712_cast")];
            tensor<int32, [4]> var_20716_begin_0 = const()[name = tensor<string, []>("op_20716_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_20716_end_0 = const()[name = tensor<string, []>("op_20716_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_20716_end_mask_0 = const()[name = tensor<string, []>("op_20716_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_20716_cast = slice_by_index(begin = var_20716_begin_0, end = var_20716_end_0, end_mask = var_20716_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20716_cast")];
            tensor<int32, [4]> var_20720_begin_0 = const()[name = tensor<string, []>("op_20720_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_20720_end_0 = const()[name = tensor<string, []>("op_20720_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_20720_end_mask_0 = const()[name = tensor<string, []>("op_20720_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_20720_cast = slice_by_index(begin = var_20720_begin_0, end = var_20720_end_0, end_mask = var_20720_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20720_cast")];
            tensor<int32, [4]> var_20724_begin_0 = const()[name = tensor<string, []>("op_20724_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_20724_end_0 = const()[name = tensor<string, []>("op_20724_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_20724_end_mask_0 = const()[name = tensor<string, []>("op_20724_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_20724_cast = slice_by_index(begin = var_20724_begin_0, end = var_20724_end_0, end_mask = var_20724_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20724_cast")];
            tensor<int32, [4]> var_20728_begin_0 = const()[name = tensor<string, []>("op_20728_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_20728_end_0 = const()[name = tensor<string, []>("op_20728_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_20728_end_mask_0 = const()[name = tensor<string, []>("op_20728_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_20728_cast = slice_by_index(begin = var_20728_begin_0, end = var_20728_end_0, end_mask = var_20728_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20728_cast")];
            tensor<int32, [4]> var_20732_begin_0 = const()[name = tensor<string, []>("op_20732_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_20732_end_0 = const()[name = tensor<string, []>("op_20732_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_20732_end_mask_0 = const()[name = tensor<string, []>("op_20732_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_20732_cast = slice_by_index(begin = var_20732_begin_0, end = var_20732_end_0, end_mask = var_20732_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20732_cast")];
            tensor<int32, [4]> var_20736_begin_0 = const()[name = tensor<string, []>("op_20736_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_20736_end_0 = const()[name = tensor<string, []>("op_20736_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_20736_end_mask_0 = const()[name = tensor<string, []>("op_20736_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_20736_cast = slice_by_index(begin = var_20736_begin_0, end = var_20736_end_0, end_mask = var_20736_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20736_cast")];
            tensor<int32, [4]> var_20740_begin_0 = const()[name = tensor<string, []>("op_20740_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_20740_end_0 = const()[name = tensor<string, []>("op_20740_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_20740_end_mask_0 = const()[name = tensor<string, []>("op_20740_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_20740_cast = slice_by_index(begin = var_20740_begin_0, end = var_20740_end_0, end_mask = var_20740_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20740_cast")];
            tensor<int32, [4]> var_20744_begin_0 = const()[name = tensor<string, []>("op_20744_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_20744_end_0 = const()[name = tensor<string, []>("op_20744_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_20744_end_mask_0 = const()[name = tensor<string, []>("op_20744_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_20744_cast = slice_by_index(begin = var_20744_begin_0, end = var_20744_end_0, end_mask = var_20744_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20744_cast")];
            tensor<int32, [4]> var_20748_begin_0 = const()[name = tensor<string, []>("op_20748_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_20748_end_0 = const()[name = tensor<string, []>("op_20748_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_20748_end_mask_0 = const()[name = tensor<string, []>("op_20748_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_20748_cast = slice_by_index(begin = var_20748_begin_0, end = var_20748_end_0, end_mask = var_20748_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20748_cast")];
            tensor<int32, [4]> var_20752_begin_0 = const()[name = tensor<string, []>("op_20752_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_20752_end_0 = const()[name = tensor<string, []>("op_20752_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_20752_end_mask_0 = const()[name = tensor<string, []>("op_20752_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_20752_cast = slice_by_index(begin = var_20752_begin_0, end = var_20752_end_0, end_mask = var_20752_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20752_cast")];
            tensor<int32, [4]> var_20756_begin_0 = const()[name = tensor<string, []>("op_20756_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_20756_end_0 = const()[name = tensor<string, []>("op_20756_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_20756_end_mask_0 = const()[name = tensor<string, []>("op_20756_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_20756_cast = slice_by_index(begin = var_20756_begin_0, end = var_20756_end_0, end_mask = var_20756_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20756_cast")];
            tensor<int32, [4]> var_20760_begin_0 = const()[name = tensor<string, []>("op_20760_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_20760_end_0 = const()[name = tensor<string, []>("op_20760_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_20760_end_mask_0 = const()[name = tensor<string, []>("op_20760_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_20760_cast = slice_by_index(begin = var_20760_begin_0, end = var_20760_end_0, end_mask = var_20760_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20760_cast")];
            tensor<int32, [4]> var_20764_begin_0 = const()[name = tensor<string, []>("op_20764_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_20764_end_0 = const()[name = tensor<string, []>("op_20764_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_20764_end_mask_0 = const()[name = tensor<string, []>("op_20764_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_20764_cast = slice_by_index(begin = var_20764_begin_0, end = var_20764_end_0, end_mask = var_20764_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20764_cast")];
            tensor<int32, [4]> var_20768_begin_0 = const()[name = tensor<string, []>("op_20768_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_20768_end_0 = const()[name = tensor<string, []>("op_20768_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_20768_end_mask_0 = const()[name = tensor<string, []>("op_20768_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_20768_cast = slice_by_index(begin = var_20768_begin_0, end = var_20768_end_0, end_mask = var_20768_end_mask_0, x = transpose_92)[name = tensor<string, []>("op_20768_cast")];
            tensor<int32, [4]> var_20770_begin_0 = const()[name = tensor<string, []>("op_20770_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_20770_end_0 = const()[name = tensor<string, []>("op_20770_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_20770_end_mask_0 = const()[name = tensor<string, []>("op_20770_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20770_cast = slice_by_index(begin = var_20770_begin_0, end = var_20770_end_0, end_mask = var_20770_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20770_cast")];
            tensor<int32, [4]> var_20774_begin_0 = const()[name = tensor<string, []>("op_20774_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_20774_end_0 = const()[name = tensor<string, []>("op_20774_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_20774_end_mask_0 = const()[name = tensor<string, []>("op_20774_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20774_cast = slice_by_index(begin = var_20774_begin_0, end = var_20774_end_0, end_mask = var_20774_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20774_cast")];
            tensor<int32, [4]> var_20778_begin_0 = const()[name = tensor<string, []>("op_20778_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_20778_end_0 = const()[name = tensor<string, []>("op_20778_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_20778_end_mask_0 = const()[name = tensor<string, []>("op_20778_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20778_cast = slice_by_index(begin = var_20778_begin_0, end = var_20778_end_0, end_mask = var_20778_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20778_cast")];
            tensor<int32, [4]> var_20782_begin_0 = const()[name = tensor<string, []>("op_20782_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_20782_end_0 = const()[name = tensor<string, []>("op_20782_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_20782_end_mask_0 = const()[name = tensor<string, []>("op_20782_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20782_cast = slice_by_index(begin = var_20782_begin_0, end = var_20782_end_0, end_mask = var_20782_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20782_cast")];
            tensor<int32, [4]> var_20786_begin_0 = const()[name = tensor<string, []>("op_20786_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_20786_end_0 = const()[name = tensor<string, []>("op_20786_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_20786_end_mask_0 = const()[name = tensor<string, []>("op_20786_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20786_cast = slice_by_index(begin = var_20786_begin_0, end = var_20786_end_0, end_mask = var_20786_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20786_cast")];
            tensor<int32, [4]> var_20790_begin_0 = const()[name = tensor<string, []>("op_20790_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_20790_end_0 = const()[name = tensor<string, []>("op_20790_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_20790_end_mask_0 = const()[name = tensor<string, []>("op_20790_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20790_cast = slice_by_index(begin = var_20790_begin_0, end = var_20790_end_0, end_mask = var_20790_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20790_cast")];
            tensor<int32, [4]> var_20794_begin_0 = const()[name = tensor<string, []>("op_20794_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_20794_end_0 = const()[name = tensor<string, []>("op_20794_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_20794_end_mask_0 = const()[name = tensor<string, []>("op_20794_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20794_cast = slice_by_index(begin = var_20794_begin_0, end = var_20794_end_0, end_mask = var_20794_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20794_cast")];
            tensor<int32, [4]> var_20798_begin_0 = const()[name = tensor<string, []>("op_20798_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_20798_end_0 = const()[name = tensor<string, []>("op_20798_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_20798_end_mask_0 = const()[name = tensor<string, []>("op_20798_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20798_cast = slice_by_index(begin = var_20798_begin_0, end = var_20798_end_0, end_mask = var_20798_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20798_cast")];
            tensor<int32, [4]> var_20802_begin_0 = const()[name = tensor<string, []>("op_20802_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_20802_end_0 = const()[name = tensor<string, []>("op_20802_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_20802_end_mask_0 = const()[name = tensor<string, []>("op_20802_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20802_cast = slice_by_index(begin = var_20802_begin_0, end = var_20802_end_0, end_mask = var_20802_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20802_cast")];
            tensor<int32, [4]> var_20806_begin_0 = const()[name = tensor<string, []>("op_20806_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_20806_end_0 = const()[name = tensor<string, []>("op_20806_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_20806_end_mask_0 = const()[name = tensor<string, []>("op_20806_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20806_cast = slice_by_index(begin = var_20806_begin_0, end = var_20806_end_0, end_mask = var_20806_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20806_cast")];
            tensor<int32, [4]> var_20810_begin_0 = const()[name = tensor<string, []>("op_20810_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_20810_end_0 = const()[name = tensor<string, []>("op_20810_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_20810_end_mask_0 = const()[name = tensor<string, []>("op_20810_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20810_cast = slice_by_index(begin = var_20810_begin_0, end = var_20810_end_0, end_mask = var_20810_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20810_cast")];
            tensor<int32, [4]> var_20814_begin_0 = const()[name = tensor<string, []>("op_20814_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_20814_end_0 = const()[name = tensor<string, []>("op_20814_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_20814_end_mask_0 = const()[name = tensor<string, []>("op_20814_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20814_cast = slice_by_index(begin = var_20814_begin_0, end = var_20814_end_0, end_mask = var_20814_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20814_cast")];
            tensor<int32, [4]> var_20818_begin_0 = const()[name = tensor<string, []>("op_20818_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_20818_end_0 = const()[name = tensor<string, []>("op_20818_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_20818_end_mask_0 = const()[name = tensor<string, []>("op_20818_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20818_cast = slice_by_index(begin = var_20818_begin_0, end = var_20818_end_0, end_mask = var_20818_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20818_cast")];
            tensor<int32, [4]> var_20822_begin_0 = const()[name = tensor<string, []>("op_20822_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_20822_end_0 = const()[name = tensor<string, []>("op_20822_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_20822_end_mask_0 = const()[name = tensor<string, []>("op_20822_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20822_cast = slice_by_index(begin = var_20822_begin_0, end = var_20822_end_0, end_mask = var_20822_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20822_cast")];
            tensor<int32, [4]> var_20826_begin_0 = const()[name = tensor<string, []>("op_20826_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_20826_end_0 = const()[name = tensor<string, []>("op_20826_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_20826_end_mask_0 = const()[name = tensor<string, []>("op_20826_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20826_cast = slice_by_index(begin = var_20826_begin_0, end = var_20826_end_0, end_mask = var_20826_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20826_cast")];
            tensor<int32, [4]> var_20830_begin_0 = const()[name = tensor<string, []>("op_20830_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_20830_end_0 = const()[name = tensor<string, []>("op_20830_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_20830_end_mask_0 = const()[name = tensor<string, []>("op_20830_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20830_cast = slice_by_index(begin = var_20830_begin_0, end = var_20830_end_0, end_mask = var_20830_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20830_cast")];
            tensor<int32, [4]> var_20834_begin_0 = const()[name = tensor<string, []>("op_20834_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_20834_end_0 = const()[name = tensor<string, []>("op_20834_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_20834_end_mask_0 = const()[name = tensor<string, []>("op_20834_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20834_cast = slice_by_index(begin = var_20834_begin_0, end = var_20834_end_0, end_mask = var_20834_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20834_cast")];
            tensor<int32, [4]> var_20838_begin_0 = const()[name = tensor<string, []>("op_20838_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_20838_end_0 = const()[name = tensor<string, []>("op_20838_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_20838_end_mask_0 = const()[name = tensor<string, []>("op_20838_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20838_cast = slice_by_index(begin = var_20838_begin_0, end = var_20838_end_0, end_mask = var_20838_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20838_cast")];
            tensor<int32, [4]> var_20842_begin_0 = const()[name = tensor<string, []>("op_20842_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_20842_end_0 = const()[name = tensor<string, []>("op_20842_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_20842_end_mask_0 = const()[name = tensor<string, []>("op_20842_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20842_cast = slice_by_index(begin = var_20842_begin_0, end = var_20842_end_0, end_mask = var_20842_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20842_cast")];
            tensor<int32, [4]> var_20846_begin_0 = const()[name = tensor<string, []>("op_20846_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_20846_end_0 = const()[name = tensor<string, []>("op_20846_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_20846_end_mask_0 = const()[name = tensor<string, []>("op_20846_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_20846_cast = slice_by_index(begin = var_20846_begin_0, end = var_20846_end_0, end_mask = var_20846_end_mask_0, x = v_95_cast)[name = tensor<string, []>("op_20846_cast")];
            tensor<string, []> var_20850_equation_0 = const()[name = tensor<string, []>("op_20850_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20850_cast = einsum(equation = var_20850_equation_0, values = (var_20692_cast, var_20609_cast))[name = tensor<string, []>("op_20850_cast")];
            tensor<fp16, []> var_20851_to_fp16 = const()[name = tensor<string, []>("op_20851_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1721_cast = mul(x = var_20850_cast, y = var_20851_to_fp16)[name = tensor<string, []>("aw_1721_cast")];
            tensor<string, []> var_20854_equation_0 = const()[name = tensor<string, []>("op_20854_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20854_cast = einsum(equation = var_20854_equation_0, values = (var_20696_cast, var_20613_cast))[name = tensor<string, []>("op_20854_cast")];
            tensor<fp16, []> var_20855_to_fp16 = const()[name = tensor<string, []>("op_20855_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1723_cast = mul(x = var_20854_cast, y = var_20855_to_fp16)[name = tensor<string, []>("aw_1723_cast")];
            tensor<string, []> var_20858_equation_0 = const()[name = tensor<string, []>("op_20858_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20858_cast = einsum(equation = var_20858_equation_0, values = (var_20700_cast, var_20617_cast))[name = tensor<string, []>("op_20858_cast")];
            tensor<fp16, []> var_20859_to_fp16 = const()[name = tensor<string, []>("op_20859_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1725_cast = mul(x = var_20858_cast, y = var_20859_to_fp16)[name = tensor<string, []>("aw_1725_cast")];
            tensor<string, []> var_20862_equation_0 = const()[name = tensor<string, []>("op_20862_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20862_cast = einsum(equation = var_20862_equation_0, values = (var_20704_cast, var_20621_cast))[name = tensor<string, []>("op_20862_cast")];
            tensor<fp16, []> var_20863_to_fp16 = const()[name = tensor<string, []>("op_20863_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1727_cast = mul(x = var_20862_cast, y = var_20863_to_fp16)[name = tensor<string, []>("aw_1727_cast")];
            tensor<string, []> var_20866_equation_0 = const()[name = tensor<string, []>("op_20866_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20866_cast = einsum(equation = var_20866_equation_0, values = (var_20708_cast, var_20625_cast))[name = tensor<string, []>("op_20866_cast")];
            tensor<fp16, []> var_20867_to_fp16 = const()[name = tensor<string, []>("op_20867_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1729_cast = mul(x = var_20866_cast, y = var_20867_to_fp16)[name = tensor<string, []>("aw_1729_cast")];
            tensor<string, []> var_20870_equation_0 = const()[name = tensor<string, []>("op_20870_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20870_cast = einsum(equation = var_20870_equation_0, values = (var_20712_cast, var_20629_cast))[name = tensor<string, []>("op_20870_cast")];
            tensor<fp16, []> var_20871_to_fp16 = const()[name = tensor<string, []>("op_20871_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1731_cast = mul(x = var_20870_cast, y = var_20871_to_fp16)[name = tensor<string, []>("aw_1731_cast")];
            tensor<string, []> var_20874_equation_0 = const()[name = tensor<string, []>("op_20874_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20874_cast = einsum(equation = var_20874_equation_0, values = (var_20716_cast, var_20633_cast))[name = tensor<string, []>("op_20874_cast")];
            tensor<fp16, []> var_20875_to_fp16 = const()[name = tensor<string, []>("op_20875_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1733_cast = mul(x = var_20874_cast, y = var_20875_to_fp16)[name = tensor<string, []>("aw_1733_cast")];
            tensor<string, []> var_20878_equation_0 = const()[name = tensor<string, []>("op_20878_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20878_cast = einsum(equation = var_20878_equation_0, values = (var_20720_cast, var_20637_cast))[name = tensor<string, []>("op_20878_cast")];
            tensor<fp16, []> var_20879_to_fp16 = const()[name = tensor<string, []>("op_20879_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1735_cast = mul(x = var_20878_cast, y = var_20879_to_fp16)[name = tensor<string, []>("aw_1735_cast")];
            tensor<string, []> var_20882_equation_0 = const()[name = tensor<string, []>("op_20882_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20882_cast = einsum(equation = var_20882_equation_0, values = (var_20724_cast, var_20641_cast))[name = tensor<string, []>("op_20882_cast")];
            tensor<fp16, []> var_20883_to_fp16 = const()[name = tensor<string, []>("op_20883_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1737_cast = mul(x = var_20882_cast, y = var_20883_to_fp16)[name = tensor<string, []>("aw_1737_cast")];
            tensor<string, []> var_20886_equation_0 = const()[name = tensor<string, []>("op_20886_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20886_cast = einsum(equation = var_20886_equation_0, values = (var_20728_cast, var_20645_cast))[name = tensor<string, []>("op_20886_cast")];
            tensor<fp16, []> var_20887_to_fp16 = const()[name = tensor<string, []>("op_20887_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1739_cast = mul(x = var_20886_cast, y = var_20887_to_fp16)[name = tensor<string, []>("aw_1739_cast")];
            tensor<string, []> var_20890_equation_0 = const()[name = tensor<string, []>("op_20890_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20890_cast = einsum(equation = var_20890_equation_0, values = (var_20732_cast, var_20649_cast))[name = tensor<string, []>("op_20890_cast")];
            tensor<fp16, []> var_20891_to_fp16 = const()[name = tensor<string, []>("op_20891_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1741_cast = mul(x = var_20890_cast, y = var_20891_to_fp16)[name = tensor<string, []>("aw_1741_cast")];
            tensor<string, []> var_20894_equation_0 = const()[name = tensor<string, []>("op_20894_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20894_cast = einsum(equation = var_20894_equation_0, values = (var_20736_cast, var_20653_cast))[name = tensor<string, []>("op_20894_cast")];
            tensor<fp16, []> var_20895_to_fp16 = const()[name = tensor<string, []>("op_20895_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1743_cast = mul(x = var_20894_cast, y = var_20895_to_fp16)[name = tensor<string, []>("aw_1743_cast")];
            tensor<string, []> var_20898_equation_0 = const()[name = tensor<string, []>("op_20898_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20898_cast = einsum(equation = var_20898_equation_0, values = (var_20740_cast, var_20657_cast))[name = tensor<string, []>("op_20898_cast")];
            tensor<fp16, []> var_20899_to_fp16 = const()[name = tensor<string, []>("op_20899_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1745_cast = mul(x = var_20898_cast, y = var_20899_to_fp16)[name = tensor<string, []>("aw_1745_cast")];
            tensor<string, []> var_20902_equation_0 = const()[name = tensor<string, []>("op_20902_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20902_cast = einsum(equation = var_20902_equation_0, values = (var_20744_cast, var_20661_cast))[name = tensor<string, []>("op_20902_cast")];
            tensor<fp16, []> var_20903_to_fp16 = const()[name = tensor<string, []>("op_20903_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1747_cast = mul(x = var_20902_cast, y = var_20903_to_fp16)[name = tensor<string, []>("aw_1747_cast")];
            tensor<string, []> var_20906_equation_0 = const()[name = tensor<string, []>("op_20906_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20906_cast = einsum(equation = var_20906_equation_0, values = (var_20748_cast, var_20665_cast))[name = tensor<string, []>("op_20906_cast")];
            tensor<fp16, []> var_20907_to_fp16 = const()[name = tensor<string, []>("op_20907_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1749_cast = mul(x = var_20906_cast, y = var_20907_to_fp16)[name = tensor<string, []>("aw_1749_cast")];
            tensor<string, []> var_20910_equation_0 = const()[name = tensor<string, []>("op_20910_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20910_cast = einsum(equation = var_20910_equation_0, values = (var_20752_cast, var_20669_cast))[name = tensor<string, []>("op_20910_cast")];
            tensor<fp16, []> var_20911_to_fp16 = const()[name = tensor<string, []>("op_20911_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1751_cast = mul(x = var_20910_cast, y = var_20911_to_fp16)[name = tensor<string, []>("aw_1751_cast")];
            tensor<string, []> var_20914_equation_0 = const()[name = tensor<string, []>("op_20914_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20914_cast = einsum(equation = var_20914_equation_0, values = (var_20756_cast, var_20673_cast))[name = tensor<string, []>("op_20914_cast")];
            tensor<fp16, []> var_20915_to_fp16 = const()[name = tensor<string, []>("op_20915_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1753_cast = mul(x = var_20914_cast, y = var_20915_to_fp16)[name = tensor<string, []>("aw_1753_cast")];
            tensor<string, []> var_20918_equation_0 = const()[name = tensor<string, []>("op_20918_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20918_cast = einsum(equation = var_20918_equation_0, values = (var_20760_cast, var_20677_cast))[name = tensor<string, []>("op_20918_cast")];
            tensor<fp16, []> var_20919_to_fp16 = const()[name = tensor<string, []>("op_20919_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1755_cast = mul(x = var_20918_cast, y = var_20919_to_fp16)[name = tensor<string, []>("aw_1755_cast")];
            tensor<string, []> var_20922_equation_0 = const()[name = tensor<string, []>("op_20922_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20922_cast = einsum(equation = var_20922_equation_0, values = (var_20764_cast, var_20681_cast))[name = tensor<string, []>("op_20922_cast")];
            tensor<fp16, []> var_20923_to_fp16 = const()[name = tensor<string, []>("op_20923_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1757_cast = mul(x = var_20922_cast, y = var_20923_to_fp16)[name = tensor<string, []>("aw_1757_cast")];
            tensor<string, []> var_20926_equation_0 = const()[name = tensor<string, []>("op_20926_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_20926_cast = einsum(equation = var_20926_equation_0, values = (var_20768_cast, var_20685_cast))[name = tensor<string, []>("op_20926_cast")];
            tensor<fp16, []> var_20927_to_fp16 = const()[name = tensor<string, []>("op_20927_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1759_cast = mul(x = var_20926_cast, y = var_20927_to_fp16)[name = tensor<string, []>("aw_1759_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20929_cast = softmax(axis = var_2634, x = aw_1721_cast)[name = tensor<string, []>("op_20929_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20930_cast = softmax(axis = var_2634, x = aw_1723_cast)[name = tensor<string, []>("op_20930_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20931_cast = softmax(axis = var_2634, x = aw_1725_cast)[name = tensor<string, []>("op_20931_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20932_cast = softmax(axis = var_2634, x = aw_1727_cast)[name = tensor<string, []>("op_20932_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20933_cast = softmax(axis = var_2634, x = aw_1729_cast)[name = tensor<string, []>("op_20933_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20934_cast = softmax(axis = var_2634, x = aw_1731_cast)[name = tensor<string, []>("op_20934_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20935_cast = softmax(axis = var_2634, x = aw_1733_cast)[name = tensor<string, []>("op_20935_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20936_cast = softmax(axis = var_2634, x = aw_1735_cast)[name = tensor<string, []>("op_20936_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20937_cast = softmax(axis = var_2634, x = aw_1737_cast)[name = tensor<string, []>("op_20937_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20938_cast = softmax(axis = var_2634, x = aw_1739_cast)[name = tensor<string, []>("op_20938_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20939_cast = softmax(axis = var_2634, x = aw_1741_cast)[name = tensor<string, []>("op_20939_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20940_cast = softmax(axis = var_2634, x = aw_1743_cast)[name = tensor<string, []>("op_20940_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20941_cast = softmax(axis = var_2634, x = aw_1745_cast)[name = tensor<string, []>("op_20941_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20942_cast = softmax(axis = var_2634, x = aw_1747_cast)[name = tensor<string, []>("op_20942_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20943_cast = softmax(axis = var_2634, x = aw_1749_cast)[name = tensor<string, []>("op_20943_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20944_cast = softmax(axis = var_2634, x = aw_1751_cast)[name = tensor<string, []>("op_20944_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20945_cast = softmax(axis = var_2634, x = aw_1753_cast)[name = tensor<string, []>("op_20945_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20946_cast = softmax(axis = var_2634, x = aw_1755_cast)[name = tensor<string, []>("op_20946_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20947_cast = softmax(axis = var_2634, x = aw_1757_cast)[name = tensor<string, []>("op_20947_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_20948_cast = softmax(axis = var_2634, x = aw_1759_cast)[name = tensor<string, []>("op_20948_cast")];
            tensor<string, []> var_20950_equation_0 = const()[name = tensor<string, []>("op_20950_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20950_cast = einsum(equation = var_20950_equation_0, values = (var_20770_cast, var_20929_cast))[name = tensor<string, []>("op_20950_cast")];
            tensor<string, []> var_20952_equation_0 = const()[name = tensor<string, []>("op_20952_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20952_cast = einsum(equation = var_20952_equation_0, values = (var_20774_cast, var_20930_cast))[name = tensor<string, []>("op_20952_cast")];
            tensor<string, []> var_20954_equation_0 = const()[name = tensor<string, []>("op_20954_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20954_cast = einsum(equation = var_20954_equation_0, values = (var_20778_cast, var_20931_cast))[name = tensor<string, []>("op_20954_cast")];
            tensor<string, []> var_20956_equation_0 = const()[name = tensor<string, []>("op_20956_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20956_cast = einsum(equation = var_20956_equation_0, values = (var_20782_cast, var_20932_cast))[name = tensor<string, []>("op_20956_cast")];
            tensor<string, []> var_20958_equation_0 = const()[name = tensor<string, []>("op_20958_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20958_cast = einsum(equation = var_20958_equation_0, values = (var_20786_cast, var_20933_cast))[name = tensor<string, []>("op_20958_cast")];
            tensor<string, []> var_20960_equation_0 = const()[name = tensor<string, []>("op_20960_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20960_cast = einsum(equation = var_20960_equation_0, values = (var_20790_cast, var_20934_cast))[name = tensor<string, []>("op_20960_cast")];
            tensor<string, []> var_20962_equation_0 = const()[name = tensor<string, []>("op_20962_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20962_cast = einsum(equation = var_20962_equation_0, values = (var_20794_cast, var_20935_cast))[name = tensor<string, []>("op_20962_cast")];
            tensor<string, []> var_20964_equation_0 = const()[name = tensor<string, []>("op_20964_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20964_cast = einsum(equation = var_20964_equation_0, values = (var_20798_cast, var_20936_cast))[name = tensor<string, []>("op_20964_cast")];
            tensor<string, []> var_20966_equation_0 = const()[name = tensor<string, []>("op_20966_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20966_cast = einsum(equation = var_20966_equation_0, values = (var_20802_cast, var_20937_cast))[name = tensor<string, []>("op_20966_cast")];
            tensor<string, []> var_20968_equation_0 = const()[name = tensor<string, []>("op_20968_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20968_cast = einsum(equation = var_20968_equation_0, values = (var_20806_cast, var_20938_cast))[name = tensor<string, []>("op_20968_cast")];
            tensor<string, []> var_20970_equation_0 = const()[name = tensor<string, []>("op_20970_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20970_cast = einsum(equation = var_20970_equation_0, values = (var_20810_cast, var_20939_cast))[name = tensor<string, []>("op_20970_cast")];
            tensor<string, []> var_20972_equation_0 = const()[name = tensor<string, []>("op_20972_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20972_cast = einsum(equation = var_20972_equation_0, values = (var_20814_cast, var_20940_cast))[name = tensor<string, []>("op_20972_cast")];
            tensor<string, []> var_20974_equation_0 = const()[name = tensor<string, []>("op_20974_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20974_cast = einsum(equation = var_20974_equation_0, values = (var_20818_cast, var_20941_cast))[name = tensor<string, []>("op_20974_cast")];
            tensor<string, []> var_20976_equation_0 = const()[name = tensor<string, []>("op_20976_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20976_cast = einsum(equation = var_20976_equation_0, values = (var_20822_cast, var_20942_cast))[name = tensor<string, []>("op_20976_cast")];
            tensor<string, []> var_20978_equation_0 = const()[name = tensor<string, []>("op_20978_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20978_cast = einsum(equation = var_20978_equation_0, values = (var_20826_cast, var_20943_cast))[name = tensor<string, []>("op_20978_cast")];
            tensor<string, []> var_20980_equation_0 = const()[name = tensor<string, []>("op_20980_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20980_cast = einsum(equation = var_20980_equation_0, values = (var_20830_cast, var_20944_cast))[name = tensor<string, []>("op_20980_cast")];
            tensor<string, []> var_20982_equation_0 = const()[name = tensor<string, []>("op_20982_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20982_cast = einsum(equation = var_20982_equation_0, values = (var_20834_cast, var_20945_cast))[name = tensor<string, []>("op_20982_cast")];
            tensor<string, []> var_20984_equation_0 = const()[name = tensor<string, []>("op_20984_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20984_cast = einsum(equation = var_20984_equation_0, values = (var_20838_cast, var_20946_cast))[name = tensor<string, []>("op_20984_cast")];
            tensor<string, []> var_20986_equation_0 = const()[name = tensor<string, []>("op_20986_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20986_cast = einsum(equation = var_20986_equation_0, values = (var_20842_cast, var_20947_cast))[name = tensor<string, []>("op_20986_cast")];
            tensor<string, []> var_20988_equation_0 = const()[name = tensor<string, []>("op_20988_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_20988_cast = einsum(equation = var_20988_equation_0, values = (var_20846_cast, var_20948_cast))[name = tensor<string, []>("op_20988_cast")];
            tensor<bool, []> input_307_interleave_0 = const()[name = tensor<string, []>("input_307_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_307_cast = concat(axis = var_2634, interleave = input_307_interleave_0, values = (var_20950_cast, var_20952_cast, var_20954_cast, var_20956_cast, var_20958_cast, var_20960_cast, var_20962_cast, var_20964_cast, var_20966_cast, var_20968_cast, var_20970_cast, var_20972_cast, var_20974_cast, var_20976_cast, var_20978_cast, var_20980_cast, var_20982_cast, var_20984_cast, var_20986_cast, var_20988_cast))[name = tensor<string, []>("input_307_cast")];
            tensor<int32, [2]> var_20994 = const()[name = tensor<string, []>("op_20994"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_20996 = const()[name = tensor<string, []>("op_20996"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_20998_pad_type_0 = const()[name = tensor<string, []>("op_20998_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_20998_pad_0 = const()[name = tensor<string, []>("op_20998_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1614637440)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1617914304)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_20998_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_out_0_bias_to_fp16, dilations = var_20996, groups = var_2634, pad = var_20998_pad_0, pad_type = var_20998_pad_type_0, strides = var_20994, weight = down_blocks_2_attentions_1_transformer_blocks_9_attn2_to_out_0_weight_to_fp16, x = input_307_cast)[name = tensor<string, []>("op_20998_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_143_cast = add(x = var_20998_cast, y = inputs_141_cast)[name = tensor<string, []>("inputs_143_cast")];
            tensor<int32, [1]> var_21002 = const()[name = tensor<string, []>("op_21002"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_143_cast = reduce_mean(axes = var_21002, keep_dims = var_2629, x = inputs_143_cast)[name = tensor<string, []>("channels_mean_143_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_143_cast = sub(x = inputs_143_cast, y = channels_mean_143_cast)[name = tensor<string, []>("zero_mean_143_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_143_cast = mul(x = zero_mean_143_cast, y = zero_mean_143_cast)[name = tensor<string, []>("zero_mean_sq_143_cast")];
            tensor<int32, [1]> var_21006 = const()[name = tensor<string, []>("op_21006"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_21007_cast = reduce_mean(axes = var_21006, keep_dims = var_2629, x = zero_mean_sq_143_cast)[name = tensor<string, []>("op_21007_cast")];
            tensor<fp16, []> var_21008_to_fp16 = const()[name = tensor<string, []>("op_21008_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_21009_cast = add(x = var_21007_cast, y = var_21008_to_fp16)[name = tensor<string, []>("op_21009_cast")];
            tensor<fp16, []> denom_143_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_143_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_143_cast = rsqrt(epsilon = denom_143_epsilon_0_to_fp16, x = var_21009_cast)[name = tensor<string, []>("denom_143_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_143_cast = mul(x = zero_mean_143_cast, y = denom_143_cast)[name = tensor<string, []>("out_143_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_21013_to_fp16 = const()[name = tensor<string, []>("op_21013_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1617916928)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_21014_cast = add(x = out_143_cast, y = var_21013_to_fp16)[name = tensor<string, []>("op_21014_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_21016_to_fp16 = const()[name = tensor<string, []>("op_21016_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1617919552)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_309_cast = mul(x = var_21014_cast, y = var_21016_to_fp16)[name = tensor<string, []>("input_309_cast")];
            tensor<int32, [2]> var_21024 = const()[name = tensor<string, []>("op_21024"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_21026 = const()[name = tensor<string, []>("op_21026"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_21028_pad_type_0 = const()[name = tensor<string, []>("op_21028_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_21028_pad_0 = const()[name = tensor<string, []>("op_21028_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_9_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_9_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1617922176)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_1_transformer_blocks_9_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_9_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1644136640)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_21028_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_9_ff_net_0_proj_bias_to_fp16, dilations = var_21026, groups = var_2634, pad = var_21028_pad_0, pad_type = var_21028_pad_type_0, strides = var_21024, weight = down_blocks_2_attentions_1_transformer_blocks_9_ff_net_0_proj_weight_to_fp16, x = input_309_cast)[name = tensor<string, []>("op_21028_cast")];
            tensor<int32, [2]> var_21029_split_sizes_0 = const()[name = tensor<string, []>("op_21029_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_21029_axis_0 = const()[name = tensor<string, []>("op_21029_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_21029_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_21029_cast_1 = split(axis = var_21029_axis_0, split_sizes = var_21029_split_sizes_0, x = var_21028_cast)[name = tensor<string, []>("op_21029_cast")];
            tensor<string, []> var_21031_mode_0 = const()[name = tensor<string, []>("op_21031_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_21031_cast = gelu(mode = var_21031_mode_0, x = var_21029_cast_1)[name = tensor<string, []>("op_21031_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_311_cast = mul(x = var_21029_cast_0, y = var_21031_cast)[name = tensor<string, []>("input_311_cast")];
            tensor<int32, [2]> var_21035 = const()[name = tensor<string, []>("op_21035"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_21037 = const()[name = tensor<string, []>("op_21037"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_21039_pad_type_0 = const()[name = tensor<string, []>("op_21039_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_21039_pad_0 = const()[name = tensor<string, []>("op_21039_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_9_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_9_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1644157184)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_9_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_9_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1657264448)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_21039_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_9_ff_net_2_bias_to_fp16, dilations = var_21037, groups = var_2634, pad = var_21039_pad_0, pad_type = var_21039_pad_type_0, strides = var_21035, weight = down_blocks_2_attentions_1_transformer_blocks_9_ff_net_2_weight_to_fp16, x = input_311_cast)[name = tensor<string, []>("op_21039_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_197_cast = add(x = var_21039_cast, y = inputs_143_cast)[name = tensor<string, []>("hidden_states_197_cast")];
            tensor<int32, [4]> var_21041 = const()[name = tensor<string, []>("op_21041"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> input_313_cast = reshape(shape = var_21041, x = hidden_states_197_cast)[name = tensor<string, []>("input_313_cast")];
            tensor<int32, [2]> var_21045 = const()[name = tensor<string, []>("op_21045"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_21047 = const()[name = tensor<string, []>("op_21047"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_199_pad_type_0 = const()[name = tensor<string, []>("hidden_states_199_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_199_pad_0 = const()[name = tensor<string, []>("hidden_states_199_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_proj_out_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1657267072)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_proj_out_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1660543936)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_199_cast = conv(bias = down_blocks_2_attentions_1_proj_out_bias_to_fp16, dilations = var_21047, groups = var_2634, pad = hidden_states_199_pad_0, pad_type = hidden_states_199_pad_type_0, strides = var_21045, weight = down_blocks_2_attentions_1_proj_out_weight_to_fp16, x = input_313_cast)[name = tensor<string, []>("hidden_states_199_cast")];
            tensor<fp16, [2, 1280, 32, 32]> sample_5_cast = add(x = hidden_states_199_cast, y = hidden_states_133_cast)[name = tensor<string, []>("sample_5_cast")];
            tensor<fp16, [2, 320, 128, 128]> res_hidden_states_cast = add(x = input_15_cast, y = additional_residual_0)[name = tensor<string, []>("res_hidden_states_cast")];
            tensor<fp16, [2, 320, 128, 128]> res_hidden_states_15_cast = add(x = input_31_cast, y = additional_residual_1)[name = tensor<string, []>("res_hidden_states_15_cast")];
            tensor<fp16, [2, 320, 128, 128]> res_hidden_states_13_cast = add(x = input_45_cast, y = additional_residual_2)[name = tensor<string, []>("res_hidden_states_13_cast")];
            tensor<fp16, [2, 320, 64, 64]> res_hidden_states_11_cast = add(x = input_47_cast, y = additional_residual_3)[name = tensor<string, []>("res_hidden_states_11_cast")];
            tensor<fp16, [2, 640, 64, 64]> res_hidden_states_9_cast = add(x = input_83_cast, y = additional_residual_4)[name = tensor<string, []>("res_hidden_states_9_cast")];
            tensor<fp16, [2, 640, 64, 64]> res_hidden_states_7_cast = add(x = input_117_cast, y = additional_residual_5)[name = tensor<string, []>("res_hidden_states_7_cast")];
            tensor<fp16, [2, 640, 32, 32]> res_hidden_states_5_cast = add(x = input_119_cast, y = additional_residual_6)[name = tensor<string, []>("res_hidden_states_5_cast")];
            tensor<fp16, [2, 1280, 32, 32]> res_hidden_states_3_cast = add(x = input_217_cast, y = additional_residual_7)[name = tensor<string, []>("res_hidden_states_3_cast")];
            tensor<fp16, [2, 1280, 32, 32]> res_hidden_states_1_cast = add(x = sample_5_cast, y = additional_residual_8)[name = tensor<string, []>("res_hidden_states_1_cast")];
            tensor<bool, []> var_21100 = const()[name = tensor<string, []>("op_21100"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_21105 = const()[name = tensor<string, []>("op_21105"), val = tensor<int32, []>(1)];
            tensor<int32, [5]> reshape_64_shape_0 = const()[name = tensor<string, []>("reshape_64_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_64_cast = reshape(shape = reshape_64_shape_0, x = sample_5_cast)[name = tensor<string, []>("reshape_64_cast")];
            tensor<int32, [3]> reduce_mean_48_axes_0 = const()[name = tensor<string, []>("reduce_mean_48_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_48_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_48_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_48_cast = reduce_mean(axes = reduce_mean_48_axes_0, keep_dims = reduce_mean_48_keep_dims_0, x = reshape_64_cast)[name = tensor<string, []>("reduce_mean_48_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_32_cast = sub(x = reshape_64_cast, y = reduce_mean_48_cast)[name = tensor<string, []>("sub_32_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_16_cast = square(x = sub_32_cast)[name = tensor<string, []>("square_16_cast")];
            tensor<int32, [3]> reduce_mean_50_axes_0 = const()[name = tensor<string, []>("reduce_mean_50_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_50_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_50_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_50_cast = reduce_mean(axes = reduce_mean_50_axes_0, keep_dims = reduce_mean_50_keep_dims_0, x = square_16_cast)[name = tensor<string, []>("reduce_mean_50_cast")];
            tensor<fp16, []> add_32_y_0_to_fp16 = const()[name = tensor<string, []>("add_32_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_32_cast = add(x = reduce_mean_50_cast, y = add_32_y_0_to_fp16)[name = tensor<string, []>("add_32_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_16_cast = sqrt(x = add_32_cast)[name = tensor<string, []>("sqrt_16_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_16_cast = real_div(x = sub_32_cast, y = sqrt_16_cast)[name = tensor<string, []>("real_div_16_cast")];
            tensor<int32, [4]> reshape_65_shape_0 = const()[name = tensor<string, []>("reshape_65_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_65_cast = reshape(shape = reshape_65_shape_0, x = real_div_16_cast)[name = tensor<string, []>("reshape_65_cast")];
            tensor<fp16, [1280]> add_33_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_33_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1660546560)))];
            tensor<fp16, [1280]> add_33_beta_0_to_fp16 = const()[name = tensor<string, []>("add_33_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1660549184)))];
            tensor<fp16, []> add_33_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_33_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 32, 32]> add_33_cast = batch_norm(beta = add_33_beta_0_to_fp16, epsilon = add_33_epsilon_0_to_fp16, gamma = add_33_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_65_cast)[name = tensor<string, []>("add_33_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_317_cast = silu(x = add_33_cast)[name = tensor<string, []>("input_317_cast")];
            tensor<int32, [2]> var_21123 = const()[name = tensor<string, []>("op_21123"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_21125 = const()[name = tensor<string, []>("op_21125"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_201_pad_type_0 = const()[name = tensor<string, []>("hidden_states_201_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_201_pad_0 = const()[name = tensor<string, []>("hidden_states_201_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> mid_block_resnets_0_conv1_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1660551808)))];
            tensor<fp16, [1280]> mid_block_resnets_0_conv1_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1690043072)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_201_cast = conv(bias = mid_block_resnets_0_conv1_bias_to_fp16, dilations = var_21125, groups = var_21105, pad = hidden_states_201_pad_0, pad_type = hidden_states_201_pad_type_0, strides = var_21123, weight = mid_block_resnets_0_conv1_weight_to_fp16, x = input_317_cast)[name = tensor<string, []>("hidden_states_201_cast")];
            tensor<int32, [2]> var_21131 = const()[name = tensor<string, []>("op_21131"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_21133 = const()[name = tensor<string, []>("op_21133"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_13_pad_type_0 = const()[name = tensor<string, []>("temb_13_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_13_pad_0 = const()[name = tensor<string, []>("temb_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1690045696)))];
            tensor<fp16, [1280]> mid_block_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1693322560)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_13_cast = conv(bias = mid_block_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_21133, groups = var_21105, pad = temb_13_pad_0, pad_type = temb_13_pad_type_0, strides = var_21131, weight = mid_block_resnets_0_time_emb_proj_weight_to_fp16, x = input_23_cast)[name = tensor<string, []>("temb_13_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_321_cast = add(x = hidden_states_201_cast, y = temb_13_cast)[name = tensor<string, []>("input_321_cast")];
            tensor<int32, [5]> reshape_68_shape_0 = const()[name = tensor<string, []>("reshape_68_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_68_cast = reshape(shape = reshape_68_shape_0, x = input_321_cast)[name = tensor<string, []>("reshape_68_cast")];
            tensor<int32, [3]> reduce_mean_51_axes_0 = const()[name = tensor<string, []>("reduce_mean_51_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_51_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_51_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_51_cast = reduce_mean(axes = reduce_mean_51_axes_0, keep_dims = reduce_mean_51_keep_dims_0, x = reshape_68_cast)[name = tensor<string, []>("reduce_mean_51_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_34_cast = sub(x = reshape_68_cast, y = reduce_mean_51_cast)[name = tensor<string, []>("sub_34_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_17_cast = square(x = sub_34_cast)[name = tensor<string, []>("square_17_cast")];
            tensor<int32, [3]> reduce_mean_53_axes_0 = const()[name = tensor<string, []>("reduce_mean_53_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_53_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_53_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_53_cast = reduce_mean(axes = reduce_mean_53_axes_0, keep_dims = reduce_mean_53_keep_dims_0, x = square_17_cast)[name = tensor<string, []>("reduce_mean_53_cast")];
            tensor<fp16, []> add_34_y_0_to_fp16 = const()[name = tensor<string, []>("add_34_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_34_cast = add(x = reduce_mean_53_cast, y = add_34_y_0_to_fp16)[name = tensor<string, []>("add_34_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_17_cast = sqrt(x = add_34_cast)[name = tensor<string, []>("sqrt_17_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_17_cast = real_div(x = sub_34_cast, y = sqrt_17_cast)[name = tensor<string, []>("real_div_17_cast")];
            tensor<int32, [4]> reshape_69_shape_0 = const()[name = tensor<string, []>("reshape_69_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_69_cast = reshape(shape = reshape_69_shape_0, x = real_div_17_cast)[name = tensor<string, []>("reshape_69_cast")];
            tensor<fp16, [1280]> add_35_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_35_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1693325184)))];
            tensor<fp16, [1280]> add_35_beta_0_to_fp16 = const()[name = tensor<string, []>("add_35_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1693327808)))];
            tensor<fp16, []> add_35_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_35_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 32, 32]> add_35_cast = batch_norm(beta = add_35_beta_0_to_fp16, epsilon = add_35_epsilon_0_to_fp16, gamma = add_35_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_69_cast)[name = tensor<string, []>("add_35_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_325_cast = silu(x = add_35_cast)[name = tensor<string, []>("input_325_cast")];
            tensor<int32, [2]> var_21143 = const()[name = tensor<string, []>("op_21143"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_21145 = const()[name = tensor<string, []>("op_21145"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_203_pad_type_0 = const()[name = tensor<string, []>("hidden_states_203_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_203_pad_0 = const()[name = tensor<string, []>("hidden_states_203_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> mid_block_resnets_0_conv2_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1693330432)))];
            tensor<fp16, [1280]> mid_block_resnets_0_conv2_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1722821696)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_203_cast = conv(bias = mid_block_resnets_0_conv2_bias_to_fp16, dilations = var_21145, groups = var_21105, pad = hidden_states_203_pad_0, pad_type = hidden_states_203_pad_type_0, strides = var_21143, weight = mid_block_resnets_0_conv2_weight_to_fp16, x = input_325_cast)[name = tensor<string, []>("hidden_states_203_cast")];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_205_cast = add(x = sample_5_cast, y = hidden_states_203_cast)[name = tensor<string, []>("hidden_states_205_cast")];
            tensor<int32, [5]> reshape_72_shape_0 = const()[name = tensor<string, []>("reshape_72_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_72_cast = reshape(shape = reshape_72_shape_0, x = hidden_states_205_cast)[name = tensor<string, []>("reshape_72_cast")];
            tensor<int32, [3]> reduce_mean_54_axes_0 = const()[name = tensor<string, []>("reduce_mean_54_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_54_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_54_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_54_cast = reduce_mean(axes = reduce_mean_54_axes_0, keep_dims = reduce_mean_54_keep_dims_0, x = reshape_72_cast)[name = tensor<string, []>("reduce_mean_54_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_36_cast = sub(x = reshape_72_cast, y = reduce_mean_54_cast)[name = tensor<string, []>("sub_36_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_18_cast = square(x = sub_36_cast)[name = tensor<string, []>("square_18_cast")];
            tensor<int32, [3]> reduce_mean_56_axes_0 = const()[name = tensor<string, []>("reduce_mean_56_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_56_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_56_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_56_cast = reduce_mean(axes = reduce_mean_56_axes_0, keep_dims = reduce_mean_56_keep_dims_0, x = square_18_cast)[name = tensor<string, []>("reduce_mean_56_cast")];
            tensor<fp16, []> add_36_y_0_to_fp16 = const()[name = tensor<string, []>("add_36_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_36_cast = add(x = reduce_mean_56_cast, y = add_36_y_0_to_fp16)[name = tensor<string, []>("add_36_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_18_cast = sqrt(x = add_36_cast)[name = tensor<string, []>("sqrt_18_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_18_cast = real_div(x = sub_36_cast, y = sqrt_18_cast)[name = tensor<string, []>("real_div_18_cast")];
            tensor<int32, [4]> reshape_73_shape_0 = const()[name = tensor<string, []>("reshape_73_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_73_cast = reshape(shape = reshape_73_shape_0, x = real_div_18_cast)[name = tensor<string, []>("reshape_73_cast")];
            tensor<fp16, [1280]> add_37_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_37_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1722824320)))];
            tensor<fp16, [1280]> add_37_beta_0_to_fp16 = const()[name = tensor<string, []>("add_37_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1722826944)))];
            tensor<fp16, []> add_37_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_37_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 32, 32]> add_37_cast = batch_norm(beta = add_37_beta_0_to_fp16, epsilon = add_37_epsilon_0_to_fp16, gamma = add_37_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_73_cast)[name = tensor<string, []>("add_37_cast")];
            tensor<int32, [2]> var_21183 = const()[name = tensor<string, []>("op_21183"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_21185 = const()[name = tensor<string, []>("op_21185"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_207_pad_type_0 = const()[name = tensor<string, []>("hidden_states_207_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_207_pad_0 = const()[name = tensor<string, []>("hidden_states_207_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_proj_in_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1722829568)))];
            tensor<fp16, [1280]> mid_block_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_proj_in_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1726106432)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_207_cast = conv(bias = mid_block_attentions_0_proj_in_bias_to_fp16, dilations = var_21185, groups = var_21105, pad = hidden_states_207_pad_0, pad_type = hidden_states_207_pad_type_0, strides = var_21183, weight = mid_block_attentions_0_proj_in_weight_to_fp16, x = add_37_cast)[name = tensor<string, []>("hidden_states_207_cast")];
            tensor<int32, [4]> var_21190 = const()[name = tensor<string, []>("op_21190"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_145_cast = reshape(shape = var_21190, x = hidden_states_207_cast)[name = tensor<string, []>("inputs_145_cast")];
            tensor<int32, [1]> var_21200 = const()[name = tensor<string, []>("op_21200"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_145_cast = reduce_mean(axes = var_21200, keep_dims = var_21100, x = inputs_145_cast)[name = tensor<string, []>("channels_mean_145_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_145_cast = sub(x = inputs_145_cast, y = channels_mean_145_cast)[name = tensor<string, []>("zero_mean_145_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_145_cast = mul(x = zero_mean_145_cast, y = zero_mean_145_cast)[name = tensor<string, []>("zero_mean_sq_145_cast")];
            tensor<int32, [1]> var_21204 = const()[name = tensor<string, []>("op_21204"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_21205_cast = reduce_mean(axes = var_21204, keep_dims = var_21100, x = zero_mean_sq_145_cast)[name = tensor<string, []>("op_21205_cast")];
            tensor<fp16, []> var_21206_to_fp16 = const()[name = tensor<string, []>("op_21206_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_21207_cast = add(x = var_21205_cast, y = var_21206_to_fp16)[name = tensor<string, []>("op_21207_cast")];
            tensor<fp16, []> denom_145_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_145_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_145_cast = rsqrt(epsilon = denom_145_epsilon_0_to_fp16, x = var_21207_cast)[name = tensor<string, []>("denom_145_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_145_cast = mul(x = zero_mean_145_cast, y = denom_145_cast)[name = tensor<string, []>("out_145_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_21211_to_fp16 = const()[name = tensor<string, []>("op_21211_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1726109056)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_21212_cast = add(x = out_145_cast, y = var_21211_to_fp16)[name = tensor<string, []>("op_21212_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_21214_to_fp16 = const()[name = tensor<string, []>("op_21214_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1726111680)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_209_cast = mul(x = var_21212_cast, y = var_21214_to_fp16)[name = tensor<string, []>("hidden_states_209_cast")];
            tensor<int32, [2]> var_21221 = const()[name = tensor<string, []>("op_21221"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_21223 = const()[name = tensor<string, []>("op_21223"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_97_pad_type_0 = const()[name = tensor<string, []>("q_97_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_97_pad_0 = const()[name = tensor<string, []>("q_97_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1726114304)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_97_cast = conv(dilations = var_21223, groups = var_21105, pad = q_97_pad_0, pad_type = q_97_pad_type_0, strides = var_21221, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_209_cast)[name = tensor<string, []>("q_97_cast")];
            tensor<int32, [2]> var_21227 = const()[name = tensor<string, []>("op_21227"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_21229 = const()[name = tensor<string, []>("op_21229"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_193_pad_type_0 = const()[name = tensor<string, []>("k_193_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_193_pad_0 = const()[name = tensor<string, []>("k_193_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1729391168)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_193_cast = conv(dilations = var_21229, groups = var_21105, pad = k_193_pad_0, pad_type = k_193_pad_type_0, strides = var_21227, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_209_cast)[name = tensor<string, []>("k_193_cast")];
            tensor<int32, [2]> var_21233 = const()[name = tensor<string, []>("op_21233"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_21235 = const()[name = tensor<string, []>("op_21235"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_97_pad_type_0 = const()[name = tensor<string, []>("v_97_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_97_pad_0 = const()[name = tensor<string, []>("v_97_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1732668032)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_97_cast = conv(dilations = var_21235, groups = var_21105, pad = v_97_pad_0, pad_type = v_97_pad_type_0, strides = var_21233, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_209_cast)[name = tensor<string, []>("v_97_cast")];
            tensor<int32, [4]> var_21239_begin_0 = const()[name = tensor<string, []>("op_21239_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_21239_end_0 = const()[name = tensor<string, []>("op_21239_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_21239_end_mask_0 = const()[name = tensor<string, []>("op_21239_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21239_cast = slice_by_index(begin = var_21239_begin_0, end = var_21239_end_0, end_mask = var_21239_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21239_cast")];
            tensor<int32, [4]> var_21243_begin_0 = const()[name = tensor<string, []>("op_21243_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_21243_end_0 = const()[name = tensor<string, []>("op_21243_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_21243_end_mask_0 = const()[name = tensor<string, []>("op_21243_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21243_cast = slice_by_index(begin = var_21243_begin_0, end = var_21243_end_0, end_mask = var_21243_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21243_cast")];
            tensor<int32, [4]> var_21247_begin_0 = const()[name = tensor<string, []>("op_21247_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_21247_end_0 = const()[name = tensor<string, []>("op_21247_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_21247_end_mask_0 = const()[name = tensor<string, []>("op_21247_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21247_cast = slice_by_index(begin = var_21247_begin_0, end = var_21247_end_0, end_mask = var_21247_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21247_cast")];
            tensor<int32, [4]> var_21251_begin_0 = const()[name = tensor<string, []>("op_21251_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_21251_end_0 = const()[name = tensor<string, []>("op_21251_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_21251_end_mask_0 = const()[name = tensor<string, []>("op_21251_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21251_cast = slice_by_index(begin = var_21251_begin_0, end = var_21251_end_0, end_mask = var_21251_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21251_cast")];
            tensor<int32, [4]> var_21255_begin_0 = const()[name = tensor<string, []>("op_21255_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_21255_end_0 = const()[name = tensor<string, []>("op_21255_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_21255_end_mask_0 = const()[name = tensor<string, []>("op_21255_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21255_cast = slice_by_index(begin = var_21255_begin_0, end = var_21255_end_0, end_mask = var_21255_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21255_cast")];
            tensor<int32, [4]> var_21259_begin_0 = const()[name = tensor<string, []>("op_21259_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_21259_end_0 = const()[name = tensor<string, []>("op_21259_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_21259_end_mask_0 = const()[name = tensor<string, []>("op_21259_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21259_cast = slice_by_index(begin = var_21259_begin_0, end = var_21259_end_0, end_mask = var_21259_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21259_cast")];
            tensor<int32, [4]> var_21263_begin_0 = const()[name = tensor<string, []>("op_21263_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_21263_end_0 = const()[name = tensor<string, []>("op_21263_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_21263_end_mask_0 = const()[name = tensor<string, []>("op_21263_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21263_cast = slice_by_index(begin = var_21263_begin_0, end = var_21263_end_0, end_mask = var_21263_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21263_cast")];
            tensor<int32, [4]> var_21267_begin_0 = const()[name = tensor<string, []>("op_21267_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_21267_end_0 = const()[name = tensor<string, []>("op_21267_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_21267_end_mask_0 = const()[name = tensor<string, []>("op_21267_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21267_cast = slice_by_index(begin = var_21267_begin_0, end = var_21267_end_0, end_mask = var_21267_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21267_cast")];
            tensor<int32, [4]> var_21271_begin_0 = const()[name = tensor<string, []>("op_21271_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_21271_end_0 = const()[name = tensor<string, []>("op_21271_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_21271_end_mask_0 = const()[name = tensor<string, []>("op_21271_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21271_cast = slice_by_index(begin = var_21271_begin_0, end = var_21271_end_0, end_mask = var_21271_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21271_cast")];
            tensor<int32, [4]> var_21275_begin_0 = const()[name = tensor<string, []>("op_21275_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_21275_end_0 = const()[name = tensor<string, []>("op_21275_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_21275_end_mask_0 = const()[name = tensor<string, []>("op_21275_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21275_cast = slice_by_index(begin = var_21275_begin_0, end = var_21275_end_0, end_mask = var_21275_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21275_cast")];
            tensor<int32, [4]> var_21279_begin_0 = const()[name = tensor<string, []>("op_21279_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_21279_end_0 = const()[name = tensor<string, []>("op_21279_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_21279_end_mask_0 = const()[name = tensor<string, []>("op_21279_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21279_cast = slice_by_index(begin = var_21279_begin_0, end = var_21279_end_0, end_mask = var_21279_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21279_cast")];
            tensor<int32, [4]> var_21283_begin_0 = const()[name = tensor<string, []>("op_21283_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_21283_end_0 = const()[name = tensor<string, []>("op_21283_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_21283_end_mask_0 = const()[name = tensor<string, []>("op_21283_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21283_cast = slice_by_index(begin = var_21283_begin_0, end = var_21283_end_0, end_mask = var_21283_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21283_cast")];
            tensor<int32, [4]> var_21287_begin_0 = const()[name = tensor<string, []>("op_21287_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_21287_end_0 = const()[name = tensor<string, []>("op_21287_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_21287_end_mask_0 = const()[name = tensor<string, []>("op_21287_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21287_cast = slice_by_index(begin = var_21287_begin_0, end = var_21287_end_0, end_mask = var_21287_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21287_cast")];
            tensor<int32, [4]> var_21291_begin_0 = const()[name = tensor<string, []>("op_21291_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_21291_end_0 = const()[name = tensor<string, []>("op_21291_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_21291_end_mask_0 = const()[name = tensor<string, []>("op_21291_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21291_cast = slice_by_index(begin = var_21291_begin_0, end = var_21291_end_0, end_mask = var_21291_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21291_cast")];
            tensor<int32, [4]> var_21295_begin_0 = const()[name = tensor<string, []>("op_21295_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_21295_end_0 = const()[name = tensor<string, []>("op_21295_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_21295_end_mask_0 = const()[name = tensor<string, []>("op_21295_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21295_cast = slice_by_index(begin = var_21295_begin_0, end = var_21295_end_0, end_mask = var_21295_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21295_cast")];
            tensor<int32, [4]> var_21299_begin_0 = const()[name = tensor<string, []>("op_21299_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_21299_end_0 = const()[name = tensor<string, []>("op_21299_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_21299_end_mask_0 = const()[name = tensor<string, []>("op_21299_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21299_cast = slice_by_index(begin = var_21299_begin_0, end = var_21299_end_0, end_mask = var_21299_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21299_cast")];
            tensor<int32, [4]> var_21303_begin_0 = const()[name = tensor<string, []>("op_21303_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_21303_end_0 = const()[name = tensor<string, []>("op_21303_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_21303_end_mask_0 = const()[name = tensor<string, []>("op_21303_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21303_cast = slice_by_index(begin = var_21303_begin_0, end = var_21303_end_0, end_mask = var_21303_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21303_cast")];
            tensor<int32, [4]> var_21307_begin_0 = const()[name = tensor<string, []>("op_21307_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_21307_end_0 = const()[name = tensor<string, []>("op_21307_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_21307_end_mask_0 = const()[name = tensor<string, []>("op_21307_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21307_cast = slice_by_index(begin = var_21307_begin_0, end = var_21307_end_0, end_mask = var_21307_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21307_cast")];
            tensor<int32, [4]> var_21311_begin_0 = const()[name = tensor<string, []>("op_21311_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_21311_end_0 = const()[name = tensor<string, []>("op_21311_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_21311_end_mask_0 = const()[name = tensor<string, []>("op_21311_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21311_cast = slice_by_index(begin = var_21311_begin_0, end = var_21311_end_0, end_mask = var_21311_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21311_cast")];
            tensor<int32, [4]> var_21315_begin_0 = const()[name = tensor<string, []>("op_21315_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_21315_end_0 = const()[name = tensor<string, []>("op_21315_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_21315_end_mask_0 = const()[name = tensor<string, []>("op_21315_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21315_cast = slice_by_index(begin = var_21315_begin_0, end = var_21315_end_0, end_mask = var_21315_end_mask_0, x = q_97_cast)[name = tensor<string, []>("op_21315_cast")];
            tensor<int32, [4]> k_195_perm_0 = const()[name = tensor<string, []>("k_195_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_21322_begin_0 = const()[name = tensor<string, []>("op_21322_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_21322_end_0 = const()[name = tensor<string, []>("op_21322_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_21322_end_mask_0 = const()[name = tensor<string, []>("op_21322_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_91 = transpose(perm = k_195_perm_0, x = k_193_cast)[name = tensor<string, []>("transpose_91")];
            tensor<fp16, [2, 1024, 1, 64]> var_21322_cast = slice_by_index(begin = var_21322_begin_0, end = var_21322_end_0, end_mask = var_21322_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21322_cast")];
            tensor<int32, [4]> var_21326_begin_0 = const()[name = tensor<string, []>("op_21326_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_21326_end_0 = const()[name = tensor<string, []>("op_21326_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_21326_end_mask_0 = const()[name = tensor<string, []>("op_21326_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_21326_cast = slice_by_index(begin = var_21326_begin_0, end = var_21326_end_0, end_mask = var_21326_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21326_cast")];
            tensor<int32, [4]> var_21330_begin_0 = const()[name = tensor<string, []>("op_21330_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_21330_end_0 = const()[name = tensor<string, []>("op_21330_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_21330_end_mask_0 = const()[name = tensor<string, []>("op_21330_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_21330_cast = slice_by_index(begin = var_21330_begin_0, end = var_21330_end_0, end_mask = var_21330_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21330_cast")];
            tensor<int32, [4]> var_21334_begin_0 = const()[name = tensor<string, []>("op_21334_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_21334_end_0 = const()[name = tensor<string, []>("op_21334_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_21334_end_mask_0 = const()[name = tensor<string, []>("op_21334_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_21334_cast = slice_by_index(begin = var_21334_begin_0, end = var_21334_end_0, end_mask = var_21334_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21334_cast")];
            tensor<int32, [4]> var_21338_begin_0 = const()[name = tensor<string, []>("op_21338_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_21338_end_0 = const()[name = tensor<string, []>("op_21338_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_21338_end_mask_0 = const()[name = tensor<string, []>("op_21338_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_21338_cast = slice_by_index(begin = var_21338_begin_0, end = var_21338_end_0, end_mask = var_21338_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21338_cast")];
            tensor<int32, [4]> var_21342_begin_0 = const()[name = tensor<string, []>("op_21342_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_21342_end_0 = const()[name = tensor<string, []>("op_21342_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_21342_end_mask_0 = const()[name = tensor<string, []>("op_21342_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_21342_cast = slice_by_index(begin = var_21342_begin_0, end = var_21342_end_0, end_mask = var_21342_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21342_cast")];
            tensor<int32, [4]> var_21346_begin_0 = const()[name = tensor<string, []>("op_21346_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_21346_end_0 = const()[name = tensor<string, []>("op_21346_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_21346_end_mask_0 = const()[name = tensor<string, []>("op_21346_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_21346_cast = slice_by_index(begin = var_21346_begin_0, end = var_21346_end_0, end_mask = var_21346_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21346_cast")];
            tensor<int32, [4]> var_21350_begin_0 = const()[name = tensor<string, []>("op_21350_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_21350_end_0 = const()[name = tensor<string, []>("op_21350_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_21350_end_mask_0 = const()[name = tensor<string, []>("op_21350_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_21350_cast = slice_by_index(begin = var_21350_begin_0, end = var_21350_end_0, end_mask = var_21350_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21350_cast")];
            tensor<int32, [4]> var_21354_begin_0 = const()[name = tensor<string, []>("op_21354_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_21354_end_0 = const()[name = tensor<string, []>("op_21354_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_21354_end_mask_0 = const()[name = tensor<string, []>("op_21354_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_21354_cast = slice_by_index(begin = var_21354_begin_0, end = var_21354_end_0, end_mask = var_21354_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21354_cast")];
            tensor<int32, [4]> var_21358_begin_0 = const()[name = tensor<string, []>("op_21358_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_21358_end_0 = const()[name = tensor<string, []>("op_21358_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_21358_end_mask_0 = const()[name = tensor<string, []>("op_21358_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_21358_cast = slice_by_index(begin = var_21358_begin_0, end = var_21358_end_0, end_mask = var_21358_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21358_cast")];
            tensor<int32, [4]> var_21362_begin_0 = const()[name = tensor<string, []>("op_21362_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_21362_end_0 = const()[name = tensor<string, []>("op_21362_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_21362_end_mask_0 = const()[name = tensor<string, []>("op_21362_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_21362_cast = slice_by_index(begin = var_21362_begin_0, end = var_21362_end_0, end_mask = var_21362_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21362_cast")];
            tensor<int32, [4]> var_21366_begin_0 = const()[name = tensor<string, []>("op_21366_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_21366_end_0 = const()[name = tensor<string, []>("op_21366_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_21366_end_mask_0 = const()[name = tensor<string, []>("op_21366_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_21366_cast = slice_by_index(begin = var_21366_begin_0, end = var_21366_end_0, end_mask = var_21366_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21366_cast")];
            tensor<int32, [4]> var_21370_begin_0 = const()[name = tensor<string, []>("op_21370_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_21370_end_0 = const()[name = tensor<string, []>("op_21370_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_21370_end_mask_0 = const()[name = tensor<string, []>("op_21370_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_21370_cast = slice_by_index(begin = var_21370_begin_0, end = var_21370_end_0, end_mask = var_21370_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21370_cast")];
            tensor<int32, [4]> var_21374_begin_0 = const()[name = tensor<string, []>("op_21374_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_21374_end_0 = const()[name = tensor<string, []>("op_21374_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_21374_end_mask_0 = const()[name = tensor<string, []>("op_21374_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_21374_cast = slice_by_index(begin = var_21374_begin_0, end = var_21374_end_0, end_mask = var_21374_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21374_cast")];
            tensor<int32, [4]> var_21378_begin_0 = const()[name = tensor<string, []>("op_21378_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_21378_end_0 = const()[name = tensor<string, []>("op_21378_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_21378_end_mask_0 = const()[name = tensor<string, []>("op_21378_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_21378_cast = slice_by_index(begin = var_21378_begin_0, end = var_21378_end_0, end_mask = var_21378_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21378_cast")];
            tensor<int32, [4]> var_21382_begin_0 = const()[name = tensor<string, []>("op_21382_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_21382_end_0 = const()[name = tensor<string, []>("op_21382_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_21382_end_mask_0 = const()[name = tensor<string, []>("op_21382_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_21382_cast = slice_by_index(begin = var_21382_begin_0, end = var_21382_end_0, end_mask = var_21382_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21382_cast")];
            tensor<int32, [4]> var_21386_begin_0 = const()[name = tensor<string, []>("op_21386_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_21386_end_0 = const()[name = tensor<string, []>("op_21386_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_21386_end_mask_0 = const()[name = tensor<string, []>("op_21386_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_21386_cast = slice_by_index(begin = var_21386_begin_0, end = var_21386_end_0, end_mask = var_21386_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21386_cast")];
            tensor<int32, [4]> var_21390_begin_0 = const()[name = tensor<string, []>("op_21390_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_21390_end_0 = const()[name = tensor<string, []>("op_21390_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_21390_end_mask_0 = const()[name = tensor<string, []>("op_21390_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_21390_cast = slice_by_index(begin = var_21390_begin_0, end = var_21390_end_0, end_mask = var_21390_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21390_cast")];
            tensor<int32, [4]> var_21394_begin_0 = const()[name = tensor<string, []>("op_21394_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_21394_end_0 = const()[name = tensor<string, []>("op_21394_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_21394_end_mask_0 = const()[name = tensor<string, []>("op_21394_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_21394_cast = slice_by_index(begin = var_21394_begin_0, end = var_21394_end_0, end_mask = var_21394_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21394_cast")];
            tensor<int32, [4]> var_21398_begin_0 = const()[name = tensor<string, []>("op_21398_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_21398_end_0 = const()[name = tensor<string, []>("op_21398_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_21398_end_mask_0 = const()[name = tensor<string, []>("op_21398_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_21398_cast = slice_by_index(begin = var_21398_begin_0, end = var_21398_end_0, end_mask = var_21398_end_mask_0, x = transpose_91)[name = tensor<string, []>("op_21398_cast")];
            tensor<int32, [4]> var_21400_begin_0 = const()[name = tensor<string, []>("op_21400_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_21400_end_0 = const()[name = tensor<string, []>("op_21400_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_21400_end_mask_0 = const()[name = tensor<string, []>("op_21400_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21400_cast = slice_by_index(begin = var_21400_begin_0, end = var_21400_end_0, end_mask = var_21400_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21400_cast")];
            tensor<int32, [4]> var_21404_begin_0 = const()[name = tensor<string, []>("op_21404_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_21404_end_0 = const()[name = tensor<string, []>("op_21404_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_21404_end_mask_0 = const()[name = tensor<string, []>("op_21404_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21404_cast = slice_by_index(begin = var_21404_begin_0, end = var_21404_end_0, end_mask = var_21404_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21404_cast")];
            tensor<int32, [4]> var_21408_begin_0 = const()[name = tensor<string, []>("op_21408_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_21408_end_0 = const()[name = tensor<string, []>("op_21408_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_21408_end_mask_0 = const()[name = tensor<string, []>("op_21408_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21408_cast = slice_by_index(begin = var_21408_begin_0, end = var_21408_end_0, end_mask = var_21408_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21408_cast")];
            tensor<int32, [4]> var_21412_begin_0 = const()[name = tensor<string, []>("op_21412_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_21412_end_0 = const()[name = tensor<string, []>("op_21412_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_21412_end_mask_0 = const()[name = tensor<string, []>("op_21412_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21412_cast = slice_by_index(begin = var_21412_begin_0, end = var_21412_end_0, end_mask = var_21412_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21412_cast")];
            tensor<int32, [4]> var_21416_begin_0 = const()[name = tensor<string, []>("op_21416_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_21416_end_0 = const()[name = tensor<string, []>("op_21416_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_21416_end_mask_0 = const()[name = tensor<string, []>("op_21416_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21416_cast = slice_by_index(begin = var_21416_begin_0, end = var_21416_end_0, end_mask = var_21416_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21416_cast")];
            tensor<int32, [4]> var_21420_begin_0 = const()[name = tensor<string, []>("op_21420_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_21420_end_0 = const()[name = tensor<string, []>("op_21420_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_21420_end_mask_0 = const()[name = tensor<string, []>("op_21420_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21420_cast = slice_by_index(begin = var_21420_begin_0, end = var_21420_end_0, end_mask = var_21420_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21420_cast")];
            tensor<int32, [4]> var_21424_begin_0 = const()[name = tensor<string, []>("op_21424_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_21424_end_0 = const()[name = tensor<string, []>("op_21424_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_21424_end_mask_0 = const()[name = tensor<string, []>("op_21424_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21424_cast = slice_by_index(begin = var_21424_begin_0, end = var_21424_end_0, end_mask = var_21424_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21424_cast")];
            tensor<int32, [4]> var_21428_begin_0 = const()[name = tensor<string, []>("op_21428_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_21428_end_0 = const()[name = tensor<string, []>("op_21428_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_21428_end_mask_0 = const()[name = tensor<string, []>("op_21428_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21428_cast = slice_by_index(begin = var_21428_begin_0, end = var_21428_end_0, end_mask = var_21428_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21428_cast")];
            tensor<int32, [4]> var_21432_begin_0 = const()[name = tensor<string, []>("op_21432_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_21432_end_0 = const()[name = tensor<string, []>("op_21432_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_21432_end_mask_0 = const()[name = tensor<string, []>("op_21432_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21432_cast = slice_by_index(begin = var_21432_begin_0, end = var_21432_end_0, end_mask = var_21432_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21432_cast")];
            tensor<int32, [4]> var_21436_begin_0 = const()[name = tensor<string, []>("op_21436_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_21436_end_0 = const()[name = tensor<string, []>("op_21436_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_21436_end_mask_0 = const()[name = tensor<string, []>("op_21436_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21436_cast = slice_by_index(begin = var_21436_begin_0, end = var_21436_end_0, end_mask = var_21436_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21436_cast")];
            tensor<int32, [4]> var_21440_begin_0 = const()[name = tensor<string, []>("op_21440_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_21440_end_0 = const()[name = tensor<string, []>("op_21440_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_21440_end_mask_0 = const()[name = tensor<string, []>("op_21440_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21440_cast = slice_by_index(begin = var_21440_begin_0, end = var_21440_end_0, end_mask = var_21440_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21440_cast")];
            tensor<int32, [4]> var_21444_begin_0 = const()[name = tensor<string, []>("op_21444_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_21444_end_0 = const()[name = tensor<string, []>("op_21444_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_21444_end_mask_0 = const()[name = tensor<string, []>("op_21444_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21444_cast = slice_by_index(begin = var_21444_begin_0, end = var_21444_end_0, end_mask = var_21444_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21444_cast")];
            tensor<int32, [4]> var_21448_begin_0 = const()[name = tensor<string, []>("op_21448_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_21448_end_0 = const()[name = tensor<string, []>("op_21448_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_21448_end_mask_0 = const()[name = tensor<string, []>("op_21448_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21448_cast = slice_by_index(begin = var_21448_begin_0, end = var_21448_end_0, end_mask = var_21448_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21448_cast")];
            tensor<int32, [4]> var_21452_begin_0 = const()[name = tensor<string, []>("op_21452_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_21452_end_0 = const()[name = tensor<string, []>("op_21452_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_21452_end_mask_0 = const()[name = tensor<string, []>("op_21452_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21452_cast = slice_by_index(begin = var_21452_begin_0, end = var_21452_end_0, end_mask = var_21452_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21452_cast")];
            tensor<int32, [4]> var_21456_begin_0 = const()[name = tensor<string, []>("op_21456_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_21456_end_0 = const()[name = tensor<string, []>("op_21456_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_21456_end_mask_0 = const()[name = tensor<string, []>("op_21456_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21456_cast = slice_by_index(begin = var_21456_begin_0, end = var_21456_end_0, end_mask = var_21456_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21456_cast")];
            tensor<int32, [4]> var_21460_begin_0 = const()[name = tensor<string, []>("op_21460_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_21460_end_0 = const()[name = tensor<string, []>("op_21460_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_21460_end_mask_0 = const()[name = tensor<string, []>("op_21460_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21460_cast = slice_by_index(begin = var_21460_begin_0, end = var_21460_end_0, end_mask = var_21460_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21460_cast")];
            tensor<int32, [4]> var_21464_begin_0 = const()[name = tensor<string, []>("op_21464_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_21464_end_0 = const()[name = tensor<string, []>("op_21464_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_21464_end_mask_0 = const()[name = tensor<string, []>("op_21464_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21464_cast = slice_by_index(begin = var_21464_begin_0, end = var_21464_end_0, end_mask = var_21464_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21464_cast")];
            tensor<int32, [4]> var_21468_begin_0 = const()[name = tensor<string, []>("op_21468_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_21468_end_0 = const()[name = tensor<string, []>("op_21468_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_21468_end_mask_0 = const()[name = tensor<string, []>("op_21468_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21468_cast = slice_by_index(begin = var_21468_begin_0, end = var_21468_end_0, end_mask = var_21468_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21468_cast")];
            tensor<int32, [4]> var_21472_begin_0 = const()[name = tensor<string, []>("op_21472_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_21472_end_0 = const()[name = tensor<string, []>("op_21472_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_21472_end_mask_0 = const()[name = tensor<string, []>("op_21472_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21472_cast = slice_by_index(begin = var_21472_begin_0, end = var_21472_end_0, end_mask = var_21472_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21472_cast")];
            tensor<int32, [4]> var_21476_begin_0 = const()[name = tensor<string, []>("op_21476_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_21476_end_0 = const()[name = tensor<string, []>("op_21476_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_21476_end_mask_0 = const()[name = tensor<string, []>("op_21476_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21476_cast = slice_by_index(begin = var_21476_begin_0, end = var_21476_end_0, end_mask = var_21476_end_mask_0, x = v_97_cast)[name = tensor<string, []>("op_21476_cast")];
            tensor<string, []> var_21480_equation_0 = const()[name = tensor<string, []>("op_21480_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21480_cast = einsum(equation = var_21480_equation_0, values = (var_21322_cast, var_21239_cast))[name = tensor<string, []>("op_21480_cast")];
            tensor<fp16, []> var_21481_to_fp16 = const()[name = tensor<string, []>("op_21481_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1761_cast = mul(x = var_21480_cast, y = var_21481_to_fp16)[name = tensor<string, []>("aw_1761_cast")];
            tensor<string, []> var_21484_equation_0 = const()[name = tensor<string, []>("op_21484_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21484_cast = einsum(equation = var_21484_equation_0, values = (var_21326_cast, var_21243_cast))[name = tensor<string, []>("op_21484_cast")];
            tensor<fp16, []> var_21485_to_fp16 = const()[name = tensor<string, []>("op_21485_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1763_cast = mul(x = var_21484_cast, y = var_21485_to_fp16)[name = tensor<string, []>("aw_1763_cast")];
            tensor<string, []> var_21488_equation_0 = const()[name = tensor<string, []>("op_21488_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21488_cast = einsum(equation = var_21488_equation_0, values = (var_21330_cast, var_21247_cast))[name = tensor<string, []>("op_21488_cast")];
            tensor<fp16, []> var_21489_to_fp16 = const()[name = tensor<string, []>("op_21489_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1765_cast = mul(x = var_21488_cast, y = var_21489_to_fp16)[name = tensor<string, []>("aw_1765_cast")];
            tensor<string, []> var_21492_equation_0 = const()[name = tensor<string, []>("op_21492_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21492_cast = einsum(equation = var_21492_equation_0, values = (var_21334_cast, var_21251_cast))[name = tensor<string, []>("op_21492_cast")];
            tensor<fp16, []> var_21493_to_fp16 = const()[name = tensor<string, []>("op_21493_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1767_cast = mul(x = var_21492_cast, y = var_21493_to_fp16)[name = tensor<string, []>("aw_1767_cast")];
            tensor<string, []> var_21496_equation_0 = const()[name = tensor<string, []>("op_21496_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21496_cast = einsum(equation = var_21496_equation_0, values = (var_21338_cast, var_21255_cast))[name = tensor<string, []>("op_21496_cast")];
            tensor<fp16, []> var_21497_to_fp16 = const()[name = tensor<string, []>("op_21497_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1769_cast = mul(x = var_21496_cast, y = var_21497_to_fp16)[name = tensor<string, []>("aw_1769_cast")];
            tensor<string, []> var_21500_equation_0 = const()[name = tensor<string, []>("op_21500_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21500_cast = einsum(equation = var_21500_equation_0, values = (var_21342_cast, var_21259_cast))[name = tensor<string, []>("op_21500_cast")];
            tensor<fp16, []> var_21501_to_fp16 = const()[name = tensor<string, []>("op_21501_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1771_cast = mul(x = var_21500_cast, y = var_21501_to_fp16)[name = tensor<string, []>("aw_1771_cast")];
            tensor<string, []> var_21504_equation_0 = const()[name = tensor<string, []>("op_21504_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21504_cast = einsum(equation = var_21504_equation_0, values = (var_21346_cast, var_21263_cast))[name = tensor<string, []>("op_21504_cast")];
            tensor<fp16, []> var_21505_to_fp16 = const()[name = tensor<string, []>("op_21505_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1773_cast = mul(x = var_21504_cast, y = var_21505_to_fp16)[name = tensor<string, []>("aw_1773_cast")];
            tensor<string, []> var_21508_equation_0 = const()[name = tensor<string, []>("op_21508_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21508_cast = einsum(equation = var_21508_equation_0, values = (var_21350_cast, var_21267_cast))[name = tensor<string, []>("op_21508_cast")];
            tensor<fp16, []> var_21509_to_fp16 = const()[name = tensor<string, []>("op_21509_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1775_cast = mul(x = var_21508_cast, y = var_21509_to_fp16)[name = tensor<string, []>("aw_1775_cast")];
            tensor<string, []> var_21512_equation_0 = const()[name = tensor<string, []>("op_21512_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21512_cast = einsum(equation = var_21512_equation_0, values = (var_21354_cast, var_21271_cast))[name = tensor<string, []>("op_21512_cast")];
            tensor<fp16, []> var_21513_to_fp16 = const()[name = tensor<string, []>("op_21513_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1777_cast = mul(x = var_21512_cast, y = var_21513_to_fp16)[name = tensor<string, []>("aw_1777_cast")];
            tensor<string, []> var_21516_equation_0 = const()[name = tensor<string, []>("op_21516_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21516_cast = einsum(equation = var_21516_equation_0, values = (var_21358_cast, var_21275_cast))[name = tensor<string, []>("op_21516_cast")];
            tensor<fp16, []> var_21517_to_fp16 = const()[name = tensor<string, []>("op_21517_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1779_cast = mul(x = var_21516_cast, y = var_21517_to_fp16)[name = tensor<string, []>("aw_1779_cast")];
            tensor<string, []> var_21520_equation_0 = const()[name = tensor<string, []>("op_21520_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21520_cast = einsum(equation = var_21520_equation_0, values = (var_21362_cast, var_21279_cast))[name = tensor<string, []>("op_21520_cast")];
            tensor<fp16, []> var_21521_to_fp16 = const()[name = tensor<string, []>("op_21521_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1781_cast = mul(x = var_21520_cast, y = var_21521_to_fp16)[name = tensor<string, []>("aw_1781_cast")];
            tensor<string, []> var_21524_equation_0 = const()[name = tensor<string, []>("op_21524_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21524_cast = einsum(equation = var_21524_equation_0, values = (var_21366_cast, var_21283_cast))[name = tensor<string, []>("op_21524_cast")];
            tensor<fp16, []> var_21525_to_fp16 = const()[name = tensor<string, []>("op_21525_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1783_cast = mul(x = var_21524_cast, y = var_21525_to_fp16)[name = tensor<string, []>("aw_1783_cast")];
            tensor<string, []> var_21528_equation_0 = const()[name = tensor<string, []>("op_21528_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21528_cast = einsum(equation = var_21528_equation_0, values = (var_21370_cast, var_21287_cast))[name = tensor<string, []>("op_21528_cast")];
            tensor<fp16, []> var_21529_to_fp16 = const()[name = tensor<string, []>("op_21529_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1785_cast = mul(x = var_21528_cast, y = var_21529_to_fp16)[name = tensor<string, []>("aw_1785_cast")];
            tensor<string, []> var_21532_equation_0 = const()[name = tensor<string, []>("op_21532_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21532_cast = einsum(equation = var_21532_equation_0, values = (var_21374_cast, var_21291_cast))[name = tensor<string, []>("op_21532_cast")];
            tensor<fp16, []> var_21533_to_fp16 = const()[name = tensor<string, []>("op_21533_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1787_cast = mul(x = var_21532_cast, y = var_21533_to_fp16)[name = tensor<string, []>("aw_1787_cast")];
            tensor<string, []> var_21536_equation_0 = const()[name = tensor<string, []>("op_21536_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21536_cast = einsum(equation = var_21536_equation_0, values = (var_21378_cast, var_21295_cast))[name = tensor<string, []>("op_21536_cast")];
            tensor<fp16, []> var_21537_to_fp16 = const()[name = tensor<string, []>("op_21537_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1789_cast = mul(x = var_21536_cast, y = var_21537_to_fp16)[name = tensor<string, []>("aw_1789_cast")];
            tensor<string, []> var_21540_equation_0 = const()[name = tensor<string, []>("op_21540_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21540_cast = einsum(equation = var_21540_equation_0, values = (var_21382_cast, var_21299_cast))[name = tensor<string, []>("op_21540_cast")];
            tensor<fp16, []> var_21541_to_fp16 = const()[name = tensor<string, []>("op_21541_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1791_cast = mul(x = var_21540_cast, y = var_21541_to_fp16)[name = tensor<string, []>("aw_1791_cast")];
            tensor<string, []> var_21544_equation_0 = const()[name = tensor<string, []>("op_21544_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21544_cast = einsum(equation = var_21544_equation_0, values = (var_21386_cast, var_21303_cast))[name = tensor<string, []>("op_21544_cast")];
            tensor<fp16, []> var_21545_to_fp16 = const()[name = tensor<string, []>("op_21545_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1793_cast = mul(x = var_21544_cast, y = var_21545_to_fp16)[name = tensor<string, []>("aw_1793_cast")];
            tensor<string, []> var_21548_equation_0 = const()[name = tensor<string, []>("op_21548_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21548_cast = einsum(equation = var_21548_equation_0, values = (var_21390_cast, var_21307_cast))[name = tensor<string, []>("op_21548_cast")];
            tensor<fp16, []> var_21549_to_fp16 = const()[name = tensor<string, []>("op_21549_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1795_cast = mul(x = var_21548_cast, y = var_21549_to_fp16)[name = tensor<string, []>("aw_1795_cast")];
            tensor<string, []> var_21552_equation_0 = const()[name = tensor<string, []>("op_21552_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21552_cast = einsum(equation = var_21552_equation_0, values = (var_21394_cast, var_21311_cast))[name = tensor<string, []>("op_21552_cast")];
            tensor<fp16, []> var_21553_to_fp16 = const()[name = tensor<string, []>("op_21553_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1797_cast = mul(x = var_21552_cast, y = var_21553_to_fp16)[name = tensor<string, []>("aw_1797_cast")];
            tensor<string, []> var_21556_equation_0 = const()[name = tensor<string, []>("op_21556_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21556_cast = einsum(equation = var_21556_equation_0, values = (var_21398_cast, var_21315_cast))[name = tensor<string, []>("op_21556_cast")];
            tensor<fp16, []> var_21557_to_fp16 = const()[name = tensor<string, []>("op_21557_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1799_cast = mul(x = var_21556_cast, y = var_21557_to_fp16)[name = tensor<string, []>("aw_1799_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21559_cast = softmax(axis = var_21105, x = aw_1761_cast)[name = tensor<string, []>("op_21559_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21560_cast = softmax(axis = var_21105, x = aw_1763_cast)[name = tensor<string, []>("op_21560_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21561_cast = softmax(axis = var_21105, x = aw_1765_cast)[name = tensor<string, []>("op_21561_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21562_cast = softmax(axis = var_21105, x = aw_1767_cast)[name = tensor<string, []>("op_21562_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21563_cast = softmax(axis = var_21105, x = aw_1769_cast)[name = tensor<string, []>("op_21563_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21564_cast = softmax(axis = var_21105, x = aw_1771_cast)[name = tensor<string, []>("op_21564_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21565_cast = softmax(axis = var_21105, x = aw_1773_cast)[name = tensor<string, []>("op_21565_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21566_cast = softmax(axis = var_21105, x = aw_1775_cast)[name = tensor<string, []>("op_21566_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21567_cast = softmax(axis = var_21105, x = aw_1777_cast)[name = tensor<string, []>("op_21567_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21568_cast = softmax(axis = var_21105, x = aw_1779_cast)[name = tensor<string, []>("op_21568_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21569_cast = softmax(axis = var_21105, x = aw_1781_cast)[name = tensor<string, []>("op_21569_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21570_cast = softmax(axis = var_21105, x = aw_1783_cast)[name = tensor<string, []>("op_21570_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21571_cast = softmax(axis = var_21105, x = aw_1785_cast)[name = tensor<string, []>("op_21571_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21572_cast = softmax(axis = var_21105, x = aw_1787_cast)[name = tensor<string, []>("op_21572_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21573_cast = softmax(axis = var_21105, x = aw_1789_cast)[name = tensor<string, []>("op_21573_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21574_cast = softmax(axis = var_21105, x = aw_1791_cast)[name = tensor<string, []>("op_21574_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21575_cast = softmax(axis = var_21105, x = aw_1793_cast)[name = tensor<string, []>("op_21575_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21576_cast = softmax(axis = var_21105, x = aw_1795_cast)[name = tensor<string, []>("op_21576_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21577_cast = softmax(axis = var_21105, x = aw_1797_cast)[name = tensor<string, []>("op_21577_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_21578_cast = softmax(axis = var_21105, x = aw_1799_cast)[name = tensor<string, []>("op_21578_cast")];
            tensor<string, []> var_21580_equation_0 = const()[name = tensor<string, []>("op_21580_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21580_cast = einsum(equation = var_21580_equation_0, values = (var_21400_cast, var_21559_cast))[name = tensor<string, []>("op_21580_cast")];
            tensor<string, []> var_21582_equation_0 = const()[name = tensor<string, []>("op_21582_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21582_cast = einsum(equation = var_21582_equation_0, values = (var_21404_cast, var_21560_cast))[name = tensor<string, []>("op_21582_cast")];
            tensor<string, []> var_21584_equation_0 = const()[name = tensor<string, []>("op_21584_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21584_cast = einsum(equation = var_21584_equation_0, values = (var_21408_cast, var_21561_cast))[name = tensor<string, []>("op_21584_cast")];
            tensor<string, []> var_21586_equation_0 = const()[name = tensor<string, []>("op_21586_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21586_cast = einsum(equation = var_21586_equation_0, values = (var_21412_cast, var_21562_cast))[name = tensor<string, []>("op_21586_cast")];
            tensor<string, []> var_21588_equation_0 = const()[name = tensor<string, []>("op_21588_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21588_cast = einsum(equation = var_21588_equation_0, values = (var_21416_cast, var_21563_cast))[name = tensor<string, []>("op_21588_cast")];
            tensor<string, []> var_21590_equation_0 = const()[name = tensor<string, []>("op_21590_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21590_cast = einsum(equation = var_21590_equation_0, values = (var_21420_cast, var_21564_cast))[name = tensor<string, []>("op_21590_cast")];
            tensor<string, []> var_21592_equation_0 = const()[name = tensor<string, []>("op_21592_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21592_cast = einsum(equation = var_21592_equation_0, values = (var_21424_cast, var_21565_cast))[name = tensor<string, []>("op_21592_cast")];
            tensor<string, []> var_21594_equation_0 = const()[name = tensor<string, []>("op_21594_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21594_cast = einsum(equation = var_21594_equation_0, values = (var_21428_cast, var_21566_cast))[name = tensor<string, []>("op_21594_cast")];
            tensor<string, []> var_21596_equation_0 = const()[name = tensor<string, []>("op_21596_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21596_cast = einsum(equation = var_21596_equation_0, values = (var_21432_cast, var_21567_cast))[name = tensor<string, []>("op_21596_cast")];
            tensor<string, []> var_21598_equation_0 = const()[name = tensor<string, []>("op_21598_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21598_cast = einsum(equation = var_21598_equation_0, values = (var_21436_cast, var_21568_cast))[name = tensor<string, []>("op_21598_cast")];
            tensor<string, []> var_21600_equation_0 = const()[name = tensor<string, []>("op_21600_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21600_cast = einsum(equation = var_21600_equation_0, values = (var_21440_cast, var_21569_cast))[name = tensor<string, []>("op_21600_cast")];
            tensor<string, []> var_21602_equation_0 = const()[name = tensor<string, []>("op_21602_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21602_cast = einsum(equation = var_21602_equation_0, values = (var_21444_cast, var_21570_cast))[name = tensor<string, []>("op_21602_cast")];
            tensor<string, []> var_21604_equation_0 = const()[name = tensor<string, []>("op_21604_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21604_cast = einsum(equation = var_21604_equation_0, values = (var_21448_cast, var_21571_cast))[name = tensor<string, []>("op_21604_cast")];
            tensor<string, []> var_21606_equation_0 = const()[name = tensor<string, []>("op_21606_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21606_cast = einsum(equation = var_21606_equation_0, values = (var_21452_cast, var_21572_cast))[name = tensor<string, []>("op_21606_cast")];
            tensor<string, []> var_21608_equation_0 = const()[name = tensor<string, []>("op_21608_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21608_cast = einsum(equation = var_21608_equation_0, values = (var_21456_cast, var_21573_cast))[name = tensor<string, []>("op_21608_cast")];
            tensor<string, []> var_21610_equation_0 = const()[name = tensor<string, []>("op_21610_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21610_cast = einsum(equation = var_21610_equation_0, values = (var_21460_cast, var_21574_cast))[name = tensor<string, []>("op_21610_cast")];
            tensor<string, []> var_21612_equation_0 = const()[name = tensor<string, []>("op_21612_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21612_cast = einsum(equation = var_21612_equation_0, values = (var_21464_cast, var_21575_cast))[name = tensor<string, []>("op_21612_cast")];
            tensor<string, []> var_21614_equation_0 = const()[name = tensor<string, []>("op_21614_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21614_cast = einsum(equation = var_21614_equation_0, values = (var_21468_cast, var_21576_cast))[name = tensor<string, []>("op_21614_cast")];
            tensor<string, []> var_21616_equation_0 = const()[name = tensor<string, []>("op_21616_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21616_cast = einsum(equation = var_21616_equation_0, values = (var_21472_cast, var_21577_cast))[name = tensor<string, []>("op_21616_cast")];
            tensor<string, []> var_21618_equation_0 = const()[name = tensor<string, []>("op_21618_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_21618_cast = einsum(equation = var_21618_equation_0, values = (var_21476_cast, var_21578_cast))[name = tensor<string, []>("op_21618_cast")];
            tensor<bool, []> input_329_interleave_0 = const()[name = tensor<string, []>("input_329_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_329_cast = concat(axis = var_21105, interleave = input_329_interleave_0, values = (var_21580_cast, var_21582_cast, var_21584_cast, var_21586_cast, var_21588_cast, var_21590_cast, var_21592_cast, var_21594_cast, var_21596_cast, var_21598_cast, var_21600_cast, var_21602_cast, var_21604_cast, var_21606_cast, var_21608_cast, var_21610_cast, var_21612_cast, var_21614_cast, var_21616_cast, var_21618_cast))[name = tensor<string, []>("input_329_cast")];
            tensor<int32, [2]> var_21624 = const()[name = tensor<string, []>("op_21624"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_21626 = const()[name = tensor<string, []>("op_21626"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_21628_pad_type_0 = const()[name = tensor<string, []>("op_21628_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_21628_pad_0 = const()[name = tensor<string, []>("op_21628_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1735944896)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1739221760)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_21628_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_21626, groups = var_21105, pad = var_21628_pad_0, pad_type = var_21628_pad_type_0, strides = var_21624, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_329_cast)[name = tensor<string, []>("op_21628_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_147_cast = add(x = var_21628_cast, y = inputs_145_cast)[name = tensor<string, []>("inputs_147_cast")];
            tensor<int32, [1]> var_21632 = const()[name = tensor<string, []>("op_21632"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_147_cast = reduce_mean(axes = var_21632, keep_dims = var_21100, x = inputs_147_cast)[name = tensor<string, []>("channels_mean_147_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_147_cast = sub(x = inputs_147_cast, y = channels_mean_147_cast)[name = tensor<string, []>("zero_mean_147_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_147_cast = mul(x = zero_mean_147_cast, y = zero_mean_147_cast)[name = tensor<string, []>("zero_mean_sq_147_cast")];
            tensor<int32, [1]> var_21636 = const()[name = tensor<string, []>("op_21636"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_21637_cast = reduce_mean(axes = var_21636, keep_dims = var_21100, x = zero_mean_sq_147_cast)[name = tensor<string, []>("op_21637_cast")];
            tensor<fp16, []> var_21638_to_fp16 = const()[name = tensor<string, []>("op_21638_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_21639_cast = add(x = var_21637_cast, y = var_21638_to_fp16)[name = tensor<string, []>("op_21639_cast")];
            tensor<fp16, []> denom_147_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_147_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_147_cast = rsqrt(epsilon = denom_147_epsilon_0_to_fp16, x = var_21639_cast)[name = tensor<string, []>("denom_147_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_147_cast = mul(x = zero_mean_147_cast, y = denom_147_cast)[name = tensor<string, []>("out_147_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_21643_to_fp16 = const()[name = tensor<string, []>("op_21643_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1739224384)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_21644_cast = add(x = out_147_cast, y = var_21643_to_fp16)[name = tensor<string, []>("op_21644_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_21646_to_fp16 = const()[name = tensor<string, []>("op_21646_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1739227008)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_211_cast = mul(x = var_21644_cast, y = var_21646_to_fp16)[name = tensor<string, []>("hidden_states_211_cast")];
            tensor<int32, [2]> var_21653 = const()[name = tensor<string, []>("op_21653"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_21655 = const()[name = tensor<string, []>("op_21655"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_99_pad_type_0 = const()[name = tensor<string, []>("q_99_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_99_pad_0 = const()[name = tensor<string, []>("q_99_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1739229632)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_99_cast = conv(dilations = var_21655, groups = var_21105, pad = q_99_pad_0, pad_type = q_99_pad_type_0, strides = var_21653, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_211_cast)[name = tensor<string, []>("q_99_cast")];
            tensor<int32, [2]> var_21659 = const()[name = tensor<string, []>("op_21659"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_21661 = const()[name = tensor<string, []>("op_21661"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_197_pad_type_0 = const()[name = tensor<string, []>("k_197_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_197_pad_0 = const()[name = tensor<string, []>("k_197_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1742506496)))];
            tensor<fp16, [2, 1280, 1, 77]> k_197_cast = conv(dilations = var_21661, groups = var_21105, pad = k_197_pad_0, pad_type = k_197_pad_type_0, strides = var_21659, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_197_cast")];
            tensor<int32, [2]> var_21665 = const()[name = tensor<string, []>("op_21665"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_21667 = const()[name = tensor<string, []>("op_21667"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_99_pad_type_0 = const()[name = tensor<string, []>("v_99_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_99_pad_0 = const()[name = tensor<string, []>("v_99_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1747749440)))];
            tensor<fp16, [2, 1280, 1, 77]> v_99_cast = conv(dilations = var_21667, groups = var_21105, pad = v_99_pad_0, pad_type = v_99_pad_type_0, strides = var_21665, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_99_cast")];
            tensor<int32, [4]> var_21671_begin_0 = const()[name = tensor<string, []>("op_21671_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_21671_end_0 = const()[name = tensor<string, []>("op_21671_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_21671_end_mask_0 = const()[name = tensor<string, []>("op_21671_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21671_cast = slice_by_index(begin = var_21671_begin_0, end = var_21671_end_0, end_mask = var_21671_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21671_cast")];
            tensor<int32, [4]> var_21675_begin_0 = const()[name = tensor<string, []>("op_21675_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_21675_end_0 = const()[name = tensor<string, []>("op_21675_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_21675_end_mask_0 = const()[name = tensor<string, []>("op_21675_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21675_cast = slice_by_index(begin = var_21675_begin_0, end = var_21675_end_0, end_mask = var_21675_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21675_cast")];
            tensor<int32, [4]> var_21679_begin_0 = const()[name = tensor<string, []>("op_21679_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_21679_end_0 = const()[name = tensor<string, []>("op_21679_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_21679_end_mask_0 = const()[name = tensor<string, []>("op_21679_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21679_cast = slice_by_index(begin = var_21679_begin_0, end = var_21679_end_0, end_mask = var_21679_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21679_cast")];
            tensor<int32, [4]> var_21683_begin_0 = const()[name = tensor<string, []>("op_21683_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_21683_end_0 = const()[name = tensor<string, []>("op_21683_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_21683_end_mask_0 = const()[name = tensor<string, []>("op_21683_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21683_cast = slice_by_index(begin = var_21683_begin_0, end = var_21683_end_0, end_mask = var_21683_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21683_cast")];
            tensor<int32, [4]> var_21687_begin_0 = const()[name = tensor<string, []>("op_21687_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_21687_end_0 = const()[name = tensor<string, []>("op_21687_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_21687_end_mask_0 = const()[name = tensor<string, []>("op_21687_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21687_cast = slice_by_index(begin = var_21687_begin_0, end = var_21687_end_0, end_mask = var_21687_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21687_cast")];
            tensor<int32, [4]> var_21691_begin_0 = const()[name = tensor<string, []>("op_21691_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_21691_end_0 = const()[name = tensor<string, []>("op_21691_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_21691_end_mask_0 = const()[name = tensor<string, []>("op_21691_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21691_cast = slice_by_index(begin = var_21691_begin_0, end = var_21691_end_0, end_mask = var_21691_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21691_cast")];
            tensor<int32, [4]> var_21695_begin_0 = const()[name = tensor<string, []>("op_21695_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_21695_end_0 = const()[name = tensor<string, []>("op_21695_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_21695_end_mask_0 = const()[name = tensor<string, []>("op_21695_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21695_cast = slice_by_index(begin = var_21695_begin_0, end = var_21695_end_0, end_mask = var_21695_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21695_cast")];
            tensor<int32, [4]> var_21699_begin_0 = const()[name = tensor<string, []>("op_21699_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_21699_end_0 = const()[name = tensor<string, []>("op_21699_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_21699_end_mask_0 = const()[name = tensor<string, []>("op_21699_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21699_cast = slice_by_index(begin = var_21699_begin_0, end = var_21699_end_0, end_mask = var_21699_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21699_cast")];
            tensor<int32, [4]> var_21703_begin_0 = const()[name = tensor<string, []>("op_21703_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_21703_end_0 = const()[name = tensor<string, []>("op_21703_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_21703_end_mask_0 = const()[name = tensor<string, []>("op_21703_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21703_cast = slice_by_index(begin = var_21703_begin_0, end = var_21703_end_0, end_mask = var_21703_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21703_cast")];
            tensor<int32, [4]> var_21707_begin_0 = const()[name = tensor<string, []>("op_21707_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_21707_end_0 = const()[name = tensor<string, []>("op_21707_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_21707_end_mask_0 = const()[name = tensor<string, []>("op_21707_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21707_cast = slice_by_index(begin = var_21707_begin_0, end = var_21707_end_0, end_mask = var_21707_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21707_cast")];
            tensor<int32, [4]> var_21711_begin_0 = const()[name = tensor<string, []>("op_21711_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_21711_end_0 = const()[name = tensor<string, []>("op_21711_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_21711_end_mask_0 = const()[name = tensor<string, []>("op_21711_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21711_cast = slice_by_index(begin = var_21711_begin_0, end = var_21711_end_0, end_mask = var_21711_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21711_cast")];
            tensor<int32, [4]> var_21715_begin_0 = const()[name = tensor<string, []>("op_21715_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_21715_end_0 = const()[name = tensor<string, []>("op_21715_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_21715_end_mask_0 = const()[name = tensor<string, []>("op_21715_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21715_cast = slice_by_index(begin = var_21715_begin_0, end = var_21715_end_0, end_mask = var_21715_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21715_cast")];
            tensor<int32, [4]> var_21719_begin_0 = const()[name = tensor<string, []>("op_21719_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_21719_end_0 = const()[name = tensor<string, []>("op_21719_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_21719_end_mask_0 = const()[name = tensor<string, []>("op_21719_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21719_cast = slice_by_index(begin = var_21719_begin_0, end = var_21719_end_0, end_mask = var_21719_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21719_cast")];
            tensor<int32, [4]> var_21723_begin_0 = const()[name = tensor<string, []>("op_21723_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_21723_end_0 = const()[name = tensor<string, []>("op_21723_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_21723_end_mask_0 = const()[name = tensor<string, []>("op_21723_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21723_cast = slice_by_index(begin = var_21723_begin_0, end = var_21723_end_0, end_mask = var_21723_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21723_cast")];
            tensor<int32, [4]> var_21727_begin_0 = const()[name = tensor<string, []>("op_21727_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_21727_end_0 = const()[name = tensor<string, []>("op_21727_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_21727_end_mask_0 = const()[name = tensor<string, []>("op_21727_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21727_cast = slice_by_index(begin = var_21727_begin_0, end = var_21727_end_0, end_mask = var_21727_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21727_cast")];
            tensor<int32, [4]> var_21731_begin_0 = const()[name = tensor<string, []>("op_21731_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_21731_end_0 = const()[name = tensor<string, []>("op_21731_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_21731_end_mask_0 = const()[name = tensor<string, []>("op_21731_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21731_cast = slice_by_index(begin = var_21731_begin_0, end = var_21731_end_0, end_mask = var_21731_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21731_cast")];
            tensor<int32, [4]> var_21735_begin_0 = const()[name = tensor<string, []>("op_21735_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_21735_end_0 = const()[name = tensor<string, []>("op_21735_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_21735_end_mask_0 = const()[name = tensor<string, []>("op_21735_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21735_cast = slice_by_index(begin = var_21735_begin_0, end = var_21735_end_0, end_mask = var_21735_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21735_cast")];
            tensor<int32, [4]> var_21739_begin_0 = const()[name = tensor<string, []>("op_21739_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_21739_end_0 = const()[name = tensor<string, []>("op_21739_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_21739_end_mask_0 = const()[name = tensor<string, []>("op_21739_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21739_cast = slice_by_index(begin = var_21739_begin_0, end = var_21739_end_0, end_mask = var_21739_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21739_cast")];
            tensor<int32, [4]> var_21743_begin_0 = const()[name = tensor<string, []>("op_21743_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_21743_end_0 = const()[name = tensor<string, []>("op_21743_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_21743_end_mask_0 = const()[name = tensor<string, []>("op_21743_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21743_cast = slice_by_index(begin = var_21743_begin_0, end = var_21743_end_0, end_mask = var_21743_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21743_cast")];
            tensor<int32, [4]> var_21747_begin_0 = const()[name = tensor<string, []>("op_21747_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_21747_end_0 = const()[name = tensor<string, []>("op_21747_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_21747_end_mask_0 = const()[name = tensor<string, []>("op_21747_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_21747_cast = slice_by_index(begin = var_21747_begin_0, end = var_21747_end_0, end_mask = var_21747_end_mask_0, x = q_99_cast)[name = tensor<string, []>("op_21747_cast")];
            tensor<int32, [4]> k_199_perm_0 = const()[name = tensor<string, []>("k_199_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_21754_begin_0 = const()[name = tensor<string, []>("op_21754_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_21754_end_0 = const()[name = tensor<string, []>("op_21754_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_21754_end_mask_0 = const()[name = tensor<string, []>("op_21754_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_90 = transpose(perm = k_199_perm_0, x = k_197_cast)[name = tensor<string, []>("transpose_90")];
            tensor<fp16, [2, 77, 1, 64]> var_21754_cast = slice_by_index(begin = var_21754_begin_0, end = var_21754_end_0, end_mask = var_21754_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21754_cast")];
            tensor<int32, [4]> var_21758_begin_0 = const()[name = tensor<string, []>("op_21758_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_21758_end_0 = const()[name = tensor<string, []>("op_21758_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_21758_end_mask_0 = const()[name = tensor<string, []>("op_21758_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_21758_cast = slice_by_index(begin = var_21758_begin_0, end = var_21758_end_0, end_mask = var_21758_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21758_cast")];
            tensor<int32, [4]> var_21762_begin_0 = const()[name = tensor<string, []>("op_21762_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_21762_end_0 = const()[name = tensor<string, []>("op_21762_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_21762_end_mask_0 = const()[name = tensor<string, []>("op_21762_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_21762_cast = slice_by_index(begin = var_21762_begin_0, end = var_21762_end_0, end_mask = var_21762_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21762_cast")];
            tensor<int32, [4]> var_21766_begin_0 = const()[name = tensor<string, []>("op_21766_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_21766_end_0 = const()[name = tensor<string, []>("op_21766_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_21766_end_mask_0 = const()[name = tensor<string, []>("op_21766_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_21766_cast = slice_by_index(begin = var_21766_begin_0, end = var_21766_end_0, end_mask = var_21766_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21766_cast")];
            tensor<int32, [4]> var_21770_begin_0 = const()[name = tensor<string, []>("op_21770_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_21770_end_0 = const()[name = tensor<string, []>("op_21770_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_21770_end_mask_0 = const()[name = tensor<string, []>("op_21770_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_21770_cast = slice_by_index(begin = var_21770_begin_0, end = var_21770_end_0, end_mask = var_21770_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21770_cast")];
            tensor<int32, [4]> var_21774_begin_0 = const()[name = tensor<string, []>("op_21774_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_21774_end_0 = const()[name = tensor<string, []>("op_21774_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_21774_end_mask_0 = const()[name = tensor<string, []>("op_21774_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_21774_cast = slice_by_index(begin = var_21774_begin_0, end = var_21774_end_0, end_mask = var_21774_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21774_cast")];
            tensor<int32, [4]> var_21778_begin_0 = const()[name = tensor<string, []>("op_21778_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_21778_end_0 = const()[name = tensor<string, []>("op_21778_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_21778_end_mask_0 = const()[name = tensor<string, []>("op_21778_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_21778_cast = slice_by_index(begin = var_21778_begin_0, end = var_21778_end_0, end_mask = var_21778_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21778_cast")];
            tensor<int32, [4]> var_21782_begin_0 = const()[name = tensor<string, []>("op_21782_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_21782_end_0 = const()[name = tensor<string, []>("op_21782_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_21782_end_mask_0 = const()[name = tensor<string, []>("op_21782_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_21782_cast = slice_by_index(begin = var_21782_begin_0, end = var_21782_end_0, end_mask = var_21782_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21782_cast")];
            tensor<int32, [4]> var_21786_begin_0 = const()[name = tensor<string, []>("op_21786_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_21786_end_0 = const()[name = tensor<string, []>("op_21786_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_21786_end_mask_0 = const()[name = tensor<string, []>("op_21786_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_21786_cast = slice_by_index(begin = var_21786_begin_0, end = var_21786_end_0, end_mask = var_21786_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21786_cast")];
            tensor<int32, [4]> var_21790_begin_0 = const()[name = tensor<string, []>("op_21790_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_21790_end_0 = const()[name = tensor<string, []>("op_21790_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_21790_end_mask_0 = const()[name = tensor<string, []>("op_21790_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_21790_cast = slice_by_index(begin = var_21790_begin_0, end = var_21790_end_0, end_mask = var_21790_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21790_cast")];
            tensor<int32, [4]> var_21794_begin_0 = const()[name = tensor<string, []>("op_21794_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_21794_end_0 = const()[name = tensor<string, []>("op_21794_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_21794_end_mask_0 = const()[name = tensor<string, []>("op_21794_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_21794_cast = slice_by_index(begin = var_21794_begin_0, end = var_21794_end_0, end_mask = var_21794_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21794_cast")];
            tensor<int32, [4]> var_21798_begin_0 = const()[name = tensor<string, []>("op_21798_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_21798_end_0 = const()[name = tensor<string, []>("op_21798_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_21798_end_mask_0 = const()[name = tensor<string, []>("op_21798_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_21798_cast = slice_by_index(begin = var_21798_begin_0, end = var_21798_end_0, end_mask = var_21798_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21798_cast")];
            tensor<int32, [4]> var_21802_begin_0 = const()[name = tensor<string, []>("op_21802_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_21802_end_0 = const()[name = tensor<string, []>("op_21802_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_21802_end_mask_0 = const()[name = tensor<string, []>("op_21802_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_21802_cast = slice_by_index(begin = var_21802_begin_0, end = var_21802_end_0, end_mask = var_21802_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21802_cast")];
            tensor<int32, [4]> var_21806_begin_0 = const()[name = tensor<string, []>("op_21806_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_21806_end_0 = const()[name = tensor<string, []>("op_21806_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_21806_end_mask_0 = const()[name = tensor<string, []>("op_21806_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_21806_cast = slice_by_index(begin = var_21806_begin_0, end = var_21806_end_0, end_mask = var_21806_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21806_cast")];
            tensor<int32, [4]> var_21810_begin_0 = const()[name = tensor<string, []>("op_21810_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_21810_end_0 = const()[name = tensor<string, []>("op_21810_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_21810_end_mask_0 = const()[name = tensor<string, []>("op_21810_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_21810_cast = slice_by_index(begin = var_21810_begin_0, end = var_21810_end_0, end_mask = var_21810_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21810_cast")];
            tensor<int32, [4]> var_21814_begin_0 = const()[name = tensor<string, []>("op_21814_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_21814_end_0 = const()[name = tensor<string, []>("op_21814_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_21814_end_mask_0 = const()[name = tensor<string, []>("op_21814_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_21814_cast = slice_by_index(begin = var_21814_begin_0, end = var_21814_end_0, end_mask = var_21814_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21814_cast")];
            tensor<int32, [4]> var_21818_begin_0 = const()[name = tensor<string, []>("op_21818_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_21818_end_0 = const()[name = tensor<string, []>("op_21818_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_21818_end_mask_0 = const()[name = tensor<string, []>("op_21818_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_21818_cast = slice_by_index(begin = var_21818_begin_0, end = var_21818_end_0, end_mask = var_21818_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21818_cast")];
            tensor<int32, [4]> var_21822_begin_0 = const()[name = tensor<string, []>("op_21822_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_21822_end_0 = const()[name = tensor<string, []>("op_21822_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_21822_end_mask_0 = const()[name = tensor<string, []>("op_21822_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_21822_cast = slice_by_index(begin = var_21822_begin_0, end = var_21822_end_0, end_mask = var_21822_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21822_cast")];
            tensor<int32, [4]> var_21826_begin_0 = const()[name = tensor<string, []>("op_21826_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_21826_end_0 = const()[name = tensor<string, []>("op_21826_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_21826_end_mask_0 = const()[name = tensor<string, []>("op_21826_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_21826_cast = slice_by_index(begin = var_21826_begin_0, end = var_21826_end_0, end_mask = var_21826_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21826_cast")];
            tensor<int32, [4]> var_21830_begin_0 = const()[name = tensor<string, []>("op_21830_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_21830_end_0 = const()[name = tensor<string, []>("op_21830_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_21830_end_mask_0 = const()[name = tensor<string, []>("op_21830_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_21830_cast = slice_by_index(begin = var_21830_begin_0, end = var_21830_end_0, end_mask = var_21830_end_mask_0, x = transpose_90)[name = tensor<string, []>("op_21830_cast")];
            tensor<int32, [4]> var_21832_begin_0 = const()[name = tensor<string, []>("op_21832_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_21832_end_0 = const()[name = tensor<string, []>("op_21832_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_21832_end_mask_0 = const()[name = tensor<string, []>("op_21832_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21832_cast = slice_by_index(begin = var_21832_begin_0, end = var_21832_end_0, end_mask = var_21832_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21832_cast")];
            tensor<int32, [4]> var_21836_begin_0 = const()[name = tensor<string, []>("op_21836_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_21836_end_0 = const()[name = tensor<string, []>("op_21836_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_21836_end_mask_0 = const()[name = tensor<string, []>("op_21836_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21836_cast = slice_by_index(begin = var_21836_begin_0, end = var_21836_end_0, end_mask = var_21836_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21836_cast")];
            tensor<int32, [4]> var_21840_begin_0 = const()[name = tensor<string, []>("op_21840_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_21840_end_0 = const()[name = tensor<string, []>("op_21840_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_21840_end_mask_0 = const()[name = tensor<string, []>("op_21840_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21840_cast = slice_by_index(begin = var_21840_begin_0, end = var_21840_end_0, end_mask = var_21840_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21840_cast")];
            tensor<int32, [4]> var_21844_begin_0 = const()[name = tensor<string, []>("op_21844_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_21844_end_0 = const()[name = tensor<string, []>("op_21844_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_21844_end_mask_0 = const()[name = tensor<string, []>("op_21844_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21844_cast = slice_by_index(begin = var_21844_begin_0, end = var_21844_end_0, end_mask = var_21844_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21844_cast")];
            tensor<int32, [4]> var_21848_begin_0 = const()[name = tensor<string, []>("op_21848_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_21848_end_0 = const()[name = tensor<string, []>("op_21848_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_21848_end_mask_0 = const()[name = tensor<string, []>("op_21848_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21848_cast = slice_by_index(begin = var_21848_begin_0, end = var_21848_end_0, end_mask = var_21848_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21848_cast")];
            tensor<int32, [4]> var_21852_begin_0 = const()[name = tensor<string, []>("op_21852_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_21852_end_0 = const()[name = tensor<string, []>("op_21852_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_21852_end_mask_0 = const()[name = tensor<string, []>("op_21852_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21852_cast = slice_by_index(begin = var_21852_begin_0, end = var_21852_end_0, end_mask = var_21852_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21852_cast")];
            tensor<int32, [4]> var_21856_begin_0 = const()[name = tensor<string, []>("op_21856_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_21856_end_0 = const()[name = tensor<string, []>("op_21856_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_21856_end_mask_0 = const()[name = tensor<string, []>("op_21856_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21856_cast = slice_by_index(begin = var_21856_begin_0, end = var_21856_end_0, end_mask = var_21856_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21856_cast")];
            tensor<int32, [4]> var_21860_begin_0 = const()[name = tensor<string, []>("op_21860_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_21860_end_0 = const()[name = tensor<string, []>("op_21860_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_21860_end_mask_0 = const()[name = tensor<string, []>("op_21860_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21860_cast = slice_by_index(begin = var_21860_begin_0, end = var_21860_end_0, end_mask = var_21860_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21860_cast")];
            tensor<int32, [4]> var_21864_begin_0 = const()[name = tensor<string, []>("op_21864_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_21864_end_0 = const()[name = tensor<string, []>("op_21864_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_21864_end_mask_0 = const()[name = tensor<string, []>("op_21864_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21864_cast = slice_by_index(begin = var_21864_begin_0, end = var_21864_end_0, end_mask = var_21864_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21864_cast")];
            tensor<int32, [4]> var_21868_begin_0 = const()[name = tensor<string, []>("op_21868_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_21868_end_0 = const()[name = tensor<string, []>("op_21868_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_21868_end_mask_0 = const()[name = tensor<string, []>("op_21868_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21868_cast = slice_by_index(begin = var_21868_begin_0, end = var_21868_end_0, end_mask = var_21868_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21868_cast")];
            tensor<int32, [4]> var_21872_begin_0 = const()[name = tensor<string, []>("op_21872_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_21872_end_0 = const()[name = tensor<string, []>("op_21872_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_21872_end_mask_0 = const()[name = tensor<string, []>("op_21872_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21872_cast = slice_by_index(begin = var_21872_begin_0, end = var_21872_end_0, end_mask = var_21872_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21872_cast")];
            tensor<int32, [4]> var_21876_begin_0 = const()[name = tensor<string, []>("op_21876_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_21876_end_0 = const()[name = tensor<string, []>("op_21876_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_21876_end_mask_0 = const()[name = tensor<string, []>("op_21876_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21876_cast = slice_by_index(begin = var_21876_begin_0, end = var_21876_end_0, end_mask = var_21876_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21876_cast")];
            tensor<int32, [4]> var_21880_begin_0 = const()[name = tensor<string, []>("op_21880_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_21880_end_0 = const()[name = tensor<string, []>("op_21880_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_21880_end_mask_0 = const()[name = tensor<string, []>("op_21880_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21880_cast = slice_by_index(begin = var_21880_begin_0, end = var_21880_end_0, end_mask = var_21880_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21880_cast")];
            tensor<int32, [4]> var_21884_begin_0 = const()[name = tensor<string, []>("op_21884_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_21884_end_0 = const()[name = tensor<string, []>("op_21884_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_21884_end_mask_0 = const()[name = tensor<string, []>("op_21884_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21884_cast = slice_by_index(begin = var_21884_begin_0, end = var_21884_end_0, end_mask = var_21884_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21884_cast")];
            tensor<int32, [4]> var_21888_begin_0 = const()[name = tensor<string, []>("op_21888_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_21888_end_0 = const()[name = tensor<string, []>("op_21888_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_21888_end_mask_0 = const()[name = tensor<string, []>("op_21888_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21888_cast = slice_by_index(begin = var_21888_begin_0, end = var_21888_end_0, end_mask = var_21888_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21888_cast")];
            tensor<int32, [4]> var_21892_begin_0 = const()[name = tensor<string, []>("op_21892_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_21892_end_0 = const()[name = tensor<string, []>("op_21892_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_21892_end_mask_0 = const()[name = tensor<string, []>("op_21892_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21892_cast = slice_by_index(begin = var_21892_begin_0, end = var_21892_end_0, end_mask = var_21892_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21892_cast")];
            tensor<int32, [4]> var_21896_begin_0 = const()[name = tensor<string, []>("op_21896_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_21896_end_0 = const()[name = tensor<string, []>("op_21896_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_21896_end_mask_0 = const()[name = tensor<string, []>("op_21896_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21896_cast = slice_by_index(begin = var_21896_begin_0, end = var_21896_end_0, end_mask = var_21896_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21896_cast")];
            tensor<int32, [4]> var_21900_begin_0 = const()[name = tensor<string, []>("op_21900_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_21900_end_0 = const()[name = tensor<string, []>("op_21900_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_21900_end_mask_0 = const()[name = tensor<string, []>("op_21900_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21900_cast = slice_by_index(begin = var_21900_begin_0, end = var_21900_end_0, end_mask = var_21900_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21900_cast")];
            tensor<int32, [4]> var_21904_begin_0 = const()[name = tensor<string, []>("op_21904_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_21904_end_0 = const()[name = tensor<string, []>("op_21904_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_21904_end_mask_0 = const()[name = tensor<string, []>("op_21904_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21904_cast = slice_by_index(begin = var_21904_begin_0, end = var_21904_end_0, end_mask = var_21904_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21904_cast")];
            tensor<int32, [4]> var_21908_begin_0 = const()[name = tensor<string, []>("op_21908_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_21908_end_0 = const()[name = tensor<string, []>("op_21908_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_21908_end_mask_0 = const()[name = tensor<string, []>("op_21908_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_21908_cast = slice_by_index(begin = var_21908_begin_0, end = var_21908_end_0, end_mask = var_21908_end_mask_0, x = v_99_cast)[name = tensor<string, []>("op_21908_cast")];
            tensor<string, []> var_21912_equation_0 = const()[name = tensor<string, []>("op_21912_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21912_cast = einsum(equation = var_21912_equation_0, values = (var_21754_cast, var_21671_cast))[name = tensor<string, []>("op_21912_cast")];
            tensor<fp16, []> var_21913_to_fp16 = const()[name = tensor<string, []>("op_21913_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1801_cast = mul(x = var_21912_cast, y = var_21913_to_fp16)[name = tensor<string, []>("aw_1801_cast")];
            tensor<string, []> var_21916_equation_0 = const()[name = tensor<string, []>("op_21916_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21916_cast = einsum(equation = var_21916_equation_0, values = (var_21758_cast, var_21675_cast))[name = tensor<string, []>("op_21916_cast")];
            tensor<fp16, []> var_21917_to_fp16 = const()[name = tensor<string, []>("op_21917_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1803_cast = mul(x = var_21916_cast, y = var_21917_to_fp16)[name = tensor<string, []>("aw_1803_cast")];
            tensor<string, []> var_21920_equation_0 = const()[name = tensor<string, []>("op_21920_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21920_cast = einsum(equation = var_21920_equation_0, values = (var_21762_cast, var_21679_cast))[name = tensor<string, []>("op_21920_cast")];
            tensor<fp16, []> var_21921_to_fp16 = const()[name = tensor<string, []>("op_21921_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1805_cast = mul(x = var_21920_cast, y = var_21921_to_fp16)[name = tensor<string, []>("aw_1805_cast")];
            tensor<string, []> var_21924_equation_0 = const()[name = tensor<string, []>("op_21924_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21924_cast = einsum(equation = var_21924_equation_0, values = (var_21766_cast, var_21683_cast))[name = tensor<string, []>("op_21924_cast")];
            tensor<fp16, []> var_21925_to_fp16 = const()[name = tensor<string, []>("op_21925_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1807_cast = mul(x = var_21924_cast, y = var_21925_to_fp16)[name = tensor<string, []>("aw_1807_cast")];
            tensor<string, []> var_21928_equation_0 = const()[name = tensor<string, []>("op_21928_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21928_cast = einsum(equation = var_21928_equation_0, values = (var_21770_cast, var_21687_cast))[name = tensor<string, []>("op_21928_cast")];
            tensor<fp16, []> var_21929_to_fp16 = const()[name = tensor<string, []>("op_21929_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1809_cast = mul(x = var_21928_cast, y = var_21929_to_fp16)[name = tensor<string, []>("aw_1809_cast")];
            tensor<string, []> var_21932_equation_0 = const()[name = tensor<string, []>("op_21932_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21932_cast = einsum(equation = var_21932_equation_0, values = (var_21774_cast, var_21691_cast))[name = tensor<string, []>("op_21932_cast")];
            tensor<fp16, []> var_21933_to_fp16 = const()[name = tensor<string, []>("op_21933_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1811_cast = mul(x = var_21932_cast, y = var_21933_to_fp16)[name = tensor<string, []>("aw_1811_cast")];
            tensor<string, []> var_21936_equation_0 = const()[name = tensor<string, []>("op_21936_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21936_cast = einsum(equation = var_21936_equation_0, values = (var_21778_cast, var_21695_cast))[name = tensor<string, []>("op_21936_cast")];
            tensor<fp16, []> var_21937_to_fp16 = const()[name = tensor<string, []>("op_21937_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1813_cast = mul(x = var_21936_cast, y = var_21937_to_fp16)[name = tensor<string, []>("aw_1813_cast")];
            tensor<string, []> var_21940_equation_0 = const()[name = tensor<string, []>("op_21940_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21940_cast = einsum(equation = var_21940_equation_0, values = (var_21782_cast, var_21699_cast))[name = tensor<string, []>("op_21940_cast")];
            tensor<fp16, []> var_21941_to_fp16 = const()[name = tensor<string, []>("op_21941_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1815_cast = mul(x = var_21940_cast, y = var_21941_to_fp16)[name = tensor<string, []>("aw_1815_cast")];
            tensor<string, []> var_21944_equation_0 = const()[name = tensor<string, []>("op_21944_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21944_cast = einsum(equation = var_21944_equation_0, values = (var_21786_cast, var_21703_cast))[name = tensor<string, []>("op_21944_cast")];
            tensor<fp16, []> var_21945_to_fp16 = const()[name = tensor<string, []>("op_21945_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1817_cast = mul(x = var_21944_cast, y = var_21945_to_fp16)[name = tensor<string, []>("aw_1817_cast")];
            tensor<string, []> var_21948_equation_0 = const()[name = tensor<string, []>("op_21948_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21948_cast = einsum(equation = var_21948_equation_0, values = (var_21790_cast, var_21707_cast))[name = tensor<string, []>("op_21948_cast")];
            tensor<fp16, []> var_21949_to_fp16 = const()[name = tensor<string, []>("op_21949_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1819_cast = mul(x = var_21948_cast, y = var_21949_to_fp16)[name = tensor<string, []>("aw_1819_cast")];
            tensor<string, []> var_21952_equation_0 = const()[name = tensor<string, []>("op_21952_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21952_cast = einsum(equation = var_21952_equation_0, values = (var_21794_cast, var_21711_cast))[name = tensor<string, []>("op_21952_cast")];
            tensor<fp16, []> var_21953_to_fp16 = const()[name = tensor<string, []>("op_21953_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1821_cast = mul(x = var_21952_cast, y = var_21953_to_fp16)[name = tensor<string, []>("aw_1821_cast")];
            tensor<string, []> var_21956_equation_0 = const()[name = tensor<string, []>("op_21956_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21956_cast = einsum(equation = var_21956_equation_0, values = (var_21798_cast, var_21715_cast))[name = tensor<string, []>("op_21956_cast")];
            tensor<fp16, []> var_21957_to_fp16 = const()[name = tensor<string, []>("op_21957_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1823_cast = mul(x = var_21956_cast, y = var_21957_to_fp16)[name = tensor<string, []>("aw_1823_cast")];
            tensor<string, []> var_21960_equation_0 = const()[name = tensor<string, []>("op_21960_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21960_cast = einsum(equation = var_21960_equation_0, values = (var_21802_cast, var_21719_cast))[name = tensor<string, []>("op_21960_cast")];
            tensor<fp16, []> var_21961_to_fp16 = const()[name = tensor<string, []>("op_21961_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1825_cast = mul(x = var_21960_cast, y = var_21961_to_fp16)[name = tensor<string, []>("aw_1825_cast")];
            tensor<string, []> var_21964_equation_0 = const()[name = tensor<string, []>("op_21964_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21964_cast = einsum(equation = var_21964_equation_0, values = (var_21806_cast, var_21723_cast))[name = tensor<string, []>("op_21964_cast")];
            tensor<fp16, []> var_21965_to_fp16 = const()[name = tensor<string, []>("op_21965_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1827_cast = mul(x = var_21964_cast, y = var_21965_to_fp16)[name = tensor<string, []>("aw_1827_cast")];
            tensor<string, []> var_21968_equation_0 = const()[name = tensor<string, []>("op_21968_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21968_cast = einsum(equation = var_21968_equation_0, values = (var_21810_cast, var_21727_cast))[name = tensor<string, []>("op_21968_cast")];
            tensor<fp16, []> var_21969_to_fp16 = const()[name = tensor<string, []>("op_21969_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1829_cast = mul(x = var_21968_cast, y = var_21969_to_fp16)[name = tensor<string, []>("aw_1829_cast")];
            tensor<string, []> var_21972_equation_0 = const()[name = tensor<string, []>("op_21972_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21972_cast = einsum(equation = var_21972_equation_0, values = (var_21814_cast, var_21731_cast))[name = tensor<string, []>("op_21972_cast")];
            tensor<fp16, []> var_21973_to_fp16 = const()[name = tensor<string, []>("op_21973_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1831_cast = mul(x = var_21972_cast, y = var_21973_to_fp16)[name = tensor<string, []>("aw_1831_cast")];
            tensor<string, []> var_21976_equation_0 = const()[name = tensor<string, []>("op_21976_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21976_cast = einsum(equation = var_21976_equation_0, values = (var_21818_cast, var_21735_cast))[name = tensor<string, []>("op_21976_cast")];
            tensor<fp16, []> var_21977_to_fp16 = const()[name = tensor<string, []>("op_21977_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1833_cast = mul(x = var_21976_cast, y = var_21977_to_fp16)[name = tensor<string, []>("aw_1833_cast")];
            tensor<string, []> var_21980_equation_0 = const()[name = tensor<string, []>("op_21980_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21980_cast = einsum(equation = var_21980_equation_0, values = (var_21822_cast, var_21739_cast))[name = tensor<string, []>("op_21980_cast")];
            tensor<fp16, []> var_21981_to_fp16 = const()[name = tensor<string, []>("op_21981_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1835_cast = mul(x = var_21980_cast, y = var_21981_to_fp16)[name = tensor<string, []>("aw_1835_cast")];
            tensor<string, []> var_21984_equation_0 = const()[name = tensor<string, []>("op_21984_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21984_cast = einsum(equation = var_21984_equation_0, values = (var_21826_cast, var_21743_cast))[name = tensor<string, []>("op_21984_cast")];
            tensor<fp16, []> var_21985_to_fp16 = const()[name = tensor<string, []>("op_21985_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1837_cast = mul(x = var_21984_cast, y = var_21985_to_fp16)[name = tensor<string, []>("aw_1837_cast")];
            tensor<string, []> var_21988_equation_0 = const()[name = tensor<string, []>("op_21988_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_21988_cast = einsum(equation = var_21988_equation_0, values = (var_21830_cast, var_21747_cast))[name = tensor<string, []>("op_21988_cast")];
            tensor<fp16, []> var_21989_to_fp16 = const()[name = tensor<string, []>("op_21989_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1839_cast = mul(x = var_21988_cast, y = var_21989_to_fp16)[name = tensor<string, []>("aw_1839_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_21991_cast = softmax(axis = var_21105, x = aw_1801_cast)[name = tensor<string, []>("op_21991_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_21992_cast = softmax(axis = var_21105, x = aw_1803_cast)[name = tensor<string, []>("op_21992_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_21993_cast = softmax(axis = var_21105, x = aw_1805_cast)[name = tensor<string, []>("op_21993_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_21994_cast = softmax(axis = var_21105, x = aw_1807_cast)[name = tensor<string, []>("op_21994_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_21995_cast = softmax(axis = var_21105, x = aw_1809_cast)[name = tensor<string, []>("op_21995_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_21996_cast = softmax(axis = var_21105, x = aw_1811_cast)[name = tensor<string, []>("op_21996_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_21997_cast = softmax(axis = var_21105, x = aw_1813_cast)[name = tensor<string, []>("op_21997_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_21998_cast = softmax(axis = var_21105, x = aw_1815_cast)[name = tensor<string, []>("op_21998_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_21999_cast = softmax(axis = var_21105, x = aw_1817_cast)[name = tensor<string, []>("op_21999_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22000_cast = softmax(axis = var_21105, x = aw_1819_cast)[name = tensor<string, []>("op_22000_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22001_cast = softmax(axis = var_21105, x = aw_1821_cast)[name = tensor<string, []>("op_22001_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22002_cast = softmax(axis = var_21105, x = aw_1823_cast)[name = tensor<string, []>("op_22002_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22003_cast = softmax(axis = var_21105, x = aw_1825_cast)[name = tensor<string, []>("op_22003_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22004_cast = softmax(axis = var_21105, x = aw_1827_cast)[name = tensor<string, []>("op_22004_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22005_cast = softmax(axis = var_21105, x = aw_1829_cast)[name = tensor<string, []>("op_22005_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22006_cast = softmax(axis = var_21105, x = aw_1831_cast)[name = tensor<string, []>("op_22006_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22007_cast = softmax(axis = var_21105, x = aw_1833_cast)[name = tensor<string, []>("op_22007_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22008_cast = softmax(axis = var_21105, x = aw_1835_cast)[name = tensor<string, []>("op_22008_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22009_cast = softmax(axis = var_21105, x = aw_1837_cast)[name = tensor<string, []>("op_22009_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22010_cast = softmax(axis = var_21105, x = aw_1839_cast)[name = tensor<string, []>("op_22010_cast")];
            tensor<string, []> var_22012_equation_0 = const()[name = tensor<string, []>("op_22012_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22012_cast = einsum(equation = var_22012_equation_0, values = (var_21832_cast, var_21991_cast))[name = tensor<string, []>("op_22012_cast")];
            tensor<string, []> var_22014_equation_0 = const()[name = tensor<string, []>("op_22014_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22014_cast = einsum(equation = var_22014_equation_0, values = (var_21836_cast, var_21992_cast))[name = tensor<string, []>("op_22014_cast")];
            tensor<string, []> var_22016_equation_0 = const()[name = tensor<string, []>("op_22016_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22016_cast = einsum(equation = var_22016_equation_0, values = (var_21840_cast, var_21993_cast))[name = tensor<string, []>("op_22016_cast")];
            tensor<string, []> var_22018_equation_0 = const()[name = tensor<string, []>("op_22018_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22018_cast = einsum(equation = var_22018_equation_0, values = (var_21844_cast, var_21994_cast))[name = tensor<string, []>("op_22018_cast")];
            tensor<string, []> var_22020_equation_0 = const()[name = tensor<string, []>("op_22020_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22020_cast = einsum(equation = var_22020_equation_0, values = (var_21848_cast, var_21995_cast))[name = tensor<string, []>("op_22020_cast")];
            tensor<string, []> var_22022_equation_0 = const()[name = tensor<string, []>("op_22022_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22022_cast = einsum(equation = var_22022_equation_0, values = (var_21852_cast, var_21996_cast))[name = tensor<string, []>("op_22022_cast")];
            tensor<string, []> var_22024_equation_0 = const()[name = tensor<string, []>("op_22024_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22024_cast = einsum(equation = var_22024_equation_0, values = (var_21856_cast, var_21997_cast))[name = tensor<string, []>("op_22024_cast")];
            tensor<string, []> var_22026_equation_0 = const()[name = tensor<string, []>("op_22026_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22026_cast = einsum(equation = var_22026_equation_0, values = (var_21860_cast, var_21998_cast))[name = tensor<string, []>("op_22026_cast")];
            tensor<string, []> var_22028_equation_0 = const()[name = tensor<string, []>("op_22028_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22028_cast = einsum(equation = var_22028_equation_0, values = (var_21864_cast, var_21999_cast))[name = tensor<string, []>("op_22028_cast")];
            tensor<string, []> var_22030_equation_0 = const()[name = tensor<string, []>("op_22030_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22030_cast = einsum(equation = var_22030_equation_0, values = (var_21868_cast, var_22000_cast))[name = tensor<string, []>("op_22030_cast")];
            tensor<string, []> var_22032_equation_0 = const()[name = tensor<string, []>("op_22032_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22032_cast = einsum(equation = var_22032_equation_0, values = (var_21872_cast, var_22001_cast))[name = tensor<string, []>("op_22032_cast")];
            tensor<string, []> var_22034_equation_0 = const()[name = tensor<string, []>("op_22034_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22034_cast = einsum(equation = var_22034_equation_0, values = (var_21876_cast, var_22002_cast))[name = tensor<string, []>("op_22034_cast")];
            tensor<string, []> var_22036_equation_0 = const()[name = tensor<string, []>("op_22036_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22036_cast = einsum(equation = var_22036_equation_0, values = (var_21880_cast, var_22003_cast))[name = tensor<string, []>("op_22036_cast")];
            tensor<string, []> var_22038_equation_0 = const()[name = tensor<string, []>("op_22038_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22038_cast = einsum(equation = var_22038_equation_0, values = (var_21884_cast, var_22004_cast))[name = tensor<string, []>("op_22038_cast")];
            tensor<string, []> var_22040_equation_0 = const()[name = tensor<string, []>("op_22040_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22040_cast = einsum(equation = var_22040_equation_0, values = (var_21888_cast, var_22005_cast))[name = tensor<string, []>("op_22040_cast")];
            tensor<string, []> var_22042_equation_0 = const()[name = tensor<string, []>("op_22042_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22042_cast = einsum(equation = var_22042_equation_0, values = (var_21892_cast, var_22006_cast))[name = tensor<string, []>("op_22042_cast")];
            tensor<string, []> var_22044_equation_0 = const()[name = tensor<string, []>("op_22044_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22044_cast = einsum(equation = var_22044_equation_0, values = (var_21896_cast, var_22007_cast))[name = tensor<string, []>("op_22044_cast")];
            tensor<string, []> var_22046_equation_0 = const()[name = tensor<string, []>("op_22046_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22046_cast = einsum(equation = var_22046_equation_0, values = (var_21900_cast, var_22008_cast))[name = tensor<string, []>("op_22046_cast")];
            tensor<string, []> var_22048_equation_0 = const()[name = tensor<string, []>("op_22048_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22048_cast = einsum(equation = var_22048_equation_0, values = (var_21904_cast, var_22009_cast))[name = tensor<string, []>("op_22048_cast")];
            tensor<string, []> var_22050_equation_0 = const()[name = tensor<string, []>("op_22050_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22050_cast = einsum(equation = var_22050_equation_0, values = (var_21908_cast, var_22010_cast))[name = tensor<string, []>("op_22050_cast")];
            tensor<bool, []> input_331_interleave_0 = const()[name = tensor<string, []>("input_331_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_331_cast = concat(axis = var_21105, interleave = input_331_interleave_0, values = (var_22012_cast, var_22014_cast, var_22016_cast, var_22018_cast, var_22020_cast, var_22022_cast, var_22024_cast, var_22026_cast, var_22028_cast, var_22030_cast, var_22032_cast, var_22034_cast, var_22036_cast, var_22038_cast, var_22040_cast, var_22042_cast, var_22044_cast, var_22046_cast, var_22048_cast, var_22050_cast))[name = tensor<string, []>("input_331_cast")];
            tensor<int32, [2]> var_22056 = const()[name = tensor<string, []>("op_22056"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_22058 = const()[name = tensor<string, []>("op_22058"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_22060_pad_type_0 = const()[name = tensor<string, []>("op_22060_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_22060_pad_0 = const()[name = tensor<string, []>("op_22060_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1752992384)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1756269248)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_22060_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_22058, groups = var_21105, pad = var_22060_pad_0, pad_type = var_22060_pad_type_0, strides = var_22056, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_331_cast)[name = tensor<string, []>("op_22060_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_149_cast = add(x = var_22060_cast, y = inputs_147_cast)[name = tensor<string, []>("inputs_149_cast")];
            tensor<int32, [1]> var_22064 = const()[name = tensor<string, []>("op_22064"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_149_cast = reduce_mean(axes = var_22064, keep_dims = var_21100, x = inputs_149_cast)[name = tensor<string, []>("channels_mean_149_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_149_cast = sub(x = inputs_149_cast, y = channels_mean_149_cast)[name = tensor<string, []>("zero_mean_149_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_149_cast = mul(x = zero_mean_149_cast, y = zero_mean_149_cast)[name = tensor<string, []>("zero_mean_sq_149_cast")];
            tensor<int32, [1]> var_22068 = const()[name = tensor<string, []>("op_22068"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_22069_cast = reduce_mean(axes = var_22068, keep_dims = var_21100, x = zero_mean_sq_149_cast)[name = tensor<string, []>("op_22069_cast")];
            tensor<fp16, []> var_22070_to_fp16 = const()[name = tensor<string, []>("op_22070_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_22071_cast = add(x = var_22069_cast, y = var_22070_to_fp16)[name = tensor<string, []>("op_22071_cast")];
            tensor<fp16, []> denom_149_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_149_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_149_cast = rsqrt(epsilon = denom_149_epsilon_0_to_fp16, x = var_22071_cast)[name = tensor<string, []>("denom_149_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_149_cast = mul(x = zero_mean_149_cast, y = denom_149_cast)[name = tensor<string, []>("out_149_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_22075_to_fp16 = const()[name = tensor<string, []>("op_22075_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1756271872)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_22076_cast = add(x = out_149_cast, y = var_22075_to_fp16)[name = tensor<string, []>("op_22076_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_22078_to_fp16 = const()[name = tensor<string, []>("op_22078_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1756274496)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_333_cast = mul(x = var_22076_cast, y = var_22078_to_fp16)[name = tensor<string, []>("input_333_cast")];
            tensor<int32, [2]> var_22086 = const()[name = tensor<string, []>("op_22086"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_22088 = const()[name = tensor<string, []>("op_22088"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_22090_pad_type_0 = const()[name = tensor<string, []>("op_22090_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_22090_pad_0 = const()[name = tensor<string, []>("op_22090_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1756277120)))];
            tensor<fp16, [10240]> mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1782491584)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_22090_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_22088, groups = var_21105, pad = var_22090_pad_0, pad_type = var_22090_pad_type_0, strides = var_22086, weight = mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_333_cast)[name = tensor<string, []>("op_22090_cast")];
            tensor<int32, [2]> var_22091_split_sizes_0 = const()[name = tensor<string, []>("op_22091_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_22091_axis_0 = const()[name = tensor<string, []>("op_22091_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_22091_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_22091_cast_1 = split(axis = var_22091_axis_0, split_sizes = var_22091_split_sizes_0, x = var_22090_cast)[name = tensor<string, []>("op_22091_cast")];
            tensor<string, []> var_22093_mode_0 = const()[name = tensor<string, []>("op_22093_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_22093_cast = gelu(mode = var_22093_mode_0, x = var_22091_cast_1)[name = tensor<string, []>("op_22093_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_335_cast = mul(x = var_22091_cast_0, y = var_22093_cast)[name = tensor<string, []>("input_335_cast")];
            tensor<int32, [2]> var_22097 = const()[name = tensor<string, []>("op_22097"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_22099 = const()[name = tensor<string, []>("op_22099"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_22101_pad_type_0 = const()[name = tensor<string, []>("op_22101_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_22101_pad_0 = const()[name = tensor<string, []>("op_22101_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1782512128)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1795619392)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_22101_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_22099, groups = var_21105, pad = var_22101_pad_0, pad_type = var_22101_pad_type_0, strides = var_22097, weight = mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_335_cast)[name = tensor<string, []>("op_22101_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_151_cast = add(x = var_22101_cast, y = inputs_149_cast)[name = tensor<string, []>("inputs_151_cast")];
            tensor<int32, [1]> var_22111 = const()[name = tensor<string, []>("op_22111"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_151_cast = reduce_mean(axes = var_22111, keep_dims = var_21100, x = inputs_151_cast)[name = tensor<string, []>("channels_mean_151_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_151_cast = sub(x = inputs_151_cast, y = channels_mean_151_cast)[name = tensor<string, []>("zero_mean_151_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_151_cast = mul(x = zero_mean_151_cast, y = zero_mean_151_cast)[name = tensor<string, []>("zero_mean_sq_151_cast")];
            tensor<int32, [1]> var_22115 = const()[name = tensor<string, []>("op_22115"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_22116_cast = reduce_mean(axes = var_22115, keep_dims = var_21100, x = zero_mean_sq_151_cast)[name = tensor<string, []>("op_22116_cast")];
            tensor<fp16, []> var_22117_to_fp16 = const()[name = tensor<string, []>("op_22117_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_22118_cast = add(x = var_22116_cast, y = var_22117_to_fp16)[name = tensor<string, []>("op_22118_cast")];
            tensor<fp16, []> denom_151_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_151_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_151_cast = rsqrt(epsilon = denom_151_epsilon_0_to_fp16, x = var_22118_cast)[name = tensor<string, []>("denom_151_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_151_cast = mul(x = zero_mean_151_cast, y = denom_151_cast)[name = tensor<string, []>("out_151_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_22122_to_fp16 = const()[name = tensor<string, []>("op_22122_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1795622016)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_22123_cast = add(x = out_151_cast, y = var_22122_to_fp16)[name = tensor<string, []>("op_22123_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_22125_to_fp16 = const()[name = tensor<string, []>("op_22125_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1795624640)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_215_cast = mul(x = var_22123_cast, y = var_22125_to_fp16)[name = tensor<string, []>("hidden_states_215_cast")];
            tensor<int32, [2]> var_22132 = const()[name = tensor<string, []>("op_22132"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_22134 = const()[name = tensor<string, []>("op_22134"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_101_pad_type_0 = const()[name = tensor<string, []>("q_101_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_101_pad_0 = const()[name = tensor<string, []>("q_101_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1795627264)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_101_cast = conv(dilations = var_22134, groups = var_21105, pad = q_101_pad_0, pad_type = q_101_pad_type_0, strides = var_22132, weight = mid_block_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_215_cast)[name = tensor<string, []>("q_101_cast")];
            tensor<int32, [2]> var_22138 = const()[name = tensor<string, []>("op_22138"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_22140 = const()[name = tensor<string, []>("op_22140"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_201_pad_type_0 = const()[name = tensor<string, []>("k_201_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_201_pad_0 = const()[name = tensor<string, []>("k_201_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1798904128)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_201_cast = conv(dilations = var_22140, groups = var_21105, pad = k_201_pad_0, pad_type = k_201_pad_type_0, strides = var_22138, weight = mid_block_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_215_cast)[name = tensor<string, []>("k_201_cast")];
            tensor<int32, [2]> var_22144 = const()[name = tensor<string, []>("op_22144"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_22146 = const()[name = tensor<string, []>("op_22146"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_101_pad_type_0 = const()[name = tensor<string, []>("v_101_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_101_pad_0 = const()[name = tensor<string, []>("v_101_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1802180992)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_101_cast = conv(dilations = var_22146, groups = var_21105, pad = v_101_pad_0, pad_type = v_101_pad_type_0, strides = var_22144, weight = mid_block_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_215_cast)[name = tensor<string, []>("v_101_cast")];
            tensor<int32, [4]> var_22150_begin_0 = const()[name = tensor<string, []>("op_22150_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_22150_end_0 = const()[name = tensor<string, []>("op_22150_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_22150_end_mask_0 = const()[name = tensor<string, []>("op_22150_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22150_cast = slice_by_index(begin = var_22150_begin_0, end = var_22150_end_0, end_mask = var_22150_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22150_cast")];
            tensor<int32, [4]> var_22154_begin_0 = const()[name = tensor<string, []>("op_22154_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_22154_end_0 = const()[name = tensor<string, []>("op_22154_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_22154_end_mask_0 = const()[name = tensor<string, []>("op_22154_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22154_cast = slice_by_index(begin = var_22154_begin_0, end = var_22154_end_0, end_mask = var_22154_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22154_cast")];
            tensor<int32, [4]> var_22158_begin_0 = const()[name = tensor<string, []>("op_22158_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_22158_end_0 = const()[name = tensor<string, []>("op_22158_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_22158_end_mask_0 = const()[name = tensor<string, []>("op_22158_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22158_cast = slice_by_index(begin = var_22158_begin_0, end = var_22158_end_0, end_mask = var_22158_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22158_cast")];
            tensor<int32, [4]> var_22162_begin_0 = const()[name = tensor<string, []>("op_22162_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_22162_end_0 = const()[name = tensor<string, []>("op_22162_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_22162_end_mask_0 = const()[name = tensor<string, []>("op_22162_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22162_cast = slice_by_index(begin = var_22162_begin_0, end = var_22162_end_0, end_mask = var_22162_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22162_cast")];
            tensor<int32, [4]> var_22166_begin_0 = const()[name = tensor<string, []>("op_22166_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_22166_end_0 = const()[name = tensor<string, []>("op_22166_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_22166_end_mask_0 = const()[name = tensor<string, []>("op_22166_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22166_cast = slice_by_index(begin = var_22166_begin_0, end = var_22166_end_0, end_mask = var_22166_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22166_cast")];
            tensor<int32, [4]> var_22170_begin_0 = const()[name = tensor<string, []>("op_22170_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_22170_end_0 = const()[name = tensor<string, []>("op_22170_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_22170_end_mask_0 = const()[name = tensor<string, []>("op_22170_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22170_cast = slice_by_index(begin = var_22170_begin_0, end = var_22170_end_0, end_mask = var_22170_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22170_cast")];
            tensor<int32, [4]> var_22174_begin_0 = const()[name = tensor<string, []>("op_22174_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_22174_end_0 = const()[name = tensor<string, []>("op_22174_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_22174_end_mask_0 = const()[name = tensor<string, []>("op_22174_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22174_cast = slice_by_index(begin = var_22174_begin_0, end = var_22174_end_0, end_mask = var_22174_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22174_cast")];
            tensor<int32, [4]> var_22178_begin_0 = const()[name = tensor<string, []>("op_22178_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_22178_end_0 = const()[name = tensor<string, []>("op_22178_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_22178_end_mask_0 = const()[name = tensor<string, []>("op_22178_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22178_cast = slice_by_index(begin = var_22178_begin_0, end = var_22178_end_0, end_mask = var_22178_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22178_cast")];
            tensor<int32, [4]> var_22182_begin_0 = const()[name = tensor<string, []>("op_22182_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_22182_end_0 = const()[name = tensor<string, []>("op_22182_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_22182_end_mask_0 = const()[name = tensor<string, []>("op_22182_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22182_cast = slice_by_index(begin = var_22182_begin_0, end = var_22182_end_0, end_mask = var_22182_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22182_cast")];
            tensor<int32, [4]> var_22186_begin_0 = const()[name = tensor<string, []>("op_22186_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_22186_end_0 = const()[name = tensor<string, []>("op_22186_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_22186_end_mask_0 = const()[name = tensor<string, []>("op_22186_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22186_cast = slice_by_index(begin = var_22186_begin_0, end = var_22186_end_0, end_mask = var_22186_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22186_cast")];
            tensor<int32, [4]> var_22190_begin_0 = const()[name = tensor<string, []>("op_22190_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_22190_end_0 = const()[name = tensor<string, []>("op_22190_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_22190_end_mask_0 = const()[name = tensor<string, []>("op_22190_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22190_cast = slice_by_index(begin = var_22190_begin_0, end = var_22190_end_0, end_mask = var_22190_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22190_cast")];
            tensor<int32, [4]> var_22194_begin_0 = const()[name = tensor<string, []>("op_22194_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_22194_end_0 = const()[name = tensor<string, []>("op_22194_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_22194_end_mask_0 = const()[name = tensor<string, []>("op_22194_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22194_cast = slice_by_index(begin = var_22194_begin_0, end = var_22194_end_0, end_mask = var_22194_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22194_cast")];
            tensor<int32, [4]> var_22198_begin_0 = const()[name = tensor<string, []>("op_22198_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_22198_end_0 = const()[name = tensor<string, []>("op_22198_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_22198_end_mask_0 = const()[name = tensor<string, []>("op_22198_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22198_cast = slice_by_index(begin = var_22198_begin_0, end = var_22198_end_0, end_mask = var_22198_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22198_cast")];
            tensor<int32, [4]> var_22202_begin_0 = const()[name = tensor<string, []>("op_22202_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_22202_end_0 = const()[name = tensor<string, []>("op_22202_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_22202_end_mask_0 = const()[name = tensor<string, []>("op_22202_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22202_cast = slice_by_index(begin = var_22202_begin_0, end = var_22202_end_0, end_mask = var_22202_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22202_cast")];
            tensor<int32, [4]> var_22206_begin_0 = const()[name = tensor<string, []>("op_22206_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_22206_end_0 = const()[name = tensor<string, []>("op_22206_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_22206_end_mask_0 = const()[name = tensor<string, []>("op_22206_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22206_cast = slice_by_index(begin = var_22206_begin_0, end = var_22206_end_0, end_mask = var_22206_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22206_cast")];
            tensor<int32, [4]> var_22210_begin_0 = const()[name = tensor<string, []>("op_22210_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_22210_end_0 = const()[name = tensor<string, []>("op_22210_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_22210_end_mask_0 = const()[name = tensor<string, []>("op_22210_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22210_cast = slice_by_index(begin = var_22210_begin_0, end = var_22210_end_0, end_mask = var_22210_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22210_cast")];
            tensor<int32, [4]> var_22214_begin_0 = const()[name = tensor<string, []>("op_22214_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_22214_end_0 = const()[name = tensor<string, []>("op_22214_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_22214_end_mask_0 = const()[name = tensor<string, []>("op_22214_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22214_cast = slice_by_index(begin = var_22214_begin_0, end = var_22214_end_0, end_mask = var_22214_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22214_cast")];
            tensor<int32, [4]> var_22218_begin_0 = const()[name = tensor<string, []>("op_22218_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_22218_end_0 = const()[name = tensor<string, []>("op_22218_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_22218_end_mask_0 = const()[name = tensor<string, []>("op_22218_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22218_cast = slice_by_index(begin = var_22218_begin_0, end = var_22218_end_0, end_mask = var_22218_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22218_cast")];
            tensor<int32, [4]> var_22222_begin_0 = const()[name = tensor<string, []>("op_22222_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_22222_end_0 = const()[name = tensor<string, []>("op_22222_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_22222_end_mask_0 = const()[name = tensor<string, []>("op_22222_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22222_cast = slice_by_index(begin = var_22222_begin_0, end = var_22222_end_0, end_mask = var_22222_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22222_cast")];
            tensor<int32, [4]> var_22226_begin_0 = const()[name = tensor<string, []>("op_22226_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_22226_end_0 = const()[name = tensor<string, []>("op_22226_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_22226_end_mask_0 = const()[name = tensor<string, []>("op_22226_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22226_cast = slice_by_index(begin = var_22226_begin_0, end = var_22226_end_0, end_mask = var_22226_end_mask_0, x = q_101_cast)[name = tensor<string, []>("op_22226_cast")];
            tensor<int32, [4]> k_203_perm_0 = const()[name = tensor<string, []>("k_203_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_22233_begin_0 = const()[name = tensor<string, []>("op_22233_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_22233_end_0 = const()[name = tensor<string, []>("op_22233_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_22233_end_mask_0 = const()[name = tensor<string, []>("op_22233_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_89 = transpose(perm = k_203_perm_0, x = k_201_cast)[name = tensor<string, []>("transpose_89")];
            tensor<fp16, [2, 1024, 1, 64]> var_22233_cast = slice_by_index(begin = var_22233_begin_0, end = var_22233_end_0, end_mask = var_22233_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22233_cast")];
            tensor<int32, [4]> var_22237_begin_0 = const()[name = tensor<string, []>("op_22237_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_22237_end_0 = const()[name = tensor<string, []>("op_22237_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_22237_end_mask_0 = const()[name = tensor<string, []>("op_22237_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_22237_cast = slice_by_index(begin = var_22237_begin_0, end = var_22237_end_0, end_mask = var_22237_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22237_cast")];
            tensor<int32, [4]> var_22241_begin_0 = const()[name = tensor<string, []>("op_22241_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_22241_end_0 = const()[name = tensor<string, []>("op_22241_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_22241_end_mask_0 = const()[name = tensor<string, []>("op_22241_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_22241_cast = slice_by_index(begin = var_22241_begin_0, end = var_22241_end_0, end_mask = var_22241_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22241_cast")];
            tensor<int32, [4]> var_22245_begin_0 = const()[name = tensor<string, []>("op_22245_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_22245_end_0 = const()[name = tensor<string, []>("op_22245_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_22245_end_mask_0 = const()[name = tensor<string, []>("op_22245_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_22245_cast = slice_by_index(begin = var_22245_begin_0, end = var_22245_end_0, end_mask = var_22245_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22245_cast")];
            tensor<int32, [4]> var_22249_begin_0 = const()[name = tensor<string, []>("op_22249_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_22249_end_0 = const()[name = tensor<string, []>("op_22249_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_22249_end_mask_0 = const()[name = tensor<string, []>("op_22249_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_22249_cast = slice_by_index(begin = var_22249_begin_0, end = var_22249_end_0, end_mask = var_22249_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22249_cast")];
            tensor<int32, [4]> var_22253_begin_0 = const()[name = tensor<string, []>("op_22253_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_22253_end_0 = const()[name = tensor<string, []>("op_22253_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_22253_end_mask_0 = const()[name = tensor<string, []>("op_22253_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_22253_cast = slice_by_index(begin = var_22253_begin_0, end = var_22253_end_0, end_mask = var_22253_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22253_cast")];
            tensor<int32, [4]> var_22257_begin_0 = const()[name = tensor<string, []>("op_22257_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_22257_end_0 = const()[name = tensor<string, []>("op_22257_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_22257_end_mask_0 = const()[name = tensor<string, []>("op_22257_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_22257_cast = slice_by_index(begin = var_22257_begin_0, end = var_22257_end_0, end_mask = var_22257_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22257_cast")];
            tensor<int32, [4]> var_22261_begin_0 = const()[name = tensor<string, []>("op_22261_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_22261_end_0 = const()[name = tensor<string, []>("op_22261_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_22261_end_mask_0 = const()[name = tensor<string, []>("op_22261_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_22261_cast = slice_by_index(begin = var_22261_begin_0, end = var_22261_end_0, end_mask = var_22261_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22261_cast")];
            tensor<int32, [4]> var_22265_begin_0 = const()[name = tensor<string, []>("op_22265_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_22265_end_0 = const()[name = tensor<string, []>("op_22265_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_22265_end_mask_0 = const()[name = tensor<string, []>("op_22265_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_22265_cast = slice_by_index(begin = var_22265_begin_0, end = var_22265_end_0, end_mask = var_22265_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22265_cast")];
            tensor<int32, [4]> var_22269_begin_0 = const()[name = tensor<string, []>("op_22269_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_22269_end_0 = const()[name = tensor<string, []>("op_22269_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_22269_end_mask_0 = const()[name = tensor<string, []>("op_22269_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_22269_cast = slice_by_index(begin = var_22269_begin_0, end = var_22269_end_0, end_mask = var_22269_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22269_cast")];
            tensor<int32, [4]> var_22273_begin_0 = const()[name = tensor<string, []>("op_22273_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_22273_end_0 = const()[name = tensor<string, []>("op_22273_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_22273_end_mask_0 = const()[name = tensor<string, []>("op_22273_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_22273_cast = slice_by_index(begin = var_22273_begin_0, end = var_22273_end_0, end_mask = var_22273_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22273_cast")];
            tensor<int32, [4]> var_22277_begin_0 = const()[name = tensor<string, []>("op_22277_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_22277_end_0 = const()[name = tensor<string, []>("op_22277_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_22277_end_mask_0 = const()[name = tensor<string, []>("op_22277_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_22277_cast = slice_by_index(begin = var_22277_begin_0, end = var_22277_end_0, end_mask = var_22277_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22277_cast")];
            tensor<int32, [4]> var_22281_begin_0 = const()[name = tensor<string, []>("op_22281_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_22281_end_0 = const()[name = tensor<string, []>("op_22281_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_22281_end_mask_0 = const()[name = tensor<string, []>("op_22281_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_22281_cast = slice_by_index(begin = var_22281_begin_0, end = var_22281_end_0, end_mask = var_22281_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22281_cast")];
            tensor<int32, [4]> var_22285_begin_0 = const()[name = tensor<string, []>("op_22285_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_22285_end_0 = const()[name = tensor<string, []>("op_22285_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_22285_end_mask_0 = const()[name = tensor<string, []>("op_22285_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_22285_cast = slice_by_index(begin = var_22285_begin_0, end = var_22285_end_0, end_mask = var_22285_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22285_cast")];
            tensor<int32, [4]> var_22289_begin_0 = const()[name = tensor<string, []>("op_22289_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_22289_end_0 = const()[name = tensor<string, []>("op_22289_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_22289_end_mask_0 = const()[name = tensor<string, []>("op_22289_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_22289_cast = slice_by_index(begin = var_22289_begin_0, end = var_22289_end_0, end_mask = var_22289_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22289_cast")];
            tensor<int32, [4]> var_22293_begin_0 = const()[name = tensor<string, []>("op_22293_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_22293_end_0 = const()[name = tensor<string, []>("op_22293_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_22293_end_mask_0 = const()[name = tensor<string, []>("op_22293_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_22293_cast = slice_by_index(begin = var_22293_begin_0, end = var_22293_end_0, end_mask = var_22293_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22293_cast")];
            tensor<int32, [4]> var_22297_begin_0 = const()[name = tensor<string, []>("op_22297_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_22297_end_0 = const()[name = tensor<string, []>("op_22297_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_22297_end_mask_0 = const()[name = tensor<string, []>("op_22297_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_22297_cast = slice_by_index(begin = var_22297_begin_0, end = var_22297_end_0, end_mask = var_22297_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22297_cast")];
            tensor<int32, [4]> var_22301_begin_0 = const()[name = tensor<string, []>("op_22301_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_22301_end_0 = const()[name = tensor<string, []>("op_22301_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_22301_end_mask_0 = const()[name = tensor<string, []>("op_22301_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_22301_cast = slice_by_index(begin = var_22301_begin_0, end = var_22301_end_0, end_mask = var_22301_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22301_cast")];
            tensor<int32, [4]> var_22305_begin_0 = const()[name = tensor<string, []>("op_22305_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_22305_end_0 = const()[name = tensor<string, []>("op_22305_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_22305_end_mask_0 = const()[name = tensor<string, []>("op_22305_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_22305_cast = slice_by_index(begin = var_22305_begin_0, end = var_22305_end_0, end_mask = var_22305_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22305_cast")];
            tensor<int32, [4]> var_22309_begin_0 = const()[name = tensor<string, []>("op_22309_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_22309_end_0 = const()[name = tensor<string, []>("op_22309_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_22309_end_mask_0 = const()[name = tensor<string, []>("op_22309_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_22309_cast = slice_by_index(begin = var_22309_begin_0, end = var_22309_end_0, end_mask = var_22309_end_mask_0, x = transpose_89)[name = tensor<string, []>("op_22309_cast")];
            tensor<int32, [4]> var_22311_begin_0 = const()[name = tensor<string, []>("op_22311_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_22311_end_0 = const()[name = tensor<string, []>("op_22311_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_22311_end_mask_0 = const()[name = tensor<string, []>("op_22311_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22311_cast = slice_by_index(begin = var_22311_begin_0, end = var_22311_end_0, end_mask = var_22311_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22311_cast")];
            tensor<int32, [4]> var_22315_begin_0 = const()[name = tensor<string, []>("op_22315_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_22315_end_0 = const()[name = tensor<string, []>("op_22315_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_22315_end_mask_0 = const()[name = tensor<string, []>("op_22315_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22315_cast = slice_by_index(begin = var_22315_begin_0, end = var_22315_end_0, end_mask = var_22315_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22315_cast")];
            tensor<int32, [4]> var_22319_begin_0 = const()[name = tensor<string, []>("op_22319_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_22319_end_0 = const()[name = tensor<string, []>("op_22319_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_22319_end_mask_0 = const()[name = tensor<string, []>("op_22319_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22319_cast = slice_by_index(begin = var_22319_begin_0, end = var_22319_end_0, end_mask = var_22319_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22319_cast")];
            tensor<int32, [4]> var_22323_begin_0 = const()[name = tensor<string, []>("op_22323_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_22323_end_0 = const()[name = tensor<string, []>("op_22323_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_22323_end_mask_0 = const()[name = tensor<string, []>("op_22323_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22323_cast = slice_by_index(begin = var_22323_begin_0, end = var_22323_end_0, end_mask = var_22323_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22323_cast")];
            tensor<int32, [4]> var_22327_begin_0 = const()[name = tensor<string, []>("op_22327_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_22327_end_0 = const()[name = tensor<string, []>("op_22327_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_22327_end_mask_0 = const()[name = tensor<string, []>("op_22327_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22327_cast = slice_by_index(begin = var_22327_begin_0, end = var_22327_end_0, end_mask = var_22327_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22327_cast")];
            tensor<int32, [4]> var_22331_begin_0 = const()[name = tensor<string, []>("op_22331_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_22331_end_0 = const()[name = tensor<string, []>("op_22331_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_22331_end_mask_0 = const()[name = tensor<string, []>("op_22331_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22331_cast = slice_by_index(begin = var_22331_begin_0, end = var_22331_end_0, end_mask = var_22331_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22331_cast")];
            tensor<int32, [4]> var_22335_begin_0 = const()[name = tensor<string, []>("op_22335_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_22335_end_0 = const()[name = tensor<string, []>("op_22335_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_22335_end_mask_0 = const()[name = tensor<string, []>("op_22335_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22335_cast = slice_by_index(begin = var_22335_begin_0, end = var_22335_end_0, end_mask = var_22335_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22335_cast")];
            tensor<int32, [4]> var_22339_begin_0 = const()[name = tensor<string, []>("op_22339_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_22339_end_0 = const()[name = tensor<string, []>("op_22339_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_22339_end_mask_0 = const()[name = tensor<string, []>("op_22339_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22339_cast = slice_by_index(begin = var_22339_begin_0, end = var_22339_end_0, end_mask = var_22339_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22339_cast")];
            tensor<int32, [4]> var_22343_begin_0 = const()[name = tensor<string, []>("op_22343_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_22343_end_0 = const()[name = tensor<string, []>("op_22343_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_22343_end_mask_0 = const()[name = tensor<string, []>("op_22343_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22343_cast = slice_by_index(begin = var_22343_begin_0, end = var_22343_end_0, end_mask = var_22343_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22343_cast")];
            tensor<int32, [4]> var_22347_begin_0 = const()[name = tensor<string, []>("op_22347_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_22347_end_0 = const()[name = tensor<string, []>("op_22347_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_22347_end_mask_0 = const()[name = tensor<string, []>("op_22347_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22347_cast = slice_by_index(begin = var_22347_begin_0, end = var_22347_end_0, end_mask = var_22347_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22347_cast")];
            tensor<int32, [4]> var_22351_begin_0 = const()[name = tensor<string, []>("op_22351_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_22351_end_0 = const()[name = tensor<string, []>("op_22351_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_22351_end_mask_0 = const()[name = tensor<string, []>("op_22351_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22351_cast = slice_by_index(begin = var_22351_begin_0, end = var_22351_end_0, end_mask = var_22351_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22351_cast")];
            tensor<int32, [4]> var_22355_begin_0 = const()[name = tensor<string, []>("op_22355_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_22355_end_0 = const()[name = tensor<string, []>("op_22355_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_22355_end_mask_0 = const()[name = tensor<string, []>("op_22355_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22355_cast = slice_by_index(begin = var_22355_begin_0, end = var_22355_end_0, end_mask = var_22355_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22355_cast")];
            tensor<int32, [4]> var_22359_begin_0 = const()[name = tensor<string, []>("op_22359_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_22359_end_0 = const()[name = tensor<string, []>("op_22359_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_22359_end_mask_0 = const()[name = tensor<string, []>("op_22359_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22359_cast = slice_by_index(begin = var_22359_begin_0, end = var_22359_end_0, end_mask = var_22359_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22359_cast")];
            tensor<int32, [4]> var_22363_begin_0 = const()[name = tensor<string, []>("op_22363_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_22363_end_0 = const()[name = tensor<string, []>("op_22363_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_22363_end_mask_0 = const()[name = tensor<string, []>("op_22363_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22363_cast = slice_by_index(begin = var_22363_begin_0, end = var_22363_end_0, end_mask = var_22363_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22363_cast")];
            tensor<int32, [4]> var_22367_begin_0 = const()[name = tensor<string, []>("op_22367_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_22367_end_0 = const()[name = tensor<string, []>("op_22367_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_22367_end_mask_0 = const()[name = tensor<string, []>("op_22367_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22367_cast = slice_by_index(begin = var_22367_begin_0, end = var_22367_end_0, end_mask = var_22367_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22367_cast")];
            tensor<int32, [4]> var_22371_begin_0 = const()[name = tensor<string, []>("op_22371_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_22371_end_0 = const()[name = tensor<string, []>("op_22371_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_22371_end_mask_0 = const()[name = tensor<string, []>("op_22371_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22371_cast = slice_by_index(begin = var_22371_begin_0, end = var_22371_end_0, end_mask = var_22371_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22371_cast")];
            tensor<int32, [4]> var_22375_begin_0 = const()[name = tensor<string, []>("op_22375_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_22375_end_0 = const()[name = tensor<string, []>("op_22375_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_22375_end_mask_0 = const()[name = tensor<string, []>("op_22375_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22375_cast = slice_by_index(begin = var_22375_begin_0, end = var_22375_end_0, end_mask = var_22375_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22375_cast")];
            tensor<int32, [4]> var_22379_begin_0 = const()[name = tensor<string, []>("op_22379_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_22379_end_0 = const()[name = tensor<string, []>("op_22379_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_22379_end_mask_0 = const()[name = tensor<string, []>("op_22379_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22379_cast = slice_by_index(begin = var_22379_begin_0, end = var_22379_end_0, end_mask = var_22379_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22379_cast")];
            tensor<int32, [4]> var_22383_begin_0 = const()[name = tensor<string, []>("op_22383_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_22383_end_0 = const()[name = tensor<string, []>("op_22383_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_22383_end_mask_0 = const()[name = tensor<string, []>("op_22383_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22383_cast = slice_by_index(begin = var_22383_begin_0, end = var_22383_end_0, end_mask = var_22383_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22383_cast")];
            tensor<int32, [4]> var_22387_begin_0 = const()[name = tensor<string, []>("op_22387_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_22387_end_0 = const()[name = tensor<string, []>("op_22387_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_22387_end_mask_0 = const()[name = tensor<string, []>("op_22387_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22387_cast = slice_by_index(begin = var_22387_begin_0, end = var_22387_end_0, end_mask = var_22387_end_mask_0, x = v_101_cast)[name = tensor<string, []>("op_22387_cast")];
            tensor<string, []> var_22391_equation_0 = const()[name = tensor<string, []>("op_22391_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22391_cast = einsum(equation = var_22391_equation_0, values = (var_22233_cast, var_22150_cast))[name = tensor<string, []>("op_22391_cast")];
            tensor<fp16, []> var_22392_to_fp16 = const()[name = tensor<string, []>("op_22392_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1841_cast = mul(x = var_22391_cast, y = var_22392_to_fp16)[name = tensor<string, []>("aw_1841_cast")];
            tensor<string, []> var_22395_equation_0 = const()[name = tensor<string, []>("op_22395_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22395_cast = einsum(equation = var_22395_equation_0, values = (var_22237_cast, var_22154_cast))[name = tensor<string, []>("op_22395_cast")];
            tensor<fp16, []> var_22396_to_fp16 = const()[name = tensor<string, []>("op_22396_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1843_cast = mul(x = var_22395_cast, y = var_22396_to_fp16)[name = tensor<string, []>("aw_1843_cast")];
            tensor<string, []> var_22399_equation_0 = const()[name = tensor<string, []>("op_22399_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22399_cast = einsum(equation = var_22399_equation_0, values = (var_22241_cast, var_22158_cast))[name = tensor<string, []>("op_22399_cast")];
            tensor<fp16, []> var_22400_to_fp16 = const()[name = tensor<string, []>("op_22400_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1845_cast = mul(x = var_22399_cast, y = var_22400_to_fp16)[name = tensor<string, []>("aw_1845_cast")];
            tensor<string, []> var_22403_equation_0 = const()[name = tensor<string, []>("op_22403_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22403_cast = einsum(equation = var_22403_equation_0, values = (var_22245_cast, var_22162_cast))[name = tensor<string, []>("op_22403_cast")];
            tensor<fp16, []> var_22404_to_fp16 = const()[name = tensor<string, []>("op_22404_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1847_cast = mul(x = var_22403_cast, y = var_22404_to_fp16)[name = tensor<string, []>("aw_1847_cast")];
            tensor<string, []> var_22407_equation_0 = const()[name = tensor<string, []>("op_22407_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22407_cast = einsum(equation = var_22407_equation_0, values = (var_22249_cast, var_22166_cast))[name = tensor<string, []>("op_22407_cast")];
            tensor<fp16, []> var_22408_to_fp16 = const()[name = tensor<string, []>("op_22408_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1849_cast = mul(x = var_22407_cast, y = var_22408_to_fp16)[name = tensor<string, []>("aw_1849_cast")];
            tensor<string, []> var_22411_equation_0 = const()[name = tensor<string, []>("op_22411_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22411_cast = einsum(equation = var_22411_equation_0, values = (var_22253_cast, var_22170_cast))[name = tensor<string, []>("op_22411_cast")];
            tensor<fp16, []> var_22412_to_fp16 = const()[name = tensor<string, []>("op_22412_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1851_cast = mul(x = var_22411_cast, y = var_22412_to_fp16)[name = tensor<string, []>("aw_1851_cast")];
            tensor<string, []> var_22415_equation_0 = const()[name = tensor<string, []>("op_22415_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22415_cast = einsum(equation = var_22415_equation_0, values = (var_22257_cast, var_22174_cast))[name = tensor<string, []>("op_22415_cast")];
            tensor<fp16, []> var_22416_to_fp16 = const()[name = tensor<string, []>("op_22416_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1853_cast = mul(x = var_22415_cast, y = var_22416_to_fp16)[name = tensor<string, []>("aw_1853_cast")];
            tensor<string, []> var_22419_equation_0 = const()[name = tensor<string, []>("op_22419_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22419_cast = einsum(equation = var_22419_equation_0, values = (var_22261_cast, var_22178_cast))[name = tensor<string, []>("op_22419_cast")];
            tensor<fp16, []> var_22420_to_fp16 = const()[name = tensor<string, []>("op_22420_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1855_cast = mul(x = var_22419_cast, y = var_22420_to_fp16)[name = tensor<string, []>("aw_1855_cast")];
            tensor<string, []> var_22423_equation_0 = const()[name = tensor<string, []>("op_22423_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22423_cast = einsum(equation = var_22423_equation_0, values = (var_22265_cast, var_22182_cast))[name = tensor<string, []>("op_22423_cast")];
            tensor<fp16, []> var_22424_to_fp16 = const()[name = tensor<string, []>("op_22424_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1857_cast = mul(x = var_22423_cast, y = var_22424_to_fp16)[name = tensor<string, []>("aw_1857_cast")];
            tensor<string, []> var_22427_equation_0 = const()[name = tensor<string, []>("op_22427_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22427_cast = einsum(equation = var_22427_equation_0, values = (var_22269_cast, var_22186_cast))[name = tensor<string, []>("op_22427_cast")];
            tensor<fp16, []> var_22428_to_fp16 = const()[name = tensor<string, []>("op_22428_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1859_cast = mul(x = var_22427_cast, y = var_22428_to_fp16)[name = tensor<string, []>("aw_1859_cast")];
            tensor<string, []> var_22431_equation_0 = const()[name = tensor<string, []>("op_22431_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22431_cast = einsum(equation = var_22431_equation_0, values = (var_22273_cast, var_22190_cast))[name = tensor<string, []>("op_22431_cast")];
            tensor<fp16, []> var_22432_to_fp16 = const()[name = tensor<string, []>("op_22432_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1861_cast = mul(x = var_22431_cast, y = var_22432_to_fp16)[name = tensor<string, []>("aw_1861_cast")];
            tensor<string, []> var_22435_equation_0 = const()[name = tensor<string, []>("op_22435_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22435_cast = einsum(equation = var_22435_equation_0, values = (var_22277_cast, var_22194_cast))[name = tensor<string, []>("op_22435_cast")];
            tensor<fp16, []> var_22436_to_fp16 = const()[name = tensor<string, []>("op_22436_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1863_cast = mul(x = var_22435_cast, y = var_22436_to_fp16)[name = tensor<string, []>("aw_1863_cast")];
            tensor<string, []> var_22439_equation_0 = const()[name = tensor<string, []>("op_22439_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22439_cast = einsum(equation = var_22439_equation_0, values = (var_22281_cast, var_22198_cast))[name = tensor<string, []>("op_22439_cast")];
            tensor<fp16, []> var_22440_to_fp16 = const()[name = tensor<string, []>("op_22440_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1865_cast = mul(x = var_22439_cast, y = var_22440_to_fp16)[name = tensor<string, []>("aw_1865_cast")];
            tensor<string, []> var_22443_equation_0 = const()[name = tensor<string, []>("op_22443_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22443_cast = einsum(equation = var_22443_equation_0, values = (var_22285_cast, var_22202_cast))[name = tensor<string, []>("op_22443_cast")];
            tensor<fp16, []> var_22444_to_fp16 = const()[name = tensor<string, []>("op_22444_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1867_cast = mul(x = var_22443_cast, y = var_22444_to_fp16)[name = tensor<string, []>("aw_1867_cast")];
            tensor<string, []> var_22447_equation_0 = const()[name = tensor<string, []>("op_22447_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22447_cast = einsum(equation = var_22447_equation_0, values = (var_22289_cast, var_22206_cast))[name = tensor<string, []>("op_22447_cast")];
            tensor<fp16, []> var_22448_to_fp16 = const()[name = tensor<string, []>("op_22448_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1869_cast = mul(x = var_22447_cast, y = var_22448_to_fp16)[name = tensor<string, []>("aw_1869_cast")];
            tensor<string, []> var_22451_equation_0 = const()[name = tensor<string, []>("op_22451_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22451_cast = einsum(equation = var_22451_equation_0, values = (var_22293_cast, var_22210_cast))[name = tensor<string, []>("op_22451_cast")];
            tensor<fp16, []> var_22452_to_fp16 = const()[name = tensor<string, []>("op_22452_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1871_cast = mul(x = var_22451_cast, y = var_22452_to_fp16)[name = tensor<string, []>("aw_1871_cast")];
            tensor<string, []> var_22455_equation_0 = const()[name = tensor<string, []>("op_22455_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22455_cast = einsum(equation = var_22455_equation_0, values = (var_22297_cast, var_22214_cast))[name = tensor<string, []>("op_22455_cast")];
            tensor<fp16, []> var_22456_to_fp16 = const()[name = tensor<string, []>("op_22456_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1873_cast = mul(x = var_22455_cast, y = var_22456_to_fp16)[name = tensor<string, []>("aw_1873_cast")];
            tensor<string, []> var_22459_equation_0 = const()[name = tensor<string, []>("op_22459_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22459_cast = einsum(equation = var_22459_equation_0, values = (var_22301_cast, var_22218_cast))[name = tensor<string, []>("op_22459_cast")];
            tensor<fp16, []> var_22460_to_fp16 = const()[name = tensor<string, []>("op_22460_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1875_cast = mul(x = var_22459_cast, y = var_22460_to_fp16)[name = tensor<string, []>("aw_1875_cast")];
            tensor<string, []> var_22463_equation_0 = const()[name = tensor<string, []>("op_22463_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22463_cast = einsum(equation = var_22463_equation_0, values = (var_22305_cast, var_22222_cast))[name = tensor<string, []>("op_22463_cast")];
            tensor<fp16, []> var_22464_to_fp16 = const()[name = tensor<string, []>("op_22464_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1877_cast = mul(x = var_22463_cast, y = var_22464_to_fp16)[name = tensor<string, []>("aw_1877_cast")];
            tensor<string, []> var_22467_equation_0 = const()[name = tensor<string, []>("op_22467_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22467_cast = einsum(equation = var_22467_equation_0, values = (var_22309_cast, var_22226_cast))[name = tensor<string, []>("op_22467_cast")];
            tensor<fp16, []> var_22468_to_fp16 = const()[name = tensor<string, []>("op_22468_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1879_cast = mul(x = var_22467_cast, y = var_22468_to_fp16)[name = tensor<string, []>("aw_1879_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22470_cast = softmax(axis = var_21105, x = aw_1841_cast)[name = tensor<string, []>("op_22470_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22471_cast = softmax(axis = var_21105, x = aw_1843_cast)[name = tensor<string, []>("op_22471_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22472_cast = softmax(axis = var_21105, x = aw_1845_cast)[name = tensor<string, []>("op_22472_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22473_cast = softmax(axis = var_21105, x = aw_1847_cast)[name = tensor<string, []>("op_22473_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22474_cast = softmax(axis = var_21105, x = aw_1849_cast)[name = tensor<string, []>("op_22474_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22475_cast = softmax(axis = var_21105, x = aw_1851_cast)[name = tensor<string, []>("op_22475_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22476_cast = softmax(axis = var_21105, x = aw_1853_cast)[name = tensor<string, []>("op_22476_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22477_cast = softmax(axis = var_21105, x = aw_1855_cast)[name = tensor<string, []>("op_22477_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22478_cast = softmax(axis = var_21105, x = aw_1857_cast)[name = tensor<string, []>("op_22478_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22479_cast = softmax(axis = var_21105, x = aw_1859_cast)[name = tensor<string, []>("op_22479_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22480_cast = softmax(axis = var_21105, x = aw_1861_cast)[name = tensor<string, []>("op_22480_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22481_cast = softmax(axis = var_21105, x = aw_1863_cast)[name = tensor<string, []>("op_22481_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22482_cast = softmax(axis = var_21105, x = aw_1865_cast)[name = tensor<string, []>("op_22482_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22483_cast = softmax(axis = var_21105, x = aw_1867_cast)[name = tensor<string, []>("op_22483_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22484_cast = softmax(axis = var_21105, x = aw_1869_cast)[name = tensor<string, []>("op_22484_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22485_cast = softmax(axis = var_21105, x = aw_1871_cast)[name = tensor<string, []>("op_22485_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22486_cast = softmax(axis = var_21105, x = aw_1873_cast)[name = tensor<string, []>("op_22486_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22487_cast = softmax(axis = var_21105, x = aw_1875_cast)[name = tensor<string, []>("op_22487_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22488_cast = softmax(axis = var_21105, x = aw_1877_cast)[name = tensor<string, []>("op_22488_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_22489_cast = softmax(axis = var_21105, x = aw_1879_cast)[name = tensor<string, []>("op_22489_cast")];
            tensor<string, []> var_22491_equation_0 = const()[name = tensor<string, []>("op_22491_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22491_cast = einsum(equation = var_22491_equation_0, values = (var_22311_cast, var_22470_cast))[name = tensor<string, []>("op_22491_cast")];
            tensor<string, []> var_22493_equation_0 = const()[name = tensor<string, []>("op_22493_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22493_cast = einsum(equation = var_22493_equation_0, values = (var_22315_cast, var_22471_cast))[name = tensor<string, []>("op_22493_cast")];
            tensor<string, []> var_22495_equation_0 = const()[name = tensor<string, []>("op_22495_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22495_cast = einsum(equation = var_22495_equation_0, values = (var_22319_cast, var_22472_cast))[name = tensor<string, []>("op_22495_cast")];
            tensor<string, []> var_22497_equation_0 = const()[name = tensor<string, []>("op_22497_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22497_cast = einsum(equation = var_22497_equation_0, values = (var_22323_cast, var_22473_cast))[name = tensor<string, []>("op_22497_cast")];
            tensor<string, []> var_22499_equation_0 = const()[name = tensor<string, []>("op_22499_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22499_cast = einsum(equation = var_22499_equation_0, values = (var_22327_cast, var_22474_cast))[name = tensor<string, []>("op_22499_cast")];
            tensor<string, []> var_22501_equation_0 = const()[name = tensor<string, []>("op_22501_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22501_cast = einsum(equation = var_22501_equation_0, values = (var_22331_cast, var_22475_cast))[name = tensor<string, []>("op_22501_cast")];
            tensor<string, []> var_22503_equation_0 = const()[name = tensor<string, []>("op_22503_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22503_cast = einsum(equation = var_22503_equation_0, values = (var_22335_cast, var_22476_cast))[name = tensor<string, []>("op_22503_cast")];
            tensor<string, []> var_22505_equation_0 = const()[name = tensor<string, []>("op_22505_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22505_cast = einsum(equation = var_22505_equation_0, values = (var_22339_cast, var_22477_cast))[name = tensor<string, []>("op_22505_cast")];
            tensor<string, []> var_22507_equation_0 = const()[name = tensor<string, []>("op_22507_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22507_cast = einsum(equation = var_22507_equation_0, values = (var_22343_cast, var_22478_cast))[name = tensor<string, []>("op_22507_cast")];
            tensor<string, []> var_22509_equation_0 = const()[name = tensor<string, []>("op_22509_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22509_cast = einsum(equation = var_22509_equation_0, values = (var_22347_cast, var_22479_cast))[name = tensor<string, []>("op_22509_cast")];
            tensor<string, []> var_22511_equation_0 = const()[name = tensor<string, []>("op_22511_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22511_cast = einsum(equation = var_22511_equation_0, values = (var_22351_cast, var_22480_cast))[name = tensor<string, []>("op_22511_cast")];
            tensor<string, []> var_22513_equation_0 = const()[name = tensor<string, []>("op_22513_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22513_cast = einsum(equation = var_22513_equation_0, values = (var_22355_cast, var_22481_cast))[name = tensor<string, []>("op_22513_cast")];
            tensor<string, []> var_22515_equation_0 = const()[name = tensor<string, []>("op_22515_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22515_cast = einsum(equation = var_22515_equation_0, values = (var_22359_cast, var_22482_cast))[name = tensor<string, []>("op_22515_cast")];
            tensor<string, []> var_22517_equation_0 = const()[name = tensor<string, []>("op_22517_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22517_cast = einsum(equation = var_22517_equation_0, values = (var_22363_cast, var_22483_cast))[name = tensor<string, []>("op_22517_cast")];
            tensor<string, []> var_22519_equation_0 = const()[name = tensor<string, []>("op_22519_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22519_cast = einsum(equation = var_22519_equation_0, values = (var_22367_cast, var_22484_cast))[name = tensor<string, []>("op_22519_cast")];
            tensor<string, []> var_22521_equation_0 = const()[name = tensor<string, []>("op_22521_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22521_cast = einsum(equation = var_22521_equation_0, values = (var_22371_cast, var_22485_cast))[name = tensor<string, []>("op_22521_cast")];
            tensor<string, []> var_22523_equation_0 = const()[name = tensor<string, []>("op_22523_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22523_cast = einsum(equation = var_22523_equation_0, values = (var_22375_cast, var_22486_cast))[name = tensor<string, []>("op_22523_cast")];
            tensor<string, []> var_22525_equation_0 = const()[name = tensor<string, []>("op_22525_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22525_cast = einsum(equation = var_22525_equation_0, values = (var_22379_cast, var_22487_cast))[name = tensor<string, []>("op_22525_cast")];
            tensor<string, []> var_22527_equation_0 = const()[name = tensor<string, []>("op_22527_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22527_cast = einsum(equation = var_22527_equation_0, values = (var_22383_cast, var_22488_cast))[name = tensor<string, []>("op_22527_cast")];
            tensor<string, []> var_22529_equation_0 = const()[name = tensor<string, []>("op_22529_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22529_cast = einsum(equation = var_22529_equation_0, values = (var_22387_cast, var_22489_cast))[name = tensor<string, []>("op_22529_cast")];
            tensor<bool, []> input_337_interleave_0 = const()[name = tensor<string, []>("input_337_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_337_cast = concat(axis = var_21105, interleave = input_337_interleave_0, values = (var_22491_cast, var_22493_cast, var_22495_cast, var_22497_cast, var_22499_cast, var_22501_cast, var_22503_cast, var_22505_cast, var_22507_cast, var_22509_cast, var_22511_cast, var_22513_cast, var_22515_cast, var_22517_cast, var_22519_cast, var_22521_cast, var_22523_cast, var_22525_cast, var_22527_cast, var_22529_cast))[name = tensor<string, []>("input_337_cast")];
            tensor<int32, [2]> var_22535 = const()[name = tensor<string, []>("op_22535"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_22537 = const()[name = tensor<string, []>("op_22537"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_22539_pad_type_0 = const()[name = tensor<string, []>("op_22539_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_22539_pad_0 = const()[name = tensor<string, []>("op_22539_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1805457856)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1808734720)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_22539_cast = conv(bias = mid_block_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_22537, groups = var_21105, pad = var_22539_pad_0, pad_type = var_22539_pad_type_0, strides = var_22535, weight = mid_block_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_337_cast)[name = tensor<string, []>("op_22539_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_153_cast = add(x = var_22539_cast, y = inputs_151_cast)[name = tensor<string, []>("inputs_153_cast")];
            tensor<int32, [1]> var_22543 = const()[name = tensor<string, []>("op_22543"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_153_cast = reduce_mean(axes = var_22543, keep_dims = var_21100, x = inputs_153_cast)[name = tensor<string, []>("channels_mean_153_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_153_cast = sub(x = inputs_153_cast, y = channels_mean_153_cast)[name = tensor<string, []>("zero_mean_153_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_153_cast = mul(x = zero_mean_153_cast, y = zero_mean_153_cast)[name = tensor<string, []>("zero_mean_sq_153_cast")];
            tensor<int32, [1]> var_22547 = const()[name = tensor<string, []>("op_22547"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_22548_cast = reduce_mean(axes = var_22547, keep_dims = var_21100, x = zero_mean_sq_153_cast)[name = tensor<string, []>("op_22548_cast")];
            tensor<fp16, []> var_22549_to_fp16 = const()[name = tensor<string, []>("op_22549_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_22550_cast = add(x = var_22548_cast, y = var_22549_to_fp16)[name = tensor<string, []>("op_22550_cast")];
            tensor<fp16, []> denom_153_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_153_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_153_cast = rsqrt(epsilon = denom_153_epsilon_0_to_fp16, x = var_22550_cast)[name = tensor<string, []>("denom_153_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_153_cast = mul(x = zero_mean_153_cast, y = denom_153_cast)[name = tensor<string, []>("out_153_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_22554_to_fp16 = const()[name = tensor<string, []>("op_22554_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1808737344)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_22555_cast = add(x = out_153_cast, y = var_22554_to_fp16)[name = tensor<string, []>("op_22555_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_22557_to_fp16 = const()[name = tensor<string, []>("op_22557_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1808739968)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_217_cast = mul(x = var_22555_cast, y = var_22557_to_fp16)[name = tensor<string, []>("hidden_states_217_cast")];
            tensor<int32, [2]> var_22564 = const()[name = tensor<string, []>("op_22564"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_22566 = const()[name = tensor<string, []>("op_22566"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_103_pad_type_0 = const()[name = tensor<string, []>("q_103_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_103_pad_0 = const()[name = tensor<string, []>("q_103_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1808742592)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_103_cast = conv(dilations = var_22566, groups = var_21105, pad = q_103_pad_0, pad_type = q_103_pad_type_0, strides = var_22564, weight = mid_block_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_217_cast)[name = tensor<string, []>("q_103_cast")];
            tensor<int32, [2]> var_22570 = const()[name = tensor<string, []>("op_22570"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_22572 = const()[name = tensor<string, []>("op_22572"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_205_pad_type_0 = const()[name = tensor<string, []>("k_205_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_205_pad_0 = const()[name = tensor<string, []>("k_205_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1812019456)))];
            tensor<fp16, [2, 1280, 1, 77]> k_205_cast = conv(dilations = var_22572, groups = var_21105, pad = k_205_pad_0, pad_type = k_205_pad_type_0, strides = var_22570, weight = mid_block_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_205_cast")];
            tensor<int32, [2]> var_22576 = const()[name = tensor<string, []>("op_22576"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_22578 = const()[name = tensor<string, []>("op_22578"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_103_pad_type_0 = const()[name = tensor<string, []>("v_103_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_103_pad_0 = const()[name = tensor<string, []>("v_103_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1817262400)))];
            tensor<fp16, [2, 1280, 1, 77]> v_103_cast = conv(dilations = var_22578, groups = var_21105, pad = v_103_pad_0, pad_type = v_103_pad_type_0, strides = var_22576, weight = mid_block_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_103_cast")];
            tensor<int32, [4]> var_22582_begin_0 = const()[name = tensor<string, []>("op_22582_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_22582_end_0 = const()[name = tensor<string, []>("op_22582_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_22582_end_mask_0 = const()[name = tensor<string, []>("op_22582_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22582_cast = slice_by_index(begin = var_22582_begin_0, end = var_22582_end_0, end_mask = var_22582_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22582_cast")];
            tensor<int32, [4]> var_22586_begin_0 = const()[name = tensor<string, []>("op_22586_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_22586_end_0 = const()[name = tensor<string, []>("op_22586_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_22586_end_mask_0 = const()[name = tensor<string, []>("op_22586_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22586_cast = slice_by_index(begin = var_22586_begin_0, end = var_22586_end_0, end_mask = var_22586_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22586_cast")];
            tensor<int32, [4]> var_22590_begin_0 = const()[name = tensor<string, []>("op_22590_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_22590_end_0 = const()[name = tensor<string, []>("op_22590_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_22590_end_mask_0 = const()[name = tensor<string, []>("op_22590_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22590_cast = slice_by_index(begin = var_22590_begin_0, end = var_22590_end_0, end_mask = var_22590_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22590_cast")];
            tensor<int32, [4]> var_22594_begin_0 = const()[name = tensor<string, []>("op_22594_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_22594_end_0 = const()[name = tensor<string, []>("op_22594_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_22594_end_mask_0 = const()[name = tensor<string, []>("op_22594_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22594_cast = slice_by_index(begin = var_22594_begin_0, end = var_22594_end_0, end_mask = var_22594_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22594_cast")];
            tensor<int32, [4]> var_22598_begin_0 = const()[name = tensor<string, []>("op_22598_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_22598_end_0 = const()[name = tensor<string, []>("op_22598_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_22598_end_mask_0 = const()[name = tensor<string, []>("op_22598_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22598_cast = slice_by_index(begin = var_22598_begin_0, end = var_22598_end_0, end_mask = var_22598_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22598_cast")];
            tensor<int32, [4]> var_22602_begin_0 = const()[name = tensor<string, []>("op_22602_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_22602_end_0 = const()[name = tensor<string, []>("op_22602_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_22602_end_mask_0 = const()[name = tensor<string, []>("op_22602_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22602_cast = slice_by_index(begin = var_22602_begin_0, end = var_22602_end_0, end_mask = var_22602_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22602_cast")];
            tensor<int32, [4]> var_22606_begin_0 = const()[name = tensor<string, []>("op_22606_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_22606_end_0 = const()[name = tensor<string, []>("op_22606_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_22606_end_mask_0 = const()[name = tensor<string, []>("op_22606_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22606_cast = slice_by_index(begin = var_22606_begin_0, end = var_22606_end_0, end_mask = var_22606_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22606_cast")];
            tensor<int32, [4]> var_22610_begin_0 = const()[name = tensor<string, []>("op_22610_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_22610_end_0 = const()[name = tensor<string, []>("op_22610_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_22610_end_mask_0 = const()[name = tensor<string, []>("op_22610_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22610_cast = slice_by_index(begin = var_22610_begin_0, end = var_22610_end_0, end_mask = var_22610_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22610_cast")];
            tensor<int32, [4]> var_22614_begin_0 = const()[name = tensor<string, []>("op_22614_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_22614_end_0 = const()[name = tensor<string, []>("op_22614_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_22614_end_mask_0 = const()[name = tensor<string, []>("op_22614_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22614_cast = slice_by_index(begin = var_22614_begin_0, end = var_22614_end_0, end_mask = var_22614_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22614_cast")];
            tensor<int32, [4]> var_22618_begin_0 = const()[name = tensor<string, []>("op_22618_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_22618_end_0 = const()[name = tensor<string, []>("op_22618_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_22618_end_mask_0 = const()[name = tensor<string, []>("op_22618_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22618_cast = slice_by_index(begin = var_22618_begin_0, end = var_22618_end_0, end_mask = var_22618_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22618_cast")];
            tensor<int32, [4]> var_22622_begin_0 = const()[name = tensor<string, []>("op_22622_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_22622_end_0 = const()[name = tensor<string, []>("op_22622_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_22622_end_mask_0 = const()[name = tensor<string, []>("op_22622_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22622_cast = slice_by_index(begin = var_22622_begin_0, end = var_22622_end_0, end_mask = var_22622_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22622_cast")];
            tensor<int32, [4]> var_22626_begin_0 = const()[name = tensor<string, []>("op_22626_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_22626_end_0 = const()[name = tensor<string, []>("op_22626_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_22626_end_mask_0 = const()[name = tensor<string, []>("op_22626_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22626_cast = slice_by_index(begin = var_22626_begin_0, end = var_22626_end_0, end_mask = var_22626_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22626_cast")];
            tensor<int32, [4]> var_22630_begin_0 = const()[name = tensor<string, []>("op_22630_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_22630_end_0 = const()[name = tensor<string, []>("op_22630_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_22630_end_mask_0 = const()[name = tensor<string, []>("op_22630_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22630_cast = slice_by_index(begin = var_22630_begin_0, end = var_22630_end_0, end_mask = var_22630_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22630_cast")];
            tensor<int32, [4]> var_22634_begin_0 = const()[name = tensor<string, []>("op_22634_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_22634_end_0 = const()[name = tensor<string, []>("op_22634_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_22634_end_mask_0 = const()[name = tensor<string, []>("op_22634_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22634_cast = slice_by_index(begin = var_22634_begin_0, end = var_22634_end_0, end_mask = var_22634_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22634_cast")];
            tensor<int32, [4]> var_22638_begin_0 = const()[name = tensor<string, []>("op_22638_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_22638_end_0 = const()[name = tensor<string, []>("op_22638_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_22638_end_mask_0 = const()[name = tensor<string, []>("op_22638_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22638_cast = slice_by_index(begin = var_22638_begin_0, end = var_22638_end_0, end_mask = var_22638_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22638_cast")];
            tensor<int32, [4]> var_22642_begin_0 = const()[name = tensor<string, []>("op_22642_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_22642_end_0 = const()[name = tensor<string, []>("op_22642_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_22642_end_mask_0 = const()[name = tensor<string, []>("op_22642_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22642_cast = slice_by_index(begin = var_22642_begin_0, end = var_22642_end_0, end_mask = var_22642_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22642_cast")];
            tensor<int32, [4]> var_22646_begin_0 = const()[name = tensor<string, []>("op_22646_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_22646_end_0 = const()[name = tensor<string, []>("op_22646_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_22646_end_mask_0 = const()[name = tensor<string, []>("op_22646_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22646_cast = slice_by_index(begin = var_22646_begin_0, end = var_22646_end_0, end_mask = var_22646_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22646_cast")];
            tensor<int32, [4]> var_22650_begin_0 = const()[name = tensor<string, []>("op_22650_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_22650_end_0 = const()[name = tensor<string, []>("op_22650_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_22650_end_mask_0 = const()[name = tensor<string, []>("op_22650_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22650_cast = slice_by_index(begin = var_22650_begin_0, end = var_22650_end_0, end_mask = var_22650_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22650_cast")];
            tensor<int32, [4]> var_22654_begin_0 = const()[name = tensor<string, []>("op_22654_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_22654_end_0 = const()[name = tensor<string, []>("op_22654_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_22654_end_mask_0 = const()[name = tensor<string, []>("op_22654_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22654_cast = slice_by_index(begin = var_22654_begin_0, end = var_22654_end_0, end_mask = var_22654_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22654_cast")];
            tensor<int32, [4]> var_22658_begin_0 = const()[name = tensor<string, []>("op_22658_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_22658_end_0 = const()[name = tensor<string, []>("op_22658_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_22658_end_mask_0 = const()[name = tensor<string, []>("op_22658_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_22658_cast = slice_by_index(begin = var_22658_begin_0, end = var_22658_end_0, end_mask = var_22658_end_mask_0, x = q_103_cast)[name = tensor<string, []>("op_22658_cast")];
            tensor<int32, [4]> k_207_perm_0 = const()[name = tensor<string, []>("k_207_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_22665_begin_0 = const()[name = tensor<string, []>("op_22665_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_22665_end_0 = const()[name = tensor<string, []>("op_22665_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_22665_end_mask_0 = const()[name = tensor<string, []>("op_22665_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_88 = transpose(perm = k_207_perm_0, x = k_205_cast)[name = tensor<string, []>("transpose_88")];
            tensor<fp16, [2, 77, 1, 64]> var_22665_cast = slice_by_index(begin = var_22665_begin_0, end = var_22665_end_0, end_mask = var_22665_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22665_cast")];
            tensor<int32, [4]> var_22669_begin_0 = const()[name = tensor<string, []>("op_22669_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_22669_end_0 = const()[name = tensor<string, []>("op_22669_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_22669_end_mask_0 = const()[name = tensor<string, []>("op_22669_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_22669_cast = slice_by_index(begin = var_22669_begin_0, end = var_22669_end_0, end_mask = var_22669_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22669_cast")];
            tensor<int32, [4]> var_22673_begin_0 = const()[name = tensor<string, []>("op_22673_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_22673_end_0 = const()[name = tensor<string, []>("op_22673_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_22673_end_mask_0 = const()[name = tensor<string, []>("op_22673_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_22673_cast = slice_by_index(begin = var_22673_begin_0, end = var_22673_end_0, end_mask = var_22673_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22673_cast")];
            tensor<int32, [4]> var_22677_begin_0 = const()[name = tensor<string, []>("op_22677_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_22677_end_0 = const()[name = tensor<string, []>("op_22677_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_22677_end_mask_0 = const()[name = tensor<string, []>("op_22677_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_22677_cast = slice_by_index(begin = var_22677_begin_0, end = var_22677_end_0, end_mask = var_22677_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22677_cast")];
            tensor<int32, [4]> var_22681_begin_0 = const()[name = tensor<string, []>("op_22681_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_22681_end_0 = const()[name = tensor<string, []>("op_22681_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_22681_end_mask_0 = const()[name = tensor<string, []>("op_22681_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_22681_cast = slice_by_index(begin = var_22681_begin_0, end = var_22681_end_0, end_mask = var_22681_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22681_cast")];
            tensor<int32, [4]> var_22685_begin_0 = const()[name = tensor<string, []>("op_22685_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_22685_end_0 = const()[name = tensor<string, []>("op_22685_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_22685_end_mask_0 = const()[name = tensor<string, []>("op_22685_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_22685_cast = slice_by_index(begin = var_22685_begin_0, end = var_22685_end_0, end_mask = var_22685_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22685_cast")];
            tensor<int32, [4]> var_22689_begin_0 = const()[name = tensor<string, []>("op_22689_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_22689_end_0 = const()[name = tensor<string, []>("op_22689_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_22689_end_mask_0 = const()[name = tensor<string, []>("op_22689_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_22689_cast = slice_by_index(begin = var_22689_begin_0, end = var_22689_end_0, end_mask = var_22689_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22689_cast")];
            tensor<int32, [4]> var_22693_begin_0 = const()[name = tensor<string, []>("op_22693_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_22693_end_0 = const()[name = tensor<string, []>("op_22693_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_22693_end_mask_0 = const()[name = tensor<string, []>("op_22693_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_22693_cast = slice_by_index(begin = var_22693_begin_0, end = var_22693_end_0, end_mask = var_22693_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22693_cast")];
            tensor<int32, [4]> var_22697_begin_0 = const()[name = tensor<string, []>("op_22697_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_22697_end_0 = const()[name = tensor<string, []>("op_22697_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_22697_end_mask_0 = const()[name = tensor<string, []>("op_22697_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_22697_cast = slice_by_index(begin = var_22697_begin_0, end = var_22697_end_0, end_mask = var_22697_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22697_cast")];
            tensor<int32, [4]> var_22701_begin_0 = const()[name = tensor<string, []>("op_22701_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_22701_end_0 = const()[name = tensor<string, []>("op_22701_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_22701_end_mask_0 = const()[name = tensor<string, []>("op_22701_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_22701_cast = slice_by_index(begin = var_22701_begin_0, end = var_22701_end_0, end_mask = var_22701_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22701_cast")];
            tensor<int32, [4]> var_22705_begin_0 = const()[name = tensor<string, []>("op_22705_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_22705_end_0 = const()[name = tensor<string, []>("op_22705_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_22705_end_mask_0 = const()[name = tensor<string, []>("op_22705_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_22705_cast = slice_by_index(begin = var_22705_begin_0, end = var_22705_end_0, end_mask = var_22705_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22705_cast")];
            tensor<int32, [4]> var_22709_begin_0 = const()[name = tensor<string, []>("op_22709_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_22709_end_0 = const()[name = tensor<string, []>("op_22709_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_22709_end_mask_0 = const()[name = tensor<string, []>("op_22709_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_22709_cast = slice_by_index(begin = var_22709_begin_0, end = var_22709_end_0, end_mask = var_22709_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22709_cast")];
            tensor<int32, [4]> var_22713_begin_0 = const()[name = tensor<string, []>("op_22713_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_22713_end_0 = const()[name = tensor<string, []>("op_22713_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_22713_end_mask_0 = const()[name = tensor<string, []>("op_22713_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_22713_cast = slice_by_index(begin = var_22713_begin_0, end = var_22713_end_0, end_mask = var_22713_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22713_cast")];
            tensor<int32, [4]> var_22717_begin_0 = const()[name = tensor<string, []>("op_22717_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_22717_end_0 = const()[name = tensor<string, []>("op_22717_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_22717_end_mask_0 = const()[name = tensor<string, []>("op_22717_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_22717_cast = slice_by_index(begin = var_22717_begin_0, end = var_22717_end_0, end_mask = var_22717_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22717_cast")];
            tensor<int32, [4]> var_22721_begin_0 = const()[name = tensor<string, []>("op_22721_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_22721_end_0 = const()[name = tensor<string, []>("op_22721_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_22721_end_mask_0 = const()[name = tensor<string, []>("op_22721_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_22721_cast = slice_by_index(begin = var_22721_begin_0, end = var_22721_end_0, end_mask = var_22721_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22721_cast")];
            tensor<int32, [4]> var_22725_begin_0 = const()[name = tensor<string, []>("op_22725_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_22725_end_0 = const()[name = tensor<string, []>("op_22725_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_22725_end_mask_0 = const()[name = tensor<string, []>("op_22725_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_22725_cast = slice_by_index(begin = var_22725_begin_0, end = var_22725_end_0, end_mask = var_22725_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22725_cast")];
            tensor<int32, [4]> var_22729_begin_0 = const()[name = tensor<string, []>("op_22729_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_22729_end_0 = const()[name = tensor<string, []>("op_22729_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_22729_end_mask_0 = const()[name = tensor<string, []>("op_22729_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_22729_cast = slice_by_index(begin = var_22729_begin_0, end = var_22729_end_0, end_mask = var_22729_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22729_cast")];
            tensor<int32, [4]> var_22733_begin_0 = const()[name = tensor<string, []>("op_22733_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_22733_end_0 = const()[name = tensor<string, []>("op_22733_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_22733_end_mask_0 = const()[name = tensor<string, []>("op_22733_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_22733_cast = slice_by_index(begin = var_22733_begin_0, end = var_22733_end_0, end_mask = var_22733_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22733_cast")];
            tensor<int32, [4]> var_22737_begin_0 = const()[name = tensor<string, []>("op_22737_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_22737_end_0 = const()[name = tensor<string, []>("op_22737_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_22737_end_mask_0 = const()[name = tensor<string, []>("op_22737_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_22737_cast = slice_by_index(begin = var_22737_begin_0, end = var_22737_end_0, end_mask = var_22737_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22737_cast")];
            tensor<int32, [4]> var_22741_begin_0 = const()[name = tensor<string, []>("op_22741_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_22741_end_0 = const()[name = tensor<string, []>("op_22741_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_22741_end_mask_0 = const()[name = tensor<string, []>("op_22741_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_22741_cast = slice_by_index(begin = var_22741_begin_0, end = var_22741_end_0, end_mask = var_22741_end_mask_0, x = transpose_88)[name = tensor<string, []>("op_22741_cast")];
            tensor<int32, [4]> var_22743_begin_0 = const()[name = tensor<string, []>("op_22743_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_22743_end_0 = const()[name = tensor<string, []>("op_22743_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_22743_end_mask_0 = const()[name = tensor<string, []>("op_22743_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22743_cast = slice_by_index(begin = var_22743_begin_0, end = var_22743_end_0, end_mask = var_22743_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22743_cast")];
            tensor<int32, [4]> var_22747_begin_0 = const()[name = tensor<string, []>("op_22747_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_22747_end_0 = const()[name = tensor<string, []>("op_22747_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_22747_end_mask_0 = const()[name = tensor<string, []>("op_22747_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22747_cast = slice_by_index(begin = var_22747_begin_0, end = var_22747_end_0, end_mask = var_22747_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22747_cast")];
            tensor<int32, [4]> var_22751_begin_0 = const()[name = tensor<string, []>("op_22751_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_22751_end_0 = const()[name = tensor<string, []>("op_22751_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_22751_end_mask_0 = const()[name = tensor<string, []>("op_22751_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22751_cast = slice_by_index(begin = var_22751_begin_0, end = var_22751_end_0, end_mask = var_22751_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22751_cast")];
            tensor<int32, [4]> var_22755_begin_0 = const()[name = tensor<string, []>("op_22755_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_22755_end_0 = const()[name = tensor<string, []>("op_22755_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_22755_end_mask_0 = const()[name = tensor<string, []>("op_22755_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22755_cast = slice_by_index(begin = var_22755_begin_0, end = var_22755_end_0, end_mask = var_22755_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22755_cast")];
            tensor<int32, [4]> var_22759_begin_0 = const()[name = tensor<string, []>("op_22759_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_22759_end_0 = const()[name = tensor<string, []>("op_22759_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_22759_end_mask_0 = const()[name = tensor<string, []>("op_22759_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22759_cast = slice_by_index(begin = var_22759_begin_0, end = var_22759_end_0, end_mask = var_22759_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22759_cast")];
            tensor<int32, [4]> var_22763_begin_0 = const()[name = tensor<string, []>("op_22763_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_22763_end_0 = const()[name = tensor<string, []>("op_22763_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_22763_end_mask_0 = const()[name = tensor<string, []>("op_22763_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22763_cast = slice_by_index(begin = var_22763_begin_0, end = var_22763_end_0, end_mask = var_22763_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22763_cast")];
            tensor<int32, [4]> var_22767_begin_0 = const()[name = tensor<string, []>("op_22767_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_22767_end_0 = const()[name = tensor<string, []>("op_22767_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_22767_end_mask_0 = const()[name = tensor<string, []>("op_22767_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22767_cast = slice_by_index(begin = var_22767_begin_0, end = var_22767_end_0, end_mask = var_22767_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22767_cast")];
            tensor<int32, [4]> var_22771_begin_0 = const()[name = tensor<string, []>("op_22771_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_22771_end_0 = const()[name = tensor<string, []>("op_22771_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_22771_end_mask_0 = const()[name = tensor<string, []>("op_22771_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22771_cast = slice_by_index(begin = var_22771_begin_0, end = var_22771_end_0, end_mask = var_22771_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22771_cast")];
            tensor<int32, [4]> var_22775_begin_0 = const()[name = tensor<string, []>("op_22775_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_22775_end_0 = const()[name = tensor<string, []>("op_22775_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_22775_end_mask_0 = const()[name = tensor<string, []>("op_22775_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22775_cast = slice_by_index(begin = var_22775_begin_0, end = var_22775_end_0, end_mask = var_22775_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22775_cast")];
            tensor<int32, [4]> var_22779_begin_0 = const()[name = tensor<string, []>("op_22779_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_22779_end_0 = const()[name = tensor<string, []>("op_22779_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_22779_end_mask_0 = const()[name = tensor<string, []>("op_22779_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22779_cast = slice_by_index(begin = var_22779_begin_0, end = var_22779_end_0, end_mask = var_22779_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22779_cast")];
            tensor<int32, [4]> var_22783_begin_0 = const()[name = tensor<string, []>("op_22783_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_22783_end_0 = const()[name = tensor<string, []>("op_22783_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_22783_end_mask_0 = const()[name = tensor<string, []>("op_22783_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22783_cast = slice_by_index(begin = var_22783_begin_0, end = var_22783_end_0, end_mask = var_22783_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22783_cast")];
            tensor<int32, [4]> var_22787_begin_0 = const()[name = tensor<string, []>("op_22787_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_22787_end_0 = const()[name = tensor<string, []>("op_22787_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_22787_end_mask_0 = const()[name = tensor<string, []>("op_22787_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22787_cast = slice_by_index(begin = var_22787_begin_0, end = var_22787_end_0, end_mask = var_22787_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22787_cast")];
            tensor<int32, [4]> var_22791_begin_0 = const()[name = tensor<string, []>("op_22791_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_22791_end_0 = const()[name = tensor<string, []>("op_22791_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_22791_end_mask_0 = const()[name = tensor<string, []>("op_22791_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22791_cast = slice_by_index(begin = var_22791_begin_0, end = var_22791_end_0, end_mask = var_22791_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22791_cast")];
            tensor<int32, [4]> var_22795_begin_0 = const()[name = tensor<string, []>("op_22795_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_22795_end_0 = const()[name = tensor<string, []>("op_22795_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_22795_end_mask_0 = const()[name = tensor<string, []>("op_22795_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22795_cast = slice_by_index(begin = var_22795_begin_0, end = var_22795_end_0, end_mask = var_22795_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22795_cast")];
            tensor<int32, [4]> var_22799_begin_0 = const()[name = tensor<string, []>("op_22799_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_22799_end_0 = const()[name = tensor<string, []>("op_22799_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_22799_end_mask_0 = const()[name = tensor<string, []>("op_22799_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22799_cast = slice_by_index(begin = var_22799_begin_0, end = var_22799_end_0, end_mask = var_22799_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22799_cast")];
            tensor<int32, [4]> var_22803_begin_0 = const()[name = tensor<string, []>("op_22803_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_22803_end_0 = const()[name = tensor<string, []>("op_22803_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_22803_end_mask_0 = const()[name = tensor<string, []>("op_22803_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22803_cast = slice_by_index(begin = var_22803_begin_0, end = var_22803_end_0, end_mask = var_22803_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22803_cast")];
            tensor<int32, [4]> var_22807_begin_0 = const()[name = tensor<string, []>("op_22807_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_22807_end_0 = const()[name = tensor<string, []>("op_22807_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_22807_end_mask_0 = const()[name = tensor<string, []>("op_22807_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22807_cast = slice_by_index(begin = var_22807_begin_0, end = var_22807_end_0, end_mask = var_22807_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22807_cast")];
            tensor<int32, [4]> var_22811_begin_0 = const()[name = tensor<string, []>("op_22811_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_22811_end_0 = const()[name = tensor<string, []>("op_22811_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_22811_end_mask_0 = const()[name = tensor<string, []>("op_22811_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22811_cast = slice_by_index(begin = var_22811_begin_0, end = var_22811_end_0, end_mask = var_22811_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22811_cast")];
            tensor<int32, [4]> var_22815_begin_0 = const()[name = tensor<string, []>("op_22815_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_22815_end_0 = const()[name = tensor<string, []>("op_22815_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_22815_end_mask_0 = const()[name = tensor<string, []>("op_22815_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22815_cast = slice_by_index(begin = var_22815_begin_0, end = var_22815_end_0, end_mask = var_22815_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22815_cast")];
            tensor<int32, [4]> var_22819_begin_0 = const()[name = tensor<string, []>("op_22819_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_22819_end_0 = const()[name = tensor<string, []>("op_22819_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_22819_end_mask_0 = const()[name = tensor<string, []>("op_22819_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_22819_cast = slice_by_index(begin = var_22819_begin_0, end = var_22819_end_0, end_mask = var_22819_end_mask_0, x = v_103_cast)[name = tensor<string, []>("op_22819_cast")];
            tensor<string, []> var_22823_equation_0 = const()[name = tensor<string, []>("op_22823_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22823_cast = einsum(equation = var_22823_equation_0, values = (var_22665_cast, var_22582_cast))[name = tensor<string, []>("op_22823_cast")];
            tensor<fp16, []> var_22824_to_fp16 = const()[name = tensor<string, []>("op_22824_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1881_cast = mul(x = var_22823_cast, y = var_22824_to_fp16)[name = tensor<string, []>("aw_1881_cast")];
            tensor<string, []> var_22827_equation_0 = const()[name = tensor<string, []>("op_22827_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22827_cast = einsum(equation = var_22827_equation_0, values = (var_22669_cast, var_22586_cast))[name = tensor<string, []>("op_22827_cast")];
            tensor<fp16, []> var_22828_to_fp16 = const()[name = tensor<string, []>("op_22828_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1883_cast = mul(x = var_22827_cast, y = var_22828_to_fp16)[name = tensor<string, []>("aw_1883_cast")];
            tensor<string, []> var_22831_equation_0 = const()[name = tensor<string, []>("op_22831_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22831_cast = einsum(equation = var_22831_equation_0, values = (var_22673_cast, var_22590_cast))[name = tensor<string, []>("op_22831_cast")];
            tensor<fp16, []> var_22832_to_fp16 = const()[name = tensor<string, []>("op_22832_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1885_cast = mul(x = var_22831_cast, y = var_22832_to_fp16)[name = tensor<string, []>("aw_1885_cast")];
            tensor<string, []> var_22835_equation_0 = const()[name = tensor<string, []>("op_22835_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22835_cast = einsum(equation = var_22835_equation_0, values = (var_22677_cast, var_22594_cast))[name = tensor<string, []>("op_22835_cast")];
            tensor<fp16, []> var_22836_to_fp16 = const()[name = tensor<string, []>("op_22836_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1887_cast = mul(x = var_22835_cast, y = var_22836_to_fp16)[name = tensor<string, []>("aw_1887_cast")];
            tensor<string, []> var_22839_equation_0 = const()[name = tensor<string, []>("op_22839_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22839_cast = einsum(equation = var_22839_equation_0, values = (var_22681_cast, var_22598_cast))[name = tensor<string, []>("op_22839_cast")];
            tensor<fp16, []> var_22840_to_fp16 = const()[name = tensor<string, []>("op_22840_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1889_cast = mul(x = var_22839_cast, y = var_22840_to_fp16)[name = tensor<string, []>("aw_1889_cast")];
            tensor<string, []> var_22843_equation_0 = const()[name = tensor<string, []>("op_22843_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22843_cast = einsum(equation = var_22843_equation_0, values = (var_22685_cast, var_22602_cast))[name = tensor<string, []>("op_22843_cast")];
            tensor<fp16, []> var_22844_to_fp16 = const()[name = tensor<string, []>("op_22844_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1891_cast = mul(x = var_22843_cast, y = var_22844_to_fp16)[name = tensor<string, []>("aw_1891_cast")];
            tensor<string, []> var_22847_equation_0 = const()[name = tensor<string, []>("op_22847_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22847_cast = einsum(equation = var_22847_equation_0, values = (var_22689_cast, var_22606_cast))[name = tensor<string, []>("op_22847_cast")];
            tensor<fp16, []> var_22848_to_fp16 = const()[name = tensor<string, []>("op_22848_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1893_cast = mul(x = var_22847_cast, y = var_22848_to_fp16)[name = tensor<string, []>("aw_1893_cast")];
            tensor<string, []> var_22851_equation_0 = const()[name = tensor<string, []>("op_22851_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22851_cast = einsum(equation = var_22851_equation_0, values = (var_22693_cast, var_22610_cast))[name = tensor<string, []>("op_22851_cast")];
            tensor<fp16, []> var_22852_to_fp16 = const()[name = tensor<string, []>("op_22852_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1895_cast = mul(x = var_22851_cast, y = var_22852_to_fp16)[name = tensor<string, []>("aw_1895_cast")];
            tensor<string, []> var_22855_equation_0 = const()[name = tensor<string, []>("op_22855_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22855_cast = einsum(equation = var_22855_equation_0, values = (var_22697_cast, var_22614_cast))[name = tensor<string, []>("op_22855_cast")];
            tensor<fp16, []> var_22856_to_fp16 = const()[name = tensor<string, []>("op_22856_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1897_cast = mul(x = var_22855_cast, y = var_22856_to_fp16)[name = tensor<string, []>("aw_1897_cast")];
            tensor<string, []> var_22859_equation_0 = const()[name = tensor<string, []>("op_22859_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22859_cast = einsum(equation = var_22859_equation_0, values = (var_22701_cast, var_22618_cast))[name = tensor<string, []>("op_22859_cast")];
            tensor<fp16, []> var_22860_to_fp16 = const()[name = tensor<string, []>("op_22860_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1899_cast = mul(x = var_22859_cast, y = var_22860_to_fp16)[name = tensor<string, []>("aw_1899_cast")];
            tensor<string, []> var_22863_equation_0 = const()[name = tensor<string, []>("op_22863_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22863_cast = einsum(equation = var_22863_equation_0, values = (var_22705_cast, var_22622_cast))[name = tensor<string, []>("op_22863_cast")];
            tensor<fp16, []> var_22864_to_fp16 = const()[name = tensor<string, []>("op_22864_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1901_cast = mul(x = var_22863_cast, y = var_22864_to_fp16)[name = tensor<string, []>("aw_1901_cast")];
            tensor<string, []> var_22867_equation_0 = const()[name = tensor<string, []>("op_22867_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22867_cast = einsum(equation = var_22867_equation_0, values = (var_22709_cast, var_22626_cast))[name = tensor<string, []>("op_22867_cast")];
            tensor<fp16, []> var_22868_to_fp16 = const()[name = tensor<string, []>("op_22868_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1903_cast = mul(x = var_22867_cast, y = var_22868_to_fp16)[name = tensor<string, []>("aw_1903_cast")];
            tensor<string, []> var_22871_equation_0 = const()[name = tensor<string, []>("op_22871_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22871_cast = einsum(equation = var_22871_equation_0, values = (var_22713_cast, var_22630_cast))[name = tensor<string, []>("op_22871_cast")];
            tensor<fp16, []> var_22872_to_fp16 = const()[name = tensor<string, []>("op_22872_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1905_cast = mul(x = var_22871_cast, y = var_22872_to_fp16)[name = tensor<string, []>("aw_1905_cast")];
            tensor<string, []> var_22875_equation_0 = const()[name = tensor<string, []>("op_22875_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22875_cast = einsum(equation = var_22875_equation_0, values = (var_22717_cast, var_22634_cast))[name = tensor<string, []>("op_22875_cast")];
            tensor<fp16, []> var_22876_to_fp16 = const()[name = tensor<string, []>("op_22876_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1907_cast = mul(x = var_22875_cast, y = var_22876_to_fp16)[name = tensor<string, []>("aw_1907_cast")];
            tensor<string, []> var_22879_equation_0 = const()[name = tensor<string, []>("op_22879_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22879_cast = einsum(equation = var_22879_equation_0, values = (var_22721_cast, var_22638_cast))[name = tensor<string, []>("op_22879_cast")];
            tensor<fp16, []> var_22880_to_fp16 = const()[name = tensor<string, []>("op_22880_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1909_cast = mul(x = var_22879_cast, y = var_22880_to_fp16)[name = tensor<string, []>("aw_1909_cast")];
            tensor<string, []> var_22883_equation_0 = const()[name = tensor<string, []>("op_22883_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22883_cast = einsum(equation = var_22883_equation_0, values = (var_22725_cast, var_22642_cast))[name = tensor<string, []>("op_22883_cast")];
            tensor<fp16, []> var_22884_to_fp16 = const()[name = tensor<string, []>("op_22884_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1911_cast = mul(x = var_22883_cast, y = var_22884_to_fp16)[name = tensor<string, []>("aw_1911_cast")];
            tensor<string, []> var_22887_equation_0 = const()[name = tensor<string, []>("op_22887_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22887_cast = einsum(equation = var_22887_equation_0, values = (var_22729_cast, var_22646_cast))[name = tensor<string, []>("op_22887_cast")];
            tensor<fp16, []> var_22888_to_fp16 = const()[name = tensor<string, []>("op_22888_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1913_cast = mul(x = var_22887_cast, y = var_22888_to_fp16)[name = tensor<string, []>("aw_1913_cast")];
            tensor<string, []> var_22891_equation_0 = const()[name = tensor<string, []>("op_22891_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22891_cast = einsum(equation = var_22891_equation_0, values = (var_22733_cast, var_22650_cast))[name = tensor<string, []>("op_22891_cast")];
            tensor<fp16, []> var_22892_to_fp16 = const()[name = tensor<string, []>("op_22892_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1915_cast = mul(x = var_22891_cast, y = var_22892_to_fp16)[name = tensor<string, []>("aw_1915_cast")];
            tensor<string, []> var_22895_equation_0 = const()[name = tensor<string, []>("op_22895_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22895_cast = einsum(equation = var_22895_equation_0, values = (var_22737_cast, var_22654_cast))[name = tensor<string, []>("op_22895_cast")];
            tensor<fp16, []> var_22896_to_fp16 = const()[name = tensor<string, []>("op_22896_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1917_cast = mul(x = var_22895_cast, y = var_22896_to_fp16)[name = tensor<string, []>("aw_1917_cast")];
            tensor<string, []> var_22899_equation_0 = const()[name = tensor<string, []>("op_22899_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_22899_cast = einsum(equation = var_22899_equation_0, values = (var_22741_cast, var_22658_cast))[name = tensor<string, []>("op_22899_cast")];
            tensor<fp16, []> var_22900_to_fp16 = const()[name = tensor<string, []>("op_22900_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1919_cast = mul(x = var_22899_cast, y = var_22900_to_fp16)[name = tensor<string, []>("aw_1919_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22902_cast = softmax(axis = var_21105, x = aw_1881_cast)[name = tensor<string, []>("op_22902_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22903_cast = softmax(axis = var_21105, x = aw_1883_cast)[name = tensor<string, []>("op_22903_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22904_cast = softmax(axis = var_21105, x = aw_1885_cast)[name = tensor<string, []>("op_22904_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22905_cast = softmax(axis = var_21105, x = aw_1887_cast)[name = tensor<string, []>("op_22905_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22906_cast = softmax(axis = var_21105, x = aw_1889_cast)[name = tensor<string, []>("op_22906_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22907_cast = softmax(axis = var_21105, x = aw_1891_cast)[name = tensor<string, []>("op_22907_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22908_cast = softmax(axis = var_21105, x = aw_1893_cast)[name = tensor<string, []>("op_22908_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22909_cast = softmax(axis = var_21105, x = aw_1895_cast)[name = tensor<string, []>("op_22909_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22910_cast = softmax(axis = var_21105, x = aw_1897_cast)[name = tensor<string, []>("op_22910_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22911_cast = softmax(axis = var_21105, x = aw_1899_cast)[name = tensor<string, []>("op_22911_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22912_cast = softmax(axis = var_21105, x = aw_1901_cast)[name = tensor<string, []>("op_22912_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22913_cast = softmax(axis = var_21105, x = aw_1903_cast)[name = tensor<string, []>("op_22913_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22914_cast = softmax(axis = var_21105, x = aw_1905_cast)[name = tensor<string, []>("op_22914_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22915_cast = softmax(axis = var_21105, x = aw_1907_cast)[name = tensor<string, []>("op_22915_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22916_cast = softmax(axis = var_21105, x = aw_1909_cast)[name = tensor<string, []>("op_22916_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22917_cast = softmax(axis = var_21105, x = aw_1911_cast)[name = tensor<string, []>("op_22917_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22918_cast = softmax(axis = var_21105, x = aw_1913_cast)[name = tensor<string, []>("op_22918_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22919_cast = softmax(axis = var_21105, x = aw_1915_cast)[name = tensor<string, []>("op_22919_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22920_cast = softmax(axis = var_21105, x = aw_1917_cast)[name = tensor<string, []>("op_22920_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_22921_cast = softmax(axis = var_21105, x = aw_1919_cast)[name = tensor<string, []>("op_22921_cast")];
            tensor<string, []> var_22923_equation_0 = const()[name = tensor<string, []>("op_22923_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22923_cast = einsum(equation = var_22923_equation_0, values = (var_22743_cast, var_22902_cast))[name = tensor<string, []>("op_22923_cast")];
            tensor<string, []> var_22925_equation_0 = const()[name = tensor<string, []>("op_22925_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22925_cast = einsum(equation = var_22925_equation_0, values = (var_22747_cast, var_22903_cast))[name = tensor<string, []>("op_22925_cast")];
            tensor<string, []> var_22927_equation_0 = const()[name = tensor<string, []>("op_22927_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22927_cast = einsum(equation = var_22927_equation_0, values = (var_22751_cast, var_22904_cast))[name = tensor<string, []>("op_22927_cast")];
            tensor<string, []> var_22929_equation_0 = const()[name = tensor<string, []>("op_22929_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22929_cast = einsum(equation = var_22929_equation_0, values = (var_22755_cast, var_22905_cast))[name = tensor<string, []>("op_22929_cast")];
            tensor<string, []> var_22931_equation_0 = const()[name = tensor<string, []>("op_22931_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22931_cast = einsum(equation = var_22931_equation_0, values = (var_22759_cast, var_22906_cast))[name = tensor<string, []>("op_22931_cast")];
            tensor<string, []> var_22933_equation_0 = const()[name = tensor<string, []>("op_22933_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22933_cast = einsum(equation = var_22933_equation_0, values = (var_22763_cast, var_22907_cast))[name = tensor<string, []>("op_22933_cast")];
            tensor<string, []> var_22935_equation_0 = const()[name = tensor<string, []>("op_22935_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22935_cast = einsum(equation = var_22935_equation_0, values = (var_22767_cast, var_22908_cast))[name = tensor<string, []>("op_22935_cast")];
            tensor<string, []> var_22937_equation_0 = const()[name = tensor<string, []>("op_22937_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22937_cast = einsum(equation = var_22937_equation_0, values = (var_22771_cast, var_22909_cast))[name = tensor<string, []>("op_22937_cast")];
            tensor<string, []> var_22939_equation_0 = const()[name = tensor<string, []>("op_22939_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22939_cast = einsum(equation = var_22939_equation_0, values = (var_22775_cast, var_22910_cast))[name = tensor<string, []>("op_22939_cast")];
            tensor<string, []> var_22941_equation_0 = const()[name = tensor<string, []>("op_22941_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22941_cast = einsum(equation = var_22941_equation_0, values = (var_22779_cast, var_22911_cast))[name = tensor<string, []>("op_22941_cast")];
            tensor<string, []> var_22943_equation_0 = const()[name = tensor<string, []>("op_22943_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22943_cast = einsum(equation = var_22943_equation_0, values = (var_22783_cast, var_22912_cast))[name = tensor<string, []>("op_22943_cast")];
            tensor<string, []> var_22945_equation_0 = const()[name = tensor<string, []>("op_22945_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22945_cast = einsum(equation = var_22945_equation_0, values = (var_22787_cast, var_22913_cast))[name = tensor<string, []>("op_22945_cast")];
            tensor<string, []> var_22947_equation_0 = const()[name = tensor<string, []>("op_22947_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22947_cast = einsum(equation = var_22947_equation_0, values = (var_22791_cast, var_22914_cast))[name = tensor<string, []>("op_22947_cast")];
            tensor<string, []> var_22949_equation_0 = const()[name = tensor<string, []>("op_22949_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22949_cast = einsum(equation = var_22949_equation_0, values = (var_22795_cast, var_22915_cast))[name = tensor<string, []>("op_22949_cast")];
            tensor<string, []> var_22951_equation_0 = const()[name = tensor<string, []>("op_22951_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22951_cast = einsum(equation = var_22951_equation_0, values = (var_22799_cast, var_22916_cast))[name = tensor<string, []>("op_22951_cast")];
            tensor<string, []> var_22953_equation_0 = const()[name = tensor<string, []>("op_22953_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22953_cast = einsum(equation = var_22953_equation_0, values = (var_22803_cast, var_22917_cast))[name = tensor<string, []>("op_22953_cast")];
            tensor<string, []> var_22955_equation_0 = const()[name = tensor<string, []>("op_22955_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22955_cast = einsum(equation = var_22955_equation_0, values = (var_22807_cast, var_22918_cast))[name = tensor<string, []>("op_22955_cast")];
            tensor<string, []> var_22957_equation_0 = const()[name = tensor<string, []>("op_22957_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22957_cast = einsum(equation = var_22957_equation_0, values = (var_22811_cast, var_22919_cast))[name = tensor<string, []>("op_22957_cast")];
            tensor<string, []> var_22959_equation_0 = const()[name = tensor<string, []>("op_22959_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22959_cast = einsum(equation = var_22959_equation_0, values = (var_22815_cast, var_22920_cast))[name = tensor<string, []>("op_22959_cast")];
            tensor<string, []> var_22961_equation_0 = const()[name = tensor<string, []>("op_22961_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_22961_cast = einsum(equation = var_22961_equation_0, values = (var_22819_cast, var_22921_cast))[name = tensor<string, []>("op_22961_cast")];
            tensor<bool, []> input_339_interleave_0 = const()[name = tensor<string, []>("input_339_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_339_cast = concat(axis = var_21105, interleave = input_339_interleave_0, values = (var_22923_cast, var_22925_cast, var_22927_cast, var_22929_cast, var_22931_cast, var_22933_cast, var_22935_cast, var_22937_cast, var_22939_cast, var_22941_cast, var_22943_cast, var_22945_cast, var_22947_cast, var_22949_cast, var_22951_cast, var_22953_cast, var_22955_cast, var_22957_cast, var_22959_cast, var_22961_cast))[name = tensor<string, []>("input_339_cast")];
            tensor<int32, [2]> var_22967 = const()[name = tensor<string, []>("op_22967"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_22969 = const()[name = tensor<string, []>("op_22969"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_22971_pad_type_0 = const()[name = tensor<string, []>("op_22971_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_22971_pad_0 = const()[name = tensor<string, []>("op_22971_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1822505344)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1825782208)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_22971_cast = conv(bias = mid_block_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_22969, groups = var_21105, pad = var_22971_pad_0, pad_type = var_22971_pad_type_0, strides = var_22967, weight = mid_block_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_339_cast)[name = tensor<string, []>("op_22971_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_155_cast = add(x = var_22971_cast, y = inputs_153_cast)[name = tensor<string, []>("inputs_155_cast")];
            tensor<int32, [1]> var_22975 = const()[name = tensor<string, []>("op_22975"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_155_cast = reduce_mean(axes = var_22975, keep_dims = var_21100, x = inputs_155_cast)[name = tensor<string, []>("channels_mean_155_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_155_cast = sub(x = inputs_155_cast, y = channels_mean_155_cast)[name = tensor<string, []>("zero_mean_155_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_155_cast = mul(x = zero_mean_155_cast, y = zero_mean_155_cast)[name = tensor<string, []>("zero_mean_sq_155_cast")];
            tensor<int32, [1]> var_22979 = const()[name = tensor<string, []>("op_22979"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_22980_cast = reduce_mean(axes = var_22979, keep_dims = var_21100, x = zero_mean_sq_155_cast)[name = tensor<string, []>("op_22980_cast")];
            tensor<fp16, []> var_22981_to_fp16 = const()[name = tensor<string, []>("op_22981_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_22982_cast = add(x = var_22980_cast, y = var_22981_to_fp16)[name = tensor<string, []>("op_22982_cast")];
            tensor<fp16, []> denom_155_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_155_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_155_cast = rsqrt(epsilon = denom_155_epsilon_0_to_fp16, x = var_22982_cast)[name = tensor<string, []>("denom_155_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_155_cast = mul(x = zero_mean_155_cast, y = denom_155_cast)[name = tensor<string, []>("out_155_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_22986_to_fp16 = const()[name = tensor<string, []>("op_22986_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1825784832)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_22987_cast = add(x = out_155_cast, y = var_22986_to_fp16)[name = tensor<string, []>("op_22987_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_22989_to_fp16 = const()[name = tensor<string, []>("op_22989_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1825787456)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_341_cast = mul(x = var_22987_cast, y = var_22989_to_fp16)[name = tensor<string, []>("input_341_cast")];
            tensor<int32, [2]> var_22997 = const()[name = tensor<string, []>("op_22997"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_22999 = const()[name = tensor<string, []>("op_22999"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_23001_pad_type_0 = const()[name = tensor<string, []>("op_23001_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_23001_pad_0 = const()[name = tensor<string, []>("op_23001_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1825790080)))];
            tensor<fp16, [10240]> mid_block_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1852004544)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_23001_cast = conv(bias = mid_block_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_22999, groups = var_21105, pad = var_23001_pad_0, pad_type = var_23001_pad_type_0, strides = var_22997, weight = mid_block_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_341_cast)[name = tensor<string, []>("op_23001_cast")];
            tensor<int32, [2]> var_23002_split_sizes_0 = const()[name = tensor<string, []>("op_23002_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_23002_axis_0 = const()[name = tensor<string, []>("op_23002_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_23002_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_23002_cast_1 = split(axis = var_23002_axis_0, split_sizes = var_23002_split_sizes_0, x = var_23001_cast)[name = tensor<string, []>("op_23002_cast")];
            tensor<string, []> var_23004_mode_0 = const()[name = tensor<string, []>("op_23004_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_23004_cast = gelu(mode = var_23004_mode_0, x = var_23002_cast_1)[name = tensor<string, []>("op_23004_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_343_cast = mul(x = var_23002_cast_0, y = var_23004_cast)[name = tensor<string, []>("input_343_cast")];
            tensor<int32, [2]> var_23008 = const()[name = tensor<string, []>("op_23008"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_23010 = const()[name = tensor<string, []>("op_23010"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_23012_pad_type_0 = const()[name = tensor<string, []>("op_23012_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_23012_pad_0 = const()[name = tensor<string, []>("op_23012_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> mid_block_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1852025088)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1865132352)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_23012_cast = conv(bias = mid_block_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_23010, groups = var_21105, pad = var_23012_pad_0, pad_type = var_23012_pad_type_0, strides = var_23008, weight = mid_block_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_343_cast)[name = tensor<string, []>("op_23012_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_157_cast = add(x = var_23012_cast, y = inputs_155_cast)[name = tensor<string, []>("inputs_157_cast")];
            tensor<int32, [1]> var_23022 = const()[name = tensor<string, []>("op_23022"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_157_cast = reduce_mean(axes = var_23022, keep_dims = var_21100, x = inputs_157_cast)[name = tensor<string, []>("channels_mean_157_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_157_cast = sub(x = inputs_157_cast, y = channels_mean_157_cast)[name = tensor<string, []>("zero_mean_157_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_157_cast = mul(x = zero_mean_157_cast, y = zero_mean_157_cast)[name = tensor<string, []>("zero_mean_sq_157_cast")];
            tensor<int32, [1]> var_23026 = const()[name = tensor<string, []>("op_23026"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_23027_cast = reduce_mean(axes = var_23026, keep_dims = var_21100, x = zero_mean_sq_157_cast)[name = tensor<string, []>("op_23027_cast")];
            tensor<fp16, []> var_23028_to_fp16 = const()[name = tensor<string, []>("op_23028_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_23029_cast = add(x = var_23027_cast, y = var_23028_to_fp16)[name = tensor<string, []>("op_23029_cast")];
            tensor<fp16, []> denom_157_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_157_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_157_cast = rsqrt(epsilon = denom_157_epsilon_0_to_fp16, x = var_23029_cast)[name = tensor<string, []>("denom_157_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_157_cast = mul(x = zero_mean_157_cast, y = denom_157_cast)[name = tensor<string, []>("out_157_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_23033_to_fp16 = const()[name = tensor<string, []>("op_23033_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1865134976)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_23034_cast = add(x = out_157_cast, y = var_23033_to_fp16)[name = tensor<string, []>("op_23034_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_23036_to_fp16 = const()[name = tensor<string, []>("op_23036_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1865137600)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_221_cast = mul(x = var_23034_cast, y = var_23036_to_fp16)[name = tensor<string, []>("hidden_states_221_cast")];
            tensor<int32, [2]> var_23043 = const()[name = tensor<string, []>("op_23043"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_23045 = const()[name = tensor<string, []>("op_23045"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_105_pad_type_0 = const()[name = tensor<string, []>("q_105_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_105_pad_0 = const()[name = tensor<string, []>("q_105_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1865140224)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_105_cast = conv(dilations = var_23045, groups = var_21105, pad = q_105_pad_0, pad_type = q_105_pad_type_0, strides = var_23043, weight = mid_block_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16, x = hidden_states_221_cast)[name = tensor<string, []>("q_105_cast")];
            tensor<int32, [2]> var_23049 = const()[name = tensor<string, []>("op_23049"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_23051 = const()[name = tensor<string, []>("op_23051"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_209_pad_type_0 = const()[name = tensor<string, []>("k_209_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_209_pad_0 = const()[name = tensor<string, []>("k_209_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1868417088)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_209_cast = conv(dilations = var_23051, groups = var_21105, pad = k_209_pad_0, pad_type = k_209_pad_type_0, strides = var_23049, weight = mid_block_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16, x = hidden_states_221_cast)[name = tensor<string, []>("k_209_cast")];
            tensor<int32, [2]> var_23055 = const()[name = tensor<string, []>("op_23055"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_23057 = const()[name = tensor<string, []>("op_23057"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_105_pad_type_0 = const()[name = tensor<string, []>("v_105_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_105_pad_0 = const()[name = tensor<string, []>("v_105_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1871693952)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_105_cast = conv(dilations = var_23057, groups = var_21105, pad = v_105_pad_0, pad_type = v_105_pad_type_0, strides = var_23055, weight = mid_block_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16, x = hidden_states_221_cast)[name = tensor<string, []>("v_105_cast")];
            tensor<int32, [4]> var_23061_begin_0 = const()[name = tensor<string, []>("op_23061_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_23061_end_0 = const()[name = tensor<string, []>("op_23061_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_23061_end_mask_0 = const()[name = tensor<string, []>("op_23061_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23061_cast = slice_by_index(begin = var_23061_begin_0, end = var_23061_end_0, end_mask = var_23061_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23061_cast")];
            tensor<int32, [4]> var_23065_begin_0 = const()[name = tensor<string, []>("op_23065_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_23065_end_0 = const()[name = tensor<string, []>("op_23065_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_23065_end_mask_0 = const()[name = tensor<string, []>("op_23065_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23065_cast = slice_by_index(begin = var_23065_begin_0, end = var_23065_end_0, end_mask = var_23065_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23065_cast")];
            tensor<int32, [4]> var_23069_begin_0 = const()[name = tensor<string, []>("op_23069_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_23069_end_0 = const()[name = tensor<string, []>("op_23069_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_23069_end_mask_0 = const()[name = tensor<string, []>("op_23069_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23069_cast = slice_by_index(begin = var_23069_begin_0, end = var_23069_end_0, end_mask = var_23069_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23069_cast")];
            tensor<int32, [4]> var_23073_begin_0 = const()[name = tensor<string, []>("op_23073_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_23073_end_0 = const()[name = tensor<string, []>("op_23073_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_23073_end_mask_0 = const()[name = tensor<string, []>("op_23073_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23073_cast = slice_by_index(begin = var_23073_begin_0, end = var_23073_end_0, end_mask = var_23073_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23073_cast")];
            tensor<int32, [4]> var_23077_begin_0 = const()[name = tensor<string, []>("op_23077_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_23077_end_0 = const()[name = tensor<string, []>("op_23077_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_23077_end_mask_0 = const()[name = tensor<string, []>("op_23077_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23077_cast = slice_by_index(begin = var_23077_begin_0, end = var_23077_end_0, end_mask = var_23077_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23077_cast")];
            tensor<int32, [4]> var_23081_begin_0 = const()[name = tensor<string, []>("op_23081_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_23081_end_0 = const()[name = tensor<string, []>("op_23081_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_23081_end_mask_0 = const()[name = tensor<string, []>("op_23081_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23081_cast = slice_by_index(begin = var_23081_begin_0, end = var_23081_end_0, end_mask = var_23081_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23081_cast")];
            tensor<int32, [4]> var_23085_begin_0 = const()[name = tensor<string, []>("op_23085_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_23085_end_0 = const()[name = tensor<string, []>("op_23085_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_23085_end_mask_0 = const()[name = tensor<string, []>("op_23085_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23085_cast = slice_by_index(begin = var_23085_begin_0, end = var_23085_end_0, end_mask = var_23085_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23085_cast")];
            tensor<int32, [4]> var_23089_begin_0 = const()[name = tensor<string, []>("op_23089_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_23089_end_0 = const()[name = tensor<string, []>("op_23089_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_23089_end_mask_0 = const()[name = tensor<string, []>("op_23089_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23089_cast = slice_by_index(begin = var_23089_begin_0, end = var_23089_end_0, end_mask = var_23089_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23089_cast")];
            tensor<int32, [4]> var_23093_begin_0 = const()[name = tensor<string, []>("op_23093_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_23093_end_0 = const()[name = tensor<string, []>("op_23093_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_23093_end_mask_0 = const()[name = tensor<string, []>("op_23093_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23093_cast = slice_by_index(begin = var_23093_begin_0, end = var_23093_end_0, end_mask = var_23093_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23093_cast")];
            tensor<int32, [4]> var_23097_begin_0 = const()[name = tensor<string, []>("op_23097_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_23097_end_0 = const()[name = tensor<string, []>("op_23097_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_23097_end_mask_0 = const()[name = tensor<string, []>("op_23097_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23097_cast = slice_by_index(begin = var_23097_begin_0, end = var_23097_end_0, end_mask = var_23097_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23097_cast")];
            tensor<int32, [4]> var_23101_begin_0 = const()[name = tensor<string, []>("op_23101_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_23101_end_0 = const()[name = tensor<string, []>("op_23101_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_23101_end_mask_0 = const()[name = tensor<string, []>("op_23101_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23101_cast = slice_by_index(begin = var_23101_begin_0, end = var_23101_end_0, end_mask = var_23101_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23101_cast")];
            tensor<int32, [4]> var_23105_begin_0 = const()[name = tensor<string, []>("op_23105_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_23105_end_0 = const()[name = tensor<string, []>("op_23105_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_23105_end_mask_0 = const()[name = tensor<string, []>("op_23105_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23105_cast = slice_by_index(begin = var_23105_begin_0, end = var_23105_end_0, end_mask = var_23105_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23105_cast")];
            tensor<int32, [4]> var_23109_begin_0 = const()[name = tensor<string, []>("op_23109_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_23109_end_0 = const()[name = tensor<string, []>("op_23109_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_23109_end_mask_0 = const()[name = tensor<string, []>("op_23109_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23109_cast = slice_by_index(begin = var_23109_begin_0, end = var_23109_end_0, end_mask = var_23109_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23109_cast")];
            tensor<int32, [4]> var_23113_begin_0 = const()[name = tensor<string, []>("op_23113_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_23113_end_0 = const()[name = tensor<string, []>("op_23113_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_23113_end_mask_0 = const()[name = tensor<string, []>("op_23113_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23113_cast = slice_by_index(begin = var_23113_begin_0, end = var_23113_end_0, end_mask = var_23113_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23113_cast")];
            tensor<int32, [4]> var_23117_begin_0 = const()[name = tensor<string, []>("op_23117_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_23117_end_0 = const()[name = tensor<string, []>("op_23117_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_23117_end_mask_0 = const()[name = tensor<string, []>("op_23117_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23117_cast = slice_by_index(begin = var_23117_begin_0, end = var_23117_end_0, end_mask = var_23117_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23117_cast")];
            tensor<int32, [4]> var_23121_begin_0 = const()[name = tensor<string, []>("op_23121_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_23121_end_0 = const()[name = tensor<string, []>("op_23121_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_23121_end_mask_0 = const()[name = tensor<string, []>("op_23121_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23121_cast = slice_by_index(begin = var_23121_begin_0, end = var_23121_end_0, end_mask = var_23121_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23121_cast")];
            tensor<int32, [4]> var_23125_begin_0 = const()[name = tensor<string, []>("op_23125_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_23125_end_0 = const()[name = tensor<string, []>("op_23125_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_23125_end_mask_0 = const()[name = tensor<string, []>("op_23125_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23125_cast = slice_by_index(begin = var_23125_begin_0, end = var_23125_end_0, end_mask = var_23125_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23125_cast")];
            tensor<int32, [4]> var_23129_begin_0 = const()[name = tensor<string, []>("op_23129_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_23129_end_0 = const()[name = tensor<string, []>("op_23129_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_23129_end_mask_0 = const()[name = tensor<string, []>("op_23129_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23129_cast = slice_by_index(begin = var_23129_begin_0, end = var_23129_end_0, end_mask = var_23129_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23129_cast")];
            tensor<int32, [4]> var_23133_begin_0 = const()[name = tensor<string, []>("op_23133_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_23133_end_0 = const()[name = tensor<string, []>("op_23133_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_23133_end_mask_0 = const()[name = tensor<string, []>("op_23133_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23133_cast = slice_by_index(begin = var_23133_begin_0, end = var_23133_end_0, end_mask = var_23133_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23133_cast")];
            tensor<int32, [4]> var_23137_begin_0 = const()[name = tensor<string, []>("op_23137_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_23137_end_0 = const()[name = tensor<string, []>("op_23137_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_23137_end_mask_0 = const()[name = tensor<string, []>("op_23137_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23137_cast = slice_by_index(begin = var_23137_begin_0, end = var_23137_end_0, end_mask = var_23137_end_mask_0, x = q_105_cast)[name = tensor<string, []>("op_23137_cast")];
            tensor<int32, [4]> k_211_perm_0 = const()[name = tensor<string, []>("k_211_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_23144_begin_0 = const()[name = tensor<string, []>("op_23144_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_23144_end_0 = const()[name = tensor<string, []>("op_23144_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_23144_end_mask_0 = const()[name = tensor<string, []>("op_23144_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_87 = transpose(perm = k_211_perm_0, x = k_209_cast)[name = tensor<string, []>("transpose_87")];
            tensor<fp16, [2, 1024, 1, 64]> var_23144_cast = slice_by_index(begin = var_23144_begin_0, end = var_23144_end_0, end_mask = var_23144_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23144_cast")];
            tensor<int32, [4]> var_23148_begin_0 = const()[name = tensor<string, []>("op_23148_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_23148_end_0 = const()[name = tensor<string, []>("op_23148_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_23148_end_mask_0 = const()[name = tensor<string, []>("op_23148_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_23148_cast = slice_by_index(begin = var_23148_begin_0, end = var_23148_end_0, end_mask = var_23148_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23148_cast")];
            tensor<int32, [4]> var_23152_begin_0 = const()[name = tensor<string, []>("op_23152_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_23152_end_0 = const()[name = tensor<string, []>("op_23152_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_23152_end_mask_0 = const()[name = tensor<string, []>("op_23152_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_23152_cast = slice_by_index(begin = var_23152_begin_0, end = var_23152_end_0, end_mask = var_23152_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23152_cast")];
            tensor<int32, [4]> var_23156_begin_0 = const()[name = tensor<string, []>("op_23156_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_23156_end_0 = const()[name = tensor<string, []>("op_23156_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_23156_end_mask_0 = const()[name = tensor<string, []>("op_23156_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_23156_cast = slice_by_index(begin = var_23156_begin_0, end = var_23156_end_0, end_mask = var_23156_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23156_cast")];
            tensor<int32, [4]> var_23160_begin_0 = const()[name = tensor<string, []>("op_23160_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_23160_end_0 = const()[name = tensor<string, []>("op_23160_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_23160_end_mask_0 = const()[name = tensor<string, []>("op_23160_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_23160_cast = slice_by_index(begin = var_23160_begin_0, end = var_23160_end_0, end_mask = var_23160_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23160_cast")];
            tensor<int32, [4]> var_23164_begin_0 = const()[name = tensor<string, []>("op_23164_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_23164_end_0 = const()[name = tensor<string, []>("op_23164_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_23164_end_mask_0 = const()[name = tensor<string, []>("op_23164_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_23164_cast = slice_by_index(begin = var_23164_begin_0, end = var_23164_end_0, end_mask = var_23164_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23164_cast")];
            tensor<int32, [4]> var_23168_begin_0 = const()[name = tensor<string, []>("op_23168_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_23168_end_0 = const()[name = tensor<string, []>("op_23168_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_23168_end_mask_0 = const()[name = tensor<string, []>("op_23168_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_23168_cast = slice_by_index(begin = var_23168_begin_0, end = var_23168_end_0, end_mask = var_23168_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23168_cast")];
            tensor<int32, [4]> var_23172_begin_0 = const()[name = tensor<string, []>("op_23172_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_23172_end_0 = const()[name = tensor<string, []>("op_23172_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_23172_end_mask_0 = const()[name = tensor<string, []>("op_23172_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_23172_cast = slice_by_index(begin = var_23172_begin_0, end = var_23172_end_0, end_mask = var_23172_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23172_cast")];
            tensor<int32, [4]> var_23176_begin_0 = const()[name = tensor<string, []>("op_23176_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_23176_end_0 = const()[name = tensor<string, []>("op_23176_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_23176_end_mask_0 = const()[name = tensor<string, []>("op_23176_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_23176_cast = slice_by_index(begin = var_23176_begin_0, end = var_23176_end_0, end_mask = var_23176_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23176_cast")];
            tensor<int32, [4]> var_23180_begin_0 = const()[name = tensor<string, []>("op_23180_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_23180_end_0 = const()[name = tensor<string, []>("op_23180_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_23180_end_mask_0 = const()[name = tensor<string, []>("op_23180_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_23180_cast = slice_by_index(begin = var_23180_begin_0, end = var_23180_end_0, end_mask = var_23180_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23180_cast")];
            tensor<int32, [4]> var_23184_begin_0 = const()[name = tensor<string, []>("op_23184_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_23184_end_0 = const()[name = tensor<string, []>("op_23184_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_23184_end_mask_0 = const()[name = tensor<string, []>("op_23184_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_23184_cast = slice_by_index(begin = var_23184_begin_0, end = var_23184_end_0, end_mask = var_23184_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23184_cast")];
            tensor<int32, [4]> var_23188_begin_0 = const()[name = tensor<string, []>("op_23188_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_23188_end_0 = const()[name = tensor<string, []>("op_23188_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_23188_end_mask_0 = const()[name = tensor<string, []>("op_23188_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_23188_cast = slice_by_index(begin = var_23188_begin_0, end = var_23188_end_0, end_mask = var_23188_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23188_cast")];
            tensor<int32, [4]> var_23192_begin_0 = const()[name = tensor<string, []>("op_23192_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_23192_end_0 = const()[name = tensor<string, []>("op_23192_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_23192_end_mask_0 = const()[name = tensor<string, []>("op_23192_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_23192_cast = slice_by_index(begin = var_23192_begin_0, end = var_23192_end_0, end_mask = var_23192_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23192_cast")];
            tensor<int32, [4]> var_23196_begin_0 = const()[name = tensor<string, []>("op_23196_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_23196_end_0 = const()[name = tensor<string, []>("op_23196_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_23196_end_mask_0 = const()[name = tensor<string, []>("op_23196_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_23196_cast = slice_by_index(begin = var_23196_begin_0, end = var_23196_end_0, end_mask = var_23196_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23196_cast")];
            tensor<int32, [4]> var_23200_begin_0 = const()[name = tensor<string, []>("op_23200_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_23200_end_0 = const()[name = tensor<string, []>("op_23200_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_23200_end_mask_0 = const()[name = tensor<string, []>("op_23200_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_23200_cast = slice_by_index(begin = var_23200_begin_0, end = var_23200_end_0, end_mask = var_23200_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23200_cast")];
            tensor<int32, [4]> var_23204_begin_0 = const()[name = tensor<string, []>("op_23204_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_23204_end_0 = const()[name = tensor<string, []>("op_23204_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_23204_end_mask_0 = const()[name = tensor<string, []>("op_23204_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_23204_cast = slice_by_index(begin = var_23204_begin_0, end = var_23204_end_0, end_mask = var_23204_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23204_cast")];
            tensor<int32, [4]> var_23208_begin_0 = const()[name = tensor<string, []>("op_23208_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_23208_end_0 = const()[name = tensor<string, []>("op_23208_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_23208_end_mask_0 = const()[name = tensor<string, []>("op_23208_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_23208_cast = slice_by_index(begin = var_23208_begin_0, end = var_23208_end_0, end_mask = var_23208_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23208_cast")];
            tensor<int32, [4]> var_23212_begin_0 = const()[name = tensor<string, []>("op_23212_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_23212_end_0 = const()[name = tensor<string, []>("op_23212_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_23212_end_mask_0 = const()[name = tensor<string, []>("op_23212_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_23212_cast = slice_by_index(begin = var_23212_begin_0, end = var_23212_end_0, end_mask = var_23212_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23212_cast")];
            tensor<int32, [4]> var_23216_begin_0 = const()[name = tensor<string, []>("op_23216_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_23216_end_0 = const()[name = tensor<string, []>("op_23216_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_23216_end_mask_0 = const()[name = tensor<string, []>("op_23216_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_23216_cast = slice_by_index(begin = var_23216_begin_0, end = var_23216_end_0, end_mask = var_23216_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23216_cast")];
            tensor<int32, [4]> var_23220_begin_0 = const()[name = tensor<string, []>("op_23220_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_23220_end_0 = const()[name = tensor<string, []>("op_23220_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_23220_end_mask_0 = const()[name = tensor<string, []>("op_23220_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_23220_cast = slice_by_index(begin = var_23220_begin_0, end = var_23220_end_0, end_mask = var_23220_end_mask_0, x = transpose_87)[name = tensor<string, []>("op_23220_cast")];
            tensor<int32, [4]> var_23222_begin_0 = const()[name = tensor<string, []>("op_23222_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_23222_end_0 = const()[name = tensor<string, []>("op_23222_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_23222_end_mask_0 = const()[name = tensor<string, []>("op_23222_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23222_cast = slice_by_index(begin = var_23222_begin_0, end = var_23222_end_0, end_mask = var_23222_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23222_cast")];
            tensor<int32, [4]> var_23226_begin_0 = const()[name = tensor<string, []>("op_23226_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_23226_end_0 = const()[name = tensor<string, []>("op_23226_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_23226_end_mask_0 = const()[name = tensor<string, []>("op_23226_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23226_cast = slice_by_index(begin = var_23226_begin_0, end = var_23226_end_0, end_mask = var_23226_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23226_cast")];
            tensor<int32, [4]> var_23230_begin_0 = const()[name = tensor<string, []>("op_23230_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_23230_end_0 = const()[name = tensor<string, []>("op_23230_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_23230_end_mask_0 = const()[name = tensor<string, []>("op_23230_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23230_cast = slice_by_index(begin = var_23230_begin_0, end = var_23230_end_0, end_mask = var_23230_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23230_cast")];
            tensor<int32, [4]> var_23234_begin_0 = const()[name = tensor<string, []>("op_23234_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_23234_end_0 = const()[name = tensor<string, []>("op_23234_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_23234_end_mask_0 = const()[name = tensor<string, []>("op_23234_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23234_cast = slice_by_index(begin = var_23234_begin_0, end = var_23234_end_0, end_mask = var_23234_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23234_cast")];
            tensor<int32, [4]> var_23238_begin_0 = const()[name = tensor<string, []>("op_23238_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_23238_end_0 = const()[name = tensor<string, []>("op_23238_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_23238_end_mask_0 = const()[name = tensor<string, []>("op_23238_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23238_cast = slice_by_index(begin = var_23238_begin_0, end = var_23238_end_0, end_mask = var_23238_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23238_cast")];
            tensor<int32, [4]> var_23242_begin_0 = const()[name = tensor<string, []>("op_23242_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_23242_end_0 = const()[name = tensor<string, []>("op_23242_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_23242_end_mask_0 = const()[name = tensor<string, []>("op_23242_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23242_cast = slice_by_index(begin = var_23242_begin_0, end = var_23242_end_0, end_mask = var_23242_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23242_cast")];
            tensor<int32, [4]> var_23246_begin_0 = const()[name = tensor<string, []>("op_23246_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_23246_end_0 = const()[name = tensor<string, []>("op_23246_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_23246_end_mask_0 = const()[name = tensor<string, []>("op_23246_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23246_cast = slice_by_index(begin = var_23246_begin_0, end = var_23246_end_0, end_mask = var_23246_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23246_cast")];
            tensor<int32, [4]> var_23250_begin_0 = const()[name = tensor<string, []>("op_23250_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_23250_end_0 = const()[name = tensor<string, []>("op_23250_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_23250_end_mask_0 = const()[name = tensor<string, []>("op_23250_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23250_cast = slice_by_index(begin = var_23250_begin_0, end = var_23250_end_0, end_mask = var_23250_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23250_cast")];
            tensor<int32, [4]> var_23254_begin_0 = const()[name = tensor<string, []>("op_23254_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_23254_end_0 = const()[name = tensor<string, []>("op_23254_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_23254_end_mask_0 = const()[name = tensor<string, []>("op_23254_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23254_cast = slice_by_index(begin = var_23254_begin_0, end = var_23254_end_0, end_mask = var_23254_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23254_cast")];
            tensor<int32, [4]> var_23258_begin_0 = const()[name = tensor<string, []>("op_23258_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_23258_end_0 = const()[name = tensor<string, []>("op_23258_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_23258_end_mask_0 = const()[name = tensor<string, []>("op_23258_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23258_cast = slice_by_index(begin = var_23258_begin_0, end = var_23258_end_0, end_mask = var_23258_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23258_cast")];
            tensor<int32, [4]> var_23262_begin_0 = const()[name = tensor<string, []>("op_23262_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_23262_end_0 = const()[name = tensor<string, []>("op_23262_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_23262_end_mask_0 = const()[name = tensor<string, []>("op_23262_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23262_cast = slice_by_index(begin = var_23262_begin_0, end = var_23262_end_0, end_mask = var_23262_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23262_cast")];
            tensor<int32, [4]> var_23266_begin_0 = const()[name = tensor<string, []>("op_23266_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_23266_end_0 = const()[name = tensor<string, []>("op_23266_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_23266_end_mask_0 = const()[name = tensor<string, []>("op_23266_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23266_cast = slice_by_index(begin = var_23266_begin_0, end = var_23266_end_0, end_mask = var_23266_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23266_cast")];
            tensor<int32, [4]> var_23270_begin_0 = const()[name = tensor<string, []>("op_23270_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_23270_end_0 = const()[name = tensor<string, []>("op_23270_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_23270_end_mask_0 = const()[name = tensor<string, []>("op_23270_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23270_cast = slice_by_index(begin = var_23270_begin_0, end = var_23270_end_0, end_mask = var_23270_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23270_cast")];
            tensor<int32, [4]> var_23274_begin_0 = const()[name = tensor<string, []>("op_23274_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_23274_end_0 = const()[name = tensor<string, []>("op_23274_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_23274_end_mask_0 = const()[name = tensor<string, []>("op_23274_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23274_cast = slice_by_index(begin = var_23274_begin_0, end = var_23274_end_0, end_mask = var_23274_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23274_cast")];
            tensor<int32, [4]> var_23278_begin_0 = const()[name = tensor<string, []>("op_23278_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_23278_end_0 = const()[name = tensor<string, []>("op_23278_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_23278_end_mask_0 = const()[name = tensor<string, []>("op_23278_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23278_cast = slice_by_index(begin = var_23278_begin_0, end = var_23278_end_0, end_mask = var_23278_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23278_cast")];
            tensor<int32, [4]> var_23282_begin_0 = const()[name = tensor<string, []>("op_23282_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_23282_end_0 = const()[name = tensor<string, []>("op_23282_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_23282_end_mask_0 = const()[name = tensor<string, []>("op_23282_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23282_cast = slice_by_index(begin = var_23282_begin_0, end = var_23282_end_0, end_mask = var_23282_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23282_cast")];
            tensor<int32, [4]> var_23286_begin_0 = const()[name = tensor<string, []>("op_23286_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_23286_end_0 = const()[name = tensor<string, []>("op_23286_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_23286_end_mask_0 = const()[name = tensor<string, []>("op_23286_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23286_cast = slice_by_index(begin = var_23286_begin_0, end = var_23286_end_0, end_mask = var_23286_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23286_cast")];
            tensor<int32, [4]> var_23290_begin_0 = const()[name = tensor<string, []>("op_23290_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_23290_end_0 = const()[name = tensor<string, []>("op_23290_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_23290_end_mask_0 = const()[name = tensor<string, []>("op_23290_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23290_cast = slice_by_index(begin = var_23290_begin_0, end = var_23290_end_0, end_mask = var_23290_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23290_cast")];
            tensor<int32, [4]> var_23294_begin_0 = const()[name = tensor<string, []>("op_23294_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_23294_end_0 = const()[name = tensor<string, []>("op_23294_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_23294_end_mask_0 = const()[name = tensor<string, []>("op_23294_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23294_cast = slice_by_index(begin = var_23294_begin_0, end = var_23294_end_0, end_mask = var_23294_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23294_cast")];
            tensor<int32, [4]> var_23298_begin_0 = const()[name = tensor<string, []>("op_23298_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_23298_end_0 = const()[name = tensor<string, []>("op_23298_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_23298_end_mask_0 = const()[name = tensor<string, []>("op_23298_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23298_cast = slice_by_index(begin = var_23298_begin_0, end = var_23298_end_0, end_mask = var_23298_end_mask_0, x = v_105_cast)[name = tensor<string, []>("op_23298_cast")];
            tensor<string, []> var_23302_equation_0 = const()[name = tensor<string, []>("op_23302_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23302_cast = einsum(equation = var_23302_equation_0, values = (var_23144_cast, var_23061_cast))[name = tensor<string, []>("op_23302_cast")];
            tensor<fp16, []> var_23303_to_fp16 = const()[name = tensor<string, []>("op_23303_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1921_cast = mul(x = var_23302_cast, y = var_23303_to_fp16)[name = tensor<string, []>("aw_1921_cast")];
            tensor<string, []> var_23306_equation_0 = const()[name = tensor<string, []>("op_23306_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23306_cast = einsum(equation = var_23306_equation_0, values = (var_23148_cast, var_23065_cast))[name = tensor<string, []>("op_23306_cast")];
            tensor<fp16, []> var_23307_to_fp16 = const()[name = tensor<string, []>("op_23307_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1923_cast = mul(x = var_23306_cast, y = var_23307_to_fp16)[name = tensor<string, []>("aw_1923_cast")];
            tensor<string, []> var_23310_equation_0 = const()[name = tensor<string, []>("op_23310_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23310_cast = einsum(equation = var_23310_equation_0, values = (var_23152_cast, var_23069_cast))[name = tensor<string, []>("op_23310_cast")];
            tensor<fp16, []> var_23311_to_fp16 = const()[name = tensor<string, []>("op_23311_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1925_cast = mul(x = var_23310_cast, y = var_23311_to_fp16)[name = tensor<string, []>("aw_1925_cast")];
            tensor<string, []> var_23314_equation_0 = const()[name = tensor<string, []>("op_23314_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23314_cast = einsum(equation = var_23314_equation_0, values = (var_23156_cast, var_23073_cast))[name = tensor<string, []>("op_23314_cast")];
            tensor<fp16, []> var_23315_to_fp16 = const()[name = tensor<string, []>("op_23315_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1927_cast = mul(x = var_23314_cast, y = var_23315_to_fp16)[name = tensor<string, []>("aw_1927_cast")];
            tensor<string, []> var_23318_equation_0 = const()[name = tensor<string, []>("op_23318_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23318_cast = einsum(equation = var_23318_equation_0, values = (var_23160_cast, var_23077_cast))[name = tensor<string, []>("op_23318_cast")];
            tensor<fp16, []> var_23319_to_fp16 = const()[name = tensor<string, []>("op_23319_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1929_cast = mul(x = var_23318_cast, y = var_23319_to_fp16)[name = tensor<string, []>("aw_1929_cast")];
            tensor<string, []> var_23322_equation_0 = const()[name = tensor<string, []>("op_23322_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23322_cast = einsum(equation = var_23322_equation_0, values = (var_23164_cast, var_23081_cast))[name = tensor<string, []>("op_23322_cast")];
            tensor<fp16, []> var_23323_to_fp16 = const()[name = tensor<string, []>("op_23323_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1931_cast = mul(x = var_23322_cast, y = var_23323_to_fp16)[name = tensor<string, []>("aw_1931_cast")];
            tensor<string, []> var_23326_equation_0 = const()[name = tensor<string, []>("op_23326_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23326_cast = einsum(equation = var_23326_equation_0, values = (var_23168_cast, var_23085_cast))[name = tensor<string, []>("op_23326_cast")];
            tensor<fp16, []> var_23327_to_fp16 = const()[name = tensor<string, []>("op_23327_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1933_cast = mul(x = var_23326_cast, y = var_23327_to_fp16)[name = tensor<string, []>("aw_1933_cast")];
            tensor<string, []> var_23330_equation_0 = const()[name = tensor<string, []>("op_23330_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23330_cast = einsum(equation = var_23330_equation_0, values = (var_23172_cast, var_23089_cast))[name = tensor<string, []>("op_23330_cast")];
            tensor<fp16, []> var_23331_to_fp16 = const()[name = tensor<string, []>("op_23331_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1935_cast = mul(x = var_23330_cast, y = var_23331_to_fp16)[name = tensor<string, []>("aw_1935_cast")];
            tensor<string, []> var_23334_equation_0 = const()[name = tensor<string, []>("op_23334_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23334_cast = einsum(equation = var_23334_equation_0, values = (var_23176_cast, var_23093_cast))[name = tensor<string, []>("op_23334_cast")];
            tensor<fp16, []> var_23335_to_fp16 = const()[name = tensor<string, []>("op_23335_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1937_cast = mul(x = var_23334_cast, y = var_23335_to_fp16)[name = tensor<string, []>("aw_1937_cast")];
            tensor<string, []> var_23338_equation_0 = const()[name = tensor<string, []>("op_23338_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23338_cast = einsum(equation = var_23338_equation_0, values = (var_23180_cast, var_23097_cast))[name = tensor<string, []>("op_23338_cast")];
            tensor<fp16, []> var_23339_to_fp16 = const()[name = tensor<string, []>("op_23339_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1939_cast = mul(x = var_23338_cast, y = var_23339_to_fp16)[name = tensor<string, []>("aw_1939_cast")];
            tensor<string, []> var_23342_equation_0 = const()[name = tensor<string, []>("op_23342_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23342_cast = einsum(equation = var_23342_equation_0, values = (var_23184_cast, var_23101_cast))[name = tensor<string, []>("op_23342_cast")];
            tensor<fp16, []> var_23343_to_fp16 = const()[name = tensor<string, []>("op_23343_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1941_cast = mul(x = var_23342_cast, y = var_23343_to_fp16)[name = tensor<string, []>("aw_1941_cast")];
            tensor<string, []> var_23346_equation_0 = const()[name = tensor<string, []>("op_23346_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23346_cast = einsum(equation = var_23346_equation_0, values = (var_23188_cast, var_23105_cast))[name = tensor<string, []>("op_23346_cast")];
            tensor<fp16, []> var_23347_to_fp16 = const()[name = tensor<string, []>("op_23347_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1943_cast = mul(x = var_23346_cast, y = var_23347_to_fp16)[name = tensor<string, []>("aw_1943_cast")];
            tensor<string, []> var_23350_equation_0 = const()[name = tensor<string, []>("op_23350_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23350_cast = einsum(equation = var_23350_equation_0, values = (var_23192_cast, var_23109_cast))[name = tensor<string, []>("op_23350_cast")];
            tensor<fp16, []> var_23351_to_fp16 = const()[name = tensor<string, []>("op_23351_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1945_cast = mul(x = var_23350_cast, y = var_23351_to_fp16)[name = tensor<string, []>("aw_1945_cast")];
            tensor<string, []> var_23354_equation_0 = const()[name = tensor<string, []>("op_23354_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23354_cast = einsum(equation = var_23354_equation_0, values = (var_23196_cast, var_23113_cast))[name = tensor<string, []>("op_23354_cast")];
            tensor<fp16, []> var_23355_to_fp16 = const()[name = tensor<string, []>("op_23355_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1947_cast = mul(x = var_23354_cast, y = var_23355_to_fp16)[name = tensor<string, []>("aw_1947_cast")];
            tensor<string, []> var_23358_equation_0 = const()[name = tensor<string, []>("op_23358_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23358_cast = einsum(equation = var_23358_equation_0, values = (var_23200_cast, var_23117_cast))[name = tensor<string, []>("op_23358_cast")];
            tensor<fp16, []> var_23359_to_fp16 = const()[name = tensor<string, []>("op_23359_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1949_cast = mul(x = var_23358_cast, y = var_23359_to_fp16)[name = tensor<string, []>("aw_1949_cast")];
            tensor<string, []> var_23362_equation_0 = const()[name = tensor<string, []>("op_23362_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23362_cast = einsum(equation = var_23362_equation_0, values = (var_23204_cast, var_23121_cast))[name = tensor<string, []>("op_23362_cast")];
            tensor<fp16, []> var_23363_to_fp16 = const()[name = tensor<string, []>("op_23363_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1951_cast = mul(x = var_23362_cast, y = var_23363_to_fp16)[name = tensor<string, []>("aw_1951_cast")];
            tensor<string, []> var_23366_equation_0 = const()[name = tensor<string, []>("op_23366_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23366_cast = einsum(equation = var_23366_equation_0, values = (var_23208_cast, var_23125_cast))[name = tensor<string, []>("op_23366_cast")];
            tensor<fp16, []> var_23367_to_fp16 = const()[name = tensor<string, []>("op_23367_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1953_cast = mul(x = var_23366_cast, y = var_23367_to_fp16)[name = tensor<string, []>("aw_1953_cast")];
            tensor<string, []> var_23370_equation_0 = const()[name = tensor<string, []>("op_23370_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23370_cast = einsum(equation = var_23370_equation_0, values = (var_23212_cast, var_23129_cast))[name = tensor<string, []>("op_23370_cast")];
            tensor<fp16, []> var_23371_to_fp16 = const()[name = tensor<string, []>("op_23371_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1955_cast = mul(x = var_23370_cast, y = var_23371_to_fp16)[name = tensor<string, []>("aw_1955_cast")];
            tensor<string, []> var_23374_equation_0 = const()[name = tensor<string, []>("op_23374_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23374_cast = einsum(equation = var_23374_equation_0, values = (var_23216_cast, var_23133_cast))[name = tensor<string, []>("op_23374_cast")];
            tensor<fp16, []> var_23375_to_fp16 = const()[name = tensor<string, []>("op_23375_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1957_cast = mul(x = var_23374_cast, y = var_23375_to_fp16)[name = tensor<string, []>("aw_1957_cast")];
            tensor<string, []> var_23378_equation_0 = const()[name = tensor<string, []>("op_23378_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23378_cast = einsum(equation = var_23378_equation_0, values = (var_23220_cast, var_23137_cast))[name = tensor<string, []>("op_23378_cast")];
            tensor<fp16, []> var_23379_to_fp16 = const()[name = tensor<string, []>("op_23379_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_1959_cast = mul(x = var_23378_cast, y = var_23379_to_fp16)[name = tensor<string, []>("aw_1959_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23381_cast = softmax(axis = var_21105, x = aw_1921_cast)[name = tensor<string, []>("op_23381_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23382_cast = softmax(axis = var_21105, x = aw_1923_cast)[name = tensor<string, []>("op_23382_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23383_cast = softmax(axis = var_21105, x = aw_1925_cast)[name = tensor<string, []>("op_23383_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23384_cast = softmax(axis = var_21105, x = aw_1927_cast)[name = tensor<string, []>("op_23384_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23385_cast = softmax(axis = var_21105, x = aw_1929_cast)[name = tensor<string, []>("op_23385_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23386_cast = softmax(axis = var_21105, x = aw_1931_cast)[name = tensor<string, []>("op_23386_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23387_cast = softmax(axis = var_21105, x = aw_1933_cast)[name = tensor<string, []>("op_23387_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23388_cast = softmax(axis = var_21105, x = aw_1935_cast)[name = tensor<string, []>("op_23388_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23389_cast = softmax(axis = var_21105, x = aw_1937_cast)[name = tensor<string, []>("op_23389_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23390_cast = softmax(axis = var_21105, x = aw_1939_cast)[name = tensor<string, []>("op_23390_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23391_cast = softmax(axis = var_21105, x = aw_1941_cast)[name = tensor<string, []>("op_23391_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23392_cast = softmax(axis = var_21105, x = aw_1943_cast)[name = tensor<string, []>("op_23392_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23393_cast = softmax(axis = var_21105, x = aw_1945_cast)[name = tensor<string, []>("op_23393_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23394_cast = softmax(axis = var_21105, x = aw_1947_cast)[name = tensor<string, []>("op_23394_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23395_cast = softmax(axis = var_21105, x = aw_1949_cast)[name = tensor<string, []>("op_23395_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23396_cast = softmax(axis = var_21105, x = aw_1951_cast)[name = tensor<string, []>("op_23396_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23397_cast = softmax(axis = var_21105, x = aw_1953_cast)[name = tensor<string, []>("op_23397_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23398_cast = softmax(axis = var_21105, x = aw_1955_cast)[name = tensor<string, []>("op_23398_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23399_cast = softmax(axis = var_21105, x = aw_1957_cast)[name = tensor<string, []>("op_23399_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_23400_cast = softmax(axis = var_21105, x = aw_1959_cast)[name = tensor<string, []>("op_23400_cast")];
            tensor<string, []> var_23402_equation_0 = const()[name = tensor<string, []>("op_23402_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23402_cast = einsum(equation = var_23402_equation_0, values = (var_23222_cast, var_23381_cast))[name = tensor<string, []>("op_23402_cast")];
            tensor<string, []> var_23404_equation_0 = const()[name = tensor<string, []>("op_23404_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23404_cast = einsum(equation = var_23404_equation_0, values = (var_23226_cast, var_23382_cast))[name = tensor<string, []>("op_23404_cast")];
            tensor<string, []> var_23406_equation_0 = const()[name = tensor<string, []>("op_23406_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23406_cast = einsum(equation = var_23406_equation_0, values = (var_23230_cast, var_23383_cast))[name = tensor<string, []>("op_23406_cast")];
            tensor<string, []> var_23408_equation_0 = const()[name = tensor<string, []>("op_23408_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23408_cast = einsum(equation = var_23408_equation_0, values = (var_23234_cast, var_23384_cast))[name = tensor<string, []>("op_23408_cast")];
            tensor<string, []> var_23410_equation_0 = const()[name = tensor<string, []>("op_23410_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23410_cast = einsum(equation = var_23410_equation_0, values = (var_23238_cast, var_23385_cast))[name = tensor<string, []>("op_23410_cast")];
            tensor<string, []> var_23412_equation_0 = const()[name = tensor<string, []>("op_23412_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23412_cast = einsum(equation = var_23412_equation_0, values = (var_23242_cast, var_23386_cast))[name = tensor<string, []>("op_23412_cast")];
            tensor<string, []> var_23414_equation_0 = const()[name = tensor<string, []>("op_23414_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23414_cast = einsum(equation = var_23414_equation_0, values = (var_23246_cast, var_23387_cast))[name = tensor<string, []>("op_23414_cast")];
            tensor<string, []> var_23416_equation_0 = const()[name = tensor<string, []>("op_23416_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23416_cast = einsum(equation = var_23416_equation_0, values = (var_23250_cast, var_23388_cast))[name = tensor<string, []>("op_23416_cast")];
            tensor<string, []> var_23418_equation_0 = const()[name = tensor<string, []>("op_23418_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23418_cast = einsum(equation = var_23418_equation_0, values = (var_23254_cast, var_23389_cast))[name = tensor<string, []>("op_23418_cast")];
            tensor<string, []> var_23420_equation_0 = const()[name = tensor<string, []>("op_23420_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23420_cast = einsum(equation = var_23420_equation_0, values = (var_23258_cast, var_23390_cast))[name = tensor<string, []>("op_23420_cast")];
            tensor<string, []> var_23422_equation_0 = const()[name = tensor<string, []>("op_23422_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23422_cast = einsum(equation = var_23422_equation_0, values = (var_23262_cast, var_23391_cast))[name = tensor<string, []>("op_23422_cast")];
            tensor<string, []> var_23424_equation_0 = const()[name = tensor<string, []>("op_23424_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23424_cast = einsum(equation = var_23424_equation_0, values = (var_23266_cast, var_23392_cast))[name = tensor<string, []>("op_23424_cast")];
            tensor<string, []> var_23426_equation_0 = const()[name = tensor<string, []>("op_23426_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23426_cast = einsum(equation = var_23426_equation_0, values = (var_23270_cast, var_23393_cast))[name = tensor<string, []>("op_23426_cast")];
            tensor<string, []> var_23428_equation_0 = const()[name = tensor<string, []>("op_23428_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23428_cast = einsum(equation = var_23428_equation_0, values = (var_23274_cast, var_23394_cast))[name = tensor<string, []>("op_23428_cast")];
            tensor<string, []> var_23430_equation_0 = const()[name = tensor<string, []>("op_23430_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23430_cast = einsum(equation = var_23430_equation_0, values = (var_23278_cast, var_23395_cast))[name = tensor<string, []>("op_23430_cast")];
            tensor<string, []> var_23432_equation_0 = const()[name = tensor<string, []>("op_23432_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23432_cast = einsum(equation = var_23432_equation_0, values = (var_23282_cast, var_23396_cast))[name = tensor<string, []>("op_23432_cast")];
            tensor<string, []> var_23434_equation_0 = const()[name = tensor<string, []>("op_23434_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23434_cast = einsum(equation = var_23434_equation_0, values = (var_23286_cast, var_23397_cast))[name = tensor<string, []>("op_23434_cast")];
            tensor<string, []> var_23436_equation_0 = const()[name = tensor<string, []>("op_23436_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23436_cast = einsum(equation = var_23436_equation_0, values = (var_23290_cast, var_23398_cast))[name = tensor<string, []>("op_23436_cast")];
            tensor<string, []> var_23438_equation_0 = const()[name = tensor<string, []>("op_23438_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23438_cast = einsum(equation = var_23438_equation_0, values = (var_23294_cast, var_23399_cast))[name = tensor<string, []>("op_23438_cast")];
            tensor<string, []> var_23440_equation_0 = const()[name = tensor<string, []>("op_23440_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23440_cast = einsum(equation = var_23440_equation_0, values = (var_23298_cast, var_23400_cast))[name = tensor<string, []>("op_23440_cast")];
            tensor<bool, []> input_345_interleave_0 = const()[name = tensor<string, []>("input_345_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_345_cast = concat(axis = var_21105, interleave = input_345_interleave_0, values = (var_23402_cast, var_23404_cast, var_23406_cast, var_23408_cast, var_23410_cast, var_23412_cast, var_23414_cast, var_23416_cast, var_23418_cast, var_23420_cast, var_23422_cast, var_23424_cast, var_23426_cast, var_23428_cast, var_23430_cast, var_23432_cast, var_23434_cast, var_23436_cast, var_23438_cast, var_23440_cast))[name = tensor<string, []>("input_345_cast")];
            tensor<int32, [2]> var_23446 = const()[name = tensor<string, []>("op_23446"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_23448 = const()[name = tensor<string, []>("op_23448"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_23450_pad_type_0 = const()[name = tensor<string, []>("op_23450_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_23450_pad_0 = const()[name = tensor<string, []>("op_23450_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1874970816)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1878247680)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_23450_cast = conv(bias = mid_block_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16, dilations = var_23448, groups = var_21105, pad = var_23450_pad_0, pad_type = var_23450_pad_type_0, strides = var_23446, weight = mid_block_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16, x = input_345_cast)[name = tensor<string, []>("op_23450_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_159_cast = add(x = var_23450_cast, y = inputs_157_cast)[name = tensor<string, []>("inputs_159_cast")];
            tensor<int32, [1]> var_23454 = const()[name = tensor<string, []>("op_23454"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_159_cast = reduce_mean(axes = var_23454, keep_dims = var_21100, x = inputs_159_cast)[name = tensor<string, []>("channels_mean_159_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_159_cast = sub(x = inputs_159_cast, y = channels_mean_159_cast)[name = tensor<string, []>("zero_mean_159_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_159_cast = mul(x = zero_mean_159_cast, y = zero_mean_159_cast)[name = tensor<string, []>("zero_mean_sq_159_cast")];
            tensor<int32, [1]> var_23458 = const()[name = tensor<string, []>("op_23458"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_23459_cast = reduce_mean(axes = var_23458, keep_dims = var_21100, x = zero_mean_sq_159_cast)[name = tensor<string, []>("op_23459_cast")];
            tensor<fp16, []> var_23460_to_fp16 = const()[name = tensor<string, []>("op_23460_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_23461_cast = add(x = var_23459_cast, y = var_23460_to_fp16)[name = tensor<string, []>("op_23461_cast")];
            tensor<fp16, []> denom_159_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_159_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_159_cast = rsqrt(epsilon = denom_159_epsilon_0_to_fp16, x = var_23461_cast)[name = tensor<string, []>("denom_159_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_159_cast = mul(x = zero_mean_159_cast, y = denom_159_cast)[name = tensor<string, []>("out_159_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_23465_to_fp16 = const()[name = tensor<string, []>("op_23465_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1878250304)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_23466_cast = add(x = out_159_cast, y = var_23465_to_fp16)[name = tensor<string, []>("op_23466_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_23468_to_fp16 = const()[name = tensor<string, []>("op_23468_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1878252928)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_223_cast = mul(x = var_23466_cast, y = var_23468_to_fp16)[name = tensor<string, []>("hidden_states_223_cast")];
            tensor<int32, [2]> var_23475 = const()[name = tensor<string, []>("op_23475"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_23477 = const()[name = tensor<string, []>("op_23477"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_107_pad_type_0 = const()[name = tensor<string, []>("q_107_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_107_pad_0 = const()[name = tensor<string, []>("q_107_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1878255552)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_107_cast = conv(dilations = var_23477, groups = var_21105, pad = q_107_pad_0, pad_type = q_107_pad_type_0, strides = var_23475, weight = mid_block_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16, x = hidden_states_223_cast)[name = tensor<string, []>("q_107_cast")];
            tensor<int32, [2]> var_23481 = const()[name = tensor<string, []>("op_23481"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_23483 = const()[name = tensor<string, []>("op_23483"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_213_pad_type_0 = const()[name = tensor<string, []>("k_213_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_213_pad_0 = const()[name = tensor<string, []>("k_213_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1881532416)))];
            tensor<fp16, [2, 1280, 1, 77]> k_213_cast = conv(dilations = var_23483, groups = var_21105, pad = k_213_pad_0, pad_type = k_213_pad_type_0, strides = var_23481, weight = mid_block_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_213_cast")];
            tensor<int32, [2]> var_23487 = const()[name = tensor<string, []>("op_23487"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_23489 = const()[name = tensor<string, []>("op_23489"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_107_pad_type_0 = const()[name = tensor<string, []>("v_107_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_107_pad_0 = const()[name = tensor<string, []>("v_107_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1886775360)))];
            tensor<fp16, [2, 1280, 1, 77]> v_107_cast = conv(dilations = var_23489, groups = var_21105, pad = v_107_pad_0, pad_type = v_107_pad_type_0, strides = var_23487, weight = mid_block_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_107_cast")];
            tensor<int32, [4]> var_23493_begin_0 = const()[name = tensor<string, []>("op_23493_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_23493_end_0 = const()[name = tensor<string, []>("op_23493_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_23493_end_mask_0 = const()[name = tensor<string, []>("op_23493_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23493_cast = slice_by_index(begin = var_23493_begin_0, end = var_23493_end_0, end_mask = var_23493_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23493_cast")];
            tensor<int32, [4]> var_23497_begin_0 = const()[name = tensor<string, []>("op_23497_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_23497_end_0 = const()[name = tensor<string, []>("op_23497_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_23497_end_mask_0 = const()[name = tensor<string, []>("op_23497_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23497_cast = slice_by_index(begin = var_23497_begin_0, end = var_23497_end_0, end_mask = var_23497_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23497_cast")];
            tensor<int32, [4]> var_23501_begin_0 = const()[name = tensor<string, []>("op_23501_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_23501_end_0 = const()[name = tensor<string, []>("op_23501_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_23501_end_mask_0 = const()[name = tensor<string, []>("op_23501_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23501_cast = slice_by_index(begin = var_23501_begin_0, end = var_23501_end_0, end_mask = var_23501_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23501_cast")];
            tensor<int32, [4]> var_23505_begin_0 = const()[name = tensor<string, []>("op_23505_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_23505_end_0 = const()[name = tensor<string, []>("op_23505_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_23505_end_mask_0 = const()[name = tensor<string, []>("op_23505_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23505_cast = slice_by_index(begin = var_23505_begin_0, end = var_23505_end_0, end_mask = var_23505_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23505_cast")];
            tensor<int32, [4]> var_23509_begin_0 = const()[name = tensor<string, []>("op_23509_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_23509_end_0 = const()[name = tensor<string, []>("op_23509_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_23509_end_mask_0 = const()[name = tensor<string, []>("op_23509_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23509_cast = slice_by_index(begin = var_23509_begin_0, end = var_23509_end_0, end_mask = var_23509_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23509_cast")];
            tensor<int32, [4]> var_23513_begin_0 = const()[name = tensor<string, []>("op_23513_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_23513_end_0 = const()[name = tensor<string, []>("op_23513_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_23513_end_mask_0 = const()[name = tensor<string, []>("op_23513_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23513_cast = slice_by_index(begin = var_23513_begin_0, end = var_23513_end_0, end_mask = var_23513_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23513_cast")];
            tensor<int32, [4]> var_23517_begin_0 = const()[name = tensor<string, []>("op_23517_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_23517_end_0 = const()[name = tensor<string, []>("op_23517_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_23517_end_mask_0 = const()[name = tensor<string, []>("op_23517_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23517_cast = slice_by_index(begin = var_23517_begin_0, end = var_23517_end_0, end_mask = var_23517_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23517_cast")];
            tensor<int32, [4]> var_23521_begin_0 = const()[name = tensor<string, []>("op_23521_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_23521_end_0 = const()[name = tensor<string, []>("op_23521_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_23521_end_mask_0 = const()[name = tensor<string, []>("op_23521_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23521_cast = slice_by_index(begin = var_23521_begin_0, end = var_23521_end_0, end_mask = var_23521_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23521_cast")];
            tensor<int32, [4]> var_23525_begin_0 = const()[name = tensor<string, []>("op_23525_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_23525_end_0 = const()[name = tensor<string, []>("op_23525_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_23525_end_mask_0 = const()[name = tensor<string, []>("op_23525_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23525_cast = slice_by_index(begin = var_23525_begin_0, end = var_23525_end_0, end_mask = var_23525_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23525_cast")];
            tensor<int32, [4]> var_23529_begin_0 = const()[name = tensor<string, []>("op_23529_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_23529_end_0 = const()[name = tensor<string, []>("op_23529_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_23529_end_mask_0 = const()[name = tensor<string, []>("op_23529_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23529_cast = slice_by_index(begin = var_23529_begin_0, end = var_23529_end_0, end_mask = var_23529_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23529_cast")];
            tensor<int32, [4]> var_23533_begin_0 = const()[name = tensor<string, []>("op_23533_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_23533_end_0 = const()[name = tensor<string, []>("op_23533_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_23533_end_mask_0 = const()[name = tensor<string, []>("op_23533_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23533_cast = slice_by_index(begin = var_23533_begin_0, end = var_23533_end_0, end_mask = var_23533_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23533_cast")];
            tensor<int32, [4]> var_23537_begin_0 = const()[name = tensor<string, []>("op_23537_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_23537_end_0 = const()[name = tensor<string, []>("op_23537_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_23537_end_mask_0 = const()[name = tensor<string, []>("op_23537_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23537_cast = slice_by_index(begin = var_23537_begin_0, end = var_23537_end_0, end_mask = var_23537_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23537_cast")];
            tensor<int32, [4]> var_23541_begin_0 = const()[name = tensor<string, []>("op_23541_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_23541_end_0 = const()[name = tensor<string, []>("op_23541_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_23541_end_mask_0 = const()[name = tensor<string, []>("op_23541_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23541_cast = slice_by_index(begin = var_23541_begin_0, end = var_23541_end_0, end_mask = var_23541_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23541_cast")];
            tensor<int32, [4]> var_23545_begin_0 = const()[name = tensor<string, []>("op_23545_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_23545_end_0 = const()[name = tensor<string, []>("op_23545_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_23545_end_mask_0 = const()[name = tensor<string, []>("op_23545_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23545_cast = slice_by_index(begin = var_23545_begin_0, end = var_23545_end_0, end_mask = var_23545_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23545_cast")];
            tensor<int32, [4]> var_23549_begin_0 = const()[name = tensor<string, []>("op_23549_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_23549_end_0 = const()[name = tensor<string, []>("op_23549_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_23549_end_mask_0 = const()[name = tensor<string, []>("op_23549_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23549_cast = slice_by_index(begin = var_23549_begin_0, end = var_23549_end_0, end_mask = var_23549_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23549_cast")];
            tensor<int32, [4]> var_23553_begin_0 = const()[name = tensor<string, []>("op_23553_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_23553_end_0 = const()[name = tensor<string, []>("op_23553_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_23553_end_mask_0 = const()[name = tensor<string, []>("op_23553_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23553_cast = slice_by_index(begin = var_23553_begin_0, end = var_23553_end_0, end_mask = var_23553_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23553_cast")];
            tensor<int32, [4]> var_23557_begin_0 = const()[name = tensor<string, []>("op_23557_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_23557_end_0 = const()[name = tensor<string, []>("op_23557_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_23557_end_mask_0 = const()[name = tensor<string, []>("op_23557_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23557_cast = slice_by_index(begin = var_23557_begin_0, end = var_23557_end_0, end_mask = var_23557_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23557_cast")];
            tensor<int32, [4]> var_23561_begin_0 = const()[name = tensor<string, []>("op_23561_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_23561_end_0 = const()[name = tensor<string, []>("op_23561_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_23561_end_mask_0 = const()[name = tensor<string, []>("op_23561_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23561_cast = slice_by_index(begin = var_23561_begin_0, end = var_23561_end_0, end_mask = var_23561_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23561_cast")];
            tensor<int32, [4]> var_23565_begin_0 = const()[name = tensor<string, []>("op_23565_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_23565_end_0 = const()[name = tensor<string, []>("op_23565_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_23565_end_mask_0 = const()[name = tensor<string, []>("op_23565_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23565_cast = slice_by_index(begin = var_23565_begin_0, end = var_23565_end_0, end_mask = var_23565_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23565_cast")];
            tensor<int32, [4]> var_23569_begin_0 = const()[name = tensor<string, []>("op_23569_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_23569_end_0 = const()[name = tensor<string, []>("op_23569_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_23569_end_mask_0 = const()[name = tensor<string, []>("op_23569_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23569_cast = slice_by_index(begin = var_23569_begin_0, end = var_23569_end_0, end_mask = var_23569_end_mask_0, x = q_107_cast)[name = tensor<string, []>("op_23569_cast")];
            tensor<int32, [4]> k_215_perm_0 = const()[name = tensor<string, []>("k_215_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_23576_begin_0 = const()[name = tensor<string, []>("op_23576_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_23576_end_0 = const()[name = tensor<string, []>("op_23576_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_23576_end_mask_0 = const()[name = tensor<string, []>("op_23576_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_86 = transpose(perm = k_215_perm_0, x = k_213_cast)[name = tensor<string, []>("transpose_86")];
            tensor<fp16, [2, 77, 1, 64]> var_23576_cast = slice_by_index(begin = var_23576_begin_0, end = var_23576_end_0, end_mask = var_23576_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23576_cast")];
            tensor<int32, [4]> var_23580_begin_0 = const()[name = tensor<string, []>("op_23580_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_23580_end_0 = const()[name = tensor<string, []>("op_23580_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_23580_end_mask_0 = const()[name = tensor<string, []>("op_23580_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_23580_cast = slice_by_index(begin = var_23580_begin_0, end = var_23580_end_0, end_mask = var_23580_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23580_cast")];
            tensor<int32, [4]> var_23584_begin_0 = const()[name = tensor<string, []>("op_23584_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_23584_end_0 = const()[name = tensor<string, []>("op_23584_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_23584_end_mask_0 = const()[name = tensor<string, []>("op_23584_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_23584_cast = slice_by_index(begin = var_23584_begin_0, end = var_23584_end_0, end_mask = var_23584_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23584_cast")];
            tensor<int32, [4]> var_23588_begin_0 = const()[name = tensor<string, []>("op_23588_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_23588_end_0 = const()[name = tensor<string, []>("op_23588_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_23588_end_mask_0 = const()[name = tensor<string, []>("op_23588_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_23588_cast = slice_by_index(begin = var_23588_begin_0, end = var_23588_end_0, end_mask = var_23588_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23588_cast")];
            tensor<int32, [4]> var_23592_begin_0 = const()[name = tensor<string, []>("op_23592_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_23592_end_0 = const()[name = tensor<string, []>("op_23592_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_23592_end_mask_0 = const()[name = tensor<string, []>("op_23592_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_23592_cast = slice_by_index(begin = var_23592_begin_0, end = var_23592_end_0, end_mask = var_23592_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23592_cast")];
            tensor<int32, [4]> var_23596_begin_0 = const()[name = tensor<string, []>("op_23596_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_23596_end_0 = const()[name = tensor<string, []>("op_23596_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_23596_end_mask_0 = const()[name = tensor<string, []>("op_23596_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_23596_cast = slice_by_index(begin = var_23596_begin_0, end = var_23596_end_0, end_mask = var_23596_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23596_cast")];
            tensor<int32, [4]> var_23600_begin_0 = const()[name = tensor<string, []>("op_23600_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_23600_end_0 = const()[name = tensor<string, []>("op_23600_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_23600_end_mask_0 = const()[name = tensor<string, []>("op_23600_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_23600_cast = slice_by_index(begin = var_23600_begin_0, end = var_23600_end_0, end_mask = var_23600_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23600_cast")];
            tensor<int32, [4]> var_23604_begin_0 = const()[name = tensor<string, []>("op_23604_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_23604_end_0 = const()[name = tensor<string, []>("op_23604_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_23604_end_mask_0 = const()[name = tensor<string, []>("op_23604_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_23604_cast = slice_by_index(begin = var_23604_begin_0, end = var_23604_end_0, end_mask = var_23604_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23604_cast")];
            tensor<int32, [4]> var_23608_begin_0 = const()[name = tensor<string, []>("op_23608_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_23608_end_0 = const()[name = tensor<string, []>("op_23608_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_23608_end_mask_0 = const()[name = tensor<string, []>("op_23608_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_23608_cast = slice_by_index(begin = var_23608_begin_0, end = var_23608_end_0, end_mask = var_23608_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23608_cast")];
            tensor<int32, [4]> var_23612_begin_0 = const()[name = tensor<string, []>("op_23612_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_23612_end_0 = const()[name = tensor<string, []>("op_23612_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_23612_end_mask_0 = const()[name = tensor<string, []>("op_23612_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_23612_cast = slice_by_index(begin = var_23612_begin_0, end = var_23612_end_0, end_mask = var_23612_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23612_cast")];
            tensor<int32, [4]> var_23616_begin_0 = const()[name = tensor<string, []>("op_23616_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_23616_end_0 = const()[name = tensor<string, []>("op_23616_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_23616_end_mask_0 = const()[name = tensor<string, []>("op_23616_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_23616_cast = slice_by_index(begin = var_23616_begin_0, end = var_23616_end_0, end_mask = var_23616_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23616_cast")];
            tensor<int32, [4]> var_23620_begin_0 = const()[name = tensor<string, []>("op_23620_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_23620_end_0 = const()[name = tensor<string, []>("op_23620_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_23620_end_mask_0 = const()[name = tensor<string, []>("op_23620_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_23620_cast = slice_by_index(begin = var_23620_begin_0, end = var_23620_end_0, end_mask = var_23620_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23620_cast")];
            tensor<int32, [4]> var_23624_begin_0 = const()[name = tensor<string, []>("op_23624_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_23624_end_0 = const()[name = tensor<string, []>("op_23624_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_23624_end_mask_0 = const()[name = tensor<string, []>("op_23624_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_23624_cast = slice_by_index(begin = var_23624_begin_0, end = var_23624_end_0, end_mask = var_23624_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23624_cast")];
            tensor<int32, [4]> var_23628_begin_0 = const()[name = tensor<string, []>("op_23628_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_23628_end_0 = const()[name = tensor<string, []>("op_23628_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_23628_end_mask_0 = const()[name = tensor<string, []>("op_23628_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_23628_cast = slice_by_index(begin = var_23628_begin_0, end = var_23628_end_0, end_mask = var_23628_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23628_cast")];
            tensor<int32, [4]> var_23632_begin_0 = const()[name = tensor<string, []>("op_23632_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_23632_end_0 = const()[name = tensor<string, []>("op_23632_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_23632_end_mask_0 = const()[name = tensor<string, []>("op_23632_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_23632_cast = slice_by_index(begin = var_23632_begin_0, end = var_23632_end_0, end_mask = var_23632_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23632_cast")];
            tensor<int32, [4]> var_23636_begin_0 = const()[name = tensor<string, []>("op_23636_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_23636_end_0 = const()[name = tensor<string, []>("op_23636_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_23636_end_mask_0 = const()[name = tensor<string, []>("op_23636_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_23636_cast = slice_by_index(begin = var_23636_begin_0, end = var_23636_end_0, end_mask = var_23636_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23636_cast")];
            tensor<int32, [4]> var_23640_begin_0 = const()[name = tensor<string, []>("op_23640_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_23640_end_0 = const()[name = tensor<string, []>("op_23640_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_23640_end_mask_0 = const()[name = tensor<string, []>("op_23640_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_23640_cast = slice_by_index(begin = var_23640_begin_0, end = var_23640_end_0, end_mask = var_23640_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23640_cast")];
            tensor<int32, [4]> var_23644_begin_0 = const()[name = tensor<string, []>("op_23644_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_23644_end_0 = const()[name = tensor<string, []>("op_23644_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_23644_end_mask_0 = const()[name = tensor<string, []>("op_23644_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_23644_cast = slice_by_index(begin = var_23644_begin_0, end = var_23644_end_0, end_mask = var_23644_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23644_cast")];
            tensor<int32, [4]> var_23648_begin_0 = const()[name = tensor<string, []>("op_23648_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_23648_end_0 = const()[name = tensor<string, []>("op_23648_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_23648_end_mask_0 = const()[name = tensor<string, []>("op_23648_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_23648_cast = slice_by_index(begin = var_23648_begin_0, end = var_23648_end_0, end_mask = var_23648_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23648_cast")];
            tensor<int32, [4]> var_23652_begin_0 = const()[name = tensor<string, []>("op_23652_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_23652_end_0 = const()[name = tensor<string, []>("op_23652_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_23652_end_mask_0 = const()[name = tensor<string, []>("op_23652_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_23652_cast = slice_by_index(begin = var_23652_begin_0, end = var_23652_end_0, end_mask = var_23652_end_mask_0, x = transpose_86)[name = tensor<string, []>("op_23652_cast")];
            tensor<int32, [4]> var_23654_begin_0 = const()[name = tensor<string, []>("op_23654_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_23654_end_0 = const()[name = tensor<string, []>("op_23654_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_23654_end_mask_0 = const()[name = tensor<string, []>("op_23654_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23654_cast = slice_by_index(begin = var_23654_begin_0, end = var_23654_end_0, end_mask = var_23654_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23654_cast")];
            tensor<int32, [4]> var_23658_begin_0 = const()[name = tensor<string, []>("op_23658_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_23658_end_0 = const()[name = tensor<string, []>("op_23658_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_23658_end_mask_0 = const()[name = tensor<string, []>("op_23658_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23658_cast = slice_by_index(begin = var_23658_begin_0, end = var_23658_end_0, end_mask = var_23658_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23658_cast")];
            tensor<int32, [4]> var_23662_begin_0 = const()[name = tensor<string, []>("op_23662_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_23662_end_0 = const()[name = tensor<string, []>("op_23662_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_23662_end_mask_0 = const()[name = tensor<string, []>("op_23662_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23662_cast = slice_by_index(begin = var_23662_begin_0, end = var_23662_end_0, end_mask = var_23662_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23662_cast")];
            tensor<int32, [4]> var_23666_begin_0 = const()[name = tensor<string, []>("op_23666_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_23666_end_0 = const()[name = tensor<string, []>("op_23666_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_23666_end_mask_0 = const()[name = tensor<string, []>("op_23666_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23666_cast = slice_by_index(begin = var_23666_begin_0, end = var_23666_end_0, end_mask = var_23666_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23666_cast")];
            tensor<int32, [4]> var_23670_begin_0 = const()[name = tensor<string, []>("op_23670_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_23670_end_0 = const()[name = tensor<string, []>("op_23670_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_23670_end_mask_0 = const()[name = tensor<string, []>("op_23670_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23670_cast = slice_by_index(begin = var_23670_begin_0, end = var_23670_end_0, end_mask = var_23670_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23670_cast")];
            tensor<int32, [4]> var_23674_begin_0 = const()[name = tensor<string, []>("op_23674_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_23674_end_0 = const()[name = tensor<string, []>("op_23674_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_23674_end_mask_0 = const()[name = tensor<string, []>("op_23674_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23674_cast = slice_by_index(begin = var_23674_begin_0, end = var_23674_end_0, end_mask = var_23674_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23674_cast")];
            tensor<int32, [4]> var_23678_begin_0 = const()[name = tensor<string, []>("op_23678_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_23678_end_0 = const()[name = tensor<string, []>("op_23678_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_23678_end_mask_0 = const()[name = tensor<string, []>("op_23678_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23678_cast = slice_by_index(begin = var_23678_begin_0, end = var_23678_end_0, end_mask = var_23678_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23678_cast")];
            tensor<int32, [4]> var_23682_begin_0 = const()[name = tensor<string, []>("op_23682_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_23682_end_0 = const()[name = tensor<string, []>("op_23682_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_23682_end_mask_0 = const()[name = tensor<string, []>("op_23682_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23682_cast = slice_by_index(begin = var_23682_begin_0, end = var_23682_end_0, end_mask = var_23682_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23682_cast")];
            tensor<int32, [4]> var_23686_begin_0 = const()[name = tensor<string, []>("op_23686_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_23686_end_0 = const()[name = tensor<string, []>("op_23686_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_23686_end_mask_0 = const()[name = tensor<string, []>("op_23686_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23686_cast = slice_by_index(begin = var_23686_begin_0, end = var_23686_end_0, end_mask = var_23686_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23686_cast")];
            tensor<int32, [4]> var_23690_begin_0 = const()[name = tensor<string, []>("op_23690_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_23690_end_0 = const()[name = tensor<string, []>("op_23690_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_23690_end_mask_0 = const()[name = tensor<string, []>("op_23690_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23690_cast = slice_by_index(begin = var_23690_begin_0, end = var_23690_end_0, end_mask = var_23690_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23690_cast")];
            tensor<int32, [4]> var_23694_begin_0 = const()[name = tensor<string, []>("op_23694_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_23694_end_0 = const()[name = tensor<string, []>("op_23694_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_23694_end_mask_0 = const()[name = tensor<string, []>("op_23694_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23694_cast = slice_by_index(begin = var_23694_begin_0, end = var_23694_end_0, end_mask = var_23694_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23694_cast")];
            tensor<int32, [4]> var_23698_begin_0 = const()[name = tensor<string, []>("op_23698_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_23698_end_0 = const()[name = tensor<string, []>("op_23698_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_23698_end_mask_0 = const()[name = tensor<string, []>("op_23698_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23698_cast = slice_by_index(begin = var_23698_begin_0, end = var_23698_end_0, end_mask = var_23698_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23698_cast")];
            tensor<int32, [4]> var_23702_begin_0 = const()[name = tensor<string, []>("op_23702_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_23702_end_0 = const()[name = tensor<string, []>("op_23702_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_23702_end_mask_0 = const()[name = tensor<string, []>("op_23702_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23702_cast = slice_by_index(begin = var_23702_begin_0, end = var_23702_end_0, end_mask = var_23702_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23702_cast")];
            tensor<int32, [4]> var_23706_begin_0 = const()[name = tensor<string, []>("op_23706_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_23706_end_0 = const()[name = tensor<string, []>("op_23706_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_23706_end_mask_0 = const()[name = tensor<string, []>("op_23706_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23706_cast = slice_by_index(begin = var_23706_begin_0, end = var_23706_end_0, end_mask = var_23706_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23706_cast")];
            tensor<int32, [4]> var_23710_begin_0 = const()[name = tensor<string, []>("op_23710_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_23710_end_0 = const()[name = tensor<string, []>("op_23710_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_23710_end_mask_0 = const()[name = tensor<string, []>("op_23710_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23710_cast = slice_by_index(begin = var_23710_begin_0, end = var_23710_end_0, end_mask = var_23710_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23710_cast")];
            tensor<int32, [4]> var_23714_begin_0 = const()[name = tensor<string, []>("op_23714_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_23714_end_0 = const()[name = tensor<string, []>("op_23714_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_23714_end_mask_0 = const()[name = tensor<string, []>("op_23714_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23714_cast = slice_by_index(begin = var_23714_begin_0, end = var_23714_end_0, end_mask = var_23714_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23714_cast")];
            tensor<int32, [4]> var_23718_begin_0 = const()[name = tensor<string, []>("op_23718_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_23718_end_0 = const()[name = tensor<string, []>("op_23718_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_23718_end_mask_0 = const()[name = tensor<string, []>("op_23718_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23718_cast = slice_by_index(begin = var_23718_begin_0, end = var_23718_end_0, end_mask = var_23718_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23718_cast")];
            tensor<int32, [4]> var_23722_begin_0 = const()[name = tensor<string, []>("op_23722_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_23722_end_0 = const()[name = tensor<string, []>("op_23722_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_23722_end_mask_0 = const()[name = tensor<string, []>("op_23722_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23722_cast = slice_by_index(begin = var_23722_begin_0, end = var_23722_end_0, end_mask = var_23722_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23722_cast")];
            tensor<int32, [4]> var_23726_begin_0 = const()[name = tensor<string, []>("op_23726_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_23726_end_0 = const()[name = tensor<string, []>("op_23726_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_23726_end_mask_0 = const()[name = tensor<string, []>("op_23726_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23726_cast = slice_by_index(begin = var_23726_begin_0, end = var_23726_end_0, end_mask = var_23726_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23726_cast")];
            tensor<int32, [4]> var_23730_begin_0 = const()[name = tensor<string, []>("op_23730_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_23730_end_0 = const()[name = tensor<string, []>("op_23730_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_23730_end_mask_0 = const()[name = tensor<string, []>("op_23730_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_23730_cast = slice_by_index(begin = var_23730_begin_0, end = var_23730_end_0, end_mask = var_23730_end_mask_0, x = v_107_cast)[name = tensor<string, []>("op_23730_cast")];
            tensor<string, []> var_23734_equation_0 = const()[name = tensor<string, []>("op_23734_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23734_cast = einsum(equation = var_23734_equation_0, values = (var_23576_cast, var_23493_cast))[name = tensor<string, []>("op_23734_cast")];
            tensor<fp16, []> var_23735_to_fp16 = const()[name = tensor<string, []>("op_23735_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1961_cast = mul(x = var_23734_cast, y = var_23735_to_fp16)[name = tensor<string, []>("aw_1961_cast")];
            tensor<string, []> var_23738_equation_0 = const()[name = tensor<string, []>("op_23738_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23738_cast = einsum(equation = var_23738_equation_0, values = (var_23580_cast, var_23497_cast))[name = tensor<string, []>("op_23738_cast")];
            tensor<fp16, []> var_23739_to_fp16 = const()[name = tensor<string, []>("op_23739_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1963_cast = mul(x = var_23738_cast, y = var_23739_to_fp16)[name = tensor<string, []>("aw_1963_cast")];
            tensor<string, []> var_23742_equation_0 = const()[name = tensor<string, []>("op_23742_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23742_cast = einsum(equation = var_23742_equation_0, values = (var_23584_cast, var_23501_cast))[name = tensor<string, []>("op_23742_cast")];
            tensor<fp16, []> var_23743_to_fp16 = const()[name = tensor<string, []>("op_23743_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1965_cast = mul(x = var_23742_cast, y = var_23743_to_fp16)[name = tensor<string, []>("aw_1965_cast")];
            tensor<string, []> var_23746_equation_0 = const()[name = tensor<string, []>("op_23746_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23746_cast = einsum(equation = var_23746_equation_0, values = (var_23588_cast, var_23505_cast))[name = tensor<string, []>("op_23746_cast")];
            tensor<fp16, []> var_23747_to_fp16 = const()[name = tensor<string, []>("op_23747_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1967_cast = mul(x = var_23746_cast, y = var_23747_to_fp16)[name = tensor<string, []>("aw_1967_cast")];
            tensor<string, []> var_23750_equation_0 = const()[name = tensor<string, []>("op_23750_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23750_cast = einsum(equation = var_23750_equation_0, values = (var_23592_cast, var_23509_cast))[name = tensor<string, []>("op_23750_cast")];
            tensor<fp16, []> var_23751_to_fp16 = const()[name = tensor<string, []>("op_23751_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1969_cast = mul(x = var_23750_cast, y = var_23751_to_fp16)[name = tensor<string, []>("aw_1969_cast")];
            tensor<string, []> var_23754_equation_0 = const()[name = tensor<string, []>("op_23754_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23754_cast = einsum(equation = var_23754_equation_0, values = (var_23596_cast, var_23513_cast))[name = tensor<string, []>("op_23754_cast")];
            tensor<fp16, []> var_23755_to_fp16 = const()[name = tensor<string, []>("op_23755_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1971_cast = mul(x = var_23754_cast, y = var_23755_to_fp16)[name = tensor<string, []>("aw_1971_cast")];
            tensor<string, []> var_23758_equation_0 = const()[name = tensor<string, []>("op_23758_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23758_cast = einsum(equation = var_23758_equation_0, values = (var_23600_cast, var_23517_cast))[name = tensor<string, []>("op_23758_cast")];
            tensor<fp16, []> var_23759_to_fp16 = const()[name = tensor<string, []>("op_23759_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1973_cast = mul(x = var_23758_cast, y = var_23759_to_fp16)[name = tensor<string, []>("aw_1973_cast")];
            tensor<string, []> var_23762_equation_0 = const()[name = tensor<string, []>("op_23762_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23762_cast = einsum(equation = var_23762_equation_0, values = (var_23604_cast, var_23521_cast))[name = tensor<string, []>("op_23762_cast")];
            tensor<fp16, []> var_23763_to_fp16 = const()[name = tensor<string, []>("op_23763_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1975_cast = mul(x = var_23762_cast, y = var_23763_to_fp16)[name = tensor<string, []>("aw_1975_cast")];
            tensor<string, []> var_23766_equation_0 = const()[name = tensor<string, []>("op_23766_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23766_cast = einsum(equation = var_23766_equation_0, values = (var_23608_cast, var_23525_cast))[name = tensor<string, []>("op_23766_cast")];
            tensor<fp16, []> var_23767_to_fp16 = const()[name = tensor<string, []>("op_23767_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1977_cast = mul(x = var_23766_cast, y = var_23767_to_fp16)[name = tensor<string, []>("aw_1977_cast")];
            tensor<string, []> var_23770_equation_0 = const()[name = tensor<string, []>("op_23770_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23770_cast = einsum(equation = var_23770_equation_0, values = (var_23612_cast, var_23529_cast))[name = tensor<string, []>("op_23770_cast")];
            tensor<fp16, []> var_23771_to_fp16 = const()[name = tensor<string, []>("op_23771_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1979_cast = mul(x = var_23770_cast, y = var_23771_to_fp16)[name = tensor<string, []>("aw_1979_cast")];
            tensor<string, []> var_23774_equation_0 = const()[name = tensor<string, []>("op_23774_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23774_cast = einsum(equation = var_23774_equation_0, values = (var_23616_cast, var_23533_cast))[name = tensor<string, []>("op_23774_cast")];
            tensor<fp16, []> var_23775_to_fp16 = const()[name = tensor<string, []>("op_23775_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1981_cast = mul(x = var_23774_cast, y = var_23775_to_fp16)[name = tensor<string, []>("aw_1981_cast")];
            tensor<string, []> var_23778_equation_0 = const()[name = tensor<string, []>("op_23778_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23778_cast = einsum(equation = var_23778_equation_0, values = (var_23620_cast, var_23537_cast))[name = tensor<string, []>("op_23778_cast")];
            tensor<fp16, []> var_23779_to_fp16 = const()[name = tensor<string, []>("op_23779_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1983_cast = mul(x = var_23778_cast, y = var_23779_to_fp16)[name = tensor<string, []>("aw_1983_cast")];
            tensor<string, []> var_23782_equation_0 = const()[name = tensor<string, []>("op_23782_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23782_cast = einsum(equation = var_23782_equation_0, values = (var_23624_cast, var_23541_cast))[name = tensor<string, []>("op_23782_cast")];
            tensor<fp16, []> var_23783_to_fp16 = const()[name = tensor<string, []>("op_23783_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1985_cast = mul(x = var_23782_cast, y = var_23783_to_fp16)[name = tensor<string, []>("aw_1985_cast")];
            tensor<string, []> var_23786_equation_0 = const()[name = tensor<string, []>("op_23786_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23786_cast = einsum(equation = var_23786_equation_0, values = (var_23628_cast, var_23545_cast))[name = tensor<string, []>("op_23786_cast")];
            tensor<fp16, []> var_23787_to_fp16 = const()[name = tensor<string, []>("op_23787_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1987_cast = mul(x = var_23786_cast, y = var_23787_to_fp16)[name = tensor<string, []>("aw_1987_cast")];
            tensor<string, []> var_23790_equation_0 = const()[name = tensor<string, []>("op_23790_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23790_cast = einsum(equation = var_23790_equation_0, values = (var_23632_cast, var_23549_cast))[name = tensor<string, []>("op_23790_cast")];
            tensor<fp16, []> var_23791_to_fp16 = const()[name = tensor<string, []>("op_23791_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1989_cast = mul(x = var_23790_cast, y = var_23791_to_fp16)[name = tensor<string, []>("aw_1989_cast")];
            tensor<string, []> var_23794_equation_0 = const()[name = tensor<string, []>("op_23794_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23794_cast = einsum(equation = var_23794_equation_0, values = (var_23636_cast, var_23553_cast))[name = tensor<string, []>("op_23794_cast")];
            tensor<fp16, []> var_23795_to_fp16 = const()[name = tensor<string, []>("op_23795_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1991_cast = mul(x = var_23794_cast, y = var_23795_to_fp16)[name = tensor<string, []>("aw_1991_cast")];
            tensor<string, []> var_23798_equation_0 = const()[name = tensor<string, []>("op_23798_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23798_cast = einsum(equation = var_23798_equation_0, values = (var_23640_cast, var_23557_cast))[name = tensor<string, []>("op_23798_cast")];
            tensor<fp16, []> var_23799_to_fp16 = const()[name = tensor<string, []>("op_23799_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1993_cast = mul(x = var_23798_cast, y = var_23799_to_fp16)[name = tensor<string, []>("aw_1993_cast")];
            tensor<string, []> var_23802_equation_0 = const()[name = tensor<string, []>("op_23802_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23802_cast = einsum(equation = var_23802_equation_0, values = (var_23644_cast, var_23561_cast))[name = tensor<string, []>("op_23802_cast")];
            tensor<fp16, []> var_23803_to_fp16 = const()[name = tensor<string, []>("op_23803_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1995_cast = mul(x = var_23802_cast, y = var_23803_to_fp16)[name = tensor<string, []>("aw_1995_cast")];
            tensor<string, []> var_23806_equation_0 = const()[name = tensor<string, []>("op_23806_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23806_cast = einsum(equation = var_23806_equation_0, values = (var_23648_cast, var_23565_cast))[name = tensor<string, []>("op_23806_cast")];
            tensor<fp16, []> var_23807_to_fp16 = const()[name = tensor<string, []>("op_23807_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1997_cast = mul(x = var_23806_cast, y = var_23807_to_fp16)[name = tensor<string, []>("aw_1997_cast")];
            tensor<string, []> var_23810_equation_0 = const()[name = tensor<string, []>("op_23810_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_23810_cast = einsum(equation = var_23810_equation_0, values = (var_23652_cast, var_23569_cast))[name = tensor<string, []>("op_23810_cast")];
            tensor<fp16, []> var_23811_to_fp16 = const()[name = tensor<string, []>("op_23811_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_1999_cast = mul(x = var_23810_cast, y = var_23811_to_fp16)[name = tensor<string, []>("aw_1999_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23813_cast = softmax(axis = var_21105, x = aw_1961_cast)[name = tensor<string, []>("op_23813_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23814_cast = softmax(axis = var_21105, x = aw_1963_cast)[name = tensor<string, []>("op_23814_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23815_cast = softmax(axis = var_21105, x = aw_1965_cast)[name = tensor<string, []>("op_23815_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23816_cast = softmax(axis = var_21105, x = aw_1967_cast)[name = tensor<string, []>("op_23816_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23817_cast = softmax(axis = var_21105, x = aw_1969_cast)[name = tensor<string, []>("op_23817_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23818_cast = softmax(axis = var_21105, x = aw_1971_cast)[name = tensor<string, []>("op_23818_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23819_cast = softmax(axis = var_21105, x = aw_1973_cast)[name = tensor<string, []>("op_23819_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23820_cast = softmax(axis = var_21105, x = aw_1975_cast)[name = tensor<string, []>("op_23820_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23821_cast = softmax(axis = var_21105, x = aw_1977_cast)[name = tensor<string, []>("op_23821_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23822_cast = softmax(axis = var_21105, x = aw_1979_cast)[name = tensor<string, []>("op_23822_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23823_cast = softmax(axis = var_21105, x = aw_1981_cast)[name = tensor<string, []>("op_23823_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23824_cast = softmax(axis = var_21105, x = aw_1983_cast)[name = tensor<string, []>("op_23824_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23825_cast = softmax(axis = var_21105, x = aw_1985_cast)[name = tensor<string, []>("op_23825_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23826_cast = softmax(axis = var_21105, x = aw_1987_cast)[name = tensor<string, []>("op_23826_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23827_cast = softmax(axis = var_21105, x = aw_1989_cast)[name = tensor<string, []>("op_23827_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23828_cast = softmax(axis = var_21105, x = aw_1991_cast)[name = tensor<string, []>("op_23828_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23829_cast = softmax(axis = var_21105, x = aw_1993_cast)[name = tensor<string, []>("op_23829_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23830_cast = softmax(axis = var_21105, x = aw_1995_cast)[name = tensor<string, []>("op_23830_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23831_cast = softmax(axis = var_21105, x = aw_1997_cast)[name = tensor<string, []>("op_23831_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_23832_cast = softmax(axis = var_21105, x = aw_1999_cast)[name = tensor<string, []>("op_23832_cast")];
            tensor<string, []> var_23834_equation_0 = const()[name = tensor<string, []>("op_23834_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23834_cast = einsum(equation = var_23834_equation_0, values = (var_23654_cast, var_23813_cast))[name = tensor<string, []>("op_23834_cast")];
            tensor<string, []> var_23836_equation_0 = const()[name = tensor<string, []>("op_23836_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23836_cast = einsum(equation = var_23836_equation_0, values = (var_23658_cast, var_23814_cast))[name = tensor<string, []>("op_23836_cast")];
            tensor<string, []> var_23838_equation_0 = const()[name = tensor<string, []>("op_23838_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23838_cast = einsum(equation = var_23838_equation_0, values = (var_23662_cast, var_23815_cast))[name = tensor<string, []>("op_23838_cast")];
            tensor<string, []> var_23840_equation_0 = const()[name = tensor<string, []>("op_23840_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23840_cast = einsum(equation = var_23840_equation_0, values = (var_23666_cast, var_23816_cast))[name = tensor<string, []>("op_23840_cast")];
            tensor<string, []> var_23842_equation_0 = const()[name = tensor<string, []>("op_23842_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23842_cast = einsum(equation = var_23842_equation_0, values = (var_23670_cast, var_23817_cast))[name = tensor<string, []>("op_23842_cast")];
            tensor<string, []> var_23844_equation_0 = const()[name = tensor<string, []>("op_23844_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23844_cast = einsum(equation = var_23844_equation_0, values = (var_23674_cast, var_23818_cast))[name = tensor<string, []>("op_23844_cast")];
            tensor<string, []> var_23846_equation_0 = const()[name = tensor<string, []>("op_23846_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23846_cast = einsum(equation = var_23846_equation_0, values = (var_23678_cast, var_23819_cast))[name = tensor<string, []>("op_23846_cast")];
            tensor<string, []> var_23848_equation_0 = const()[name = tensor<string, []>("op_23848_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23848_cast = einsum(equation = var_23848_equation_0, values = (var_23682_cast, var_23820_cast))[name = tensor<string, []>("op_23848_cast")];
            tensor<string, []> var_23850_equation_0 = const()[name = tensor<string, []>("op_23850_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23850_cast = einsum(equation = var_23850_equation_0, values = (var_23686_cast, var_23821_cast))[name = tensor<string, []>("op_23850_cast")];
            tensor<string, []> var_23852_equation_0 = const()[name = tensor<string, []>("op_23852_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23852_cast = einsum(equation = var_23852_equation_0, values = (var_23690_cast, var_23822_cast))[name = tensor<string, []>("op_23852_cast")];
            tensor<string, []> var_23854_equation_0 = const()[name = tensor<string, []>("op_23854_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23854_cast = einsum(equation = var_23854_equation_0, values = (var_23694_cast, var_23823_cast))[name = tensor<string, []>("op_23854_cast")];
            tensor<string, []> var_23856_equation_0 = const()[name = tensor<string, []>("op_23856_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23856_cast = einsum(equation = var_23856_equation_0, values = (var_23698_cast, var_23824_cast))[name = tensor<string, []>("op_23856_cast")];
            tensor<string, []> var_23858_equation_0 = const()[name = tensor<string, []>("op_23858_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23858_cast = einsum(equation = var_23858_equation_0, values = (var_23702_cast, var_23825_cast))[name = tensor<string, []>("op_23858_cast")];
            tensor<string, []> var_23860_equation_0 = const()[name = tensor<string, []>("op_23860_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23860_cast = einsum(equation = var_23860_equation_0, values = (var_23706_cast, var_23826_cast))[name = tensor<string, []>("op_23860_cast")];
            tensor<string, []> var_23862_equation_0 = const()[name = tensor<string, []>("op_23862_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23862_cast = einsum(equation = var_23862_equation_0, values = (var_23710_cast, var_23827_cast))[name = tensor<string, []>("op_23862_cast")];
            tensor<string, []> var_23864_equation_0 = const()[name = tensor<string, []>("op_23864_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23864_cast = einsum(equation = var_23864_equation_0, values = (var_23714_cast, var_23828_cast))[name = tensor<string, []>("op_23864_cast")];
            tensor<string, []> var_23866_equation_0 = const()[name = tensor<string, []>("op_23866_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23866_cast = einsum(equation = var_23866_equation_0, values = (var_23718_cast, var_23829_cast))[name = tensor<string, []>("op_23866_cast")];
            tensor<string, []> var_23868_equation_0 = const()[name = tensor<string, []>("op_23868_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23868_cast = einsum(equation = var_23868_equation_0, values = (var_23722_cast, var_23830_cast))[name = tensor<string, []>("op_23868_cast")];
            tensor<string, []> var_23870_equation_0 = const()[name = tensor<string, []>("op_23870_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23870_cast = einsum(equation = var_23870_equation_0, values = (var_23726_cast, var_23831_cast))[name = tensor<string, []>("op_23870_cast")];
            tensor<string, []> var_23872_equation_0 = const()[name = tensor<string, []>("op_23872_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_23872_cast = einsum(equation = var_23872_equation_0, values = (var_23730_cast, var_23832_cast))[name = tensor<string, []>("op_23872_cast")];
            tensor<bool, []> input_347_interleave_0 = const()[name = tensor<string, []>("input_347_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_347_cast = concat(axis = var_21105, interleave = input_347_interleave_0, values = (var_23834_cast, var_23836_cast, var_23838_cast, var_23840_cast, var_23842_cast, var_23844_cast, var_23846_cast, var_23848_cast, var_23850_cast, var_23852_cast, var_23854_cast, var_23856_cast, var_23858_cast, var_23860_cast, var_23862_cast, var_23864_cast, var_23866_cast, var_23868_cast, var_23870_cast, var_23872_cast))[name = tensor<string, []>("input_347_cast")];
            tensor<int32, [2]> var_23878 = const()[name = tensor<string, []>("op_23878"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_23880 = const()[name = tensor<string, []>("op_23880"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_23882_pad_type_0 = const()[name = tensor<string, []>("op_23882_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_23882_pad_0 = const()[name = tensor<string, []>("op_23882_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1892018304)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1895295168)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_23882_cast = conv(bias = mid_block_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16, dilations = var_23880, groups = var_21105, pad = var_23882_pad_0, pad_type = var_23882_pad_type_0, strides = var_23878, weight = mid_block_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16, x = input_347_cast)[name = tensor<string, []>("op_23882_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_161_cast = add(x = var_23882_cast, y = inputs_159_cast)[name = tensor<string, []>("inputs_161_cast")];
            tensor<int32, [1]> var_23886 = const()[name = tensor<string, []>("op_23886"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_161_cast = reduce_mean(axes = var_23886, keep_dims = var_21100, x = inputs_161_cast)[name = tensor<string, []>("channels_mean_161_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_161_cast = sub(x = inputs_161_cast, y = channels_mean_161_cast)[name = tensor<string, []>("zero_mean_161_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_161_cast = mul(x = zero_mean_161_cast, y = zero_mean_161_cast)[name = tensor<string, []>("zero_mean_sq_161_cast")];
            tensor<int32, [1]> var_23890 = const()[name = tensor<string, []>("op_23890"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_23891_cast = reduce_mean(axes = var_23890, keep_dims = var_21100, x = zero_mean_sq_161_cast)[name = tensor<string, []>("op_23891_cast")];
            tensor<fp16, []> var_23892_to_fp16 = const()[name = tensor<string, []>("op_23892_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_23893_cast = add(x = var_23891_cast, y = var_23892_to_fp16)[name = tensor<string, []>("op_23893_cast")];
            tensor<fp16, []> denom_161_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_161_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_161_cast = rsqrt(epsilon = denom_161_epsilon_0_to_fp16, x = var_23893_cast)[name = tensor<string, []>("denom_161_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_161_cast = mul(x = zero_mean_161_cast, y = denom_161_cast)[name = tensor<string, []>("out_161_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_23897_to_fp16 = const()[name = tensor<string, []>("op_23897_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1895297792)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_23898_cast = add(x = out_161_cast, y = var_23897_to_fp16)[name = tensor<string, []>("op_23898_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_23900_to_fp16 = const()[name = tensor<string, []>("op_23900_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1895300416)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_349_cast = mul(x = var_23898_cast, y = var_23900_to_fp16)[name = tensor<string, []>("input_349_cast")];
            tensor<int32, [2]> var_23908 = const()[name = tensor<string, []>("op_23908"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_23910 = const()[name = tensor<string, []>("op_23910"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_23912_pad_type_0 = const()[name = tensor<string, []>("op_23912_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_23912_pad_0 = const()[name = tensor<string, []>("op_23912_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1895303040)))];
            tensor<fp16, [10240]> mid_block_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1921517504)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_23912_cast = conv(bias = mid_block_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16, dilations = var_23910, groups = var_21105, pad = var_23912_pad_0, pad_type = var_23912_pad_type_0, strides = var_23908, weight = mid_block_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16, x = input_349_cast)[name = tensor<string, []>("op_23912_cast")];
            tensor<int32, [2]> var_23913_split_sizes_0 = const()[name = tensor<string, []>("op_23913_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_23913_axis_0 = const()[name = tensor<string, []>("op_23913_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_23913_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_23913_cast_1 = split(axis = var_23913_axis_0, split_sizes = var_23913_split_sizes_0, x = var_23912_cast)[name = tensor<string, []>("op_23913_cast")];
            tensor<string, []> var_23915_mode_0 = const()[name = tensor<string, []>("op_23915_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_23915_cast = gelu(mode = var_23915_mode_0, x = var_23913_cast_1)[name = tensor<string, []>("op_23915_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_351_cast = mul(x = var_23913_cast_0, y = var_23915_cast)[name = tensor<string, []>("input_351_cast")];
            tensor<int32, [2]> var_23919 = const()[name = tensor<string, []>("op_23919"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_23921 = const()[name = tensor<string, []>("op_23921"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_23923_pad_type_0 = const()[name = tensor<string, []>("op_23923_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_23923_pad_0 = const()[name = tensor<string, []>("op_23923_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> mid_block_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1921538048)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1934645312)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_23923_cast = conv(bias = mid_block_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16, dilations = var_23921, groups = var_21105, pad = var_23923_pad_0, pad_type = var_23923_pad_type_0, strides = var_23919, weight = mid_block_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16, x = input_351_cast)[name = tensor<string, []>("op_23923_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_163_cast = add(x = var_23923_cast, y = inputs_161_cast)[name = tensor<string, []>("inputs_163_cast")];
            tensor<int32, [1]> var_23933 = const()[name = tensor<string, []>("op_23933"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_163_cast = reduce_mean(axes = var_23933, keep_dims = var_21100, x = inputs_163_cast)[name = tensor<string, []>("channels_mean_163_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_163_cast = sub(x = inputs_163_cast, y = channels_mean_163_cast)[name = tensor<string, []>("zero_mean_163_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_163_cast = mul(x = zero_mean_163_cast, y = zero_mean_163_cast)[name = tensor<string, []>("zero_mean_sq_163_cast")];
            tensor<int32, [1]> var_23937 = const()[name = tensor<string, []>("op_23937"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_23938_cast = reduce_mean(axes = var_23937, keep_dims = var_21100, x = zero_mean_sq_163_cast)[name = tensor<string, []>("op_23938_cast")];
            tensor<fp16, []> var_23939_to_fp16 = const()[name = tensor<string, []>("op_23939_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_23940_cast = add(x = var_23938_cast, y = var_23939_to_fp16)[name = tensor<string, []>("op_23940_cast")];
            tensor<fp16, []> denom_163_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_163_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_163_cast = rsqrt(epsilon = denom_163_epsilon_0_to_fp16, x = var_23940_cast)[name = tensor<string, []>("denom_163_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_163_cast = mul(x = zero_mean_163_cast, y = denom_163_cast)[name = tensor<string, []>("out_163_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_23944_to_fp16 = const()[name = tensor<string, []>("op_23944_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1934647936)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_23945_cast = add(x = out_163_cast, y = var_23944_to_fp16)[name = tensor<string, []>("op_23945_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_23947_to_fp16 = const()[name = tensor<string, []>("op_23947_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1934650560)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_227_cast = mul(x = var_23945_cast, y = var_23947_to_fp16)[name = tensor<string, []>("hidden_states_227_cast")];
            tensor<int32, [2]> var_23954 = const()[name = tensor<string, []>("op_23954"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_23956 = const()[name = tensor<string, []>("op_23956"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_109_pad_type_0 = const()[name = tensor<string, []>("q_109_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_109_pad_0 = const()[name = tensor<string, []>("q_109_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1934653184)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_109_cast = conv(dilations = var_23956, groups = var_21105, pad = q_109_pad_0, pad_type = q_109_pad_type_0, strides = var_23954, weight = mid_block_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16, x = hidden_states_227_cast)[name = tensor<string, []>("q_109_cast")];
            tensor<int32, [2]> var_23960 = const()[name = tensor<string, []>("op_23960"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_23962 = const()[name = tensor<string, []>("op_23962"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_217_pad_type_0 = const()[name = tensor<string, []>("k_217_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_217_pad_0 = const()[name = tensor<string, []>("k_217_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1937930048)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_217_cast = conv(dilations = var_23962, groups = var_21105, pad = k_217_pad_0, pad_type = k_217_pad_type_0, strides = var_23960, weight = mid_block_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16, x = hidden_states_227_cast)[name = tensor<string, []>("k_217_cast")];
            tensor<int32, [2]> var_23966 = const()[name = tensor<string, []>("op_23966"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_23968 = const()[name = tensor<string, []>("op_23968"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_109_pad_type_0 = const()[name = tensor<string, []>("v_109_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_109_pad_0 = const()[name = tensor<string, []>("v_109_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1941206912)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_109_cast = conv(dilations = var_23968, groups = var_21105, pad = v_109_pad_0, pad_type = v_109_pad_type_0, strides = var_23966, weight = mid_block_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16, x = hidden_states_227_cast)[name = tensor<string, []>("v_109_cast")];
            tensor<int32, [4]> var_23972_begin_0 = const()[name = tensor<string, []>("op_23972_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_23972_end_0 = const()[name = tensor<string, []>("op_23972_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_23972_end_mask_0 = const()[name = tensor<string, []>("op_23972_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23972_cast = slice_by_index(begin = var_23972_begin_0, end = var_23972_end_0, end_mask = var_23972_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_23972_cast")];
            tensor<int32, [4]> var_23976_begin_0 = const()[name = tensor<string, []>("op_23976_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_23976_end_0 = const()[name = tensor<string, []>("op_23976_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_23976_end_mask_0 = const()[name = tensor<string, []>("op_23976_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23976_cast = slice_by_index(begin = var_23976_begin_0, end = var_23976_end_0, end_mask = var_23976_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_23976_cast")];
            tensor<int32, [4]> var_23980_begin_0 = const()[name = tensor<string, []>("op_23980_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_23980_end_0 = const()[name = tensor<string, []>("op_23980_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_23980_end_mask_0 = const()[name = tensor<string, []>("op_23980_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23980_cast = slice_by_index(begin = var_23980_begin_0, end = var_23980_end_0, end_mask = var_23980_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_23980_cast")];
            tensor<int32, [4]> var_23984_begin_0 = const()[name = tensor<string, []>("op_23984_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_23984_end_0 = const()[name = tensor<string, []>("op_23984_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_23984_end_mask_0 = const()[name = tensor<string, []>("op_23984_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23984_cast = slice_by_index(begin = var_23984_begin_0, end = var_23984_end_0, end_mask = var_23984_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_23984_cast")];
            tensor<int32, [4]> var_23988_begin_0 = const()[name = tensor<string, []>("op_23988_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_23988_end_0 = const()[name = tensor<string, []>("op_23988_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_23988_end_mask_0 = const()[name = tensor<string, []>("op_23988_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23988_cast = slice_by_index(begin = var_23988_begin_0, end = var_23988_end_0, end_mask = var_23988_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_23988_cast")];
            tensor<int32, [4]> var_23992_begin_0 = const()[name = tensor<string, []>("op_23992_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_23992_end_0 = const()[name = tensor<string, []>("op_23992_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_23992_end_mask_0 = const()[name = tensor<string, []>("op_23992_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23992_cast = slice_by_index(begin = var_23992_begin_0, end = var_23992_end_0, end_mask = var_23992_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_23992_cast")];
            tensor<int32, [4]> var_23996_begin_0 = const()[name = tensor<string, []>("op_23996_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_23996_end_0 = const()[name = tensor<string, []>("op_23996_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_23996_end_mask_0 = const()[name = tensor<string, []>("op_23996_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_23996_cast = slice_by_index(begin = var_23996_begin_0, end = var_23996_end_0, end_mask = var_23996_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_23996_cast")];
            tensor<int32, [4]> var_24000_begin_0 = const()[name = tensor<string, []>("op_24000_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_24000_end_0 = const()[name = tensor<string, []>("op_24000_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_24000_end_mask_0 = const()[name = tensor<string, []>("op_24000_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24000_cast = slice_by_index(begin = var_24000_begin_0, end = var_24000_end_0, end_mask = var_24000_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_24000_cast")];
            tensor<int32, [4]> var_24004_begin_0 = const()[name = tensor<string, []>("op_24004_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_24004_end_0 = const()[name = tensor<string, []>("op_24004_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_24004_end_mask_0 = const()[name = tensor<string, []>("op_24004_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24004_cast = slice_by_index(begin = var_24004_begin_0, end = var_24004_end_0, end_mask = var_24004_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_24004_cast")];
            tensor<int32, [4]> var_24008_begin_0 = const()[name = tensor<string, []>("op_24008_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_24008_end_0 = const()[name = tensor<string, []>("op_24008_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_24008_end_mask_0 = const()[name = tensor<string, []>("op_24008_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24008_cast = slice_by_index(begin = var_24008_begin_0, end = var_24008_end_0, end_mask = var_24008_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_24008_cast")];
            tensor<int32, [4]> var_24012_begin_0 = const()[name = tensor<string, []>("op_24012_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_24012_end_0 = const()[name = tensor<string, []>("op_24012_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_24012_end_mask_0 = const()[name = tensor<string, []>("op_24012_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24012_cast = slice_by_index(begin = var_24012_begin_0, end = var_24012_end_0, end_mask = var_24012_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_24012_cast")];
            tensor<int32, [4]> var_24016_begin_0 = const()[name = tensor<string, []>("op_24016_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_24016_end_0 = const()[name = tensor<string, []>("op_24016_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_24016_end_mask_0 = const()[name = tensor<string, []>("op_24016_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24016_cast = slice_by_index(begin = var_24016_begin_0, end = var_24016_end_0, end_mask = var_24016_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_24016_cast")];
            tensor<int32, [4]> var_24020_begin_0 = const()[name = tensor<string, []>("op_24020_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_24020_end_0 = const()[name = tensor<string, []>("op_24020_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_24020_end_mask_0 = const()[name = tensor<string, []>("op_24020_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24020_cast = slice_by_index(begin = var_24020_begin_0, end = var_24020_end_0, end_mask = var_24020_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_24020_cast")];
            tensor<int32, [4]> var_24024_begin_0 = const()[name = tensor<string, []>("op_24024_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_24024_end_0 = const()[name = tensor<string, []>("op_24024_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_24024_end_mask_0 = const()[name = tensor<string, []>("op_24024_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24024_cast = slice_by_index(begin = var_24024_begin_0, end = var_24024_end_0, end_mask = var_24024_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_24024_cast")];
            tensor<int32, [4]> var_24028_begin_0 = const()[name = tensor<string, []>("op_24028_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_24028_end_0 = const()[name = tensor<string, []>("op_24028_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_24028_end_mask_0 = const()[name = tensor<string, []>("op_24028_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24028_cast = slice_by_index(begin = var_24028_begin_0, end = var_24028_end_0, end_mask = var_24028_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_24028_cast")];
            tensor<int32, [4]> var_24032_begin_0 = const()[name = tensor<string, []>("op_24032_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_24032_end_0 = const()[name = tensor<string, []>("op_24032_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_24032_end_mask_0 = const()[name = tensor<string, []>("op_24032_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24032_cast = slice_by_index(begin = var_24032_begin_0, end = var_24032_end_0, end_mask = var_24032_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_24032_cast")];
            tensor<int32, [4]> var_24036_begin_0 = const()[name = tensor<string, []>("op_24036_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_24036_end_0 = const()[name = tensor<string, []>("op_24036_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_24036_end_mask_0 = const()[name = tensor<string, []>("op_24036_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24036_cast = slice_by_index(begin = var_24036_begin_0, end = var_24036_end_0, end_mask = var_24036_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_24036_cast")];
            tensor<int32, [4]> var_24040_begin_0 = const()[name = tensor<string, []>("op_24040_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_24040_end_0 = const()[name = tensor<string, []>("op_24040_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_24040_end_mask_0 = const()[name = tensor<string, []>("op_24040_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24040_cast = slice_by_index(begin = var_24040_begin_0, end = var_24040_end_0, end_mask = var_24040_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_24040_cast")];
            tensor<int32, [4]> var_24044_begin_0 = const()[name = tensor<string, []>("op_24044_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_24044_end_0 = const()[name = tensor<string, []>("op_24044_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_24044_end_mask_0 = const()[name = tensor<string, []>("op_24044_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24044_cast = slice_by_index(begin = var_24044_begin_0, end = var_24044_end_0, end_mask = var_24044_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_24044_cast")];
            tensor<int32, [4]> var_24048_begin_0 = const()[name = tensor<string, []>("op_24048_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_24048_end_0 = const()[name = tensor<string, []>("op_24048_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_24048_end_mask_0 = const()[name = tensor<string, []>("op_24048_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24048_cast = slice_by_index(begin = var_24048_begin_0, end = var_24048_end_0, end_mask = var_24048_end_mask_0, x = q_109_cast)[name = tensor<string, []>("op_24048_cast")];
            tensor<int32, [4]> k_219_perm_0 = const()[name = tensor<string, []>("k_219_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_24055_begin_0 = const()[name = tensor<string, []>("op_24055_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_24055_end_0 = const()[name = tensor<string, []>("op_24055_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_24055_end_mask_0 = const()[name = tensor<string, []>("op_24055_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_85 = transpose(perm = k_219_perm_0, x = k_217_cast)[name = tensor<string, []>("transpose_85")];
            tensor<fp16, [2, 1024, 1, 64]> var_24055_cast = slice_by_index(begin = var_24055_begin_0, end = var_24055_end_0, end_mask = var_24055_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24055_cast")];
            tensor<int32, [4]> var_24059_begin_0 = const()[name = tensor<string, []>("op_24059_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_24059_end_0 = const()[name = tensor<string, []>("op_24059_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_24059_end_mask_0 = const()[name = tensor<string, []>("op_24059_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24059_cast = slice_by_index(begin = var_24059_begin_0, end = var_24059_end_0, end_mask = var_24059_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24059_cast")];
            tensor<int32, [4]> var_24063_begin_0 = const()[name = tensor<string, []>("op_24063_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_24063_end_0 = const()[name = tensor<string, []>("op_24063_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_24063_end_mask_0 = const()[name = tensor<string, []>("op_24063_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24063_cast = slice_by_index(begin = var_24063_begin_0, end = var_24063_end_0, end_mask = var_24063_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24063_cast")];
            tensor<int32, [4]> var_24067_begin_0 = const()[name = tensor<string, []>("op_24067_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_24067_end_0 = const()[name = tensor<string, []>("op_24067_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_24067_end_mask_0 = const()[name = tensor<string, []>("op_24067_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24067_cast = slice_by_index(begin = var_24067_begin_0, end = var_24067_end_0, end_mask = var_24067_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24067_cast")];
            tensor<int32, [4]> var_24071_begin_0 = const()[name = tensor<string, []>("op_24071_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_24071_end_0 = const()[name = tensor<string, []>("op_24071_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_24071_end_mask_0 = const()[name = tensor<string, []>("op_24071_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24071_cast = slice_by_index(begin = var_24071_begin_0, end = var_24071_end_0, end_mask = var_24071_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24071_cast")];
            tensor<int32, [4]> var_24075_begin_0 = const()[name = tensor<string, []>("op_24075_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_24075_end_0 = const()[name = tensor<string, []>("op_24075_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_24075_end_mask_0 = const()[name = tensor<string, []>("op_24075_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24075_cast = slice_by_index(begin = var_24075_begin_0, end = var_24075_end_0, end_mask = var_24075_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24075_cast")];
            tensor<int32, [4]> var_24079_begin_0 = const()[name = tensor<string, []>("op_24079_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_24079_end_0 = const()[name = tensor<string, []>("op_24079_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_24079_end_mask_0 = const()[name = tensor<string, []>("op_24079_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24079_cast = slice_by_index(begin = var_24079_begin_0, end = var_24079_end_0, end_mask = var_24079_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24079_cast")];
            tensor<int32, [4]> var_24083_begin_0 = const()[name = tensor<string, []>("op_24083_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_24083_end_0 = const()[name = tensor<string, []>("op_24083_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_24083_end_mask_0 = const()[name = tensor<string, []>("op_24083_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24083_cast = slice_by_index(begin = var_24083_begin_0, end = var_24083_end_0, end_mask = var_24083_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24083_cast")];
            tensor<int32, [4]> var_24087_begin_0 = const()[name = tensor<string, []>("op_24087_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_24087_end_0 = const()[name = tensor<string, []>("op_24087_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_24087_end_mask_0 = const()[name = tensor<string, []>("op_24087_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24087_cast = slice_by_index(begin = var_24087_begin_0, end = var_24087_end_0, end_mask = var_24087_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24087_cast")];
            tensor<int32, [4]> var_24091_begin_0 = const()[name = tensor<string, []>("op_24091_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_24091_end_0 = const()[name = tensor<string, []>("op_24091_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_24091_end_mask_0 = const()[name = tensor<string, []>("op_24091_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24091_cast = slice_by_index(begin = var_24091_begin_0, end = var_24091_end_0, end_mask = var_24091_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24091_cast")];
            tensor<int32, [4]> var_24095_begin_0 = const()[name = tensor<string, []>("op_24095_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_24095_end_0 = const()[name = tensor<string, []>("op_24095_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_24095_end_mask_0 = const()[name = tensor<string, []>("op_24095_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24095_cast = slice_by_index(begin = var_24095_begin_0, end = var_24095_end_0, end_mask = var_24095_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24095_cast")];
            tensor<int32, [4]> var_24099_begin_0 = const()[name = tensor<string, []>("op_24099_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_24099_end_0 = const()[name = tensor<string, []>("op_24099_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_24099_end_mask_0 = const()[name = tensor<string, []>("op_24099_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24099_cast = slice_by_index(begin = var_24099_begin_0, end = var_24099_end_0, end_mask = var_24099_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24099_cast")];
            tensor<int32, [4]> var_24103_begin_0 = const()[name = tensor<string, []>("op_24103_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_24103_end_0 = const()[name = tensor<string, []>("op_24103_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_24103_end_mask_0 = const()[name = tensor<string, []>("op_24103_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24103_cast = slice_by_index(begin = var_24103_begin_0, end = var_24103_end_0, end_mask = var_24103_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24103_cast")];
            tensor<int32, [4]> var_24107_begin_0 = const()[name = tensor<string, []>("op_24107_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_24107_end_0 = const()[name = tensor<string, []>("op_24107_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_24107_end_mask_0 = const()[name = tensor<string, []>("op_24107_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24107_cast = slice_by_index(begin = var_24107_begin_0, end = var_24107_end_0, end_mask = var_24107_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24107_cast")];
            tensor<int32, [4]> var_24111_begin_0 = const()[name = tensor<string, []>("op_24111_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_24111_end_0 = const()[name = tensor<string, []>("op_24111_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_24111_end_mask_0 = const()[name = tensor<string, []>("op_24111_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24111_cast = slice_by_index(begin = var_24111_begin_0, end = var_24111_end_0, end_mask = var_24111_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24111_cast")];
            tensor<int32, [4]> var_24115_begin_0 = const()[name = tensor<string, []>("op_24115_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_24115_end_0 = const()[name = tensor<string, []>("op_24115_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_24115_end_mask_0 = const()[name = tensor<string, []>("op_24115_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24115_cast = slice_by_index(begin = var_24115_begin_0, end = var_24115_end_0, end_mask = var_24115_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24115_cast")];
            tensor<int32, [4]> var_24119_begin_0 = const()[name = tensor<string, []>("op_24119_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_24119_end_0 = const()[name = tensor<string, []>("op_24119_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_24119_end_mask_0 = const()[name = tensor<string, []>("op_24119_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24119_cast = slice_by_index(begin = var_24119_begin_0, end = var_24119_end_0, end_mask = var_24119_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24119_cast")];
            tensor<int32, [4]> var_24123_begin_0 = const()[name = tensor<string, []>("op_24123_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_24123_end_0 = const()[name = tensor<string, []>("op_24123_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_24123_end_mask_0 = const()[name = tensor<string, []>("op_24123_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24123_cast = slice_by_index(begin = var_24123_begin_0, end = var_24123_end_0, end_mask = var_24123_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24123_cast")];
            tensor<int32, [4]> var_24127_begin_0 = const()[name = tensor<string, []>("op_24127_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_24127_end_0 = const()[name = tensor<string, []>("op_24127_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_24127_end_mask_0 = const()[name = tensor<string, []>("op_24127_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24127_cast = slice_by_index(begin = var_24127_begin_0, end = var_24127_end_0, end_mask = var_24127_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24127_cast")];
            tensor<int32, [4]> var_24131_begin_0 = const()[name = tensor<string, []>("op_24131_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_24131_end_0 = const()[name = tensor<string, []>("op_24131_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_24131_end_mask_0 = const()[name = tensor<string, []>("op_24131_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24131_cast = slice_by_index(begin = var_24131_begin_0, end = var_24131_end_0, end_mask = var_24131_end_mask_0, x = transpose_85)[name = tensor<string, []>("op_24131_cast")];
            tensor<int32, [4]> var_24133_begin_0 = const()[name = tensor<string, []>("op_24133_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_24133_end_0 = const()[name = tensor<string, []>("op_24133_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_24133_end_mask_0 = const()[name = tensor<string, []>("op_24133_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24133_cast = slice_by_index(begin = var_24133_begin_0, end = var_24133_end_0, end_mask = var_24133_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24133_cast")];
            tensor<int32, [4]> var_24137_begin_0 = const()[name = tensor<string, []>("op_24137_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_24137_end_0 = const()[name = tensor<string, []>("op_24137_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_24137_end_mask_0 = const()[name = tensor<string, []>("op_24137_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24137_cast = slice_by_index(begin = var_24137_begin_0, end = var_24137_end_0, end_mask = var_24137_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24137_cast")];
            tensor<int32, [4]> var_24141_begin_0 = const()[name = tensor<string, []>("op_24141_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_24141_end_0 = const()[name = tensor<string, []>("op_24141_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_24141_end_mask_0 = const()[name = tensor<string, []>("op_24141_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24141_cast = slice_by_index(begin = var_24141_begin_0, end = var_24141_end_0, end_mask = var_24141_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24141_cast")];
            tensor<int32, [4]> var_24145_begin_0 = const()[name = tensor<string, []>("op_24145_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_24145_end_0 = const()[name = tensor<string, []>("op_24145_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_24145_end_mask_0 = const()[name = tensor<string, []>("op_24145_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24145_cast = slice_by_index(begin = var_24145_begin_0, end = var_24145_end_0, end_mask = var_24145_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24145_cast")];
            tensor<int32, [4]> var_24149_begin_0 = const()[name = tensor<string, []>("op_24149_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_24149_end_0 = const()[name = tensor<string, []>("op_24149_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_24149_end_mask_0 = const()[name = tensor<string, []>("op_24149_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24149_cast = slice_by_index(begin = var_24149_begin_0, end = var_24149_end_0, end_mask = var_24149_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24149_cast")];
            tensor<int32, [4]> var_24153_begin_0 = const()[name = tensor<string, []>("op_24153_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_24153_end_0 = const()[name = tensor<string, []>("op_24153_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_24153_end_mask_0 = const()[name = tensor<string, []>("op_24153_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24153_cast = slice_by_index(begin = var_24153_begin_0, end = var_24153_end_0, end_mask = var_24153_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24153_cast")];
            tensor<int32, [4]> var_24157_begin_0 = const()[name = tensor<string, []>("op_24157_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_24157_end_0 = const()[name = tensor<string, []>("op_24157_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_24157_end_mask_0 = const()[name = tensor<string, []>("op_24157_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24157_cast = slice_by_index(begin = var_24157_begin_0, end = var_24157_end_0, end_mask = var_24157_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24157_cast")];
            tensor<int32, [4]> var_24161_begin_0 = const()[name = tensor<string, []>("op_24161_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_24161_end_0 = const()[name = tensor<string, []>("op_24161_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_24161_end_mask_0 = const()[name = tensor<string, []>("op_24161_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24161_cast = slice_by_index(begin = var_24161_begin_0, end = var_24161_end_0, end_mask = var_24161_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24161_cast")];
            tensor<int32, [4]> var_24165_begin_0 = const()[name = tensor<string, []>("op_24165_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_24165_end_0 = const()[name = tensor<string, []>("op_24165_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_24165_end_mask_0 = const()[name = tensor<string, []>("op_24165_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24165_cast = slice_by_index(begin = var_24165_begin_0, end = var_24165_end_0, end_mask = var_24165_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24165_cast")];
            tensor<int32, [4]> var_24169_begin_0 = const()[name = tensor<string, []>("op_24169_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_24169_end_0 = const()[name = tensor<string, []>("op_24169_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_24169_end_mask_0 = const()[name = tensor<string, []>("op_24169_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24169_cast = slice_by_index(begin = var_24169_begin_0, end = var_24169_end_0, end_mask = var_24169_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24169_cast")];
            tensor<int32, [4]> var_24173_begin_0 = const()[name = tensor<string, []>("op_24173_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_24173_end_0 = const()[name = tensor<string, []>("op_24173_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_24173_end_mask_0 = const()[name = tensor<string, []>("op_24173_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24173_cast = slice_by_index(begin = var_24173_begin_0, end = var_24173_end_0, end_mask = var_24173_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24173_cast")];
            tensor<int32, [4]> var_24177_begin_0 = const()[name = tensor<string, []>("op_24177_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_24177_end_0 = const()[name = tensor<string, []>("op_24177_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_24177_end_mask_0 = const()[name = tensor<string, []>("op_24177_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24177_cast = slice_by_index(begin = var_24177_begin_0, end = var_24177_end_0, end_mask = var_24177_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24177_cast")];
            tensor<int32, [4]> var_24181_begin_0 = const()[name = tensor<string, []>("op_24181_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_24181_end_0 = const()[name = tensor<string, []>("op_24181_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_24181_end_mask_0 = const()[name = tensor<string, []>("op_24181_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24181_cast = slice_by_index(begin = var_24181_begin_0, end = var_24181_end_0, end_mask = var_24181_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24181_cast")];
            tensor<int32, [4]> var_24185_begin_0 = const()[name = tensor<string, []>("op_24185_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_24185_end_0 = const()[name = tensor<string, []>("op_24185_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_24185_end_mask_0 = const()[name = tensor<string, []>("op_24185_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24185_cast = slice_by_index(begin = var_24185_begin_0, end = var_24185_end_0, end_mask = var_24185_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24185_cast")];
            tensor<int32, [4]> var_24189_begin_0 = const()[name = tensor<string, []>("op_24189_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_24189_end_0 = const()[name = tensor<string, []>("op_24189_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_24189_end_mask_0 = const()[name = tensor<string, []>("op_24189_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24189_cast = slice_by_index(begin = var_24189_begin_0, end = var_24189_end_0, end_mask = var_24189_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24189_cast")];
            tensor<int32, [4]> var_24193_begin_0 = const()[name = tensor<string, []>("op_24193_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_24193_end_0 = const()[name = tensor<string, []>("op_24193_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_24193_end_mask_0 = const()[name = tensor<string, []>("op_24193_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24193_cast = slice_by_index(begin = var_24193_begin_0, end = var_24193_end_0, end_mask = var_24193_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24193_cast")];
            tensor<int32, [4]> var_24197_begin_0 = const()[name = tensor<string, []>("op_24197_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_24197_end_0 = const()[name = tensor<string, []>("op_24197_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_24197_end_mask_0 = const()[name = tensor<string, []>("op_24197_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24197_cast = slice_by_index(begin = var_24197_begin_0, end = var_24197_end_0, end_mask = var_24197_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24197_cast")];
            tensor<int32, [4]> var_24201_begin_0 = const()[name = tensor<string, []>("op_24201_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_24201_end_0 = const()[name = tensor<string, []>("op_24201_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_24201_end_mask_0 = const()[name = tensor<string, []>("op_24201_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24201_cast = slice_by_index(begin = var_24201_begin_0, end = var_24201_end_0, end_mask = var_24201_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24201_cast")];
            tensor<int32, [4]> var_24205_begin_0 = const()[name = tensor<string, []>("op_24205_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_24205_end_0 = const()[name = tensor<string, []>("op_24205_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_24205_end_mask_0 = const()[name = tensor<string, []>("op_24205_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24205_cast = slice_by_index(begin = var_24205_begin_0, end = var_24205_end_0, end_mask = var_24205_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24205_cast")];
            tensor<int32, [4]> var_24209_begin_0 = const()[name = tensor<string, []>("op_24209_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_24209_end_0 = const()[name = tensor<string, []>("op_24209_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_24209_end_mask_0 = const()[name = tensor<string, []>("op_24209_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24209_cast = slice_by_index(begin = var_24209_begin_0, end = var_24209_end_0, end_mask = var_24209_end_mask_0, x = v_109_cast)[name = tensor<string, []>("op_24209_cast")];
            tensor<string, []> var_24213_equation_0 = const()[name = tensor<string, []>("op_24213_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24213_cast = einsum(equation = var_24213_equation_0, values = (var_24055_cast, var_23972_cast))[name = tensor<string, []>("op_24213_cast")];
            tensor<fp16, []> var_24214_to_fp16 = const()[name = tensor<string, []>("op_24214_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2001_cast = mul(x = var_24213_cast, y = var_24214_to_fp16)[name = tensor<string, []>("aw_2001_cast")];
            tensor<string, []> var_24217_equation_0 = const()[name = tensor<string, []>("op_24217_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24217_cast = einsum(equation = var_24217_equation_0, values = (var_24059_cast, var_23976_cast))[name = tensor<string, []>("op_24217_cast")];
            tensor<fp16, []> var_24218_to_fp16 = const()[name = tensor<string, []>("op_24218_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2003_cast = mul(x = var_24217_cast, y = var_24218_to_fp16)[name = tensor<string, []>("aw_2003_cast")];
            tensor<string, []> var_24221_equation_0 = const()[name = tensor<string, []>("op_24221_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24221_cast = einsum(equation = var_24221_equation_0, values = (var_24063_cast, var_23980_cast))[name = tensor<string, []>("op_24221_cast")];
            tensor<fp16, []> var_24222_to_fp16 = const()[name = tensor<string, []>("op_24222_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2005_cast = mul(x = var_24221_cast, y = var_24222_to_fp16)[name = tensor<string, []>("aw_2005_cast")];
            tensor<string, []> var_24225_equation_0 = const()[name = tensor<string, []>("op_24225_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24225_cast = einsum(equation = var_24225_equation_0, values = (var_24067_cast, var_23984_cast))[name = tensor<string, []>("op_24225_cast")];
            tensor<fp16, []> var_24226_to_fp16 = const()[name = tensor<string, []>("op_24226_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2007_cast = mul(x = var_24225_cast, y = var_24226_to_fp16)[name = tensor<string, []>("aw_2007_cast")];
            tensor<string, []> var_24229_equation_0 = const()[name = tensor<string, []>("op_24229_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24229_cast = einsum(equation = var_24229_equation_0, values = (var_24071_cast, var_23988_cast))[name = tensor<string, []>("op_24229_cast")];
            tensor<fp16, []> var_24230_to_fp16 = const()[name = tensor<string, []>("op_24230_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2009_cast = mul(x = var_24229_cast, y = var_24230_to_fp16)[name = tensor<string, []>("aw_2009_cast")];
            tensor<string, []> var_24233_equation_0 = const()[name = tensor<string, []>("op_24233_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24233_cast = einsum(equation = var_24233_equation_0, values = (var_24075_cast, var_23992_cast))[name = tensor<string, []>("op_24233_cast")];
            tensor<fp16, []> var_24234_to_fp16 = const()[name = tensor<string, []>("op_24234_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2011_cast = mul(x = var_24233_cast, y = var_24234_to_fp16)[name = tensor<string, []>("aw_2011_cast")];
            tensor<string, []> var_24237_equation_0 = const()[name = tensor<string, []>("op_24237_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24237_cast = einsum(equation = var_24237_equation_0, values = (var_24079_cast, var_23996_cast))[name = tensor<string, []>("op_24237_cast")];
            tensor<fp16, []> var_24238_to_fp16 = const()[name = tensor<string, []>("op_24238_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2013_cast = mul(x = var_24237_cast, y = var_24238_to_fp16)[name = tensor<string, []>("aw_2013_cast")];
            tensor<string, []> var_24241_equation_0 = const()[name = tensor<string, []>("op_24241_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24241_cast = einsum(equation = var_24241_equation_0, values = (var_24083_cast, var_24000_cast))[name = tensor<string, []>("op_24241_cast")];
            tensor<fp16, []> var_24242_to_fp16 = const()[name = tensor<string, []>("op_24242_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2015_cast = mul(x = var_24241_cast, y = var_24242_to_fp16)[name = tensor<string, []>("aw_2015_cast")];
            tensor<string, []> var_24245_equation_0 = const()[name = tensor<string, []>("op_24245_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24245_cast = einsum(equation = var_24245_equation_0, values = (var_24087_cast, var_24004_cast))[name = tensor<string, []>("op_24245_cast")];
            tensor<fp16, []> var_24246_to_fp16 = const()[name = tensor<string, []>("op_24246_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2017_cast = mul(x = var_24245_cast, y = var_24246_to_fp16)[name = tensor<string, []>("aw_2017_cast")];
            tensor<string, []> var_24249_equation_0 = const()[name = tensor<string, []>("op_24249_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24249_cast = einsum(equation = var_24249_equation_0, values = (var_24091_cast, var_24008_cast))[name = tensor<string, []>("op_24249_cast")];
            tensor<fp16, []> var_24250_to_fp16 = const()[name = tensor<string, []>("op_24250_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2019_cast = mul(x = var_24249_cast, y = var_24250_to_fp16)[name = tensor<string, []>("aw_2019_cast")];
            tensor<string, []> var_24253_equation_0 = const()[name = tensor<string, []>("op_24253_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24253_cast = einsum(equation = var_24253_equation_0, values = (var_24095_cast, var_24012_cast))[name = tensor<string, []>("op_24253_cast")];
            tensor<fp16, []> var_24254_to_fp16 = const()[name = tensor<string, []>("op_24254_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2021_cast = mul(x = var_24253_cast, y = var_24254_to_fp16)[name = tensor<string, []>("aw_2021_cast")];
            tensor<string, []> var_24257_equation_0 = const()[name = tensor<string, []>("op_24257_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24257_cast = einsum(equation = var_24257_equation_0, values = (var_24099_cast, var_24016_cast))[name = tensor<string, []>("op_24257_cast")];
            tensor<fp16, []> var_24258_to_fp16 = const()[name = tensor<string, []>("op_24258_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2023_cast = mul(x = var_24257_cast, y = var_24258_to_fp16)[name = tensor<string, []>("aw_2023_cast")];
            tensor<string, []> var_24261_equation_0 = const()[name = tensor<string, []>("op_24261_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24261_cast = einsum(equation = var_24261_equation_0, values = (var_24103_cast, var_24020_cast))[name = tensor<string, []>("op_24261_cast")];
            tensor<fp16, []> var_24262_to_fp16 = const()[name = tensor<string, []>("op_24262_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2025_cast = mul(x = var_24261_cast, y = var_24262_to_fp16)[name = tensor<string, []>("aw_2025_cast")];
            tensor<string, []> var_24265_equation_0 = const()[name = tensor<string, []>("op_24265_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24265_cast = einsum(equation = var_24265_equation_0, values = (var_24107_cast, var_24024_cast))[name = tensor<string, []>("op_24265_cast")];
            tensor<fp16, []> var_24266_to_fp16 = const()[name = tensor<string, []>("op_24266_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2027_cast = mul(x = var_24265_cast, y = var_24266_to_fp16)[name = tensor<string, []>("aw_2027_cast")];
            tensor<string, []> var_24269_equation_0 = const()[name = tensor<string, []>("op_24269_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24269_cast = einsum(equation = var_24269_equation_0, values = (var_24111_cast, var_24028_cast))[name = tensor<string, []>("op_24269_cast")];
            tensor<fp16, []> var_24270_to_fp16 = const()[name = tensor<string, []>("op_24270_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2029_cast = mul(x = var_24269_cast, y = var_24270_to_fp16)[name = tensor<string, []>("aw_2029_cast")];
            tensor<string, []> var_24273_equation_0 = const()[name = tensor<string, []>("op_24273_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24273_cast = einsum(equation = var_24273_equation_0, values = (var_24115_cast, var_24032_cast))[name = tensor<string, []>("op_24273_cast")];
            tensor<fp16, []> var_24274_to_fp16 = const()[name = tensor<string, []>("op_24274_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2031_cast = mul(x = var_24273_cast, y = var_24274_to_fp16)[name = tensor<string, []>("aw_2031_cast")];
            tensor<string, []> var_24277_equation_0 = const()[name = tensor<string, []>("op_24277_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24277_cast = einsum(equation = var_24277_equation_0, values = (var_24119_cast, var_24036_cast))[name = tensor<string, []>("op_24277_cast")];
            tensor<fp16, []> var_24278_to_fp16 = const()[name = tensor<string, []>("op_24278_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2033_cast = mul(x = var_24277_cast, y = var_24278_to_fp16)[name = tensor<string, []>("aw_2033_cast")];
            tensor<string, []> var_24281_equation_0 = const()[name = tensor<string, []>("op_24281_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24281_cast = einsum(equation = var_24281_equation_0, values = (var_24123_cast, var_24040_cast))[name = tensor<string, []>("op_24281_cast")];
            tensor<fp16, []> var_24282_to_fp16 = const()[name = tensor<string, []>("op_24282_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2035_cast = mul(x = var_24281_cast, y = var_24282_to_fp16)[name = tensor<string, []>("aw_2035_cast")];
            tensor<string, []> var_24285_equation_0 = const()[name = tensor<string, []>("op_24285_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24285_cast = einsum(equation = var_24285_equation_0, values = (var_24127_cast, var_24044_cast))[name = tensor<string, []>("op_24285_cast")];
            tensor<fp16, []> var_24286_to_fp16 = const()[name = tensor<string, []>("op_24286_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2037_cast = mul(x = var_24285_cast, y = var_24286_to_fp16)[name = tensor<string, []>("aw_2037_cast")];
            tensor<string, []> var_24289_equation_0 = const()[name = tensor<string, []>("op_24289_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24289_cast = einsum(equation = var_24289_equation_0, values = (var_24131_cast, var_24048_cast))[name = tensor<string, []>("op_24289_cast")];
            tensor<fp16, []> var_24290_to_fp16 = const()[name = tensor<string, []>("op_24290_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2039_cast = mul(x = var_24289_cast, y = var_24290_to_fp16)[name = tensor<string, []>("aw_2039_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24292_cast = softmax(axis = var_21105, x = aw_2001_cast)[name = tensor<string, []>("op_24292_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24293_cast = softmax(axis = var_21105, x = aw_2003_cast)[name = tensor<string, []>("op_24293_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24294_cast = softmax(axis = var_21105, x = aw_2005_cast)[name = tensor<string, []>("op_24294_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24295_cast = softmax(axis = var_21105, x = aw_2007_cast)[name = tensor<string, []>("op_24295_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24296_cast = softmax(axis = var_21105, x = aw_2009_cast)[name = tensor<string, []>("op_24296_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24297_cast = softmax(axis = var_21105, x = aw_2011_cast)[name = tensor<string, []>("op_24297_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24298_cast = softmax(axis = var_21105, x = aw_2013_cast)[name = tensor<string, []>("op_24298_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24299_cast = softmax(axis = var_21105, x = aw_2015_cast)[name = tensor<string, []>("op_24299_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24300_cast = softmax(axis = var_21105, x = aw_2017_cast)[name = tensor<string, []>("op_24300_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24301_cast = softmax(axis = var_21105, x = aw_2019_cast)[name = tensor<string, []>("op_24301_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24302_cast = softmax(axis = var_21105, x = aw_2021_cast)[name = tensor<string, []>("op_24302_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24303_cast = softmax(axis = var_21105, x = aw_2023_cast)[name = tensor<string, []>("op_24303_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24304_cast = softmax(axis = var_21105, x = aw_2025_cast)[name = tensor<string, []>("op_24304_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24305_cast = softmax(axis = var_21105, x = aw_2027_cast)[name = tensor<string, []>("op_24305_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24306_cast = softmax(axis = var_21105, x = aw_2029_cast)[name = tensor<string, []>("op_24306_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24307_cast = softmax(axis = var_21105, x = aw_2031_cast)[name = tensor<string, []>("op_24307_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24308_cast = softmax(axis = var_21105, x = aw_2033_cast)[name = tensor<string, []>("op_24308_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24309_cast = softmax(axis = var_21105, x = aw_2035_cast)[name = tensor<string, []>("op_24309_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24310_cast = softmax(axis = var_21105, x = aw_2037_cast)[name = tensor<string, []>("op_24310_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_24311_cast = softmax(axis = var_21105, x = aw_2039_cast)[name = tensor<string, []>("op_24311_cast")];
            tensor<string, []> var_24313_equation_0 = const()[name = tensor<string, []>("op_24313_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24313_cast = einsum(equation = var_24313_equation_0, values = (var_24133_cast, var_24292_cast))[name = tensor<string, []>("op_24313_cast")];
            tensor<string, []> var_24315_equation_0 = const()[name = tensor<string, []>("op_24315_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24315_cast = einsum(equation = var_24315_equation_0, values = (var_24137_cast, var_24293_cast))[name = tensor<string, []>("op_24315_cast")];
            tensor<string, []> var_24317_equation_0 = const()[name = tensor<string, []>("op_24317_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24317_cast = einsum(equation = var_24317_equation_0, values = (var_24141_cast, var_24294_cast))[name = tensor<string, []>("op_24317_cast")];
            tensor<string, []> var_24319_equation_0 = const()[name = tensor<string, []>("op_24319_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24319_cast = einsum(equation = var_24319_equation_0, values = (var_24145_cast, var_24295_cast))[name = tensor<string, []>("op_24319_cast")];
            tensor<string, []> var_24321_equation_0 = const()[name = tensor<string, []>("op_24321_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24321_cast = einsum(equation = var_24321_equation_0, values = (var_24149_cast, var_24296_cast))[name = tensor<string, []>("op_24321_cast")];
            tensor<string, []> var_24323_equation_0 = const()[name = tensor<string, []>("op_24323_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24323_cast = einsum(equation = var_24323_equation_0, values = (var_24153_cast, var_24297_cast))[name = tensor<string, []>("op_24323_cast")];
            tensor<string, []> var_24325_equation_0 = const()[name = tensor<string, []>("op_24325_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24325_cast = einsum(equation = var_24325_equation_0, values = (var_24157_cast, var_24298_cast))[name = tensor<string, []>("op_24325_cast")];
            tensor<string, []> var_24327_equation_0 = const()[name = tensor<string, []>("op_24327_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24327_cast = einsum(equation = var_24327_equation_0, values = (var_24161_cast, var_24299_cast))[name = tensor<string, []>("op_24327_cast")];
            tensor<string, []> var_24329_equation_0 = const()[name = tensor<string, []>("op_24329_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24329_cast = einsum(equation = var_24329_equation_0, values = (var_24165_cast, var_24300_cast))[name = tensor<string, []>("op_24329_cast")];
            tensor<string, []> var_24331_equation_0 = const()[name = tensor<string, []>("op_24331_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24331_cast = einsum(equation = var_24331_equation_0, values = (var_24169_cast, var_24301_cast))[name = tensor<string, []>("op_24331_cast")];
            tensor<string, []> var_24333_equation_0 = const()[name = tensor<string, []>("op_24333_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24333_cast = einsum(equation = var_24333_equation_0, values = (var_24173_cast, var_24302_cast))[name = tensor<string, []>("op_24333_cast")];
            tensor<string, []> var_24335_equation_0 = const()[name = tensor<string, []>("op_24335_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24335_cast = einsum(equation = var_24335_equation_0, values = (var_24177_cast, var_24303_cast))[name = tensor<string, []>("op_24335_cast")];
            tensor<string, []> var_24337_equation_0 = const()[name = tensor<string, []>("op_24337_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24337_cast = einsum(equation = var_24337_equation_0, values = (var_24181_cast, var_24304_cast))[name = tensor<string, []>("op_24337_cast")];
            tensor<string, []> var_24339_equation_0 = const()[name = tensor<string, []>("op_24339_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24339_cast = einsum(equation = var_24339_equation_0, values = (var_24185_cast, var_24305_cast))[name = tensor<string, []>("op_24339_cast")];
            tensor<string, []> var_24341_equation_0 = const()[name = tensor<string, []>("op_24341_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24341_cast = einsum(equation = var_24341_equation_0, values = (var_24189_cast, var_24306_cast))[name = tensor<string, []>("op_24341_cast")];
            tensor<string, []> var_24343_equation_0 = const()[name = tensor<string, []>("op_24343_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24343_cast = einsum(equation = var_24343_equation_0, values = (var_24193_cast, var_24307_cast))[name = tensor<string, []>("op_24343_cast")];
            tensor<string, []> var_24345_equation_0 = const()[name = tensor<string, []>("op_24345_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24345_cast = einsum(equation = var_24345_equation_0, values = (var_24197_cast, var_24308_cast))[name = tensor<string, []>("op_24345_cast")];
            tensor<string, []> var_24347_equation_0 = const()[name = tensor<string, []>("op_24347_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24347_cast = einsum(equation = var_24347_equation_0, values = (var_24201_cast, var_24309_cast))[name = tensor<string, []>("op_24347_cast")];
            tensor<string, []> var_24349_equation_0 = const()[name = tensor<string, []>("op_24349_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24349_cast = einsum(equation = var_24349_equation_0, values = (var_24205_cast, var_24310_cast))[name = tensor<string, []>("op_24349_cast")];
            tensor<string, []> var_24351_equation_0 = const()[name = tensor<string, []>("op_24351_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24351_cast = einsum(equation = var_24351_equation_0, values = (var_24209_cast, var_24311_cast))[name = tensor<string, []>("op_24351_cast")];
            tensor<bool, []> input_353_interleave_0 = const()[name = tensor<string, []>("input_353_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_353_cast = concat(axis = var_21105, interleave = input_353_interleave_0, values = (var_24313_cast, var_24315_cast, var_24317_cast, var_24319_cast, var_24321_cast, var_24323_cast, var_24325_cast, var_24327_cast, var_24329_cast, var_24331_cast, var_24333_cast, var_24335_cast, var_24337_cast, var_24339_cast, var_24341_cast, var_24343_cast, var_24345_cast, var_24347_cast, var_24349_cast, var_24351_cast))[name = tensor<string, []>("input_353_cast")];
            tensor<int32, [2]> var_24357 = const()[name = tensor<string, []>("op_24357"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_24359 = const()[name = tensor<string, []>("op_24359"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_24361_pad_type_0 = const()[name = tensor<string, []>("op_24361_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_24361_pad_0 = const()[name = tensor<string, []>("op_24361_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1944483776)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1947760640)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_24361_cast = conv(bias = mid_block_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16, dilations = var_24359, groups = var_21105, pad = var_24361_pad_0, pad_type = var_24361_pad_type_0, strides = var_24357, weight = mid_block_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16, x = input_353_cast)[name = tensor<string, []>("op_24361_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_165_cast = add(x = var_24361_cast, y = inputs_163_cast)[name = tensor<string, []>("inputs_165_cast")];
            tensor<int32, [1]> var_24365 = const()[name = tensor<string, []>("op_24365"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_165_cast = reduce_mean(axes = var_24365, keep_dims = var_21100, x = inputs_165_cast)[name = tensor<string, []>("channels_mean_165_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_165_cast = sub(x = inputs_165_cast, y = channels_mean_165_cast)[name = tensor<string, []>("zero_mean_165_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_165_cast = mul(x = zero_mean_165_cast, y = zero_mean_165_cast)[name = tensor<string, []>("zero_mean_sq_165_cast")];
            tensor<int32, [1]> var_24369 = const()[name = tensor<string, []>("op_24369"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_24370_cast = reduce_mean(axes = var_24369, keep_dims = var_21100, x = zero_mean_sq_165_cast)[name = tensor<string, []>("op_24370_cast")];
            tensor<fp16, []> var_24371_to_fp16 = const()[name = tensor<string, []>("op_24371_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_24372_cast = add(x = var_24370_cast, y = var_24371_to_fp16)[name = tensor<string, []>("op_24372_cast")];
            tensor<fp16, []> denom_165_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_165_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_165_cast = rsqrt(epsilon = denom_165_epsilon_0_to_fp16, x = var_24372_cast)[name = tensor<string, []>("denom_165_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_165_cast = mul(x = zero_mean_165_cast, y = denom_165_cast)[name = tensor<string, []>("out_165_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_24376_to_fp16 = const()[name = tensor<string, []>("op_24376_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1947763264)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_24377_cast = add(x = out_165_cast, y = var_24376_to_fp16)[name = tensor<string, []>("op_24377_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_24379_to_fp16 = const()[name = tensor<string, []>("op_24379_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1947765888)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_229_cast = mul(x = var_24377_cast, y = var_24379_to_fp16)[name = tensor<string, []>("hidden_states_229_cast")];
            tensor<int32, [2]> var_24386 = const()[name = tensor<string, []>("op_24386"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_24388 = const()[name = tensor<string, []>("op_24388"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_111_pad_type_0 = const()[name = tensor<string, []>("q_111_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_111_pad_0 = const()[name = tensor<string, []>("q_111_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1947768512)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_111_cast = conv(dilations = var_24388, groups = var_21105, pad = q_111_pad_0, pad_type = q_111_pad_type_0, strides = var_24386, weight = mid_block_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16, x = hidden_states_229_cast)[name = tensor<string, []>("q_111_cast")];
            tensor<int32, [2]> var_24392 = const()[name = tensor<string, []>("op_24392"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_24394 = const()[name = tensor<string, []>("op_24394"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_221_pad_type_0 = const()[name = tensor<string, []>("k_221_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_221_pad_0 = const()[name = tensor<string, []>("k_221_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1951045376)))];
            tensor<fp16, [2, 1280, 1, 77]> k_221_cast = conv(dilations = var_24394, groups = var_21105, pad = k_221_pad_0, pad_type = k_221_pad_type_0, strides = var_24392, weight = mid_block_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_221_cast")];
            tensor<int32, [2]> var_24398 = const()[name = tensor<string, []>("op_24398"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_24400 = const()[name = tensor<string, []>("op_24400"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_111_pad_type_0 = const()[name = tensor<string, []>("v_111_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_111_pad_0 = const()[name = tensor<string, []>("v_111_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1956288320)))];
            tensor<fp16, [2, 1280, 1, 77]> v_111_cast = conv(dilations = var_24400, groups = var_21105, pad = v_111_pad_0, pad_type = v_111_pad_type_0, strides = var_24398, weight = mid_block_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_111_cast")];
            tensor<int32, [4]> var_24404_begin_0 = const()[name = tensor<string, []>("op_24404_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_24404_end_0 = const()[name = tensor<string, []>("op_24404_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_24404_end_mask_0 = const()[name = tensor<string, []>("op_24404_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24404_cast = slice_by_index(begin = var_24404_begin_0, end = var_24404_end_0, end_mask = var_24404_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24404_cast")];
            tensor<int32, [4]> var_24408_begin_0 = const()[name = tensor<string, []>("op_24408_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_24408_end_0 = const()[name = tensor<string, []>("op_24408_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_24408_end_mask_0 = const()[name = tensor<string, []>("op_24408_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24408_cast = slice_by_index(begin = var_24408_begin_0, end = var_24408_end_0, end_mask = var_24408_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24408_cast")];
            tensor<int32, [4]> var_24412_begin_0 = const()[name = tensor<string, []>("op_24412_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_24412_end_0 = const()[name = tensor<string, []>("op_24412_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_24412_end_mask_0 = const()[name = tensor<string, []>("op_24412_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24412_cast = slice_by_index(begin = var_24412_begin_0, end = var_24412_end_0, end_mask = var_24412_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24412_cast")];
            tensor<int32, [4]> var_24416_begin_0 = const()[name = tensor<string, []>("op_24416_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_24416_end_0 = const()[name = tensor<string, []>("op_24416_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_24416_end_mask_0 = const()[name = tensor<string, []>("op_24416_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24416_cast = slice_by_index(begin = var_24416_begin_0, end = var_24416_end_0, end_mask = var_24416_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24416_cast")];
            tensor<int32, [4]> var_24420_begin_0 = const()[name = tensor<string, []>("op_24420_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_24420_end_0 = const()[name = tensor<string, []>("op_24420_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_24420_end_mask_0 = const()[name = tensor<string, []>("op_24420_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24420_cast = slice_by_index(begin = var_24420_begin_0, end = var_24420_end_0, end_mask = var_24420_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24420_cast")];
            tensor<int32, [4]> var_24424_begin_0 = const()[name = tensor<string, []>("op_24424_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_24424_end_0 = const()[name = tensor<string, []>("op_24424_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_24424_end_mask_0 = const()[name = tensor<string, []>("op_24424_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24424_cast = slice_by_index(begin = var_24424_begin_0, end = var_24424_end_0, end_mask = var_24424_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24424_cast")];
            tensor<int32, [4]> var_24428_begin_0 = const()[name = tensor<string, []>("op_24428_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_24428_end_0 = const()[name = tensor<string, []>("op_24428_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_24428_end_mask_0 = const()[name = tensor<string, []>("op_24428_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24428_cast = slice_by_index(begin = var_24428_begin_0, end = var_24428_end_0, end_mask = var_24428_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24428_cast")];
            tensor<int32, [4]> var_24432_begin_0 = const()[name = tensor<string, []>("op_24432_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_24432_end_0 = const()[name = tensor<string, []>("op_24432_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_24432_end_mask_0 = const()[name = tensor<string, []>("op_24432_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24432_cast = slice_by_index(begin = var_24432_begin_0, end = var_24432_end_0, end_mask = var_24432_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24432_cast")];
            tensor<int32, [4]> var_24436_begin_0 = const()[name = tensor<string, []>("op_24436_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_24436_end_0 = const()[name = tensor<string, []>("op_24436_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_24436_end_mask_0 = const()[name = tensor<string, []>("op_24436_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24436_cast = slice_by_index(begin = var_24436_begin_0, end = var_24436_end_0, end_mask = var_24436_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24436_cast")];
            tensor<int32, [4]> var_24440_begin_0 = const()[name = tensor<string, []>("op_24440_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_24440_end_0 = const()[name = tensor<string, []>("op_24440_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_24440_end_mask_0 = const()[name = tensor<string, []>("op_24440_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24440_cast = slice_by_index(begin = var_24440_begin_0, end = var_24440_end_0, end_mask = var_24440_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24440_cast")];
            tensor<int32, [4]> var_24444_begin_0 = const()[name = tensor<string, []>("op_24444_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_24444_end_0 = const()[name = tensor<string, []>("op_24444_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_24444_end_mask_0 = const()[name = tensor<string, []>("op_24444_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24444_cast = slice_by_index(begin = var_24444_begin_0, end = var_24444_end_0, end_mask = var_24444_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24444_cast")];
            tensor<int32, [4]> var_24448_begin_0 = const()[name = tensor<string, []>("op_24448_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_24448_end_0 = const()[name = tensor<string, []>("op_24448_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_24448_end_mask_0 = const()[name = tensor<string, []>("op_24448_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24448_cast = slice_by_index(begin = var_24448_begin_0, end = var_24448_end_0, end_mask = var_24448_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24448_cast")];
            tensor<int32, [4]> var_24452_begin_0 = const()[name = tensor<string, []>("op_24452_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_24452_end_0 = const()[name = tensor<string, []>("op_24452_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_24452_end_mask_0 = const()[name = tensor<string, []>("op_24452_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24452_cast = slice_by_index(begin = var_24452_begin_0, end = var_24452_end_0, end_mask = var_24452_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24452_cast")];
            tensor<int32, [4]> var_24456_begin_0 = const()[name = tensor<string, []>("op_24456_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_24456_end_0 = const()[name = tensor<string, []>("op_24456_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_24456_end_mask_0 = const()[name = tensor<string, []>("op_24456_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24456_cast = slice_by_index(begin = var_24456_begin_0, end = var_24456_end_0, end_mask = var_24456_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24456_cast")];
            tensor<int32, [4]> var_24460_begin_0 = const()[name = tensor<string, []>("op_24460_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_24460_end_0 = const()[name = tensor<string, []>("op_24460_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_24460_end_mask_0 = const()[name = tensor<string, []>("op_24460_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24460_cast = slice_by_index(begin = var_24460_begin_0, end = var_24460_end_0, end_mask = var_24460_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24460_cast")];
            tensor<int32, [4]> var_24464_begin_0 = const()[name = tensor<string, []>("op_24464_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_24464_end_0 = const()[name = tensor<string, []>("op_24464_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_24464_end_mask_0 = const()[name = tensor<string, []>("op_24464_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24464_cast = slice_by_index(begin = var_24464_begin_0, end = var_24464_end_0, end_mask = var_24464_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24464_cast")];
            tensor<int32, [4]> var_24468_begin_0 = const()[name = tensor<string, []>("op_24468_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_24468_end_0 = const()[name = tensor<string, []>("op_24468_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_24468_end_mask_0 = const()[name = tensor<string, []>("op_24468_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24468_cast = slice_by_index(begin = var_24468_begin_0, end = var_24468_end_0, end_mask = var_24468_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24468_cast")];
            tensor<int32, [4]> var_24472_begin_0 = const()[name = tensor<string, []>("op_24472_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_24472_end_0 = const()[name = tensor<string, []>("op_24472_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_24472_end_mask_0 = const()[name = tensor<string, []>("op_24472_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24472_cast = slice_by_index(begin = var_24472_begin_0, end = var_24472_end_0, end_mask = var_24472_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24472_cast")];
            tensor<int32, [4]> var_24476_begin_0 = const()[name = tensor<string, []>("op_24476_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_24476_end_0 = const()[name = tensor<string, []>("op_24476_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_24476_end_mask_0 = const()[name = tensor<string, []>("op_24476_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24476_cast = slice_by_index(begin = var_24476_begin_0, end = var_24476_end_0, end_mask = var_24476_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24476_cast")];
            tensor<int32, [4]> var_24480_begin_0 = const()[name = tensor<string, []>("op_24480_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_24480_end_0 = const()[name = tensor<string, []>("op_24480_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_24480_end_mask_0 = const()[name = tensor<string, []>("op_24480_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24480_cast = slice_by_index(begin = var_24480_begin_0, end = var_24480_end_0, end_mask = var_24480_end_mask_0, x = q_111_cast)[name = tensor<string, []>("op_24480_cast")];
            tensor<int32, [4]> k_223_perm_0 = const()[name = tensor<string, []>("k_223_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_24487_begin_0 = const()[name = tensor<string, []>("op_24487_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_24487_end_0 = const()[name = tensor<string, []>("op_24487_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_24487_end_mask_0 = const()[name = tensor<string, []>("op_24487_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_84 = transpose(perm = k_223_perm_0, x = k_221_cast)[name = tensor<string, []>("transpose_84")];
            tensor<fp16, [2, 77, 1, 64]> var_24487_cast = slice_by_index(begin = var_24487_begin_0, end = var_24487_end_0, end_mask = var_24487_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24487_cast")];
            tensor<int32, [4]> var_24491_begin_0 = const()[name = tensor<string, []>("op_24491_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_24491_end_0 = const()[name = tensor<string, []>("op_24491_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_24491_end_mask_0 = const()[name = tensor<string, []>("op_24491_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_24491_cast = slice_by_index(begin = var_24491_begin_0, end = var_24491_end_0, end_mask = var_24491_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24491_cast")];
            tensor<int32, [4]> var_24495_begin_0 = const()[name = tensor<string, []>("op_24495_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_24495_end_0 = const()[name = tensor<string, []>("op_24495_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_24495_end_mask_0 = const()[name = tensor<string, []>("op_24495_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_24495_cast = slice_by_index(begin = var_24495_begin_0, end = var_24495_end_0, end_mask = var_24495_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24495_cast")];
            tensor<int32, [4]> var_24499_begin_0 = const()[name = tensor<string, []>("op_24499_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_24499_end_0 = const()[name = tensor<string, []>("op_24499_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_24499_end_mask_0 = const()[name = tensor<string, []>("op_24499_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_24499_cast = slice_by_index(begin = var_24499_begin_0, end = var_24499_end_0, end_mask = var_24499_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24499_cast")];
            tensor<int32, [4]> var_24503_begin_0 = const()[name = tensor<string, []>("op_24503_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_24503_end_0 = const()[name = tensor<string, []>("op_24503_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_24503_end_mask_0 = const()[name = tensor<string, []>("op_24503_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_24503_cast = slice_by_index(begin = var_24503_begin_0, end = var_24503_end_0, end_mask = var_24503_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24503_cast")];
            tensor<int32, [4]> var_24507_begin_0 = const()[name = tensor<string, []>("op_24507_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_24507_end_0 = const()[name = tensor<string, []>("op_24507_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_24507_end_mask_0 = const()[name = tensor<string, []>("op_24507_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_24507_cast = slice_by_index(begin = var_24507_begin_0, end = var_24507_end_0, end_mask = var_24507_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24507_cast")];
            tensor<int32, [4]> var_24511_begin_0 = const()[name = tensor<string, []>("op_24511_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_24511_end_0 = const()[name = tensor<string, []>("op_24511_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_24511_end_mask_0 = const()[name = tensor<string, []>("op_24511_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_24511_cast = slice_by_index(begin = var_24511_begin_0, end = var_24511_end_0, end_mask = var_24511_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24511_cast")];
            tensor<int32, [4]> var_24515_begin_0 = const()[name = tensor<string, []>("op_24515_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_24515_end_0 = const()[name = tensor<string, []>("op_24515_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_24515_end_mask_0 = const()[name = tensor<string, []>("op_24515_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_24515_cast = slice_by_index(begin = var_24515_begin_0, end = var_24515_end_0, end_mask = var_24515_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24515_cast")];
            tensor<int32, [4]> var_24519_begin_0 = const()[name = tensor<string, []>("op_24519_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_24519_end_0 = const()[name = tensor<string, []>("op_24519_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_24519_end_mask_0 = const()[name = tensor<string, []>("op_24519_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_24519_cast = slice_by_index(begin = var_24519_begin_0, end = var_24519_end_0, end_mask = var_24519_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24519_cast")];
            tensor<int32, [4]> var_24523_begin_0 = const()[name = tensor<string, []>("op_24523_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_24523_end_0 = const()[name = tensor<string, []>("op_24523_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_24523_end_mask_0 = const()[name = tensor<string, []>("op_24523_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_24523_cast = slice_by_index(begin = var_24523_begin_0, end = var_24523_end_0, end_mask = var_24523_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24523_cast")];
            tensor<int32, [4]> var_24527_begin_0 = const()[name = tensor<string, []>("op_24527_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_24527_end_0 = const()[name = tensor<string, []>("op_24527_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_24527_end_mask_0 = const()[name = tensor<string, []>("op_24527_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_24527_cast = slice_by_index(begin = var_24527_begin_0, end = var_24527_end_0, end_mask = var_24527_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24527_cast")];
            tensor<int32, [4]> var_24531_begin_0 = const()[name = tensor<string, []>("op_24531_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_24531_end_0 = const()[name = tensor<string, []>("op_24531_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_24531_end_mask_0 = const()[name = tensor<string, []>("op_24531_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_24531_cast = slice_by_index(begin = var_24531_begin_0, end = var_24531_end_0, end_mask = var_24531_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24531_cast")];
            tensor<int32, [4]> var_24535_begin_0 = const()[name = tensor<string, []>("op_24535_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_24535_end_0 = const()[name = tensor<string, []>("op_24535_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_24535_end_mask_0 = const()[name = tensor<string, []>("op_24535_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_24535_cast = slice_by_index(begin = var_24535_begin_0, end = var_24535_end_0, end_mask = var_24535_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24535_cast")];
            tensor<int32, [4]> var_24539_begin_0 = const()[name = tensor<string, []>("op_24539_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_24539_end_0 = const()[name = tensor<string, []>("op_24539_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_24539_end_mask_0 = const()[name = tensor<string, []>("op_24539_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_24539_cast = slice_by_index(begin = var_24539_begin_0, end = var_24539_end_0, end_mask = var_24539_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24539_cast")];
            tensor<int32, [4]> var_24543_begin_0 = const()[name = tensor<string, []>("op_24543_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_24543_end_0 = const()[name = tensor<string, []>("op_24543_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_24543_end_mask_0 = const()[name = tensor<string, []>("op_24543_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_24543_cast = slice_by_index(begin = var_24543_begin_0, end = var_24543_end_0, end_mask = var_24543_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24543_cast")];
            tensor<int32, [4]> var_24547_begin_0 = const()[name = tensor<string, []>("op_24547_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_24547_end_0 = const()[name = tensor<string, []>("op_24547_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_24547_end_mask_0 = const()[name = tensor<string, []>("op_24547_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_24547_cast = slice_by_index(begin = var_24547_begin_0, end = var_24547_end_0, end_mask = var_24547_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24547_cast")];
            tensor<int32, [4]> var_24551_begin_0 = const()[name = tensor<string, []>("op_24551_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_24551_end_0 = const()[name = tensor<string, []>("op_24551_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_24551_end_mask_0 = const()[name = tensor<string, []>("op_24551_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_24551_cast = slice_by_index(begin = var_24551_begin_0, end = var_24551_end_0, end_mask = var_24551_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24551_cast")];
            tensor<int32, [4]> var_24555_begin_0 = const()[name = tensor<string, []>("op_24555_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_24555_end_0 = const()[name = tensor<string, []>("op_24555_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_24555_end_mask_0 = const()[name = tensor<string, []>("op_24555_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_24555_cast = slice_by_index(begin = var_24555_begin_0, end = var_24555_end_0, end_mask = var_24555_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24555_cast")];
            tensor<int32, [4]> var_24559_begin_0 = const()[name = tensor<string, []>("op_24559_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_24559_end_0 = const()[name = tensor<string, []>("op_24559_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_24559_end_mask_0 = const()[name = tensor<string, []>("op_24559_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_24559_cast = slice_by_index(begin = var_24559_begin_0, end = var_24559_end_0, end_mask = var_24559_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24559_cast")];
            tensor<int32, [4]> var_24563_begin_0 = const()[name = tensor<string, []>("op_24563_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_24563_end_0 = const()[name = tensor<string, []>("op_24563_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_24563_end_mask_0 = const()[name = tensor<string, []>("op_24563_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_24563_cast = slice_by_index(begin = var_24563_begin_0, end = var_24563_end_0, end_mask = var_24563_end_mask_0, x = transpose_84)[name = tensor<string, []>("op_24563_cast")];
            tensor<int32, [4]> var_24565_begin_0 = const()[name = tensor<string, []>("op_24565_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_24565_end_0 = const()[name = tensor<string, []>("op_24565_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_24565_end_mask_0 = const()[name = tensor<string, []>("op_24565_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24565_cast = slice_by_index(begin = var_24565_begin_0, end = var_24565_end_0, end_mask = var_24565_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24565_cast")];
            tensor<int32, [4]> var_24569_begin_0 = const()[name = tensor<string, []>("op_24569_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_24569_end_0 = const()[name = tensor<string, []>("op_24569_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_24569_end_mask_0 = const()[name = tensor<string, []>("op_24569_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24569_cast = slice_by_index(begin = var_24569_begin_0, end = var_24569_end_0, end_mask = var_24569_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24569_cast")];
            tensor<int32, [4]> var_24573_begin_0 = const()[name = tensor<string, []>("op_24573_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_24573_end_0 = const()[name = tensor<string, []>("op_24573_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_24573_end_mask_0 = const()[name = tensor<string, []>("op_24573_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24573_cast = slice_by_index(begin = var_24573_begin_0, end = var_24573_end_0, end_mask = var_24573_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24573_cast")];
            tensor<int32, [4]> var_24577_begin_0 = const()[name = tensor<string, []>("op_24577_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_24577_end_0 = const()[name = tensor<string, []>("op_24577_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_24577_end_mask_0 = const()[name = tensor<string, []>("op_24577_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24577_cast = slice_by_index(begin = var_24577_begin_0, end = var_24577_end_0, end_mask = var_24577_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24577_cast")];
            tensor<int32, [4]> var_24581_begin_0 = const()[name = tensor<string, []>("op_24581_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_24581_end_0 = const()[name = tensor<string, []>("op_24581_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_24581_end_mask_0 = const()[name = tensor<string, []>("op_24581_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24581_cast = slice_by_index(begin = var_24581_begin_0, end = var_24581_end_0, end_mask = var_24581_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24581_cast")];
            tensor<int32, [4]> var_24585_begin_0 = const()[name = tensor<string, []>("op_24585_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_24585_end_0 = const()[name = tensor<string, []>("op_24585_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_24585_end_mask_0 = const()[name = tensor<string, []>("op_24585_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24585_cast = slice_by_index(begin = var_24585_begin_0, end = var_24585_end_0, end_mask = var_24585_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24585_cast")];
            tensor<int32, [4]> var_24589_begin_0 = const()[name = tensor<string, []>("op_24589_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_24589_end_0 = const()[name = tensor<string, []>("op_24589_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_24589_end_mask_0 = const()[name = tensor<string, []>("op_24589_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24589_cast = slice_by_index(begin = var_24589_begin_0, end = var_24589_end_0, end_mask = var_24589_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24589_cast")];
            tensor<int32, [4]> var_24593_begin_0 = const()[name = tensor<string, []>("op_24593_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_24593_end_0 = const()[name = tensor<string, []>("op_24593_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_24593_end_mask_0 = const()[name = tensor<string, []>("op_24593_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24593_cast = slice_by_index(begin = var_24593_begin_0, end = var_24593_end_0, end_mask = var_24593_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24593_cast")];
            tensor<int32, [4]> var_24597_begin_0 = const()[name = tensor<string, []>("op_24597_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_24597_end_0 = const()[name = tensor<string, []>("op_24597_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_24597_end_mask_0 = const()[name = tensor<string, []>("op_24597_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24597_cast = slice_by_index(begin = var_24597_begin_0, end = var_24597_end_0, end_mask = var_24597_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24597_cast")];
            tensor<int32, [4]> var_24601_begin_0 = const()[name = tensor<string, []>("op_24601_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_24601_end_0 = const()[name = tensor<string, []>("op_24601_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_24601_end_mask_0 = const()[name = tensor<string, []>("op_24601_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24601_cast = slice_by_index(begin = var_24601_begin_0, end = var_24601_end_0, end_mask = var_24601_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24601_cast")];
            tensor<int32, [4]> var_24605_begin_0 = const()[name = tensor<string, []>("op_24605_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_24605_end_0 = const()[name = tensor<string, []>("op_24605_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_24605_end_mask_0 = const()[name = tensor<string, []>("op_24605_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24605_cast = slice_by_index(begin = var_24605_begin_0, end = var_24605_end_0, end_mask = var_24605_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24605_cast")];
            tensor<int32, [4]> var_24609_begin_0 = const()[name = tensor<string, []>("op_24609_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_24609_end_0 = const()[name = tensor<string, []>("op_24609_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_24609_end_mask_0 = const()[name = tensor<string, []>("op_24609_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24609_cast = slice_by_index(begin = var_24609_begin_0, end = var_24609_end_0, end_mask = var_24609_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24609_cast")];
            tensor<int32, [4]> var_24613_begin_0 = const()[name = tensor<string, []>("op_24613_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_24613_end_0 = const()[name = tensor<string, []>("op_24613_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_24613_end_mask_0 = const()[name = tensor<string, []>("op_24613_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24613_cast = slice_by_index(begin = var_24613_begin_0, end = var_24613_end_0, end_mask = var_24613_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24613_cast")];
            tensor<int32, [4]> var_24617_begin_0 = const()[name = tensor<string, []>("op_24617_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_24617_end_0 = const()[name = tensor<string, []>("op_24617_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_24617_end_mask_0 = const()[name = tensor<string, []>("op_24617_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24617_cast = slice_by_index(begin = var_24617_begin_0, end = var_24617_end_0, end_mask = var_24617_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24617_cast")];
            tensor<int32, [4]> var_24621_begin_0 = const()[name = tensor<string, []>("op_24621_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_24621_end_0 = const()[name = tensor<string, []>("op_24621_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_24621_end_mask_0 = const()[name = tensor<string, []>("op_24621_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24621_cast = slice_by_index(begin = var_24621_begin_0, end = var_24621_end_0, end_mask = var_24621_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24621_cast")];
            tensor<int32, [4]> var_24625_begin_0 = const()[name = tensor<string, []>("op_24625_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_24625_end_0 = const()[name = tensor<string, []>("op_24625_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_24625_end_mask_0 = const()[name = tensor<string, []>("op_24625_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24625_cast = slice_by_index(begin = var_24625_begin_0, end = var_24625_end_0, end_mask = var_24625_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24625_cast")];
            tensor<int32, [4]> var_24629_begin_0 = const()[name = tensor<string, []>("op_24629_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_24629_end_0 = const()[name = tensor<string, []>("op_24629_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_24629_end_mask_0 = const()[name = tensor<string, []>("op_24629_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24629_cast = slice_by_index(begin = var_24629_begin_0, end = var_24629_end_0, end_mask = var_24629_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24629_cast")];
            tensor<int32, [4]> var_24633_begin_0 = const()[name = tensor<string, []>("op_24633_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_24633_end_0 = const()[name = tensor<string, []>("op_24633_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_24633_end_mask_0 = const()[name = tensor<string, []>("op_24633_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24633_cast = slice_by_index(begin = var_24633_begin_0, end = var_24633_end_0, end_mask = var_24633_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24633_cast")];
            tensor<int32, [4]> var_24637_begin_0 = const()[name = tensor<string, []>("op_24637_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_24637_end_0 = const()[name = tensor<string, []>("op_24637_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_24637_end_mask_0 = const()[name = tensor<string, []>("op_24637_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24637_cast = slice_by_index(begin = var_24637_begin_0, end = var_24637_end_0, end_mask = var_24637_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24637_cast")];
            tensor<int32, [4]> var_24641_begin_0 = const()[name = tensor<string, []>("op_24641_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_24641_end_0 = const()[name = tensor<string, []>("op_24641_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_24641_end_mask_0 = const()[name = tensor<string, []>("op_24641_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_24641_cast = slice_by_index(begin = var_24641_begin_0, end = var_24641_end_0, end_mask = var_24641_end_mask_0, x = v_111_cast)[name = tensor<string, []>("op_24641_cast")];
            tensor<string, []> var_24645_equation_0 = const()[name = tensor<string, []>("op_24645_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24645_cast = einsum(equation = var_24645_equation_0, values = (var_24487_cast, var_24404_cast))[name = tensor<string, []>("op_24645_cast")];
            tensor<fp16, []> var_24646_to_fp16 = const()[name = tensor<string, []>("op_24646_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2041_cast = mul(x = var_24645_cast, y = var_24646_to_fp16)[name = tensor<string, []>("aw_2041_cast")];
            tensor<string, []> var_24649_equation_0 = const()[name = tensor<string, []>("op_24649_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24649_cast = einsum(equation = var_24649_equation_0, values = (var_24491_cast, var_24408_cast))[name = tensor<string, []>("op_24649_cast")];
            tensor<fp16, []> var_24650_to_fp16 = const()[name = tensor<string, []>("op_24650_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2043_cast = mul(x = var_24649_cast, y = var_24650_to_fp16)[name = tensor<string, []>("aw_2043_cast")];
            tensor<string, []> var_24653_equation_0 = const()[name = tensor<string, []>("op_24653_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24653_cast = einsum(equation = var_24653_equation_0, values = (var_24495_cast, var_24412_cast))[name = tensor<string, []>("op_24653_cast")];
            tensor<fp16, []> var_24654_to_fp16 = const()[name = tensor<string, []>("op_24654_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2045_cast = mul(x = var_24653_cast, y = var_24654_to_fp16)[name = tensor<string, []>("aw_2045_cast")];
            tensor<string, []> var_24657_equation_0 = const()[name = tensor<string, []>("op_24657_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24657_cast = einsum(equation = var_24657_equation_0, values = (var_24499_cast, var_24416_cast))[name = tensor<string, []>("op_24657_cast")];
            tensor<fp16, []> var_24658_to_fp16 = const()[name = tensor<string, []>("op_24658_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2047_cast = mul(x = var_24657_cast, y = var_24658_to_fp16)[name = tensor<string, []>("aw_2047_cast")];
            tensor<string, []> var_24661_equation_0 = const()[name = tensor<string, []>("op_24661_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24661_cast = einsum(equation = var_24661_equation_0, values = (var_24503_cast, var_24420_cast))[name = tensor<string, []>("op_24661_cast")];
            tensor<fp16, []> var_24662_to_fp16 = const()[name = tensor<string, []>("op_24662_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2049_cast = mul(x = var_24661_cast, y = var_24662_to_fp16)[name = tensor<string, []>("aw_2049_cast")];
            tensor<string, []> var_24665_equation_0 = const()[name = tensor<string, []>("op_24665_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24665_cast = einsum(equation = var_24665_equation_0, values = (var_24507_cast, var_24424_cast))[name = tensor<string, []>("op_24665_cast")];
            tensor<fp16, []> var_24666_to_fp16 = const()[name = tensor<string, []>("op_24666_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2051_cast = mul(x = var_24665_cast, y = var_24666_to_fp16)[name = tensor<string, []>("aw_2051_cast")];
            tensor<string, []> var_24669_equation_0 = const()[name = tensor<string, []>("op_24669_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24669_cast = einsum(equation = var_24669_equation_0, values = (var_24511_cast, var_24428_cast))[name = tensor<string, []>("op_24669_cast")];
            tensor<fp16, []> var_24670_to_fp16 = const()[name = tensor<string, []>("op_24670_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2053_cast = mul(x = var_24669_cast, y = var_24670_to_fp16)[name = tensor<string, []>("aw_2053_cast")];
            tensor<string, []> var_24673_equation_0 = const()[name = tensor<string, []>("op_24673_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24673_cast = einsum(equation = var_24673_equation_0, values = (var_24515_cast, var_24432_cast))[name = tensor<string, []>("op_24673_cast")];
            tensor<fp16, []> var_24674_to_fp16 = const()[name = tensor<string, []>("op_24674_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2055_cast = mul(x = var_24673_cast, y = var_24674_to_fp16)[name = tensor<string, []>("aw_2055_cast")];
            tensor<string, []> var_24677_equation_0 = const()[name = tensor<string, []>("op_24677_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24677_cast = einsum(equation = var_24677_equation_0, values = (var_24519_cast, var_24436_cast))[name = tensor<string, []>("op_24677_cast")];
            tensor<fp16, []> var_24678_to_fp16 = const()[name = tensor<string, []>("op_24678_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2057_cast = mul(x = var_24677_cast, y = var_24678_to_fp16)[name = tensor<string, []>("aw_2057_cast")];
            tensor<string, []> var_24681_equation_0 = const()[name = tensor<string, []>("op_24681_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24681_cast = einsum(equation = var_24681_equation_0, values = (var_24523_cast, var_24440_cast))[name = tensor<string, []>("op_24681_cast")];
            tensor<fp16, []> var_24682_to_fp16 = const()[name = tensor<string, []>("op_24682_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2059_cast = mul(x = var_24681_cast, y = var_24682_to_fp16)[name = tensor<string, []>("aw_2059_cast")];
            tensor<string, []> var_24685_equation_0 = const()[name = tensor<string, []>("op_24685_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24685_cast = einsum(equation = var_24685_equation_0, values = (var_24527_cast, var_24444_cast))[name = tensor<string, []>("op_24685_cast")];
            tensor<fp16, []> var_24686_to_fp16 = const()[name = tensor<string, []>("op_24686_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2061_cast = mul(x = var_24685_cast, y = var_24686_to_fp16)[name = tensor<string, []>("aw_2061_cast")];
            tensor<string, []> var_24689_equation_0 = const()[name = tensor<string, []>("op_24689_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24689_cast = einsum(equation = var_24689_equation_0, values = (var_24531_cast, var_24448_cast))[name = tensor<string, []>("op_24689_cast")];
            tensor<fp16, []> var_24690_to_fp16 = const()[name = tensor<string, []>("op_24690_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2063_cast = mul(x = var_24689_cast, y = var_24690_to_fp16)[name = tensor<string, []>("aw_2063_cast")];
            tensor<string, []> var_24693_equation_0 = const()[name = tensor<string, []>("op_24693_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24693_cast = einsum(equation = var_24693_equation_0, values = (var_24535_cast, var_24452_cast))[name = tensor<string, []>("op_24693_cast")];
            tensor<fp16, []> var_24694_to_fp16 = const()[name = tensor<string, []>("op_24694_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2065_cast = mul(x = var_24693_cast, y = var_24694_to_fp16)[name = tensor<string, []>("aw_2065_cast")];
            tensor<string, []> var_24697_equation_0 = const()[name = tensor<string, []>("op_24697_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24697_cast = einsum(equation = var_24697_equation_0, values = (var_24539_cast, var_24456_cast))[name = tensor<string, []>("op_24697_cast")];
            tensor<fp16, []> var_24698_to_fp16 = const()[name = tensor<string, []>("op_24698_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2067_cast = mul(x = var_24697_cast, y = var_24698_to_fp16)[name = tensor<string, []>("aw_2067_cast")];
            tensor<string, []> var_24701_equation_0 = const()[name = tensor<string, []>("op_24701_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24701_cast = einsum(equation = var_24701_equation_0, values = (var_24543_cast, var_24460_cast))[name = tensor<string, []>("op_24701_cast")];
            tensor<fp16, []> var_24702_to_fp16 = const()[name = tensor<string, []>("op_24702_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2069_cast = mul(x = var_24701_cast, y = var_24702_to_fp16)[name = tensor<string, []>("aw_2069_cast")];
            tensor<string, []> var_24705_equation_0 = const()[name = tensor<string, []>("op_24705_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24705_cast = einsum(equation = var_24705_equation_0, values = (var_24547_cast, var_24464_cast))[name = tensor<string, []>("op_24705_cast")];
            tensor<fp16, []> var_24706_to_fp16 = const()[name = tensor<string, []>("op_24706_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2071_cast = mul(x = var_24705_cast, y = var_24706_to_fp16)[name = tensor<string, []>("aw_2071_cast")];
            tensor<string, []> var_24709_equation_0 = const()[name = tensor<string, []>("op_24709_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24709_cast = einsum(equation = var_24709_equation_0, values = (var_24551_cast, var_24468_cast))[name = tensor<string, []>("op_24709_cast")];
            tensor<fp16, []> var_24710_to_fp16 = const()[name = tensor<string, []>("op_24710_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2073_cast = mul(x = var_24709_cast, y = var_24710_to_fp16)[name = tensor<string, []>("aw_2073_cast")];
            tensor<string, []> var_24713_equation_0 = const()[name = tensor<string, []>("op_24713_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24713_cast = einsum(equation = var_24713_equation_0, values = (var_24555_cast, var_24472_cast))[name = tensor<string, []>("op_24713_cast")];
            tensor<fp16, []> var_24714_to_fp16 = const()[name = tensor<string, []>("op_24714_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2075_cast = mul(x = var_24713_cast, y = var_24714_to_fp16)[name = tensor<string, []>("aw_2075_cast")];
            tensor<string, []> var_24717_equation_0 = const()[name = tensor<string, []>("op_24717_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24717_cast = einsum(equation = var_24717_equation_0, values = (var_24559_cast, var_24476_cast))[name = tensor<string, []>("op_24717_cast")];
            tensor<fp16, []> var_24718_to_fp16 = const()[name = tensor<string, []>("op_24718_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2077_cast = mul(x = var_24717_cast, y = var_24718_to_fp16)[name = tensor<string, []>("aw_2077_cast")];
            tensor<string, []> var_24721_equation_0 = const()[name = tensor<string, []>("op_24721_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_24721_cast = einsum(equation = var_24721_equation_0, values = (var_24563_cast, var_24480_cast))[name = tensor<string, []>("op_24721_cast")];
            tensor<fp16, []> var_24722_to_fp16 = const()[name = tensor<string, []>("op_24722_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2079_cast = mul(x = var_24721_cast, y = var_24722_to_fp16)[name = tensor<string, []>("aw_2079_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24724_cast = softmax(axis = var_21105, x = aw_2041_cast)[name = tensor<string, []>("op_24724_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24725_cast = softmax(axis = var_21105, x = aw_2043_cast)[name = tensor<string, []>("op_24725_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24726_cast = softmax(axis = var_21105, x = aw_2045_cast)[name = tensor<string, []>("op_24726_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24727_cast = softmax(axis = var_21105, x = aw_2047_cast)[name = tensor<string, []>("op_24727_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24728_cast = softmax(axis = var_21105, x = aw_2049_cast)[name = tensor<string, []>("op_24728_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24729_cast = softmax(axis = var_21105, x = aw_2051_cast)[name = tensor<string, []>("op_24729_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24730_cast = softmax(axis = var_21105, x = aw_2053_cast)[name = tensor<string, []>("op_24730_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24731_cast = softmax(axis = var_21105, x = aw_2055_cast)[name = tensor<string, []>("op_24731_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24732_cast = softmax(axis = var_21105, x = aw_2057_cast)[name = tensor<string, []>("op_24732_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24733_cast = softmax(axis = var_21105, x = aw_2059_cast)[name = tensor<string, []>("op_24733_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24734_cast = softmax(axis = var_21105, x = aw_2061_cast)[name = tensor<string, []>("op_24734_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24735_cast = softmax(axis = var_21105, x = aw_2063_cast)[name = tensor<string, []>("op_24735_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24736_cast = softmax(axis = var_21105, x = aw_2065_cast)[name = tensor<string, []>("op_24736_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24737_cast = softmax(axis = var_21105, x = aw_2067_cast)[name = tensor<string, []>("op_24737_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24738_cast = softmax(axis = var_21105, x = aw_2069_cast)[name = tensor<string, []>("op_24738_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24739_cast = softmax(axis = var_21105, x = aw_2071_cast)[name = tensor<string, []>("op_24739_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24740_cast = softmax(axis = var_21105, x = aw_2073_cast)[name = tensor<string, []>("op_24740_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24741_cast = softmax(axis = var_21105, x = aw_2075_cast)[name = tensor<string, []>("op_24741_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24742_cast = softmax(axis = var_21105, x = aw_2077_cast)[name = tensor<string, []>("op_24742_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_24743_cast = softmax(axis = var_21105, x = aw_2079_cast)[name = tensor<string, []>("op_24743_cast")];
            tensor<string, []> var_24745_equation_0 = const()[name = tensor<string, []>("op_24745_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24745_cast = einsum(equation = var_24745_equation_0, values = (var_24565_cast, var_24724_cast))[name = tensor<string, []>("op_24745_cast")];
            tensor<string, []> var_24747_equation_0 = const()[name = tensor<string, []>("op_24747_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24747_cast = einsum(equation = var_24747_equation_0, values = (var_24569_cast, var_24725_cast))[name = tensor<string, []>("op_24747_cast")];
            tensor<string, []> var_24749_equation_0 = const()[name = tensor<string, []>("op_24749_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24749_cast = einsum(equation = var_24749_equation_0, values = (var_24573_cast, var_24726_cast))[name = tensor<string, []>("op_24749_cast")];
            tensor<string, []> var_24751_equation_0 = const()[name = tensor<string, []>("op_24751_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24751_cast = einsum(equation = var_24751_equation_0, values = (var_24577_cast, var_24727_cast))[name = tensor<string, []>("op_24751_cast")];
            tensor<string, []> var_24753_equation_0 = const()[name = tensor<string, []>("op_24753_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24753_cast = einsum(equation = var_24753_equation_0, values = (var_24581_cast, var_24728_cast))[name = tensor<string, []>("op_24753_cast")];
            tensor<string, []> var_24755_equation_0 = const()[name = tensor<string, []>("op_24755_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24755_cast = einsum(equation = var_24755_equation_0, values = (var_24585_cast, var_24729_cast))[name = tensor<string, []>("op_24755_cast")];
            tensor<string, []> var_24757_equation_0 = const()[name = tensor<string, []>("op_24757_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24757_cast = einsum(equation = var_24757_equation_0, values = (var_24589_cast, var_24730_cast))[name = tensor<string, []>("op_24757_cast")];
            tensor<string, []> var_24759_equation_0 = const()[name = tensor<string, []>("op_24759_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24759_cast = einsum(equation = var_24759_equation_0, values = (var_24593_cast, var_24731_cast))[name = tensor<string, []>("op_24759_cast")];
            tensor<string, []> var_24761_equation_0 = const()[name = tensor<string, []>("op_24761_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24761_cast = einsum(equation = var_24761_equation_0, values = (var_24597_cast, var_24732_cast))[name = tensor<string, []>("op_24761_cast")];
            tensor<string, []> var_24763_equation_0 = const()[name = tensor<string, []>("op_24763_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24763_cast = einsum(equation = var_24763_equation_0, values = (var_24601_cast, var_24733_cast))[name = tensor<string, []>("op_24763_cast")];
            tensor<string, []> var_24765_equation_0 = const()[name = tensor<string, []>("op_24765_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24765_cast = einsum(equation = var_24765_equation_0, values = (var_24605_cast, var_24734_cast))[name = tensor<string, []>("op_24765_cast")];
            tensor<string, []> var_24767_equation_0 = const()[name = tensor<string, []>("op_24767_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24767_cast = einsum(equation = var_24767_equation_0, values = (var_24609_cast, var_24735_cast))[name = tensor<string, []>("op_24767_cast")];
            tensor<string, []> var_24769_equation_0 = const()[name = tensor<string, []>("op_24769_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24769_cast = einsum(equation = var_24769_equation_0, values = (var_24613_cast, var_24736_cast))[name = tensor<string, []>("op_24769_cast")];
            tensor<string, []> var_24771_equation_0 = const()[name = tensor<string, []>("op_24771_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24771_cast = einsum(equation = var_24771_equation_0, values = (var_24617_cast, var_24737_cast))[name = tensor<string, []>("op_24771_cast")];
            tensor<string, []> var_24773_equation_0 = const()[name = tensor<string, []>("op_24773_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24773_cast = einsum(equation = var_24773_equation_0, values = (var_24621_cast, var_24738_cast))[name = tensor<string, []>("op_24773_cast")];
            tensor<string, []> var_24775_equation_0 = const()[name = tensor<string, []>("op_24775_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24775_cast = einsum(equation = var_24775_equation_0, values = (var_24625_cast, var_24739_cast))[name = tensor<string, []>("op_24775_cast")];
            tensor<string, []> var_24777_equation_0 = const()[name = tensor<string, []>("op_24777_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24777_cast = einsum(equation = var_24777_equation_0, values = (var_24629_cast, var_24740_cast))[name = tensor<string, []>("op_24777_cast")];
            tensor<string, []> var_24779_equation_0 = const()[name = tensor<string, []>("op_24779_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24779_cast = einsum(equation = var_24779_equation_0, values = (var_24633_cast, var_24741_cast))[name = tensor<string, []>("op_24779_cast")];
            tensor<string, []> var_24781_equation_0 = const()[name = tensor<string, []>("op_24781_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24781_cast = einsum(equation = var_24781_equation_0, values = (var_24637_cast, var_24742_cast))[name = tensor<string, []>("op_24781_cast")];
            tensor<string, []> var_24783_equation_0 = const()[name = tensor<string, []>("op_24783_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_24783_cast = einsum(equation = var_24783_equation_0, values = (var_24641_cast, var_24743_cast))[name = tensor<string, []>("op_24783_cast")];
            tensor<bool, []> input_355_interleave_0 = const()[name = tensor<string, []>("input_355_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_355_cast = concat(axis = var_21105, interleave = input_355_interleave_0, values = (var_24745_cast, var_24747_cast, var_24749_cast, var_24751_cast, var_24753_cast, var_24755_cast, var_24757_cast, var_24759_cast, var_24761_cast, var_24763_cast, var_24765_cast, var_24767_cast, var_24769_cast, var_24771_cast, var_24773_cast, var_24775_cast, var_24777_cast, var_24779_cast, var_24781_cast, var_24783_cast))[name = tensor<string, []>("input_355_cast")];
            tensor<int32, [2]> var_24789 = const()[name = tensor<string, []>("op_24789"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_24791 = const()[name = tensor<string, []>("op_24791"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_24793_pad_type_0 = const()[name = tensor<string, []>("op_24793_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_24793_pad_0 = const()[name = tensor<string, []>("op_24793_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1961531264)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1964808128)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_24793_cast = conv(bias = mid_block_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16, dilations = var_24791, groups = var_21105, pad = var_24793_pad_0, pad_type = var_24793_pad_type_0, strides = var_24789, weight = mid_block_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16, x = input_355_cast)[name = tensor<string, []>("op_24793_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_167_cast = add(x = var_24793_cast, y = inputs_165_cast)[name = tensor<string, []>("inputs_167_cast")];
            tensor<int32, [1]> var_24797 = const()[name = tensor<string, []>("op_24797"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_167_cast = reduce_mean(axes = var_24797, keep_dims = var_21100, x = inputs_167_cast)[name = tensor<string, []>("channels_mean_167_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_167_cast = sub(x = inputs_167_cast, y = channels_mean_167_cast)[name = tensor<string, []>("zero_mean_167_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_167_cast = mul(x = zero_mean_167_cast, y = zero_mean_167_cast)[name = tensor<string, []>("zero_mean_sq_167_cast")];
            tensor<int32, [1]> var_24801 = const()[name = tensor<string, []>("op_24801"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_24802_cast = reduce_mean(axes = var_24801, keep_dims = var_21100, x = zero_mean_sq_167_cast)[name = tensor<string, []>("op_24802_cast")];
            tensor<fp16, []> var_24803_to_fp16 = const()[name = tensor<string, []>("op_24803_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_24804_cast = add(x = var_24802_cast, y = var_24803_to_fp16)[name = tensor<string, []>("op_24804_cast")];
            tensor<fp16, []> denom_167_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_167_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_167_cast = rsqrt(epsilon = denom_167_epsilon_0_to_fp16, x = var_24804_cast)[name = tensor<string, []>("denom_167_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_167_cast = mul(x = zero_mean_167_cast, y = denom_167_cast)[name = tensor<string, []>("out_167_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_24808_to_fp16 = const()[name = tensor<string, []>("op_24808_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1964810752)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_24809_cast = add(x = out_167_cast, y = var_24808_to_fp16)[name = tensor<string, []>("op_24809_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_24811_to_fp16 = const()[name = tensor<string, []>("op_24811_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1964813376)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_357_cast = mul(x = var_24809_cast, y = var_24811_to_fp16)[name = tensor<string, []>("input_357_cast")];
            tensor<int32, [2]> var_24819 = const()[name = tensor<string, []>("op_24819"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_24821 = const()[name = tensor<string, []>("op_24821"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_24823_pad_type_0 = const()[name = tensor<string, []>("op_24823_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_24823_pad_0 = const()[name = tensor<string, []>("op_24823_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1964816000)))];
            tensor<fp16, [10240]> mid_block_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1991030464)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_24823_cast = conv(bias = mid_block_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16, dilations = var_24821, groups = var_21105, pad = var_24823_pad_0, pad_type = var_24823_pad_type_0, strides = var_24819, weight = mid_block_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16, x = input_357_cast)[name = tensor<string, []>("op_24823_cast")];
            tensor<int32, [2]> var_24824_split_sizes_0 = const()[name = tensor<string, []>("op_24824_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_24824_axis_0 = const()[name = tensor<string, []>("op_24824_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_24824_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_24824_cast_1 = split(axis = var_24824_axis_0, split_sizes = var_24824_split_sizes_0, x = var_24823_cast)[name = tensor<string, []>("op_24824_cast")];
            tensor<string, []> var_24826_mode_0 = const()[name = tensor<string, []>("op_24826_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_24826_cast = gelu(mode = var_24826_mode_0, x = var_24824_cast_1)[name = tensor<string, []>("op_24826_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_359_cast = mul(x = var_24824_cast_0, y = var_24826_cast)[name = tensor<string, []>("input_359_cast")];
            tensor<int32, [2]> var_24830 = const()[name = tensor<string, []>("op_24830"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_24832 = const()[name = tensor<string, []>("op_24832"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_24834_pad_type_0 = const()[name = tensor<string, []>("op_24834_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_24834_pad_0 = const()[name = tensor<string, []>("op_24834_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> mid_block_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1991051008)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2004158272)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_24834_cast = conv(bias = mid_block_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16, dilations = var_24832, groups = var_21105, pad = var_24834_pad_0, pad_type = var_24834_pad_type_0, strides = var_24830, weight = mid_block_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16, x = input_359_cast)[name = tensor<string, []>("op_24834_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_169_cast = add(x = var_24834_cast, y = inputs_167_cast)[name = tensor<string, []>("inputs_169_cast")];
            tensor<int32, [1]> var_24844 = const()[name = tensor<string, []>("op_24844"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_169_cast = reduce_mean(axes = var_24844, keep_dims = var_21100, x = inputs_169_cast)[name = tensor<string, []>("channels_mean_169_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_169_cast = sub(x = inputs_169_cast, y = channels_mean_169_cast)[name = tensor<string, []>("zero_mean_169_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_169_cast = mul(x = zero_mean_169_cast, y = zero_mean_169_cast)[name = tensor<string, []>("zero_mean_sq_169_cast")];
            tensor<int32, [1]> var_24848 = const()[name = tensor<string, []>("op_24848"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_24849_cast = reduce_mean(axes = var_24848, keep_dims = var_21100, x = zero_mean_sq_169_cast)[name = tensor<string, []>("op_24849_cast")];
            tensor<fp16, []> var_24850_to_fp16 = const()[name = tensor<string, []>("op_24850_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_24851_cast = add(x = var_24849_cast, y = var_24850_to_fp16)[name = tensor<string, []>("op_24851_cast")];
            tensor<fp16, []> denom_169_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_169_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_169_cast = rsqrt(epsilon = denom_169_epsilon_0_to_fp16, x = var_24851_cast)[name = tensor<string, []>("denom_169_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_169_cast = mul(x = zero_mean_169_cast, y = denom_169_cast)[name = tensor<string, []>("out_169_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_24855_to_fp16 = const()[name = tensor<string, []>("op_24855_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2004160896)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_24856_cast = add(x = out_169_cast, y = var_24855_to_fp16)[name = tensor<string, []>("op_24856_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_24858_to_fp16 = const()[name = tensor<string, []>("op_24858_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2004163520)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_233_cast = mul(x = var_24856_cast, y = var_24858_to_fp16)[name = tensor<string, []>("hidden_states_233_cast")];
            tensor<int32, [2]> var_24865 = const()[name = tensor<string, []>("op_24865"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_24867 = const()[name = tensor<string, []>("op_24867"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_113_pad_type_0 = const()[name = tensor<string, []>("q_113_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_113_pad_0 = const()[name = tensor<string, []>("q_113_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2004166144)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_113_cast = conv(dilations = var_24867, groups = var_21105, pad = q_113_pad_0, pad_type = q_113_pad_type_0, strides = var_24865, weight = mid_block_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16, x = hidden_states_233_cast)[name = tensor<string, []>("q_113_cast")];
            tensor<int32, [2]> var_24871 = const()[name = tensor<string, []>("op_24871"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_24873 = const()[name = tensor<string, []>("op_24873"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_225_pad_type_0 = const()[name = tensor<string, []>("k_225_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_225_pad_0 = const()[name = tensor<string, []>("k_225_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2007443008)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_225_cast = conv(dilations = var_24873, groups = var_21105, pad = k_225_pad_0, pad_type = k_225_pad_type_0, strides = var_24871, weight = mid_block_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16, x = hidden_states_233_cast)[name = tensor<string, []>("k_225_cast")];
            tensor<int32, [2]> var_24877 = const()[name = tensor<string, []>("op_24877"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_24879 = const()[name = tensor<string, []>("op_24879"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_113_pad_type_0 = const()[name = tensor<string, []>("v_113_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_113_pad_0 = const()[name = tensor<string, []>("v_113_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2010719872)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_113_cast = conv(dilations = var_24879, groups = var_21105, pad = v_113_pad_0, pad_type = v_113_pad_type_0, strides = var_24877, weight = mid_block_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16, x = hidden_states_233_cast)[name = tensor<string, []>("v_113_cast")];
            tensor<int32, [4]> var_24883_begin_0 = const()[name = tensor<string, []>("op_24883_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_24883_end_0 = const()[name = tensor<string, []>("op_24883_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_24883_end_mask_0 = const()[name = tensor<string, []>("op_24883_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24883_cast = slice_by_index(begin = var_24883_begin_0, end = var_24883_end_0, end_mask = var_24883_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24883_cast")];
            tensor<int32, [4]> var_24887_begin_0 = const()[name = tensor<string, []>("op_24887_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_24887_end_0 = const()[name = tensor<string, []>("op_24887_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_24887_end_mask_0 = const()[name = tensor<string, []>("op_24887_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24887_cast = slice_by_index(begin = var_24887_begin_0, end = var_24887_end_0, end_mask = var_24887_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24887_cast")];
            tensor<int32, [4]> var_24891_begin_0 = const()[name = tensor<string, []>("op_24891_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_24891_end_0 = const()[name = tensor<string, []>("op_24891_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_24891_end_mask_0 = const()[name = tensor<string, []>("op_24891_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24891_cast = slice_by_index(begin = var_24891_begin_0, end = var_24891_end_0, end_mask = var_24891_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24891_cast")];
            tensor<int32, [4]> var_24895_begin_0 = const()[name = tensor<string, []>("op_24895_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_24895_end_0 = const()[name = tensor<string, []>("op_24895_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_24895_end_mask_0 = const()[name = tensor<string, []>("op_24895_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24895_cast = slice_by_index(begin = var_24895_begin_0, end = var_24895_end_0, end_mask = var_24895_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24895_cast")];
            tensor<int32, [4]> var_24899_begin_0 = const()[name = tensor<string, []>("op_24899_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_24899_end_0 = const()[name = tensor<string, []>("op_24899_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_24899_end_mask_0 = const()[name = tensor<string, []>("op_24899_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24899_cast = slice_by_index(begin = var_24899_begin_0, end = var_24899_end_0, end_mask = var_24899_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24899_cast")];
            tensor<int32, [4]> var_24903_begin_0 = const()[name = tensor<string, []>("op_24903_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_24903_end_0 = const()[name = tensor<string, []>("op_24903_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_24903_end_mask_0 = const()[name = tensor<string, []>("op_24903_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24903_cast = slice_by_index(begin = var_24903_begin_0, end = var_24903_end_0, end_mask = var_24903_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24903_cast")];
            tensor<int32, [4]> var_24907_begin_0 = const()[name = tensor<string, []>("op_24907_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_24907_end_0 = const()[name = tensor<string, []>("op_24907_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_24907_end_mask_0 = const()[name = tensor<string, []>("op_24907_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24907_cast = slice_by_index(begin = var_24907_begin_0, end = var_24907_end_0, end_mask = var_24907_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24907_cast")];
            tensor<int32, [4]> var_24911_begin_0 = const()[name = tensor<string, []>("op_24911_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_24911_end_0 = const()[name = tensor<string, []>("op_24911_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_24911_end_mask_0 = const()[name = tensor<string, []>("op_24911_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24911_cast = slice_by_index(begin = var_24911_begin_0, end = var_24911_end_0, end_mask = var_24911_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24911_cast")];
            tensor<int32, [4]> var_24915_begin_0 = const()[name = tensor<string, []>("op_24915_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_24915_end_0 = const()[name = tensor<string, []>("op_24915_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_24915_end_mask_0 = const()[name = tensor<string, []>("op_24915_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24915_cast = slice_by_index(begin = var_24915_begin_0, end = var_24915_end_0, end_mask = var_24915_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24915_cast")];
            tensor<int32, [4]> var_24919_begin_0 = const()[name = tensor<string, []>("op_24919_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_24919_end_0 = const()[name = tensor<string, []>("op_24919_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_24919_end_mask_0 = const()[name = tensor<string, []>("op_24919_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24919_cast = slice_by_index(begin = var_24919_begin_0, end = var_24919_end_0, end_mask = var_24919_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24919_cast")];
            tensor<int32, [4]> var_24923_begin_0 = const()[name = tensor<string, []>("op_24923_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_24923_end_0 = const()[name = tensor<string, []>("op_24923_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_24923_end_mask_0 = const()[name = tensor<string, []>("op_24923_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24923_cast = slice_by_index(begin = var_24923_begin_0, end = var_24923_end_0, end_mask = var_24923_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24923_cast")];
            tensor<int32, [4]> var_24927_begin_0 = const()[name = tensor<string, []>("op_24927_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_24927_end_0 = const()[name = tensor<string, []>("op_24927_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_24927_end_mask_0 = const()[name = tensor<string, []>("op_24927_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24927_cast = slice_by_index(begin = var_24927_begin_0, end = var_24927_end_0, end_mask = var_24927_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24927_cast")];
            tensor<int32, [4]> var_24931_begin_0 = const()[name = tensor<string, []>("op_24931_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_24931_end_0 = const()[name = tensor<string, []>("op_24931_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_24931_end_mask_0 = const()[name = tensor<string, []>("op_24931_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24931_cast = slice_by_index(begin = var_24931_begin_0, end = var_24931_end_0, end_mask = var_24931_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24931_cast")];
            tensor<int32, [4]> var_24935_begin_0 = const()[name = tensor<string, []>("op_24935_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_24935_end_0 = const()[name = tensor<string, []>("op_24935_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_24935_end_mask_0 = const()[name = tensor<string, []>("op_24935_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24935_cast = slice_by_index(begin = var_24935_begin_0, end = var_24935_end_0, end_mask = var_24935_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24935_cast")];
            tensor<int32, [4]> var_24939_begin_0 = const()[name = tensor<string, []>("op_24939_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_24939_end_0 = const()[name = tensor<string, []>("op_24939_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_24939_end_mask_0 = const()[name = tensor<string, []>("op_24939_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24939_cast = slice_by_index(begin = var_24939_begin_0, end = var_24939_end_0, end_mask = var_24939_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24939_cast")];
            tensor<int32, [4]> var_24943_begin_0 = const()[name = tensor<string, []>("op_24943_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_24943_end_0 = const()[name = tensor<string, []>("op_24943_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_24943_end_mask_0 = const()[name = tensor<string, []>("op_24943_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24943_cast = slice_by_index(begin = var_24943_begin_0, end = var_24943_end_0, end_mask = var_24943_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24943_cast")];
            tensor<int32, [4]> var_24947_begin_0 = const()[name = tensor<string, []>("op_24947_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_24947_end_0 = const()[name = tensor<string, []>("op_24947_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_24947_end_mask_0 = const()[name = tensor<string, []>("op_24947_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24947_cast = slice_by_index(begin = var_24947_begin_0, end = var_24947_end_0, end_mask = var_24947_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24947_cast")];
            tensor<int32, [4]> var_24951_begin_0 = const()[name = tensor<string, []>("op_24951_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_24951_end_0 = const()[name = tensor<string, []>("op_24951_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_24951_end_mask_0 = const()[name = tensor<string, []>("op_24951_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24951_cast = slice_by_index(begin = var_24951_begin_0, end = var_24951_end_0, end_mask = var_24951_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24951_cast")];
            tensor<int32, [4]> var_24955_begin_0 = const()[name = tensor<string, []>("op_24955_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_24955_end_0 = const()[name = tensor<string, []>("op_24955_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_24955_end_mask_0 = const()[name = tensor<string, []>("op_24955_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24955_cast = slice_by_index(begin = var_24955_begin_0, end = var_24955_end_0, end_mask = var_24955_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24955_cast")];
            tensor<int32, [4]> var_24959_begin_0 = const()[name = tensor<string, []>("op_24959_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_24959_end_0 = const()[name = tensor<string, []>("op_24959_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_24959_end_mask_0 = const()[name = tensor<string, []>("op_24959_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_24959_cast = slice_by_index(begin = var_24959_begin_0, end = var_24959_end_0, end_mask = var_24959_end_mask_0, x = q_113_cast)[name = tensor<string, []>("op_24959_cast")];
            tensor<int32, [4]> k_227_perm_0 = const()[name = tensor<string, []>("k_227_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_24966_begin_0 = const()[name = tensor<string, []>("op_24966_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_24966_end_0 = const()[name = tensor<string, []>("op_24966_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_24966_end_mask_0 = const()[name = tensor<string, []>("op_24966_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_83 = transpose(perm = k_227_perm_0, x = k_225_cast)[name = tensor<string, []>("transpose_83")];
            tensor<fp16, [2, 1024, 1, 64]> var_24966_cast = slice_by_index(begin = var_24966_begin_0, end = var_24966_end_0, end_mask = var_24966_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_24966_cast")];
            tensor<int32, [4]> var_24970_begin_0 = const()[name = tensor<string, []>("op_24970_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_24970_end_0 = const()[name = tensor<string, []>("op_24970_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_24970_end_mask_0 = const()[name = tensor<string, []>("op_24970_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24970_cast = slice_by_index(begin = var_24970_begin_0, end = var_24970_end_0, end_mask = var_24970_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_24970_cast")];
            tensor<int32, [4]> var_24974_begin_0 = const()[name = tensor<string, []>("op_24974_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_24974_end_0 = const()[name = tensor<string, []>("op_24974_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_24974_end_mask_0 = const()[name = tensor<string, []>("op_24974_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24974_cast = slice_by_index(begin = var_24974_begin_0, end = var_24974_end_0, end_mask = var_24974_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_24974_cast")];
            tensor<int32, [4]> var_24978_begin_0 = const()[name = tensor<string, []>("op_24978_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_24978_end_0 = const()[name = tensor<string, []>("op_24978_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_24978_end_mask_0 = const()[name = tensor<string, []>("op_24978_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24978_cast = slice_by_index(begin = var_24978_begin_0, end = var_24978_end_0, end_mask = var_24978_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_24978_cast")];
            tensor<int32, [4]> var_24982_begin_0 = const()[name = tensor<string, []>("op_24982_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_24982_end_0 = const()[name = tensor<string, []>("op_24982_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_24982_end_mask_0 = const()[name = tensor<string, []>("op_24982_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24982_cast = slice_by_index(begin = var_24982_begin_0, end = var_24982_end_0, end_mask = var_24982_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_24982_cast")];
            tensor<int32, [4]> var_24986_begin_0 = const()[name = tensor<string, []>("op_24986_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_24986_end_0 = const()[name = tensor<string, []>("op_24986_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_24986_end_mask_0 = const()[name = tensor<string, []>("op_24986_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24986_cast = slice_by_index(begin = var_24986_begin_0, end = var_24986_end_0, end_mask = var_24986_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_24986_cast")];
            tensor<int32, [4]> var_24990_begin_0 = const()[name = tensor<string, []>("op_24990_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_24990_end_0 = const()[name = tensor<string, []>("op_24990_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_24990_end_mask_0 = const()[name = tensor<string, []>("op_24990_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24990_cast = slice_by_index(begin = var_24990_begin_0, end = var_24990_end_0, end_mask = var_24990_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_24990_cast")];
            tensor<int32, [4]> var_24994_begin_0 = const()[name = tensor<string, []>("op_24994_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_24994_end_0 = const()[name = tensor<string, []>("op_24994_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_24994_end_mask_0 = const()[name = tensor<string, []>("op_24994_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24994_cast = slice_by_index(begin = var_24994_begin_0, end = var_24994_end_0, end_mask = var_24994_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_24994_cast")];
            tensor<int32, [4]> var_24998_begin_0 = const()[name = tensor<string, []>("op_24998_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_24998_end_0 = const()[name = tensor<string, []>("op_24998_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_24998_end_mask_0 = const()[name = tensor<string, []>("op_24998_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_24998_cast = slice_by_index(begin = var_24998_begin_0, end = var_24998_end_0, end_mask = var_24998_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_24998_cast")];
            tensor<int32, [4]> var_25002_begin_0 = const()[name = tensor<string, []>("op_25002_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_25002_end_0 = const()[name = tensor<string, []>("op_25002_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_25002_end_mask_0 = const()[name = tensor<string, []>("op_25002_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25002_cast = slice_by_index(begin = var_25002_begin_0, end = var_25002_end_0, end_mask = var_25002_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_25002_cast")];
            tensor<int32, [4]> var_25006_begin_0 = const()[name = tensor<string, []>("op_25006_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_25006_end_0 = const()[name = tensor<string, []>("op_25006_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_25006_end_mask_0 = const()[name = tensor<string, []>("op_25006_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25006_cast = slice_by_index(begin = var_25006_begin_0, end = var_25006_end_0, end_mask = var_25006_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_25006_cast")];
            tensor<int32, [4]> var_25010_begin_0 = const()[name = tensor<string, []>("op_25010_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_25010_end_0 = const()[name = tensor<string, []>("op_25010_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_25010_end_mask_0 = const()[name = tensor<string, []>("op_25010_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25010_cast = slice_by_index(begin = var_25010_begin_0, end = var_25010_end_0, end_mask = var_25010_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_25010_cast")];
            tensor<int32, [4]> var_25014_begin_0 = const()[name = tensor<string, []>("op_25014_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_25014_end_0 = const()[name = tensor<string, []>("op_25014_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_25014_end_mask_0 = const()[name = tensor<string, []>("op_25014_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25014_cast = slice_by_index(begin = var_25014_begin_0, end = var_25014_end_0, end_mask = var_25014_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_25014_cast")];
            tensor<int32, [4]> var_25018_begin_0 = const()[name = tensor<string, []>("op_25018_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_25018_end_0 = const()[name = tensor<string, []>("op_25018_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_25018_end_mask_0 = const()[name = tensor<string, []>("op_25018_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25018_cast = slice_by_index(begin = var_25018_begin_0, end = var_25018_end_0, end_mask = var_25018_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_25018_cast")];
            tensor<int32, [4]> var_25022_begin_0 = const()[name = tensor<string, []>("op_25022_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_25022_end_0 = const()[name = tensor<string, []>("op_25022_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_25022_end_mask_0 = const()[name = tensor<string, []>("op_25022_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25022_cast = slice_by_index(begin = var_25022_begin_0, end = var_25022_end_0, end_mask = var_25022_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_25022_cast")];
            tensor<int32, [4]> var_25026_begin_0 = const()[name = tensor<string, []>("op_25026_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_25026_end_0 = const()[name = tensor<string, []>("op_25026_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_25026_end_mask_0 = const()[name = tensor<string, []>("op_25026_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25026_cast = slice_by_index(begin = var_25026_begin_0, end = var_25026_end_0, end_mask = var_25026_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_25026_cast")];
            tensor<int32, [4]> var_25030_begin_0 = const()[name = tensor<string, []>("op_25030_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_25030_end_0 = const()[name = tensor<string, []>("op_25030_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_25030_end_mask_0 = const()[name = tensor<string, []>("op_25030_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25030_cast = slice_by_index(begin = var_25030_begin_0, end = var_25030_end_0, end_mask = var_25030_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_25030_cast")];
            tensor<int32, [4]> var_25034_begin_0 = const()[name = tensor<string, []>("op_25034_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_25034_end_0 = const()[name = tensor<string, []>("op_25034_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_25034_end_mask_0 = const()[name = tensor<string, []>("op_25034_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25034_cast = slice_by_index(begin = var_25034_begin_0, end = var_25034_end_0, end_mask = var_25034_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_25034_cast")];
            tensor<int32, [4]> var_25038_begin_0 = const()[name = tensor<string, []>("op_25038_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_25038_end_0 = const()[name = tensor<string, []>("op_25038_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_25038_end_mask_0 = const()[name = tensor<string, []>("op_25038_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25038_cast = slice_by_index(begin = var_25038_begin_0, end = var_25038_end_0, end_mask = var_25038_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_25038_cast")];
            tensor<int32, [4]> var_25042_begin_0 = const()[name = tensor<string, []>("op_25042_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_25042_end_0 = const()[name = tensor<string, []>("op_25042_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_25042_end_mask_0 = const()[name = tensor<string, []>("op_25042_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25042_cast = slice_by_index(begin = var_25042_begin_0, end = var_25042_end_0, end_mask = var_25042_end_mask_0, x = transpose_83)[name = tensor<string, []>("op_25042_cast")];
            tensor<int32, [4]> var_25044_begin_0 = const()[name = tensor<string, []>("op_25044_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_25044_end_0 = const()[name = tensor<string, []>("op_25044_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_25044_end_mask_0 = const()[name = tensor<string, []>("op_25044_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25044_cast = slice_by_index(begin = var_25044_begin_0, end = var_25044_end_0, end_mask = var_25044_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25044_cast")];
            tensor<int32, [4]> var_25048_begin_0 = const()[name = tensor<string, []>("op_25048_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_25048_end_0 = const()[name = tensor<string, []>("op_25048_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_25048_end_mask_0 = const()[name = tensor<string, []>("op_25048_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25048_cast = slice_by_index(begin = var_25048_begin_0, end = var_25048_end_0, end_mask = var_25048_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25048_cast")];
            tensor<int32, [4]> var_25052_begin_0 = const()[name = tensor<string, []>("op_25052_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_25052_end_0 = const()[name = tensor<string, []>("op_25052_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_25052_end_mask_0 = const()[name = tensor<string, []>("op_25052_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25052_cast = slice_by_index(begin = var_25052_begin_0, end = var_25052_end_0, end_mask = var_25052_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25052_cast")];
            tensor<int32, [4]> var_25056_begin_0 = const()[name = tensor<string, []>("op_25056_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_25056_end_0 = const()[name = tensor<string, []>("op_25056_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_25056_end_mask_0 = const()[name = tensor<string, []>("op_25056_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25056_cast = slice_by_index(begin = var_25056_begin_0, end = var_25056_end_0, end_mask = var_25056_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25056_cast")];
            tensor<int32, [4]> var_25060_begin_0 = const()[name = tensor<string, []>("op_25060_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_25060_end_0 = const()[name = tensor<string, []>("op_25060_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_25060_end_mask_0 = const()[name = tensor<string, []>("op_25060_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25060_cast = slice_by_index(begin = var_25060_begin_0, end = var_25060_end_0, end_mask = var_25060_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25060_cast")];
            tensor<int32, [4]> var_25064_begin_0 = const()[name = tensor<string, []>("op_25064_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_25064_end_0 = const()[name = tensor<string, []>("op_25064_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_25064_end_mask_0 = const()[name = tensor<string, []>("op_25064_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25064_cast = slice_by_index(begin = var_25064_begin_0, end = var_25064_end_0, end_mask = var_25064_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25064_cast")];
            tensor<int32, [4]> var_25068_begin_0 = const()[name = tensor<string, []>("op_25068_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_25068_end_0 = const()[name = tensor<string, []>("op_25068_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_25068_end_mask_0 = const()[name = tensor<string, []>("op_25068_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25068_cast = slice_by_index(begin = var_25068_begin_0, end = var_25068_end_0, end_mask = var_25068_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25068_cast")];
            tensor<int32, [4]> var_25072_begin_0 = const()[name = tensor<string, []>("op_25072_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_25072_end_0 = const()[name = tensor<string, []>("op_25072_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_25072_end_mask_0 = const()[name = tensor<string, []>("op_25072_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25072_cast = slice_by_index(begin = var_25072_begin_0, end = var_25072_end_0, end_mask = var_25072_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25072_cast")];
            tensor<int32, [4]> var_25076_begin_0 = const()[name = tensor<string, []>("op_25076_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_25076_end_0 = const()[name = tensor<string, []>("op_25076_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_25076_end_mask_0 = const()[name = tensor<string, []>("op_25076_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25076_cast = slice_by_index(begin = var_25076_begin_0, end = var_25076_end_0, end_mask = var_25076_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25076_cast")];
            tensor<int32, [4]> var_25080_begin_0 = const()[name = tensor<string, []>("op_25080_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_25080_end_0 = const()[name = tensor<string, []>("op_25080_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_25080_end_mask_0 = const()[name = tensor<string, []>("op_25080_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25080_cast = slice_by_index(begin = var_25080_begin_0, end = var_25080_end_0, end_mask = var_25080_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25080_cast")];
            tensor<int32, [4]> var_25084_begin_0 = const()[name = tensor<string, []>("op_25084_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_25084_end_0 = const()[name = tensor<string, []>("op_25084_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_25084_end_mask_0 = const()[name = tensor<string, []>("op_25084_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25084_cast = slice_by_index(begin = var_25084_begin_0, end = var_25084_end_0, end_mask = var_25084_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25084_cast")];
            tensor<int32, [4]> var_25088_begin_0 = const()[name = tensor<string, []>("op_25088_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_25088_end_0 = const()[name = tensor<string, []>("op_25088_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_25088_end_mask_0 = const()[name = tensor<string, []>("op_25088_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25088_cast = slice_by_index(begin = var_25088_begin_0, end = var_25088_end_0, end_mask = var_25088_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25088_cast")];
            tensor<int32, [4]> var_25092_begin_0 = const()[name = tensor<string, []>("op_25092_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_25092_end_0 = const()[name = tensor<string, []>("op_25092_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_25092_end_mask_0 = const()[name = tensor<string, []>("op_25092_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25092_cast = slice_by_index(begin = var_25092_begin_0, end = var_25092_end_0, end_mask = var_25092_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25092_cast")];
            tensor<int32, [4]> var_25096_begin_0 = const()[name = tensor<string, []>("op_25096_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_25096_end_0 = const()[name = tensor<string, []>("op_25096_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_25096_end_mask_0 = const()[name = tensor<string, []>("op_25096_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25096_cast = slice_by_index(begin = var_25096_begin_0, end = var_25096_end_0, end_mask = var_25096_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25096_cast")];
            tensor<int32, [4]> var_25100_begin_0 = const()[name = tensor<string, []>("op_25100_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_25100_end_0 = const()[name = tensor<string, []>("op_25100_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_25100_end_mask_0 = const()[name = tensor<string, []>("op_25100_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25100_cast = slice_by_index(begin = var_25100_begin_0, end = var_25100_end_0, end_mask = var_25100_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25100_cast")];
            tensor<int32, [4]> var_25104_begin_0 = const()[name = tensor<string, []>("op_25104_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_25104_end_0 = const()[name = tensor<string, []>("op_25104_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_25104_end_mask_0 = const()[name = tensor<string, []>("op_25104_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25104_cast = slice_by_index(begin = var_25104_begin_0, end = var_25104_end_0, end_mask = var_25104_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25104_cast")];
            tensor<int32, [4]> var_25108_begin_0 = const()[name = tensor<string, []>("op_25108_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_25108_end_0 = const()[name = tensor<string, []>("op_25108_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_25108_end_mask_0 = const()[name = tensor<string, []>("op_25108_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25108_cast = slice_by_index(begin = var_25108_begin_0, end = var_25108_end_0, end_mask = var_25108_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25108_cast")];
            tensor<int32, [4]> var_25112_begin_0 = const()[name = tensor<string, []>("op_25112_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_25112_end_0 = const()[name = tensor<string, []>("op_25112_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_25112_end_mask_0 = const()[name = tensor<string, []>("op_25112_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25112_cast = slice_by_index(begin = var_25112_begin_0, end = var_25112_end_0, end_mask = var_25112_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25112_cast")];
            tensor<int32, [4]> var_25116_begin_0 = const()[name = tensor<string, []>("op_25116_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_25116_end_0 = const()[name = tensor<string, []>("op_25116_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_25116_end_mask_0 = const()[name = tensor<string, []>("op_25116_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25116_cast = slice_by_index(begin = var_25116_begin_0, end = var_25116_end_0, end_mask = var_25116_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25116_cast")];
            tensor<int32, [4]> var_25120_begin_0 = const()[name = tensor<string, []>("op_25120_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_25120_end_0 = const()[name = tensor<string, []>("op_25120_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_25120_end_mask_0 = const()[name = tensor<string, []>("op_25120_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25120_cast = slice_by_index(begin = var_25120_begin_0, end = var_25120_end_0, end_mask = var_25120_end_mask_0, x = v_113_cast)[name = tensor<string, []>("op_25120_cast")];
            tensor<string, []> var_25124_equation_0 = const()[name = tensor<string, []>("op_25124_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25124_cast = einsum(equation = var_25124_equation_0, values = (var_24966_cast, var_24883_cast))[name = tensor<string, []>("op_25124_cast")];
            tensor<fp16, []> var_25125_to_fp16 = const()[name = tensor<string, []>("op_25125_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2081_cast = mul(x = var_25124_cast, y = var_25125_to_fp16)[name = tensor<string, []>("aw_2081_cast")];
            tensor<string, []> var_25128_equation_0 = const()[name = tensor<string, []>("op_25128_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25128_cast = einsum(equation = var_25128_equation_0, values = (var_24970_cast, var_24887_cast))[name = tensor<string, []>("op_25128_cast")];
            tensor<fp16, []> var_25129_to_fp16 = const()[name = tensor<string, []>("op_25129_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2083_cast = mul(x = var_25128_cast, y = var_25129_to_fp16)[name = tensor<string, []>("aw_2083_cast")];
            tensor<string, []> var_25132_equation_0 = const()[name = tensor<string, []>("op_25132_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25132_cast = einsum(equation = var_25132_equation_0, values = (var_24974_cast, var_24891_cast))[name = tensor<string, []>("op_25132_cast")];
            tensor<fp16, []> var_25133_to_fp16 = const()[name = tensor<string, []>("op_25133_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2085_cast = mul(x = var_25132_cast, y = var_25133_to_fp16)[name = tensor<string, []>("aw_2085_cast")];
            tensor<string, []> var_25136_equation_0 = const()[name = tensor<string, []>("op_25136_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25136_cast = einsum(equation = var_25136_equation_0, values = (var_24978_cast, var_24895_cast))[name = tensor<string, []>("op_25136_cast")];
            tensor<fp16, []> var_25137_to_fp16 = const()[name = tensor<string, []>("op_25137_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2087_cast = mul(x = var_25136_cast, y = var_25137_to_fp16)[name = tensor<string, []>("aw_2087_cast")];
            tensor<string, []> var_25140_equation_0 = const()[name = tensor<string, []>("op_25140_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25140_cast = einsum(equation = var_25140_equation_0, values = (var_24982_cast, var_24899_cast))[name = tensor<string, []>("op_25140_cast")];
            tensor<fp16, []> var_25141_to_fp16 = const()[name = tensor<string, []>("op_25141_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2089_cast = mul(x = var_25140_cast, y = var_25141_to_fp16)[name = tensor<string, []>("aw_2089_cast")];
            tensor<string, []> var_25144_equation_0 = const()[name = tensor<string, []>("op_25144_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25144_cast = einsum(equation = var_25144_equation_0, values = (var_24986_cast, var_24903_cast))[name = tensor<string, []>("op_25144_cast")];
            tensor<fp16, []> var_25145_to_fp16 = const()[name = tensor<string, []>("op_25145_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2091_cast = mul(x = var_25144_cast, y = var_25145_to_fp16)[name = tensor<string, []>("aw_2091_cast")];
            tensor<string, []> var_25148_equation_0 = const()[name = tensor<string, []>("op_25148_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25148_cast = einsum(equation = var_25148_equation_0, values = (var_24990_cast, var_24907_cast))[name = tensor<string, []>("op_25148_cast")];
            tensor<fp16, []> var_25149_to_fp16 = const()[name = tensor<string, []>("op_25149_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2093_cast = mul(x = var_25148_cast, y = var_25149_to_fp16)[name = tensor<string, []>("aw_2093_cast")];
            tensor<string, []> var_25152_equation_0 = const()[name = tensor<string, []>("op_25152_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25152_cast = einsum(equation = var_25152_equation_0, values = (var_24994_cast, var_24911_cast))[name = tensor<string, []>("op_25152_cast")];
            tensor<fp16, []> var_25153_to_fp16 = const()[name = tensor<string, []>("op_25153_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2095_cast = mul(x = var_25152_cast, y = var_25153_to_fp16)[name = tensor<string, []>("aw_2095_cast")];
            tensor<string, []> var_25156_equation_0 = const()[name = tensor<string, []>("op_25156_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25156_cast = einsum(equation = var_25156_equation_0, values = (var_24998_cast, var_24915_cast))[name = tensor<string, []>("op_25156_cast")];
            tensor<fp16, []> var_25157_to_fp16 = const()[name = tensor<string, []>("op_25157_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2097_cast = mul(x = var_25156_cast, y = var_25157_to_fp16)[name = tensor<string, []>("aw_2097_cast")];
            tensor<string, []> var_25160_equation_0 = const()[name = tensor<string, []>("op_25160_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25160_cast = einsum(equation = var_25160_equation_0, values = (var_25002_cast, var_24919_cast))[name = tensor<string, []>("op_25160_cast")];
            tensor<fp16, []> var_25161_to_fp16 = const()[name = tensor<string, []>("op_25161_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2099_cast = mul(x = var_25160_cast, y = var_25161_to_fp16)[name = tensor<string, []>("aw_2099_cast")];
            tensor<string, []> var_25164_equation_0 = const()[name = tensor<string, []>("op_25164_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25164_cast = einsum(equation = var_25164_equation_0, values = (var_25006_cast, var_24923_cast))[name = tensor<string, []>("op_25164_cast")];
            tensor<fp16, []> var_25165_to_fp16 = const()[name = tensor<string, []>("op_25165_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2101_cast = mul(x = var_25164_cast, y = var_25165_to_fp16)[name = tensor<string, []>("aw_2101_cast")];
            tensor<string, []> var_25168_equation_0 = const()[name = tensor<string, []>("op_25168_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25168_cast = einsum(equation = var_25168_equation_0, values = (var_25010_cast, var_24927_cast))[name = tensor<string, []>("op_25168_cast")];
            tensor<fp16, []> var_25169_to_fp16 = const()[name = tensor<string, []>("op_25169_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2103_cast = mul(x = var_25168_cast, y = var_25169_to_fp16)[name = tensor<string, []>("aw_2103_cast")];
            tensor<string, []> var_25172_equation_0 = const()[name = tensor<string, []>("op_25172_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25172_cast = einsum(equation = var_25172_equation_0, values = (var_25014_cast, var_24931_cast))[name = tensor<string, []>("op_25172_cast")];
            tensor<fp16, []> var_25173_to_fp16 = const()[name = tensor<string, []>("op_25173_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2105_cast = mul(x = var_25172_cast, y = var_25173_to_fp16)[name = tensor<string, []>("aw_2105_cast")];
            tensor<string, []> var_25176_equation_0 = const()[name = tensor<string, []>("op_25176_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25176_cast = einsum(equation = var_25176_equation_0, values = (var_25018_cast, var_24935_cast))[name = tensor<string, []>("op_25176_cast")];
            tensor<fp16, []> var_25177_to_fp16 = const()[name = tensor<string, []>("op_25177_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2107_cast = mul(x = var_25176_cast, y = var_25177_to_fp16)[name = tensor<string, []>("aw_2107_cast")];
            tensor<string, []> var_25180_equation_0 = const()[name = tensor<string, []>("op_25180_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25180_cast = einsum(equation = var_25180_equation_0, values = (var_25022_cast, var_24939_cast))[name = tensor<string, []>("op_25180_cast")];
            tensor<fp16, []> var_25181_to_fp16 = const()[name = tensor<string, []>("op_25181_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2109_cast = mul(x = var_25180_cast, y = var_25181_to_fp16)[name = tensor<string, []>("aw_2109_cast")];
            tensor<string, []> var_25184_equation_0 = const()[name = tensor<string, []>("op_25184_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25184_cast = einsum(equation = var_25184_equation_0, values = (var_25026_cast, var_24943_cast))[name = tensor<string, []>("op_25184_cast")];
            tensor<fp16, []> var_25185_to_fp16 = const()[name = tensor<string, []>("op_25185_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2111_cast = mul(x = var_25184_cast, y = var_25185_to_fp16)[name = tensor<string, []>("aw_2111_cast")];
            tensor<string, []> var_25188_equation_0 = const()[name = tensor<string, []>("op_25188_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25188_cast = einsum(equation = var_25188_equation_0, values = (var_25030_cast, var_24947_cast))[name = tensor<string, []>("op_25188_cast")];
            tensor<fp16, []> var_25189_to_fp16 = const()[name = tensor<string, []>("op_25189_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2113_cast = mul(x = var_25188_cast, y = var_25189_to_fp16)[name = tensor<string, []>("aw_2113_cast")];
            tensor<string, []> var_25192_equation_0 = const()[name = tensor<string, []>("op_25192_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25192_cast = einsum(equation = var_25192_equation_0, values = (var_25034_cast, var_24951_cast))[name = tensor<string, []>("op_25192_cast")];
            tensor<fp16, []> var_25193_to_fp16 = const()[name = tensor<string, []>("op_25193_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2115_cast = mul(x = var_25192_cast, y = var_25193_to_fp16)[name = tensor<string, []>("aw_2115_cast")];
            tensor<string, []> var_25196_equation_0 = const()[name = tensor<string, []>("op_25196_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25196_cast = einsum(equation = var_25196_equation_0, values = (var_25038_cast, var_24955_cast))[name = tensor<string, []>("op_25196_cast")];
            tensor<fp16, []> var_25197_to_fp16 = const()[name = tensor<string, []>("op_25197_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2117_cast = mul(x = var_25196_cast, y = var_25197_to_fp16)[name = tensor<string, []>("aw_2117_cast")];
            tensor<string, []> var_25200_equation_0 = const()[name = tensor<string, []>("op_25200_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25200_cast = einsum(equation = var_25200_equation_0, values = (var_25042_cast, var_24959_cast))[name = tensor<string, []>("op_25200_cast")];
            tensor<fp16, []> var_25201_to_fp16 = const()[name = tensor<string, []>("op_25201_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2119_cast = mul(x = var_25200_cast, y = var_25201_to_fp16)[name = tensor<string, []>("aw_2119_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25203_cast = softmax(axis = var_21105, x = aw_2081_cast)[name = tensor<string, []>("op_25203_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25204_cast = softmax(axis = var_21105, x = aw_2083_cast)[name = tensor<string, []>("op_25204_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25205_cast = softmax(axis = var_21105, x = aw_2085_cast)[name = tensor<string, []>("op_25205_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25206_cast = softmax(axis = var_21105, x = aw_2087_cast)[name = tensor<string, []>("op_25206_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25207_cast = softmax(axis = var_21105, x = aw_2089_cast)[name = tensor<string, []>("op_25207_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25208_cast = softmax(axis = var_21105, x = aw_2091_cast)[name = tensor<string, []>("op_25208_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25209_cast = softmax(axis = var_21105, x = aw_2093_cast)[name = tensor<string, []>("op_25209_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25210_cast = softmax(axis = var_21105, x = aw_2095_cast)[name = tensor<string, []>("op_25210_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25211_cast = softmax(axis = var_21105, x = aw_2097_cast)[name = tensor<string, []>("op_25211_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25212_cast = softmax(axis = var_21105, x = aw_2099_cast)[name = tensor<string, []>("op_25212_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25213_cast = softmax(axis = var_21105, x = aw_2101_cast)[name = tensor<string, []>("op_25213_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25214_cast = softmax(axis = var_21105, x = aw_2103_cast)[name = tensor<string, []>("op_25214_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25215_cast = softmax(axis = var_21105, x = aw_2105_cast)[name = tensor<string, []>("op_25215_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25216_cast = softmax(axis = var_21105, x = aw_2107_cast)[name = tensor<string, []>("op_25216_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25217_cast = softmax(axis = var_21105, x = aw_2109_cast)[name = tensor<string, []>("op_25217_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25218_cast = softmax(axis = var_21105, x = aw_2111_cast)[name = tensor<string, []>("op_25218_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25219_cast = softmax(axis = var_21105, x = aw_2113_cast)[name = tensor<string, []>("op_25219_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25220_cast = softmax(axis = var_21105, x = aw_2115_cast)[name = tensor<string, []>("op_25220_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25221_cast = softmax(axis = var_21105, x = aw_2117_cast)[name = tensor<string, []>("op_25221_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_25222_cast = softmax(axis = var_21105, x = aw_2119_cast)[name = tensor<string, []>("op_25222_cast")];
            tensor<string, []> var_25224_equation_0 = const()[name = tensor<string, []>("op_25224_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25224_cast = einsum(equation = var_25224_equation_0, values = (var_25044_cast, var_25203_cast))[name = tensor<string, []>("op_25224_cast")];
            tensor<string, []> var_25226_equation_0 = const()[name = tensor<string, []>("op_25226_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25226_cast = einsum(equation = var_25226_equation_0, values = (var_25048_cast, var_25204_cast))[name = tensor<string, []>("op_25226_cast")];
            tensor<string, []> var_25228_equation_0 = const()[name = tensor<string, []>("op_25228_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25228_cast = einsum(equation = var_25228_equation_0, values = (var_25052_cast, var_25205_cast))[name = tensor<string, []>("op_25228_cast")];
            tensor<string, []> var_25230_equation_0 = const()[name = tensor<string, []>("op_25230_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25230_cast = einsum(equation = var_25230_equation_0, values = (var_25056_cast, var_25206_cast))[name = tensor<string, []>("op_25230_cast")];
            tensor<string, []> var_25232_equation_0 = const()[name = tensor<string, []>("op_25232_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25232_cast = einsum(equation = var_25232_equation_0, values = (var_25060_cast, var_25207_cast))[name = tensor<string, []>("op_25232_cast")];
            tensor<string, []> var_25234_equation_0 = const()[name = tensor<string, []>("op_25234_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25234_cast = einsum(equation = var_25234_equation_0, values = (var_25064_cast, var_25208_cast))[name = tensor<string, []>("op_25234_cast")];
            tensor<string, []> var_25236_equation_0 = const()[name = tensor<string, []>("op_25236_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25236_cast = einsum(equation = var_25236_equation_0, values = (var_25068_cast, var_25209_cast))[name = tensor<string, []>("op_25236_cast")];
            tensor<string, []> var_25238_equation_0 = const()[name = tensor<string, []>("op_25238_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25238_cast = einsum(equation = var_25238_equation_0, values = (var_25072_cast, var_25210_cast))[name = tensor<string, []>("op_25238_cast")];
            tensor<string, []> var_25240_equation_0 = const()[name = tensor<string, []>("op_25240_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25240_cast = einsum(equation = var_25240_equation_0, values = (var_25076_cast, var_25211_cast))[name = tensor<string, []>("op_25240_cast")];
            tensor<string, []> var_25242_equation_0 = const()[name = tensor<string, []>("op_25242_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25242_cast = einsum(equation = var_25242_equation_0, values = (var_25080_cast, var_25212_cast))[name = tensor<string, []>("op_25242_cast")];
            tensor<string, []> var_25244_equation_0 = const()[name = tensor<string, []>("op_25244_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25244_cast = einsum(equation = var_25244_equation_0, values = (var_25084_cast, var_25213_cast))[name = tensor<string, []>("op_25244_cast")];
            tensor<string, []> var_25246_equation_0 = const()[name = tensor<string, []>("op_25246_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25246_cast = einsum(equation = var_25246_equation_0, values = (var_25088_cast, var_25214_cast))[name = tensor<string, []>("op_25246_cast")];
            tensor<string, []> var_25248_equation_0 = const()[name = tensor<string, []>("op_25248_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25248_cast = einsum(equation = var_25248_equation_0, values = (var_25092_cast, var_25215_cast))[name = tensor<string, []>("op_25248_cast")];
            tensor<string, []> var_25250_equation_0 = const()[name = tensor<string, []>("op_25250_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25250_cast = einsum(equation = var_25250_equation_0, values = (var_25096_cast, var_25216_cast))[name = tensor<string, []>("op_25250_cast")];
            tensor<string, []> var_25252_equation_0 = const()[name = tensor<string, []>("op_25252_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25252_cast = einsum(equation = var_25252_equation_0, values = (var_25100_cast, var_25217_cast))[name = tensor<string, []>("op_25252_cast")];
            tensor<string, []> var_25254_equation_0 = const()[name = tensor<string, []>("op_25254_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25254_cast = einsum(equation = var_25254_equation_0, values = (var_25104_cast, var_25218_cast))[name = tensor<string, []>("op_25254_cast")];
            tensor<string, []> var_25256_equation_0 = const()[name = tensor<string, []>("op_25256_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25256_cast = einsum(equation = var_25256_equation_0, values = (var_25108_cast, var_25219_cast))[name = tensor<string, []>("op_25256_cast")];
            tensor<string, []> var_25258_equation_0 = const()[name = tensor<string, []>("op_25258_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25258_cast = einsum(equation = var_25258_equation_0, values = (var_25112_cast, var_25220_cast))[name = tensor<string, []>("op_25258_cast")];
            tensor<string, []> var_25260_equation_0 = const()[name = tensor<string, []>("op_25260_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25260_cast = einsum(equation = var_25260_equation_0, values = (var_25116_cast, var_25221_cast))[name = tensor<string, []>("op_25260_cast")];
            tensor<string, []> var_25262_equation_0 = const()[name = tensor<string, []>("op_25262_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25262_cast = einsum(equation = var_25262_equation_0, values = (var_25120_cast, var_25222_cast))[name = tensor<string, []>("op_25262_cast")];
            tensor<bool, []> input_361_interleave_0 = const()[name = tensor<string, []>("input_361_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_361_cast = concat(axis = var_21105, interleave = input_361_interleave_0, values = (var_25224_cast, var_25226_cast, var_25228_cast, var_25230_cast, var_25232_cast, var_25234_cast, var_25236_cast, var_25238_cast, var_25240_cast, var_25242_cast, var_25244_cast, var_25246_cast, var_25248_cast, var_25250_cast, var_25252_cast, var_25254_cast, var_25256_cast, var_25258_cast, var_25260_cast, var_25262_cast))[name = tensor<string, []>("input_361_cast")];
            tensor<int32, [2]> var_25268 = const()[name = tensor<string, []>("op_25268"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_25270 = const()[name = tensor<string, []>("op_25270"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_25272_pad_type_0 = const()[name = tensor<string, []>("op_25272_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_25272_pad_0 = const()[name = tensor<string, []>("op_25272_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2013996736)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2017273600)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_25272_cast = conv(bias = mid_block_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16, dilations = var_25270, groups = var_21105, pad = var_25272_pad_0, pad_type = var_25272_pad_type_0, strides = var_25268, weight = mid_block_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16, x = input_361_cast)[name = tensor<string, []>("op_25272_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_171_cast = add(x = var_25272_cast, y = inputs_169_cast)[name = tensor<string, []>("inputs_171_cast")];
            tensor<int32, [1]> var_25276 = const()[name = tensor<string, []>("op_25276"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_171_cast = reduce_mean(axes = var_25276, keep_dims = var_21100, x = inputs_171_cast)[name = tensor<string, []>("channels_mean_171_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_171_cast = sub(x = inputs_171_cast, y = channels_mean_171_cast)[name = tensor<string, []>("zero_mean_171_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_171_cast = mul(x = zero_mean_171_cast, y = zero_mean_171_cast)[name = tensor<string, []>("zero_mean_sq_171_cast")];
            tensor<int32, [1]> var_25280 = const()[name = tensor<string, []>("op_25280"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_25281_cast = reduce_mean(axes = var_25280, keep_dims = var_21100, x = zero_mean_sq_171_cast)[name = tensor<string, []>("op_25281_cast")];
            tensor<fp16, []> var_25282_to_fp16 = const()[name = tensor<string, []>("op_25282_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_25283_cast = add(x = var_25281_cast, y = var_25282_to_fp16)[name = tensor<string, []>("op_25283_cast")];
            tensor<fp16, []> denom_171_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_171_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_171_cast = rsqrt(epsilon = denom_171_epsilon_0_to_fp16, x = var_25283_cast)[name = tensor<string, []>("denom_171_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_171_cast = mul(x = zero_mean_171_cast, y = denom_171_cast)[name = tensor<string, []>("out_171_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_25287_to_fp16 = const()[name = tensor<string, []>("op_25287_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2017276224)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_25288_cast = add(x = out_171_cast, y = var_25287_to_fp16)[name = tensor<string, []>("op_25288_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_25290_to_fp16 = const()[name = tensor<string, []>("op_25290_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2017278848)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_235_cast = mul(x = var_25288_cast, y = var_25290_to_fp16)[name = tensor<string, []>("hidden_states_235_cast")];
            tensor<int32, [2]> var_25297 = const()[name = tensor<string, []>("op_25297"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_25299 = const()[name = tensor<string, []>("op_25299"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_115_pad_type_0 = const()[name = tensor<string, []>("q_115_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_115_pad_0 = const()[name = tensor<string, []>("q_115_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2017281472)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_115_cast = conv(dilations = var_25299, groups = var_21105, pad = q_115_pad_0, pad_type = q_115_pad_type_0, strides = var_25297, weight = mid_block_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16, x = hidden_states_235_cast)[name = tensor<string, []>("q_115_cast")];
            tensor<int32, [2]> var_25303 = const()[name = tensor<string, []>("op_25303"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_25305 = const()[name = tensor<string, []>("op_25305"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_229_pad_type_0 = const()[name = tensor<string, []>("k_229_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_229_pad_0 = const()[name = tensor<string, []>("k_229_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2020558336)))];
            tensor<fp16, [2, 1280, 1, 77]> k_229_cast = conv(dilations = var_25305, groups = var_21105, pad = k_229_pad_0, pad_type = k_229_pad_type_0, strides = var_25303, weight = mid_block_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_229_cast")];
            tensor<int32, [2]> var_25309 = const()[name = tensor<string, []>("op_25309"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_25311 = const()[name = tensor<string, []>("op_25311"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_115_pad_type_0 = const()[name = tensor<string, []>("v_115_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_115_pad_0 = const()[name = tensor<string, []>("v_115_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2025801280)))];
            tensor<fp16, [2, 1280, 1, 77]> v_115_cast = conv(dilations = var_25311, groups = var_21105, pad = v_115_pad_0, pad_type = v_115_pad_type_0, strides = var_25309, weight = mid_block_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_115_cast")];
            tensor<int32, [4]> var_25315_begin_0 = const()[name = tensor<string, []>("op_25315_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_25315_end_0 = const()[name = tensor<string, []>("op_25315_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_25315_end_mask_0 = const()[name = tensor<string, []>("op_25315_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25315_cast = slice_by_index(begin = var_25315_begin_0, end = var_25315_end_0, end_mask = var_25315_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25315_cast")];
            tensor<int32, [4]> var_25319_begin_0 = const()[name = tensor<string, []>("op_25319_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_25319_end_0 = const()[name = tensor<string, []>("op_25319_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_25319_end_mask_0 = const()[name = tensor<string, []>("op_25319_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25319_cast = slice_by_index(begin = var_25319_begin_0, end = var_25319_end_0, end_mask = var_25319_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25319_cast")];
            tensor<int32, [4]> var_25323_begin_0 = const()[name = tensor<string, []>("op_25323_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_25323_end_0 = const()[name = tensor<string, []>("op_25323_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_25323_end_mask_0 = const()[name = tensor<string, []>("op_25323_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25323_cast = slice_by_index(begin = var_25323_begin_0, end = var_25323_end_0, end_mask = var_25323_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25323_cast")];
            tensor<int32, [4]> var_25327_begin_0 = const()[name = tensor<string, []>("op_25327_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_25327_end_0 = const()[name = tensor<string, []>("op_25327_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_25327_end_mask_0 = const()[name = tensor<string, []>("op_25327_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25327_cast = slice_by_index(begin = var_25327_begin_0, end = var_25327_end_0, end_mask = var_25327_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25327_cast")];
            tensor<int32, [4]> var_25331_begin_0 = const()[name = tensor<string, []>("op_25331_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_25331_end_0 = const()[name = tensor<string, []>("op_25331_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_25331_end_mask_0 = const()[name = tensor<string, []>("op_25331_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25331_cast = slice_by_index(begin = var_25331_begin_0, end = var_25331_end_0, end_mask = var_25331_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25331_cast")];
            tensor<int32, [4]> var_25335_begin_0 = const()[name = tensor<string, []>("op_25335_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_25335_end_0 = const()[name = tensor<string, []>("op_25335_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_25335_end_mask_0 = const()[name = tensor<string, []>("op_25335_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25335_cast = slice_by_index(begin = var_25335_begin_0, end = var_25335_end_0, end_mask = var_25335_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25335_cast")];
            tensor<int32, [4]> var_25339_begin_0 = const()[name = tensor<string, []>("op_25339_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_25339_end_0 = const()[name = tensor<string, []>("op_25339_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_25339_end_mask_0 = const()[name = tensor<string, []>("op_25339_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25339_cast = slice_by_index(begin = var_25339_begin_0, end = var_25339_end_0, end_mask = var_25339_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25339_cast")];
            tensor<int32, [4]> var_25343_begin_0 = const()[name = tensor<string, []>("op_25343_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_25343_end_0 = const()[name = tensor<string, []>("op_25343_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_25343_end_mask_0 = const()[name = tensor<string, []>("op_25343_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25343_cast = slice_by_index(begin = var_25343_begin_0, end = var_25343_end_0, end_mask = var_25343_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25343_cast")];
            tensor<int32, [4]> var_25347_begin_0 = const()[name = tensor<string, []>("op_25347_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_25347_end_0 = const()[name = tensor<string, []>("op_25347_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_25347_end_mask_0 = const()[name = tensor<string, []>("op_25347_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25347_cast = slice_by_index(begin = var_25347_begin_0, end = var_25347_end_0, end_mask = var_25347_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25347_cast")];
            tensor<int32, [4]> var_25351_begin_0 = const()[name = tensor<string, []>("op_25351_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_25351_end_0 = const()[name = tensor<string, []>("op_25351_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_25351_end_mask_0 = const()[name = tensor<string, []>("op_25351_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25351_cast = slice_by_index(begin = var_25351_begin_0, end = var_25351_end_0, end_mask = var_25351_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25351_cast")];
            tensor<int32, [4]> var_25355_begin_0 = const()[name = tensor<string, []>("op_25355_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_25355_end_0 = const()[name = tensor<string, []>("op_25355_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_25355_end_mask_0 = const()[name = tensor<string, []>("op_25355_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25355_cast = slice_by_index(begin = var_25355_begin_0, end = var_25355_end_0, end_mask = var_25355_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25355_cast")];
            tensor<int32, [4]> var_25359_begin_0 = const()[name = tensor<string, []>("op_25359_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_25359_end_0 = const()[name = tensor<string, []>("op_25359_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_25359_end_mask_0 = const()[name = tensor<string, []>("op_25359_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25359_cast = slice_by_index(begin = var_25359_begin_0, end = var_25359_end_0, end_mask = var_25359_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25359_cast")];
            tensor<int32, [4]> var_25363_begin_0 = const()[name = tensor<string, []>("op_25363_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_25363_end_0 = const()[name = tensor<string, []>("op_25363_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_25363_end_mask_0 = const()[name = tensor<string, []>("op_25363_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25363_cast = slice_by_index(begin = var_25363_begin_0, end = var_25363_end_0, end_mask = var_25363_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25363_cast")];
            tensor<int32, [4]> var_25367_begin_0 = const()[name = tensor<string, []>("op_25367_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_25367_end_0 = const()[name = tensor<string, []>("op_25367_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_25367_end_mask_0 = const()[name = tensor<string, []>("op_25367_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25367_cast = slice_by_index(begin = var_25367_begin_0, end = var_25367_end_0, end_mask = var_25367_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25367_cast")];
            tensor<int32, [4]> var_25371_begin_0 = const()[name = tensor<string, []>("op_25371_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_25371_end_0 = const()[name = tensor<string, []>("op_25371_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_25371_end_mask_0 = const()[name = tensor<string, []>("op_25371_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25371_cast = slice_by_index(begin = var_25371_begin_0, end = var_25371_end_0, end_mask = var_25371_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25371_cast")];
            tensor<int32, [4]> var_25375_begin_0 = const()[name = tensor<string, []>("op_25375_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_25375_end_0 = const()[name = tensor<string, []>("op_25375_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_25375_end_mask_0 = const()[name = tensor<string, []>("op_25375_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25375_cast = slice_by_index(begin = var_25375_begin_0, end = var_25375_end_0, end_mask = var_25375_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25375_cast")];
            tensor<int32, [4]> var_25379_begin_0 = const()[name = tensor<string, []>("op_25379_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_25379_end_0 = const()[name = tensor<string, []>("op_25379_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_25379_end_mask_0 = const()[name = tensor<string, []>("op_25379_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25379_cast = slice_by_index(begin = var_25379_begin_0, end = var_25379_end_0, end_mask = var_25379_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25379_cast")];
            tensor<int32, [4]> var_25383_begin_0 = const()[name = tensor<string, []>("op_25383_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_25383_end_0 = const()[name = tensor<string, []>("op_25383_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_25383_end_mask_0 = const()[name = tensor<string, []>("op_25383_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25383_cast = slice_by_index(begin = var_25383_begin_0, end = var_25383_end_0, end_mask = var_25383_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25383_cast")];
            tensor<int32, [4]> var_25387_begin_0 = const()[name = tensor<string, []>("op_25387_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_25387_end_0 = const()[name = tensor<string, []>("op_25387_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_25387_end_mask_0 = const()[name = tensor<string, []>("op_25387_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25387_cast = slice_by_index(begin = var_25387_begin_0, end = var_25387_end_0, end_mask = var_25387_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25387_cast")];
            tensor<int32, [4]> var_25391_begin_0 = const()[name = tensor<string, []>("op_25391_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_25391_end_0 = const()[name = tensor<string, []>("op_25391_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_25391_end_mask_0 = const()[name = tensor<string, []>("op_25391_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25391_cast = slice_by_index(begin = var_25391_begin_0, end = var_25391_end_0, end_mask = var_25391_end_mask_0, x = q_115_cast)[name = tensor<string, []>("op_25391_cast")];
            tensor<int32, [4]> k_231_perm_0 = const()[name = tensor<string, []>("k_231_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_25398_begin_0 = const()[name = tensor<string, []>("op_25398_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_25398_end_0 = const()[name = tensor<string, []>("op_25398_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_25398_end_mask_0 = const()[name = tensor<string, []>("op_25398_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_82 = transpose(perm = k_231_perm_0, x = k_229_cast)[name = tensor<string, []>("transpose_82")];
            tensor<fp16, [2, 77, 1, 64]> var_25398_cast = slice_by_index(begin = var_25398_begin_0, end = var_25398_end_0, end_mask = var_25398_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25398_cast")];
            tensor<int32, [4]> var_25402_begin_0 = const()[name = tensor<string, []>("op_25402_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_25402_end_0 = const()[name = tensor<string, []>("op_25402_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_25402_end_mask_0 = const()[name = tensor<string, []>("op_25402_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_25402_cast = slice_by_index(begin = var_25402_begin_0, end = var_25402_end_0, end_mask = var_25402_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25402_cast")];
            tensor<int32, [4]> var_25406_begin_0 = const()[name = tensor<string, []>("op_25406_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_25406_end_0 = const()[name = tensor<string, []>("op_25406_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_25406_end_mask_0 = const()[name = tensor<string, []>("op_25406_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_25406_cast = slice_by_index(begin = var_25406_begin_0, end = var_25406_end_0, end_mask = var_25406_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25406_cast")];
            tensor<int32, [4]> var_25410_begin_0 = const()[name = tensor<string, []>("op_25410_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_25410_end_0 = const()[name = tensor<string, []>("op_25410_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_25410_end_mask_0 = const()[name = tensor<string, []>("op_25410_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_25410_cast = slice_by_index(begin = var_25410_begin_0, end = var_25410_end_0, end_mask = var_25410_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25410_cast")];
            tensor<int32, [4]> var_25414_begin_0 = const()[name = tensor<string, []>("op_25414_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_25414_end_0 = const()[name = tensor<string, []>("op_25414_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_25414_end_mask_0 = const()[name = tensor<string, []>("op_25414_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_25414_cast = slice_by_index(begin = var_25414_begin_0, end = var_25414_end_0, end_mask = var_25414_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25414_cast")];
            tensor<int32, [4]> var_25418_begin_0 = const()[name = tensor<string, []>("op_25418_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_25418_end_0 = const()[name = tensor<string, []>("op_25418_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_25418_end_mask_0 = const()[name = tensor<string, []>("op_25418_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_25418_cast = slice_by_index(begin = var_25418_begin_0, end = var_25418_end_0, end_mask = var_25418_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25418_cast")];
            tensor<int32, [4]> var_25422_begin_0 = const()[name = tensor<string, []>("op_25422_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_25422_end_0 = const()[name = tensor<string, []>("op_25422_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_25422_end_mask_0 = const()[name = tensor<string, []>("op_25422_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_25422_cast = slice_by_index(begin = var_25422_begin_0, end = var_25422_end_0, end_mask = var_25422_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25422_cast")];
            tensor<int32, [4]> var_25426_begin_0 = const()[name = tensor<string, []>("op_25426_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_25426_end_0 = const()[name = tensor<string, []>("op_25426_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_25426_end_mask_0 = const()[name = tensor<string, []>("op_25426_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_25426_cast = slice_by_index(begin = var_25426_begin_0, end = var_25426_end_0, end_mask = var_25426_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25426_cast")];
            tensor<int32, [4]> var_25430_begin_0 = const()[name = tensor<string, []>("op_25430_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_25430_end_0 = const()[name = tensor<string, []>("op_25430_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_25430_end_mask_0 = const()[name = tensor<string, []>("op_25430_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_25430_cast = slice_by_index(begin = var_25430_begin_0, end = var_25430_end_0, end_mask = var_25430_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25430_cast")];
            tensor<int32, [4]> var_25434_begin_0 = const()[name = tensor<string, []>("op_25434_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_25434_end_0 = const()[name = tensor<string, []>("op_25434_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_25434_end_mask_0 = const()[name = tensor<string, []>("op_25434_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_25434_cast = slice_by_index(begin = var_25434_begin_0, end = var_25434_end_0, end_mask = var_25434_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25434_cast")];
            tensor<int32, [4]> var_25438_begin_0 = const()[name = tensor<string, []>("op_25438_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_25438_end_0 = const()[name = tensor<string, []>("op_25438_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_25438_end_mask_0 = const()[name = tensor<string, []>("op_25438_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_25438_cast = slice_by_index(begin = var_25438_begin_0, end = var_25438_end_0, end_mask = var_25438_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25438_cast")];
            tensor<int32, [4]> var_25442_begin_0 = const()[name = tensor<string, []>("op_25442_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_25442_end_0 = const()[name = tensor<string, []>("op_25442_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_25442_end_mask_0 = const()[name = tensor<string, []>("op_25442_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_25442_cast = slice_by_index(begin = var_25442_begin_0, end = var_25442_end_0, end_mask = var_25442_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25442_cast")];
            tensor<int32, [4]> var_25446_begin_0 = const()[name = tensor<string, []>("op_25446_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_25446_end_0 = const()[name = tensor<string, []>("op_25446_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_25446_end_mask_0 = const()[name = tensor<string, []>("op_25446_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_25446_cast = slice_by_index(begin = var_25446_begin_0, end = var_25446_end_0, end_mask = var_25446_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25446_cast")];
            tensor<int32, [4]> var_25450_begin_0 = const()[name = tensor<string, []>("op_25450_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_25450_end_0 = const()[name = tensor<string, []>("op_25450_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_25450_end_mask_0 = const()[name = tensor<string, []>("op_25450_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_25450_cast = slice_by_index(begin = var_25450_begin_0, end = var_25450_end_0, end_mask = var_25450_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25450_cast")];
            tensor<int32, [4]> var_25454_begin_0 = const()[name = tensor<string, []>("op_25454_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_25454_end_0 = const()[name = tensor<string, []>("op_25454_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_25454_end_mask_0 = const()[name = tensor<string, []>("op_25454_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_25454_cast = slice_by_index(begin = var_25454_begin_0, end = var_25454_end_0, end_mask = var_25454_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25454_cast")];
            tensor<int32, [4]> var_25458_begin_0 = const()[name = tensor<string, []>("op_25458_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_25458_end_0 = const()[name = tensor<string, []>("op_25458_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_25458_end_mask_0 = const()[name = tensor<string, []>("op_25458_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_25458_cast = slice_by_index(begin = var_25458_begin_0, end = var_25458_end_0, end_mask = var_25458_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25458_cast")];
            tensor<int32, [4]> var_25462_begin_0 = const()[name = tensor<string, []>("op_25462_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_25462_end_0 = const()[name = tensor<string, []>("op_25462_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_25462_end_mask_0 = const()[name = tensor<string, []>("op_25462_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_25462_cast = slice_by_index(begin = var_25462_begin_0, end = var_25462_end_0, end_mask = var_25462_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25462_cast")];
            tensor<int32, [4]> var_25466_begin_0 = const()[name = tensor<string, []>("op_25466_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_25466_end_0 = const()[name = tensor<string, []>("op_25466_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_25466_end_mask_0 = const()[name = tensor<string, []>("op_25466_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_25466_cast = slice_by_index(begin = var_25466_begin_0, end = var_25466_end_0, end_mask = var_25466_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25466_cast")];
            tensor<int32, [4]> var_25470_begin_0 = const()[name = tensor<string, []>("op_25470_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_25470_end_0 = const()[name = tensor<string, []>("op_25470_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_25470_end_mask_0 = const()[name = tensor<string, []>("op_25470_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_25470_cast = slice_by_index(begin = var_25470_begin_0, end = var_25470_end_0, end_mask = var_25470_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25470_cast")];
            tensor<int32, [4]> var_25474_begin_0 = const()[name = tensor<string, []>("op_25474_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_25474_end_0 = const()[name = tensor<string, []>("op_25474_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_25474_end_mask_0 = const()[name = tensor<string, []>("op_25474_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_25474_cast = slice_by_index(begin = var_25474_begin_0, end = var_25474_end_0, end_mask = var_25474_end_mask_0, x = transpose_82)[name = tensor<string, []>("op_25474_cast")];
            tensor<int32, [4]> var_25476_begin_0 = const()[name = tensor<string, []>("op_25476_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_25476_end_0 = const()[name = tensor<string, []>("op_25476_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_25476_end_mask_0 = const()[name = tensor<string, []>("op_25476_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25476_cast = slice_by_index(begin = var_25476_begin_0, end = var_25476_end_0, end_mask = var_25476_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25476_cast")];
            tensor<int32, [4]> var_25480_begin_0 = const()[name = tensor<string, []>("op_25480_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_25480_end_0 = const()[name = tensor<string, []>("op_25480_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_25480_end_mask_0 = const()[name = tensor<string, []>("op_25480_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25480_cast = slice_by_index(begin = var_25480_begin_0, end = var_25480_end_0, end_mask = var_25480_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25480_cast")];
            tensor<int32, [4]> var_25484_begin_0 = const()[name = tensor<string, []>("op_25484_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_25484_end_0 = const()[name = tensor<string, []>("op_25484_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_25484_end_mask_0 = const()[name = tensor<string, []>("op_25484_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25484_cast = slice_by_index(begin = var_25484_begin_0, end = var_25484_end_0, end_mask = var_25484_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25484_cast")];
            tensor<int32, [4]> var_25488_begin_0 = const()[name = tensor<string, []>("op_25488_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_25488_end_0 = const()[name = tensor<string, []>("op_25488_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_25488_end_mask_0 = const()[name = tensor<string, []>("op_25488_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25488_cast = slice_by_index(begin = var_25488_begin_0, end = var_25488_end_0, end_mask = var_25488_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25488_cast")];
            tensor<int32, [4]> var_25492_begin_0 = const()[name = tensor<string, []>("op_25492_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_25492_end_0 = const()[name = tensor<string, []>("op_25492_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_25492_end_mask_0 = const()[name = tensor<string, []>("op_25492_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25492_cast = slice_by_index(begin = var_25492_begin_0, end = var_25492_end_0, end_mask = var_25492_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25492_cast")];
            tensor<int32, [4]> var_25496_begin_0 = const()[name = tensor<string, []>("op_25496_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_25496_end_0 = const()[name = tensor<string, []>("op_25496_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_25496_end_mask_0 = const()[name = tensor<string, []>("op_25496_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25496_cast = slice_by_index(begin = var_25496_begin_0, end = var_25496_end_0, end_mask = var_25496_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25496_cast")];
            tensor<int32, [4]> var_25500_begin_0 = const()[name = tensor<string, []>("op_25500_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_25500_end_0 = const()[name = tensor<string, []>("op_25500_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_25500_end_mask_0 = const()[name = tensor<string, []>("op_25500_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25500_cast = slice_by_index(begin = var_25500_begin_0, end = var_25500_end_0, end_mask = var_25500_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25500_cast")];
            tensor<int32, [4]> var_25504_begin_0 = const()[name = tensor<string, []>("op_25504_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_25504_end_0 = const()[name = tensor<string, []>("op_25504_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_25504_end_mask_0 = const()[name = tensor<string, []>("op_25504_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25504_cast = slice_by_index(begin = var_25504_begin_0, end = var_25504_end_0, end_mask = var_25504_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25504_cast")];
            tensor<int32, [4]> var_25508_begin_0 = const()[name = tensor<string, []>("op_25508_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_25508_end_0 = const()[name = tensor<string, []>("op_25508_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_25508_end_mask_0 = const()[name = tensor<string, []>("op_25508_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25508_cast = slice_by_index(begin = var_25508_begin_0, end = var_25508_end_0, end_mask = var_25508_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25508_cast")];
            tensor<int32, [4]> var_25512_begin_0 = const()[name = tensor<string, []>("op_25512_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_25512_end_0 = const()[name = tensor<string, []>("op_25512_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_25512_end_mask_0 = const()[name = tensor<string, []>("op_25512_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25512_cast = slice_by_index(begin = var_25512_begin_0, end = var_25512_end_0, end_mask = var_25512_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25512_cast")];
            tensor<int32, [4]> var_25516_begin_0 = const()[name = tensor<string, []>("op_25516_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_25516_end_0 = const()[name = tensor<string, []>("op_25516_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_25516_end_mask_0 = const()[name = tensor<string, []>("op_25516_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25516_cast = slice_by_index(begin = var_25516_begin_0, end = var_25516_end_0, end_mask = var_25516_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25516_cast")];
            tensor<int32, [4]> var_25520_begin_0 = const()[name = tensor<string, []>("op_25520_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_25520_end_0 = const()[name = tensor<string, []>("op_25520_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_25520_end_mask_0 = const()[name = tensor<string, []>("op_25520_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25520_cast = slice_by_index(begin = var_25520_begin_0, end = var_25520_end_0, end_mask = var_25520_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25520_cast")];
            tensor<int32, [4]> var_25524_begin_0 = const()[name = tensor<string, []>("op_25524_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_25524_end_0 = const()[name = tensor<string, []>("op_25524_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_25524_end_mask_0 = const()[name = tensor<string, []>("op_25524_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25524_cast = slice_by_index(begin = var_25524_begin_0, end = var_25524_end_0, end_mask = var_25524_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25524_cast")];
            tensor<int32, [4]> var_25528_begin_0 = const()[name = tensor<string, []>("op_25528_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_25528_end_0 = const()[name = tensor<string, []>("op_25528_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_25528_end_mask_0 = const()[name = tensor<string, []>("op_25528_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25528_cast = slice_by_index(begin = var_25528_begin_0, end = var_25528_end_0, end_mask = var_25528_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25528_cast")];
            tensor<int32, [4]> var_25532_begin_0 = const()[name = tensor<string, []>("op_25532_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_25532_end_0 = const()[name = tensor<string, []>("op_25532_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_25532_end_mask_0 = const()[name = tensor<string, []>("op_25532_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25532_cast = slice_by_index(begin = var_25532_begin_0, end = var_25532_end_0, end_mask = var_25532_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25532_cast")];
            tensor<int32, [4]> var_25536_begin_0 = const()[name = tensor<string, []>("op_25536_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_25536_end_0 = const()[name = tensor<string, []>("op_25536_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_25536_end_mask_0 = const()[name = tensor<string, []>("op_25536_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25536_cast = slice_by_index(begin = var_25536_begin_0, end = var_25536_end_0, end_mask = var_25536_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25536_cast")];
            tensor<int32, [4]> var_25540_begin_0 = const()[name = tensor<string, []>("op_25540_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_25540_end_0 = const()[name = tensor<string, []>("op_25540_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_25540_end_mask_0 = const()[name = tensor<string, []>("op_25540_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25540_cast = slice_by_index(begin = var_25540_begin_0, end = var_25540_end_0, end_mask = var_25540_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25540_cast")];
            tensor<int32, [4]> var_25544_begin_0 = const()[name = tensor<string, []>("op_25544_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_25544_end_0 = const()[name = tensor<string, []>("op_25544_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_25544_end_mask_0 = const()[name = tensor<string, []>("op_25544_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25544_cast = slice_by_index(begin = var_25544_begin_0, end = var_25544_end_0, end_mask = var_25544_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25544_cast")];
            tensor<int32, [4]> var_25548_begin_0 = const()[name = tensor<string, []>("op_25548_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_25548_end_0 = const()[name = tensor<string, []>("op_25548_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_25548_end_mask_0 = const()[name = tensor<string, []>("op_25548_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25548_cast = slice_by_index(begin = var_25548_begin_0, end = var_25548_end_0, end_mask = var_25548_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25548_cast")];
            tensor<int32, [4]> var_25552_begin_0 = const()[name = tensor<string, []>("op_25552_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_25552_end_0 = const()[name = tensor<string, []>("op_25552_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_25552_end_mask_0 = const()[name = tensor<string, []>("op_25552_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_25552_cast = slice_by_index(begin = var_25552_begin_0, end = var_25552_end_0, end_mask = var_25552_end_mask_0, x = v_115_cast)[name = tensor<string, []>("op_25552_cast")];
            tensor<string, []> var_25556_equation_0 = const()[name = tensor<string, []>("op_25556_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25556_cast = einsum(equation = var_25556_equation_0, values = (var_25398_cast, var_25315_cast))[name = tensor<string, []>("op_25556_cast")];
            tensor<fp16, []> var_25557_to_fp16 = const()[name = tensor<string, []>("op_25557_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2121_cast = mul(x = var_25556_cast, y = var_25557_to_fp16)[name = tensor<string, []>("aw_2121_cast")];
            tensor<string, []> var_25560_equation_0 = const()[name = tensor<string, []>("op_25560_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25560_cast = einsum(equation = var_25560_equation_0, values = (var_25402_cast, var_25319_cast))[name = tensor<string, []>("op_25560_cast")];
            tensor<fp16, []> var_25561_to_fp16 = const()[name = tensor<string, []>("op_25561_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2123_cast = mul(x = var_25560_cast, y = var_25561_to_fp16)[name = tensor<string, []>("aw_2123_cast")];
            tensor<string, []> var_25564_equation_0 = const()[name = tensor<string, []>("op_25564_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25564_cast = einsum(equation = var_25564_equation_0, values = (var_25406_cast, var_25323_cast))[name = tensor<string, []>("op_25564_cast")];
            tensor<fp16, []> var_25565_to_fp16 = const()[name = tensor<string, []>("op_25565_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2125_cast = mul(x = var_25564_cast, y = var_25565_to_fp16)[name = tensor<string, []>("aw_2125_cast")];
            tensor<string, []> var_25568_equation_0 = const()[name = tensor<string, []>("op_25568_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25568_cast = einsum(equation = var_25568_equation_0, values = (var_25410_cast, var_25327_cast))[name = tensor<string, []>("op_25568_cast")];
            tensor<fp16, []> var_25569_to_fp16 = const()[name = tensor<string, []>("op_25569_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2127_cast = mul(x = var_25568_cast, y = var_25569_to_fp16)[name = tensor<string, []>("aw_2127_cast")];
            tensor<string, []> var_25572_equation_0 = const()[name = tensor<string, []>("op_25572_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25572_cast = einsum(equation = var_25572_equation_0, values = (var_25414_cast, var_25331_cast))[name = tensor<string, []>("op_25572_cast")];
            tensor<fp16, []> var_25573_to_fp16 = const()[name = tensor<string, []>("op_25573_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2129_cast = mul(x = var_25572_cast, y = var_25573_to_fp16)[name = tensor<string, []>("aw_2129_cast")];
            tensor<string, []> var_25576_equation_0 = const()[name = tensor<string, []>("op_25576_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25576_cast = einsum(equation = var_25576_equation_0, values = (var_25418_cast, var_25335_cast))[name = tensor<string, []>("op_25576_cast")];
            tensor<fp16, []> var_25577_to_fp16 = const()[name = tensor<string, []>("op_25577_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2131_cast = mul(x = var_25576_cast, y = var_25577_to_fp16)[name = tensor<string, []>("aw_2131_cast")];
            tensor<string, []> var_25580_equation_0 = const()[name = tensor<string, []>("op_25580_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25580_cast = einsum(equation = var_25580_equation_0, values = (var_25422_cast, var_25339_cast))[name = tensor<string, []>("op_25580_cast")];
            tensor<fp16, []> var_25581_to_fp16 = const()[name = tensor<string, []>("op_25581_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2133_cast = mul(x = var_25580_cast, y = var_25581_to_fp16)[name = tensor<string, []>("aw_2133_cast")];
            tensor<string, []> var_25584_equation_0 = const()[name = tensor<string, []>("op_25584_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25584_cast = einsum(equation = var_25584_equation_0, values = (var_25426_cast, var_25343_cast))[name = tensor<string, []>("op_25584_cast")];
            tensor<fp16, []> var_25585_to_fp16 = const()[name = tensor<string, []>("op_25585_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2135_cast = mul(x = var_25584_cast, y = var_25585_to_fp16)[name = tensor<string, []>("aw_2135_cast")];
            tensor<string, []> var_25588_equation_0 = const()[name = tensor<string, []>("op_25588_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25588_cast = einsum(equation = var_25588_equation_0, values = (var_25430_cast, var_25347_cast))[name = tensor<string, []>("op_25588_cast")];
            tensor<fp16, []> var_25589_to_fp16 = const()[name = tensor<string, []>("op_25589_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2137_cast = mul(x = var_25588_cast, y = var_25589_to_fp16)[name = tensor<string, []>("aw_2137_cast")];
            tensor<string, []> var_25592_equation_0 = const()[name = tensor<string, []>("op_25592_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25592_cast = einsum(equation = var_25592_equation_0, values = (var_25434_cast, var_25351_cast))[name = tensor<string, []>("op_25592_cast")];
            tensor<fp16, []> var_25593_to_fp16 = const()[name = tensor<string, []>("op_25593_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2139_cast = mul(x = var_25592_cast, y = var_25593_to_fp16)[name = tensor<string, []>("aw_2139_cast")];
            tensor<string, []> var_25596_equation_0 = const()[name = tensor<string, []>("op_25596_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25596_cast = einsum(equation = var_25596_equation_0, values = (var_25438_cast, var_25355_cast))[name = tensor<string, []>("op_25596_cast")];
            tensor<fp16, []> var_25597_to_fp16 = const()[name = tensor<string, []>("op_25597_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2141_cast = mul(x = var_25596_cast, y = var_25597_to_fp16)[name = tensor<string, []>("aw_2141_cast")];
            tensor<string, []> var_25600_equation_0 = const()[name = tensor<string, []>("op_25600_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25600_cast = einsum(equation = var_25600_equation_0, values = (var_25442_cast, var_25359_cast))[name = tensor<string, []>("op_25600_cast")];
            tensor<fp16, []> var_25601_to_fp16 = const()[name = tensor<string, []>("op_25601_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2143_cast = mul(x = var_25600_cast, y = var_25601_to_fp16)[name = tensor<string, []>("aw_2143_cast")];
            tensor<string, []> var_25604_equation_0 = const()[name = tensor<string, []>("op_25604_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25604_cast = einsum(equation = var_25604_equation_0, values = (var_25446_cast, var_25363_cast))[name = tensor<string, []>("op_25604_cast")];
            tensor<fp16, []> var_25605_to_fp16 = const()[name = tensor<string, []>("op_25605_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2145_cast = mul(x = var_25604_cast, y = var_25605_to_fp16)[name = tensor<string, []>("aw_2145_cast")];
            tensor<string, []> var_25608_equation_0 = const()[name = tensor<string, []>("op_25608_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25608_cast = einsum(equation = var_25608_equation_0, values = (var_25450_cast, var_25367_cast))[name = tensor<string, []>("op_25608_cast")];
            tensor<fp16, []> var_25609_to_fp16 = const()[name = tensor<string, []>("op_25609_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2147_cast = mul(x = var_25608_cast, y = var_25609_to_fp16)[name = tensor<string, []>("aw_2147_cast")];
            tensor<string, []> var_25612_equation_0 = const()[name = tensor<string, []>("op_25612_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25612_cast = einsum(equation = var_25612_equation_0, values = (var_25454_cast, var_25371_cast))[name = tensor<string, []>("op_25612_cast")];
            tensor<fp16, []> var_25613_to_fp16 = const()[name = tensor<string, []>("op_25613_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2149_cast = mul(x = var_25612_cast, y = var_25613_to_fp16)[name = tensor<string, []>("aw_2149_cast")];
            tensor<string, []> var_25616_equation_0 = const()[name = tensor<string, []>("op_25616_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25616_cast = einsum(equation = var_25616_equation_0, values = (var_25458_cast, var_25375_cast))[name = tensor<string, []>("op_25616_cast")];
            tensor<fp16, []> var_25617_to_fp16 = const()[name = tensor<string, []>("op_25617_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2151_cast = mul(x = var_25616_cast, y = var_25617_to_fp16)[name = tensor<string, []>("aw_2151_cast")];
            tensor<string, []> var_25620_equation_0 = const()[name = tensor<string, []>("op_25620_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25620_cast = einsum(equation = var_25620_equation_0, values = (var_25462_cast, var_25379_cast))[name = tensor<string, []>("op_25620_cast")];
            tensor<fp16, []> var_25621_to_fp16 = const()[name = tensor<string, []>("op_25621_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2153_cast = mul(x = var_25620_cast, y = var_25621_to_fp16)[name = tensor<string, []>("aw_2153_cast")];
            tensor<string, []> var_25624_equation_0 = const()[name = tensor<string, []>("op_25624_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25624_cast = einsum(equation = var_25624_equation_0, values = (var_25466_cast, var_25383_cast))[name = tensor<string, []>("op_25624_cast")];
            tensor<fp16, []> var_25625_to_fp16 = const()[name = tensor<string, []>("op_25625_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2155_cast = mul(x = var_25624_cast, y = var_25625_to_fp16)[name = tensor<string, []>("aw_2155_cast")];
            tensor<string, []> var_25628_equation_0 = const()[name = tensor<string, []>("op_25628_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25628_cast = einsum(equation = var_25628_equation_0, values = (var_25470_cast, var_25387_cast))[name = tensor<string, []>("op_25628_cast")];
            tensor<fp16, []> var_25629_to_fp16 = const()[name = tensor<string, []>("op_25629_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2157_cast = mul(x = var_25628_cast, y = var_25629_to_fp16)[name = tensor<string, []>("aw_2157_cast")];
            tensor<string, []> var_25632_equation_0 = const()[name = tensor<string, []>("op_25632_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_25632_cast = einsum(equation = var_25632_equation_0, values = (var_25474_cast, var_25391_cast))[name = tensor<string, []>("op_25632_cast")];
            tensor<fp16, []> var_25633_to_fp16 = const()[name = tensor<string, []>("op_25633_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2159_cast = mul(x = var_25632_cast, y = var_25633_to_fp16)[name = tensor<string, []>("aw_2159_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25635_cast = softmax(axis = var_21105, x = aw_2121_cast)[name = tensor<string, []>("op_25635_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25636_cast = softmax(axis = var_21105, x = aw_2123_cast)[name = tensor<string, []>("op_25636_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25637_cast = softmax(axis = var_21105, x = aw_2125_cast)[name = tensor<string, []>("op_25637_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25638_cast = softmax(axis = var_21105, x = aw_2127_cast)[name = tensor<string, []>("op_25638_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25639_cast = softmax(axis = var_21105, x = aw_2129_cast)[name = tensor<string, []>("op_25639_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25640_cast = softmax(axis = var_21105, x = aw_2131_cast)[name = tensor<string, []>("op_25640_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25641_cast = softmax(axis = var_21105, x = aw_2133_cast)[name = tensor<string, []>("op_25641_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25642_cast = softmax(axis = var_21105, x = aw_2135_cast)[name = tensor<string, []>("op_25642_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25643_cast = softmax(axis = var_21105, x = aw_2137_cast)[name = tensor<string, []>("op_25643_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25644_cast = softmax(axis = var_21105, x = aw_2139_cast)[name = tensor<string, []>("op_25644_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25645_cast = softmax(axis = var_21105, x = aw_2141_cast)[name = tensor<string, []>("op_25645_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25646_cast = softmax(axis = var_21105, x = aw_2143_cast)[name = tensor<string, []>("op_25646_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25647_cast = softmax(axis = var_21105, x = aw_2145_cast)[name = tensor<string, []>("op_25647_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25648_cast = softmax(axis = var_21105, x = aw_2147_cast)[name = tensor<string, []>("op_25648_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25649_cast = softmax(axis = var_21105, x = aw_2149_cast)[name = tensor<string, []>("op_25649_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25650_cast = softmax(axis = var_21105, x = aw_2151_cast)[name = tensor<string, []>("op_25650_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25651_cast = softmax(axis = var_21105, x = aw_2153_cast)[name = tensor<string, []>("op_25651_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25652_cast = softmax(axis = var_21105, x = aw_2155_cast)[name = tensor<string, []>("op_25652_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25653_cast = softmax(axis = var_21105, x = aw_2157_cast)[name = tensor<string, []>("op_25653_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_25654_cast = softmax(axis = var_21105, x = aw_2159_cast)[name = tensor<string, []>("op_25654_cast")];
            tensor<string, []> var_25656_equation_0 = const()[name = tensor<string, []>("op_25656_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25656_cast = einsum(equation = var_25656_equation_0, values = (var_25476_cast, var_25635_cast))[name = tensor<string, []>("op_25656_cast")];
            tensor<string, []> var_25658_equation_0 = const()[name = tensor<string, []>("op_25658_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25658_cast = einsum(equation = var_25658_equation_0, values = (var_25480_cast, var_25636_cast))[name = tensor<string, []>("op_25658_cast")];
            tensor<string, []> var_25660_equation_0 = const()[name = tensor<string, []>("op_25660_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25660_cast = einsum(equation = var_25660_equation_0, values = (var_25484_cast, var_25637_cast))[name = tensor<string, []>("op_25660_cast")];
            tensor<string, []> var_25662_equation_0 = const()[name = tensor<string, []>("op_25662_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25662_cast = einsum(equation = var_25662_equation_0, values = (var_25488_cast, var_25638_cast))[name = tensor<string, []>("op_25662_cast")];
            tensor<string, []> var_25664_equation_0 = const()[name = tensor<string, []>("op_25664_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25664_cast = einsum(equation = var_25664_equation_0, values = (var_25492_cast, var_25639_cast))[name = tensor<string, []>("op_25664_cast")];
            tensor<string, []> var_25666_equation_0 = const()[name = tensor<string, []>("op_25666_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25666_cast = einsum(equation = var_25666_equation_0, values = (var_25496_cast, var_25640_cast))[name = tensor<string, []>("op_25666_cast")];
            tensor<string, []> var_25668_equation_0 = const()[name = tensor<string, []>("op_25668_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25668_cast = einsum(equation = var_25668_equation_0, values = (var_25500_cast, var_25641_cast))[name = tensor<string, []>("op_25668_cast")];
            tensor<string, []> var_25670_equation_0 = const()[name = tensor<string, []>("op_25670_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25670_cast = einsum(equation = var_25670_equation_0, values = (var_25504_cast, var_25642_cast))[name = tensor<string, []>("op_25670_cast")];
            tensor<string, []> var_25672_equation_0 = const()[name = tensor<string, []>("op_25672_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25672_cast = einsum(equation = var_25672_equation_0, values = (var_25508_cast, var_25643_cast))[name = tensor<string, []>("op_25672_cast")];
            tensor<string, []> var_25674_equation_0 = const()[name = tensor<string, []>("op_25674_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25674_cast = einsum(equation = var_25674_equation_0, values = (var_25512_cast, var_25644_cast))[name = tensor<string, []>("op_25674_cast")];
            tensor<string, []> var_25676_equation_0 = const()[name = tensor<string, []>("op_25676_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25676_cast = einsum(equation = var_25676_equation_0, values = (var_25516_cast, var_25645_cast))[name = tensor<string, []>("op_25676_cast")];
            tensor<string, []> var_25678_equation_0 = const()[name = tensor<string, []>("op_25678_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25678_cast = einsum(equation = var_25678_equation_0, values = (var_25520_cast, var_25646_cast))[name = tensor<string, []>("op_25678_cast")];
            tensor<string, []> var_25680_equation_0 = const()[name = tensor<string, []>("op_25680_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25680_cast = einsum(equation = var_25680_equation_0, values = (var_25524_cast, var_25647_cast))[name = tensor<string, []>("op_25680_cast")];
            tensor<string, []> var_25682_equation_0 = const()[name = tensor<string, []>("op_25682_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25682_cast = einsum(equation = var_25682_equation_0, values = (var_25528_cast, var_25648_cast))[name = tensor<string, []>("op_25682_cast")];
            tensor<string, []> var_25684_equation_0 = const()[name = tensor<string, []>("op_25684_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25684_cast = einsum(equation = var_25684_equation_0, values = (var_25532_cast, var_25649_cast))[name = tensor<string, []>("op_25684_cast")];
            tensor<string, []> var_25686_equation_0 = const()[name = tensor<string, []>("op_25686_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25686_cast = einsum(equation = var_25686_equation_0, values = (var_25536_cast, var_25650_cast))[name = tensor<string, []>("op_25686_cast")];
            tensor<string, []> var_25688_equation_0 = const()[name = tensor<string, []>("op_25688_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25688_cast = einsum(equation = var_25688_equation_0, values = (var_25540_cast, var_25651_cast))[name = tensor<string, []>("op_25688_cast")];
            tensor<string, []> var_25690_equation_0 = const()[name = tensor<string, []>("op_25690_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25690_cast = einsum(equation = var_25690_equation_0, values = (var_25544_cast, var_25652_cast))[name = tensor<string, []>("op_25690_cast")];
            tensor<string, []> var_25692_equation_0 = const()[name = tensor<string, []>("op_25692_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25692_cast = einsum(equation = var_25692_equation_0, values = (var_25548_cast, var_25653_cast))[name = tensor<string, []>("op_25692_cast")];
            tensor<string, []> var_25694_equation_0 = const()[name = tensor<string, []>("op_25694_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_25694_cast = einsum(equation = var_25694_equation_0, values = (var_25552_cast, var_25654_cast))[name = tensor<string, []>("op_25694_cast")];
            tensor<bool, []> input_363_interleave_0 = const()[name = tensor<string, []>("input_363_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_363_cast = concat(axis = var_21105, interleave = input_363_interleave_0, values = (var_25656_cast, var_25658_cast, var_25660_cast, var_25662_cast, var_25664_cast, var_25666_cast, var_25668_cast, var_25670_cast, var_25672_cast, var_25674_cast, var_25676_cast, var_25678_cast, var_25680_cast, var_25682_cast, var_25684_cast, var_25686_cast, var_25688_cast, var_25690_cast, var_25692_cast, var_25694_cast))[name = tensor<string, []>("input_363_cast")];
            tensor<int32, [2]> var_25700 = const()[name = tensor<string, []>("op_25700"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_25702 = const()[name = tensor<string, []>("op_25702"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_25704_pad_type_0 = const()[name = tensor<string, []>("op_25704_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_25704_pad_0 = const()[name = tensor<string, []>("op_25704_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2031044224)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2034321088)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_25704_cast = conv(bias = mid_block_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16, dilations = var_25702, groups = var_21105, pad = var_25704_pad_0, pad_type = var_25704_pad_type_0, strides = var_25700, weight = mid_block_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16, x = input_363_cast)[name = tensor<string, []>("op_25704_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_173_cast = add(x = var_25704_cast, y = inputs_171_cast)[name = tensor<string, []>("inputs_173_cast")];
            tensor<int32, [1]> var_25708 = const()[name = tensor<string, []>("op_25708"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_173_cast = reduce_mean(axes = var_25708, keep_dims = var_21100, x = inputs_173_cast)[name = tensor<string, []>("channels_mean_173_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_173_cast = sub(x = inputs_173_cast, y = channels_mean_173_cast)[name = tensor<string, []>("zero_mean_173_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_173_cast = mul(x = zero_mean_173_cast, y = zero_mean_173_cast)[name = tensor<string, []>("zero_mean_sq_173_cast")];
            tensor<int32, [1]> var_25712 = const()[name = tensor<string, []>("op_25712"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_25713_cast = reduce_mean(axes = var_25712, keep_dims = var_21100, x = zero_mean_sq_173_cast)[name = tensor<string, []>("op_25713_cast")];
            tensor<fp16, []> var_25714_to_fp16 = const()[name = tensor<string, []>("op_25714_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_25715_cast = add(x = var_25713_cast, y = var_25714_to_fp16)[name = tensor<string, []>("op_25715_cast")];
            tensor<fp16, []> denom_173_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_173_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_173_cast = rsqrt(epsilon = denom_173_epsilon_0_to_fp16, x = var_25715_cast)[name = tensor<string, []>("denom_173_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_173_cast = mul(x = zero_mean_173_cast, y = denom_173_cast)[name = tensor<string, []>("out_173_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_25719_to_fp16 = const()[name = tensor<string, []>("op_25719_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2034323712)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_25720_cast = add(x = out_173_cast, y = var_25719_to_fp16)[name = tensor<string, []>("op_25720_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_25722_to_fp16 = const()[name = tensor<string, []>("op_25722_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2034326336)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_365_cast = mul(x = var_25720_cast, y = var_25722_to_fp16)[name = tensor<string, []>("input_365_cast")];
            tensor<int32, [2]> var_25730 = const()[name = tensor<string, []>("op_25730"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_25732 = const()[name = tensor<string, []>("op_25732"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_25734_pad_type_0 = const()[name = tensor<string, []>("op_25734_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_25734_pad_0 = const()[name = tensor<string, []>("op_25734_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2034328960)))];
            tensor<fp16, [10240]> mid_block_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2060543424)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_25734_cast = conv(bias = mid_block_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16, dilations = var_25732, groups = var_21105, pad = var_25734_pad_0, pad_type = var_25734_pad_type_0, strides = var_25730, weight = mid_block_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16, x = input_365_cast)[name = tensor<string, []>("op_25734_cast")];
            tensor<int32, [2]> var_25735_split_sizes_0 = const()[name = tensor<string, []>("op_25735_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_25735_axis_0 = const()[name = tensor<string, []>("op_25735_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_25735_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_25735_cast_1 = split(axis = var_25735_axis_0, split_sizes = var_25735_split_sizes_0, x = var_25734_cast)[name = tensor<string, []>("op_25735_cast")];
            tensor<string, []> var_25737_mode_0 = const()[name = tensor<string, []>("op_25737_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_25737_cast = gelu(mode = var_25737_mode_0, x = var_25735_cast_1)[name = tensor<string, []>("op_25737_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_367_cast = mul(x = var_25735_cast_0, y = var_25737_cast)[name = tensor<string, []>("input_367_cast")];
            tensor<int32, [2]> var_25741 = const()[name = tensor<string, []>("op_25741"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_25743 = const()[name = tensor<string, []>("op_25743"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_25745_pad_type_0 = const()[name = tensor<string, []>("op_25745_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_25745_pad_0 = const()[name = tensor<string, []>("op_25745_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> mid_block_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2060563968)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2073671232)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_25745_cast = conv(bias = mid_block_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16, dilations = var_25743, groups = var_21105, pad = var_25745_pad_0, pad_type = var_25745_pad_type_0, strides = var_25741, weight = mid_block_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16, x = input_367_cast)[name = tensor<string, []>("op_25745_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_175_cast = add(x = var_25745_cast, y = inputs_173_cast)[name = tensor<string, []>("inputs_175_cast")];
            tensor<int32, [1]> var_25755 = const()[name = tensor<string, []>("op_25755"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_175_cast = reduce_mean(axes = var_25755, keep_dims = var_21100, x = inputs_175_cast)[name = tensor<string, []>("channels_mean_175_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_175_cast = sub(x = inputs_175_cast, y = channels_mean_175_cast)[name = tensor<string, []>("zero_mean_175_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_175_cast = mul(x = zero_mean_175_cast, y = zero_mean_175_cast)[name = tensor<string, []>("zero_mean_sq_175_cast")];
            tensor<int32, [1]> var_25759 = const()[name = tensor<string, []>("op_25759"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_25760_cast = reduce_mean(axes = var_25759, keep_dims = var_21100, x = zero_mean_sq_175_cast)[name = tensor<string, []>("op_25760_cast")];
            tensor<fp16, []> var_25761_to_fp16 = const()[name = tensor<string, []>("op_25761_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_25762_cast = add(x = var_25760_cast, y = var_25761_to_fp16)[name = tensor<string, []>("op_25762_cast")];
            tensor<fp16, []> denom_175_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_175_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_175_cast = rsqrt(epsilon = denom_175_epsilon_0_to_fp16, x = var_25762_cast)[name = tensor<string, []>("denom_175_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_175_cast = mul(x = zero_mean_175_cast, y = denom_175_cast)[name = tensor<string, []>("out_175_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_25766_to_fp16 = const()[name = tensor<string, []>("op_25766_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2073673856)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_25767_cast = add(x = out_175_cast, y = var_25766_to_fp16)[name = tensor<string, []>("op_25767_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_25769_to_fp16 = const()[name = tensor<string, []>("op_25769_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2073676480)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_239_cast = mul(x = var_25767_cast, y = var_25769_to_fp16)[name = tensor<string, []>("hidden_states_239_cast")];
            tensor<int32, [2]> var_25776 = const()[name = tensor<string, []>("op_25776"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_25778 = const()[name = tensor<string, []>("op_25778"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_117_pad_type_0 = const()[name = tensor<string, []>("q_117_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_117_pad_0 = const()[name = tensor<string, []>("q_117_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2073679104)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_117_cast = conv(dilations = var_25778, groups = var_21105, pad = q_117_pad_0, pad_type = q_117_pad_type_0, strides = var_25776, weight = mid_block_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16, x = hidden_states_239_cast)[name = tensor<string, []>("q_117_cast")];
            tensor<int32, [2]> var_25782 = const()[name = tensor<string, []>("op_25782"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_25784 = const()[name = tensor<string, []>("op_25784"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_233_pad_type_0 = const()[name = tensor<string, []>("k_233_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_233_pad_0 = const()[name = tensor<string, []>("k_233_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2076955968)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_233_cast = conv(dilations = var_25784, groups = var_21105, pad = k_233_pad_0, pad_type = k_233_pad_type_0, strides = var_25782, weight = mid_block_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16, x = hidden_states_239_cast)[name = tensor<string, []>("k_233_cast")];
            tensor<int32, [2]> var_25788 = const()[name = tensor<string, []>("op_25788"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_25790 = const()[name = tensor<string, []>("op_25790"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_117_pad_type_0 = const()[name = tensor<string, []>("v_117_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_117_pad_0 = const()[name = tensor<string, []>("v_117_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2080232832)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_117_cast = conv(dilations = var_25790, groups = var_21105, pad = v_117_pad_0, pad_type = v_117_pad_type_0, strides = var_25788, weight = mid_block_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16, x = hidden_states_239_cast)[name = tensor<string, []>("v_117_cast")];
            tensor<int32, [4]> var_25794_begin_0 = const()[name = tensor<string, []>("op_25794_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_25794_end_0 = const()[name = tensor<string, []>("op_25794_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_25794_end_mask_0 = const()[name = tensor<string, []>("op_25794_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25794_cast = slice_by_index(begin = var_25794_begin_0, end = var_25794_end_0, end_mask = var_25794_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25794_cast")];
            tensor<int32, [4]> var_25798_begin_0 = const()[name = tensor<string, []>("op_25798_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_25798_end_0 = const()[name = tensor<string, []>("op_25798_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_25798_end_mask_0 = const()[name = tensor<string, []>("op_25798_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25798_cast = slice_by_index(begin = var_25798_begin_0, end = var_25798_end_0, end_mask = var_25798_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25798_cast")];
            tensor<int32, [4]> var_25802_begin_0 = const()[name = tensor<string, []>("op_25802_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_25802_end_0 = const()[name = tensor<string, []>("op_25802_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_25802_end_mask_0 = const()[name = tensor<string, []>("op_25802_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25802_cast = slice_by_index(begin = var_25802_begin_0, end = var_25802_end_0, end_mask = var_25802_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25802_cast")];
            tensor<int32, [4]> var_25806_begin_0 = const()[name = tensor<string, []>("op_25806_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_25806_end_0 = const()[name = tensor<string, []>("op_25806_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_25806_end_mask_0 = const()[name = tensor<string, []>("op_25806_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25806_cast = slice_by_index(begin = var_25806_begin_0, end = var_25806_end_0, end_mask = var_25806_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25806_cast")];
            tensor<int32, [4]> var_25810_begin_0 = const()[name = tensor<string, []>("op_25810_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_25810_end_0 = const()[name = tensor<string, []>("op_25810_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_25810_end_mask_0 = const()[name = tensor<string, []>("op_25810_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25810_cast = slice_by_index(begin = var_25810_begin_0, end = var_25810_end_0, end_mask = var_25810_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25810_cast")];
            tensor<int32, [4]> var_25814_begin_0 = const()[name = tensor<string, []>("op_25814_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_25814_end_0 = const()[name = tensor<string, []>("op_25814_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_25814_end_mask_0 = const()[name = tensor<string, []>("op_25814_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25814_cast = slice_by_index(begin = var_25814_begin_0, end = var_25814_end_0, end_mask = var_25814_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25814_cast")];
            tensor<int32, [4]> var_25818_begin_0 = const()[name = tensor<string, []>("op_25818_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_25818_end_0 = const()[name = tensor<string, []>("op_25818_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_25818_end_mask_0 = const()[name = tensor<string, []>("op_25818_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25818_cast = slice_by_index(begin = var_25818_begin_0, end = var_25818_end_0, end_mask = var_25818_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25818_cast")];
            tensor<int32, [4]> var_25822_begin_0 = const()[name = tensor<string, []>("op_25822_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_25822_end_0 = const()[name = tensor<string, []>("op_25822_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_25822_end_mask_0 = const()[name = tensor<string, []>("op_25822_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25822_cast = slice_by_index(begin = var_25822_begin_0, end = var_25822_end_0, end_mask = var_25822_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25822_cast")];
            tensor<int32, [4]> var_25826_begin_0 = const()[name = tensor<string, []>("op_25826_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_25826_end_0 = const()[name = tensor<string, []>("op_25826_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_25826_end_mask_0 = const()[name = tensor<string, []>("op_25826_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25826_cast = slice_by_index(begin = var_25826_begin_0, end = var_25826_end_0, end_mask = var_25826_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25826_cast")];
            tensor<int32, [4]> var_25830_begin_0 = const()[name = tensor<string, []>("op_25830_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_25830_end_0 = const()[name = tensor<string, []>("op_25830_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_25830_end_mask_0 = const()[name = tensor<string, []>("op_25830_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25830_cast = slice_by_index(begin = var_25830_begin_0, end = var_25830_end_0, end_mask = var_25830_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25830_cast")];
            tensor<int32, [4]> var_25834_begin_0 = const()[name = tensor<string, []>("op_25834_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_25834_end_0 = const()[name = tensor<string, []>("op_25834_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_25834_end_mask_0 = const()[name = tensor<string, []>("op_25834_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25834_cast = slice_by_index(begin = var_25834_begin_0, end = var_25834_end_0, end_mask = var_25834_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25834_cast")];
            tensor<int32, [4]> var_25838_begin_0 = const()[name = tensor<string, []>("op_25838_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_25838_end_0 = const()[name = tensor<string, []>("op_25838_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_25838_end_mask_0 = const()[name = tensor<string, []>("op_25838_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25838_cast = slice_by_index(begin = var_25838_begin_0, end = var_25838_end_0, end_mask = var_25838_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25838_cast")];
            tensor<int32, [4]> var_25842_begin_0 = const()[name = tensor<string, []>("op_25842_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_25842_end_0 = const()[name = tensor<string, []>("op_25842_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_25842_end_mask_0 = const()[name = tensor<string, []>("op_25842_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25842_cast = slice_by_index(begin = var_25842_begin_0, end = var_25842_end_0, end_mask = var_25842_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25842_cast")];
            tensor<int32, [4]> var_25846_begin_0 = const()[name = tensor<string, []>("op_25846_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_25846_end_0 = const()[name = tensor<string, []>("op_25846_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_25846_end_mask_0 = const()[name = tensor<string, []>("op_25846_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25846_cast = slice_by_index(begin = var_25846_begin_0, end = var_25846_end_0, end_mask = var_25846_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25846_cast")];
            tensor<int32, [4]> var_25850_begin_0 = const()[name = tensor<string, []>("op_25850_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_25850_end_0 = const()[name = tensor<string, []>("op_25850_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_25850_end_mask_0 = const()[name = tensor<string, []>("op_25850_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25850_cast = slice_by_index(begin = var_25850_begin_0, end = var_25850_end_0, end_mask = var_25850_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25850_cast")];
            tensor<int32, [4]> var_25854_begin_0 = const()[name = tensor<string, []>("op_25854_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_25854_end_0 = const()[name = tensor<string, []>("op_25854_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_25854_end_mask_0 = const()[name = tensor<string, []>("op_25854_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25854_cast = slice_by_index(begin = var_25854_begin_0, end = var_25854_end_0, end_mask = var_25854_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25854_cast")];
            tensor<int32, [4]> var_25858_begin_0 = const()[name = tensor<string, []>("op_25858_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_25858_end_0 = const()[name = tensor<string, []>("op_25858_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_25858_end_mask_0 = const()[name = tensor<string, []>("op_25858_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25858_cast = slice_by_index(begin = var_25858_begin_0, end = var_25858_end_0, end_mask = var_25858_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25858_cast")];
            tensor<int32, [4]> var_25862_begin_0 = const()[name = tensor<string, []>("op_25862_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_25862_end_0 = const()[name = tensor<string, []>("op_25862_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_25862_end_mask_0 = const()[name = tensor<string, []>("op_25862_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25862_cast = slice_by_index(begin = var_25862_begin_0, end = var_25862_end_0, end_mask = var_25862_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25862_cast")];
            tensor<int32, [4]> var_25866_begin_0 = const()[name = tensor<string, []>("op_25866_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_25866_end_0 = const()[name = tensor<string, []>("op_25866_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_25866_end_mask_0 = const()[name = tensor<string, []>("op_25866_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25866_cast = slice_by_index(begin = var_25866_begin_0, end = var_25866_end_0, end_mask = var_25866_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25866_cast")];
            tensor<int32, [4]> var_25870_begin_0 = const()[name = tensor<string, []>("op_25870_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_25870_end_0 = const()[name = tensor<string, []>("op_25870_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_25870_end_mask_0 = const()[name = tensor<string, []>("op_25870_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25870_cast = slice_by_index(begin = var_25870_begin_0, end = var_25870_end_0, end_mask = var_25870_end_mask_0, x = q_117_cast)[name = tensor<string, []>("op_25870_cast")];
            tensor<int32, [4]> k_235_perm_0 = const()[name = tensor<string, []>("k_235_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_25877_begin_0 = const()[name = tensor<string, []>("op_25877_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_25877_end_0 = const()[name = tensor<string, []>("op_25877_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_25877_end_mask_0 = const()[name = tensor<string, []>("op_25877_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_81 = transpose(perm = k_235_perm_0, x = k_233_cast)[name = tensor<string, []>("transpose_81")];
            tensor<fp16, [2, 1024, 1, 64]> var_25877_cast = slice_by_index(begin = var_25877_begin_0, end = var_25877_end_0, end_mask = var_25877_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25877_cast")];
            tensor<int32, [4]> var_25881_begin_0 = const()[name = tensor<string, []>("op_25881_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_25881_end_0 = const()[name = tensor<string, []>("op_25881_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_25881_end_mask_0 = const()[name = tensor<string, []>("op_25881_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25881_cast = slice_by_index(begin = var_25881_begin_0, end = var_25881_end_0, end_mask = var_25881_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25881_cast")];
            tensor<int32, [4]> var_25885_begin_0 = const()[name = tensor<string, []>("op_25885_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_25885_end_0 = const()[name = tensor<string, []>("op_25885_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_25885_end_mask_0 = const()[name = tensor<string, []>("op_25885_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25885_cast = slice_by_index(begin = var_25885_begin_0, end = var_25885_end_0, end_mask = var_25885_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25885_cast")];
            tensor<int32, [4]> var_25889_begin_0 = const()[name = tensor<string, []>("op_25889_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_25889_end_0 = const()[name = tensor<string, []>("op_25889_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_25889_end_mask_0 = const()[name = tensor<string, []>("op_25889_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25889_cast = slice_by_index(begin = var_25889_begin_0, end = var_25889_end_0, end_mask = var_25889_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25889_cast")];
            tensor<int32, [4]> var_25893_begin_0 = const()[name = tensor<string, []>("op_25893_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_25893_end_0 = const()[name = tensor<string, []>("op_25893_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_25893_end_mask_0 = const()[name = tensor<string, []>("op_25893_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25893_cast = slice_by_index(begin = var_25893_begin_0, end = var_25893_end_0, end_mask = var_25893_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25893_cast")];
            tensor<int32, [4]> var_25897_begin_0 = const()[name = tensor<string, []>("op_25897_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_25897_end_0 = const()[name = tensor<string, []>("op_25897_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_25897_end_mask_0 = const()[name = tensor<string, []>("op_25897_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25897_cast = slice_by_index(begin = var_25897_begin_0, end = var_25897_end_0, end_mask = var_25897_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25897_cast")];
            tensor<int32, [4]> var_25901_begin_0 = const()[name = tensor<string, []>("op_25901_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_25901_end_0 = const()[name = tensor<string, []>("op_25901_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_25901_end_mask_0 = const()[name = tensor<string, []>("op_25901_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25901_cast = slice_by_index(begin = var_25901_begin_0, end = var_25901_end_0, end_mask = var_25901_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25901_cast")];
            tensor<int32, [4]> var_25905_begin_0 = const()[name = tensor<string, []>("op_25905_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_25905_end_0 = const()[name = tensor<string, []>("op_25905_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_25905_end_mask_0 = const()[name = tensor<string, []>("op_25905_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25905_cast = slice_by_index(begin = var_25905_begin_0, end = var_25905_end_0, end_mask = var_25905_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25905_cast")];
            tensor<int32, [4]> var_25909_begin_0 = const()[name = tensor<string, []>("op_25909_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_25909_end_0 = const()[name = tensor<string, []>("op_25909_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_25909_end_mask_0 = const()[name = tensor<string, []>("op_25909_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25909_cast = slice_by_index(begin = var_25909_begin_0, end = var_25909_end_0, end_mask = var_25909_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25909_cast")];
            tensor<int32, [4]> var_25913_begin_0 = const()[name = tensor<string, []>("op_25913_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_25913_end_0 = const()[name = tensor<string, []>("op_25913_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_25913_end_mask_0 = const()[name = tensor<string, []>("op_25913_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25913_cast = slice_by_index(begin = var_25913_begin_0, end = var_25913_end_0, end_mask = var_25913_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25913_cast")];
            tensor<int32, [4]> var_25917_begin_0 = const()[name = tensor<string, []>("op_25917_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_25917_end_0 = const()[name = tensor<string, []>("op_25917_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_25917_end_mask_0 = const()[name = tensor<string, []>("op_25917_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25917_cast = slice_by_index(begin = var_25917_begin_0, end = var_25917_end_0, end_mask = var_25917_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25917_cast")];
            tensor<int32, [4]> var_25921_begin_0 = const()[name = tensor<string, []>("op_25921_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_25921_end_0 = const()[name = tensor<string, []>("op_25921_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_25921_end_mask_0 = const()[name = tensor<string, []>("op_25921_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25921_cast = slice_by_index(begin = var_25921_begin_0, end = var_25921_end_0, end_mask = var_25921_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25921_cast")];
            tensor<int32, [4]> var_25925_begin_0 = const()[name = tensor<string, []>("op_25925_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_25925_end_0 = const()[name = tensor<string, []>("op_25925_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_25925_end_mask_0 = const()[name = tensor<string, []>("op_25925_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25925_cast = slice_by_index(begin = var_25925_begin_0, end = var_25925_end_0, end_mask = var_25925_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25925_cast")];
            tensor<int32, [4]> var_25929_begin_0 = const()[name = tensor<string, []>("op_25929_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_25929_end_0 = const()[name = tensor<string, []>("op_25929_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_25929_end_mask_0 = const()[name = tensor<string, []>("op_25929_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25929_cast = slice_by_index(begin = var_25929_begin_0, end = var_25929_end_0, end_mask = var_25929_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25929_cast")];
            tensor<int32, [4]> var_25933_begin_0 = const()[name = tensor<string, []>("op_25933_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_25933_end_0 = const()[name = tensor<string, []>("op_25933_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_25933_end_mask_0 = const()[name = tensor<string, []>("op_25933_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25933_cast = slice_by_index(begin = var_25933_begin_0, end = var_25933_end_0, end_mask = var_25933_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25933_cast")];
            tensor<int32, [4]> var_25937_begin_0 = const()[name = tensor<string, []>("op_25937_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_25937_end_0 = const()[name = tensor<string, []>("op_25937_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_25937_end_mask_0 = const()[name = tensor<string, []>("op_25937_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25937_cast = slice_by_index(begin = var_25937_begin_0, end = var_25937_end_0, end_mask = var_25937_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25937_cast")];
            tensor<int32, [4]> var_25941_begin_0 = const()[name = tensor<string, []>("op_25941_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_25941_end_0 = const()[name = tensor<string, []>("op_25941_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_25941_end_mask_0 = const()[name = tensor<string, []>("op_25941_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25941_cast = slice_by_index(begin = var_25941_begin_0, end = var_25941_end_0, end_mask = var_25941_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25941_cast")];
            tensor<int32, [4]> var_25945_begin_0 = const()[name = tensor<string, []>("op_25945_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_25945_end_0 = const()[name = tensor<string, []>("op_25945_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_25945_end_mask_0 = const()[name = tensor<string, []>("op_25945_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25945_cast = slice_by_index(begin = var_25945_begin_0, end = var_25945_end_0, end_mask = var_25945_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25945_cast")];
            tensor<int32, [4]> var_25949_begin_0 = const()[name = tensor<string, []>("op_25949_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_25949_end_0 = const()[name = tensor<string, []>("op_25949_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_25949_end_mask_0 = const()[name = tensor<string, []>("op_25949_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25949_cast = slice_by_index(begin = var_25949_begin_0, end = var_25949_end_0, end_mask = var_25949_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25949_cast")];
            tensor<int32, [4]> var_25953_begin_0 = const()[name = tensor<string, []>("op_25953_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_25953_end_0 = const()[name = tensor<string, []>("op_25953_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_25953_end_mask_0 = const()[name = tensor<string, []>("op_25953_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_25953_cast = slice_by_index(begin = var_25953_begin_0, end = var_25953_end_0, end_mask = var_25953_end_mask_0, x = transpose_81)[name = tensor<string, []>("op_25953_cast")];
            tensor<int32, [4]> var_25955_begin_0 = const()[name = tensor<string, []>("op_25955_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_25955_end_0 = const()[name = tensor<string, []>("op_25955_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_25955_end_mask_0 = const()[name = tensor<string, []>("op_25955_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25955_cast = slice_by_index(begin = var_25955_begin_0, end = var_25955_end_0, end_mask = var_25955_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_25955_cast")];
            tensor<int32, [4]> var_25959_begin_0 = const()[name = tensor<string, []>("op_25959_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_25959_end_0 = const()[name = tensor<string, []>("op_25959_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_25959_end_mask_0 = const()[name = tensor<string, []>("op_25959_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25959_cast = slice_by_index(begin = var_25959_begin_0, end = var_25959_end_0, end_mask = var_25959_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_25959_cast")];
            tensor<int32, [4]> var_25963_begin_0 = const()[name = tensor<string, []>("op_25963_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_25963_end_0 = const()[name = tensor<string, []>("op_25963_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_25963_end_mask_0 = const()[name = tensor<string, []>("op_25963_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25963_cast = slice_by_index(begin = var_25963_begin_0, end = var_25963_end_0, end_mask = var_25963_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_25963_cast")];
            tensor<int32, [4]> var_25967_begin_0 = const()[name = tensor<string, []>("op_25967_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_25967_end_0 = const()[name = tensor<string, []>("op_25967_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_25967_end_mask_0 = const()[name = tensor<string, []>("op_25967_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25967_cast = slice_by_index(begin = var_25967_begin_0, end = var_25967_end_0, end_mask = var_25967_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_25967_cast")];
            tensor<int32, [4]> var_25971_begin_0 = const()[name = tensor<string, []>("op_25971_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_25971_end_0 = const()[name = tensor<string, []>("op_25971_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_25971_end_mask_0 = const()[name = tensor<string, []>("op_25971_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25971_cast = slice_by_index(begin = var_25971_begin_0, end = var_25971_end_0, end_mask = var_25971_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_25971_cast")];
            tensor<int32, [4]> var_25975_begin_0 = const()[name = tensor<string, []>("op_25975_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_25975_end_0 = const()[name = tensor<string, []>("op_25975_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_25975_end_mask_0 = const()[name = tensor<string, []>("op_25975_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25975_cast = slice_by_index(begin = var_25975_begin_0, end = var_25975_end_0, end_mask = var_25975_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_25975_cast")];
            tensor<int32, [4]> var_25979_begin_0 = const()[name = tensor<string, []>("op_25979_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_25979_end_0 = const()[name = tensor<string, []>("op_25979_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_25979_end_mask_0 = const()[name = tensor<string, []>("op_25979_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25979_cast = slice_by_index(begin = var_25979_begin_0, end = var_25979_end_0, end_mask = var_25979_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_25979_cast")];
            tensor<int32, [4]> var_25983_begin_0 = const()[name = tensor<string, []>("op_25983_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_25983_end_0 = const()[name = tensor<string, []>("op_25983_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_25983_end_mask_0 = const()[name = tensor<string, []>("op_25983_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25983_cast = slice_by_index(begin = var_25983_begin_0, end = var_25983_end_0, end_mask = var_25983_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_25983_cast")];
            tensor<int32, [4]> var_25987_begin_0 = const()[name = tensor<string, []>("op_25987_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_25987_end_0 = const()[name = tensor<string, []>("op_25987_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_25987_end_mask_0 = const()[name = tensor<string, []>("op_25987_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25987_cast = slice_by_index(begin = var_25987_begin_0, end = var_25987_end_0, end_mask = var_25987_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_25987_cast")];
            tensor<int32, [4]> var_25991_begin_0 = const()[name = tensor<string, []>("op_25991_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_25991_end_0 = const()[name = tensor<string, []>("op_25991_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_25991_end_mask_0 = const()[name = tensor<string, []>("op_25991_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25991_cast = slice_by_index(begin = var_25991_begin_0, end = var_25991_end_0, end_mask = var_25991_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_25991_cast")];
            tensor<int32, [4]> var_25995_begin_0 = const()[name = tensor<string, []>("op_25995_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_25995_end_0 = const()[name = tensor<string, []>("op_25995_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_25995_end_mask_0 = const()[name = tensor<string, []>("op_25995_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25995_cast = slice_by_index(begin = var_25995_begin_0, end = var_25995_end_0, end_mask = var_25995_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_25995_cast")];
            tensor<int32, [4]> var_25999_begin_0 = const()[name = tensor<string, []>("op_25999_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_25999_end_0 = const()[name = tensor<string, []>("op_25999_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_25999_end_mask_0 = const()[name = tensor<string, []>("op_25999_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_25999_cast = slice_by_index(begin = var_25999_begin_0, end = var_25999_end_0, end_mask = var_25999_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_25999_cast")];
            tensor<int32, [4]> var_26003_begin_0 = const()[name = tensor<string, []>("op_26003_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_26003_end_0 = const()[name = tensor<string, []>("op_26003_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_26003_end_mask_0 = const()[name = tensor<string, []>("op_26003_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26003_cast = slice_by_index(begin = var_26003_begin_0, end = var_26003_end_0, end_mask = var_26003_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_26003_cast")];
            tensor<int32, [4]> var_26007_begin_0 = const()[name = tensor<string, []>("op_26007_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_26007_end_0 = const()[name = tensor<string, []>("op_26007_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_26007_end_mask_0 = const()[name = tensor<string, []>("op_26007_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26007_cast = slice_by_index(begin = var_26007_begin_0, end = var_26007_end_0, end_mask = var_26007_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_26007_cast")];
            tensor<int32, [4]> var_26011_begin_0 = const()[name = tensor<string, []>("op_26011_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_26011_end_0 = const()[name = tensor<string, []>("op_26011_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_26011_end_mask_0 = const()[name = tensor<string, []>("op_26011_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26011_cast = slice_by_index(begin = var_26011_begin_0, end = var_26011_end_0, end_mask = var_26011_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_26011_cast")];
            tensor<int32, [4]> var_26015_begin_0 = const()[name = tensor<string, []>("op_26015_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_26015_end_0 = const()[name = tensor<string, []>("op_26015_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_26015_end_mask_0 = const()[name = tensor<string, []>("op_26015_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26015_cast = slice_by_index(begin = var_26015_begin_0, end = var_26015_end_0, end_mask = var_26015_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_26015_cast")];
            tensor<int32, [4]> var_26019_begin_0 = const()[name = tensor<string, []>("op_26019_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_26019_end_0 = const()[name = tensor<string, []>("op_26019_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_26019_end_mask_0 = const()[name = tensor<string, []>("op_26019_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26019_cast = slice_by_index(begin = var_26019_begin_0, end = var_26019_end_0, end_mask = var_26019_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_26019_cast")];
            tensor<int32, [4]> var_26023_begin_0 = const()[name = tensor<string, []>("op_26023_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_26023_end_0 = const()[name = tensor<string, []>("op_26023_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_26023_end_mask_0 = const()[name = tensor<string, []>("op_26023_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26023_cast = slice_by_index(begin = var_26023_begin_0, end = var_26023_end_0, end_mask = var_26023_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_26023_cast")];
            tensor<int32, [4]> var_26027_begin_0 = const()[name = tensor<string, []>("op_26027_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_26027_end_0 = const()[name = tensor<string, []>("op_26027_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_26027_end_mask_0 = const()[name = tensor<string, []>("op_26027_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26027_cast = slice_by_index(begin = var_26027_begin_0, end = var_26027_end_0, end_mask = var_26027_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_26027_cast")];
            tensor<int32, [4]> var_26031_begin_0 = const()[name = tensor<string, []>("op_26031_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_26031_end_0 = const()[name = tensor<string, []>("op_26031_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_26031_end_mask_0 = const()[name = tensor<string, []>("op_26031_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26031_cast = slice_by_index(begin = var_26031_begin_0, end = var_26031_end_0, end_mask = var_26031_end_mask_0, x = v_117_cast)[name = tensor<string, []>("op_26031_cast")];
            tensor<string, []> var_26035_equation_0 = const()[name = tensor<string, []>("op_26035_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26035_cast = einsum(equation = var_26035_equation_0, values = (var_25877_cast, var_25794_cast))[name = tensor<string, []>("op_26035_cast")];
            tensor<fp16, []> var_26036_to_fp16 = const()[name = tensor<string, []>("op_26036_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2161_cast = mul(x = var_26035_cast, y = var_26036_to_fp16)[name = tensor<string, []>("aw_2161_cast")];
            tensor<string, []> var_26039_equation_0 = const()[name = tensor<string, []>("op_26039_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26039_cast = einsum(equation = var_26039_equation_0, values = (var_25881_cast, var_25798_cast))[name = tensor<string, []>("op_26039_cast")];
            tensor<fp16, []> var_26040_to_fp16 = const()[name = tensor<string, []>("op_26040_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2163_cast = mul(x = var_26039_cast, y = var_26040_to_fp16)[name = tensor<string, []>("aw_2163_cast")];
            tensor<string, []> var_26043_equation_0 = const()[name = tensor<string, []>("op_26043_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26043_cast = einsum(equation = var_26043_equation_0, values = (var_25885_cast, var_25802_cast))[name = tensor<string, []>("op_26043_cast")];
            tensor<fp16, []> var_26044_to_fp16 = const()[name = tensor<string, []>("op_26044_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2165_cast = mul(x = var_26043_cast, y = var_26044_to_fp16)[name = tensor<string, []>("aw_2165_cast")];
            tensor<string, []> var_26047_equation_0 = const()[name = tensor<string, []>("op_26047_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26047_cast = einsum(equation = var_26047_equation_0, values = (var_25889_cast, var_25806_cast))[name = tensor<string, []>("op_26047_cast")];
            tensor<fp16, []> var_26048_to_fp16 = const()[name = tensor<string, []>("op_26048_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2167_cast = mul(x = var_26047_cast, y = var_26048_to_fp16)[name = tensor<string, []>("aw_2167_cast")];
            tensor<string, []> var_26051_equation_0 = const()[name = tensor<string, []>("op_26051_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26051_cast = einsum(equation = var_26051_equation_0, values = (var_25893_cast, var_25810_cast))[name = tensor<string, []>("op_26051_cast")];
            tensor<fp16, []> var_26052_to_fp16 = const()[name = tensor<string, []>("op_26052_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2169_cast = mul(x = var_26051_cast, y = var_26052_to_fp16)[name = tensor<string, []>("aw_2169_cast")];
            tensor<string, []> var_26055_equation_0 = const()[name = tensor<string, []>("op_26055_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26055_cast = einsum(equation = var_26055_equation_0, values = (var_25897_cast, var_25814_cast))[name = tensor<string, []>("op_26055_cast")];
            tensor<fp16, []> var_26056_to_fp16 = const()[name = tensor<string, []>("op_26056_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2171_cast = mul(x = var_26055_cast, y = var_26056_to_fp16)[name = tensor<string, []>("aw_2171_cast")];
            tensor<string, []> var_26059_equation_0 = const()[name = tensor<string, []>("op_26059_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26059_cast = einsum(equation = var_26059_equation_0, values = (var_25901_cast, var_25818_cast))[name = tensor<string, []>("op_26059_cast")];
            tensor<fp16, []> var_26060_to_fp16 = const()[name = tensor<string, []>("op_26060_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2173_cast = mul(x = var_26059_cast, y = var_26060_to_fp16)[name = tensor<string, []>("aw_2173_cast")];
            tensor<string, []> var_26063_equation_0 = const()[name = tensor<string, []>("op_26063_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26063_cast = einsum(equation = var_26063_equation_0, values = (var_25905_cast, var_25822_cast))[name = tensor<string, []>("op_26063_cast")];
            tensor<fp16, []> var_26064_to_fp16 = const()[name = tensor<string, []>("op_26064_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2175_cast = mul(x = var_26063_cast, y = var_26064_to_fp16)[name = tensor<string, []>("aw_2175_cast")];
            tensor<string, []> var_26067_equation_0 = const()[name = tensor<string, []>("op_26067_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26067_cast = einsum(equation = var_26067_equation_0, values = (var_25909_cast, var_25826_cast))[name = tensor<string, []>("op_26067_cast")];
            tensor<fp16, []> var_26068_to_fp16 = const()[name = tensor<string, []>("op_26068_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2177_cast = mul(x = var_26067_cast, y = var_26068_to_fp16)[name = tensor<string, []>("aw_2177_cast")];
            tensor<string, []> var_26071_equation_0 = const()[name = tensor<string, []>("op_26071_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26071_cast = einsum(equation = var_26071_equation_0, values = (var_25913_cast, var_25830_cast))[name = tensor<string, []>("op_26071_cast")];
            tensor<fp16, []> var_26072_to_fp16 = const()[name = tensor<string, []>("op_26072_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2179_cast = mul(x = var_26071_cast, y = var_26072_to_fp16)[name = tensor<string, []>("aw_2179_cast")];
            tensor<string, []> var_26075_equation_0 = const()[name = tensor<string, []>("op_26075_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26075_cast = einsum(equation = var_26075_equation_0, values = (var_25917_cast, var_25834_cast))[name = tensor<string, []>("op_26075_cast")];
            tensor<fp16, []> var_26076_to_fp16 = const()[name = tensor<string, []>("op_26076_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2181_cast = mul(x = var_26075_cast, y = var_26076_to_fp16)[name = tensor<string, []>("aw_2181_cast")];
            tensor<string, []> var_26079_equation_0 = const()[name = tensor<string, []>("op_26079_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26079_cast = einsum(equation = var_26079_equation_0, values = (var_25921_cast, var_25838_cast))[name = tensor<string, []>("op_26079_cast")];
            tensor<fp16, []> var_26080_to_fp16 = const()[name = tensor<string, []>("op_26080_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2183_cast = mul(x = var_26079_cast, y = var_26080_to_fp16)[name = tensor<string, []>("aw_2183_cast")];
            tensor<string, []> var_26083_equation_0 = const()[name = tensor<string, []>("op_26083_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26083_cast = einsum(equation = var_26083_equation_0, values = (var_25925_cast, var_25842_cast))[name = tensor<string, []>("op_26083_cast")];
            tensor<fp16, []> var_26084_to_fp16 = const()[name = tensor<string, []>("op_26084_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2185_cast = mul(x = var_26083_cast, y = var_26084_to_fp16)[name = tensor<string, []>("aw_2185_cast")];
            tensor<string, []> var_26087_equation_0 = const()[name = tensor<string, []>("op_26087_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26087_cast = einsum(equation = var_26087_equation_0, values = (var_25929_cast, var_25846_cast))[name = tensor<string, []>("op_26087_cast")];
            tensor<fp16, []> var_26088_to_fp16 = const()[name = tensor<string, []>("op_26088_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2187_cast = mul(x = var_26087_cast, y = var_26088_to_fp16)[name = tensor<string, []>("aw_2187_cast")];
            tensor<string, []> var_26091_equation_0 = const()[name = tensor<string, []>("op_26091_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26091_cast = einsum(equation = var_26091_equation_0, values = (var_25933_cast, var_25850_cast))[name = tensor<string, []>("op_26091_cast")];
            tensor<fp16, []> var_26092_to_fp16 = const()[name = tensor<string, []>("op_26092_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2189_cast = mul(x = var_26091_cast, y = var_26092_to_fp16)[name = tensor<string, []>("aw_2189_cast")];
            tensor<string, []> var_26095_equation_0 = const()[name = tensor<string, []>("op_26095_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26095_cast = einsum(equation = var_26095_equation_0, values = (var_25937_cast, var_25854_cast))[name = tensor<string, []>("op_26095_cast")];
            tensor<fp16, []> var_26096_to_fp16 = const()[name = tensor<string, []>("op_26096_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2191_cast = mul(x = var_26095_cast, y = var_26096_to_fp16)[name = tensor<string, []>("aw_2191_cast")];
            tensor<string, []> var_26099_equation_0 = const()[name = tensor<string, []>("op_26099_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26099_cast = einsum(equation = var_26099_equation_0, values = (var_25941_cast, var_25858_cast))[name = tensor<string, []>("op_26099_cast")];
            tensor<fp16, []> var_26100_to_fp16 = const()[name = tensor<string, []>("op_26100_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2193_cast = mul(x = var_26099_cast, y = var_26100_to_fp16)[name = tensor<string, []>("aw_2193_cast")];
            tensor<string, []> var_26103_equation_0 = const()[name = tensor<string, []>("op_26103_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26103_cast = einsum(equation = var_26103_equation_0, values = (var_25945_cast, var_25862_cast))[name = tensor<string, []>("op_26103_cast")];
            tensor<fp16, []> var_26104_to_fp16 = const()[name = tensor<string, []>("op_26104_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2195_cast = mul(x = var_26103_cast, y = var_26104_to_fp16)[name = tensor<string, []>("aw_2195_cast")];
            tensor<string, []> var_26107_equation_0 = const()[name = tensor<string, []>("op_26107_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26107_cast = einsum(equation = var_26107_equation_0, values = (var_25949_cast, var_25866_cast))[name = tensor<string, []>("op_26107_cast")];
            tensor<fp16, []> var_26108_to_fp16 = const()[name = tensor<string, []>("op_26108_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2197_cast = mul(x = var_26107_cast, y = var_26108_to_fp16)[name = tensor<string, []>("aw_2197_cast")];
            tensor<string, []> var_26111_equation_0 = const()[name = tensor<string, []>("op_26111_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26111_cast = einsum(equation = var_26111_equation_0, values = (var_25953_cast, var_25870_cast))[name = tensor<string, []>("op_26111_cast")];
            tensor<fp16, []> var_26112_to_fp16 = const()[name = tensor<string, []>("op_26112_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2199_cast = mul(x = var_26111_cast, y = var_26112_to_fp16)[name = tensor<string, []>("aw_2199_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26114_cast = softmax(axis = var_21105, x = aw_2161_cast)[name = tensor<string, []>("op_26114_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26115_cast = softmax(axis = var_21105, x = aw_2163_cast)[name = tensor<string, []>("op_26115_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26116_cast = softmax(axis = var_21105, x = aw_2165_cast)[name = tensor<string, []>("op_26116_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26117_cast = softmax(axis = var_21105, x = aw_2167_cast)[name = tensor<string, []>("op_26117_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26118_cast = softmax(axis = var_21105, x = aw_2169_cast)[name = tensor<string, []>("op_26118_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26119_cast = softmax(axis = var_21105, x = aw_2171_cast)[name = tensor<string, []>("op_26119_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26120_cast = softmax(axis = var_21105, x = aw_2173_cast)[name = tensor<string, []>("op_26120_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26121_cast = softmax(axis = var_21105, x = aw_2175_cast)[name = tensor<string, []>("op_26121_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26122_cast = softmax(axis = var_21105, x = aw_2177_cast)[name = tensor<string, []>("op_26122_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26123_cast = softmax(axis = var_21105, x = aw_2179_cast)[name = tensor<string, []>("op_26123_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26124_cast = softmax(axis = var_21105, x = aw_2181_cast)[name = tensor<string, []>("op_26124_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26125_cast = softmax(axis = var_21105, x = aw_2183_cast)[name = tensor<string, []>("op_26125_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26126_cast = softmax(axis = var_21105, x = aw_2185_cast)[name = tensor<string, []>("op_26126_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26127_cast = softmax(axis = var_21105, x = aw_2187_cast)[name = tensor<string, []>("op_26127_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26128_cast = softmax(axis = var_21105, x = aw_2189_cast)[name = tensor<string, []>("op_26128_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26129_cast = softmax(axis = var_21105, x = aw_2191_cast)[name = tensor<string, []>("op_26129_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26130_cast = softmax(axis = var_21105, x = aw_2193_cast)[name = tensor<string, []>("op_26130_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26131_cast = softmax(axis = var_21105, x = aw_2195_cast)[name = tensor<string, []>("op_26131_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26132_cast = softmax(axis = var_21105, x = aw_2197_cast)[name = tensor<string, []>("op_26132_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26133_cast = softmax(axis = var_21105, x = aw_2199_cast)[name = tensor<string, []>("op_26133_cast")];
            tensor<string, []> var_26135_equation_0 = const()[name = tensor<string, []>("op_26135_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26135_cast = einsum(equation = var_26135_equation_0, values = (var_25955_cast, var_26114_cast))[name = tensor<string, []>("op_26135_cast")];
            tensor<string, []> var_26137_equation_0 = const()[name = tensor<string, []>("op_26137_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26137_cast = einsum(equation = var_26137_equation_0, values = (var_25959_cast, var_26115_cast))[name = tensor<string, []>("op_26137_cast")];
            tensor<string, []> var_26139_equation_0 = const()[name = tensor<string, []>("op_26139_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26139_cast = einsum(equation = var_26139_equation_0, values = (var_25963_cast, var_26116_cast))[name = tensor<string, []>("op_26139_cast")];
            tensor<string, []> var_26141_equation_0 = const()[name = tensor<string, []>("op_26141_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26141_cast = einsum(equation = var_26141_equation_0, values = (var_25967_cast, var_26117_cast))[name = tensor<string, []>("op_26141_cast")];
            tensor<string, []> var_26143_equation_0 = const()[name = tensor<string, []>("op_26143_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26143_cast = einsum(equation = var_26143_equation_0, values = (var_25971_cast, var_26118_cast))[name = tensor<string, []>("op_26143_cast")];
            tensor<string, []> var_26145_equation_0 = const()[name = tensor<string, []>("op_26145_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26145_cast = einsum(equation = var_26145_equation_0, values = (var_25975_cast, var_26119_cast))[name = tensor<string, []>("op_26145_cast")];
            tensor<string, []> var_26147_equation_0 = const()[name = tensor<string, []>("op_26147_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26147_cast = einsum(equation = var_26147_equation_0, values = (var_25979_cast, var_26120_cast))[name = tensor<string, []>("op_26147_cast")];
            tensor<string, []> var_26149_equation_0 = const()[name = tensor<string, []>("op_26149_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26149_cast = einsum(equation = var_26149_equation_0, values = (var_25983_cast, var_26121_cast))[name = tensor<string, []>("op_26149_cast")];
            tensor<string, []> var_26151_equation_0 = const()[name = tensor<string, []>("op_26151_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26151_cast = einsum(equation = var_26151_equation_0, values = (var_25987_cast, var_26122_cast))[name = tensor<string, []>("op_26151_cast")];
            tensor<string, []> var_26153_equation_0 = const()[name = tensor<string, []>("op_26153_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26153_cast = einsum(equation = var_26153_equation_0, values = (var_25991_cast, var_26123_cast))[name = tensor<string, []>("op_26153_cast")];
            tensor<string, []> var_26155_equation_0 = const()[name = tensor<string, []>("op_26155_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26155_cast = einsum(equation = var_26155_equation_0, values = (var_25995_cast, var_26124_cast))[name = tensor<string, []>("op_26155_cast")];
            tensor<string, []> var_26157_equation_0 = const()[name = tensor<string, []>("op_26157_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26157_cast = einsum(equation = var_26157_equation_0, values = (var_25999_cast, var_26125_cast))[name = tensor<string, []>("op_26157_cast")];
            tensor<string, []> var_26159_equation_0 = const()[name = tensor<string, []>("op_26159_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26159_cast = einsum(equation = var_26159_equation_0, values = (var_26003_cast, var_26126_cast))[name = tensor<string, []>("op_26159_cast")];
            tensor<string, []> var_26161_equation_0 = const()[name = tensor<string, []>("op_26161_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26161_cast = einsum(equation = var_26161_equation_0, values = (var_26007_cast, var_26127_cast))[name = tensor<string, []>("op_26161_cast")];
            tensor<string, []> var_26163_equation_0 = const()[name = tensor<string, []>("op_26163_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26163_cast = einsum(equation = var_26163_equation_0, values = (var_26011_cast, var_26128_cast))[name = tensor<string, []>("op_26163_cast")];
            tensor<string, []> var_26165_equation_0 = const()[name = tensor<string, []>("op_26165_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26165_cast = einsum(equation = var_26165_equation_0, values = (var_26015_cast, var_26129_cast))[name = tensor<string, []>("op_26165_cast")];
            tensor<string, []> var_26167_equation_0 = const()[name = tensor<string, []>("op_26167_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26167_cast = einsum(equation = var_26167_equation_0, values = (var_26019_cast, var_26130_cast))[name = tensor<string, []>("op_26167_cast")];
            tensor<string, []> var_26169_equation_0 = const()[name = tensor<string, []>("op_26169_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26169_cast = einsum(equation = var_26169_equation_0, values = (var_26023_cast, var_26131_cast))[name = tensor<string, []>("op_26169_cast")];
            tensor<string, []> var_26171_equation_0 = const()[name = tensor<string, []>("op_26171_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26171_cast = einsum(equation = var_26171_equation_0, values = (var_26027_cast, var_26132_cast))[name = tensor<string, []>("op_26171_cast")];
            tensor<string, []> var_26173_equation_0 = const()[name = tensor<string, []>("op_26173_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26173_cast = einsum(equation = var_26173_equation_0, values = (var_26031_cast, var_26133_cast))[name = tensor<string, []>("op_26173_cast")];
            tensor<bool, []> input_369_interleave_0 = const()[name = tensor<string, []>("input_369_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_369_cast = concat(axis = var_21105, interleave = input_369_interleave_0, values = (var_26135_cast, var_26137_cast, var_26139_cast, var_26141_cast, var_26143_cast, var_26145_cast, var_26147_cast, var_26149_cast, var_26151_cast, var_26153_cast, var_26155_cast, var_26157_cast, var_26159_cast, var_26161_cast, var_26163_cast, var_26165_cast, var_26167_cast, var_26169_cast, var_26171_cast, var_26173_cast))[name = tensor<string, []>("input_369_cast")];
            tensor<int32, [2]> var_26179 = const()[name = tensor<string, []>("op_26179"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_26181 = const()[name = tensor<string, []>("op_26181"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_26183_pad_type_0 = const()[name = tensor<string, []>("op_26183_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_26183_pad_0 = const()[name = tensor<string, []>("op_26183_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2083509696)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2086786560)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_26183_cast = conv(bias = mid_block_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16, dilations = var_26181, groups = var_21105, pad = var_26183_pad_0, pad_type = var_26183_pad_type_0, strides = var_26179, weight = mid_block_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16, x = input_369_cast)[name = tensor<string, []>("op_26183_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_177_cast = add(x = var_26183_cast, y = inputs_175_cast)[name = tensor<string, []>("inputs_177_cast")];
            tensor<int32, [1]> var_26187 = const()[name = tensor<string, []>("op_26187"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_177_cast = reduce_mean(axes = var_26187, keep_dims = var_21100, x = inputs_177_cast)[name = tensor<string, []>("channels_mean_177_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_177_cast = sub(x = inputs_177_cast, y = channels_mean_177_cast)[name = tensor<string, []>("zero_mean_177_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_177_cast = mul(x = zero_mean_177_cast, y = zero_mean_177_cast)[name = tensor<string, []>("zero_mean_sq_177_cast")];
            tensor<int32, [1]> var_26191 = const()[name = tensor<string, []>("op_26191"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_26192_cast = reduce_mean(axes = var_26191, keep_dims = var_21100, x = zero_mean_sq_177_cast)[name = tensor<string, []>("op_26192_cast")];
            tensor<fp16, []> var_26193_to_fp16 = const()[name = tensor<string, []>("op_26193_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_26194_cast = add(x = var_26192_cast, y = var_26193_to_fp16)[name = tensor<string, []>("op_26194_cast")];
            tensor<fp16, []> denom_177_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_177_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_177_cast = rsqrt(epsilon = denom_177_epsilon_0_to_fp16, x = var_26194_cast)[name = tensor<string, []>("denom_177_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_177_cast = mul(x = zero_mean_177_cast, y = denom_177_cast)[name = tensor<string, []>("out_177_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_26198_to_fp16 = const()[name = tensor<string, []>("op_26198_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2086789184)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_26199_cast = add(x = out_177_cast, y = var_26198_to_fp16)[name = tensor<string, []>("op_26199_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_26201_to_fp16 = const()[name = tensor<string, []>("op_26201_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2086791808)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_241_cast = mul(x = var_26199_cast, y = var_26201_to_fp16)[name = tensor<string, []>("hidden_states_241_cast")];
            tensor<int32, [2]> var_26208 = const()[name = tensor<string, []>("op_26208"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_26210 = const()[name = tensor<string, []>("op_26210"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_119_pad_type_0 = const()[name = tensor<string, []>("q_119_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_119_pad_0 = const()[name = tensor<string, []>("q_119_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2086794432)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_119_cast = conv(dilations = var_26210, groups = var_21105, pad = q_119_pad_0, pad_type = q_119_pad_type_0, strides = var_26208, weight = mid_block_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16, x = hidden_states_241_cast)[name = tensor<string, []>("q_119_cast")];
            tensor<int32, [2]> var_26214 = const()[name = tensor<string, []>("op_26214"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_26216 = const()[name = tensor<string, []>("op_26216"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_237_pad_type_0 = const()[name = tensor<string, []>("k_237_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_237_pad_0 = const()[name = tensor<string, []>("k_237_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2090071296)))];
            tensor<fp16, [2, 1280, 1, 77]> k_237_cast = conv(dilations = var_26216, groups = var_21105, pad = k_237_pad_0, pad_type = k_237_pad_type_0, strides = var_26214, weight = mid_block_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_237_cast")];
            tensor<int32, [2]> var_26220 = const()[name = tensor<string, []>("op_26220"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_26222 = const()[name = tensor<string, []>("op_26222"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_119_pad_type_0 = const()[name = tensor<string, []>("v_119_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_119_pad_0 = const()[name = tensor<string, []>("v_119_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2095314240)))];
            tensor<fp16, [2, 1280, 1, 77]> v_119_cast = conv(dilations = var_26222, groups = var_21105, pad = v_119_pad_0, pad_type = v_119_pad_type_0, strides = var_26220, weight = mid_block_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_119_cast")];
            tensor<int32, [4]> var_26226_begin_0 = const()[name = tensor<string, []>("op_26226_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_26226_end_0 = const()[name = tensor<string, []>("op_26226_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_26226_end_mask_0 = const()[name = tensor<string, []>("op_26226_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26226_cast = slice_by_index(begin = var_26226_begin_0, end = var_26226_end_0, end_mask = var_26226_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26226_cast")];
            tensor<int32, [4]> var_26230_begin_0 = const()[name = tensor<string, []>("op_26230_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_26230_end_0 = const()[name = tensor<string, []>("op_26230_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_26230_end_mask_0 = const()[name = tensor<string, []>("op_26230_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26230_cast = slice_by_index(begin = var_26230_begin_0, end = var_26230_end_0, end_mask = var_26230_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26230_cast")];
            tensor<int32, [4]> var_26234_begin_0 = const()[name = tensor<string, []>("op_26234_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_26234_end_0 = const()[name = tensor<string, []>("op_26234_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_26234_end_mask_0 = const()[name = tensor<string, []>("op_26234_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26234_cast = slice_by_index(begin = var_26234_begin_0, end = var_26234_end_0, end_mask = var_26234_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26234_cast")];
            tensor<int32, [4]> var_26238_begin_0 = const()[name = tensor<string, []>("op_26238_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_26238_end_0 = const()[name = tensor<string, []>("op_26238_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_26238_end_mask_0 = const()[name = tensor<string, []>("op_26238_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26238_cast = slice_by_index(begin = var_26238_begin_0, end = var_26238_end_0, end_mask = var_26238_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26238_cast")];
            tensor<int32, [4]> var_26242_begin_0 = const()[name = tensor<string, []>("op_26242_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_26242_end_0 = const()[name = tensor<string, []>("op_26242_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_26242_end_mask_0 = const()[name = tensor<string, []>("op_26242_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26242_cast = slice_by_index(begin = var_26242_begin_0, end = var_26242_end_0, end_mask = var_26242_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26242_cast")];
            tensor<int32, [4]> var_26246_begin_0 = const()[name = tensor<string, []>("op_26246_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_26246_end_0 = const()[name = tensor<string, []>("op_26246_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_26246_end_mask_0 = const()[name = tensor<string, []>("op_26246_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26246_cast = slice_by_index(begin = var_26246_begin_0, end = var_26246_end_0, end_mask = var_26246_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26246_cast")];
            tensor<int32, [4]> var_26250_begin_0 = const()[name = tensor<string, []>("op_26250_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_26250_end_0 = const()[name = tensor<string, []>("op_26250_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_26250_end_mask_0 = const()[name = tensor<string, []>("op_26250_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26250_cast = slice_by_index(begin = var_26250_begin_0, end = var_26250_end_0, end_mask = var_26250_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26250_cast")];
            tensor<int32, [4]> var_26254_begin_0 = const()[name = tensor<string, []>("op_26254_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_26254_end_0 = const()[name = tensor<string, []>("op_26254_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_26254_end_mask_0 = const()[name = tensor<string, []>("op_26254_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26254_cast = slice_by_index(begin = var_26254_begin_0, end = var_26254_end_0, end_mask = var_26254_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26254_cast")];
            tensor<int32, [4]> var_26258_begin_0 = const()[name = tensor<string, []>("op_26258_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_26258_end_0 = const()[name = tensor<string, []>("op_26258_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_26258_end_mask_0 = const()[name = tensor<string, []>("op_26258_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26258_cast = slice_by_index(begin = var_26258_begin_0, end = var_26258_end_0, end_mask = var_26258_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26258_cast")];
            tensor<int32, [4]> var_26262_begin_0 = const()[name = tensor<string, []>("op_26262_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_26262_end_0 = const()[name = tensor<string, []>("op_26262_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_26262_end_mask_0 = const()[name = tensor<string, []>("op_26262_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26262_cast = slice_by_index(begin = var_26262_begin_0, end = var_26262_end_0, end_mask = var_26262_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26262_cast")];
            tensor<int32, [4]> var_26266_begin_0 = const()[name = tensor<string, []>("op_26266_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_26266_end_0 = const()[name = tensor<string, []>("op_26266_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_26266_end_mask_0 = const()[name = tensor<string, []>("op_26266_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26266_cast = slice_by_index(begin = var_26266_begin_0, end = var_26266_end_0, end_mask = var_26266_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26266_cast")];
            tensor<int32, [4]> var_26270_begin_0 = const()[name = tensor<string, []>("op_26270_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_26270_end_0 = const()[name = tensor<string, []>("op_26270_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_26270_end_mask_0 = const()[name = tensor<string, []>("op_26270_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26270_cast = slice_by_index(begin = var_26270_begin_0, end = var_26270_end_0, end_mask = var_26270_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26270_cast")];
            tensor<int32, [4]> var_26274_begin_0 = const()[name = tensor<string, []>("op_26274_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_26274_end_0 = const()[name = tensor<string, []>("op_26274_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_26274_end_mask_0 = const()[name = tensor<string, []>("op_26274_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26274_cast = slice_by_index(begin = var_26274_begin_0, end = var_26274_end_0, end_mask = var_26274_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26274_cast")];
            tensor<int32, [4]> var_26278_begin_0 = const()[name = tensor<string, []>("op_26278_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_26278_end_0 = const()[name = tensor<string, []>("op_26278_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_26278_end_mask_0 = const()[name = tensor<string, []>("op_26278_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26278_cast = slice_by_index(begin = var_26278_begin_0, end = var_26278_end_0, end_mask = var_26278_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26278_cast")];
            tensor<int32, [4]> var_26282_begin_0 = const()[name = tensor<string, []>("op_26282_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_26282_end_0 = const()[name = tensor<string, []>("op_26282_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_26282_end_mask_0 = const()[name = tensor<string, []>("op_26282_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26282_cast = slice_by_index(begin = var_26282_begin_0, end = var_26282_end_0, end_mask = var_26282_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26282_cast")];
            tensor<int32, [4]> var_26286_begin_0 = const()[name = tensor<string, []>("op_26286_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_26286_end_0 = const()[name = tensor<string, []>("op_26286_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_26286_end_mask_0 = const()[name = tensor<string, []>("op_26286_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26286_cast = slice_by_index(begin = var_26286_begin_0, end = var_26286_end_0, end_mask = var_26286_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26286_cast")];
            tensor<int32, [4]> var_26290_begin_0 = const()[name = tensor<string, []>("op_26290_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_26290_end_0 = const()[name = tensor<string, []>("op_26290_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_26290_end_mask_0 = const()[name = tensor<string, []>("op_26290_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26290_cast = slice_by_index(begin = var_26290_begin_0, end = var_26290_end_0, end_mask = var_26290_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26290_cast")];
            tensor<int32, [4]> var_26294_begin_0 = const()[name = tensor<string, []>("op_26294_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_26294_end_0 = const()[name = tensor<string, []>("op_26294_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_26294_end_mask_0 = const()[name = tensor<string, []>("op_26294_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26294_cast = slice_by_index(begin = var_26294_begin_0, end = var_26294_end_0, end_mask = var_26294_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26294_cast")];
            tensor<int32, [4]> var_26298_begin_0 = const()[name = tensor<string, []>("op_26298_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_26298_end_0 = const()[name = tensor<string, []>("op_26298_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_26298_end_mask_0 = const()[name = tensor<string, []>("op_26298_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26298_cast = slice_by_index(begin = var_26298_begin_0, end = var_26298_end_0, end_mask = var_26298_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26298_cast")];
            tensor<int32, [4]> var_26302_begin_0 = const()[name = tensor<string, []>("op_26302_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_26302_end_0 = const()[name = tensor<string, []>("op_26302_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_26302_end_mask_0 = const()[name = tensor<string, []>("op_26302_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26302_cast = slice_by_index(begin = var_26302_begin_0, end = var_26302_end_0, end_mask = var_26302_end_mask_0, x = q_119_cast)[name = tensor<string, []>("op_26302_cast")];
            tensor<int32, [4]> k_239_perm_0 = const()[name = tensor<string, []>("k_239_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_26309_begin_0 = const()[name = tensor<string, []>("op_26309_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_26309_end_0 = const()[name = tensor<string, []>("op_26309_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_26309_end_mask_0 = const()[name = tensor<string, []>("op_26309_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_80 = transpose(perm = k_239_perm_0, x = k_237_cast)[name = tensor<string, []>("transpose_80")];
            tensor<fp16, [2, 77, 1, 64]> var_26309_cast = slice_by_index(begin = var_26309_begin_0, end = var_26309_end_0, end_mask = var_26309_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26309_cast")];
            tensor<int32, [4]> var_26313_begin_0 = const()[name = tensor<string, []>("op_26313_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_26313_end_0 = const()[name = tensor<string, []>("op_26313_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_26313_end_mask_0 = const()[name = tensor<string, []>("op_26313_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_26313_cast = slice_by_index(begin = var_26313_begin_0, end = var_26313_end_0, end_mask = var_26313_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26313_cast")];
            tensor<int32, [4]> var_26317_begin_0 = const()[name = tensor<string, []>("op_26317_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_26317_end_0 = const()[name = tensor<string, []>("op_26317_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_26317_end_mask_0 = const()[name = tensor<string, []>("op_26317_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_26317_cast = slice_by_index(begin = var_26317_begin_0, end = var_26317_end_0, end_mask = var_26317_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26317_cast")];
            tensor<int32, [4]> var_26321_begin_0 = const()[name = tensor<string, []>("op_26321_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_26321_end_0 = const()[name = tensor<string, []>("op_26321_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_26321_end_mask_0 = const()[name = tensor<string, []>("op_26321_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_26321_cast = slice_by_index(begin = var_26321_begin_0, end = var_26321_end_0, end_mask = var_26321_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26321_cast")];
            tensor<int32, [4]> var_26325_begin_0 = const()[name = tensor<string, []>("op_26325_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_26325_end_0 = const()[name = tensor<string, []>("op_26325_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_26325_end_mask_0 = const()[name = tensor<string, []>("op_26325_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_26325_cast = slice_by_index(begin = var_26325_begin_0, end = var_26325_end_0, end_mask = var_26325_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26325_cast")];
            tensor<int32, [4]> var_26329_begin_0 = const()[name = tensor<string, []>("op_26329_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_26329_end_0 = const()[name = tensor<string, []>("op_26329_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_26329_end_mask_0 = const()[name = tensor<string, []>("op_26329_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_26329_cast = slice_by_index(begin = var_26329_begin_0, end = var_26329_end_0, end_mask = var_26329_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26329_cast")];
            tensor<int32, [4]> var_26333_begin_0 = const()[name = tensor<string, []>("op_26333_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_26333_end_0 = const()[name = tensor<string, []>("op_26333_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_26333_end_mask_0 = const()[name = tensor<string, []>("op_26333_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_26333_cast = slice_by_index(begin = var_26333_begin_0, end = var_26333_end_0, end_mask = var_26333_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26333_cast")];
            tensor<int32, [4]> var_26337_begin_0 = const()[name = tensor<string, []>("op_26337_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_26337_end_0 = const()[name = tensor<string, []>("op_26337_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_26337_end_mask_0 = const()[name = tensor<string, []>("op_26337_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_26337_cast = slice_by_index(begin = var_26337_begin_0, end = var_26337_end_0, end_mask = var_26337_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26337_cast")];
            tensor<int32, [4]> var_26341_begin_0 = const()[name = tensor<string, []>("op_26341_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_26341_end_0 = const()[name = tensor<string, []>("op_26341_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_26341_end_mask_0 = const()[name = tensor<string, []>("op_26341_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_26341_cast = slice_by_index(begin = var_26341_begin_0, end = var_26341_end_0, end_mask = var_26341_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26341_cast")];
            tensor<int32, [4]> var_26345_begin_0 = const()[name = tensor<string, []>("op_26345_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_26345_end_0 = const()[name = tensor<string, []>("op_26345_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_26345_end_mask_0 = const()[name = tensor<string, []>("op_26345_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_26345_cast = slice_by_index(begin = var_26345_begin_0, end = var_26345_end_0, end_mask = var_26345_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26345_cast")];
            tensor<int32, [4]> var_26349_begin_0 = const()[name = tensor<string, []>("op_26349_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_26349_end_0 = const()[name = tensor<string, []>("op_26349_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_26349_end_mask_0 = const()[name = tensor<string, []>("op_26349_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_26349_cast = slice_by_index(begin = var_26349_begin_0, end = var_26349_end_0, end_mask = var_26349_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26349_cast")];
            tensor<int32, [4]> var_26353_begin_0 = const()[name = tensor<string, []>("op_26353_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_26353_end_0 = const()[name = tensor<string, []>("op_26353_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_26353_end_mask_0 = const()[name = tensor<string, []>("op_26353_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_26353_cast = slice_by_index(begin = var_26353_begin_0, end = var_26353_end_0, end_mask = var_26353_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26353_cast")];
            tensor<int32, [4]> var_26357_begin_0 = const()[name = tensor<string, []>("op_26357_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_26357_end_0 = const()[name = tensor<string, []>("op_26357_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_26357_end_mask_0 = const()[name = tensor<string, []>("op_26357_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_26357_cast = slice_by_index(begin = var_26357_begin_0, end = var_26357_end_0, end_mask = var_26357_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26357_cast")];
            tensor<int32, [4]> var_26361_begin_0 = const()[name = tensor<string, []>("op_26361_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_26361_end_0 = const()[name = tensor<string, []>("op_26361_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_26361_end_mask_0 = const()[name = tensor<string, []>("op_26361_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_26361_cast = slice_by_index(begin = var_26361_begin_0, end = var_26361_end_0, end_mask = var_26361_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26361_cast")];
            tensor<int32, [4]> var_26365_begin_0 = const()[name = tensor<string, []>("op_26365_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_26365_end_0 = const()[name = tensor<string, []>("op_26365_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_26365_end_mask_0 = const()[name = tensor<string, []>("op_26365_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_26365_cast = slice_by_index(begin = var_26365_begin_0, end = var_26365_end_0, end_mask = var_26365_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26365_cast")];
            tensor<int32, [4]> var_26369_begin_0 = const()[name = tensor<string, []>("op_26369_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_26369_end_0 = const()[name = tensor<string, []>("op_26369_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_26369_end_mask_0 = const()[name = tensor<string, []>("op_26369_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_26369_cast = slice_by_index(begin = var_26369_begin_0, end = var_26369_end_0, end_mask = var_26369_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26369_cast")];
            tensor<int32, [4]> var_26373_begin_0 = const()[name = tensor<string, []>("op_26373_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_26373_end_0 = const()[name = tensor<string, []>("op_26373_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_26373_end_mask_0 = const()[name = tensor<string, []>("op_26373_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_26373_cast = slice_by_index(begin = var_26373_begin_0, end = var_26373_end_0, end_mask = var_26373_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26373_cast")];
            tensor<int32, [4]> var_26377_begin_0 = const()[name = tensor<string, []>("op_26377_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_26377_end_0 = const()[name = tensor<string, []>("op_26377_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_26377_end_mask_0 = const()[name = tensor<string, []>("op_26377_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_26377_cast = slice_by_index(begin = var_26377_begin_0, end = var_26377_end_0, end_mask = var_26377_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26377_cast")];
            tensor<int32, [4]> var_26381_begin_0 = const()[name = tensor<string, []>("op_26381_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_26381_end_0 = const()[name = tensor<string, []>("op_26381_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_26381_end_mask_0 = const()[name = tensor<string, []>("op_26381_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_26381_cast = slice_by_index(begin = var_26381_begin_0, end = var_26381_end_0, end_mask = var_26381_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26381_cast")];
            tensor<int32, [4]> var_26385_begin_0 = const()[name = tensor<string, []>("op_26385_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_26385_end_0 = const()[name = tensor<string, []>("op_26385_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_26385_end_mask_0 = const()[name = tensor<string, []>("op_26385_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_26385_cast = slice_by_index(begin = var_26385_begin_0, end = var_26385_end_0, end_mask = var_26385_end_mask_0, x = transpose_80)[name = tensor<string, []>("op_26385_cast")];
            tensor<int32, [4]> var_26387_begin_0 = const()[name = tensor<string, []>("op_26387_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_26387_end_0 = const()[name = tensor<string, []>("op_26387_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_26387_end_mask_0 = const()[name = tensor<string, []>("op_26387_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26387_cast = slice_by_index(begin = var_26387_begin_0, end = var_26387_end_0, end_mask = var_26387_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26387_cast")];
            tensor<int32, [4]> var_26391_begin_0 = const()[name = tensor<string, []>("op_26391_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_26391_end_0 = const()[name = tensor<string, []>("op_26391_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_26391_end_mask_0 = const()[name = tensor<string, []>("op_26391_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26391_cast = slice_by_index(begin = var_26391_begin_0, end = var_26391_end_0, end_mask = var_26391_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26391_cast")];
            tensor<int32, [4]> var_26395_begin_0 = const()[name = tensor<string, []>("op_26395_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_26395_end_0 = const()[name = tensor<string, []>("op_26395_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_26395_end_mask_0 = const()[name = tensor<string, []>("op_26395_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26395_cast = slice_by_index(begin = var_26395_begin_0, end = var_26395_end_0, end_mask = var_26395_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26395_cast")];
            tensor<int32, [4]> var_26399_begin_0 = const()[name = tensor<string, []>("op_26399_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_26399_end_0 = const()[name = tensor<string, []>("op_26399_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_26399_end_mask_0 = const()[name = tensor<string, []>("op_26399_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26399_cast = slice_by_index(begin = var_26399_begin_0, end = var_26399_end_0, end_mask = var_26399_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26399_cast")];
            tensor<int32, [4]> var_26403_begin_0 = const()[name = tensor<string, []>("op_26403_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_26403_end_0 = const()[name = tensor<string, []>("op_26403_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_26403_end_mask_0 = const()[name = tensor<string, []>("op_26403_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26403_cast = slice_by_index(begin = var_26403_begin_0, end = var_26403_end_0, end_mask = var_26403_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26403_cast")];
            tensor<int32, [4]> var_26407_begin_0 = const()[name = tensor<string, []>("op_26407_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_26407_end_0 = const()[name = tensor<string, []>("op_26407_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_26407_end_mask_0 = const()[name = tensor<string, []>("op_26407_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26407_cast = slice_by_index(begin = var_26407_begin_0, end = var_26407_end_0, end_mask = var_26407_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26407_cast")];
            tensor<int32, [4]> var_26411_begin_0 = const()[name = tensor<string, []>("op_26411_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_26411_end_0 = const()[name = tensor<string, []>("op_26411_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_26411_end_mask_0 = const()[name = tensor<string, []>("op_26411_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26411_cast = slice_by_index(begin = var_26411_begin_0, end = var_26411_end_0, end_mask = var_26411_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26411_cast")];
            tensor<int32, [4]> var_26415_begin_0 = const()[name = tensor<string, []>("op_26415_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_26415_end_0 = const()[name = tensor<string, []>("op_26415_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_26415_end_mask_0 = const()[name = tensor<string, []>("op_26415_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26415_cast = slice_by_index(begin = var_26415_begin_0, end = var_26415_end_0, end_mask = var_26415_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26415_cast")];
            tensor<int32, [4]> var_26419_begin_0 = const()[name = tensor<string, []>("op_26419_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_26419_end_0 = const()[name = tensor<string, []>("op_26419_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_26419_end_mask_0 = const()[name = tensor<string, []>("op_26419_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26419_cast = slice_by_index(begin = var_26419_begin_0, end = var_26419_end_0, end_mask = var_26419_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26419_cast")];
            tensor<int32, [4]> var_26423_begin_0 = const()[name = tensor<string, []>("op_26423_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_26423_end_0 = const()[name = tensor<string, []>("op_26423_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_26423_end_mask_0 = const()[name = tensor<string, []>("op_26423_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26423_cast = slice_by_index(begin = var_26423_begin_0, end = var_26423_end_0, end_mask = var_26423_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26423_cast")];
            tensor<int32, [4]> var_26427_begin_0 = const()[name = tensor<string, []>("op_26427_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_26427_end_0 = const()[name = tensor<string, []>("op_26427_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_26427_end_mask_0 = const()[name = tensor<string, []>("op_26427_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26427_cast = slice_by_index(begin = var_26427_begin_0, end = var_26427_end_0, end_mask = var_26427_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26427_cast")];
            tensor<int32, [4]> var_26431_begin_0 = const()[name = tensor<string, []>("op_26431_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_26431_end_0 = const()[name = tensor<string, []>("op_26431_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_26431_end_mask_0 = const()[name = tensor<string, []>("op_26431_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26431_cast = slice_by_index(begin = var_26431_begin_0, end = var_26431_end_0, end_mask = var_26431_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26431_cast")];
            tensor<int32, [4]> var_26435_begin_0 = const()[name = tensor<string, []>("op_26435_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_26435_end_0 = const()[name = tensor<string, []>("op_26435_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_26435_end_mask_0 = const()[name = tensor<string, []>("op_26435_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26435_cast = slice_by_index(begin = var_26435_begin_0, end = var_26435_end_0, end_mask = var_26435_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26435_cast")];
            tensor<int32, [4]> var_26439_begin_0 = const()[name = tensor<string, []>("op_26439_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_26439_end_0 = const()[name = tensor<string, []>("op_26439_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_26439_end_mask_0 = const()[name = tensor<string, []>("op_26439_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26439_cast = slice_by_index(begin = var_26439_begin_0, end = var_26439_end_0, end_mask = var_26439_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26439_cast")];
            tensor<int32, [4]> var_26443_begin_0 = const()[name = tensor<string, []>("op_26443_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_26443_end_0 = const()[name = tensor<string, []>("op_26443_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_26443_end_mask_0 = const()[name = tensor<string, []>("op_26443_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26443_cast = slice_by_index(begin = var_26443_begin_0, end = var_26443_end_0, end_mask = var_26443_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26443_cast")];
            tensor<int32, [4]> var_26447_begin_0 = const()[name = tensor<string, []>("op_26447_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_26447_end_0 = const()[name = tensor<string, []>("op_26447_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_26447_end_mask_0 = const()[name = tensor<string, []>("op_26447_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26447_cast = slice_by_index(begin = var_26447_begin_0, end = var_26447_end_0, end_mask = var_26447_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26447_cast")];
            tensor<int32, [4]> var_26451_begin_0 = const()[name = tensor<string, []>("op_26451_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_26451_end_0 = const()[name = tensor<string, []>("op_26451_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_26451_end_mask_0 = const()[name = tensor<string, []>("op_26451_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26451_cast = slice_by_index(begin = var_26451_begin_0, end = var_26451_end_0, end_mask = var_26451_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26451_cast")];
            tensor<int32, [4]> var_26455_begin_0 = const()[name = tensor<string, []>("op_26455_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_26455_end_0 = const()[name = tensor<string, []>("op_26455_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_26455_end_mask_0 = const()[name = tensor<string, []>("op_26455_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26455_cast = slice_by_index(begin = var_26455_begin_0, end = var_26455_end_0, end_mask = var_26455_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26455_cast")];
            tensor<int32, [4]> var_26459_begin_0 = const()[name = tensor<string, []>("op_26459_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_26459_end_0 = const()[name = tensor<string, []>("op_26459_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_26459_end_mask_0 = const()[name = tensor<string, []>("op_26459_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26459_cast = slice_by_index(begin = var_26459_begin_0, end = var_26459_end_0, end_mask = var_26459_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26459_cast")];
            tensor<int32, [4]> var_26463_begin_0 = const()[name = tensor<string, []>("op_26463_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_26463_end_0 = const()[name = tensor<string, []>("op_26463_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_26463_end_mask_0 = const()[name = tensor<string, []>("op_26463_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_26463_cast = slice_by_index(begin = var_26463_begin_0, end = var_26463_end_0, end_mask = var_26463_end_mask_0, x = v_119_cast)[name = tensor<string, []>("op_26463_cast")];
            tensor<string, []> var_26467_equation_0 = const()[name = tensor<string, []>("op_26467_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26467_cast = einsum(equation = var_26467_equation_0, values = (var_26309_cast, var_26226_cast))[name = tensor<string, []>("op_26467_cast")];
            tensor<fp16, []> var_26468_to_fp16 = const()[name = tensor<string, []>("op_26468_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2201_cast = mul(x = var_26467_cast, y = var_26468_to_fp16)[name = tensor<string, []>("aw_2201_cast")];
            tensor<string, []> var_26471_equation_0 = const()[name = tensor<string, []>("op_26471_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26471_cast = einsum(equation = var_26471_equation_0, values = (var_26313_cast, var_26230_cast))[name = tensor<string, []>("op_26471_cast")];
            tensor<fp16, []> var_26472_to_fp16 = const()[name = tensor<string, []>("op_26472_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2203_cast = mul(x = var_26471_cast, y = var_26472_to_fp16)[name = tensor<string, []>("aw_2203_cast")];
            tensor<string, []> var_26475_equation_0 = const()[name = tensor<string, []>("op_26475_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26475_cast = einsum(equation = var_26475_equation_0, values = (var_26317_cast, var_26234_cast))[name = tensor<string, []>("op_26475_cast")];
            tensor<fp16, []> var_26476_to_fp16 = const()[name = tensor<string, []>("op_26476_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2205_cast = mul(x = var_26475_cast, y = var_26476_to_fp16)[name = tensor<string, []>("aw_2205_cast")];
            tensor<string, []> var_26479_equation_0 = const()[name = tensor<string, []>("op_26479_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26479_cast = einsum(equation = var_26479_equation_0, values = (var_26321_cast, var_26238_cast))[name = tensor<string, []>("op_26479_cast")];
            tensor<fp16, []> var_26480_to_fp16 = const()[name = tensor<string, []>("op_26480_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2207_cast = mul(x = var_26479_cast, y = var_26480_to_fp16)[name = tensor<string, []>("aw_2207_cast")];
            tensor<string, []> var_26483_equation_0 = const()[name = tensor<string, []>("op_26483_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26483_cast = einsum(equation = var_26483_equation_0, values = (var_26325_cast, var_26242_cast))[name = tensor<string, []>("op_26483_cast")];
            tensor<fp16, []> var_26484_to_fp16 = const()[name = tensor<string, []>("op_26484_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2209_cast = mul(x = var_26483_cast, y = var_26484_to_fp16)[name = tensor<string, []>("aw_2209_cast")];
            tensor<string, []> var_26487_equation_0 = const()[name = tensor<string, []>("op_26487_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26487_cast = einsum(equation = var_26487_equation_0, values = (var_26329_cast, var_26246_cast))[name = tensor<string, []>("op_26487_cast")];
            tensor<fp16, []> var_26488_to_fp16 = const()[name = tensor<string, []>("op_26488_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2211_cast = mul(x = var_26487_cast, y = var_26488_to_fp16)[name = tensor<string, []>("aw_2211_cast")];
            tensor<string, []> var_26491_equation_0 = const()[name = tensor<string, []>("op_26491_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26491_cast = einsum(equation = var_26491_equation_0, values = (var_26333_cast, var_26250_cast))[name = tensor<string, []>("op_26491_cast")];
            tensor<fp16, []> var_26492_to_fp16 = const()[name = tensor<string, []>("op_26492_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2213_cast = mul(x = var_26491_cast, y = var_26492_to_fp16)[name = tensor<string, []>("aw_2213_cast")];
            tensor<string, []> var_26495_equation_0 = const()[name = tensor<string, []>("op_26495_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26495_cast = einsum(equation = var_26495_equation_0, values = (var_26337_cast, var_26254_cast))[name = tensor<string, []>("op_26495_cast")];
            tensor<fp16, []> var_26496_to_fp16 = const()[name = tensor<string, []>("op_26496_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2215_cast = mul(x = var_26495_cast, y = var_26496_to_fp16)[name = tensor<string, []>("aw_2215_cast")];
            tensor<string, []> var_26499_equation_0 = const()[name = tensor<string, []>("op_26499_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26499_cast = einsum(equation = var_26499_equation_0, values = (var_26341_cast, var_26258_cast))[name = tensor<string, []>("op_26499_cast")];
            tensor<fp16, []> var_26500_to_fp16 = const()[name = tensor<string, []>("op_26500_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2217_cast = mul(x = var_26499_cast, y = var_26500_to_fp16)[name = tensor<string, []>("aw_2217_cast")];
            tensor<string, []> var_26503_equation_0 = const()[name = tensor<string, []>("op_26503_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26503_cast = einsum(equation = var_26503_equation_0, values = (var_26345_cast, var_26262_cast))[name = tensor<string, []>("op_26503_cast")];
            tensor<fp16, []> var_26504_to_fp16 = const()[name = tensor<string, []>("op_26504_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2219_cast = mul(x = var_26503_cast, y = var_26504_to_fp16)[name = tensor<string, []>("aw_2219_cast")];
            tensor<string, []> var_26507_equation_0 = const()[name = tensor<string, []>("op_26507_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26507_cast = einsum(equation = var_26507_equation_0, values = (var_26349_cast, var_26266_cast))[name = tensor<string, []>("op_26507_cast")];
            tensor<fp16, []> var_26508_to_fp16 = const()[name = tensor<string, []>("op_26508_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2221_cast = mul(x = var_26507_cast, y = var_26508_to_fp16)[name = tensor<string, []>("aw_2221_cast")];
            tensor<string, []> var_26511_equation_0 = const()[name = tensor<string, []>("op_26511_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26511_cast = einsum(equation = var_26511_equation_0, values = (var_26353_cast, var_26270_cast))[name = tensor<string, []>("op_26511_cast")];
            tensor<fp16, []> var_26512_to_fp16 = const()[name = tensor<string, []>("op_26512_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2223_cast = mul(x = var_26511_cast, y = var_26512_to_fp16)[name = tensor<string, []>("aw_2223_cast")];
            tensor<string, []> var_26515_equation_0 = const()[name = tensor<string, []>("op_26515_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26515_cast = einsum(equation = var_26515_equation_0, values = (var_26357_cast, var_26274_cast))[name = tensor<string, []>("op_26515_cast")];
            tensor<fp16, []> var_26516_to_fp16 = const()[name = tensor<string, []>("op_26516_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2225_cast = mul(x = var_26515_cast, y = var_26516_to_fp16)[name = tensor<string, []>("aw_2225_cast")];
            tensor<string, []> var_26519_equation_0 = const()[name = tensor<string, []>("op_26519_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26519_cast = einsum(equation = var_26519_equation_0, values = (var_26361_cast, var_26278_cast))[name = tensor<string, []>("op_26519_cast")];
            tensor<fp16, []> var_26520_to_fp16 = const()[name = tensor<string, []>("op_26520_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2227_cast = mul(x = var_26519_cast, y = var_26520_to_fp16)[name = tensor<string, []>("aw_2227_cast")];
            tensor<string, []> var_26523_equation_0 = const()[name = tensor<string, []>("op_26523_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26523_cast = einsum(equation = var_26523_equation_0, values = (var_26365_cast, var_26282_cast))[name = tensor<string, []>("op_26523_cast")];
            tensor<fp16, []> var_26524_to_fp16 = const()[name = tensor<string, []>("op_26524_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2229_cast = mul(x = var_26523_cast, y = var_26524_to_fp16)[name = tensor<string, []>("aw_2229_cast")];
            tensor<string, []> var_26527_equation_0 = const()[name = tensor<string, []>("op_26527_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26527_cast = einsum(equation = var_26527_equation_0, values = (var_26369_cast, var_26286_cast))[name = tensor<string, []>("op_26527_cast")];
            tensor<fp16, []> var_26528_to_fp16 = const()[name = tensor<string, []>("op_26528_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2231_cast = mul(x = var_26527_cast, y = var_26528_to_fp16)[name = tensor<string, []>("aw_2231_cast")];
            tensor<string, []> var_26531_equation_0 = const()[name = tensor<string, []>("op_26531_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26531_cast = einsum(equation = var_26531_equation_0, values = (var_26373_cast, var_26290_cast))[name = tensor<string, []>("op_26531_cast")];
            tensor<fp16, []> var_26532_to_fp16 = const()[name = tensor<string, []>("op_26532_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2233_cast = mul(x = var_26531_cast, y = var_26532_to_fp16)[name = tensor<string, []>("aw_2233_cast")];
            tensor<string, []> var_26535_equation_0 = const()[name = tensor<string, []>("op_26535_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26535_cast = einsum(equation = var_26535_equation_0, values = (var_26377_cast, var_26294_cast))[name = tensor<string, []>("op_26535_cast")];
            tensor<fp16, []> var_26536_to_fp16 = const()[name = tensor<string, []>("op_26536_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2235_cast = mul(x = var_26535_cast, y = var_26536_to_fp16)[name = tensor<string, []>("aw_2235_cast")];
            tensor<string, []> var_26539_equation_0 = const()[name = tensor<string, []>("op_26539_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26539_cast = einsum(equation = var_26539_equation_0, values = (var_26381_cast, var_26298_cast))[name = tensor<string, []>("op_26539_cast")];
            tensor<fp16, []> var_26540_to_fp16 = const()[name = tensor<string, []>("op_26540_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2237_cast = mul(x = var_26539_cast, y = var_26540_to_fp16)[name = tensor<string, []>("aw_2237_cast")];
            tensor<string, []> var_26543_equation_0 = const()[name = tensor<string, []>("op_26543_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_26543_cast = einsum(equation = var_26543_equation_0, values = (var_26385_cast, var_26302_cast))[name = tensor<string, []>("op_26543_cast")];
            tensor<fp16, []> var_26544_to_fp16 = const()[name = tensor<string, []>("op_26544_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2239_cast = mul(x = var_26543_cast, y = var_26544_to_fp16)[name = tensor<string, []>("aw_2239_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26546_cast = softmax(axis = var_21105, x = aw_2201_cast)[name = tensor<string, []>("op_26546_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26547_cast = softmax(axis = var_21105, x = aw_2203_cast)[name = tensor<string, []>("op_26547_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26548_cast = softmax(axis = var_21105, x = aw_2205_cast)[name = tensor<string, []>("op_26548_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26549_cast = softmax(axis = var_21105, x = aw_2207_cast)[name = tensor<string, []>("op_26549_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26550_cast = softmax(axis = var_21105, x = aw_2209_cast)[name = tensor<string, []>("op_26550_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26551_cast = softmax(axis = var_21105, x = aw_2211_cast)[name = tensor<string, []>("op_26551_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26552_cast = softmax(axis = var_21105, x = aw_2213_cast)[name = tensor<string, []>("op_26552_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26553_cast = softmax(axis = var_21105, x = aw_2215_cast)[name = tensor<string, []>("op_26553_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26554_cast = softmax(axis = var_21105, x = aw_2217_cast)[name = tensor<string, []>("op_26554_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26555_cast = softmax(axis = var_21105, x = aw_2219_cast)[name = tensor<string, []>("op_26555_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26556_cast = softmax(axis = var_21105, x = aw_2221_cast)[name = tensor<string, []>("op_26556_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26557_cast = softmax(axis = var_21105, x = aw_2223_cast)[name = tensor<string, []>("op_26557_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26558_cast = softmax(axis = var_21105, x = aw_2225_cast)[name = tensor<string, []>("op_26558_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26559_cast = softmax(axis = var_21105, x = aw_2227_cast)[name = tensor<string, []>("op_26559_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26560_cast = softmax(axis = var_21105, x = aw_2229_cast)[name = tensor<string, []>("op_26560_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26561_cast = softmax(axis = var_21105, x = aw_2231_cast)[name = tensor<string, []>("op_26561_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26562_cast = softmax(axis = var_21105, x = aw_2233_cast)[name = tensor<string, []>("op_26562_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26563_cast = softmax(axis = var_21105, x = aw_2235_cast)[name = tensor<string, []>("op_26563_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26564_cast = softmax(axis = var_21105, x = aw_2237_cast)[name = tensor<string, []>("op_26564_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_26565_cast = softmax(axis = var_21105, x = aw_2239_cast)[name = tensor<string, []>("op_26565_cast")];
            tensor<string, []> var_26567_equation_0 = const()[name = tensor<string, []>("op_26567_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26567_cast = einsum(equation = var_26567_equation_0, values = (var_26387_cast, var_26546_cast))[name = tensor<string, []>("op_26567_cast")];
            tensor<string, []> var_26569_equation_0 = const()[name = tensor<string, []>("op_26569_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26569_cast = einsum(equation = var_26569_equation_0, values = (var_26391_cast, var_26547_cast))[name = tensor<string, []>("op_26569_cast")];
            tensor<string, []> var_26571_equation_0 = const()[name = tensor<string, []>("op_26571_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26571_cast = einsum(equation = var_26571_equation_0, values = (var_26395_cast, var_26548_cast))[name = tensor<string, []>("op_26571_cast")];
            tensor<string, []> var_26573_equation_0 = const()[name = tensor<string, []>("op_26573_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26573_cast = einsum(equation = var_26573_equation_0, values = (var_26399_cast, var_26549_cast))[name = tensor<string, []>("op_26573_cast")];
            tensor<string, []> var_26575_equation_0 = const()[name = tensor<string, []>("op_26575_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26575_cast = einsum(equation = var_26575_equation_0, values = (var_26403_cast, var_26550_cast))[name = tensor<string, []>("op_26575_cast")];
            tensor<string, []> var_26577_equation_0 = const()[name = tensor<string, []>("op_26577_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26577_cast = einsum(equation = var_26577_equation_0, values = (var_26407_cast, var_26551_cast))[name = tensor<string, []>("op_26577_cast")];
            tensor<string, []> var_26579_equation_0 = const()[name = tensor<string, []>("op_26579_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26579_cast = einsum(equation = var_26579_equation_0, values = (var_26411_cast, var_26552_cast))[name = tensor<string, []>("op_26579_cast")];
            tensor<string, []> var_26581_equation_0 = const()[name = tensor<string, []>("op_26581_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26581_cast = einsum(equation = var_26581_equation_0, values = (var_26415_cast, var_26553_cast))[name = tensor<string, []>("op_26581_cast")];
            tensor<string, []> var_26583_equation_0 = const()[name = tensor<string, []>("op_26583_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26583_cast = einsum(equation = var_26583_equation_0, values = (var_26419_cast, var_26554_cast))[name = tensor<string, []>("op_26583_cast")];
            tensor<string, []> var_26585_equation_0 = const()[name = tensor<string, []>("op_26585_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26585_cast = einsum(equation = var_26585_equation_0, values = (var_26423_cast, var_26555_cast))[name = tensor<string, []>("op_26585_cast")];
            tensor<string, []> var_26587_equation_0 = const()[name = tensor<string, []>("op_26587_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26587_cast = einsum(equation = var_26587_equation_0, values = (var_26427_cast, var_26556_cast))[name = tensor<string, []>("op_26587_cast")];
            tensor<string, []> var_26589_equation_0 = const()[name = tensor<string, []>("op_26589_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26589_cast = einsum(equation = var_26589_equation_0, values = (var_26431_cast, var_26557_cast))[name = tensor<string, []>("op_26589_cast")];
            tensor<string, []> var_26591_equation_0 = const()[name = tensor<string, []>("op_26591_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26591_cast = einsum(equation = var_26591_equation_0, values = (var_26435_cast, var_26558_cast))[name = tensor<string, []>("op_26591_cast")];
            tensor<string, []> var_26593_equation_0 = const()[name = tensor<string, []>("op_26593_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26593_cast = einsum(equation = var_26593_equation_0, values = (var_26439_cast, var_26559_cast))[name = tensor<string, []>("op_26593_cast")];
            tensor<string, []> var_26595_equation_0 = const()[name = tensor<string, []>("op_26595_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26595_cast = einsum(equation = var_26595_equation_0, values = (var_26443_cast, var_26560_cast))[name = tensor<string, []>("op_26595_cast")];
            tensor<string, []> var_26597_equation_0 = const()[name = tensor<string, []>("op_26597_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26597_cast = einsum(equation = var_26597_equation_0, values = (var_26447_cast, var_26561_cast))[name = tensor<string, []>("op_26597_cast")];
            tensor<string, []> var_26599_equation_0 = const()[name = tensor<string, []>("op_26599_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26599_cast = einsum(equation = var_26599_equation_0, values = (var_26451_cast, var_26562_cast))[name = tensor<string, []>("op_26599_cast")];
            tensor<string, []> var_26601_equation_0 = const()[name = tensor<string, []>("op_26601_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26601_cast = einsum(equation = var_26601_equation_0, values = (var_26455_cast, var_26563_cast))[name = tensor<string, []>("op_26601_cast")];
            tensor<string, []> var_26603_equation_0 = const()[name = tensor<string, []>("op_26603_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26603_cast = einsum(equation = var_26603_equation_0, values = (var_26459_cast, var_26564_cast))[name = tensor<string, []>("op_26603_cast")];
            tensor<string, []> var_26605_equation_0 = const()[name = tensor<string, []>("op_26605_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_26605_cast = einsum(equation = var_26605_equation_0, values = (var_26463_cast, var_26565_cast))[name = tensor<string, []>("op_26605_cast")];
            tensor<bool, []> input_371_interleave_0 = const()[name = tensor<string, []>("input_371_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_371_cast = concat(axis = var_21105, interleave = input_371_interleave_0, values = (var_26567_cast, var_26569_cast, var_26571_cast, var_26573_cast, var_26575_cast, var_26577_cast, var_26579_cast, var_26581_cast, var_26583_cast, var_26585_cast, var_26587_cast, var_26589_cast, var_26591_cast, var_26593_cast, var_26595_cast, var_26597_cast, var_26599_cast, var_26601_cast, var_26603_cast, var_26605_cast))[name = tensor<string, []>("input_371_cast")];
            tensor<int32, [2]> var_26611 = const()[name = tensor<string, []>("op_26611"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_26613 = const()[name = tensor<string, []>("op_26613"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_26615_pad_type_0 = const()[name = tensor<string, []>("op_26615_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_26615_pad_0 = const()[name = tensor<string, []>("op_26615_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2100557184)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2103834048)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_26615_cast = conv(bias = mid_block_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16, dilations = var_26613, groups = var_21105, pad = var_26615_pad_0, pad_type = var_26615_pad_type_0, strides = var_26611, weight = mid_block_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16, x = input_371_cast)[name = tensor<string, []>("op_26615_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_179_cast = add(x = var_26615_cast, y = inputs_177_cast)[name = tensor<string, []>("inputs_179_cast")];
            tensor<int32, [1]> var_26619 = const()[name = tensor<string, []>("op_26619"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_179_cast = reduce_mean(axes = var_26619, keep_dims = var_21100, x = inputs_179_cast)[name = tensor<string, []>("channels_mean_179_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_179_cast = sub(x = inputs_179_cast, y = channels_mean_179_cast)[name = tensor<string, []>("zero_mean_179_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_179_cast = mul(x = zero_mean_179_cast, y = zero_mean_179_cast)[name = tensor<string, []>("zero_mean_sq_179_cast")];
            tensor<int32, [1]> var_26623 = const()[name = tensor<string, []>("op_26623"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_26624_cast = reduce_mean(axes = var_26623, keep_dims = var_21100, x = zero_mean_sq_179_cast)[name = tensor<string, []>("op_26624_cast")];
            tensor<fp16, []> var_26625_to_fp16 = const()[name = tensor<string, []>("op_26625_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_26626_cast = add(x = var_26624_cast, y = var_26625_to_fp16)[name = tensor<string, []>("op_26626_cast")];
            tensor<fp16, []> denom_179_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_179_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_179_cast = rsqrt(epsilon = denom_179_epsilon_0_to_fp16, x = var_26626_cast)[name = tensor<string, []>("denom_179_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_179_cast = mul(x = zero_mean_179_cast, y = denom_179_cast)[name = tensor<string, []>("out_179_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_26630_to_fp16 = const()[name = tensor<string, []>("op_26630_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2103836672)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_26631_cast = add(x = out_179_cast, y = var_26630_to_fp16)[name = tensor<string, []>("op_26631_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_26633_to_fp16 = const()[name = tensor<string, []>("op_26633_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2103839296)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_373_cast = mul(x = var_26631_cast, y = var_26633_to_fp16)[name = tensor<string, []>("input_373_cast")];
            tensor<int32, [2]> var_26641 = const()[name = tensor<string, []>("op_26641"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_26643 = const()[name = tensor<string, []>("op_26643"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_26645_pad_type_0 = const()[name = tensor<string, []>("op_26645_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_26645_pad_0 = const()[name = tensor<string, []>("op_26645_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2103841920)))];
            tensor<fp16, [10240]> mid_block_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2130056384)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_26645_cast = conv(bias = mid_block_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16, dilations = var_26643, groups = var_21105, pad = var_26645_pad_0, pad_type = var_26645_pad_type_0, strides = var_26641, weight = mid_block_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16, x = input_373_cast)[name = tensor<string, []>("op_26645_cast")];
            tensor<int32, [2]> var_26646_split_sizes_0 = const()[name = tensor<string, []>("op_26646_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_26646_axis_0 = const()[name = tensor<string, []>("op_26646_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_26646_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_26646_cast_1 = split(axis = var_26646_axis_0, split_sizes = var_26646_split_sizes_0, x = var_26645_cast)[name = tensor<string, []>("op_26646_cast")];
            tensor<string, []> var_26648_mode_0 = const()[name = tensor<string, []>("op_26648_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_26648_cast = gelu(mode = var_26648_mode_0, x = var_26646_cast_1)[name = tensor<string, []>("op_26648_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_375_cast = mul(x = var_26646_cast_0, y = var_26648_cast)[name = tensor<string, []>("input_375_cast")];
            tensor<int32, [2]> var_26652 = const()[name = tensor<string, []>("op_26652"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_26654 = const()[name = tensor<string, []>("op_26654"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_26656_pad_type_0 = const()[name = tensor<string, []>("op_26656_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_26656_pad_0 = const()[name = tensor<string, []>("op_26656_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> mid_block_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2130076928)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2143184192)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_26656_cast = conv(bias = mid_block_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16, dilations = var_26654, groups = var_21105, pad = var_26656_pad_0, pad_type = var_26656_pad_type_0, strides = var_26652, weight = mid_block_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16, x = input_375_cast)[name = tensor<string, []>("op_26656_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_181_cast = add(x = var_26656_cast, y = inputs_179_cast)[name = tensor<string, []>("inputs_181_cast")];
            tensor<int32, [1]> var_26666 = const()[name = tensor<string, []>("op_26666"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_181_cast = reduce_mean(axes = var_26666, keep_dims = var_21100, x = inputs_181_cast)[name = tensor<string, []>("channels_mean_181_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_181_cast = sub(x = inputs_181_cast, y = channels_mean_181_cast)[name = tensor<string, []>("zero_mean_181_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_181_cast = mul(x = zero_mean_181_cast, y = zero_mean_181_cast)[name = tensor<string, []>("zero_mean_sq_181_cast")];
            tensor<int32, [1]> var_26670 = const()[name = tensor<string, []>("op_26670"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_26671_cast = reduce_mean(axes = var_26670, keep_dims = var_21100, x = zero_mean_sq_181_cast)[name = tensor<string, []>("op_26671_cast")];
            tensor<fp16, []> var_26672_to_fp16 = const()[name = tensor<string, []>("op_26672_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_26673_cast = add(x = var_26671_cast, y = var_26672_to_fp16)[name = tensor<string, []>("op_26673_cast")];
            tensor<fp16, []> denom_181_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_181_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_181_cast = rsqrt(epsilon = denom_181_epsilon_0_to_fp16, x = var_26673_cast)[name = tensor<string, []>("denom_181_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_181_cast = mul(x = zero_mean_181_cast, y = denom_181_cast)[name = tensor<string, []>("out_181_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_26677_to_fp16 = const()[name = tensor<string, []>("op_26677_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2143186816)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_26678_cast = add(x = out_181_cast, y = var_26677_to_fp16)[name = tensor<string, []>("op_26678_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_26680_to_fp16 = const()[name = tensor<string, []>("op_26680_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2143189440)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_245_cast = mul(x = var_26678_cast, y = var_26680_to_fp16)[name = tensor<string, []>("hidden_states_245_cast")];
            tensor<int32, [2]> var_26687 = const()[name = tensor<string, []>("op_26687"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_26689 = const()[name = tensor<string, []>("op_26689"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_121_pad_type_0 = const()[name = tensor<string, []>("q_121_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_121_pad_0 = const()[name = tensor<string, []>("q_121_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2143192064)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_121_cast = conv(dilations = var_26689, groups = var_21105, pad = q_121_pad_0, pad_type = q_121_pad_type_0, strides = var_26687, weight = mid_block_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16, x = hidden_states_245_cast)[name = tensor<string, []>("q_121_cast")];
            tensor<int32, [2]> var_26693 = const()[name = tensor<string, []>("op_26693"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_26695 = const()[name = tensor<string, []>("op_26695"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_241_pad_type_0 = const()[name = tensor<string, []>("k_241_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_241_pad_0 = const()[name = tensor<string, []>("k_241_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2146468928)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_241_cast = conv(dilations = var_26695, groups = var_21105, pad = k_241_pad_0, pad_type = k_241_pad_type_0, strides = var_26693, weight = mid_block_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16, x = hidden_states_245_cast)[name = tensor<string, []>("k_241_cast")];
            tensor<int32, [2]> var_26699 = const()[name = tensor<string, []>("op_26699"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_26701 = const()[name = tensor<string, []>("op_26701"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_121_pad_type_0 = const()[name = tensor<string, []>("v_121_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_121_pad_0 = const()[name = tensor<string, []>("v_121_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2149745792)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_121_cast = conv(dilations = var_26701, groups = var_21105, pad = v_121_pad_0, pad_type = v_121_pad_type_0, strides = var_26699, weight = mid_block_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16, x = hidden_states_245_cast)[name = tensor<string, []>("v_121_cast")];
            tensor<int32, [4]> var_26705_begin_0 = const()[name = tensor<string, []>("op_26705_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_26705_end_0 = const()[name = tensor<string, []>("op_26705_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_26705_end_mask_0 = const()[name = tensor<string, []>("op_26705_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26705_cast = slice_by_index(begin = var_26705_begin_0, end = var_26705_end_0, end_mask = var_26705_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26705_cast")];
            tensor<int32, [4]> var_26709_begin_0 = const()[name = tensor<string, []>("op_26709_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_26709_end_0 = const()[name = tensor<string, []>("op_26709_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_26709_end_mask_0 = const()[name = tensor<string, []>("op_26709_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26709_cast = slice_by_index(begin = var_26709_begin_0, end = var_26709_end_0, end_mask = var_26709_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26709_cast")];
            tensor<int32, [4]> var_26713_begin_0 = const()[name = tensor<string, []>("op_26713_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_26713_end_0 = const()[name = tensor<string, []>("op_26713_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_26713_end_mask_0 = const()[name = tensor<string, []>("op_26713_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26713_cast = slice_by_index(begin = var_26713_begin_0, end = var_26713_end_0, end_mask = var_26713_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26713_cast")];
            tensor<int32, [4]> var_26717_begin_0 = const()[name = tensor<string, []>("op_26717_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_26717_end_0 = const()[name = tensor<string, []>("op_26717_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_26717_end_mask_0 = const()[name = tensor<string, []>("op_26717_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26717_cast = slice_by_index(begin = var_26717_begin_0, end = var_26717_end_0, end_mask = var_26717_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26717_cast")];
            tensor<int32, [4]> var_26721_begin_0 = const()[name = tensor<string, []>("op_26721_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_26721_end_0 = const()[name = tensor<string, []>("op_26721_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_26721_end_mask_0 = const()[name = tensor<string, []>("op_26721_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26721_cast = slice_by_index(begin = var_26721_begin_0, end = var_26721_end_0, end_mask = var_26721_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26721_cast")];
            tensor<int32, [4]> var_26725_begin_0 = const()[name = tensor<string, []>("op_26725_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_26725_end_0 = const()[name = tensor<string, []>("op_26725_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_26725_end_mask_0 = const()[name = tensor<string, []>("op_26725_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26725_cast = slice_by_index(begin = var_26725_begin_0, end = var_26725_end_0, end_mask = var_26725_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26725_cast")];
            tensor<int32, [4]> var_26729_begin_0 = const()[name = tensor<string, []>("op_26729_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_26729_end_0 = const()[name = tensor<string, []>("op_26729_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_26729_end_mask_0 = const()[name = tensor<string, []>("op_26729_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26729_cast = slice_by_index(begin = var_26729_begin_0, end = var_26729_end_0, end_mask = var_26729_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26729_cast")];
            tensor<int32, [4]> var_26733_begin_0 = const()[name = tensor<string, []>("op_26733_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_26733_end_0 = const()[name = tensor<string, []>("op_26733_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_26733_end_mask_0 = const()[name = tensor<string, []>("op_26733_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26733_cast = slice_by_index(begin = var_26733_begin_0, end = var_26733_end_0, end_mask = var_26733_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26733_cast")];
            tensor<int32, [4]> var_26737_begin_0 = const()[name = tensor<string, []>("op_26737_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_26737_end_0 = const()[name = tensor<string, []>("op_26737_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_26737_end_mask_0 = const()[name = tensor<string, []>("op_26737_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26737_cast = slice_by_index(begin = var_26737_begin_0, end = var_26737_end_0, end_mask = var_26737_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26737_cast")];
            tensor<int32, [4]> var_26741_begin_0 = const()[name = tensor<string, []>("op_26741_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_26741_end_0 = const()[name = tensor<string, []>("op_26741_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_26741_end_mask_0 = const()[name = tensor<string, []>("op_26741_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26741_cast = slice_by_index(begin = var_26741_begin_0, end = var_26741_end_0, end_mask = var_26741_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26741_cast")];
            tensor<int32, [4]> var_26745_begin_0 = const()[name = tensor<string, []>("op_26745_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_26745_end_0 = const()[name = tensor<string, []>("op_26745_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_26745_end_mask_0 = const()[name = tensor<string, []>("op_26745_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26745_cast = slice_by_index(begin = var_26745_begin_0, end = var_26745_end_0, end_mask = var_26745_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26745_cast")];
            tensor<int32, [4]> var_26749_begin_0 = const()[name = tensor<string, []>("op_26749_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_26749_end_0 = const()[name = tensor<string, []>("op_26749_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_26749_end_mask_0 = const()[name = tensor<string, []>("op_26749_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26749_cast = slice_by_index(begin = var_26749_begin_0, end = var_26749_end_0, end_mask = var_26749_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26749_cast")];
            tensor<int32, [4]> var_26753_begin_0 = const()[name = tensor<string, []>("op_26753_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_26753_end_0 = const()[name = tensor<string, []>("op_26753_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_26753_end_mask_0 = const()[name = tensor<string, []>("op_26753_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26753_cast = slice_by_index(begin = var_26753_begin_0, end = var_26753_end_0, end_mask = var_26753_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26753_cast")];
            tensor<int32, [4]> var_26757_begin_0 = const()[name = tensor<string, []>("op_26757_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_26757_end_0 = const()[name = tensor<string, []>("op_26757_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_26757_end_mask_0 = const()[name = tensor<string, []>("op_26757_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26757_cast = slice_by_index(begin = var_26757_begin_0, end = var_26757_end_0, end_mask = var_26757_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26757_cast")];
            tensor<int32, [4]> var_26761_begin_0 = const()[name = tensor<string, []>("op_26761_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_26761_end_0 = const()[name = tensor<string, []>("op_26761_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_26761_end_mask_0 = const()[name = tensor<string, []>("op_26761_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26761_cast = slice_by_index(begin = var_26761_begin_0, end = var_26761_end_0, end_mask = var_26761_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26761_cast")];
            tensor<int32, [4]> var_26765_begin_0 = const()[name = tensor<string, []>("op_26765_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_26765_end_0 = const()[name = tensor<string, []>("op_26765_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_26765_end_mask_0 = const()[name = tensor<string, []>("op_26765_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26765_cast = slice_by_index(begin = var_26765_begin_0, end = var_26765_end_0, end_mask = var_26765_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26765_cast")];
            tensor<int32, [4]> var_26769_begin_0 = const()[name = tensor<string, []>("op_26769_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_26769_end_0 = const()[name = tensor<string, []>("op_26769_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_26769_end_mask_0 = const()[name = tensor<string, []>("op_26769_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26769_cast = slice_by_index(begin = var_26769_begin_0, end = var_26769_end_0, end_mask = var_26769_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26769_cast")];
            tensor<int32, [4]> var_26773_begin_0 = const()[name = tensor<string, []>("op_26773_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_26773_end_0 = const()[name = tensor<string, []>("op_26773_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_26773_end_mask_0 = const()[name = tensor<string, []>("op_26773_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26773_cast = slice_by_index(begin = var_26773_begin_0, end = var_26773_end_0, end_mask = var_26773_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26773_cast")];
            tensor<int32, [4]> var_26777_begin_0 = const()[name = tensor<string, []>("op_26777_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_26777_end_0 = const()[name = tensor<string, []>("op_26777_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_26777_end_mask_0 = const()[name = tensor<string, []>("op_26777_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26777_cast = slice_by_index(begin = var_26777_begin_0, end = var_26777_end_0, end_mask = var_26777_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26777_cast")];
            tensor<int32, [4]> var_26781_begin_0 = const()[name = tensor<string, []>("op_26781_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_26781_end_0 = const()[name = tensor<string, []>("op_26781_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_26781_end_mask_0 = const()[name = tensor<string, []>("op_26781_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26781_cast = slice_by_index(begin = var_26781_begin_0, end = var_26781_end_0, end_mask = var_26781_end_mask_0, x = q_121_cast)[name = tensor<string, []>("op_26781_cast")];
            tensor<int32, [4]> k_243_perm_0 = const()[name = tensor<string, []>("k_243_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_26788_begin_0 = const()[name = tensor<string, []>("op_26788_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_26788_end_0 = const()[name = tensor<string, []>("op_26788_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_26788_end_mask_0 = const()[name = tensor<string, []>("op_26788_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_79 = transpose(perm = k_243_perm_0, x = k_241_cast)[name = tensor<string, []>("transpose_79")];
            tensor<fp16, [2, 1024, 1, 64]> var_26788_cast = slice_by_index(begin = var_26788_begin_0, end = var_26788_end_0, end_mask = var_26788_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26788_cast")];
            tensor<int32, [4]> var_26792_begin_0 = const()[name = tensor<string, []>("op_26792_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_26792_end_0 = const()[name = tensor<string, []>("op_26792_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_26792_end_mask_0 = const()[name = tensor<string, []>("op_26792_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_26792_cast = slice_by_index(begin = var_26792_begin_0, end = var_26792_end_0, end_mask = var_26792_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26792_cast")];
            tensor<int32, [4]> var_26796_begin_0 = const()[name = tensor<string, []>("op_26796_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_26796_end_0 = const()[name = tensor<string, []>("op_26796_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_26796_end_mask_0 = const()[name = tensor<string, []>("op_26796_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_26796_cast = slice_by_index(begin = var_26796_begin_0, end = var_26796_end_0, end_mask = var_26796_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26796_cast")];
            tensor<int32, [4]> var_26800_begin_0 = const()[name = tensor<string, []>("op_26800_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_26800_end_0 = const()[name = tensor<string, []>("op_26800_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_26800_end_mask_0 = const()[name = tensor<string, []>("op_26800_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_26800_cast = slice_by_index(begin = var_26800_begin_0, end = var_26800_end_0, end_mask = var_26800_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26800_cast")];
            tensor<int32, [4]> var_26804_begin_0 = const()[name = tensor<string, []>("op_26804_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_26804_end_0 = const()[name = tensor<string, []>("op_26804_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_26804_end_mask_0 = const()[name = tensor<string, []>("op_26804_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_26804_cast = slice_by_index(begin = var_26804_begin_0, end = var_26804_end_0, end_mask = var_26804_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26804_cast")];
            tensor<int32, [4]> var_26808_begin_0 = const()[name = tensor<string, []>("op_26808_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_26808_end_0 = const()[name = tensor<string, []>("op_26808_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_26808_end_mask_0 = const()[name = tensor<string, []>("op_26808_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_26808_cast = slice_by_index(begin = var_26808_begin_0, end = var_26808_end_0, end_mask = var_26808_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26808_cast")];
            tensor<int32, [4]> var_26812_begin_0 = const()[name = tensor<string, []>("op_26812_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_26812_end_0 = const()[name = tensor<string, []>("op_26812_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_26812_end_mask_0 = const()[name = tensor<string, []>("op_26812_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_26812_cast = slice_by_index(begin = var_26812_begin_0, end = var_26812_end_0, end_mask = var_26812_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26812_cast")];
            tensor<int32, [4]> var_26816_begin_0 = const()[name = tensor<string, []>("op_26816_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_26816_end_0 = const()[name = tensor<string, []>("op_26816_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_26816_end_mask_0 = const()[name = tensor<string, []>("op_26816_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_26816_cast = slice_by_index(begin = var_26816_begin_0, end = var_26816_end_0, end_mask = var_26816_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26816_cast")];
            tensor<int32, [4]> var_26820_begin_0 = const()[name = tensor<string, []>("op_26820_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_26820_end_0 = const()[name = tensor<string, []>("op_26820_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_26820_end_mask_0 = const()[name = tensor<string, []>("op_26820_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_26820_cast = slice_by_index(begin = var_26820_begin_0, end = var_26820_end_0, end_mask = var_26820_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26820_cast")];
            tensor<int32, [4]> var_26824_begin_0 = const()[name = tensor<string, []>("op_26824_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_26824_end_0 = const()[name = tensor<string, []>("op_26824_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_26824_end_mask_0 = const()[name = tensor<string, []>("op_26824_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_26824_cast = slice_by_index(begin = var_26824_begin_0, end = var_26824_end_0, end_mask = var_26824_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26824_cast")];
            tensor<int32, [4]> var_26828_begin_0 = const()[name = tensor<string, []>("op_26828_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_26828_end_0 = const()[name = tensor<string, []>("op_26828_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_26828_end_mask_0 = const()[name = tensor<string, []>("op_26828_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_26828_cast = slice_by_index(begin = var_26828_begin_0, end = var_26828_end_0, end_mask = var_26828_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26828_cast")];
            tensor<int32, [4]> var_26832_begin_0 = const()[name = tensor<string, []>("op_26832_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_26832_end_0 = const()[name = tensor<string, []>("op_26832_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_26832_end_mask_0 = const()[name = tensor<string, []>("op_26832_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_26832_cast = slice_by_index(begin = var_26832_begin_0, end = var_26832_end_0, end_mask = var_26832_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26832_cast")];
            tensor<int32, [4]> var_26836_begin_0 = const()[name = tensor<string, []>("op_26836_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_26836_end_0 = const()[name = tensor<string, []>("op_26836_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_26836_end_mask_0 = const()[name = tensor<string, []>("op_26836_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_26836_cast = slice_by_index(begin = var_26836_begin_0, end = var_26836_end_0, end_mask = var_26836_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26836_cast")];
            tensor<int32, [4]> var_26840_begin_0 = const()[name = tensor<string, []>("op_26840_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_26840_end_0 = const()[name = tensor<string, []>("op_26840_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_26840_end_mask_0 = const()[name = tensor<string, []>("op_26840_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_26840_cast = slice_by_index(begin = var_26840_begin_0, end = var_26840_end_0, end_mask = var_26840_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26840_cast")];
            tensor<int32, [4]> var_26844_begin_0 = const()[name = tensor<string, []>("op_26844_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_26844_end_0 = const()[name = tensor<string, []>("op_26844_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_26844_end_mask_0 = const()[name = tensor<string, []>("op_26844_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_26844_cast = slice_by_index(begin = var_26844_begin_0, end = var_26844_end_0, end_mask = var_26844_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26844_cast")];
            tensor<int32, [4]> var_26848_begin_0 = const()[name = tensor<string, []>("op_26848_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_26848_end_0 = const()[name = tensor<string, []>("op_26848_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_26848_end_mask_0 = const()[name = tensor<string, []>("op_26848_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_26848_cast = slice_by_index(begin = var_26848_begin_0, end = var_26848_end_0, end_mask = var_26848_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26848_cast")];
            tensor<int32, [4]> var_26852_begin_0 = const()[name = tensor<string, []>("op_26852_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_26852_end_0 = const()[name = tensor<string, []>("op_26852_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_26852_end_mask_0 = const()[name = tensor<string, []>("op_26852_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_26852_cast = slice_by_index(begin = var_26852_begin_0, end = var_26852_end_0, end_mask = var_26852_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26852_cast")];
            tensor<int32, [4]> var_26856_begin_0 = const()[name = tensor<string, []>("op_26856_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_26856_end_0 = const()[name = tensor<string, []>("op_26856_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_26856_end_mask_0 = const()[name = tensor<string, []>("op_26856_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_26856_cast = slice_by_index(begin = var_26856_begin_0, end = var_26856_end_0, end_mask = var_26856_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26856_cast")];
            tensor<int32, [4]> var_26860_begin_0 = const()[name = tensor<string, []>("op_26860_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_26860_end_0 = const()[name = tensor<string, []>("op_26860_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_26860_end_mask_0 = const()[name = tensor<string, []>("op_26860_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_26860_cast = slice_by_index(begin = var_26860_begin_0, end = var_26860_end_0, end_mask = var_26860_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26860_cast")];
            tensor<int32, [4]> var_26864_begin_0 = const()[name = tensor<string, []>("op_26864_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_26864_end_0 = const()[name = tensor<string, []>("op_26864_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_26864_end_mask_0 = const()[name = tensor<string, []>("op_26864_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_26864_cast = slice_by_index(begin = var_26864_begin_0, end = var_26864_end_0, end_mask = var_26864_end_mask_0, x = transpose_79)[name = tensor<string, []>("op_26864_cast")];
            tensor<int32, [4]> var_26866_begin_0 = const()[name = tensor<string, []>("op_26866_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_26866_end_0 = const()[name = tensor<string, []>("op_26866_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_26866_end_mask_0 = const()[name = tensor<string, []>("op_26866_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26866_cast = slice_by_index(begin = var_26866_begin_0, end = var_26866_end_0, end_mask = var_26866_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26866_cast")];
            tensor<int32, [4]> var_26870_begin_0 = const()[name = tensor<string, []>("op_26870_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_26870_end_0 = const()[name = tensor<string, []>("op_26870_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_26870_end_mask_0 = const()[name = tensor<string, []>("op_26870_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26870_cast = slice_by_index(begin = var_26870_begin_0, end = var_26870_end_0, end_mask = var_26870_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26870_cast")];
            tensor<int32, [4]> var_26874_begin_0 = const()[name = tensor<string, []>("op_26874_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_26874_end_0 = const()[name = tensor<string, []>("op_26874_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_26874_end_mask_0 = const()[name = tensor<string, []>("op_26874_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26874_cast = slice_by_index(begin = var_26874_begin_0, end = var_26874_end_0, end_mask = var_26874_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26874_cast")];
            tensor<int32, [4]> var_26878_begin_0 = const()[name = tensor<string, []>("op_26878_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_26878_end_0 = const()[name = tensor<string, []>("op_26878_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_26878_end_mask_0 = const()[name = tensor<string, []>("op_26878_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26878_cast = slice_by_index(begin = var_26878_begin_0, end = var_26878_end_0, end_mask = var_26878_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26878_cast")];
            tensor<int32, [4]> var_26882_begin_0 = const()[name = tensor<string, []>("op_26882_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_26882_end_0 = const()[name = tensor<string, []>("op_26882_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_26882_end_mask_0 = const()[name = tensor<string, []>("op_26882_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26882_cast = slice_by_index(begin = var_26882_begin_0, end = var_26882_end_0, end_mask = var_26882_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26882_cast")];
            tensor<int32, [4]> var_26886_begin_0 = const()[name = tensor<string, []>("op_26886_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_26886_end_0 = const()[name = tensor<string, []>("op_26886_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_26886_end_mask_0 = const()[name = tensor<string, []>("op_26886_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26886_cast = slice_by_index(begin = var_26886_begin_0, end = var_26886_end_0, end_mask = var_26886_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26886_cast")];
            tensor<int32, [4]> var_26890_begin_0 = const()[name = tensor<string, []>("op_26890_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_26890_end_0 = const()[name = tensor<string, []>("op_26890_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_26890_end_mask_0 = const()[name = tensor<string, []>("op_26890_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26890_cast = slice_by_index(begin = var_26890_begin_0, end = var_26890_end_0, end_mask = var_26890_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26890_cast")];
            tensor<int32, [4]> var_26894_begin_0 = const()[name = tensor<string, []>("op_26894_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_26894_end_0 = const()[name = tensor<string, []>("op_26894_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_26894_end_mask_0 = const()[name = tensor<string, []>("op_26894_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26894_cast = slice_by_index(begin = var_26894_begin_0, end = var_26894_end_0, end_mask = var_26894_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26894_cast")];
            tensor<int32, [4]> var_26898_begin_0 = const()[name = tensor<string, []>("op_26898_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_26898_end_0 = const()[name = tensor<string, []>("op_26898_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_26898_end_mask_0 = const()[name = tensor<string, []>("op_26898_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26898_cast = slice_by_index(begin = var_26898_begin_0, end = var_26898_end_0, end_mask = var_26898_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26898_cast")];
            tensor<int32, [4]> var_26902_begin_0 = const()[name = tensor<string, []>("op_26902_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_26902_end_0 = const()[name = tensor<string, []>("op_26902_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_26902_end_mask_0 = const()[name = tensor<string, []>("op_26902_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26902_cast = slice_by_index(begin = var_26902_begin_0, end = var_26902_end_0, end_mask = var_26902_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26902_cast")];
            tensor<int32, [4]> var_26906_begin_0 = const()[name = tensor<string, []>("op_26906_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_26906_end_0 = const()[name = tensor<string, []>("op_26906_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_26906_end_mask_0 = const()[name = tensor<string, []>("op_26906_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26906_cast = slice_by_index(begin = var_26906_begin_0, end = var_26906_end_0, end_mask = var_26906_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26906_cast")];
            tensor<int32, [4]> var_26910_begin_0 = const()[name = tensor<string, []>("op_26910_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_26910_end_0 = const()[name = tensor<string, []>("op_26910_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_26910_end_mask_0 = const()[name = tensor<string, []>("op_26910_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26910_cast = slice_by_index(begin = var_26910_begin_0, end = var_26910_end_0, end_mask = var_26910_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26910_cast")];
            tensor<int32, [4]> var_26914_begin_0 = const()[name = tensor<string, []>("op_26914_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_26914_end_0 = const()[name = tensor<string, []>("op_26914_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_26914_end_mask_0 = const()[name = tensor<string, []>("op_26914_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26914_cast = slice_by_index(begin = var_26914_begin_0, end = var_26914_end_0, end_mask = var_26914_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26914_cast")];
            tensor<int32, [4]> var_26918_begin_0 = const()[name = tensor<string, []>("op_26918_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_26918_end_0 = const()[name = tensor<string, []>("op_26918_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_26918_end_mask_0 = const()[name = tensor<string, []>("op_26918_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26918_cast = slice_by_index(begin = var_26918_begin_0, end = var_26918_end_0, end_mask = var_26918_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26918_cast")];
            tensor<int32, [4]> var_26922_begin_0 = const()[name = tensor<string, []>("op_26922_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_26922_end_0 = const()[name = tensor<string, []>("op_26922_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_26922_end_mask_0 = const()[name = tensor<string, []>("op_26922_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26922_cast = slice_by_index(begin = var_26922_begin_0, end = var_26922_end_0, end_mask = var_26922_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26922_cast")];
            tensor<int32, [4]> var_26926_begin_0 = const()[name = tensor<string, []>("op_26926_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_26926_end_0 = const()[name = tensor<string, []>("op_26926_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_26926_end_mask_0 = const()[name = tensor<string, []>("op_26926_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26926_cast = slice_by_index(begin = var_26926_begin_0, end = var_26926_end_0, end_mask = var_26926_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26926_cast")];
            tensor<int32, [4]> var_26930_begin_0 = const()[name = tensor<string, []>("op_26930_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_26930_end_0 = const()[name = tensor<string, []>("op_26930_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_26930_end_mask_0 = const()[name = tensor<string, []>("op_26930_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26930_cast = slice_by_index(begin = var_26930_begin_0, end = var_26930_end_0, end_mask = var_26930_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26930_cast")];
            tensor<int32, [4]> var_26934_begin_0 = const()[name = tensor<string, []>("op_26934_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_26934_end_0 = const()[name = tensor<string, []>("op_26934_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_26934_end_mask_0 = const()[name = tensor<string, []>("op_26934_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26934_cast = slice_by_index(begin = var_26934_begin_0, end = var_26934_end_0, end_mask = var_26934_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26934_cast")];
            tensor<int32, [4]> var_26938_begin_0 = const()[name = tensor<string, []>("op_26938_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_26938_end_0 = const()[name = tensor<string, []>("op_26938_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_26938_end_mask_0 = const()[name = tensor<string, []>("op_26938_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26938_cast = slice_by_index(begin = var_26938_begin_0, end = var_26938_end_0, end_mask = var_26938_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26938_cast")];
            tensor<int32, [4]> var_26942_begin_0 = const()[name = tensor<string, []>("op_26942_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_26942_end_0 = const()[name = tensor<string, []>("op_26942_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_26942_end_mask_0 = const()[name = tensor<string, []>("op_26942_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_26942_cast = slice_by_index(begin = var_26942_begin_0, end = var_26942_end_0, end_mask = var_26942_end_mask_0, x = v_121_cast)[name = tensor<string, []>("op_26942_cast")];
            tensor<string, []> var_26946_equation_0 = const()[name = tensor<string, []>("op_26946_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26946_cast = einsum(equation = var_26946_equation_0, values = (var_26788_cast, var_26705_cast))[name = tensor<string, []>("op_26946_cast")];
            tensor<fp16, []> var_26947_to_fp16 = const()[name = tensor<string, []>("op_26947_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2241_cast = mul(x = var_26946_cast, y = var_26947_to_fp16)[name = tensor<string, []>("aw_2241_cast")];
            tensor<string, []> var_26950_equation_0 = const()[name = tensor<string, []>("op_26950_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26950_cast = einsum(equation = var_26950_equation_0, values = (var_26792_cast, var_26709_cast))[name = tensor<string, []>("op_26950_cast")];
            tensor<fp16, []> var_26951_to_fp16 = const()[name = tensor<string, []>("op_26951_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2243_cast = mul(x = var_26950_cast, y = var_26951_to_fp16)[name = tensor<string, []>("aw_2243_cast")];
            tensor<string, []> var_26954_equation_0 = const()[name = tensor<string, []>("op_26954_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26954_cast = einsum(equation = var_26954_equation_0, values = (var_26796_cast, var_26713_cast))[name = tensor<string, []>("op_26954_cast")];
            tensor<fp16, []> var_26955_to_fp16 = const()[name = tensor<string, []>("op_26955_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2245_cast = mul(x = var_26954_cast, y = var_26955_to_fp16)[name = tensor<string, []>("aw_2245_cast")];
            tensor<string, []> var_26958_equation_0 = const()[name = tensor<string, []>("op_26958_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26958_cast = einsum(equation = var_26958_equation_0, values = (var_26800_cast, var_26717_cast))[name = tensor<string, []>("op_26958_cast")];
            tensor<fp16, []> var_26959_to_fp16 = const()[name = tensor<string, []>("op_26959_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2247_cast = mul(x = var_26958_cast, y = var_26959_to_fp16)[name = tensor<string, []>("aw_2247_cast")];
            tensor<string, []> var_26962_equation_0 = const()[name = tensor<string, []>("op_26962_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26962_cast = einsum(equation = var_26962_equation_0, values = (var_26804_cast, var_26721_cast))[name = tensor<string, []>("op_26962_cast")];
            tensor<fp16, []> var_26963_to_fp16 = const()[name = tensor<string, []>("op_26963_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2249_cast = mul(x = var_26962_cast, y = var_26963_to_fp16)[name = tensor<string, []>("aw_2249_cast")];
            tensor<string, []> var_26966_equation_0 = const()[name = tensor<string, []>("op_26966_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26966_cast = einsum(equation = var_26966_equation_0, values = (var_26808_cast, var_26725_cast))[name = tensor<string, []>("op_26966_cast")];
            tensor<fp16, []> var_26967_to_fp16 = const()[name = tensor<string, []>("op_26967_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2251_cast = mul(x = var_26966_cast, y = var_26967_to_fp16)[name = tensor<string, []>("aw_2251_cast")];
            tensor<string, []> var_26970_equation_0 = const()[name = tensor<string, []>("op_26970_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26970_cast = einsum(equation = var_26970_equation_0, values = (var_26812_cast, var_26729_cast))[name = tensor<string, []>("op_26970_cast")];
            tensor<fp16, []> var_26971_to_fp16 = const()[name = tensor<string, []>("op_26971_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2253_cast = mul(x = var_26970_cast, y = var_26971_to_fp16)[name = tensor<string, []>("aw_2253_cast")];
            tensor<string, []> var_26974_equation_0 = const()[name = tensor<string, []>("op_26974_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26974_cast = einsum(equation = var_26974_equation_0, values = (var_26816_cast, var_26733_cast))[name = tensor<string, []>("op_26974_cast")];
            tensor<fp16, []> var_26975_to_fp16 = const()[name = tensor<string, []>("op_26975_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2255_cast = mul(x = var_26974_cast, y = var_26975_to_fp16)[name = tensor<string, []>("aw_2255_cast")];
            tensor<string, []> var_26978_equation_0 = const()[name = tensor<string, []>("op_26978_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26978_cast = einsum(equation = var_26978_equation_0, values = (var_26820_cast, var_26737_cast))[name = tensor<string, []>("op_26978_cast")];
            tensor<fp16, []> var_26979_to_fp16 = const()[name = tensor<string, []>("op_26979_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2257_cast = mul(x = var_26978_cast, y = var_26979_to_fp16)[name = tensor<string, []>("aw_2257_cast")];
            tensor<string, []> var_26982_equation_0 = const()[name = tensor<string, []>("op_26982_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26982_cast = einsum(equation = var_26982_equation_0, values = (var_26824_cast, var_26741_cast))[name = tensor<string, []>("op_26982_cast")];
            tensor<fp16, []> var_26983_to_fp16 = const()[name = tensor<string, []>("op_26983_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2259_cast = mul(x = var_26982_cast, y = var_26983_to_fp16)[name = tensor<string, []>("aw_2259_cast")];
            tensor<string, []> var_26986_equation_0 = const()[name = tensor<string, []>("op_26986_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26986_cast = einsum(equation = var_26986_equation_0, values = (var_26828_cast, var_26745_cast))[name = tensor<string, []>("op_26986_cast")];
            tensor<fp16, []> var_26987_to_fp16 = const()[name = tensor<string, []>("op_26987_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2261_cast = mul(x = var_26986_cast, y = var_26987_to_fp16)[name = tensor<string, []>("aw_2261_cast")];
            tensor<string, []> var_26990_equation_0 = const()[name = tensor<string, []>("op_26990_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26990_cast = einsum(equation = var_26990_equation_0, values = (var_26832_cast, var_26749_cast))[name = tensor<string, []>("op_26990_cast")];
            tensor<fp16, []> var_26991_to_fp16 = const()[name = tensor<string, []>("op_26991_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2263_cast = mul(x = var_26990_cast, y = var_26991_to_fp16)[name = tensor<string, []>("aw_2263_cast")];
            tensor<string, []> var_26994_equation_0 = const()[name = tensor<string, []>("op_26994_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26994_cast = einsum(equation = var_26994_equation_0, values = (var_26836_cast, var_26753_cast))[name = tensor<string, []>("op_26994_cast")];
            tensor<fp16, []> var_26995_to_fp16 = const()[name = tensor<string, []>("op_26995_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2265_cast = mul(x = var_26994_cast, y = var_26995_to_fp16)[name = tensor<string, []>("aw_2265_cast")];
            tensor<string, []> var_26998_equation_0 = const()[name = tensor<string, []>("op_26998_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_26998_cast = einsum(equation = var_26998_equation_0, values = (var_26840_cast, var_26757_cast))[name = tensor<string, []>("op_26998_cast")];
            tensor<fp16, []> var_26999_to_fp16 = const()[name = tensor<string, []>("op_26999_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2267_cast = mul(x = var_26998_cast, y = var_26999_to_fp16)[name = tensor<string, []>("aw_2267_cast")];
            tensor<string, []> var_27002_equation_0 = const()[name = tensor<string, []>("op_27002_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27002_cast = einsum(equation = var_27002_equation_0, values = (var_26844_cast, var_26761_cast))[name = tensor<string, []>("op_27002_cast")];
            tensor<fp16, []> var_27003_to_fp16 = const()[name = tensor<string, []>("op_27003_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2269_cast = mul(x = var_27002_cast, y = var_27003_to_fp16)[name = tensor<string, []>("aw_2269_cast")];
            tensor<string, []> var_27006_equation_0 = const()[name = tensor<string, []>("op_27006_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27006_cast = einsum(equation = var_27006_equation_0, values = (var_26848_cast, var_26765_cast))[name = tensor<string, []>("op_27006_cast")];
            tensor<fp16, []> var_27007_to_fp16 = const()[name = tensor<string, []>("op_27007_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2271_cast = mul(x = var_27006_cast, y = var_27007_to_fp16)[name = tensor<string, []>("aw_2271_cast")];
            tensor<string, []> var_27010_equation_0 = const()[name = tensor<string, []>("op_27010_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27010_cast = einsum(equation = var_27010_equation_0, values = (var_26852_cast, var_26769_cast))[name = tensor<string, []>("op_27010_cast")];
            tensor<fp16, []> var_27011_to_fp16 = const()[name = tensor<string, []>("op_27011_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2273_cast = mul(x = var_27010_cast, y = var_27011_to_fp16)[name = tensor<string, []>("aw_2273_cast")];
            tensor<string, []> var_27014_equation_0 = const()[name = tensor<string, []>("op_27014_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27014_cast = einsum(equation = var_27014_equation_0, values = (var_26856_cast, var_26773_cast))[name = tensor<string, []>("op_27014_cast")];
            tensor<fp16, []> var_27015_to_fp16 = const()[name = tensor<string, []>("op_27015_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2275_cast = mul(x = var_27014_cast, y = var_27015_to_fp16)[name = tensor<string, []>("aw_2275_cast")];
            tensor<string, []> var_27018_equation_0 = const()[name = tensor<string, []>("op_27018_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27018_cast = einsum(equation = var_27018_equation_0, values = (var_26860_cast, var_26777_cast))[name = tensor<string, []>("op_27018_cast")];
            tensor<fp16, []> var_27019_to_fp16 = const()[name = tensor<string, []>("op_27019_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2277_cast = mul(x = var_27018_cast, y = var_27019_to_fp16)[name = tensor<string, []>("aw_2277_cast")];
            tensor<string, []> var_27022_equation_0 = const()[name = tensor<string, []>("op_27022_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27022_cast = einsum(equation = var_27022_equation_0, values = (var_26864_cast, var_26781_cast))[name = tensor<string, []>("op_27022_cast")];
            tensor<fp16, []> var_27023_to_fp16 = const()[name = tensor<string, []>("op_27023_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2279_cast = mul(x = var_27022_cast, y = var_27023_to_fp16)[name = tensor<string, []>("aw_2279_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27025_cast = softmax(axis = var_21105, x = aw_2241_cast)[name = tensor<string, []>("op_27025_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27026_cast = softmax(axis = var_21105, x = aw_2243_cast)[name = tensor<string, []>("op_27026_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27027_cast = softmax(axis = var_21105, x = aw_2245_cast)[name = tensor<string, []>("op_27027_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27028_cast = softmax(axis = var_21105, x = aw_2247_cast)[name = tensor<string, []>("op_27028_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27029_cast = softmax(axis = var_21105, x = aw_2249_cast)[name = tensor<string, []>("op_27029_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27030_cast = softmax(axis = var_21105, x = aw_2251_cast)[name = tensor<string, []>("op_27030_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27031_cast = softmax(axis = var_21105, x = aw_2253_cast)[name = tensor<string, []>("op_27031_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27032_cast = softmax(axis = var_21105, x = aw_2255_cast)[name = tensor<string, []>("op_27032_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27033_cast = softmax(axis = var_21105, x = aw_2257_cast)[name = tensor<string, []>("op_27033_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27034_cast = softmax(axis = var_21105, x = aw_2259_cast)[name = tensor<string, []>("op_27034_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27035_cast = softmax(axis = var_21105, x = aw_2261_cast)[name = tensor<string, []>("op_27035_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27036_cast = softmax(axis = var_21105, x = aw_2263_cast)[name = tensor<string, []>("op_27036_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27037_cast = softmax(axis = var_21105, x = aw_2265_cast)[name = tensor<string, []>("op_27037_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27038_cast = softmax(axis = var_21105, x = aw_2267_cast)[name = tensor<string, []>("op_27038_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27039_cast = softmax(axis = var_21105, x = aw_2269_cast)[name = tensor<string, []>("op_27039_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27040_cast = softmax(axis = var_21105, x = aw_2271_cast)[name = tensor<string, []>("op_27040_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27041_cast = softmax(axis = var_21105, x = aw_2273_cast)[name = tensor<string, []>("op_27041_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27042_cast = softmax(axis = var_21105, x = aw_2275_cast)[name = tensor<string, []>("op_27042_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27043_cast = softmax(axis = var_21105, x = aw_2277_cast)[name = tensor<string, []>("op_27043_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27044_cast = softmax(axis = var_21105, x = aw_2279_cast)[name = tensor<string, []>("op_27044_cast")];
            tensor<string, []> var_27046_equation_0 = const()[name = tensor<string, []>("op_27046_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27046_cast = einsum(equation = var_27046_equation_0, values = (var_26866_cast, var_27025_cast))[name = tensor<string, []>("op_27046_cast")];
            tensor<string, []> var_27048_equation_0 = const()[name = tensor<string, []>("op_27048_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27048_cast = einsum(equation = var_27048_equation_0, values = (var_26870_cast, var_27026_cast))[name = tensor<string, []>("op_27048_cast")];
            tensor<string, []> var_27050_equation_0 = const()[name = tensor<string, []>("op_27050_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27050_cast = einsum(equation = var_27050_equation_0, values = (var_26874_cast, var_27027_cast))[name = tensor<string, []>("op_27050_cast")];
            tensor<string, []> var_27052_equation_0 = const()[name = tensor<string, []>("op_27052_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27052_cast = einsum(equation = var_27052_equation_0, values = (var_26878_cast, var_27028_cast))[name = tensor<string, []>("op_27052_cast")];
            tensor<string, []> var_27054_equation_0 = const()[name = tensor<string, []>("op_27054_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27054_cast = einsum(equation = var_27054_equation_0, values = (var_26882_cast, var_27029_cast))[name = tensor<string, []>("op_27054_cast")];
            tensor<string, []> var_27056_equation_0 = const()[name = tensor<string, []>("op_27056_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27056_cast = einsum(equation = var_27056_equation_0, values = (var_26886_cast, var_27030_cast))[name = tensor<string, []>("op_27056_cast")];
            tensor<string, []> var_27058_equation_0 = const()[name = tensor<string, []>("op_27058_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27058_cast = einsum(equation = var_27058_equation_0, values = (var_26890_cast, var_27031_cast))[name = tensor<string, []>("op_27058_cast")];
            tensor<string, []> var_27060_equation_0 = const()[name = tensor<string, []>("op_27060_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27060_cast = einsum(equation = var_27060_equation_0, values = (var_26894_cast, var_27032_cast))[name = tensor<string, []>("op_27060_cast")];
            tensor<string, []> var_27062_equation_0 = const()[name = tensor<string, []>("op_27062_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27062_cast = einsum(equation = var_27062_equation_0, values = (var_26898_cast, var_27033_cast))[name = tensor<string, []>("op_27062_cast")];
            tensor<string, []> var_27064_equation_0 = const()[name = tensor<string, []>("op_27064_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27064_cast = einsum(equation = var_27064_equation_0, values = (var_26902_cast, var_27034_cast))[name = tensor<string, []>("op_27064_cast")];
            tensor<string, []> var_27066_equation_0 = const()[name = tensor<string, []>("op_27066_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27066_cast = einsum(equation = var_27066_equation_0, values = (var_26906_cast, var_27035_cast))[name = tensor<string, []>("op_27066_cast")];
            tensor<string, []> var_27068_equation_0 = const()[name = tensor<string, []>("op_27068_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27068_cast = einsum(equation = var_27068_equation_0, values = (var_26910_cast, var_27036_cast))[name = tensor<string, []>("op_27068_cast")];
            tensor<string, []> var_27070_equation_0 = const()[name = tensor<string, []>("op_27070_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27070_cast = einsum(equation = var_27070_equation_0, values = (var_26914_cast, var_27037_cast))[name = tensor<string, []>("op_27070_cast")];
            tensor<string, []> var_27072_equation_0 = const()[name = tensor<string, []>("op_27072_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27072_cast = einsum(equation = var_27072_equation_0, values = (var_26918_cast, var_27038_cast))[name = tensor<string, []>("op_27072_cast")];
            tensor<string, []> var_27074_equation_0 = const()[name = tensor<string, []>("op_27074_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27074_cast = einsum(equation = var_27074_equation_0, values = (var_26922_cast, var_27039_cast))[name = tensor<string, []>("op_27074_cast")];
            tensor<string, []> var_27076_equation_0 = const()[name = tensor<string, []>("op_27076_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27076_cast = einsum(equation = var_27076_equation_0, values = (var_26926_cast, var_27040_cast))[name = tensor<string, []>("op_27076_cast")];
            tensor<string, []> var_27078_equation_0 = const()[name = tensor<string, []>("op_27078_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27078_cast = einsum(equation = var_27078_equation_0, values = (var_26930_cast, var_27041_cast))[name = tensor<string, []>("op_27078_cast")];
            tensor<string, []> var_27080_equation_0 = const()[name = tensor<string, []>("op_27080_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27080_cast = einsum(equation = var_27080_equation_0, values = (var_26934_cast, var_27042_cast))[name = tensor<string, []>("op_27080_cast")];
            tensor<string, []> var_27082_equation_0 = const()[name = tensor<string, []>("op_27082_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27082_cast = einsum(equation = var_27082_equation_0, values = (var_26938_cast, var_27043_cast))[name = tensor<string, []>("op_27082_cast")];
            tensor<string, []> var_27084_equation_0 = const()[name = tensor<string, []>("op_27084_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27084_cast = einsum(equation = var_27084_equation_0, values = (var_26942_cast, var_27044_cast))[name = tensor<string, []>("op_27084_cast")];
            tensor<bool, []> input_377_interleave_0 = const()[name = tensor<string, []>("input_377_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_377_cast = concat(axis = var_21105, interleave = input_377_interleave_0, values = (var_27046_cast, var_27048_cast, var_27050_cast, var_27052_cast, var_27054_cast, var_27056_cast, var_27058_cast, var_27060_cast, var_27062_cast, var_27064_cast, var_27066_cast, var_27068_cast, var_27070_cast, var_27072_cast, var_27074_cast, var_27076_cast, var_27078_cast, var_27080_cast, var_27082_cast, var_27084_cast))[name = tensor<string, []>("input_377_cast")];
            tensor<int32, [2]> var_27090 = const()[name = tensor<string, []>("op_27090"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_27092 = const()[name = tensor<string, []>("op_27092"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_27094_pad_type_0 = const()[name = tensor<string, []>("op_27094_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_27094_pad_0 = const()[name = tensor<string, []>("op_27094_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2153022656)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2156299520)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_27094_cast = conv(bias = mid_block_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16, dilations = var_27092, groups = var_21105, pad = var_27094_pad_0, pad_type = var_27094_pad_type_0, strides = var_27090, weight = mid_block_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16, x = input_377_cast)[name = tensor<string, []>("op_27094_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_183_cast = add(x = var_27094_cast, y = inputs_181_cast)[name = tensor<string, []>("inputs_183_cast")];
            tensor<int32, [1]> var_27098 = const()[name = tensor<string, []>("op_27098"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_183_cast = reduce_mean(axes = var_27098, keep_dims = var_21100, x = inputs_183_cast)[name = tensor<string, []>("channels_mean_183_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_183_cast = sub(x = inputs_183_cast, y = channels_mean_183_cast)[name = tensor<string, []>("zero_mean_183_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_183_cast = mul(x = zero_mean_183_cast, y = zero_mean_183_cast)[name = tensor<string, []>("zero_mean_sq_183_cast")];
            tensor<int32, [1]> var_27102 = const()[name = tensor<string, []>("op_27102"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_27103_cast = reduce_mean(axes = var_27102, keep_dims = var_21100, x = zero_mean_sq_183_cast)[name = tensor<string, []>("op_27103_cast")];
            tensor<fp16, []> var_27104_to_fp16 = const()[name = tensor<string, []>("op_27104_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_27105_cast = add(x = var_27103_cast, y = var_27104_to_fp16)[name = tensor<string, []>("op_27105_cast")];
            tensor<fp16, []> denom_183_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_183_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_183_cast = rsqrt(epsilon = denom_183_epsilon_0_to_fp16, x = var_27105_cast)[name = tensor<string, []>("denom_183_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_183_cast = mul(x = zero_mean_183_cast, y = denom_183_cast)[name = tensor<string, []>("out_183_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_27109_to_fp16 = const()[name = tensor<string, []>("op_27109_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2156302144)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_27110_cast = add(x = out_183_cast, y = var_27109_to_fp16)[name = tensor<string, []>("op_27110_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_27112_to_fp16 = const()[name = tensor<string, []>("op_27112_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2156304768)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_247_cast = mul(x = var_27110_cast, y = var_27112_to_fp16)[name = tensor<string, []>("hidden_states_247_cast")];
            tensor<int32, [2]> var_27119 = const()[name = tensor<string, []>("op_27119"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_27121 = const()[name = tensor<string, []>("op_27121"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_123_pad_type_0 = const()[name = tensor<string, []>("q_123_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_123_pad_0 = const()[name = tensor<string, []>("q_123_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2156307392)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_123_cast = conv(dilations = var_27121, groups = var_21105, pad = q_123_pad_0, pad_type = q_123_pad_type_0, strides = var_27119, weight = mid_block_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16, x = hidden_states_247_cast)[name = tensor<string, []>("q_123_cast")];
            tensor<int32, [2]> var_27125 = const()[name = tensor<string, []>("op_27125"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_27127 = const()[name = tensor<string, []>("op_27127"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_245_pad_type_0 = const()[name = tensor<string, []>("k_245_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_245_pad_0 = const()[name = tensor<string, []>("k_245_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2159584256)))];
            tensor<fp16, [2, 1280, 1, 77]> k_245_cast = conv(dilations = var_27127, groups = var_21105, pad = k_245_pad_0, pad_type = k_245_pad_type_0, strides = var_27125, weight = mid_block_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_245_cast")];
            tensor<int32, [2]> var_27131 = const()[name = tensor<string, []>("op_27131"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_27133 = const()[name = tensor<string, []>("op_27133"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_123_pad_type_0 = const()[name = tensor<string, []>("v_123_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_123_pad_0 = const()[name = tensor<string, []>("v_123_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2164827200)))];
            tensor<fp16, [2, 1280, 1, 77]> v_123_cast = conv(dilations = var_27133, groups = var_21105, pad = v_123_pad_0, pad_type = v_123_pad_type_0, strides = var_27131, weight = mid_block_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_123_cast")];
            tensor<int32, [4]> var_27137_begin_0 = const()[name = tensor<string, []>("op_27137_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_27137_end_0 = const()[name = tensor<string, []>("op_27137_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_27137_end_mask_0 = const()[name = tensor<string, []>("op_27137_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27137_cast = slice_by_index(begin = var_27137_begin_0, end = var_27137_end_0, end_mask = var_27137_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27137_cast")];
            tensor<int32, [4]> var_27141_begin_0 = const()[name = tensor<string, []>("op_27141_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_27141_end_0 = const()[name = tensor<string, []>("op_27141_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_27141_end_mask_0 = const()[name = tensor<string, []>("op_27141_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27141_cast = slice_by_index(begin = var_27141_begin_0, end = var_27141_end_0, end_mask = var_27141_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27141_cast")];
            tensor<int32, [4]> var_27145_begin_0 = const()[name = tensor<string, []>("op_27145_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_27145_end_0 = const()[name = tensor<string, []>("op_27145_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_27145_end_mask_0 = const()[name = tensor<string, []>("op_27145_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27145_cast = slice_by_index(begin = var_27145_begin_0, end = var_27145_end_0, end_mask = var_27145_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27145_cast")];
            tensor<int32, [4]> var_27149_begin_0 = const()[name = tensor<string, []>("op_27149_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_27149_end_0 = const()[name = tensor<string, []>("op_27149_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_27149_end_mask_0 = const()[name = tensor<string, []>("op_27149_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27149_cast = slice_by_index(begin = var_27149_begin_0, end = var_27149_end_0, end_mask = var_27149_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27149_cast")];
            tensor<int32, [4]> var_27153_begin_0 = const()[name = tensor<string, []>("op_27153_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_27153_end_0 = const()[name = tensor<string, []>("op_27153_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_27153_end_mask_0 = const()[name = tensor<string, []>("op_27153_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27153_cast = slice_by_index(begin = var_27153_begin_0, end = var_27153_end_0, end_mask = var_27153_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27153_cast")];
            tensor<int32, [4]> var_27157_begin_0 = const()[name = tensor<string, []>("op_27157_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_27157_end_0 = const()[name = tensor<string, []>("op_27157_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_27157_end_mask_0 = const()[name = tensor<string, []>("op_27157_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27157_cast = slice_by_index(begin = var_27157_begin_0, end = var_27157_end_0, end_mask = var_27157_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27157_cast")];
            tensor<int32, [4]> var_27161_begin_0 = const()[name = tensor<string, []>("op_27161_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_27161_end_0 = const()[name = tensor<string, []>("op_27161_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_27161_end_mask_0 = const()[name = tensor<string, []>("op_27161_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27161_cast = slice_by_index(begin = var_27161_begin_0, end = var_27161_end_0, end_mask = var_27161_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27161_cast")];
            tensor<int32, [4]> var_27165_begin_0 = const()[name = tensor<string, []>("op_27165_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_27165_end_0 = const()[name = tensor<string, []>("op_27165_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_27165_end_mask_0 = const()[name = tensor<string, []>("op_27165_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27165_cast = slice_by_index(begin = var_27165_begin_0, end = var_27165_end_0, end_mask = var_27165_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27165_cast")];
            tensor<int32, [4]> var_27169_begin_0 = const()[name = tensor<string, []>("op_27169_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_27169_end_0 = const()[name = tensor<string, []>("op_27169_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_27169_end_mask_0 = const()[name = tensor<string, []>("op_27169_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27169_cast = slice_by_index(begin = var_27169_begin_0, end = var_27169_end_0, end_mask = var_27169_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27169_cast")];
            tensor<int32, [4]> var_27173_begin_0 = const()[name = tensor<string, []>("op_27173_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_27173_end_0 = const()[name = tensor<string, []>("op_27173_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_27173_end_mask_0 = const()[name = tensor<string, []>("op_27173_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27173_cast = slice_by_index(begin = var_27173_begin_0, end = var_27173_end_0, end_mask = var_27173_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27173_cast")];
            tensor<int32, [4]> var_27177_begin_0 = const()[name = tensor<string, []>("op_27177_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_27177_end_0 = const()[name = tensor<string, []>("op_27177_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_27177_end_mask_0 = const()[name = tensor<string, []>("op_27177_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27177_cast = slice_by_index(begin = var_27177_begin_0, end = var_27177_end_0, end_mask = var_27177_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27177_cast")];
            tensor<int32, [4]> var_27181_begin_0 = const()[name = tensor<string, []>("op_27181_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_27181_end_0 = const()[name = tensor<string, []>("op_27181_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_27181_end_mask_0 = const()[name = tensor<string, []>("op_27181_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27181_cast = slice_by_index(begin = var_27181_begin_0, end = var_27181_end_0, end_mask = var_27181_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27181_cast")];
            tensor<int32, [4]> var_27185_begin_0 = const()[name = tensor<string, []>("op_27185_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_27185_end_0 = const()[name = tensor<string, []>("op_27185_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_27185_end_mask_0 = const()[name = tensor<string, []>("op_27185_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27185_cast = slice_by_index(begin = var_27185_begin_0, end = var_27185_end_0, end_mask = var_27185_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27185_cast")];
            tensor<int32, [4]> var_27189_begin_0 = const()[name = tensor<string, []>("op_27189_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_27189_end_0 = const()[name = tensor<string, []>("op_27189_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_27189_end_mask_0 = const()[name = tensor<string, []>("op_27189_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27189_cast = slice_by_index(begin = var_27189_begin_0, end = var_27189_end_0, end_mask = var_27189_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27189_cast")];
            tensor<int32, [4]> var_27193_begin_0 = const()[name = tensor<string, []>("op_27193_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_27193_end_0 = const()[name = tensor<string, []>("op_27193_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_27193_end_mask_0 = const()[name = tensor<string, []>("op_27193_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27193_cast = slice_by_index(begin = var_27193_begin_0, end = var_27193_end_0, end_mask = var_27193_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27193_cast")];
            tensor<int32, [4]> var_27197_begin_0 = const()[name = tensor<string, []>("op_27197_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_27197_end_0 = const()[name = tensor<string, []>("op_27197_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_27197_end_mask_0 = const()[name = tensor<string, []>("op_27197_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27197_cast = slice_by_index(begin = var_27197_begin_0, end = var_27197_end_0, end_mask = var_27197_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27197_cast")];
            tensor<int32, [4]> var_27201_begin_0 = const()[name = tensor<string, []>("op_27201_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_27201_end_0 = const()[name = tensor<string, []>("op_27201_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_27201_end_mask_0 = const()[name = tensor<string, []>("op_27201_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27201_cast = slice_by_index(begin = var_27201_begin_0, end = var_27201_end_0, end_mask = var_27201_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27201_cast")];
            tensor<int32, [4]> var_27205_begin_0 = const()[name = tensor<string, []>("op_27205_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_27205_end_0 = const()[name = tensor<string, []>("op_27205_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_27205_end_mask_0 = const()[name = tensor<string, []>("op_27205_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27205_cast = slice_by_index(begin = var_27205_begin_0, end = var_27205_end_0, end_mask = var_27205_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27205_cast")];
            tensor<int32, [4]> var_27209_begin_0 = const()[name = tensor<string, []>("op_27209_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_27209_end_0 = const()[name = tensor<string, []>("op_27209_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_27209_end_mask_0 = const()[name = tensor<string, []>("op_27209_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27209_cast = slice_by_index(begin = var_27209_begin_0, end = var_27209_end_0, end_mask = var_27209_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27209_cast")];
            tensor<int32, [4]> var_27213_begin_0 = const()[name = tensor<string, []>("op_27213_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_27213_end_0 = const()[name = tensor<string, []>("op_27213_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_27213_end_mask_0 = const()[name = tensor<string, []>("op_27213_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27213_cast = slice_by_index(begin = var_27213_begin_0, end = var_27213_end_0, end_mask = var_27213_end_mask_0, x = q_123_cast)[name = tensor<string, []>("op_27213_cast")];
            tensor<int32, [4]> k_247_perm_0 = const()[name = tensor<string, []>("k_247_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_27220_begin_0 = const()[name = tensor<string, []>("op_27220_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_27220_end_0 = const()[name = tensor<string, []>("op_27220_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_27220_end_mask_0 = const()[name = tensor<string, []>("op_27220_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_78 = transpose(perm = k_247_perm_0, x = k_245_cast)[name = tensor<string, []>("transpose_78")];
            tensor<fp16, [2, 77, 1, 64]> var_27220_cast = slice_by_index(begin = var_27220_begin_0, end = var_27220_end_0, end_mask = var_27220_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27220_cast")];
            tensor<int32, [4]> var_27224_begin_0 = const()[name = tensor<string, []>("op_27224_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_27224_end_0 = const()[name = tensor<string, []>("op_27224_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_27224_end_mask_0 = const()[name = tensor<string, []>("op_27224_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_27224_cast = slice_by_index(begin = var_27224_begin_0, end = var_27224_end_0, end_mask = var_27224_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27224_cast")];
            tensor<int32, [4]> var_27228_begin_0 = const()[name = tensor<string, []>("op_27228_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_27228_end_0 = const()[name = tensor<string, []>("op_27228_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_27228_end_mask_0 = const()[name = tensor<string, []>("op_27228_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_27228_cast = slice_by_index(begin = var_27228_begin_0, end = var_27228_end_0, end_mask = var_27228_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27228_cast")];
            tensor<int32, [4]> var_27232_begin_0 = const()[name = tensor<string, []>("op_27232_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_27232_end_0 = const()[name = tensor<string, []>("op_27232_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_27232_end_mask_0 = const()[name = tensor<string, []>("op_27232_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_27232_cast = slice_by_index(begin = var_27232_begin_0, end = var_27232_end_0, end_mask = var_27232_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27232_cast")];
            tensor<int32, [4]> var_27236_begin_0 = const()[name = tensor<string, []>("op_27236_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_27236_end_0 = const()[name = tensor<string, []>("op_27236_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_27236_end_mask_0 = const()[name = tensor<string, []>("op_27236_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_27236_cast = slice_by_index(begin = var_27236_begin_0, end = var_27236_end_0, end_mask = var_27236_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27236_cast")];
            tensor<int32, [4]> var_27240_begin_0 = const()[name = tensor<string, []>("op_27240_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_27240_end_0 = const()[name = tensor<string, []>("op_27240_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_27240_end_mask_0 = const()[name = tensor<string, []>("op_27240_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_27240_cast = slice_by_index(begin = var_27240_begin_0, end = var_27240_end_0, end_mask = var_27240_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27240_cast")];
            tensor<int32, [4]> var_27244_begin_0 = const()[name = tensor<string, []>("op_27244_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_27244_end_0 = const()[name = tensor<string, []>("op_27244_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_27244_end_mask_0 = const()[name = tensor<string, []>("op_27244_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_27244_cast = slice_by_index(begin = var_27244_begin_0, end = var_27244_end_0, end_mask = var_27244_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27244_cast")];
            tensor<int32, [4]> var_27248_begin_0 = const()[name = tensor<string, []>("op_27248_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_27248_end_0 = const()[name = tensor<string, []>("op_27248_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_27248_end_mask_0 = const()[name = tensor<string, []>("op_27248_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_27248_cast = slice_by_index(begin = var_27248_begin_0, end = var_27248_end_0, end_mask = var_27248_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27248_cast")];
            tensor<int32, [4]> var_27252_begin_0 = const()[name = tensor<string, []>("op_27252_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_27252_end_0 = const()[name = tensor<string, []>("op_27252_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_27252_end_mask_0 = const()[name = tensor<string, []>("op_27252_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_27252_cast = slice_by_index(begin = var_27252_begin_0, end = var_27252_end_0, end_mask = var_27252_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27252_cast")];
            tensor<int32, [4]> var_27256_begin_0 = const()[name = tensor<string, []>("op_27256_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_27256_end_0 = const()[name = tensor<string, []>("op_27256_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_27256_end_mask_0 = const()[name = tensor<string, []>("op_27256_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_27256_cast = slice_by_index(begin = var_27256_begin_0, end = var_27256_end_0, end_mask = var_27256_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27256_cast")];
            tensor<int32, [4]> var_27260_begin_0 = const()[name = tensor<string, []>("op_27260_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_27260_end_0 = const()[name = tensor<string, []>("op_27260_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_27260_end_mask_0 = const()[name = tensor<string, []>("op_27260_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_27260_cast = slice_by_index(begin = var_27260_begin_0, end = var_27260_end_0, end_mask = var_27260_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27260_cast")];
            tensor<int32, [4]> var_27264_begin_0 = const()[name = tensor<string, []>("op_27264_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_27264_end_0 = const()[name = tensor<string, []>("op_27264_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_27264_end_mask_0 = const()[name = tensor<string, []>("op_27264_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_27264_cast = slice_by_index(begin = var_27264_begin_0, end = var_27264_end_0, end_mask = var_27264_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27264_cast")];
            tensor<int32, [4]> var_27268_begin_0 = const()[name = tensor<string, []>("op_27268_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_27268_end_0 = const()[name = tensor<string, []>("op_27268_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_27268_end_mask_0 = const()[name = tensor<string, []>("op_27268_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_27268_cast = slice_by_index(begin = var_27268_begin_0, end = var_27268_end_0, end_mask = var_27268_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27268_cast")];
            tensor<int32, [4]> var_27272_begin_0 = const()[name = tensor<string, []>("op_27272_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_27272_end_0 = const()[name = tensor<string, []>("op_27272_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_27272_end_mask_0 = const()[name = tensor<string, []>("op_27272_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_27272_cast = slice_by_index(begin = var_27272_begin_0, end = var_27272_end_0, end_mask = var_27272_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27272_cast")];
            tensor<int32, [4]> var_27276_begin_0 = const()[name = tensor<string, []>("op_27276_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_27276_end_0 = const()[name = tensor<string, []>("op_27276_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_27276_end_mask_0 = const()[name = tensor<string, []>("op_27276_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_27276_cast = slice_by_index(begin = var_27276_begin_0, end = var_27276_end_0, end_mask = var_27276_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27276_cast")];
            tensor<int32, [4]> var_27280_begin_0 = const()[name = tensor<string, []>("op_27280_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_27280_end_0 = const()[name = tensor<string, []>("op_27280_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_27280_end_mask_0 = const()[name = tensor<string, []>("op_27280_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_27280_cast = slice_by_index(begin = var_27280_begin_0, end = var_27280_end_0, end_mask = var_27280_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27280_cast")];
            tensor<int32, [4]> var_27284_begin_0 = const()[name = tensor<string, []>("op_27284_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_27284_end_0 = const()[name = tensor<string, []>("op_27284_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_27284_end_mask_0 = const()[name = tensor<string, []>("op_27284_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_27284_cast = slice_by_index(begin = var_27284_begin_0, end = var_27284_end_0, end_mask = var_27284_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27284_cast")];
            tensor<int32, [4]> var_27288_begin_0 = const()[name = tensor<string, []>("op_27288_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_27288_end_0 = const()[name = tensor<string, []>("op_27288_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_27288_end_mask_0 = const()[name = tensor<string, []>("op_27288_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_27288_cast = slice_by_index(begin = var_27288_begin_0, end = var_27288_end_0, end_mask = var_27288_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27288_cast")];
            tensor<int32, [4]> var_27292_begin_0 = const()[name = tensor<string, []>("op_27292_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_27292_end_0 = const()[name = tensor<string, []>("op_27292_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_27292_end_mask_0 = const()[name = tensor<string, []>("op_27292_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_27292_cast = slice_by_index(begin = var_27292_begin_0, end = var_27292_end_0, end_mask = var_27292_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27292_cast")];
            tensor<int32, [4]> var_27296_begin_0 = const()[name = tensor<string, []>("op_27296_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_27296_end_0 = const()[name = tensor<string, []>("op_27296_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_27296_end_mask_0 = const()[name = tensor<string, []>("op_27296_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_27296_cast = slice_by_index(begin = var_27296_begin_0, end = var_27296_end_0, end_mask = var_27296_end_mask_0, x = transpose_78)[name = tensor<string, []>("op_27296_cast")];
            tensor<int32, [4]> var_27298_begin_0 = const()[name = tensor<string, []>("op_27298_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_27298_end_0 = const()[name = tensor<string, []>("op_27298_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_27298_end_mask_0 = const()[name = tensor<string, []>("op_27298_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27298_cast = slice_by_index(begin = var_27298_begin_0, end = var_27298_end_0, end_mask = var_27298_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27298_cast")];
            tensor<int32, [4]> var_27302_begin_0 = const()[name = tensor<string, []>("op_27302_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_27302_end_0 = const()[name = tensor<string, []>("op_27302_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_27302_end_mask_0 = const()[name = tensor<string, []>("op_27302_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27302_cast = slice_by_index(begin = var_27302_begin_0, end = var_27302_end_0, end_mask = var_27302_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27302_cast")];
            tensor<int32, [4]> var_27306_begin_0 = const()[name = tensor<string, []>("op_27306_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_27306_end_0 = const()[name = tensor<string, []>("op_27306_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_27306_end_mask_0 = const()[name = tensor<string, []>("op_27306_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27306_cast = slice_by_index(begin = var_27306_begin_0, end = var_27306_end_0, end_mask = var_27306_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27306_cast")];
            tensor<int32, [4]> var_27310_begin_0 = const()[name = tensor<string, []>("op_27310_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_27310_end_0 = const()[name = tensor<string, []>("op_27310_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_27310_end_mask_0 = const()[name = tensor<string, []>("op_27310_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27310_cast = slice_by_index(begin = var_27310_begin_0, end = var_27310_end_0, end_mask = var_27310_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27310_cast")];
            tensor<int32, [4]> var_27314_begin_0 = const()[name = tensor<string, []>("op_27314_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_27314_end_0 = const()[name = tensor<string, []>("op_27314_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_27314_end_mask_0 = const()[name = tensor<string, []>("op_27314_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27314_cast = slice_by_index(begin = var_27314_begin_0, end = var_27314_end_0, end_mask = var_27314_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27314_cast")];
            tensor<int32, [4]> var_27318_begin_0 = const()[name = tensor<string, []>("op_27318_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_27318_end_0 = const()[name = tensor<string, []>("op_27318_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_27318_end_mask_0 = const()[name = tensor<string, []>("op_27318_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27318_cast = slice_by_index(begin = var_27318_begin_0, end = var_27318_end_0, end_mask = var_27318_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27318_cast")];
            tensor<int32, [4]> var_27322_begin_0 = const()[name = tensor<string, []>("op_27322_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_27322_end_0 = const()[name = tensor<string, []>("op_27322_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_27322_end_mask_0 = const()[name = tensor<string, []>("op_27322_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27322_cast = slice_by_index(begin = var_27322_begin_0, end = var_27322_end_0, end_mask = var_27322_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27322_cast")];
            tensor<int32, [4]> var_27326_begin_0 = const()[name = tensor<string, []>("op_27326_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_27326_end_0 = const()[name = tensor<string, []>("op_27326_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_27326_end_mask_0 = const()[name = tensor<string, []>("op_27326_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27326_cast = slice_by_index(begin = var_27326_begin_0, end = var_27326_end_0, end_mask = var_27326_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27326_cast")];
            tensor<int32, [4]> var_27330_begin_0 = const()[name = tensor<string, []>("op_27330_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_27330_end_0 = const()[name = tensor<string, []>("op_27330_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_27330_end_mask_0 = const()[name = tensor<string, []>("op_27330_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27330_cast = slice_by_index(begin = var_27330_begin_0, end = var_27330_end_0, end_mask = var_27330_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27330_cast")];
            tensor<int32, [4]> var_27334_begin_0 = const()[name = tensor<string, []>("op_27334_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_27334_end_0 = const()[name = tensor<string, []>("op_27334_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_27334_end_mask_0 = const()[name = tensor<string, []>("op_27334_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27334_cast = slice_by_index(begin = var_27334_begin_0, end = var_27334_end_0, end_mask = var_27334_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27334_cast")];
            tensor<int32, [4]> var_27338_begin_0 = const()[name = tensor<string, []>("op_27338_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_27338_end_0 = const()[name = tensor<string, []>("op_27338_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_27338_end_mask_0 = const()[name = tensor<string, []>("op_27338_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27338_cast = slice_by_index(begin = var_27338_begin_0, end = var_27338_end_0, end_mask = var_27338_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27338_cast")];
            tensor<int32, [4]> var_27342_begin_0 = const()[name = tensor<string, []>("op_27342_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_27342_end_0 = const()[name = tensor<string, []>("op_27342_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_27342_end_mask_0 = const()[name = tensor<string, []>("op_27342_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27342_cast = slice_by_index(begin = var_27342_begin_0, end = var_27342_end_0, end_mask = var_27342_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27342_cast")];
            tensor<int32, [4]> var_27346_begin_0 = const()[name = tensor<string, []>("op_27346_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_27346_end_0 = const()[name = tensor<string, []>("op_27346_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_27346_end_mask_0 = const()[name = tensor<string, []>("op_27346_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27346_cast = slice_by_index(begin = var_27346_begin_0, end = var_27346_end_0, end_mask = var_27346_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27346_cast")];
            tensor<int32, [4]> var_27350_begin_0 = const()[name = tensor<string, []>("op_27350_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_27350_end_0 = const()[name = tensor<string, []>("op_27350_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_27350_end_mask_0 = const()[name = tensor<string, []>("op_27350_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27350_cast = slice_by_index(begin = var_27350_begin_0, end = var_27350_end_0, end_mask = var_27350_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27350_cast")];
            tensor<int32, [4]> var_27354_begin_0 = const()[name = tensor<string, []>("op_27354_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_27354_end_0 = const()[name = tensor<string, []>("op_27354_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_27354_end_mask_0 = const()[name = tensor<string, []>("op_27354_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27354_cast = slice_by_index(begin = var_27354_begin_0, end = var_27354_end_0, end_mask = var_27354_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27354_cast")];
            tensor<int32, [4]> var_27358_begin_0 = const()[name = tensor<string, []>("op_27358_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_27358_end_0 = const()[name = tensor<string, []>("op_27358_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_27358_end_mask_0 = const()[name = tensor<string, []>("op_27358_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27358_cast = slice_by_index(begin = var_27358_begin_0, end = var_27358_end_0, end_mask = var_27358_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27358_cast")];
            tensor<int32, [4]> var_27362_begin_0 = const()[name = tensor<string, []>("op_27362_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_27362_end_0 = const()[name = tensor<string, []>("op_27362_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_27362_end_mask_0 = const()[name = tensor<string, []>("op_27362_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27362_cast = slice_by_index(begin = var_27362_begin_0, end = var_27362_end_0, end_mask = var_27362_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27362_cast")];
            tensor<int32, [4]> var_27366_begin_0 = const()[name = tensor<string, []>("op_27366_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_27366_end_0 = const()[name = tensor<string, []>("op_27366_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_27366_end_mask_0 = const()[name = tensor<string, []>("op_27366_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27366_cast = slice_by_index(begin = var_27366_begin_0, end = var_27366_end_0, end_mask = var_27366_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27366_cast")];
            tensor<int32, [4]> var_27370_begin_0 = const()[name = tensor<string, []>("op_27370_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_27370_end_0 = const()[name = tensor<string, []>("op_27370_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_27370_end_mask_0 = const()[name = tensor<string, []>("op_27370_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27370_cast = slice_by_index(begin = var_27370_begin_0, end = var_27370_end_0, end_mask = var_27370_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27370_cast")];
            tensor<int32, [4]> var_27374_begin_0 = const()[name = tensor<string, []>("op_27374_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_27374_end_0 = const()[name = tensor<string, []>("op_27374_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_27374_end_mask_0 = const()[name = tensor<string, []>("op_27374_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_27374_cast = slice_by_index(begin = var_27374_begin_0, end = var_27374_end_0, end_mask = var_27374_end_mask_0, x = v_123_cast)[name = tensor<string, []>("op_27374_cast")];
            tensor<string, []> var_27378_equation_0 = const()[name = tensor<string, []>("op_27378_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27378_cast = einsum(equation = var_27378_equation_0, values = (var_27220_cast, var_27137_cast))[name = tensor<string, []>("op_27378_cast")];
            tensor<fp16, []> var_27379_to_fp16 = const()[name = tensor<string, []>("op_27379_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2281_cast = mul(x = var_27378_cast, y = var_27379_to_fp16)[name = tensor<string, []>("aw_2281_cast")];
            tensor<string, []> var_27382_equation_0 = const()[name = tensor<string, []>("op_27382_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27382_cast = einsum(equation = var_27382_equation_0, values = (var_27224_cast, var_27141_cast))[name = tensor<string, []>("op_27382_cast")];
            tensor<fp16, []> var_27383_to_fp16 = const()[name = tensor<string, []>("op_27383_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2283_cast = mul(x = var_27382_cast, y = var_27383_to_fp16)[name = tensor<string, []>("aw_2283_cast")];
            tensor<string, []> var_27386_equation_0 = const()[name = tensor<string, []>("op_27386_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27386_cast = einsum(equation = var_27386_equation_0, values = (var_27228_cast, var_27145_cast))[name = tensor<string, []>("op_27386_cast")];
            tensor<fp16, []> var_27387_to_fp16 = const()[name = tensor<string, []>("op_27387_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2285_cast = mul(x = var_27386_cast, y = var_27387_to_fp16)[name = tensor<string, []>("aw_2285_cast")];
            tensor<string, []> var_27390_equation_0 = const()[name = tensor<string, []>("op_27390_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27390_cast = einsum(equation = var_27390_equation_0, values = (var_27232_cast, var_27149_cast))[name = tensor<string, []>("op_27390_cast")];
            tensor<fp16, []> var_27391_to_fp16 = const()[name = tensor<string, []>("op_27391_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2287_cast = mul(x = var_27390_cast, y = var_27391_to_fp16)[name = tensor<string, []>("aw_2287_cast")];
            tensor<string, []> var_27394_equation_0 = const()[name = tensor<string, []>("op_27394_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27394_cast = einsum(equation = var_27394_equation_0, values = (var_27236_cast, var_27153_cast))[name = tensor<string, []>("op_27394_cast")];
            tensor<fp16, []> var_27395_to_fp16 = const()[name = tensor<string, []>("op_27395_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2289_cast = mul(x = var_27394_cast, y = var_27395_to_fp16)[name = tensor<string, []>("aw_2289_cast")];
            tensor<string, []> var_27398_equation_0 = const()[name = tensor<string, []>("op_27398_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27398_cast = einsum(equation = var_27398_equation_0, values = (var_27240_cast, var_27157_cast))[name = tensor<string, []>("op_27398_cast")];
            tensor<fp16, []> var_27399_to_fp16 = const()[name = tensor<string, []>("op_27399_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2291_cast = mul(x = var_27398_cast, y = var_27399_to_fp16)[name = tensor<string, []>("aw_2291_cast")];
            tensor<string, []> var_27402_equation_0 = const()[name = tensor<string, []>("op_27402_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27402_cast = einsum(equation = var_27402_equation_0, values = (var_27244_cast, var_27161_cast))[name = tensor<string, []>("op_27402_cast")];
            tensor<fp16, []> var_27403_to_fp16 = const()[name = tensor<string, []>("op_27403_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2293_cast = mul(x = var_27402_cast, y = var_27403_to_fp16)[name = tensor<string, []>("aw_2293_cast")];
            tensor<string, []> var_27406_equation_0 = const()[name = tensor<string, []>("op_27406_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27406_cast = einsum(equation = var_27406_equation_0, values = (var_27248_cast, var_27165_cast))[name = tensor<string, []>("op_27406_cast")];
            tensor<fp16, []> var_27407_to_fp16 = const()[name = tensor<string, []>("op_27407_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2295_cast = mul(x = var_27406_cast, y = var_27407_to_fp16)[name = tensor<string, []>("aw_2295_cast")];
            tensor<string, []> var_27410_equation_0 = const()[name = tensor<string, []>("op_27410_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27410_cast = einsum(equation = var_27410_equation_0, values = (var_27252_cast, var_27169_cast))[name = tensor<string, []>("op_27410_cast")];
            tensor<fp16, []> var_27411_to_fp16 = const()[name = tensor<string, []>("op_27411_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2297_cast = mul(x = var_27410_cast, y = var_27411_to_fp16)[name = tensor<string, []>("aw_2297_cast")];
            tensor<string, []> var_27414_equation_0 = const()[name = tensor<string, []>("op_27414_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27414_cast = einsum(equation = var_27414_equation_0, values = (var_27256_cast, var_27173_cast))[name = tensor<string, []>("op_27414_cast")];
            tensor<fp16, []> var_27415_to_fp16 = const()[name = tensor<string, []>("op_27415_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2299_cast = mul(x = var_27414_cast, y = var_27415_to_fp16)[name = tensor<string, []>("aw_2299_cast")];
            tensor<string, []> var_27418_equation_0 = const()[name = tensor<string, []>("op_27418_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27418_cast = einsum(equation = var_27418_equation_0, values = (var_27260_cast, var_27177_cast))[name = tensor<string, []>("op_27418_cast")];
            tensor<fp16, []> var_27419_to_fp16 = const()[name = tensor<string, []>("op_27419_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2301_cast = mul(x = var_27418_cast, y = var_27419_to_fp16)[name = tensor<string, []>("aw_2301_cast")];
            tensor<string, []> var_27422_equation_0 = const()[name = tensor<string, []>("op_27422_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27422_cast = einsum(equation = var_27422_equation_0, values = (var_27264_cast, var_27181_cast))[name = tensor<string, []>("op_27422_cast")];
            tensor<fp16, []> var_27423_to_fp16 = const()[name = tensor<string, []>("op_27423_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2303_cast = mul(x = var_27422_cast, y = var_27423_to_fp16)[name = tensor<string, []>("aw_2303_cast")];
            tensor<string, []> var_27426_equation_0 = const()[name = tensor<string, []>("op_27426_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27426_cast = einsum(equation = var_27426_equation_0, values = (var_27268_cast, var_27185_cast))[name = tensor<string, []>("op_27426_cast")];
            tensor<fp16, []> var_27427_to_fp16 = const()[name = tensor<string, []>("op_27427_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2305_cast = mul(x = var_27426_cast, y = var_27427_to_fp16)[name = tensor<string, []>("aw_2305_cast")];
            tensor<string, []> var_27430_equation_0 = const()[name = tensor<string, []>("op_27430_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27430_cast = einsum(equation = var_27430_equation_0, values = (var_27272_cast, var_27189_cast))[name = tensor<string, []>("op_27430_cast")];
            tensor<fp16, []> var_27431_to_fp16 = const()[name = tensor<string, []>("op_27431_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2307_cast = mul(x = var_27430_cast, y = var_27431_to_fp16)[name = tensor<string, []>("aw_2307_cast")];
            tensor<string, []> var_27434_equation_0 = const()[name = tensor<string, []>("op_27434_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27434_cast = einsum(equation = var_27434_equation_0, values = (var_27276_cast, var_27193_cast))[name = tensor<string, []>("op_27434_cast")];
            tensor<fp16, []> var_27435_to_fp16 = const()[name = tensor<string, []>("op_27435_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2309_cast = mul(x = var_27434_cast, y = var_27435_to_fp16)[name = tensor<string, []>("aw_2309_cast")];
            tensor<string, []> var_27438_equation_0 = const()[name = tensor<string, []>("op_27438_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27438_cast = einsum(equation = var_27438_equation_0, values = (var_27280_cast, var_27197_cast))[name = tensor<string, []>("op_27438_cast")];
            tensor<fp16, []> var_27439_to_fp16 = const()[name = tensor<string, []>("op_27439_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2311_cast = mul(x = var_27438_cast, y = var_27439_to_fp16)[name = tensor<string, []>("aw_2311_cast")];
            tensor<string, []> var_27442_equation_0 = const()[name = tensor<string, []>("op_27442_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27442_cast = einsum(equation = var_27442_equation_0, values = (var_27284_cast, var_27201_cast))[name = tensor<string, []>("op_27442_cast")];
            tensor<fp16, []> var_27443_to_fp16 = const()[name = tensor<string, []>("op_27443_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2313_cast = mul(x = var_27442_cast, y = var_27443_to_fp16)[name = tensor<string, []>("aw_2313_cast")];
            tensor<string, []> var_27446_equation_0 = const()[name = tensor<string, []>("op_27446_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27446_cast = einsum(equation = var_27446_equation_0, values = (var_27288_cast, var_27205_cast))[name = tensor<string, []>("op_27446_cast")];
            tensor<fp16, []> var_27447_to_fp16 = const()[name = tensor<string, []>("op_27447_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2315_cast = mul(x = var_27446_cast, y = var_27447_to_fp16)[name = tensor<string, []>("aw_2315_cast")];
            tensor<string, []> var_27450_equation_0 = const()[name = tensor<string, []>("op_27450_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27450_cast = einsum(equation = var_27450_equation_0, values = (var_27292_cast, var_27209_cast))[name = tensor<string, []>("op_27450_cast")];
            tensor<fp16, []> var_27451_to_fp16 = const()[name = tensor<string, []>("op_27451_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2317_cast = mul(x = var_27450_cast, y = var_27451_to_fp16)[name = tensor<string, []>("aw_2317_cast")];
            tensor<string, []> var_27454_equation_0 = const()[name = tensor<string, []>("op_27454_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_27454_cast = einsum(equation = var_27454_equation_0, values = (var_27296_cast, var_27213_cast))[name = tensor<string, []>("op_27454_cast")];
            tensor<fp16, []> var_27455_to_fp16 = const()[name = tensor<string, []>("op_27455_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2319_cast = mul(x = var_27454_cast, y = var_27455_to_fp16)[name = tensor<string, []>("aw_2319_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27457_cast = softmax(axis = var_21105, x = aw_2281_cast)[name = tensor<string, []>("op_27457_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27458_cast = softmax(axis = var_21105, x = aw_2283_cast)[name = tensor<string, []>("op_27458_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27459_cast = softmax(axis = var_21105, x = aw_2285_cast)[name = tensor<string, []>("op_27459_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27460_cast = softmax(axis = var_21105, x = aw_2287_cast)[name = tensor<string, []>("op_27460_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27461_cast = softmax(axis = var_21105, x = aw_2289_cast)[name = tensor<string, []>("op_27461_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27462_cast = softmax(axis = var_21105, x = aw_2291_cast)[name = tensor<string, []>("op_27462_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27463_cast = softmax(axis = var_21105, x = aw_2293_cast)[name = tensor<string, []>("op_27463_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27464_cast = softmax(axis = var_21105, x = aw_2295_cast)[name = tensor<string, []>("op_27464_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27465_cast = softmax(axis = var_21105, x = aw_2297_cast)[name = tensor<string, []>("op_27465_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27466_cast = softmax(axis = var_21105, x = aw_2299_cast)[name = tensor<string, []>("op_27466_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27467_cast = softmax(axis = var_21105, x = aw_2301_cast)[name = tensor<string, []>("op_27467_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27468_cast = softmax(axis = var_21105, x = aw_2303_cast)[name = tensor<string, []>("op_27468_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27469_cast = softmax(axis = var_21105, x = aw_2305_cast)[name = tensor<string, []>("op_27469_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27470_cast = softmax(axis = var_21105, x = aw_2307_cast)[name = tensor<string, []>("op_27470_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27471_cast = softmax(axis = var_21105, x = aw_2309_cast)[name = tensor<string, []>("op_27471_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27472_cast = softmax(axis = var_21105, x = aw_2311_cast)[name = tensor<string, []>("op_27472_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27473_cast = softmax(axis = var_21105, x = aw_2313_cast)[name = tensor<string, []>("op_27473_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27474_cast = softmax(axis = var_21105, x = aw_2315_cast)[name = tensor<string, []>("op_27474_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27475_cast = softmax(axis = var_21105, x = aw_2317_cast)[name = tensor<string, []>("op_27475_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_27476_cast = softmax(axis = var_21105, x = aw_2319_cast)[name = tensor<string, []>("op_27476_cast")];
            tensor<string, []> var_27478_equation_0 = const()[name = tensor<string, []>("op_27478_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27478_cast = einsum(equation = var_27478_equation_0, values = (var_27298_cast, var_27457_cast))[name = tensor<string, []>("op_27478_cast")];
            tensor<string, []> var_27480_equation_0 = const()[name = tensor<string, []>("op_27480_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27480_cast = einsum(equation = var_27480_equation_0, values = (var_27302_cast, var_27458_cast))[name = tensor<string, []>("op_27480_cast")];
            tensor<string, []> var_27482_equation_0 = const()[name = tensor<string, []>("op_27482_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27482_cast = einsum(equation = var_27482_equation_0, values = (var_27306_cast, var_27459_cast))[name = tensor<string, []>("op_27482_cast")];
            tensor<string, []> var_27484_equation_0 = const()[name = tensor<string, []>("op_27484_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27484_cast = einsum(equation = var_27484_equation_0, values = (var_27310_cast, var_27460_cast))[name = tensor<string, []>("op_27484_cast")];
            tensor<string, []> var_27486_equation_0 = const()[name = tensor<string, []>("op_27486_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27486_cast = einsum(equation = var_27486_equation_0, values = (var_27314_cast, var_27461_cast))[name = tensor<string, []>("op_27486_cast")];
            tensor<string, []> var_27488_equation_0 = const()[name = tensor<string, []>("op_27488_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27488_cast = einsum(equation = var_27488_equation_0, values = (var_27318_cast, var_27462_cast))[name = tensor<string, []>("op_27488_cast")];
            tensor<string, []> var_27490_equation_0 = const()[name = tensor<string, []>("op_27490_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27490_cast = einsum(equation = var_27490_equation_0, values = (var_27322_cast, var_27463_cast))[name = tensor<string, []>("op_27490_cast")];
            tensor<string, []> var_27492_equation_0 = const()[name = tensor<string, []>("op_27492_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27492_cast = einsum(equation = var_27492_equation_0, values = (var_27326_cast, var_27464_cast))[name = tensor<string, []>("op_27492_cast")];
            tensor<string, []> var_27494_equation_0 = const()[name = tensor<string, []>("op_27494_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27494_cast = einsum(equation = var_27494_equation_0, values = (var_27330_cast, var_27465_cast))[name = tensor<string, []>("op_27494_cast")];
            tensor<string, []> var_27496_equation_0 = const()[name = tensor<string, []>("op_27496_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27496_cast = einsum(equation = var_27496_equation_0, values = (var_27334_cast, var_27466_cast))[name = tensor<string, []>("op_27496_cast")];
            tensor<string, []> var_27498_equation_0 = const()[name = tensor<string, []>("op_27498_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27498_cast = einsum(equation = var_27498_equation_0, values = (var_27338_cast, var_27467_cast))[name = tensor<string, []>("op_27498_cast")];
            tensor<string, []> var_27500_equation_0 = const()[name = tensor<string, []>("op_27500_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27500_cast = einsum(equation = var_27500_equation_0, values = (var_27342_cast, var_27468_cast))[name = tensor<string, []>("op_27500_cast")];
            tensor<string, []> var_27502_equation_0 = const()[name = tensor<string, []>("op_27502_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27502_cast = einsum(equation = var_27502_equation_0, values = (var_27346_cast, var_27469_cast))[name = tensor<string, []>("op_27502_cast")];
            tensor<string, []> var_27504_equation_0 = const()[name = tensor<string, []>("op_27504_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27504_cast = einsum(equation = var_27504_equation_0, values = (var_27350_cast, var_27470_cast))[name = tensor<string, []>("op_27504_cast")];
            tensor<string, []> var_27506_equation_0 = const()[name = tensor<string, []>("op_27506_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27506_cast = einsum(equation = var_27506_equation_0, values = (var_27354_cast, var_27471_cast))[name = tensor<string, []>("op_27506_cast")];
            tensor<string, []> var_27508_equation_0 = const()[name = tensor<string, []>("op_27508_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27508_cast = einsum(equation = var_27508_equation_0, values = (var_27358_cast, var_27472_cast))[name = tensor<string, []>("op_27508_cast")];
            tensor<string, []> var_27510_equation_0 = const()[name = tensor<string, []>("op_27510_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27510_cast = einsum(equation = var_27510_equation_0, values = (var_27362_cast, var_27473_cast))[name = tensor<string, []>("op_27510_cast")];
            tensor<string, []> var_27512_equation_0 = const()[name = tensor<string, []>("op_27512_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27512_cast = einsum(equation = var_27512_equation_0, values = (var_27366_cast, var_27474_cast))[name = tensor<string, []>("op_27512_cast")];
            tensor<string, []> var_27514_equation_0 = const()[name = tensor<string, []>("op_27514_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27514_cast = einsum(equation = var_27514_equation_0, values = (var_27370_cast, var_27475_cast))[name = tensor<string, []>("op_27514_cast")];
            tensor<string, []> var_27516_equation_0 = const()[name = tensor<string, []>("op_27516_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27516_cast = einsum(equation = var_27516_equation_0, values = (var_27374_cast, var_27476_cast))[name = tensor<string, []>("op_27516_cast")];
            tensor<bool, []> input_379_interleave_0 = const()[name = tensor<string, []>("input_379_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_379_cast = concat(axis = var_21105, interleave = input_379_interleave_0, values = (var_27478_cast, var_27480_cast, var_27482_cast, var_27484_cast, var_27486_cast, var_27488_cast, var_27490_cast, var_27492_cast, var_27494_cast, var_27496_cast, var_27498_cast, var_27500_cast, var_27502_cast, var_27504_cast, var_27506_cast, var_27508_cast, var_27510_cast, var_27512_cast, var_27514_cast, var_27516_cast))[name = tensor<string, []>("input_379_cast")];
            tensor<int32, [2]> var_27522 = const()[name = tensor<string, []>("op_27522"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_27524 = const()[name = tensor<string, []>("op_27524"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_27526_pad_type_0 = const()[name = tensor<string, []>("op_27526_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_27526_pad_0 = const()[name = tensor<string, []>("op_27526_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2170070144)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2173347008)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_27526_cast = conv(bias = mid_block_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16, dilations = var_27524, groups = var_21105, pad = var_27526_pad_0, pad_type = var_27526_pad_type_0, strides = var_27522, weight = mid_block_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16, x = input_379_cast)[name = tensor<string, []>("op_27526_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_185_cast = add(x = var_27526_cast, y = inputs_183_cast)[name = tensor<string, []>("inputs_185_cast")];
            tensor<int32, [1]> var_27530 = const()[name = tensor<string, []>("op_27530"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_185_cast = reduce_mean(axes = var_27530, keep_dims = var_21100, x = inputs_185_cast)[name = tensor<string, []>("channels_mean_185_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_185_cast = sub(x = inputs_185_cast, y = channels_mean_185_cast)[name = tensor<string, []>("zero_mean_185_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_185_cast = mul(x = zero_mean_185_cast, y = zero_mean_185_cast)[name = tensor<string, []>("zero_mean_sq_185_cast")];
            tensor<int32, [1]> var_27534 = const()[name = tensor<string, []>("op_27534"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_27535_cast = reduce_mean(axes = var_27534, keep_dims = var_21100, x = zero_mean_sq_185_cast)[name = tensor<string, []>("op_27535_cast")];
            tensor<fp16, []> var_27536_to_fp16 = const()[name = tensor<string, []>("op_27536_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_27537_cast = add(x = var_27535_cast, y = var_27536_to_fp16)[name = tensor<string, []>("op_27537_cast")];
            tensor<fp16, []> denom_185_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_185_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_185_cast = rsqrt(epsilon = denom_185_epsilon_0_to_fp16, x = var_27537_cast)[name = tensor<string, []>("denom_185_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_185_cast = mul(x = zero_mean_185_cast, y = denom_185_cast)[name = tensor<string, []>("out_185_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_27541_to_fp16 = const()[name = tensor<string, []>("op_27541_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2173349632)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_27542_cast = add(x = out_185_cast, y = var_27541_to_fp16)[name = tensor<string, []>("op_27542_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_27544_to_fp16 = const()[name = tensor<string, []>("op_27544_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2173352256)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_381_cast = mul(x = var_27542_cast, y = var_27544_to_fp16)[name = tensor<string, []>("input_381_cast")];
            tensor<int32, [2]> var_27552 = const()[name = tensor<string, []>("op_27552"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_27554 = const()[name = tensor<string, []>("op_27554"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_27556_pad_type_0 = const()[name = tensor<string, []>("op_27556_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_27556_pad_0 = const()[name = tensor<string, []>("op_27556_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2173354880)))];
            tensor<fp16, [10240]> mid_block_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2199569344)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_27556_cast = conv(bias = mid_block_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16, dilations = var_27554, groups = var_21105, pad = var_27556_pad_0, pad_type = var_27556_pad_type_0, strides = var_27552, weight = mid_block_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16, x = input_381_cast)[name = tensor<string, []>("op_27556_cast")];
            tensor<int32, [2]> var_27557_split_sizes_0 = const()[name = tensor<string, []>("op_27557_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_27557_axis_0 = const()[name = tensor<string, []>("op_27557_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_27557_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_27557_cast_1 = split(axis = var_27557_axis_0, split_sizes = var_27557_split_sizes_0, x = var_27556_cast)[name = tensor<string, []>("op_27557_cast")];
            tensor<string, []> var_27559_mode_0 = const()[name = tensor<string, []>("op_27559_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_27559_cast = gelu(mode = var_27559_mode_0, x = var_27557_cast_1)[name = tensor<string, []>("op_27559_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_383_cast = mul(x = var_27557_cast_0, y = var_27559_cast)[name = tensor<string, []>("input_383_cast")];
            tensor<int32, [2]> var_27563 = const()[name = tensor<string, []>("op_27563"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_27565 = const()[name = tensor<string, []>("op_27565"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_27567_pad_type_0 = const()[name = tensor<string, []>("op_27567_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_27567_pad_0 = const()[name = tensor<string, []>("op_27567_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> mid_block_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2199589888)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2212697152)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_27567_cast = conv(bias = mid_block_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16, dilations = var_27565, groups = var_21105, pad = var_27567_pad_0, pad_type = var_27567_pad_type_0, strides = var_27563, weight = mid_block_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16, x = input_383_cast)[name = tensor<string, []>("op_27567_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_187_cast = add(x = var_27567_cast, y = inputs_185_cast)[name = tensor<string, []>("inputs_187_cast")];
            tensor<int32, [1]> var_27577 = const()[name = tensor<string, []>("op_27577"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_187_cast = reduce_mean(axes = var_27577, keep_dims = var_21100, x = inputs_187_cast)[name = tensor<string, []>("channels_mean_187_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_187_cast = sub(x = inputs_187_cast, y = channels_mean_187_cast)[name = tensor<string, []>("zero_mean_187_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_187_cast = mul(x = zero_mean_187_cast, y = zero_mean_187_cast)[name = tensor<string, []>("zero_mean_sq_187_cast")];
            tensor<int32, [1]> var_27581 = const()[name = tensor<string, []>("op_27581"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_27582_cast = reduce_mean(axes = var_27581, keep_dims = var_21100, x = zero_mean_sq_187_cast)[name = tensor<string, []>("op_27582_cast")];
            tensor<fp16, []> var_27583_to_fp16 = const()[name = tensor<string, []>("op_27583_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_27584_cast = add(x = var_27582_cast, y = var_27583_to_fp16)[name = tensor<string, []>("op_27584_cast")];
            tensor<fp16, []> denom_187_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_187_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_187_cast = rsqrt(epsilon = denom_187_epsilon_0_to_fp16, x = var_27584_cast)[name = tensor<string, []>("denom_187_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_187_cast = mul(x = zero_mean_187_cast, y = denom_187_cast)[name = tensor<string, []>("out_187_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_27588_to_fp16 = const()[name = tensor<string, []>("op_27588_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2212699776)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_27589_cast = add(x = out_187_cast, y = var_27588_to_fp16)[name = tensor<string, []>("op_27589_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_27591_to_fp16 = const()[name = tensor<string, []>("op_27591_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2212702400)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_251_cast = mul(x = var_27589_cast, y = var_27591_to_fp16)[name = tensor<string, []>("hidden_states_251_cast")];
            tensor<int32, [2]> var_27598 = const()[name = tensor<string, []>("op_27598"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_27600 = const()[name = tensor<string, []>("op_27600"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_125_pad_type_0 = const()[name = tensor<string, []>("q_125_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_125_pad_0 = const()[name = tensor<string, []>("q_125_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2212705024)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_125_cast = conv(dilations = var_27600, groups = var_21105, pad = q_125_pad_0, pad_type = q_125_pad_type_0, strides = var_27598, weight = mid_block_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16, x = hidden_states_251_cast)[name = tensor<string, []>("q_125_cast")];
            tensor<int32, [2]> var_27604 = const()[name = tensor<string, []>("op_27604"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_27606 = const()[name = tensor<string, []>("op_27606"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_249_pad_type_0 = const()[name = tensor<string, []>("k_249_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_249_pad_0 = const()[name = tensor<string, []>("k_249_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2215981888)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_249_cast = conv(dilations = var_27606, groups = var_21105, pad = k_249_pad_0, pad_type = k_249_pad_type_0, strides = var_27604, weight = mid_block_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16, x = hidden_states_251_cast)[name = tensor<string, []>("k_249_cast")];
            tensor<int32, [2]> var_27610 = const()[name = tensor<string, []>("op_27610"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_27612 = const()[name = tensor<string, []>("op_27612"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_125_pad_type_0 = const()[name = tensor<string, []>("v_125_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_125_pad_0 = const()[name = tensor<string, []>("v_125_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2219258752)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_125_cast = conv(dilations = var_27612, groups = var_21105, pad = v_125_pad_0, pad_type = v_125_pad_type_0, strides = var_27610, weight = mid_block_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16, x = hidden_states_251_cast)[name = tensor<string, []>("v_125_cast")];
            tensor<int32, [4]> var_27616_begin_0 = const()[name = tensor<string, []>("op_27616_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_27616_end_0 = const()[name = tensor<string, []>("op_27616_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_27616_end_mask_0 = const()[name = tensor<string, []>("op_27616_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27616_cast = slice_by_index(begin = var_27616_begin_0, end = var_27616_end_0, end_mask = var_27616_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27616_cast")];
            tensor<int32, [4]> var_27620_begin_0 = const()[name = tensor<string, []>("op_27620_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_27620_end_0 = const()[name = tensor<string, []>("op_27620_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_27620_end_mask_0 = const()[name = tensor<string, []>("op_27620_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27620_cast = slice_by_index(begin = var_27620_begin_0, end = var_27620_end_0, end_mask = var_27620_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27620_cast")];
            tensor<int32, [4]> var_27624_begin_0 = const()[name = tensor<string, []>("op_27624_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_27624_end_0 = const()[name = tensor<string, []>("op_27624_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_27624_end_mask_0 = const()[name = tensor<string, []>("op_27624_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27624_cast = slice_by_index(begin = var_27624_begin_0, end = var_27624_end_0, end_mask = var_27624_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27624_cast")];
            tensor<int32, [4]> var_27628_begin_0 = const()[name = tensor<string, []>("op_27628_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_27628_end_0 = const()[name = tensor<string, []>("op_27628_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_27628_end_mask_0 = const()[name = tensor<string, []>("op_27628_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27628_cast = slice_by_index(begin = var_27628_begin_0, end = var_27628_end_0, end_mask = var_27628_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27628_cast")];
            tensor<int32, [4]> var_27632_begin_0 = const()[name = tensor<string, []>("op_27632_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_27632_end_0 = const()[name = tensor<string, []>("op_27632_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_27632_end_mask_0 = const()[name = tensor<string, []>("op_27632_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27632_cast = slice_by_index(begin = var_27632_begin_0, end = var_27632_end_0, end_mask = var_27632_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27632_cast")];
            tensor<int32, [4]> var_27636_begin_0 = const()[name = tensor<string, []>("op_27636_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_27636_end_0 = const()[name = tensor<string, []>("op_27636_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_27636_end_mask_0 = const()[name = tensor<string, []>("op_27636_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27636_cast = slice_by_index(begin = var_27636_begin_0, end = var_27636_end_0, end_mask = var_27636_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27636_cast")];
            tensor<int32, [4]> var_27640_begin_0 = const()[name = tensor<string, []>("op_27640_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_27640_end_0 = const()[name = tensor<string, []>("op_27640_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_27640_end_mask_0 = const()[name = tensor<string, []>("op_27640_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27640_cast = slice_by_index(begin = var_27640_begin_0, end = var_27640_end_0, end_mask = var_27640_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27640_cast")];
            tensor<int32, [4]> var_27644_begin_0 = const()[name = tensor<string, []>("op_27644_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_27644_end_0 = const()[name = tensor<string, []>("op_27644_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_27644_end_mask_0 = const()[name = tensor<string, []>("op_27644_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27644_cast = slice_by_index(begin = var_27644_begin_0, end = var_27644_end_0, end_mask = var_27644_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27644_cast")];
            tensor<int32, [4]> var_27648_begin_0 = const()[name = tensor<string, []>("op_27648_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_27648_end_0 = const()[name = tensor<string, []>("op_27648_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_27648_end_mask_0 = const()[name = tensor<string, []>("op_27648_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27648_cast = slice_by_index(begin = var_27648_begin_0, end = var_27648_end_0, end_mask = var_27648_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27648_cast")];
            tensor<int32, [4]> var_27652_begin_0 = const()[name = tensor<string, []>("op_27652_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_27652_end_0 = const()[name = tensor<string, []>("op_27652_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_27652_end_mask_0 = const()[name = tensor<string, []>("op_27652_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27652_cast = slice_by_index(begin = var_27652_begin_0, end = var_27652_end_0, end_mask = var_27652_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27652_cast")];
            tensor<int32, [4]> var_27656_begin_0 = const()[name = tensor<string, []>("op_27656_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_27656_end_0 = const()[name = tensor<string, []>("op_27656_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_27656_end_mask_0 = const()[name = tensor<string, []>("op_27656_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27656_cast = slice_by_index(begin = var_27656_begin_0, end = var_27656_end_0, end_mask = var_27656_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27656_cast")];
            tensor<int32, [4]> var_27660_begin_0 = const()[name = tensor<string, []>("op_27660_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_27660_end_0 = const()[name = tensor<string, []>("op_27660_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_27660_end_mask_0 = const()[name = tensor<string, []>("op_27660_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27660_cast = slice_by_index(begin = var_27660_begin_0, end = var_27660_end_0, end_mask = var_27660_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27660_cast")];
            tensor<int32, [4]> var_27664_begin_0 = const()[name = tensor<string, []>("op_27664_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_27664_end_0 = const()[name = tensor<string, []>("op_27664_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_27664_end_mask_0 = const()[name = tensor<string, []>("op_27664_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27664_cast = slice_by_index(begin = var_27664_begin_0, end = var_27664_end_0, end_mask = var_27664_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27664_cast")];
            tensor<int32, [4]> var_27668_begin_0 = const()[name = tensor<string, []>("op_27668_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_27668_end_0 = const()[name = tensor<string, []>("op_27668_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_27668_end_mask_0 = const()[name = tensor<string, []>("op_27668_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27668_cast = slice_by_index(begin = var_27668_begin_0, end = var_27668_end_0, end_mask = var_27668_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27668_cast")];
            tensor<int32, [4]> var_27672_begin_0 = const()[name = tensor<string, []>("op_27672_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_27672_end_0 = const()[name = tensor<string, []>("op_27672_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_27672_end_mask_0 = const()[name = tensor<string, []>("op_27672_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27672_cast = slice_by_index(begin = var_27672_begin_0, end = var_27672_end_0, end_mask = var_27672_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27672_cast")];
            tensor<int32, [4]> var_27676_begin_0 = const()[name = tensor<string, []>("op_27676_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_27676_end_0 = const()[name = tensor<string, []>("op_27676_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_27676_end_mask_0 = const()[name = tensor<string, []>("op_27676_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27676_cast = slice_by_index(begin = var_27676_begin_0, end = var_27676_end_0, end_mask = var_27676_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27676_cast")];
            tensor<int32, [4]> var_27680_begin_0 = const()[name = tensor<string, []>("op_27680_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_27680_end_0 = const()[name = tensor<string, []>("op_27680_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_27680_end_mask_0 = const()[name = tensor<string, []>("op_27680_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27680_cast = slice_by_index(begin = var_27680_begin_0, end = var_27680_end_0, end_mask = var_27680_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27680_cast")];
            tensor<int32, [4]> var_27684_begin_0 = const()[name = tensor<string, []>("op_27684_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_27684_end_0 = const()[name = tensor<string, []>("op_27684_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_27684_end_mask_0 = const()[name = tensor<string, []>("op_27684_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27684_cast = slice_by_index(begin = var_27684_begin_0, end = var_27684_end_0, end_mask = var_27684_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27684_cast")];
            tensor<int32, [4]> var_27688_begin_0 = const()[name = tensor<string, []>("op_27688_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_27688_end_0 = const()[name = tensor<string, []>("op_27688_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_27688_end_mask_0 = const()[name = tensor<string, []>("op_27688_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27688_cast = slice_by_index(begin = var_27688_begin_0, end = var_27688_end_0, end_mask = var_27688_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27688_cast")];
            tensor<int32, [4]> var_27692_begin_0 = const()[name = tensor<string, []>("op_27692_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_27692_end_0 = const()[name = tensor<string, []>("op_27692_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_27692_end_mask_0 = const()[name = tensor<string, []>("op_27692_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27692_cast = slice_by_index(begin = var_27692_begin_0, end = var_27692_end_0, end_mask = var_27692_end_mask_0, x = q_125_cast)[name = tensor<string, []>("op_27692_cast")];
            tensor<int32, [4]> k_251_perm_0 = const()[name = tensor<string, []>("k_251_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_27699_begin_0 = const()[name = tensor<string, []>("op_27699_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_27699_end_0 = const()[name = tensor<string, []>("op_27699_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_27699_end_mask_0 = const()[name = tensor<string, []>("op_27699_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_77 = transpose(perm = k_251_perm_0, x = k_249_cast)[name = tensor<string, []>("transpose_77")];
            tensor<fp16, [2, 1024, 1, 64]> var_27699_cast = slice_by_index(begin = var_27699_begin_0, end = var_27699_end_0, end_mask = var_27699_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27699_cast")];
            tensor<int32, [4]> var_27703_begin_0 = const()[name = tensor<string, []>("op_27703_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_27703_end_0 = const()[name = tensor<string, []>("op_27703_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_27703_end_mask_0 = const()[name = tensor<string, []>("op_27703_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_27703_cast = slice_by_index(begin = var_27703_begin_0, end = var_27703_end_0, end_mask = var_27703_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27703_cast")];
            tensor<int32, [4]> var_27707_begin_0 = const()[name = tensor<string, []>("op_27707_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_27707_end_0 = const()[name = tensor<string, []>("op_27707_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_27707_end_mask_0 = const()[name = tensor<string, []>("op_27707_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_27707_cast = slice_by_index(begin = var_27707_begin_0, end = var_27707_end_0, end_mask = var_27707_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27707_cast")];
            tensor<int32, [4]> var_27711_begin_0 = const()[name = tensor<string, []>("op_27711_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_27711_end_0 = const()[name = tensor<string, []>("op_27711_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_27711_end_mask_0 = const()[name = tensor<string, []>("op_27711_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_27711_cast = slice_by_index(begin = var_27711_begin_0, end = var_27711_end_0, end_mask = var_27711_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27711_cast")];
            tensor<int32, [4]> var_27715_begin_0 = const()[name = tensor<string, []>("op_27715_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_27715_end_0 = const()[name = tensor<string, []>("op_27715_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_27715_end_mask_0 = const()[name = tensor<string, []>("op_27715_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_27715_cast = slice_by_index(begin = var_27715_begin_0, end = var_27715_end_0, end_mask = var_27715_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27715_cast")];
            tensor<int32, [4]> var_27719_begin_0 = const()[name = tensor<string, []>("op_27719_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_27719_end_0 = const()[name = tensor<string, []>("op_27719_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_27719_end_mask_0 = const()[name = tensor<string, []>("op_27719_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_27719_cast = slice_by_index(begin = var_27719_begin_0, end = var_27719_end_0, end_mask = var_27719_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27719_cast")];
            tensor<int32, [4]> var_27723_begin_0 = const()[name = tensor<string, []>("op_27723_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_27723_end_0 = const()[name = tensor<string, []>("op_27723_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_27723_end_mask_0 = const()[name = tensor<string, []>("op_27723_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_27723_cast = slice_by_index(begin = var_27723_begin_0, end = var_27723_end_0, end_mask = var_27723_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27723_cast")];
            tensor<int32, [4]> var_27727_begin_0 = const()[name = tensor<string, []>("op_27727_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_27727_end_0 = const()[name = tensor<string, []>("op_27727_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_27727_end_mask_0 = const()[name = tensor<string, []>("op_27727_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_27727_cast = slice_by_index(begin = var_27727_begin_0, end = var_27727_end_0, end_mask = var_27727_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27727_cast")];
            tensor<int32, [4]> var_27731_begin_0 = const()[name = tensor<string, []>("op_27731_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_27731_end_0 = const()[name = tensor<string, []>("op_27731_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_27731_end_mask_0 = const()[name = tensor<string, []>("op_27731_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_27731_cast = slice_by_index(begin = var_27731_begin_0, end = var_27731_end_0, end_mask = var_27731_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27731_cast")];
            tensor<int32, [4]> var_27735_begin_0 = const()[name = tensor<string, []>("op_27735_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_27735_end_0 = const()[name = tensor<string, []>("op_27735_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_27735_end_mask_0 = const()[name = tensor<string, []>("op_27735_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_27735_cast = slice_by_index(begin = var_27735_begin_0, end = var_27735_end_0, end_mask = var_27735_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27735_cast")];
            tensor<int32, [4]> var_27739_begin_0 = const()[name = tensor<string, []>("op_27739_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_27739_end_0 = const()[name = tensor<string, []>("op_27739_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_27739_end_mask_0 = const()[name = tensor<string, []>("op_27739_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_27739_cast = slice_by_index(begin = var_27739_begin_0, end = var_27739_end_0, end_mask = var_27739_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27739_cast")];
            tensor<int32, [4]> var_27743_begin_0 = const()[name = tensor<string, []>("op_27743_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_27743_end_0 = const()[name = tensor<string, []>("op_27743_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_27743_end_mask_0 = const()[name = tensor<string, []>("op_27743_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_27743_cast = slice_by_index(begin = var_27743_begin_0, end = var_27743_end_0, end_mask = var_27743_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27743_cast")];
            tensor<int32, [4]> var_27747_begin_0 = const()[name = tensor<string, []>("op_27747_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_27747_end_0 = const()[name = tensor<string, []>("op_27747_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_27747_end_mask_0 = const()[name = tensor<string, []>("op_27747_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_27747_cast = slice_by_index(begin = var_27747_begin_0, end = var_27747_end_0, end_mask = var_27747_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27747_cast")];
            tensor<int32, [4]> var_27751_begin_0 = const()[name = tensor<string, []>("op_27751_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_27751_end_0 = const()[name = tensor<string, []>("op_27751_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_27751_end_mask_0 = const()[name = tensor<string, []>("op_27751_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_27751_cast = slice_by_index(begin = var_27751_begin_0, end = var_27751_end_0, end_mask = var_27751_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27751_cast")];
            tensor<int32, [4]> var_27755_begin_0 = const()[name = tensor<string, []>("op_27755_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_27755_end_0 = const()[name = tensor<string, []>("op_27755_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_27755_end_mask_0 = const()[name = tensor<string, []>("op_27755_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_27755_cast = slice_by_index(begin = var_27755_begin_0, end = var_27755_end_0, end_mask = var_27755_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27755_cast")];
            tensor<int32, [4]> var_27759_begin_0 = const()[name = tensor<string, []>("op_27759_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_27759_end_0 = const()[name = tensor<string, []>("op_27759_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_27759_end_mask_0 = const()[name = tensor<string, []>("op_27759_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_27759_cast = slice_by_index(begin = var_27759_begin_0, end = var_27759_end_0, end_mask = var_27759_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27759_cast")];
            tensor<int32, [4]> var_27763_begin_0 = const()[name = tensor<string, []>("op_27763_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_27763_end_0 = const()[name = tensor<string, []>("op_27763_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_27763_end_mask_0 = const()[name = tensor<string, []>("op_27763_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_27763_cast = slice_by_index(begin = var_27763_begin_0, end = var_27763_end_0, end_mask = var_27763_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27763_cast")];
            tensor<int32, [4]> var_27767_begin_0 = const()[name = tensor<string, []>("op_27767_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_27767_end_0 = const()[name = tensor<string, []>("op_27767_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_27767_end_mask_0 = const()[name = tensor<string, []>("op_27767_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_27767_cast = slice_by_index(begin = var_27767_begin_0, end = var_27767_end_0, end_mask = var_27767_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27767_cast")];
            tensor<int32, [4]> var_27771_begin_0 = const()[name = tensor<string, []>("op_27771_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_27771_end_0 = const()[name = tensor<string, []>("op_27771_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_27771_end_mask_0 = const()[name = tensor<string, []>("op_27771_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_27771_cast = slice_by_index(begin = var_27771_begin_0, end = var_27771_end_0, end_mask = var_27771_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27771_cast")];
            tensor<int32, [4]> var_27775_begin_0 = const()[name = tensor<string, []>("op_27775_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_27775_end_0 = const()[name = tensor<string, []>("op_27775_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_27775_end_mask_0 = const()[name = tensor<string, []>("op_27775_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_27775_cast = slice_by_index(begin = var_27775_begin_0, end = var_27775_end_0, end_mask = var_27775_end_mask_0, x = transpose_77)[name = tensor<string, []>("op_27775_cast")];
            tensor<int32, [4]> var_27777_begin_0 = const()[name = tensor<string, []>("op_27777_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_27777_end_0 = const()[name = tensor<string, []>("op_27777_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_27777_end_mask_0 = const()[name = tensor<string, []>("op_27777_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27777_cast = slice_by_index(begin = var_27777_begin_0, end = var_27777_end_0, end_mask = var_27777_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27777_cast")];
            tensor<int32, [4]> var_27781_begin_0 = const()[name = tensor<string, []>("op_27781_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_27781_end_0 = const()[name = tensor<string, []>("op_27781_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_27781_end_mask_0 = const()[name = tensor<string, []>("op_27781_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27781_cast = slice_by_index(begin = var_27781_begin_0, end = var_27781_end_0, end_mask = var_27781_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27781_cast")];
            tensor<int32, [4]> var_27785_begin_0 = const()[name = tensor<string, []>("op_27785_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_27785_end_0 = const()[name = tensor<string, []>("op_27785_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_27785_end_mask_0 = const()[name = tensor<string, []>("op_27785_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27785_cast = slice_by_index(begin = var_27785_begin_0, end = var_27785_end_0, end_mask = var_27785_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27785_cast")];
            tensor<int32, [4]> var_27789_begin_0 = const()[name = tensor<string, []>("op_27789_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_27789_end_0 = const()[name = tensor<string, []>("op_27789_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_27789_end_mask_0 = const()[name = tensor<string, []>("op_27789_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27789_cast = slice_by_index(begin = var_27789_begin_0, end = var_27789_end_0, end_mask = var_27789_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27789_cast")];
            tensor<int32, [4]> var_27793_begin_0 = const()[name = tensor<string, []>("op_27793_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_27793_end_0 = const()[name = tensor<string, []>("op_27793_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_27793_end_mask_0 = const()[name = tensor<string, []>("op_27793_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27793_cast = slice_by_index(begin = var_27793_begin_0, end = var_27793_end_0, end_mask = var_27793_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27793_cast")];
            tensor<int32, [4]> var_27797_begin_0 = const()[name = tensor<string, []>("op_27797_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_27797_end_0 = const()[name = tensor<string, []>("op_27797_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_27797_end_mask_0 = const()[name = tensor<string, []>("op_27797_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27797_cast = slice_by_index(begin = var_27797_begin_0, end = var_27797_end_0, end_mask = var_27797_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27797_cast")];
            tensor<int32, [4]> var_27801_begin_0 = const()[name = tensor<string, []>("op_27801_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_27801_end_0 = const()[name = tensor<string, []>("op_27801_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_27801_end_mask_0 = const()[name = tensor<string, []>("op_27801_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27801_cast = slice_by_index(begin = var_27801_begin_0, end = var_27801_end_0, end_mask = var_27801_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27801_cast")];
            tensor<int32, [4]> var_27805_begin_0 = const()[name = tensor<string, []>("op_27805_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_27805_end_0 = const()[name = tensor<string, []>("op_27805_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_27805_end_mask_0 = const()[name = tensor<string, []>("op_27805_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27805_cast = slice_by_index(begin = var_27805_begin_0, end = var_27805_end_0, end_mask = var_27805_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27805_cast")];
            tensor<int32, [4]> var_27809_begin_0 = const()[name = tensor<string, []>("op_27809_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_27809_end_0 = const()[name = tensor<string, []>("op_27809_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_27809_end_mask_0 = const()[name = tensor<string, []>("op_27809_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27809_cast = slice_by_index(begin = var_27809_begin_0, end = var_27809_end_0, end_mask = var_27809_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27809_cast")];
            tensor<int32, [4]> var_27813_begin_0 = const()[name = tensor<string, []>("op_27813_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_27813_end_0 = const()[name = tensor<string, []>("op_27813_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_27813_end_mask_0 = const()[name = tensor<string, []>("op_27813_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27813_cast = slice_by_index(begin = var_27813_begin_0, end = var_27813_end_0, end_mask = var_27813_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27813_cast")];
            tensor<int32, [4]> var_27817_begin_0 = const()[name = tensor<string, []>("op_27817_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_27817_end_0 = const()[name = tensor<string, []>("op_27817_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_27817_end_mask_0 = const()[name = tensor<string, []>("op_27817_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27817_cast = slice_by_index(begin = var_27817_begin_0, end = var_27817_end_0, end_mask = var_27817_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27817_cast")];
            tensor<int32, [4]> var_27821_begin_0 = const()[name = tensor<string, []>("op_27821_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_27821_end_0 = const()[name = tensor<string, []>("op_27821_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_27821_end_mask_0 = const()[name = tensor<string, []>("op_27821_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27821_cast = slice_by_index(begin = var_27821_begin_0, end = var_27821_end_0, end_mask = var_27821_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27821_cast")];
            tensor<int32, [4]> var_27825_begin_0 = const()[name = tensor<string, []>("op_27825_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_27825_end_0 = const()[name = tensor<string, []>("op_27825_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_27825_end_mask_0 = const()[name = tensor<string, []>("op_27825_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27825_cast = slice_by_index(begin = var_27825_begin_0, end = var_27825_end_0, end_mask = var_27825_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27825_cast")];
            tensor<int32, [4]> var_27829_begin_0 = const()[name = tensor<string, []>("op_27829_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_27829_end_0 = const()[name = tensor<string, []>("op_27829_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_27829_end_mask_0 = const()[name = tensor<string, []>("op_27829_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27829_cast = slice_by_index(begin = var_27829_begin_0, end = var_27829_end_0, end_mask = var_27829_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27829_cast")];
            tensor<int32, [4]> var_27833_begin_0 = const()[name = tensor<string, []>("op_27833_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_27833_end_0 = const()[name = tensor<string, []>("op_27833_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_27833_end_mask_0 = const()[name = tensor<string, []>("op_27833_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27833_cast = slice_by_index(begin = var_27833_begin_0, end = var_27833_end_0, end_mask = var_27833_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27833_cast")];
            tensor<int32, [4]> var_27837_begin_0 = const()[name = tensor<string, []>("op_27837_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_27837_end_0 = const()[name = tensor<string, []>("op_27837_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_27837_end_mask_0 = const()[name = tensor<string, []>("op_27837_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27837_cast = slice_by_index(begin = var_27837_begin_0, end = var_27837_end_0, end_mask = var_27837_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27837_cast")];
            tensor<int32, [4]> var_27841_begin_0 = const()[name = tensor<string, []>("op_27841_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_27841_end_0 = const()[name = tensor<string, []>("op_27841_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_27841_end_mask_0 = const()[name = tensor<string, []>("op_27841_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27841_cast = slice_by_index(begin = var_27841_begin_0, end = var_27841_end_0, end_mask = var_27841_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27841_cast")];
            tensor<int32, [4]> var_27845_begin_0 = const()[name = tensor<string, []>("op_27845_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_27845_end_0 = const()[name = tensor<string, []>("op_27845_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_27845_end_mask_0 = const()[name = tensor<string, []>("op_27845_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27845_cast = slice_by_index(begin = var_27845_begin_0, end = var_27845_end_0, end_mask = var_27845_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27845_cast")];
            tensor<int32, [4]> var_27849_begin_0 = const()[name = tensor<string, []>("op_27849_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_27849_end_0 = const()[name = tensor<string, []>("op_27849_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_27849_end_mask_0 = const()[name = tensor<string, []>("op_27849_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27849_cast = slice_by_index(begin = var_27849_begin_0, end = var_27849_end_0, end_mask = var_27849_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27849_cast")];
            tensor<int32, [4]> var_27853_begin_0 = const()[name = tensor<string, []>("op_27853_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_27853_end_0 = const()[name = tensor<string, []>("op_27853_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_27853_end_mask_0 = const()[name = tensor<string, []>("op_27853_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_27853_cast = slice_by_index(begin = var_27853_begin_0, end = var_27853_end_0, end_mask = var_27853_end_mask_0, x = v_125_cast)[name = tensor<string, []>("op_27853_cast")];
            tensor<string, []> var_27857_equation_0 = const()[name = tensor<string, []>("op_27857_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27857_cast = einsum(equation = var_27857_equation_0, values = (var_27699_cast, var_27616_cast))[name = tensor<string, []>("op_27857_cast")];
            tensor<fp16, []> var_27858_to_fp16 = const()[name = tensor<string, []>("op_27858_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2321_cast = mul(x = var_27857_cast, y = var_27858_to_fp16)[name = tensor<string, []>("aw_2321_cast")];
            tensor<string, []> var_27861_equation_0 = const()[name = tensor<string, []>("op_27861_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27861_cast = einsum(equation = var_27861_equation_0, values = (var_27703_cast, var_27620_cast))[name = tensor<string, []>("op_27861_cast")];
            tensor<fp16, []> var_27862_to_fp16 = const()[name = tensor<string, []>("op_27862_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2323_cast = mul(x = var_27861_cast, y = var_27862_to_fp16)[name = tensor<string, []>("aw_2323_cast")];
            tensor<string, []> var_27865_equation_0 = const()[name = tensor<string, []>("op_27865_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27865_cast = einsum(equation = var_27865_equation_0, values = (var_27707_cast, var_27624_cast))[name = tensor<string, []>("op_27865_cast")];
            tensor<fp16, []> var_27866_to_fp16 = const()[name = tensor<string, []>("op_27866_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2325_cast = mul(x = var_27865_cast, y = var_27866_to_fp16)[name = tensor<string, []>("aw_2325_cast")];
            tensor<string, []> var_27869_equation_0 = const()[name = tensor<string, []>("op_27869_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27869_cast = einsum(equation = var_27869_equation_0, values = (var_27711_cast, var_27628_cast))[name = tensor<string, []>("op_27869_cast")];
            tensor<fp16, []> var_27870_to_fp16 = const()[name = tensor<string, []>("op_27870_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2327_cast = mul(x = var_27869_cast, y = var_27870_to_fp16)[name = tensor<string, []>("aw_2327_cast")];
            tensor<string, []> var_27873_equation_0 = const()[name = tensor<string, []>("op_27873_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27873_cast = einsum(equation = var_27873_equation_0, values = (var_27715_cast, var_27632_cast))[name = tensor<string, []>("op_27873_cast")];
            tensor<fp16, []> var_27874_to_fp16 = const()[name = tensor<string, []>("op_27874_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2329_cast = mul(x = var_27873_cast, y = var_27874_to_fp16)[name = tensor<string, []>("aw_2329_cast")];
            tensor<string, []> var_27877_equation_0 = const()[name = tensor<string, []>("op_27877_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27877_cast = einsum(equation = var_27877_equation_0, values = (var_27719_cast, var_27636_cast))[name = tensor<string, []>("op_27877_cast")];
            tensor<fp16, []> var_27878_to_fp16 = const()[name = tensor<string, []>("op_27878_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2331_cast = mul(x = var_27877_cast, y = var_27878_to_fp16)[name = tensor<string, []>("aw_2331_cast")];
            tensor<string, []> var_27881_equation_0 = const()[name = tensor<string, []>("op_27881_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27881_cast = einsum(equation = var_27881_equation_0, values = (var_27723_cast, var_27640_cast))[name = tensor<string, []>("op_27881_cast")];
            tensor<fp16, []> var_27882_to_fp16 = const()[name = tensor<string, []>("op_27882_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2333_cast = mul(x = var_27881_cast, y = var_27882_to_fp16)[name = tensor<string, []>("aw_2333_cast")];
            tensor<string, []> var_27885_equation_0 = const()[name = tensor<string, []>("op_27885_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27885_cast = einsum(equation = var_27885_equation_0, values = (var_27727_cast, var_27644_cast))[name = tensor<string, []>("op_27885_cast")];
            tensor<fp16, []> var_27886_to_fp16 = const()[name = tensor<string, []>("op_27886_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2335_cast = mul(x = var_27885_cast, y = var_27886_to_fp16)[name = tensor<string, []>("aw_2335_cast")];
            tensor<string, []> var_27889_equation_0 = const()[name = tensor<string, []>("op_27889_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27889_cast = einsum(equation = var_27889_equation_0, values = (var_27731_cast, var_27648_cast))[name = tensor<string, []>("op_27889_cast")];
            tensor<fp16, []> var_27890_to_fp16 = const()[name = tensor<string, []>("op_27890_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2337_cast = mul(x = var_27889_cast, y = var_27890_to_fp16)[name = tensor<string, []>("aw_2337_cast")];
            tensor<string, []> var_27893_equation_0 = const()[name = tensor<string, []>("op_27893_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27893_cast = einsum(equation = var_27893_equation_0, values = (var_27735_cast, var_27652_cast))[name = tensor<string, []>("op_27893_cast")];
            tensor<fp16, []> var_27894_to_fp16 = const()[name = tensor<string, []>("op_27894_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2339_cast = mul(x = var_27893_cast, y = var_27894_to_fp16)[name = tensor<string, []>("aw_2339_cast")];
            tensor<string, []> var_27897_equation_0 = const()[name = tensor<string, []>("op_27897_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27897_cast = einsum(equation = var_27897_equation_0, values = (var_27739_cast, var_27656_cast))[name = tensor<string, []>("op_27897_cast")];
            tensor<fp16, []> var_27898_to_fp16 = const()[name = tensor<string, []>("op_27898_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2341_cast = mul(x = var_27897_cast, y = var_27898_to_fp16)[name = tensor<string, []>("aw_2341_cast")];
            tensor<string, []> var_27901_equation_0 = const()[name = tensor<string, []>("op_27901_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27901_cast = einsum(equation = var_27901_equation_0, values = (var_27743_cast, var_27660_cast))[name = tensor<string, []>("op_27901_cast")];
            tensor<fp16, []> var_27902_to_fp16 = const()[name = tensor<string, []>("op_27902_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2343_cast = mul(x = var_27901_cast, y = var_27902_to_fp16)[name = tensor<string, []>("aw_2343_cast")];
            tensor<string, []> var_27905_equation_0 = const()[name = tensor<string, []>("op_27905_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27905_cast = einsum(equation = var_27905_equation_0, values = (var_27747_cast, var_27664_cast))[name = tensor<string, []>("op_27905_cast")];
            tensor<fp16, []> var_27906_to_fp16 = const()[name = tensor<string, []>("op_27906_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2345_cast = mul(x = var_27905_cast, y = var_27906_to_fp16)[name = tensor<string, []>("aw_2345_cast")];
            tensor<string, []> var_27909_equation_0 = const()[name = tensor<string, []>("op_27909_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27909_cast = einsum(equation = var_27909_equation_0, values = (var_27751_cast, var_27668_cast))[name = tensor<string, []>("op_27909_cast")];
            tensor<fp16, []> var_27910_to_fp16 = const()[name = tensor<string, []>("op_27910_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2347_cast = mul(x = var_27909_cast, y = var_27910_to_fp16)[name = tensor<string, []>("aw_2347_cast")];
            tensor<string, []> var_27913_equation_0 = const()[name = tensor<string, []>("op_27913_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27913_cast = einsum(equation = var_27913_equation_0, values = (var_27755_cast, var_27672_cast))[name = tensor<string, []>("op_27913_cast")];
            tensor<fp16, []> var_27914_to_fp16 = const()[name = tensor<string, []>("op_27914_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2349_cast = mul(x = var_27913_cast, y = var_27914_to_fp16)[name = tensor<string, []>("aw_2349_cast")];
            tensor<string, []> var_27917_equation_0 = const()[name = tensor<string, []>("op_27917_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27917_cast = einsum(equation = var_27917_equation_0, values = (var_27759_cast, var_27676_cast))[name = tensor<string, []>("op_27917_cast")];
            tensor<fp16, []> var_27918_to_fp16 = const()[name = tensor<string, []>("op_27918_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2351_cast = mul(x = var_27917_cast, y = var_27918_to_fp16)[name = tensor<string, []>("aw_2351_cast")];
            tensor<string, []> var_27921_equation_0 = const()[name = tensor<string, []>("op_27921_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27921_cast = einsum(equation = var_27921_equation_0, values = (var_27763_cast, var_27680_cast))[name = tensor<string, []>("op_27921_cast")];
            tensor<fp16, []> var_27922_to_fp16 = const()[name = tensor<string, []>("op_27922_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2353_cast = mul(x = var_27921_cast, y = var_27922_to_fp16)[name = tensor<string, []>("aw_2353_cast")];
            tensor<string, []> var_27925_equation_0 = const()[name = tensor<string, []>("op_27925_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27925_cast = einsum(equation = var_27925_equation_0, values = (var_27767_cast, var_27684_cast))[name = tensor<string, []>("op_27925_cast")];
            tensor<fp16, []> var_27926_to_fp16 = const()[name = tensor<string, []>("op_27926_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2355_cast = mul(x = var_27925_cast, y = var_27926_to_fp16)[name = tensor<string, []>("aw_2355_cast")];
            tensor<string, []> var_27929_equation_0 = const()[name = tensor<string, []>("op_27929_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27929_cast = einsum(equation = var_27929_equation_0, values = (var_27771_cast, var_27688_cast))[name = tensor<string, []>("op_27929_cast")];
            tensor<fp16, []> var_27930_to_fp16 = const()[name = tensor<string, []>("op_27930_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2357_cast = mul(x = var_27929_cast, y = var_27930_to_fp16)[name = tensor<string, []>("aw_2357_cast")];
            tensor<string, []> var_27933_equation_0 = const()[name = tensor<string, []>("op_27933_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27933_cast = einsum(equation = var_27933_equation_0, values = (var_27775_cast, var_27692_cast))[name = tensor<string, []>("op_27933_cast")];
            tensor<fp16, []> var_27934_to_fp16 = const()[name = tensor<string, []>("op_27934_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2359_cast = mul(x = var_27933_cast, y = var_27934_to_fp16)[name = tensor<string, []>("aw_2359_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27936_cast = softmax(axis = var_21105, x = aw_2321_cast)[name = tensor<string, []>("op_27936_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27937_cast = softmax(axis = var_21105, x = aw_2323_cast)[name = tensor<string, []>("op_27937_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27938_cast = softmax(axis = var_21105, x = aw_2325_cast)[name = tensor<string, []>("op_27938_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27939_cast = softmax(axis = var_21105, x = aw_2327_cast)[name = tensor<string, []>("op_27939_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27940_cast = softmax(axis = var_21105, x = aw_2329_cast)[name = tensor<string, []>("op_27940_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27941_cast = softmax(axis = var_21105, x = aw_2331_cast)[name = tensor<string, []>("op_27941_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27942_cast = softmax(axis = var_21105, x = aw_2333_cast)[name = tensor<string, []>("op_27942_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27943_cast = softmax(axis = var_21105, x = aw_2335_cast)[name = tensor<string, []>("op_27943_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27944_cast = softmax(axis = var_21105, x = aw_2337_cast)[name = tensor<string, []>("op_27944_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27945_cast = softmax(axis = var_21105, x = aw_2339_cast)[name = tensor<string, []>("op_27945_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27946_cast = softmax(axis = var_21105, x = aw_2341_cast)[name = tensor<string, []>("op_27946_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27947_cast = softmax(axis = var_21105, x = aw_2343_cast)[name = tensor<string, []>("op_27947_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27948_cast = softmax(axis = var_21105, x = aw_2345_cast)[name = tensor<string, []>("op_27948_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27949_cast = softmax(axis = var_21105, x = aw_2347_cast)[name = tensor<string, []>("op_27949_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27950_cast = softmax(axis = var_21105, x = aw_2349_cast)[name = tensor<string, []>("op_27950_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27951_cast = softmax(axis = var_21105, x = aw_2351_cast)[name = tensor<string, []>("op_27951_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27952_cast = softmax(axis = var_21105, x = aw_2353_cast)[name = tensor<string, []>("op_27952_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27953_cast = softmax(axis = var_21105, x = aw_2355_cast)[name = tensor<string, []>("op_27953_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27954_cast = softmax(axis = var_21105, x = aw_2357_cast)[name = tensor<string, []>("op_27954_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_27955_cast = softmax(axis = var_21105, x = aw_2359_cast)[name = tensor<string, []>("op_27955_cast")];
            tensor<string, []> var_27957_equation_0 = const()[name = tensor<string, []>("op_27957_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27957_cast = einsum(equation = var_27957_equation_0, values = (var_27777_cast, var_27936_cast))[name = tensor<string, []>("op_27957_cast")];
            tensor<string, []> var_27959_equation_0 = const()[name = tensor<string, []>("op_27959_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27959_cast = einsum(equation = var_27959_equation_0, values = (var_27781_cast, var_27937_cast))[name = tensor<string, []>("op_27959_cast")];
            tensor<string, []> var_27961_equation_0 = const()[name = tensor<string, []>("op_27961_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27961_cast = einsum(equation = var_27961_equation_0, values = (var_27785_cast, var_27938_cast))[name = tensor<string, []>("op_27961_cast")];
            tensor<string, []> var_27963_equation_0 = const()[name = tensor<string, []>("op_27963_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27963_cast = einsum(equation = var_27963_equation_0, values = (var_27789_cast, var_27939_cast))[name = tensor<string, []>("op_27963_cast")];
            tensor<string, []> var_27965_equation_0 = const()[name = tensor<string, []>("op_27965_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27965_cast = einsum(equation = var_27965_equation_0, values = (var_27793_cast, var_27940_cast))[name = tensor<string, []>("op_27965_cast")];
            tensor<string, []> var_27967_equation_0 = const()[name = tensor<string, []>("op_27967_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27967_cast = einsum(equation = var_27967_equation_0, values = (var_27797_cast, var_27941_cast))[name = tensor<string, []>("op_27967_cast")];
            tensor<string, []> var_27969_equation_0 = const()[name = tensor<string, []>("op_27969_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27969_cast = einsum(equation = var_27969_equation_0, values = (var_27801_cast, var_27942_cast))[name = tensor<string, []>("op_27969_cast")];
            tensor<string, []> var_27971_equation_0 = const()[name = tensor<string, []>("op_27971_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27971_cast = einsum(equation = var_27971_equation_0, values = (var_27805_cast, var_27943_cast))[name = tensor<string, []>("op_27971_cast")];
            tensor<string, []> var_27973_equation_0 = const()[name = tensor<string, []>("op_27973_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27973_cast = einsum(equation = var_27973_equation_0, values = (var_27809_cast, var_27944_cast))[name = tensor<string, []>("op_27973_cast")];
            tensor<string, []> var_27975_equation_0 = const()[name = tensor<string, []>("op_27975_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27975_cast = einsum(equation = var_27975_equation_0, values = (var_27813_cast, var_27945_cast))[name = tensor<string, []>("op_27975_cast")];
            tensor<string, []> var_27977_equation_0 = const()[name = tensor<string, []>("op_27977_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27977_cast = einsum(equation = var_27977_equation_0, values = (var_27817_cast, var_27946_cast))[name = tensor<string, []>("op_27977_cast")];
            tensor<string, []> var_27979_equation_0 = const()[name = tensor<string, []>("op_27979_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27979_cast = einsum(equation = var_27979_equation_0, values = (var_27821_cast, var_27947_cast))[name = tensor<string, []>("op_27979_cast")];
            tensor<string, []> var_27981_equation_0 = const()[name = tensor<string, []>("op_27981_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27981_cast = einsum(equation = var_27981_equation_0, values = (var_27825_cast, var_27948_cast))[name = tensor<string, []>("op_27981_cast")];
            tensor<string, []> var_27983_equation_0 = const()[name = tensor<string, []>("op_27983_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27983_cast = einsum(equation = var_27983_equation_0, values = (var_27829_cast, var_27949_cast))[name = tensor<string, []>("op_27983_cast")];
            tensor<string, []> var_27985_equation_0 = const()[name = tensor<string, []>("op_27985_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27985_cast = einsum(equation = var_27985_equation_0, values = (var_27833_cast, var_27950_cast))[name = tensor<string, []>("op_27985_cast")];
            tensor<string, []> var_27987_equation_0 = const()[name = tensor<string, []>("op_27987_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27987_cast = einsum(equation = var_27987_equation_0, values = (var_27837_cast, var_27951_cast))[name = tensor<string, []>("op_27987_cast")];
            tensor<string, []> var_27989_equation_0 = const()[name = tensor<string, []>("op_27989_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27989_cast = einsum(equation = var_27989_equation_0, values = (var_27841_cast, var_27952_cast))[name = tensor<string, []>("op_27989_cast")];
            tensor<string, []> var_27991_equation_0 = const()[name = tensor<string, []>("op_27991_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27991_cast = einsum(equation = var_27991_equation_0, values = (var_27845_cast, var_27953_cast))[name = tensor<string, []>("op_27991_cast")];
            tensor<string, []> var_27993_equation_0 = const()[name = tensor<string, []>("op_27993_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27993_cast = einsum(equation = var_27993_equation_0, values = (var_27849_cast, var_27954_cast))[name = tensor<string, []>("op_27993_cast")];
            tensor<string, []> var_27995_equation_0 = const()[name = tensor<string, []>("op_27995_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_27995_cast = einsum(equation = var_27995_equation_0, values = (var_27853_cast, var_27955_cast))[name = tensor<string, []>("op_27995_cast")];
            tensor<bool, []> input_385_interleave_0 = const()[name = tensor<string, []>("input_385_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_385_cast = concat(axis = var_21105, interleave = input_385_interleave_0, values = (var_27957_cast, var_27959_cast, var_27961_cast, var_27963_cast, var_27965_cast, var_27967_cast, var_27969_cast, var_27971_cast, var_27973_cast, var_27975_cast, var_27977_cast, var_27979_cast, var_27981_cast, var_27983_cast, var_27985_cast, var_27987_cast, var_27989_cast, var_27991_cast, var_27993_cast, var_27995_cast))[name = tensor<string, []>("input_385_cast")];
            tensor<int32, [2]> var_28001 = const()[name = tensor<string, []>("op_28001"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_28003 = const()[name = tensor<string, []>("op_28003"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_28005_pad_type_0 = const()[name = tensor<string, []>("op_28005_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_28005_pad_0 = const()[name = tensor<string, []>("op_28005_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2222535616)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2225812480)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_28005_cast = conv(bias = mid_block_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16, dilations = var_28003, groups = var_21105, pad = var_28005_pad_0, pad_type = var_28005_pad_type_0, strides = var_28001, weight = mid_block_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16, x = input_385_cast)[name = tensor<string, []>("op_28005_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_189_cast = add(x = var_28005_cast, y = inputs_187_cast)[name = tensor<string, []>("inputs_189_cast")];
            tensor<int32, [1]> var_28009 = const()[name = tensor<string, []>("op_28009"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_189_cast = reduce_mean(axes = var_28009, keep_dims = var_21100, x = inputs_189_cast)[name = tensor<string, []>("channels_mean_189_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_189_cast = sub(x = inputs_189_cast, y = channels_mean_189_cast)[name = tensor<string, []>("zero_mean_189_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_189_cast = mul(x = zero_mean_189_cast, y = zero_mean_189_cast)[name = tensor<string, []>("zero_mean_sq_189_cast")];
            tensor<int32, [1]> var_28013 = const()[name = tensor<string, []>("op_28013"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_28014_cast = reduce_mean(axes = var_28013, keep_dims = var_21100, x = zero_mean_sq_189_cast)[name = tensor<string, []>("op_28014_cast")];
            tensor<fp16, []> var_28015_to_fp16 = const()[name = tensor<string, []>("op_28015_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_28016_cast = add(x = var_28014_cast, y = var_28015_to_fp16)[name = tensor<string, []>("op_28016_cast")];
            tensor<fp16, []> denom_189_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_189_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_189_cast = rsqrt(epsilon = denom_189_epsilon_0_to_fp16, x = var_28016_cast)[name = tensor<string, []>("denom_189_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_189_cast = mul(x = zero_mean_189_cast, y = denom_189_cast)[name = tensor<string, []>("out_189_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_28020_to_fp16 = const()[name = tensor<string, []>("op_28020_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2225815104)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_28021_cast = add(x = out_189_cast, y = var_28020_to_fp16)[name = tensor<string, []>("op_28021_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_28023_to_fp16 = const()[name = tensor<string, []>("op_28023_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2225817728)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_253_cast = mul(x = var_28021_cast, y = var_28023_to_fp16)[name = tensor<string, []>("hidden_states_253_cast")];
            tensor<int32, [2]> var_28030 = const()[name = tensor<string, []>("op_28030"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_28032 = const()[name = tensor<string, []>("op_28032"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_127_pad_type_0 = const()[name = tensor<string, []>("q_127_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_127_pad_0 = const()[name = tensor<string, []>("q_127_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2225820352)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_127_cast = conv(dilations = var_28032, groups = var_21105, pad = q_127_pad_0, pad_type = q_127_pad_type_0, strides = var_28030, weight = mid_block_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16, x = hidden_states_253_cast)[name = tensor<string, []>("q_127_cast")];
            tensor<int32, [2]> var_28036 = const()[name = tensor<string, []>("op_28036"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_28038 = const()[name = tensor<string, []>("op_28038"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_253_pad_type_0 = const()[name = tensor<string, []>("k_253_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_253_pad_0 = const()[name = tensor<string, []>("k_253_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2229097216)))];
            tensor<fp16, [2, 1280, 1, 77]> k_253_cast = conv(dilations = var_28038, groups = var_21105, pad = k_253_pad_0, pad_type = k_253_pad_type_0, strides = var_28036, weight = mid_block_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_253_cast")];
            tensor<int32, [2]> var_28042 = const()[name = tensor<string, []>("op_28042"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_28044 = const()[name = tensor<string, []>("op_28044"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_127_pad_type_0 = const()[name = tensor<string, []>("v_127_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_127_pad_0 = const()[name = tensor<string, []>("v_127_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2234340160)))];
            tensor<fp16, [2, 1280, 1, 77]> v_127_cast = conv(dilations = var_28044, groups = var_21105, pad = v_127_pad_0, pad_type = v_127_pad_type_0, strides = var_28042, weight = mid_block_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_127_cast")];
            tensor<int32, [4]> var_28048_begin_0 = const()[name = tensor<string, []>("op_28048_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_28048_end_0 = const()[name = tensor<string, []>("op_28048_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_28048_end_mask_0 = const()[name = tensor<string, []>("op_28048_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28048_cast = slice_by_index(begin = var_28048_begin_0, end = var_28048_end_0, end_mask = var_28048_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28048_cast")];
            tensor<int32, [4]> var_28052_begin_0 = const()[name = tensor<string, []>("op_28052_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_28052_end_0 = const()[name = tensor<string, []>("op_28052_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_28052_end_mask_0 = const()[name = tensor<string, []>("op_28052_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28052_cast = slice_by_index(begin = var_28052_begin_0, end = var_28052_end_0, end_mask = var_28052_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28052_cast")];
            tensor<int32, [4]> var_28056_begin_0 = const()[name = tensor<string, []>("op_28056_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_28056_end_0 = const()[name = tensor<string, []>("op_28056_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_28056_end_mask_0 = const()[name = tensor<string, []>("op_28056_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28056_cast = slice_by_index(begin = var_28056_begin_0, end = var_28056_end_0, end_mask = var_28056_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28056_cast")];
            tensor<int32, [4]> var_28060_begin_0 = const()[name = tensor<string, []>("op_28060_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_28060_end_0 = const()[name = tensor<string, []>("op_28060_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_28060_end_mask_0 = const()[name = tensor<string, []>("op_28060_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28060_cast = slice_by_index(begin = var_28060_begin_0, end = var_28060_end_0, end_mask = var_28060_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28060_cast")];
            tensor<int32, [4]> var_28064_begin_0 = const()[name = tensor<string, []>("op_28064_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_28064_end_0 = const()[name = tensor<string, []>("op_28064_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_28064_end_mask_0 = const()[name = tensor<string, []>("op_28064_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28064_cast = slice_by_index(begin = var_28064_begin_0, end = var_28064_end_0, end_mask = var_28064_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28064_cast")];
            tensor<int32, [4]> var_28068_begin_0 = const()[name = tensor<string, []>("op_28068_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_28068_end_0 = const()[name = tensor<string, []>("op_28068_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_28068_end_mask_0 = const()[name = tensor<string, []>("op_28068_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28068_cast = slice_by_index(begin = var_28068_begin_0, end = var_28068_end_0, end_mask = var_28068_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28068_cast")];
            tensor<int32, [4]> var_28072_begin_0 = const()[name = tensor<string, []>("op_28072_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_28072_end_0 = const()[name = tensor<string, []>("op_28072_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_28072_end_mask_0 = const()[name = tensor<string, []>("op_28072_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28072_cast = slice_by_index(begin = var_28072_begin_0, end = var_28072_end_0, end_mask = var_28072_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28072_cast")];
            tensor<int32, [4]> var_28076_begin_0 = const()[name = tensor<string, []>("op_28076_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_28076_end_0 = const()[name = tensor<string, []>("op_28076_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_28076_end_mask_0 = const()[name = tensor<string, []>("op_28076_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28076_cast = slice_by_index(begin = var_28076_begin_0, end = var_28076_end_0, end_mask = var_28076_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28076_cast")];
            tensor<int32, [4]> var_28080_begin_0 = const()[name = tensor<string, []>("op_28080_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_28080_end_0 = const()[name = tensor<string, []>("op_28080_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_28080_end_mask_0 = const()[name = tensor<string, []>("op_28080_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28080_cast = slice_by_index(begin = var_28080_begin_0, end = var_28080_end_0, end_mask = var_28080_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28080_cast")];
            tensor<int32, [4]> var_28084_begin_0 = const()[name = tensor<string, []>("op_28084_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_28084_end_0 = const()[name = tensor<string, []>("op_28084_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_28084_end_mask_0 = const()[name = tensor<string, []>("op_28084_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28084_cast = slice_by_index(begin = var_28084_begin_0, end = var_28084_end_0, end_mask = var_28084_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28084_cast")];
            tensor<int32, [4]> var_28088_begin_0 = const()[name = tensor<string, []>("op_28088_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_28088_end_0 = const()[name = tensor<string, []>("op_28088_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_28088_end_mask_0 = const()[name = tensor<string, []>("op_28088_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28088_cast = slice_by_index(begin = var_28088_begin_0, end = var_28088_end_0, end_mask = var_28088_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28088_cast")];
            tensor<int32, [4]> var_28092_begin_0 = const()[name = tensor<string, []>("op_28092_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_28092_end_0 = const()[name = tensor<string, []>("op_28092_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_28092_end_mask_0 = const()[name = tensor<string, []>("op_28092_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28092_cast = slice_by_index(begin = var_28092_begin_0, end = var_28092_end_0, end_mask = var_28092_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28092_cast")];
            tensor<int32, [4]> var_28096_begin_0 = const()[name = tensor<string, []>("op_28096_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_28096_end_0 = const()[name = tensor<string, []>("op_28096_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_28096_end_mask_0 = const()[name = tensor<string, []>("op_28096_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28096_cast = slice_by_index(begin = var_28096_begin_0, end = var_28096_end_0, end_mask = var_28096_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28096_cast")];
            tensor<int32, [4]> var_28100_begin_0 = const()[name = tensor<string, []>("op_28100_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_28100_end_0 = const()[name = tensor<string, []>("op_28100_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_28100_end_mask_0 = const()[name = tensor<string, []>("op_28100_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28100_cast = slice_by_index(begin = var_28100_begin_0, end = var_28100_end_0, end_mask = var_28100_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28100_cast")];
            tensor<int32, [4]> var_28104_begin_0 = const()[name = tensor<string, []>("op_28104_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_28104_end_0 = const()[name = tensor<string, []>("op_28104_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_28104_end_mask_0 = const()[name = tensor<string, []>("op_28104_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28104_cast = slice_by_index(begin = var_28104_begin_0, end = var_28104_end_0, end_mask = var_28104_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28104_cast")];
            tensor<int32, [4]> var_28108_begin_0 = const()[name = tensor<string, []>("op_28108_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_28108_end_0 = const()[name = tensor<string, []>("op_28108_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_28108_end_mask_0 = const()[name = tensor<string, []>("op_28108_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28108_cast = slice_by_index(begin = var_28108_begin_0, end = var_28108_end_0, end_mask = var_28108_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28108_cast")];
            tensor<int32, [4]> var_28112_begin_0 = const()[name = tensor<string, []>("op_28112_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_28112_end_0 = const()[name = tensor<string, []>("op_28112_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_28112_end_mask_0 = const()[name = tensor<string, []>("op_28112_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28112_cast = slice_by_index(begin = var_28112_begin_0, end = var_28112_end_0, end_mask = var_28112_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28112_cast")];
            tensor<int32, [4]> var_28116_begin_0 = const()[name = tensor<string, []>("op_28116_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_28116_end_0 = const()[name = tensor<string, []>("op_28116_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_28116_end_mask_0 = const()[name = tensor<string, []>("op_28116_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28116_cast = slice_by_index(begin = var_28116_begin_0, end = var_28116_end_0, end_mask = var_28116_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28116_cast")];
            tensor<int32, [4]> var_28120_begin_0 = const()[name = tensor<string, []>("op_28120_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_28120_end_0 = const()[name = tensor<string, []>("op_28120_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_28120_end_mask_0 = const()[name = tensor<string, []>("op_28120_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28120_cast = slice_by_index(begin = var_28120_begin_0, end = var_28120_end_0, end_mask = var_28120_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28120_cast")];
            tensor<int32, [4]> var_28124_begin_0 = const()[name = tensor<string, []>("op_28124_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_28124_end_0 = const()[name = tensor<string, []>("op_28124_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_28124_end_mask_0 = const()[name = tensor<string, []>("op_28124_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28124_cast = slice_by_index(begin = var_28124_begin_0, end = var_28124_end_0, end_mask = var_28124_end_mask_0, x = q_127_cast)[name = tensor<string, []>("op_28124_cast")];
            tensor<int32, [4]> k_255_perm_0 = const()[name = tensor<string, []>("k_255_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_28131_begin_0 = const()[name = tensor<string, []>("op_28131_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_28131_end_0 = const()[name = tensor<string, []>("op_28131_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_28131_end_mask_0 = const()[name = tensor<string, []>("op_28131_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_76 = transpose(perm = k_255_perm_0, x = k_253_cast)[name = tensor<string, []>("transpose_76")];
            tensor<fp16, [2, 77, 1, 64]> var_28131_cast = slice_by_index(begin = var_28131_begin_0, end = var_28131_end_0, end_mask = var_28131_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28131_cast")];
            tensor<int32, [4]> var_28135_begin_0 = const()[name = tensor<string, []>("op_28135_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_28135_end_0 = const()[name = tensor<string, []>("op_28135_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_28135_end_mask_0 = const()[name = tensor<string, []>("op_28135_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_28135_cast = slice_by_index(begin = var_28135_begin_0, end = var_28135_end_0, end_mask = var_28135_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28135_cast")];
            tensor<int32, [4]> var_28139_begin_0 = const()[name = tensor<string, []>("op_28139_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_28139_end_0 = const()[name = tensor<string, []>("op_28139_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_28139_end_mask_0 = const()[name = tensor<string, []>("op_28139_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_28139_cast = slice_by_index(begin = var_28139_begin_0, end = var_28139_end_0, end_mask = var_28139_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28139_cast")];
            tensor<int32, [4]> var_28143_begin_0 = const()[name = tensor<string, []>("op_28143_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_28143_end_0 = const()[name = tensor<string, []>("op_28143_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_28143_end_mask_0 = const()[name = tensor<string, []>("op_28143_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_28143_cast = slice_by_index(begin = var_28143_begin_0, end = var_28143_end_0, end_mask = var_28143_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28143_cast")];
            tensor<int32, [4]> var_28147_begin_0 = const()[name = tensor<string, []>("op_28147_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_28147_end_0 = const()[name = tensor<string, []>("op_28147_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_28147_end_mask_0 = const()[name = tensor<string, []>("op_28147_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_28147_cast = slice_by_index(begin = var_28147_begin_0, end = var_28147_end_0, end_mask = var_28147_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28147_cast")];
            tensor<int32, [4]> var_28151_begin_0 = const()[name = tensor<string, []>("op_28151_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_28151_end_0 = const()[name = tensor<string, []>("op_28151_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_28151_end_mask_0 = const()[name = tensor<string, []>("op_28151_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_28151_cast = slice_by_index(begin = var_28151_begin_0, end = var_28151_end_0, end_mask = var_28151_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28151_cast")];
            tensor<int32, [4]> var_28155_begin_0 = const()[name = tensor<string, []>("op_28155_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_28155_end_0 = const()[name = tensor<string, []>("op_28155_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_28155_end_mask_0 = const()[name = tensor<string, []>("op_28155_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_28155_cast = slice_by_index(begin = var_28155_begin_0, end = var_28155_end_0, end_mask = var_28155_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28155_cast")];
            tensor<int32, [4]> var_28159_begin_0 = const()[name = tensor<string, []>("op_28159_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_28159_end_0 = const()[name = tensor<string, []>("op_28159_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_28159_end_mask_0 = const()[name = tensor<string, []>("op_28159_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_28159_cast = slice_by_index(begin = var_28159_begin_0, end = var_28159_end_0, end_mask = var_28159_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28159_cast")];
            tensor<int32, [4]> var_28163_begin_0 = const()[name = tensor<string, []>("op_28163_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_28163_end_0 = const()[name = tensor<string, []>("op_28163_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_28163_end_mask_0 = const()[name = tensor<string, []>("op_28163_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_28163_cast = slice_by_index(begin = var_28163_begin_0, end = var_28163_end_0, end_mask = var_28163_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28163_cast")];
            tensor<int32, [4]> var_28167_begin_0 = const()[name = tensor<string, []>("op_28167_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_28167_end_0 = const()[name = tensor<string, []>("op_28167_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_28167_end_mask_0 = const()[name = tensor<string, []>("op_28167_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_28167_cast = slice_by_index(begin = var_28167_begin_0, end = var_28167_end_0, end_mask = var_28167_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28167_cast")];
            tensor<int32, [4]> var_28171_begin_0 = const()[name = tensor<string, []>("op_28171_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_28171_end_0 = const()[name = tensor<string, []>("op_28171_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_28171_end_mask_0 = const()[name = tensor<string, []>("op_28171_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_28171_cast = slice_by_index(begin = var_28171_begin_0, end = var_28171_end_0, end_mask = var_28171_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28171_cast")];
            tensor<int32, [4]> var_28175_begin_0 = const()[name = tensor<string, []>("op_28175_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_28175_end_0 = const()[name = tensor<string, []>("op_28175_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_28175_end_mask_0 = const()[name = tensor<string, []>("op_28175_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_28175_cast = slice_by_index(begin = var_28175_begin_0, end = var_28175_end_0, end_mask = var_28175_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28175_cast")];
            tensor<int32, [4]> var_28179_begin_0 = const()[name = tensor<string, []>("op_28179_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_28179_end_0 = const()[name = tensor<string, []>("op_28179_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_28179_end_mask_0 = const()[name = tensor<string, []>("op_28179_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_28179_cast = slice_by_index(begin = var_28179_begin_0, end = var_28179_end_0, end_mask = var_28179_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28179_cast")];
            tensor<int32, [4]> var_28183_begin_0 = const()[name = tensor<string, []>("op_28183_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_28183_end_0 = const()[name = tensor<string, []>("op_28183_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_28183_end_mask_0 = const()[name = tensor<string, []>("op_28183_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_28183_cast = slice_by_index(begin = var_28183_begin_0, end = var_28183_end_0, end_mask = var_28183_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28183_cast")];
            tensor<int32, [4]> var_28187_begin_0 = const()[name = tensor<string, []>("op_28187_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_28187_end_0 = const()[name = tensor<string, []>("op_28187_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_28187_end_mask_0 = const()[name = tensor<string, []>("op_28187_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_28187_cast = slice_by_index(begin = var_28187_begin_0, end = var_28187_end_0, end_mask = var_28187_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28187_cast")];
            tensor<int32, [4]> var_28191_begin_0 = const()[name = tensor<string, []>("op_28191_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_28191_end_0 = const()[name = tensor<string, []>("op_28191_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_28191_end_mask_0 = const()[name = tensor<string, []>("op_28191_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_28191_cast = slice_by_index(begin = var_28191_begin_0, end = var_28191_end_0, end_mask = var_28191_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28191_cast")];
            tensor<int32, [4]> var_28195_begin_0 = const()[name = tensor<string, []>("op_28195_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_28195_end_0 = const()[name = tensor<string, []>("op_28195_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_28195_end_mask_0 = const()[name = tensor<string, []>("op_28195_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_28195_cast = slice_by_index(begin = var_28195_begin_0, end = var_28195_end_0, end_mask = var_28195_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28195_cast")];
            tensor<int32, [4]> var_28199_begin_0 = const()[name = tensor<string, []>("op_28199_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_28199_end_0 = const()[name = tensor<string, []>("op_28199_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_28199_end_mask_0 = const()[name = tensor<string, []>("op_28199_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_28199_cast = slice_by_index(begin = var_28199_begin_0, end = var_28199_end_0, end_mask = var_28199_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28199_cast")];
            tensor<int32, [4]> var_28203_begin_0 = const()[name = tensor<string, []>("op_28203_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_28203_end_0 = const()[name = tensor<string, []>("op_28203_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_28203_end_mask_0 = const()[name = tensor<string, []>("op_28203_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_28203_cast = slice_by_index(begin = var_28203_begin_0, end = var_28203_end_0, end_mask = var_28203_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28203_cast")];
            tensor<int32, [4]> var_28207_begin_0 = const()[name = tensor<string, []>("op_28207_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_28207_end_0 = const()[name = tensor<string, []>("op_28207_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_28207_end_mask_0 = const()[name = tensor<string, []>("op_28207_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_28207_cast = slice_by_index(begin = var_28207_begin_0, end = var_28207_end_0, end_mask = var_28207_end_mask_0, x = transpose_76)[name = tensor<string, []>("op_28207_cast")];
            tensor<int32, [4]> var_28209_begin_0 = const()[name = tensor<string, []>("op_28209_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_28209_end_0 = const()[name = tensor<string, []>("op_28209_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_28209_end_mask_0 = const()[name = tensor<string, []>("op_28209_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28209_cast = slice_by_index(begin = var_28209_begin_0, end = var_28209_end_0, end_mask = var_28209_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28209_cast")];
            tensor<int32, [4]> var_28213_begin_0 = const()[name = tensor<string, []>("op_28213_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_28213_end_0 = const()[name = tensor<string, []>("op_28213_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_28213_end_mask_0 = const()[name = tensor<string, []>("op_28213_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28213_cast = slice_by_index(begin = var_28213_begin_0, end = var_28213_end_0, end_mask = var_28213_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28213_cast")];
            tensor<int32, [4]> var_28217_begin_0 = const()[name = tensor<string, []>("op_28217_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_28217_end_0 = const()[name = tensor<string, []>("op_28217_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_28217_end_mask_0 = const()[name = tensor<string, []>("op_28217_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28217_cast = slice_by_index(begin = var_28217_begin_0, end = var_28217_end_0, end_mask = var_28217_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28217_cast")];
            tensor<int32, [4]> var_28221_begin_0 = const()[name = tensor<string, []>("op_28221_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_28221_end_0 = const()[name = tensor<string, []>("op_28221_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_28221_end_mask_0 = const()[name = tensor<string, []>("op_28221_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28221_cast = slice_by_index(begin = var_28221_begin_0, end = var_28221_end_0, end_mask = var_28221_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28221_cast")];
            tensor<int32, [4]> var_28225_begin_0 = const()[name = tensor<string, []>("op_28225_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_28225_end_0 = const()[name = tensor<string, []>("op_28225_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_28225_end_mask_0 = const()[name = tensor<string, []>("op_28225_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28225_cast = slice_by_index(begin = var_28225_begin_0, end = var_28225_end_0, end_mask = var_28225_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28225_cast")];
            tensor<int32, [4]> var_28229_begin_0 = const()[name = tensor<string, []>("op_28229_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_28229_end_0 = const()[name = tensor<string, []>("op_28229_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_28229_end_mask_0 = const()[name = tensor<string, []>("op_28229_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28229_cast = slice_by_index(begin = var_28229_begin_0, end = var_28229_end_0, end_mask = var_28229_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28229_cast")];
            tensor<int32, [4]> var_28233_begin_0 = const()[name = tensor<string, []>("op_28233_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_28233_end_0 = const()[name = tensor<string, []>("op_28233_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_28233_end_mask_0 = const()[name = tensor<string, []>("op_28233_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28233_cast = slice_by_index(begin = var_28233_begin_0, end = var_28233_end_0, end_mask = var_28233_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28233_cast")];
            tensor<int32, [4]> var_28237_begin_0 = const()[name = tensor<string, []>("op_28237_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_28237_end_0 = const()[name = tensor<string, []>("op_28237_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_28237_end_mask_0 = const()[name = tensor<string, []>("op_28237_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28237_cast = slice_by_index(begin = var_28237_begin_0, end = var_28237_end_0, end_mask = var_28237_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28237_cast")];
            tensor<int32, [4]> var_28241_begin_0 = const()[name = tensor<string, []>("op_28241_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_28241_end_0 = const()[name = tensor<string, []>("op_28241_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_28241_end_mask_0 = const()[name = tensor<string, []>("op_28241_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28241_cast = slice_by_index(begin = var_28241_begin_0, end = var_28241_end_0, end_mask = var_28241_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28241_cast")];
            tensor<int32, [4]> var_28245_begin_0 = const()[name = tensor<string, []>("op_28245_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_28245_end_0 = const()[name = tensor<string, []>("op_28245_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_28245_end_mask_0 = const()[name = tensor<string, []>("op_28245_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28245_cast = slice_by_index(begin = var_28245_begin_0, end = var_28245_end_0, end_mask = var_28245_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28245_cast")];
            tensor<int32, [4]> var_28249_begin_0 = const()[name = tensor<string, []>("op_28249_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_28249_end_0 = const()[name = tensor<string, []>("op_28249_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_28249_end_mask_0 = const()[name = tensor<string, []>("op_28249_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28249_cast = slice_by_index(begin = var_28249_begin_0, end = var_28249_end_0, end_mask = var_28249_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28249_cast")];
            tensor<int32, [4]> var_28253_begin_0 = const()[name = tensor<string, []>("op_28253_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_28253_end_0 = const()[name = tensor<string, []>("op_28253_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_28253_end_mask_0 = const()[name = tensor<string, []>("op_28253_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28253_cast = slice_by_index(begin = var_28253_begin_0, end = var_28253_end_0, end_mask = var_28253_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28253_cast")];
            tensor<int32, [4]> var_28257_begin_0 = const()[name = tensor<string, []>("op_28257_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_28257_end_0 = const()[name = tensor<string, []>("op_28257_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_28257_end_mask_0 = const()[name = tensor<string, []>("op_28257_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28257_cast = slice_by_index(begin = var_28257_begin_0, end = var_28257_end_0, end_mask = var_28257_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28257_cast")];
            tensor<int32, [4]> var_28261_begin_0 = const()[name = tensor<string, []>("op_28261_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_28261_end_0 = const()[name = tensor<string, []>("op_28261_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_28261_end_mask_0 = const()[name = tensor<string, []>("op_28261_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28261_cast = slice_by_index(begin = var_28261_begin_0, end = var_28261_end_0, end_mask = var_28261_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28261_cast")];
            tensor<int32, [4]> var_28265_begin_0 = const()[name = tensor<string, []>("op_28265_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_28265_end_0 = const()[name = tensor<string, []>("op_28265_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_28265_end_mask_0 = const()[name = tensor<string, []>("op_28265_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28265_cast = slice_by_index(begin = var_28265_begin_0, end = var_28265_end_0, end_mask = var_28265_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28265_cast")];
            tensor<int32, [4]> var_28269_begin_0 = const()[name = tensor<string, []>("op_28269_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_28269_end_0 = const()[name = tensor<string, []>("op_28269_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_28269_end_mask_0 = const()[name = tensor<string, []>("op_28269_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28269_cast = slice_by_index(begin = var_28269_begin_0, end = var_28269_end_0, end_mask = var_28269_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28269_cast")];
            tensor<int32, [4]> var_28273_begin_0 = const()[name = tensor<string, []>("op_28273_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_28273_end_0 = const()[name = tensor<string, []>("op_28273_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_28273_end_mask_0 = const()[name = tensor<string, []>("op_28273_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28273_cast = slice_by_index(begin = var_28273_begin_0, end = var_28273_end_0, end_mask = var_28273_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28273_cast")];
            tensor<int32, [4]> var_28277_begin_0 = const()[name = tensor<string, []>("op_28277_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_28277_end_0 = const()[name = tensor<string, []>("op_28277_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_28277_end_mask_0 = const()[name = tensor<string, []>("op_28277_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28277_cast = slice_by_index(begin = var_28277_begin_0, end = var_28277_end_0, end_mask = var_28277_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28277_cast")];
            tensor<int32, [4]> var_28281_begin_0 = const()[name = tensor<string, []>("op_28281_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_28281_end_0 = const()[name = tensor<string, []>("op_28281_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_28281_end_mask_0 = const()[name = tensor<string, []>("op_28281_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28281_cast = slice_by_index(begin = var_28281_begin_0, end = var_28281_end_0, end_mask = var_28281_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28281_cast")];
            tensor<int32, [4]> var_28285_begin_0 = const()[name = tensor<string, []>("op_28285_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_28285_end_0 = const()[name = tensor<string, []>("op_28285_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_28285_end_mask_0 = const()[name = tensor<string, []>("op_28285_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_28285_cast = slice_by_index(begin = var_28285_begin_0, end = var_28285_end_0, end_mask = var_28285_end_mask_0, x = v_127_cast)[name = tensor<string, []>("op_28285_cast")];
            tensor<string, []> var_28289_equation_0 = const()[name = tensor<string, []>("op_28289_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28289_cast = einsum(equation = var_28289_equation_0, values = (var_28131_cast, var_28048_cast))[name = tensor<string, []>("op_28289_cast")];
            tensor<fp16, []> var_28290_to_fp16 = const()[name = tensor<string, []>("op_28290_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2361_cast = mul(x = var_28289_cast, y = var_28290_to_fp16)[name = tensor<string, []>("aw_2361_cast")];
            tensor<string, []> var_28293_equation_0 = const()[name = tensor<string, []>("op_28293_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28293_cast = einsum(equation = var_28293_equation_0, values = (var_28135_cast, var_28052_cast))[name = tensor<string, []>("op_28293_cast")];
            tensor<fp16, []> var_28294_to_fp16 = const()[name = tensor<string, []>("op_28294_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2363_cast = mul(x = var_28293_cast, y = var_28294_to_fp16)[name = tensor<string, []>("aw_2363_cast")];
            tensor<string, []> var_28297_equation_0 = const()[name = tensor<string, []>("op_28297_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28297_cast = einsum(equation = var_28297_equation_0, values = (var_28139_cast, var_28056_cast))[name = tensor<string, []>("op_28297_cast")];
            tensor<fp16, []> var_28298_to_fp16 = const()[name = tensor<string, []>("op_28298_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2365_cast = mul(x = var_28297_cast, y = var_28298_to_fp16)[name = tensor<string, []>("aw_2365_cast")];
            tensor<string, []> var_28301_equation_0 = const()[name = tensor<string, []>("op_28301_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28301_cast = einsum(equation = var_28301_equation_0, values = (var_28143_cast, var_28060_cast))[name = tensor<string, []>("op_28301_cast")];
            tensor<fp16, []> var_28302_to_fp16 = const()[name = tensor<string, []>("op_28302_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2367_cast = mul(x = var_28301_cast, y = var_28302_to_fp16)[name = tensor<string, []>("aw_2367_cast")];
            tensor<string, []> var_28305_equation_0 = const()[name = tensor<string, []>("op_28305_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28305_cast = einsum(equation = var_28305_equation_0, values = (var_28147_cast, var_28064_cast))[name = tensor<string, []>("op_28305_cast")];
            tensor<fp16, []> var_28306_to_fp16 = const()[name = tensor<string, []>("op_28306_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2369_cast = mul(x = var_28305_cast, y = var_28306_to_fp16)[name = tensor<string, []>("aw_2369_cast")];
            tensor<string, []> var_28309_equation_0 = const()[name = tensor<string, []>("op_28309_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28309_cast = einsum(equation = var_28309_equation_0, values = (var_28151_cast, var_28068_cast))[name = tensor<string, []>("op_28309_cast")];
            tensor<fp16, []> var_28310_to_fp16 = const()[name = tensor<string, []>("op_28310_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2371_cast = mul(x = var_28309_cast, y = var_28310_to_fp16)[name = tensor<string, []>("aw_2371_cast")];
            tensor<string, []> var_28313_equation_0 = const()[name = tensor<string, []>("op_28313_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28313_cast = einsum(equation = var_28313_equation_0, values = (var_28155_cast, var_28072_cast))[name = tensor<string, []>("op_28313_cast")];
            tensor<fp16, []> var_28314_to_fp16 = const()[name = tensor<string, []>("op_28314_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2373_cast = mul(x = var_28313_cast, y = var_28314_to_fp16)[name = tensor<string, []>("aw_2373_cast")];
            tensor<string, []> var_28317_equation_0 = const()[name = tensor<string, []>("op_28317_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28317_cast = einsum(equation = var_28317_equation_0, values = (var_28159_cast, var_28076_cast))[name = tensor<string, []>("op_28317_cast")];
            tensor<fp16, []> var_28318_to_fp16 = const()[name = tensor<string, []>("op_28318_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2375_cast = mul(x = var_28317_cast, y = var_28318_to_fp16)[name = tensor<string, []>("aw_2375_cast")];
            tensor<string, []> var_28321_equation_0 = const()[name = tensor<string, []>("op_28321_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28321_cast = einsum(equation = var_28321_equation_0, values = (var_28163_cast, var_28080_cast))[name = tensor<string, []>("op_28321_cast")];
            tensor<fp16, []> var_28322_to_fp16 = const()[name = tensor<string, []>("op_28322_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2377_cast = mul(x = var_28321_cast, y = var_28322_to_fp16)[name = tensor<string, []>("aw_2377_cast")];
            tensor<string, []> var_28325_equation_0 = const()[name = tensor<string, []>("op_28325_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28325_cast = einsum(equation = var_28325_equation_0, values = (var_28167_cast, var_28084_cast))[name = tensor<string, []>("op_28325_cast")];
            tensor<fp16, []> var_28326_to_fp16 = const()[name = tensor<string, []>("op_28326_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2379_cast = mul(x = var_28325_cast, y = var_28326_to_fp16)[name = tensor<string, []>("aw_2379_cast")];
            tensor<string, []> var_28329_equation_0 = const()[name = tensor<string, []>("op_28329_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28329_cast = einsum(equation = var_28329_equation_0, values = (var_28171_cast, var_28088_cast))[name = tensor<string, []>("op_28329_cast")];
            tensor<fp16, []> var_28330_to_fp16 = const()[name = tensor<string, []>("op_28330_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2381_cast = mul(x = var_28329_cast, y = var_28330_to_fp16)[name = tensor<string, []>("aw_2381_cast")];
            tensor<string, []> var_28333_equation_0 = const()[name = tensor<string, []>("op_28333_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28333_cast = einsum(equation = var_28333_equation_0, values = (var_28175_cast, var_28092_cast))[name = tensor<string, []>("op_28333_cast")];
            tensor<fp16, []> var_28334_to_fp16 = const()[name = tensor<string, []>("op_28334_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2383_cast = mul(x = var_28333_cast, y = var_28334_to_fp16)[name = tensor<string, []>("aw_2383_cast")];
            tensor<string, []> var_28337_equation_0 = const()[name = tensor<string, []>("op_28337_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28337_cast = einsum(equation = var_28337_equation_0, values = (var_28179_cast, var_28096_cast))[name = tensor<string, []>("op_28337_cast")];
            tensor<fp16, []> var_28338_to_fp16 = const()[name = tensor<string, []>("op_28338_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2385_cast = mul(x = var_28337_cast, y = var_28338_to_fp16)[name = tensor<string, []>("aw_2385_cast")];
            tensor<string, []> var_28341_equation_0 = const()[name = tensor<string, []>("op_28341_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28341_cast = einsum(equation = var_28341_equation_0, values = (var_28183_cast, var_28100_cast))[name = tensor<string, []>("op_28341_cast")];
            tensor<fp16, []> var_28342_to_fp16 = const()[name = tensor<string, []>("op_28342_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2387_cast = mul(x = var_28341_cast, y = var_28342_to_fp16)[name = tensor<string, []>("aw_2387_cast")];
            tensor<string, []> var_28345_equation_0 = const()[name = tensor<string, []>("op_28345_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28345_cast = einsum(equation = var_28345_equation_0, values = (var_28187_cast, var_28104_cast))[name = tensor<string, []>("op_28345_cast")];
            tensor<fp16, []> var_28346_to_fp16 = const()[name = tensor<string, []>("op_28346_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2389_cast = mul(x = var_28345_cast, y = var_28346_to_fp16)[name = tensor<string, []>("aw_2389_cast")];
            tensor<string, []> var_28349_equation_0 = const()[name = tensor<string, []>("op_28349_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28349_cast = einsum(equation = var_28349_equation_0, values = (var_28191_cast, var_28108_cast))[name = tensor<string, []>("op_28349_cast")];
            tensor<fp16, []> var_28350_to_fp16 = const()[name = tensor<string, []>("op_28350_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2391_cast = mul(x = var_28349_cast, y = var_28350_to_fp16)[name = tensor<string, []>("aw_2391_cast")];
            tensor<string, []> var_28353_equation_0 = const()[name = tensor<string, []>("op_28353_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28353_cast = einsum(equation = var_28353_equation_0, values = (var_28195_cast, var_28112_cast))[name = tensor<string, []>("op_28353_cast")];
            tensor<fp16, []> var_28354_to_fp16 = const()[name = tensor<string, []>("op_28354_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2393_cast = mul(x = var_28353_cast, y = var_28354_to_fp16)[name = tensor<string, []>("aw_2393_cast")];
            tensor<string, []> var_28357_equation_0 = const()[name = tensor<string, []>("op_28357_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28357_cast = einsum(equation = var_28357_equation_0, values = (var_28199_cast, var_28116_cast))[name = tensor<string, []>("op_28357_cast")];
            tensor<fp16, []> var_28358_to_fp16 = const()[name = tensor<string, []>("op_28358_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2395_cast = mul(x = var_28357_cast, y = var_28358_to_fp16)[name = tensor<string, []>("aw_2395_cast")];
            tensor<string, []> var_28361_equation_0 = const()[name = tensor<string, []>("op_28361_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28361_cast = einsum(equation = var_28361_equation_0, values = (var_28203_cast, var_28120_cast))[name = tensor<string, []>("op_28361_cast")];
            tensor<fp16, []> var_28362_to_fp16 = const()[name = tensor<string, []>("op_28362_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2397_cast = mul(x = var_28361_cast, y = var_28362_to_fp16)[name = tensor<string, []>("aw_2397_cast")];
            tensor<string, []> var_28365_equation_0 = const()[name = tensor<string, []>("op_28365_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_28365_cast = einsum(equation = var_28365_equation_0, values = (var_28207_cast, var_28124_cast))[name = tensor<string, []>("op_28365_cast")];
            tensor<fp16, []> var_28366_to_fp16 = const()[name = tensor<string, []>("op_28366_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2399_cast = mul(x = var_28365_cast, y = var_28366_to_fp16)[name = tensor<string, []>("aw_2399_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28368_cast = softmax(axis = var_21105, x = aw_2361_cast)[name = tensor<string, []>("op_28368_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28369_cast = softmax(axis = var_21105, x = aw_2363_cast)[name = tensor<string, []>("op_28369_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28370_cast = softmax(axis = var_21105, x = aw_2365_cast)[name = tensor<string, []>("op_28370_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28371_cast = softmax(axis = var_21105, x = aw_2367_cast)[name = tensor<string, []>("op_28371_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28372_cast = softmax(axis = var_21105, x = aw_2369_cast)[name = tensor<string, []>("op_28372_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28373_cast = softmax(axis = var_21105, x = aw_2371_cast)[name = tensor<string, []>("op_28373_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28374_cast = softmax(axis = var_21105, x = aw_2373_cast)[name = tensor<string, []>("op_28374_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28375_cast = softmax(axis = var_21105, x = aw_2375_cast)[name = tensor<string, []>("op_28375_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28376_cast = softmax(axis = var_21105, x = aw_2377_cast)[name = tensor<string, []>("op_28376_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28377_cast = softmax(axis = var_21105, x = aw_2379_cast)[name = tensor<string, []>("op_28377_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28378_cast = softmax(axis = var_21105, x = aw_2381_cast)[name = tensor<string, []>("op_28378_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28379_cast = softmax(axis = var_21105, x = aw_2383_cast)[name = tensor<string, []>("op_28379_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28380_cast = softmax(axis = var_21105, x = aw_2385_cast)[name = tensor<string, []>("op_28380_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28381_cast = softmax(axis = var_21105, x = aw_2387_cast)[name = tensor<string, []>("op_28381_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28382_cast = softmax(axis = var_21105, x = aw_2389_cast)[name = tensor<string, []>("op_28382_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28383_cast = softmax(axis = var_21105, x = aw_2391_cast)[name = tensor<string, []>("op_28383_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28384_cast = softmax(axis = var_21105, x = aw_2393_cast)[name = tensor<string, []>("op_28384_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28385_cast = softmax(axis = var_21105, x = aw_2395_cast)[name = tensor<string, []>("op_28385_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28386_cast = softmax(axis = var_21105, x = aw_2397_cast)[name = tensor<string, []>("op_28386_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_28387_cast = softmax(axis = var_21105, x = aw_2399_cast)[name = tensor<string, []>("op_28387_cast")];
            tensor<string, []> var_28389_equation_0 = const()[name = tensor<string, []>("op_28389_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28389_cast = einsum(equation = var_28389_equation_0, values = (var_28209_cast, var_28368_cast))[name = tensor<string, []>("op_28389_cast")];
            tensor<string, []> var_28391_equation_0 = const()[name = tensor<string, []>("op_28391_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28391_cast = einsum(equation = var_28391_equation_0, values = (var_28213_cast, var_28369_cast))[name = tensor<string, []>("op_28391_cast")];
            tensor<string, []> var_28393_equation_0 = const()[name = tensor<string, []>("op_28393_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28393_cast = einsum(equation = var_28393_equation_0, values = (var_28217_cast, var_28370_cast))[name = tensor<string, []>("op_28393_cast")];
            tensor<string, []> var_28395_equation_0 = const()[name = tensor<string, []>("op_28395_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28395_cast = einsum(equation = var_28395_equation_0, values = (var_28221_cast, var_28371_cast))[name = tensor<string, []>("op_28395_cast")];
            tensor<string, []> var_28397_equation_0 = const()[name = tensor<string, []>("op_28397_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28397_cast = einsum(equation = var_28397_equation_0, values = (var_28225_cast, var_28372_cast))[name = tensor<string, []>("op_28397_cast")];
            tensor<string, []> var_28399_equation_0 = const()[name = tensor<string, []>("op_28399_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28399_cast = einsum(equation = var_28399_equation_0, values = (var_28229_cast, var_28373_cast))[name = tensor<string, []>("op_28399_cast")];
            tensor<string, []> var_28401_equation_0 = const()[name = tensor<string, []>("op_28401_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28401_cast = einsum(equation = var_28401_equation_0, values = (var_28233_cast, var_28374_cast))[name = tensor<string, []>("op_28401_cast")];
            tensor<string, []> var_28403_equation_0 = const()[name = tensor<string, []>("op_28403_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28403_cast = einsum(equation = var_28403_equation_0, values = (var_28237_cast, var_28375_cast))[name = tensor<string, []>("op_28403_cast")];
            tensor<string, []> var_28405_equation_0 = const()[name = tensor<string, []>("op_28405_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28405_cast = einsum(equation = var_28405_equation_0, values = (var_28241_cast, var_28376_cast))[name = tensor<string, []>("op_28405_cast")];
            tensor<string, []> var_28407_equation_0 = const()[name = tensor<string, []>("op_28407_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28407_cast = einsum(equation = var_28407_equation_0, values = (var_28245_cast, var_28377_cast))[name = tensor<string, []>("op_28407_cast")];
            tensor<string, []> var_28409_equation_0 = const()[name = tensor<string, []>("op_28409_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28409_cast = einsum(equation = var_28409_equation_0, values = (var_28249_cast, var_28378_cast))[name = tensor<string, []>("op_28409_cast")];
            tensor<string, []> var_28411_equation_0 = const()[name = tensor<string, []>("op_28411_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28411_cast = einsum(equation = var_28411_equation_0, values = (var_28253_cast, var_28379_cast))[name = tensor<string, []>("op_28411_cast")];
            tensor<string, []> var_28413_equation_0 = const()[name = tensor<string, []>("op_28413_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28413_cast = einsum(equation = var_28413_equation_0, values = (var_28257_cast, var_28380_cast))[name = tensor<string, []>("op_28413_cast")];
            tensor<string, []> var_28415_equation_0 = const()[name = tensor<string, []>("op_28415_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28415_cast = einsum(equation = var_28415_equation_0, values = (var_28261_cast, var_28381_cast))[name = tensor<string, []>("op_28415_cast")];
            tensor<string, []> var_28417_equation_0 = const()[name = tensor<string, []>("op_28417_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28417_cast = einsum(equation = var_28417_equation_0, values = (var_28265_cast, var_28382_cast))[name = tensor<string, []>("op_28417_cast")];
            tensor<string, []> var_28419_equation_0 = const()[name = tensor<string, []>("op_28419_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28419_cast = einsum(equation = var_28419_equation_0, values = (var_28269_cast, var_28383_cast))[name = tensor<string, []>("op_28419_cast")];
            tensor<string, []> var_28421_equation_0 = const()[name = tensor<string, []>("op_28421_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28421_cast = einsum(equation = var_28421_equation_0, values = (var_28273_cast, var_28384_cast))[name = tensor<string, []>("op_28421_cast")];
            tensor<string, []> var_28423_equation_0 = const()[name = tensor<string, []>("op_28423_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28423_cast = einsum(equation = var_28423_equation_0, values = (var_28277_cast, var_28385_cast))[name = tensor<string, []>("op_28423_cast")];
            tensor<string, []> var_28425_equation_0 = const()[name = tensor<string, []>("op_28425_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28425_cast = einsum(equation = var_28425_equation_0, values = (var_28281_cast, var_28386_cast))[name = tensor<string, []>("op_28425_cast")];
            tensor<string, []> var_28427_equation_0 = const()[name = tensor<string, []>("op_28427_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28427_cast = einsum(equation = var_28427_equation_0, values = (var_28285_cast, var_28387_cast))[name = tensor<string, []>("op_28427_cast")];
            tensor<bool, []> input_387_interleave_0 = const()[name = tensor<string, []>("input_387_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_387_cast = concat(axis = var_21105, interleave = input_387_interleave_0, values = (var_28389_cast, var_28391_cast, var_28393_cast, var_28395_cast, var_28397_cast, var_28399_cast, var_28401_cast, var_28403_cast, var_28405_cast, var_28407_cast, var_28409_cast, var_28411_cast, var_28413_cast, var_28415_cast, var_28417_cast, var_28419_cast, var_28421_cast, var_28423_cast, var_28425_cast, var_28427_cast))[name = tensor<string, []>("input_387_cast")];
            tensor<int32, [2]> var_28433 = const()[name = tensor<string, []>("op_28433"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_28435 = const()[name = tensor<string, []>("op_28435"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_28437_pad_type_0 = const()[name = tensor<string, []>("op_28437_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_28437_pad_0 = const()[name = tensor<string, []>("op_28437_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2239583104)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2242859968)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_28437_cast = conv(bias = mid_block_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16, dilations = var_28435, groups = var_21105, pad = var_28437_pad_0, pad_type = var_28437_pad_type_0, strides = var_28433, weight = mid_block_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16, x = input_387_cast)[name = tensor<string, []>("op_28437_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_191_cast = add(x = var_28437_cast, y = inputs_189_cast)[name = tensor<string, []>("inputs_191_cast")];
            tensor<int32, [1]> var_28441 = const()[name = tensor<string, []>("op_28441"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_191_cast = reduce_mean(axes = var_28441, keep_dims = var_21100, x = inputs_191_cast)[name = tensor<string, []>("channels_mean_191_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_191_cast = sub(x = inputs_191_cast, y = channels_mean_191_cast)[name = tensor<string, []>("zero_mean_191_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_191_cast = mul(x = zero_mean_191_cast, y = zero_mean_191_cast)[name = tensor<string, []>("zero_mean_sq_191_cast")];
            tensor<int32, [1]> var_28445 = const()[name = tensor<string, []>("op_28445"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_28446_cast = reduce_mean(axes = var_28445, keep_dims = var_21100, x = zero_mean_sq_191_cast)[name = tensor<string, []>("op_28446_cast")];
            tensor<fp16, []> var_28447_to_fp16 = const()[name = tensor<string, []>("op_28447_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_28448_cast = add(x = var_28446_cast, y = var_28447_to_fp16)[name = tensor<string, []>("op_28448_cast")];
            tensor<fp16, []> denom_191_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_191_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_191_cast = rsqrt(epsilon = denom_191_epsilon_0_to_fp16, x = var_28448_cast)[name = tensor<string, []>("denom_191_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_191_cast = mul(x = zero_mean_191_cast, y = denom_191_cast)[name = tensor<string, []>("out_191_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_28452_to_fp16 = const()[name = tensor<string, []>("op_28452_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2242862592)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_28453_cast = add(x = out_191_cast, y = var_28452_to_fp16)[name = tensor<string, []>("op_28453_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_28455_to_fp16 = const()[name = tensor<string, []>("op_28455_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2242865216)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_389_cast = mul(x = var_28453_cast, y = var_28455_to_fp16)[name = tensor<string, []>("input_389_cast")];
            tensor<int32, [2]> var_28463 = const()[name = tensor<string, []>("op_28463"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_28465 = const()[name = tensor<string, []>("op_28465"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_28467_pad_type_0 = const()[name = tensor<string, []>("op_28467_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_28467_pad_0 = const()[name = tensor<string, []>("op_28467_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2242867840)))];
            tensor<fp16, [10240]> mid_block_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2269082304)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_28467_cast = conv(bias = mid_block_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16, dilations = var_28465, groups = var_21105, pad = var_28467_pad_0, pad_type = var_28467_pad_type_0, strides = var_28463, weight = mid_block_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16, x = input_389_cast)[name = tensor<string, []>("op_28467_cast")];
            tensor<int32, [2]> var_28468_split_sizes_0 = const()[name = tensor<string, []>("op_28468_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_28468_axis_0 = const()[name = tensor<string, []>("op_28468_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_28468_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_28468_cast_1 = split(axis = var_28468_axis_0, split_sizes = var_28468_split_sizes_0, x = var_28467_cast)[name = tensor<string, []>("op_28468_cast")];
            tensor<string, []> var_28470_mode_0 = const()[name = tensor<string, []>("op_28470_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_28470_cast = gelu(mode = var_28470_mode_0, x = var_28468_cast_1)[name = tensor<string, []>("op_28470_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_391_cast = mul(x = var_28468_cast_0, y = var_28470_cast)[name = tensor<string, []>("input_391_cast")];
            tensor<int32, [2]> var_28474 = const()[name = tensor<string, []>("op_28474"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_28476 = const()[name = tensor<string, []>("op_28476"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_28478_pad_type_0 = const()[name = tensor<string, []>("op_28478_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_28478_pad_0 = const()[name = tensor<string, []>("op_28478_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> mid_block_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2269102848)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2282210112)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_28478_cast = conv(bias = mid_block_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16, dilations = var_28476, groups = var_21105, pad = var_28478_pad_0, pad_type = var_28478_pad_type_0, strides = var_28474, weight = mid_block_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16, x = input_391_cast)[name = tensor<string, []>("op_28478_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_193_cast = add(x = var_28478_cast, y = inputs_191_cast)[name = tensor<string, []>("inputs_193_cast")];
            tensor<int32, [1]> var_28488 = const()[name = tensor<string, []>("op_28488"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_193_cast = reduce_mean(axes = var_28488, keep_dims = var_21100, x = inputs_193_cast)[name = tensor<string, []>("channels_mean_193_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_193_cast = sub(x = inputs_193_cast, y = channels_mean_193_cast)[name = tensor<string, []>("zero_mean_193_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_193_cast = mul(x = zero_mean_193_cast, y = zero_mean_193_cast)[name = tensor<string, []>("zero_mean_sq_193_cast")];
            tensor<int32, [1]> var_28492 = const()[name = tensor<string, []>("op_28492"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_28493_cast = reduce_mean(axes = var_28492, keep_dims = var_21100, x = zero_mean_sq_193_cast)[name = tensor<string, []>("op_28493_cast")];
            tensor<fp16, []> var_28494_to_fp16 = const()[name = tensor<string, []>("op_28494_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_28495_cast = add(x = var_28493_cast, y = var_28494_to_fp16)[name = tensor<string, []>("op_28495_cast")];
            tensor<fp16, []> denom_193_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_193_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_193_cast = rsqrt(epsilon = denom_193_epsilon_0_to_fp16, x = var_28495_cast)[name = tensor<string, []>("denom_193_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_193_cast = mul(x = zero_mean_193_cast, y = denom_193_cast)[name = tensor<string, []>("out_193_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_28499_to_fp16 = const()[name = tensor<string, []>("op_28499_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2282212736)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_28500_cast = add(x = out_193_cast, y = var_28499_to_fp16)[name = tensor<string, []>("op_28500_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_28502_to_fp16 = const()[name = tensor<string, []>("op_28502_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2282215360)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_257_cast = mul(x = var_28500_cast, y = var_28502_to_fp16)[name = tensor<string, []>("hidden_states_257_cast")];
            tensor<int32, [2]> var_28509 = const()[name = tensor<string, []>("op_28509"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_28511 = const()[name = tensor<string, []>("op_28511"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_129_pad_type_0 = const()[name = tensor<string, []>("q_129_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_129_pad_0 = const()[name = tensor<string, []>("q_129_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2282217984)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_129_cast = conv(dilations = var_28511, groups = var_21105, pad = q_129_pad_0, pad_type = q_129_pad_type_0, strides = var_28509, weight = mid_block_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16, x = hidden_states_257_cast)[name = tensor<string, []>("q_129_cast")];
            tensor<int32, [2]> var_28515 = const()[name = tensor<string, []>("op_28515"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_28517 = const()[name = tensor<string, []>("op_28517"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_257_pad_type_0 = const()[name = tensor<string, []>("k_257_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_257_pad_0 = const()[name = tensor<string, []>("k_257_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2285494848)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_257_cast = conv(dilations = var_28517, groups = var_21105, pad = k_257_pad_0, pad_type = k_257_pad_type_0, strides = var_28515, weight = mid_block_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16, x = hidden_states_257_cast)[name = tensor<string, []>("k_257_cast")];
            tensor<int32, [2]> var_28521 = const()[name = tensor<string, []>("op_28521"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_28523 = const()[name = tensor<string, []>("op_28523"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_129_pad_type_0 = const()[name = tensor<string, []>("v_129_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_129_pad_0 = const()[name = tensor<string, []>("v_129_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2288771712)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_129_cast = conv(dilations = var_28523, groups = var_21105, pad = v_129_pad_0, pad_type = v_129_pad_type_0, strides = var_28521, weight = mid_block_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16, x = hidden_states_257_cast)[name = tensor<string, []>("v_129_cast")];
            tensor<int32, [4]> var_28527_begin_0 = const()[name = tensor<string, []>("op_28527_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_28527_end_0 = const()[name = tensor<string, []>("op_28527_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_28527_end_mask_0 = const()[name = tensor<string, []>("op_28527_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28527_cast = slice_by_index(begin = var_28527_begin_0, end = var_28527_end_0, end_mask = var_28527_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28527_cast")];
            tensor<int32, [4]> var_28531_begin_0 = const()[name = tensor<string, []>("op_28531_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_28531_end_0 = const()[name = tensor<string, []>("op_28531_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_28531_end_mask_0 = const()[name = tensor<string, []>("op_28531_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28531_cast = slice_by_index(begin = var_28531_begin_0, end = var_28531_end_0, end_mask = var_28531_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28531_cast")];
            tensor<int32, [4]> var_28535_begin_0 = const()[name = tensor<string, []>("op_28535_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_28535_end_0 = const()[name = tensor<string, []>("op_28535_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_28535_end_mask_0 = const()[name = tensor<string, []>("op_28535_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28535_cast = slice_by_index(begin = var_28535_begin_0, end = var_28535_end_0, end_mask = var_28535_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28535_cast")];
            tensor<int32, [4]> var_28539_begin_0 = const()[name = tensor<string, []>("op_28539_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_28539_end_0 = const()[name = tensor<string, []>("op_28539_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_28539_end_mask_0 = const()[name = tensor<string, []>("op_28539_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28539_cast = slice_by_index(begin = var_28539_begin_0, end = var_28539_end_0, end_mask = var_28539_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28539_cast")];
            tensor<int32, [4]> var_28543_begin_0 = const()[name = tensor<string, []>("op_28543_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_28543_end_0 = const()[name = tensor<string, []>("op_28543_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_28543_end_mask_0 = const()[name = tensor<string, []>("op_28543_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28543_cast = slice_by_index(begin = var_28543_begin_0, end = var_28543_end_0, end_mask = var_28543_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28543_cast")];
            tensor<int32, [4]> var_28547_begin_0 = const()[name = tensor<string, []>("op_28547_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_28547_end_0 = const()[name = tensor<string, []>("op_28547_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_28547_end_mask_0 = const()[name = tensor<string, []>("op_28547_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28547_cast = slice_by_index(begin = var_28547_begin_0, end = var_28547_end_0, end_mask = var_28547_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28547_cast")];
            tensor<int32, [4]> var_28551_begin_0 = const()[name = tensor<string, []>("op_28551_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_28551_end_0 = const()[name = tensor<string, []>("op_28551_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_28551_end_mask_0 = const()[name = tensor<string, []>("op_28551_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28551_cast = slice_by_index(begin = var_28551_begin_0, end = var_28551_end_0, end_mask = var_28551_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28551_cast")];
            tensor<int32, [4]> var_28555_begin_0 = const()[name = tensor<string, []>("op_28555_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_28555_end_0 = const()[name = tensor<string, []>("op_28555_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_28555_end_mask_0 = const()[name = tensor<string, []>("op_28555_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28555_cast = slice_by_index(begin = var_28555_begin_0, end = var_28555_end_0, end_mask = var_28555_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28555_cast")];
            tensor<int32, [4]> var_28559_begin_0 = const()[name = tensor<string, []>("op_28559_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_28559_end_0 = const()[name = tensor<string, []>("op_28559_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_28559_end_mask_0 = const()[name = tensor<string, []>("op_28559_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28559_cast = slice_by_index(begin = var_28559_begin_0, end = var_28559_end_0, end_mask = var_28559_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28559_cast")];
            tensor<int32, [4]> var_28563_begin_0 = const()[name = tensor<string, []>("op_28563_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_28563_end_0 = const()[name = tensor<string, []>("op_28563_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_28563_end_mask_0 = const()[name = tensor<string, []>("op_28563_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28563_cast = slice_by_index(begin = var_28563_begin_0, end = var_28563_end_0, end_mask = var_28563_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28563_cast")];
            tensor<int32, [4]> var_28567_begin_0 = const()[name = tensor<string, []>("op_28567_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_28567_end_0 = const()[name = tensor<string, []>("op_28567_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_28567_end_mask_0 = const()[name = tensor<string, []>("op_28567_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28567_cast = slice_by_index(begin = var_28567_begin_0, end = var_28567_end_0, end_mask = var_28567_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28567_cast")];
            tensor<int32, [4]> var_28571_begin_0 = const()[name = tensor<string, []>("op_28571_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_28571_end_0 = const()[name = tensor<string, []>("op_28571_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_28571_end_mask_0 = const()[name = tensor<string, []>("op_28571_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28571_cast = slice_by_index(begin = var_28571_begin_0, end = var_28571_end_0, end_mask = var_28571_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28571_cast")];
            tensor<int32, [4]> var_28575_begin_0 = const()[name = tensor<string, []>("op_28575_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_28575_end_0 = const()[name = tensor<string, []>("op_28575_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_28575_end_mask_0 = const()[name = tensor<string, []>("op_28575_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28575_cast = slice_by_index(begin = var_28575_begin_0, end = var_28575_end_0, end_mask = var_28575_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28575_cast")];
            tensor<int32, [4]> var_28579_begin_0 = const()[name = tensor<string, []>("op_28579_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_28579_end_0 = const()[name = tensor<string, []>("op_28579_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_28579_end_mask_0 = const()[name = tensor<string, []>("op_28579_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28579_cast = slice_by_index(begin = var_28579_begin_0, end = var_28579_end_0, end_mask = var_28579_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28579_cast")];
            tensor<int32, [4]> var_28583_begin_0 = const()[name = tensor<string, []>("op_28583_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_28583_end_0 = const()[name = tensor<string, []>("op_28583_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_28583_end_mask_0 = const()[name = tensor<string, []>("op_28583_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28583_cast = slice_by_index(begin = var_28583_begin_0, end = var_28583_end_0, end_mask = var_28583_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28583_cast")];
            tensor<int32, [4]> var_28587_begin_0 = const()[name = tensor<string, []>("op_28587_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_28587_end_0 = const()[name = tensor<string, []>("op_28587_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_28587_end_mask_0 = const()[name = tensor<string, []>("op_28587_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28587_cast = slice_by_index(begin = var_28587_begin_0, end = var_28587_end_0, end_mask = var_28587_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28587_cast")];
            tensor<int32, [4]> var_28591_begin_0 = const()[name = tensor<string, []>("op_28591_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_28591_end_0 = const()[name = tensor<string, []>("op_28591_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_28591_end_mask_0 = const()[name = tensor<string, []>("op_28591_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28591_cast = slice_by_index(begin = var_28591_begin_0, end = var_28591_end_0, end_mask = var_28591_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28591_cast")];
            tensor<int32, [4]> var_28595_begin_0 = const()[name = tensor<string, []>("op_28595_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_28595_end_0 = const()[name = tensor<string, []>("op_28595_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_28595_end_mask_0 = const()[name = tensor<string, []>("op_28595_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28595_cast = slice_by_index(begin = var_28595_begin_0, end = var_28595_end_0, end_mask = var_28595_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28595_cast")];
            tensor<int32, [4]> var_28599_begin_0 = const()[name = tensor<string, []>("op_28599_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_28599_end_0 = const()[name = tensor<string, []>("op_28599_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_28599_end_mask_0 = const()[name = tensor<string, []>("op_28599_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28599_cast = slice_by_index(begin = var_28599_begin_0, end = var_28599_end_0, end_mask = var_28599_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28599_cast")];
            tensor<int32, [4]> var_28603_begin_0 = const()[name = tensor<string, []>("op_28603_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_28603_end_0 = const()[name = tensor<string, []>("op_28603_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_28603_end_mask_0 = const()[name = tensor<string, []>("op_28603_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28603_cast = slice_by_index(begin = var_28603_begin_0, end = var_28603_end_0, end_mask = var_28603_end_mask_0, x = q_129_cast)[name = tensor<string, []>("op_28603_cast")];
            tensor<int32, [4]> k_259_perm_0 = const()[name = tensor<string, []>("k_259_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_28610_begin_0 = const()[name = tensor<string, []>("op_28610_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_28610_end_0 = const()[name = tensor<string, []>("op_28610_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_28610_end_mask_0 = const()[name = tensor<string, []>("op_28610_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_75 = transpose(perm = k_259_perm_0, x = k_257_cast)[name = tensor<string, []>("transpose_75")];
            tensor<fp16, [2, 1024, 1, 64]> var_28610_cast = slice_by_index(begin = var_28610_begin_0, end = var_28610_end_0, end_mask = var_28610_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28610_cast")];
            tensor<int32, [4]> var_28614_begin_0 = const()[name = tensor<string, []>("op_28614_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_28614_end_0 = const()[name = tensor<string, []>("op_28614_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_28614_end_mask_0 = const()[name = tensor<string, []>("op_28614_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_28614_cast = slice_by_index(begin = var_28614_begin_0, end = var_28614_end_0, end_mask = var_28614_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28614_cast")];
            tensor<int32, [4]> var_28618_begin_0 = const()[name = tensor<string, []>("op_28618_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_28618_end_0 = const()[name = tensor<string, []>("op_28618_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_28618_end_mask_0 = const()[name = tensor<string, []>("op_28618_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_28618_cast = slice_by_index(begin = var_28618_begin_0, end = var_28618_end_0, end_mask = var_28618_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28618_cast")];
            tensor<int32, [4]> var_28622_begin_0 = const()[name = tensor<string, []>("op_28622_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_28622_end_0 = const()[name = tensor<string, []>("op_28622_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_28622_end_mask_0 = const()[name = tensor<string, []>("op_28622_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_28622_cast = slice_by_index(begin = var_28622_begin_0, end = var_28622_end_0, end_mask = var_28622_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28622_cast")];
            tensor<int32, [4]> var_28626_begin_0 = const()[name = tensor<string, []>("op_28626_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_28626_end_0 = const()[name = tensor<string, []>("op_28626_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_28626_end_mask_0 = const()[name = tensor<string, []>("op_28626_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_28626_cast = slice_by_index(begin = var_28626_begin_0, end = var_28626_end_0, end_mask = var_28626_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28626_cast")];
            tensor<int32, [4]> var_28630_begin_0 = const()[name = tensor<string, []>("op_28630_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_28630_end_0 = const()[name = tensor<string, []>("op_28630_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_28630_end_mask_0 = const()[name = tensor<string, []>("op_28630_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_28630_cast = slice_by_index(begin = var_28630_begin_0, end = var_28630_end_0, end_mask = var_28630_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28630_cast")];
            tensor<int32, [4]> var_28634_begin_0 = const()[name = tensor<string, []>("op_28634_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_28634_end_0 = const()[name = tensor<string, []>("op_28634_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_28634_end_mask_0 = const()[name = tensor<string, []>("op_28634_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_28634_cast = slice_by_index(begin = var_28634_begin_0, end = var_28634_end_0, end_mask = var_28634_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28634_cast")];
            tensor<int32, [4]> var_28638_begin_0 = const()[name = tensor<string, []>("op_28638_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_28638_end_0 = const()[name = tensor<string, []>("op_28638_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_28638_end_mask_0 = const()[name = tensor<string, []>("op_28638_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_28638_cast = slice_by_index(begin = var_28638_begin_0, end = var_28638_end_0, end_mask = var_28638_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28638_cast")];
            tensor<int32, [4]> var_28642_begin_0 = const()[name = tensor<string, []>("op_28642_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_28642_end_0 = const()[name = tensor<string, []>("op_28642_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_28642_end_mask_0 = const()[name = tensor<string, []>("op_28642_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_28642_cast = slice_by_index(begin = var_28642_begin_0, end = var_28642_end_0, end_mask = var_28642_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28642_cast")];
            tensor<int32, [4]> var_28646_begin_0 = const()[name = tensor<string, []>("op_28646_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_28646_end_0 = const()[name = tensor<string, []>("op_28646_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_28646_end_mask_0 = const()[name = tensor<string, []>("op_28646_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_28646_cast = slice_by_index(begin = var_28646_begin_0, end = var_28646_end_0, end_mask = var_28646_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28646_cast")];
            tensor<int32, [4]> var_28650_begin_0 = const()[name = tensor<string, []>("op_28650_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_28650_end_0 = const()[name = tensor<string, []>("op_28650_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_28650_end_mask_0 = const()[name = tensor<string, []>("op_28650_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_28650_cast = slice_by_index(begin = var_28650_begin_0, end = var_28650_end_0, end_mask = var_28650_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28650_cast")];
            tensor<int32, [4]> var_28654_begin_0 = const()[name = tensor<string, []>("op_28654_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_28654_end_0 = const()[name = tensor<string, []>("op_28654_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_28654_end_mask_0 = const()[name = tensor<string, []>("op_28654_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_28654_cast = slice_by_index(begin = var_28654_begin_0, end = var_28654_end_0, end_mask = var_28654_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28654_cast")];
            tensor<int32, [4]> var_28658_begin_0 = const()[name = tensor<string, []>("op_28658_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_28658_end_0 = const()[name = tensor<string, []>("op_28658_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_28658_end_mask_0 = const()[name = tensor<string, []>("op_28658_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_28658_cast = slice_by_index(begin = var_28658_begin_0, end = var_28658_end_0, end_mask = var_28658_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28658_cast")];
            tensor<int32, [4]> var_28662_begin_0 = const()[name = tensor<string, []>("op_28662_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_28662_end_0 = const()[name = tensor<string, []>("op_28662_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_28662_end_mask_0 = const()[name = tensor<string, []>("op_28662_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_28662_cast = slice_by_index(begin = var_28662_begin_0, end = var_28662_end_0, end_mask = var_28662_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28662_cast")];
            tensor<int32, [4]> var_28666_begin_0 = const()[name = tensor<string, []>("op_28666_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_28666_end_0 = const()[name = tensor<string, []>("op_28666_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_28666_end_mask_0 = const()[name = tensor<string, []>("op_28666_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_28666_cast = slice_by_index(begin = var_28666_begin_0, end = var_28666_end_0, end_mask = var_28666_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28666_cast")];
            tensor<int32, [4]> var_28670_begin_0 = const()[name = tensor<string, []>("op_28670_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_28670_end_0 = const()[name = tensor<string, []>("op_28670_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_28670_end_mask_0 = const()[name = tensor<string, []>("op_28670_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_28670_cast = slice_by_index(begin = var_28670_begin_0, end = var_28670_end_0, end_mask = var_28670_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28670_cast")];
            tensor<int32, [4]> var_28674_begin_0 = const()[name = tensor<string, []>("op_28674_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_28674_end_0 = const()[name = tensor<string, []>("op_28674_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_28674_end_mask_0 = const()[name = tensor<string, []>("op_28674_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_28674_cast = slice_by_index(begin = var_28674_begin_0, end = var_28674_end_0, end_mask = var_28674_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28674_cast")];
            tensor<int32, [4]> var_28678_begin_0 = const()[name = tensor<string, []>("op_28678_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_28678_end_0 = const()[name = tensor<string, []>("op_28678_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_28678_end_mask_0 = const()[name = tensor<string, []>("op_28678_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_28678_cast = slice_by_index(begin = var_28678_begin_0, end = var_28678_end_0, end_mask = var_28678_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28678_cast")];
            tensor<int32, [4]> var_28682_begin_0 = const()[name = tensor<string, []>("op_28682_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_28682_end_0 = const()[name = tensor<string, []>("op_28682_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_28682_end_mask_0 = const()[name = tensor<string, []>("op_28682_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_28682_cast = slice_by_index(begin = var_28682_begin_0, end = var_28682_end_0, end_mask = var_28682_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28682_cast")];
            tensor<int32, [4]> var_28686_begin_0 = const()[name = tensor<string, []>("op_28686_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_28686_end_0 = const()[name = tensor<string, []>("op_28686_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_28686_end_mask_0 = const()[name = tensor<string, []>("op_28686_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_28686_cast = slice_by_index(begin = var_28686_begin_0, end = var_28686_end_0, end_mask = var_28686_end_mask_0, x = transpose_75)[name = tensor<string, []>("op_28686_cast")];
            tensor<int32, [4]> var_28688_begin_0 = const()[name = tensor<string, []>("op_28688_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_28688_end_0 = const()[name = tensor<string, []>("op_28688_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_28688_end_mask_0 = const()[name = tensor<string, []>("op_28688_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28688_cast = slice_by_index(begin = var_28688_begin_0, end = var_28688_end_0, end_mask = var_28688_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28688_cast")];
            tensor<int32, [4]> var_28692_begin_0 = const()[name = tensor<string, []>("op_28692_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_28692_end_0 = const()[name = tensor<string, []>("op_28692_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_28692_end_mask_0 = const()[name = tensor<string, []>("op_28692_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28692_cast = slice_by_index(begin = var_28692_begin_0, end = var_28692_end_0, end_mask = var_28692_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28692_cast")];
            tensor<int32, [4]> var_28696_begin_0 = const()[name = tensor<string, []>("op_28696_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_28696_end_0 = const()[name = tensor<string, []>("op_28696_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_28696_end_mask_0 = const()[name = tensor<string, []>("op_28696_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28696_cast = slice_by_index(begin = var_28696_begin_0, end = var_28696_end_0, end_mask = var_28696_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28696_cast")];
            tensor<int32, [4]> var_28700_begin_0 = const()[name = tensor<string, []>("op_28700_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_28700_end_0 = const()[name = tensor<string, []>("op_28700_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_28700_end_mask_0 = const()[name = tensor<string, []>("op_28700_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28700_cast = slice_by_index(begin = var_28700_begin_0, end = var_28700_end_0, end_mask = var_28700_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28700_cast")];
            tensor<int32, [4]> var_28704_begin_0 = const()[name = tensor<string, []>("op_28704_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_28704_end_0 = const()[name = tensor<string, []>("op_28704_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_28704_end_mask_0 = const()[name = tensor<string, []>("op_28704_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28704_cast = slice_by_index(begin = var_28704_begin_0, end = var_28704_end_0, end_mask = var_28704_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28704_cast")];
            tensor<int32, [4]> var_28708_begin_0 = const()[name = tensor<string, []>("op_28708_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_28708_end_0 = const()[name = tensor<string, []>("op_28708_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_28708_end_mask_0 = const()[name = tensor<string, []>("op_28708_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28708_cast = slice_by_index(begin = var_28708_begin_0, end = var_28708_end_0, end_mask = var_28708_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28708_cast")];
            tensor<int32, [4]> var_28712_begin_0 = const()[name = tensor<string, []>("op_28712_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_28712_end_0 = const()[name = tensor<string, []>("op_28712_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_28712_end_mask_0 = const()[name = tensor<string, []>("op_28712_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28712_cast = slice_by_index(begin = var_28712_begin_0, end = var_28712_end_0, end_mask = var_28712_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28712_cast")];
            tensor<int32, [4]> var_28716_begin_0 = const()[name = tensor<string, []>("op_28716_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_28716_end_0 = const()[name = tensor<string, []>("op_28716_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_28716_end_mask_0 = const()[name = tensor<string, []>("op_28716_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28716_cast = slice_by_index(begin = var_28716_begin_0, end = var_28716_end_0, end_mask = var_28716_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28716_cast")];
            tensor<int32, [4]> var_28720_begin_0 = const()[name = tensor<string, []>("op_28720_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_28720_end_0 = const()[name = tensor<string, []>("op_28720_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_28720_end_mask_0 = const()[name = tensor<string, []>("op_28720_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28720_cast = slice_by_index(begin = var_28720_begin_0, end = var_28720_end_0, end_mask = var_28720_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28720_cast")];
            tensor<int32, [4]> var_28724_begin_0 = const()[name = tensor<string, []>("op_28724_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_28724_end_0 = const()[name = tensor<string, []>("op_28724_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_28724_end_mask_0 = const()[name = tensor<string, []>("op_28724_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28724_cast = slice_by_index(begin = var_28724_begin_0, end = var_28724_end_0, end_mask = var_28724_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28724_cast")];
            tensor<int32, [4]> var_28728_begin_0 = const()[name = tensor<string, []>("op_28728_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_28728_end_0 = const()[name = tensor<string, []>("op_28728_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_28728_end_mask_0 = const()[name = tensor<string, []>("op_28728_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28728_cast = slice_by_index(begin = var_28728_begin_0, end = var_28728_end_0, end_mask = var_28728_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28728_cast")];
            tensor<int32, [4]> var_28732_begin_0 = const()[name = tensor<string, []>("op_28732_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_28732_end_0 = const()[name = tensor<string, []>("op_28732_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_28732_end_mask_0 = const()[name = tensor<string, []>("op_28732_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28732_cast = slice_by_index(begin = var_28732_begin_0, end = var_28732_end_0, end_mask = var_28732_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28732_cast")];
            tensor<int32, [4]> var_28736_begin_0 = const()[name = tensor<string, []>("op_28736_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_28736_end_0 = const()[name = tensor<string, []>("op_28736_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_28736_end_mask_0 = const()[name = tensor<string, []>("op_28736_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28736_cast = slice_by_index(begin = var_28736_begin_0, end = var_28736_end_0, end_mask = var_28736_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28736_cast")];
            tensor<int32, [4]> var_28740_begin_0 = const()[name = tensor<string, []>("op_28740_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_28740_end_0 = const()[name = tensor<string, []>("op_28740_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_28740_end_mask_0 = const()[name = tensor<string, []>("op_28740_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28740_cast = slice_by_index(begin = var_28740_begin_0, end = var_28740_end_0, end_mask = var_28740_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28740_cast")];
            tensor<int32, [4]> var_28744_begin_0 = const()[name = tensor<string, []>("op_28744_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_28744_end_0 = const()[name = tensor<string, []>("op_28744_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_28744_end_mask_0 = const()[name = tensor<string, []>("op_28744_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28744_cast = slice_by_index(begin = var_28744_begin_0, end = var_28744_end_0, end_mask = var_28744_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28744_cast")];
            tensor<int32, [4]> var_28748_begin_0 = const()[name = tensor<string, []>("op_28748_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_28748_end_0 = const()[name = tensor<string, []>("op_28748_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_28748_end_mask_0 = const()[name = tensor<string, []>("op_28748_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28748_cast = slice_by_index(begin = var_28748_begin_0, end = var_28748_end_0, end_mask = var_28748_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28748_cast")];
            tensor<int32, [4]> var_28752_begin_0 = const()[name = tensor<string, []>("op_28752_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_28752_end_0 = const()[name = tensor<string, []>("op_28752_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_28752_end_mask_0 = const()[name = tensor<string, []>("op_28752_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28752_cast = slice_by_index(begin = var_28752_begin_0, end = var_28752_end_0, end_mask = var_28752_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28752_cast")];
            tensor<int32, [4]> var_28756_begin_0 = const()[name = tensor<string, []>("op_28756_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_28756_end_0 = const()[name = tensor<string, []>("op_28756_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_28756_end_mask_0 = const()[name = tensor<string, []>("op_28756_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28756_cast = slice_by_index(begin = var_28756_begin_0, end = var_28756_end_0, end_mask = var_28756_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28756_cast")];
            tensor<int32, [4]> var_28760_begin_0 = const()[name = tensor<string, []>("op_28760_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_28760_end_0 = const()[name = tensor<string, []>("op_28760_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_28760_end_mask_0 = const()[name = tensor<string, []>("op_28760_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28760_cast = slice_by_index(begin = var_28760_begin_0, end = var_28760_end_0, end_mask = var_28760_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28760_cast")];
            tensor<int32, [4]> var_28764_begin_0 = const()[name = tensor<string, []>("op_28764_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_28764_end_0 = const()[name = tensor<string, []>("op_28764_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_28764_end_mask_0 = const()[name = tensor<string, []>("op_28764_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28764_cast = slice_by_index(begin = var_28764_begin_0, end = var_28764_end_0, end_mask = var_28764_end_mask_0, x = v_129_cast)[name = tensor<string, []>("op_28764_cast")];
            tensor<string, []> var_28768_equation_0 = const()[name = tensor<string, []>("op_28768_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28768_cast = einsum(equation = var_28768_equation_0, values = (var_28610_cast, var_28527_cast))[name = tensor<string, []>("op_28768_cast")];
            tensor<fp16, []> var_28769_to_fp16 = const()[name = tensor<string, []>("op_28769_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2401_cast = mul(x = var_28768_cast, y = var_28769_to_fp16)[name = tensor<string, []>("aw_2401_cast")];
            tensor<string, []> var_28772_equation_0 = const()[name = tensor<string, []>("op_28772_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28772_cast = einsum(equation = var_28772_equation_0, values = (var_28614_cast, var_28531_cast))[name = tensor<string, []>("op_28772_cast")];
            tensor<fp16, []> var_28773_to_fp16 = const()[name = tensor<string, []>("op_28773_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2403_cast = mul(x = var_28772_cast, y = var_28773_to_fp16)[name = tensor<string, []>("aw_2403_cast")];
            tensor<string, []> var_28776_equation_0 = const()[name = tensor<string, []>("op_28776_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28776_cast = einsum(equation = var_28776_equation_0, values = (var_28618_cast, var_28535_cast))[name = tensor<string, []>("op_28776_cast")];
            tensor<fp16, []> var_28777_to_fp16 = const()[name = tensor<string, []>("op_28777_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2405_cast = mul(x = var_28776_cast, y = var_28777_to_fp16)[name = tensor<string, []>("aw_2405_cast")];
            tensor<string, []> var_28780_equation_0 = const()[name = tensor<string, []>("op_28780_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28780_cast = einsum(equation = var_28780_equation_0, values = (var_28622_cast, var_28539_cast))[name = tensor<string, []>("op_28780_cast")];
            tensor<fp16, []> var_28781_to_fp16 = const()[name = tensor<string, []>("op_28781_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2407_cast = mul(x = var_28780_cast, y = var_28781_to_fp16)[name = tensor<string, []>("aw_2407_cast")];
            tensor<string, []> var_28784_equation_0 = const()[name = tensor<string, []>("op_28784_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28784_cast = einsum(equation = var_28784_equation_0, values = (var_28626_cast, var_28543_cast))[name = tensor<string, []>("op_28784_cast")];
            tensor<fp16, []> var_28785_to_fp16 = const()[name = tensor<string, []>("op_28785_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2409_cast = mul(x = var_28784_cast, y = var_28785_to_fp16)[name = tensor<string, []>("aw_2409_cast")];
            tensor<string, []> var_28788_equation_0 = const()[name = tensor<string, []>("op_28788_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28788_cast = einsum(equation = var_28788_equation_0, values = (var_28630_cast, var_28547_cast))[name = tensor<string, []>("op_28788_cast")];
            tensor<fp16, []> var_28789_to_fp16 = const()[name = tensor<string, []>("op_28789_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2411_cast = mul(x = var_28788_cast, y = var_28789_to_fp16)[name = tensor<string, []>("aw_2411_cast")];
            tensor<string, []> var_28792_equation_0 = const()[name = tensor<string, []>("op_28792_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28792_cast = einsum(equation = var_28792_equation_0, values = (var_28634_cast, var_28551_cast))[name = tensor<string, []>("op_28792_cast")];
            tensor<fp16, []> var_28793_to_fp16 = const()[name = tensor<string, []>("op_28793_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2413_cast = mul(x = var_28792_cast, y = var_28793_to_fp16)[name = tensor<string, []>("aw_2413_cast")];
            tensor<string, []> var_28796_equation_0 = const()[name = tensor<string, []>("op_28796_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28796_cast = einsum(equation = var_28796_equation_0, values = (var_28638_cast, var_28555_cast))[name = tensor<string, []>("op_28796_cast")];
            tensor<fp16, []> var_28797_to_fp16 = const()[name = tensor<string, []>("op_28797_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2415_cast = mul(x = var_28796_cast, y = var_28797_to_fp16)[name = tensor<string, []>("aw_2415_cast")];
            tensor<string, []> var_28800_equation_0 = const()[name = tensor<string, []>("op_28800_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28800_cast = einsum(equation = var_28800_equation_0, values = (var_28642_cast, var_28559_cast))[name = tensor<string, []>("op_28800_cast")];
            tensor<fp16, []> var_28801_to_fp16 = const()[name = tensor<string, []>("op_28801_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2417_cast = mul(x = var_28800_cast, y = var_28801_to_fp16)[name = tensor<string, []>("aw_2417_cast")];
            tensor<string, []> var_28804_equation_0 = const()[name = tensor<string, []>("op_28804_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28804_cast = einsum(equation = var_28804_equation_0, values = (var_28646_cast, var_28563_cast))[name = tensor<string, []>("op_28804_cast")];
            tensor<fp16, []> var_28805_to_fp16 = const()[name = tensor<string, []>("op_28805_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2419_cast = mul(x = var_28804_cast, y = var_28805_to_fp16)[name = tensor<string, []>("aw_2419_cast")];
            tensor<string, []> var_28808_equation_0 = const()[name = tensor<string, []>("op_28808_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28808_cast = einsum(equation = var_28808_equation_0, values = (var_28650_cast, var_28567_cast))[name = tensor<string, []>("op_28808_cast")];
            tensor<fp16, []> var_28809_to_fp16 = const()[name = tensor<string, []>("op_28809_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2421_cast = mul(x = var_28808_cast, y = var_28809_to_fp16)[name = tensor<string, []>("aw_2421_cast")];
            tensor<string, []> var_28812_equation_0 = const()[name = tensor<string, []>("op_28812_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28812_cast = einsum(equation = var_28812_equation_0, values = (var_28654_cast, var_28571_cast))[name = tensor<string, []>("op_28812_cast")];
            tensor<fp16, []> var_28813_to_fp16 = const()[name = tensor<string, []>("op_28813_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2423_cast = mul(x = var_28812_cast, y = var_28813_to_fp16)[name = tensor<string, []>("aw_2423_cast")];
            tensor<string, []> var_28816_equation_0 = const()[name = tensor<string, []>("op_28816_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28816_cast = einsum(equation = var_28816_equation_0, values = (var_28658_cast, var_28575_cast))[name = tensor<string, []>("op_28816_cast")];
            tensor<fp16, []> var_28817_to_fp16 = const()[name = tensor<string, []>("op_28817_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2425_cast = mul(x = var_28816_cast, y = var_28817_to_fp16)[name = tensor<string, []>("aw_2425_cast")];
            tensor<string, []> var_28820_equation_0 = const()[name = tensor<string, []>("op_28820_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28820_cast = einsum(equation = var_28820_equation_0, values = (var_28662_cast, var_28579_cast))[name = tensor<string, []>("op_28820_cast")];
            tensor<fp16, []> var_28821_to_fp16 = const()[name = tensor<string, []>("op_28821_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2427_cast = mul(x = var_28820_cast, y = var_28821_to_fp16)[name = tensor<string, []>("aw_2427_cast")];
            tensor<string, []> var_28824_equation_0 = const()[name = tensor<string, []>("op_28824_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28824_cast = einsum(equation = var_28824_equation_0, values = (var_28666_cast, var_28583_cast))[name = tensor<string, []>("op_28824_cast")];
            tensor<fp16, []> var_28825_to_fp16 = const()[name = tensor<string, []>("op_28825_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2429_cast = mul(x = var_28824_cast, y = var_28825_to_fp16)[name = tensor<string, []>("aw_2429_cast")];
            tensor<string, []> var_28828_equation_0 = const()[name = tensor<string, []>("op_28828_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28828_cast = einsum(equation = var_28828_equation_0, values = (var_28670_cast, var_28587_cast))[name = tensor<string, []>("op_28828_cast")];
            tensor<fp16, []> var_28829_to_fp16 = const()[name = tensor<string, []>("op_28829_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2431_cast = mul(x = var_28828_cast, y = var_28829_to_fp16)[name = tensor<string, []>("aw_2431_cast")];
            tensor<string, []> var_28832_equation_0 = const()[name = tensor<string, []>("op_28832_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28832_cast = einsum(equation = var_28832_equation_0, values = (var_28674_cast, var_28591_cast))[name = tensor<string, []>("op_28832_cast")];
            tensor<fp16, []> var_28833_to_fp16 = const()[name = tensor<string, []>("op_28833_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2433_cast = mul(x = var_28832_cast, y = var_28833_to_fp16)[name = tensor<string, []>("aw_2433_cast")];
            tensor<string, []> var_28836_equation_0 = const()[name = tensor<string, []>("op_28836_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28836_cast = einsum(equation = var_28836_equation_0, values = (var_28678_cast, var_28595_cast))[name = tensor<string, []>("op_28836_cast")];
            tensor<fp16, []> var_28837_to_fp16 = const()[name = tensor<string, []>("op_28837_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2435_cast = mul(x = var_28836_cast, y = var_28837_to_fp16)[name = tensor<string, []>("aw_2435_cast")];
            tensor<string, []> var_28840_equation_0 = const()[name = tensor<string, []>("op_28840_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28840_cast = einsum(equation = var_28840_equation_0, values = (var_28682_cast, var_28599_cast))[name = tensor<string, []>("op_28840_cast")];
            tensor<fp16, []> var_28841_to_fp16 = const()[name = tensor<string, []>("op_28841_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2437_cast = mul(x = var_28840_cast, y = var_28841_to_fp16)[name = tensor<string, []>("aw_2437_cast")];
            tensor<string, []> var_28844_equation_0 = const()[name = tensor<string, []>("op_28844_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28844_cast = einsum(equation = var_28844_equation_0, values = (var_28686_cast, var_28603_cast))[name = tensor<string, []>("op_28844_cast")];
            tensor<fp16, []> var_28845_to_fp16 = const()[name = tensor<string, []>("op_28845_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2439_cast = mul(x = var_28844_cast, y = var_28845_to_fp16)[name = tensor<string, []>("aw_2439_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28847_cast = softmax(axis = var_21105, x = aw_2401_cast)[name = tensor<string, []>("op_28847_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28848_cast = softmax(axis = var_21105, x = aw_2403_cast)[name = tensor<string, []>("op_28848_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28849_cast = softmax(axis = var_21105, x = aw_2405_cast)[name = tensor<string, []>("op_28849_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28850_cast = softmax(axis = var_21105, x = aw_2407_cast)[name = tensor<string, []>("op_28850_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28851_cast = softmax(axis = var_21105, x = aw_2409_cast)[name = tensor<string, []>("op_28851_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28852_cast = softmax(axis = var_21105, x = aw_2411_cast)[name = tensor<string, []>("op_28852_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28853_cast = softmax(axis = var_21105, x = aw_2413_cast)[name = tensor<string, []>("op_28853_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28854_cast = softmax(axis = var_21105, x = aw_2415_cast)[name = tensor<string, []>("op_28854_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28855_cast = softmax(axis = var_21105, x = aw_2417_cast)[name = tensor<string, []>("op_28855_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28856_cast = softmax(axis = var_21105, x = aw_2419_cast)[name = tensor<string, []>("op_28856_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28857_cast = softmax(axis = var_21105, x = aw_2421_cast)[name = tensor<string, []>("op_28857_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28858_cast = softmax(axis = var_21105, x = aw_2423_cast)[name = tensor<string, []>("op_28858_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28859_cast = softmax(axis = var_21105, x = aw_2425_cast)[name = tensor<string, []>("op_28859_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28860_cast = softmax(axis = var_21105, x = aw_2427_cast)[name = tensor<string, []>("op_28860_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28861_cast = softmax(axis = var_21105, x = aw_2429_cast)[name = tensor<string, []>("op_28861_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28862_cast = softmax(axis = var_21105, x = aw_2431_cast)[name = tensor<string, []>("op_28862_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28863_cast = softmax(axis = var_21105, x = aw_2433_cast)[name = tensor<string, []>("op_28863_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28864_cast = softmax(axis = var_21105, x = aw_2435_cast)[name = tensor<string, []>("op_28864_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28865_cast = softmax(axis = var_21105, x = aw_2437_cast)[name = tensor<string, []>("op_28865_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_28866_cast = softmax(axis = var_21105, x = aw_2439_cast)[name = tensor<string, []>("op_28866_cast")];
            tensor<string, []> var_28868_equation_0 = const()[name = tensor<string, []>("op_28868_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28868_cast = einsum(equation = var_28868_equation_0, values = (var_28688_cast, var_28847_cast))[name = tensor<string, []>("op_28868_cast")];
            tensor<string, []> var_28870_equation_0 = const()[name = tensor<string, []>("op_28870_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28870_cast = einsum(equation = var_28870_equation_0, values = (var_28692_cast, var_28848_cast))[name = tensor<string, []>("op_28870_cast")];
            tensor<string, []> var_28872_equation_0 = const()[name = tensor<string, []>("op_28872_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28872_cast = einsum(equation = var_28872_equation_0, values = (var_28696_cast, var_28849_cast))[name = tensor<string, []>("op_28872_cast")];
            tensor<string, []> var_28874_equation_0 = const()[name = tensor<string, []>("op_28874_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28874_cast = einsum(equation = var_28874_equation_0, values = (var_28700_cast, var_28850_cast))[name = tensor<string, []>("op_28874_cast")];
            tensor<string, []> var_28876_equation_0 = const()[name = tensor<string, []>("op_28876_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28876_cast = einsum(equation = var_28876_equation_0, values = (var_28704_cast, var_28851_cast))[name = tensor<string, []>("op_28876_cast")];
            tensor<string, []> var_28878_equation_0 = const()[name = tensor<string, []>("op_28878_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28878_cast = einsum(equation = var_28878_equation_0, values = (var_28708_cast, var_28852_cast))[name = tensor<string, []>("op_28878_cast")];
            tensor<string, []> var_28880_equation_0 = const()[name = tensor<string, []>("op_28880_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28880_cast = einsum(equation = var_28880_equation_0, values = (var_28712_cast, var_28853_cast))[name = tensor<string, []>("op_28880_cast")];
            tensor<string, []> var_28882_equation_0 = const()[name = tensor<string, []>("op_28882_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28882_cast = einsum(equation = var_28882_equation_0, values = (var_28716_cast, var_28854_cast))[name = tensor<string, []>("op_28882_cast")];
            tensor<string, []> var_28884_equation_0 = const()[name = tensor<string, []>("op_28884_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28884_cast = einsum(equation = var_28884_equation_0, values = (var_28720_cast, var_28855_cast))[name = tensor<string, []>("op_28884_cast")];
            tensor<string, []> var_28886_equation_0 = const()[name = tensor<string, []>("op_28886_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28886_cast = einsum(equation = var_28886_equation_0, values = (var_28724_cast, var_28856_cast))[name = tensor<string, []>("op_28886_cast")];
            tensor<string, []> var_28888_equation_0 = const()[name = tensor<string, []>("op_28888_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28888_cast = einsum(equation = var_28888_equation_0, values = (var_28728_cast, var_28857_cast))[name = tensor<string, []>("op_28888_cast")];
            tensor<string, []> var_28890_equation_0 = const()[name = tensor<string, []>("op_28890_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28890_cast = einsum(equation = var_28890_equation_0, values = (var_28732_cast, var_28858_cast))[name = tensor<string, []>("op_28890_cast")];
            tensor<string, []> var_28892_equation_0 = const()[name = tensor<string, []>("op_28892_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28892_cast = einsum(equation = var_28892_equation_0, values = (var_28736_cast, var_28859_cast))[name = tensor<string, []>("op_28892_cast")];
            tensor<string, []> var_28894_equation_0 = const()[name = tensor<string, []>("op_28894_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28894_cast = einsum(equation = var_28894_equation_0, values = (var_28740_cast, var_28860_cast))[name = tensor<string, []>("op_28894_cast")];
            tensor<string, []> var_28896_equation_0 = const()[name = tensor<string, []>("op_28896_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28896_cast = einsum(equation = var_28896_equation_0, values = (var_28744_cast, var_28861_cast))[name = tensor<string, []>("op_28896_cast")];
            tensor<string, []> var_28898_equation_0 = const()[name = tensor<string, []>("op_28898_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28898_cast = einsum(equation = var_28898_equation_0, values = (var_28748_cast, var_28862_cast))[name = tensor<string, []>("op_28898_cast")];
            tensor<string, []> var_28900_equation_0 = const()[name = tensor<string, []>("op_28900_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28900_cast = einsum(equation = var_28900_equation_0, values = (var_28752_cast, var_28863_cast))[name = tensor<string, []>("op_28900_cast")];
            tensor<string, []> var_28902_equation_0 = const()[name = tensor<string, []>("op_28902_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28902_cast = einsum(equation = var_28902_equation_0, values = (var_28756_cast, var_28864_cast))[name = tensor<string, []>("op_28902_cast")];
            tensor<string, []> var_28904_equation_0 = const()[name = tensor<string, []>("op_28904_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28904_cast = einsum(equation = var_28904_equation_0, values = (var_28760_cast, var_28865_cast))[name = tensor<string, []>("op_28904_cast")];
            tensor<string, []> var_28906_equation_0 = const()[name = tensor<string, []>("op_28906_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_28906_cast = einsum(equation = var_28906_equation_0, values = (var_28764_cast, var_28866_cast))[name = tensor<string, []>("op_28906_cast")];
            tensor<bool, []> input_393_interleave_0 = const()[name = tensor<string, []>("input_393_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_393_cast = concat(axis = var_21105, interleave = input_393_interleave_0, values = (var_28868_cast, var_28870_cast, var_28872_cast, var_28874_cast, var_28876_cast, var_28878_cast, var_28880_cast, var_28882_cast, var_28884_cast, var_28886_cast, var_28888_cast, var_28890_cast, var_28892_cast, var_28894_cast, var_28896_cast, var_28898_cast, var_28900_cast, var_28902_cast, var_28904_cast, var_28906_cast))[name = tensor<string, []>("input_393_cast")];
            tensor<int32, [2]> var_28912 = const()[name = tensor<string, []>("op_28912"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_28914 = const()[name = tensor<string, []>("op_28914"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_28916_pad_type_0 = const()[name = tensor<string, []>("op_28916_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_28916_pad_0 = const()[name = tensor<string, []>("op_28916_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2292048576)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2295325440)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_28916_cast = conv(bias = mid_block_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16, dilations = var_28914, groups = var_21105, pad = var_28916_pad_0, pad_type = var_28916_pad_type_0, strides = var_28912, weight = mid_block_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16, x = input_393_cast)[name = tensor<string, []>("op_28916_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_195_cast = add(x = var_28916_cast, y = inputs_193_cast)[name = tensor<string, []>("inputs_195_cast")];
            tensor<int32, [1]> var_28920 = const()[name = tensor<string, []>("op_28920"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_195_cast = reduce_mean(axes = var_28920, keep_dims = var_21100, x = inputs_195_cast)[name = tensor<string, []>("channels_mean_195_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_195_cast = sub(x = inputs_195_cast, y = channels_mean_195_cast)[name = tensor<string, []>("zero_mean_195_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_195_cast = mul(x = zero_mean_195_cast, y = zero_mean_195_cast)[name = tensor<string, []>("zero_mean_sq_195_cast")];
            tensor<int32, [1]> var_28924 = const()[name = tensor<string, []>("op_28924"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_28925_cast = reduce_mean(axes = var_28924, keep_dims = var_21100, x = zero_mean_sq_195_cast)[name = tensor<string, []>("op_28925_cast")];
            tensor<fp16, []> var_28926_to_fp16 = const()[name = tensor<string, []>("op_28926_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_28927_cast = add(x = var_28925_cast, y = var_28926_to_fp16)[name = tensor<string, []>("op_28927_cast")];
            tensor<fp16, []> denom_195_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_195_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_195_cast = rsqrt(epsilon = denom_195_epsilon_0_to_fp16, x = var_28927_cast)[name = tensor<string, []>("denom_195_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_195_cast = mul(x = zero_mean_195_cast, y = denom_195_cast)[name = tensor<string, []>("out_195_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_28931_to_fp16 = const()[name = tensor<string, []>("op_28931_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2295328064)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_28932_cast = add(x = out_195_cast, y = var_28931_to_fp16)[name = tensor<string, []>("op_28932_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_28934_to_fp16 = const()[name = tensor<string, []>("op_28934_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2295330688)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_259_cast = mul(x = var_28932_cast, y = var_28934_to_fp16)[name = tensor<string, []>("hidden_states_259_cast")];
            tensor<int32, [2]> var_28941 = const()[name = tensor<string, []>("op_28941"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_28943 = const()[name = tensor<string, []>("op_28943"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_131_pad_type_0 = const()[name = tensor<string, []>("q_131_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_131_pad_0 = const()[name = tensor<string, []>("q_131_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2295333312)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_131_cast = conv(dilations = var_28943, groups = var_21105, pad = q_131_pad_0, pad_type = q_131_pad_type_0, strides = var_28941, weight = mid_block_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16, x = hidden_states_259_cast)[name = tensor<string, []>("q_131_cast")];
            tensor<int32, [2]> var_28947 = const()[name = tensor<string, []>("op_28947"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_28949 = const()[name = tensor<string, []>("op_28949"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_261_pad_type_0 = const()[name = tensor<string, []>("k_261_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_261_pad_0 = const()[name = tensor<string, []>("k_261_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2298610176)))];
            tensor<fp16, [2, 1280, 1, 77]> k_261_cast = conv(dilations = var_28949, groups = var_21105, pad = k_261_pad_0, pad_type = k_261_pad_type_0, strides = var_28947, weight = mid_block_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_261_cast")];
            tensor<int32, [2]> var_28953 = const()[name = tensor<string, []>("op_28953"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_28955 = const()[name = tensor<string, []>("op_28955"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_131_pad_type_0 = const()[name = tensor<string, []>("v_131_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_131_pad_0 = const()[name = tensor<string, []>("v_131_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2303853120)))];
            tensor<fp16, [2, 1280, 1, 77]> v_131_cast = conv(dilations = var_28955, groups = var_21105, pad = v_131_pad_0, pad_type = v_131_pad_type_0, strides = var_28953, weight = mid_block_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_131_cast")];
            tensor<int32, [4]> var_28959_begin_0 = const()[name = tensor<string, []>("op_28959_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_28959_end_0 = const()[name = tensor<string, []>("op_28959_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_28959_end_mask_0 = const()[name = tensor<string, []>("op_28959_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28959_cast = slice_by_index(begin = var_28959_begin_0, end = var_28959_end_0, end_mask = var_28959_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_28959_cast")];
            tensor<int32, [4]> var_28963_begin_0 = const()[name = tensor<string, []>("op_28963_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_28963_end_0 = const()[name = tensor<string, []>("op_28963_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_28963_end_mask_0 = const()[name = tensor<string, []>("op_28963_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28963_cast = slice_by_index(begin = var_28963_begin_0, end = var_28963_end_0, end_mask = var_28963_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_28963_cast")];
            tensor<int32, [4]> var_28967_begin_0 = const()[name = tensor<string, []>("op_28967_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_28967_end_0 = const()[name = tensor<string, []>("op_28967_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_28967_end_mask_0 = const()[name = tensor<string, []>("op_28967_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28967_cast = slice_by_index(begin = var_28967_begin_0, end = var_28967_end_0, end_mask = var_28967_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_28967_cast")];
            tensor<int32, [4]> var_28971_begin_0 = const()[name = tensor<string, []>("op_28971_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_28971_end_0 = const()[name = tensor<string, []>("op_28971_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_28971_end_mask_0 = const()[name = tensor<string, []>("op_28971_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28971_cast = slice_by_index(begin = var_28971_begin_0, end = var_28971_end_0, end_mask = var_28971_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_28971_cast")];
            tensor<int32, [4]> var_28975_begin_0 = const()[name = tensor<string, []>("op_28975_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_28975_end_0 = const()[name = tensor<string, []>("op_28975_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_28975_end_mask_0 = const()[name = tensor<string, []>("op_28975_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28975_cast = slice_by_index(begin = var_28975_begin_0, end = var_28975_end_0, end_mask = var_28975_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_28975_cast")];
            tensor<int32, [4]> var_28979_begin_0 = const()[name = tensor<string, []>("op_28979_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_28979_end_0 = const()[name = tensor<string, []>("op_28979_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_28979_end_mask_0 = const()[name = tensor<string, []>("op_28979_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28979_cast = slice_by_index(begin = var_28979_begin_0, end = var_28979_end_0, end_mask = var_28979_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_28979_cast")];
            tensor<int32, [4]> var_28983_begin_0 = const()[name = tensor<string, []>("op_28983_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_28983_end_0 = const()[name = tensor<string, []>("op_28983_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_28983_end_mask_0 = const()[name = tensor<string, []>("op_28983_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28983_cast = slice_by_index(begin = var_28983_begin_0, end = var_28983_end_0, end_mask = var_28983_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_28983_cast")];
            tensor<int32, [4]> var_28987_begin_0 = const()[name = tensor<string, []>("op_28987_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_28987_end_0 = const()[name = tensor<string, []>("op_28987_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_28987_end_mask_0 = const()[name = tensor<string, []>("op_28987_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28987_cast = slice_by_index(begin = var_28987_begin_0, end = var_28987_end_0, end_mask = var_28987_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_28987_cast")];
            tensor<int32, [4]> var_28991_begin_0 = const()[name = tensor<string, []>("op_28991_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_28991_end_0 = const()[name = tensor<string, []>("op_28991_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_28991_end_mask_0 = const()[name = tensor<string, []>("op_28991_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28991_cast = slice_by_index(begin = var_28991_begin_0, end = var_28991_end_0, end_mask = var_28991_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_28991_cast")];
            tensor<int32, [4]> var_28995_begin_0 = const()[name = tensor<string, []>("op_28995_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_28995_end_0 = const()[name = tensor<string, []>("op_28995_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_28995_end_mask_0 = const()[name = tensor<string, []>("op_28995_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28995_cast = slice_by_index(begin = var_28995_begin_0, end = var_28995_end_0, end_mask = var_28995_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_28995_cast")];
            tensor<int32, [4]> var_28999_begin_0 = const()[name = tensor<string, []>("op_28999_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_28999_end_0 = const()[name = tensor<string, []>("op_28999_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_28999_end_mask_0 = const()[name = tensor<string, []>("op_28999_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_28999_cast = slice_by_index(begin = var_28999_begin_0, end = var_28999_end_0, end_mask = var_28999_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_28999_cast")];
            tensor<int32, [4]> var_29003_begin_0 = const()[name = tensor<string, []>("op_29003_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_29003_end_0 = const()[name = tensor<string, []>("op_29003_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_29003_end_mask_0 = const()[name = tensor<string, []>("op_29003_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29003_cast = slice_by_index(begin = var_29003_begin_0, end = var_29003_end_0, end_mask = var_29003_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_29003_cast")];
            tensor<int32, [4]> var_29007_begin_0 = const()[name = tensor<string, []>("op_29007_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_29007_end_0 = const()[name = tensor<string, []>("op_29007_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_29007_end_mask_0 = const()[name = tensor<string, []>("op_29007_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29007_cast = slice_by_index(begin = var_29007_begin_0, end = var_29007_end_0, end_mask = var_29007_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_29007_cast")];
            tensor<int32, [4]> var_29011_begin_0 = const()[name = tensor<string, []>("op_29011_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_29011_end_0 = const()[name = tensor<string, []>("op_29011_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_29011_end_mask_0 = const()[name = tensor<string, []>("op_29011_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29011_cast = slice_by_index(begin = var_29011_begin_0, end = var_29011_end_0, end_mask = var_29011_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_29011_cast")];
            tensor<int32, [4]> var_29015_begin_0 = const()[name = tensor<string, []>("op_29015_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_29015_end_0 = const()[name = tensor<string, []>("op_29015_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_29015_end_mask_0 = const()[name = tensor<string, []>("op_29015_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29015_cast = slice_by_index(begin = var_29015_begin_0, end = var_29015_end_0, end_mask = var_29015_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_29015_cast")];
            tensor<int32, [4]> var_29019_begin_0 = const()[name = tensor<string, []>("op_29019_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_29019_end_0 = const()[name = tensor<string, []>("op_29019_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_29019_end_mask_0 = const()[name = tensor<string, []>("op_29019_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29019_cast = slice_by_index(begin = var_29019_begin_0, end = var_29019_end_0, end_mask = var_29019_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_29019_cast")];
            tensor<int32, [4]> var_29023_begin_0 = const()[name = tensor<string, []>("op_29023_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_29023_end_0 = const()[name = tensor<string, []>("op_29023_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_29023_end_mask_0 = const()[name = tensor<string, []>("op_29023_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29023_cast = slice_by_index(begin = var_29023_begin_0, end = var_29023_end_0, end_mask = var_29023_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_29023_cast")];
            tensor<int32, [4]> var_29027_begin_0 = const()[name = tensor<string, []>("op_29027_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_29027_end_0 = const()[name = tensor<string, []>("op_29027_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_29027_end_mask_0 = const()[name = tensor<string, []>("op_29027_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29027_cast = slice_by_index(begin = var_29027_begin_0, end = var_29027_end_0, end_mask = var_29027_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_29027_cast")];
            tensor<int32, [4]> var_29031_begin_0 = const()[name = tensor<string, []>("op_29031_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_29031_end_0 = const()[name = tensor<string, []>("op_29031_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_29031_end_mask_0 = const()[name = tensor<string, []>("op_29031_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29031_cast = slice_by_index(begin = var_29031_begin_0, end = var_29031_end_0, end_mask = var_29031_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_29031_cast")];
            tensor<int32, [4]> var_29035_begin_0 = const()[name = tensor<string, []>("op_29035_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_29035_end_0 = const()[name = tensor<string, []>("op_29035_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_29035_end_mask_0 = const()[name = tensor<string, []>("op_29035_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29035_cast = slice_by_index(begin = var_29035_begin_0, end = var_29035_end_0, end_mask = var_29035_end_mask_0, x = q_131_cast)[name = tensor<string, []>("op_29035_cast")];
            tensor<int32, [4]> k_263_perm_0 = const()[name = tensor<string, []>("k_263_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_29042_begin_0 = const()[name = tensor<string, []>("op_29042_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_29042_end_0 = const()[name = tensor<string, []>("op_29042_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_29042_end_mask_0 = const()[name = tensor<string, []>("op_29042_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_74 = transpose(perm = k_263_perm_0, x = k_261_cast)[name = tensor<string, []>("transpose_74")];
            tensor<fp16, [2, 77, 1, 64]> var_29042_cast = slice_by_index(begin = var_29042_begin_0, end = var_29042_end_0, end_mask = var_29042_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29042_cast")];
            tensor<int32, [4]> var_29046_begin_0 = const()[name = tensor<string, []>("op_29046_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_29046_end_0 = const()[name = tensor<string, []>("op_29046_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_29046_end_mask_0 = const()[name = tensor<string, []>("op_29046_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29046_cast = slice_by_index(begin = var_29046_begin_0, end = var_29046_end_0, end_mask = var_29046_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29046_cast")];
            tensor<int32, [4]> var_29050_begin_0 = const()[name = tensor<string, []>("op_29050_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_29050_end_0 = const()[name = tensor<string, []>("op_29050_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_29050_end_mask_0 = const()[name = tensor<string, []>("op_29050_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29050_cast = slice_by_index(begin = var_29050_begin_0, end = var_29050_end_0, end_mask = var_29050_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29050_cast")];
            tensor<int32, [4]> var_29054_begin_0 = const()[name = tensor<string, []>("op_29054_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_29054_end_0 = const()[name = tensor<string, []>("op_29054_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_29054_end_mask_0 = const()[name = tensor<string, []>("op_29054_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29054_cast = slice_by_index(begin = var_29054_begin_0, end = var_29054_end_0, end_mask = var_29054_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29054_cast")];
            tensor<int32, [4]> var_29058_begin_0 = const()[name = tensor<string, []>("op_29058_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_29058_end_0 = const()[name = tensor<string, []>("op_29058_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_29058_end_mask_0 = const()[name = tensor<string, []>("op_29058_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29058_cast = slice_by_index(begin = var_29058_begin_0, end = var_29058_end_0, end_mask = var_29058_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29058_cast")];
            tensor<int32, [4]> var_29062_begin_0 = const()[name = tensor<string, []>("op_29062_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_29062_end_0 = const()[name = tensor<string, []>("op_29062_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_29062_end_mask_0 = const()[name = tensor<string, []>("op_29062_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29062_cast = slice_by_index(begin = var_29062_begin_0, end = var_29062_end_0, end_mask = var_29062_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29062_cast")];
            tensor<int32, [4]> var_29066_begin_0 = const()[name = tensor<string, []>("op_29066_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_29066_end_0 = const()[name = tensor<string, []>("op_29066_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_29066_end_mask_0 = const()[name = tensor<string, []>("op_29066_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29066_cast = slice_by_index(begin = var_29066_begin_0, end = var_29066_end_0, end_mask = var_29066_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29066_cast")];
            tensor<int32, [4]> var_29070_begin_0 = const()[name = tensor<string, []>("op_29070_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_29070_end_0 = const()[name = tensor<string, []>("op_29070_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_29070_end_mask_0 = const()[name = tensor<string, []>("op_29070_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29070_cast = slice_by_index(begin = var_29070_begin_0, end = var_29070_end_0, end_mask = var_29070_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29070_cast")];
            tensor<int32, [4]> var_29074_begin_0 = const()[name = tensor<string, []>("op_29074_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_29074_end_0 = const()[name = tensor<string, []>("op_29074_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_29074_end_mask_0 = const()[name = tensor<string, []>("op_29074_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29074_cast = slice_by_index(begin = var_29074_begin_0, end = var_29074_end_0, end_mask = var_29074_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29074_cast")];
            tensor<int32, [4]> var_29078_begin_0 = const()[name = tensor<string, []>("op_29078_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_29078_end_0 = const()[name = tensor<string, []>("op_29078_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_29078_end_mask_0 = const()[name = tensor<string, []>("op_29078_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29078_cast = slice_by_index(begin = var_29078_begin_0, end = var_29078_end_0, end_mask = var_29078_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29078_cast")];
            tensor<int32, [4]> var_29082_begin_0 = const()[name = tensor<string, []>("op_29082_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_29082_end_0 = const()[name = tensor<string, []>("op_29082_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_29082_end_mask_0 = const()[name = tensor<string, []>("op_29082_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29082_cast = slice_by_index(begin = var_29082_begin_0, end = var_29082_end_0, end_mask = var_29082_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29082_cast")];
            tensor<int32, [4]> var_29086_begin_0 = const()[name = tensor<string, []>("op_29086_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_29086_end_0 = const()[name = tensor<string, []>("op_29086_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_29086_end_mask_0 = const()[name = tensor<string, []>("op_29086_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29086_cast = slice_by_index(begin = var_29086_begin_0, end = var_29086_end_0, end_mask = var_29086_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29086_cast")];
            tensor<int32, [4]> var_29090_begin_0 = const()[name = tensor<string, []>("op_29090_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_29090_end_0 = const()[name = tensor<string, []>("op_29090_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_29090_end_mask_0 = const()[name = tensor<string, []>("op_29090_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29090_cast = slice_by_index(begin = var_29090_begin_0, end = var_29090_end_0, end_mask = var_29090_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29090_cast")];
            tensor<int32, [4]> var_29094_begin_0 = const()[name = tensor<string, []>("op_29094_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_29094_end_0 = const()[name = tensor<string, []>("op_29094_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_29094_end_mask_0 = const()[name = tensor<string, []>("op_29094_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29094_cast = slice_by_index(begin = var_29094_begin_0, end = var_29094_end_0, end_mask = var_29094_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29094_cast")];
            tensor<int32, [4]> var_29098_begin_0 = const()[name = tensor<string, []>("op_29098_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_29098_end_0 = const()[name = tensor<string, []>("op_29098_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_29098_end_mask_0 = const()[name = tensor<string, []>("op_29098_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29098_cast = slice_by_index(begin = var_29098_begin_0, end = var_29098_end_0, end_mask = var_29098_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29098_cast")];
            tensor<int32, [4]> var_29102_begin_0 = const()[name = tensor<string, []>("op_29102_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_29102_end_0 = const()[name = tensor<string, []>("op_29102_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_29102_end_mask_0 = const()[name = tensor<string, []>("op_29102_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29102_cast = slice_by_index(begin = var_29102_begin_0, end = var_29102_end_0, end_mask = var_29102_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29102_cast")];
            tensor<int32, [4]> var_29106_begin_0 = const()[name = tensor<string, []>("op_29106_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_29106_end_0 = const()[name = tensor<string, []>("op_29106_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_29106_end_mask_0 = const()[name = tensor<string, []>("op_29106_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29106_cast = slice_by_index(begin = var_29106_begin_0, end = var_29106_end_0, end_mask = var_29106_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29106_cast")];
            tensor<int32, [4]> var_29110_begin_0 = const()[name = tensor<string, []>("op_29110_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_29110_end_0 = const()[name = tensor<string, []>("op_29110_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_29110_end_mask_0 = const()[name = tensor<string, []>("op_29110_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29110_cast = slice_by_index(begin = var_29110_begin_0, end = var_29110_end_0, end_mask = var_29110_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29110_cast")];
            tensor<int32, [4]> var_29114_begin_0 = const()[name = tensor<string, []>("op_29114_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_29114_end_0 = const()[name = tensor<string, []>("op_29114_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_29114_end_mask_0 = const()[name = tensor<string, []>("op_29114_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29114_cast = slice_by_index(begin = var_29114_begin_0, end = var_29114_end_0, end_mask = var_29114_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29114_cast")];
            tensor<int32, [4]> var_29118_begin_0 = const()[name = tensor<string, []>("op_29118_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_29118_end_0 = const()[name = tensor<string, []>("op_29118_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_29118_end_mask_0 = const()[name = tensor<string, []>("op_29118_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29118_cast = slice_by_index(begin = var_29118_begin_0, end = var_29118_end_0, end_mask = var_29118_end_mask_0, x = transpose_74)[name = tensor<string, []>("op_29118_cast")];
            tensor<int32, [4]> var_29120_begin_0 = const()[name = tensor<string, []>("op_29120_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_29120_end_0 = const()[name = tensor<string, []>("op_29120_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_29120_end_mask_0 = const()[name = tensor<string, []>("op_29120_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29120_cast = slice_by_index(begin = var_29120_begin_0, end = var_29120_end_0, end_mask = var_29120_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29120_cast")];
            tensor<int32, [4]> var_29124_begin_0 = const()[name = tensor<string, []>("op_29124_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_29124_end_0 = const()[name = tensor<string, []>("op_29124_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_29124_end_mask_0 = const()[name = tensor<string, []>("op_29124_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29124_cast = slice_by_index(begin = var_29124_begin_0, end = var_29124_end_0, end_mask = var_29124_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29124_cast")];
            tensor<int32, [4]> var_29128_begin_0 = const()[name = tensor<string, []>("op_29128_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_29128_end_0 = const()[name = tensor<string, []>("op_29128_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_29128_end_mask_0 = const()[name = tensor<string, []>("op_29128_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29128_cast = slice_by_index(begin = var_29128_begin_0, end = var_29128_end_0, end_mask = var_29128_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29128_cast")];
            tensor<int32, [4]> var_29132_begin_0 = const()[name = tensor<string, []>("op_29132_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_29132_end_0 = const()[name = tensor<string, []>("op_29132_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_29132_end_mask_0 = const()[name = tensor<string, []>("op_29132_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29132_cast = slice_by_index(begin = var_29132_begin_0, end = var_29132_end_0, end_mask = var_29132_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29132_cast")];
            tensor<int32, [4]> var_29136_begin_0 = const()[name = tensor<string, []>("op_29136_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_29136_end_0 = const()[name = tensor<string, []>("op_29136_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_29136_end_mask_0 = const()[name = tensor<string, []>("op_29136_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29136_cast = slice_by_index(begin = var_29136_begin_0, end = var_29136_end_0, end_mask = var_29136_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29136_cast")];
            tensor<int32, [4]> var_29140_begin_0 = const()[name = tensor<string, []>("op_29140_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_29140_end_0 = const()[name = tensor<string, []>("op_29140_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_29140_end_mask_0 = const()[name = tensor<string, []>("op_29140_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29140_cast = slice_by_index(begin = var_29140_begin_0, end = var_29140_end_0, end_mask = var_29140_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29140_cast")];
            tensor<int32, [4]> var_29144_begin_0 = const()[name = tensor<string, []>("op_29144_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_29144_end_0 = const()[name = tensor<string, []>("op_29144_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_29144_end_mask_0 = const()[name = tensor<string, []>("op_29144_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29144_cast = slice_by_index(begin = var_29144_begin_0, end = var_29144_end_0, end_mask = var_29144_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29144_cast")];
            tensor<int32, [4]> var_29148_begin_0 = const()[name = tensor<string, []>("op_29148_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_29148_end_0 = const()[name = tensor<string, []>("op_29148_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_29148_end_mask_0 = const()[name = tensor<string, []>("op_29148_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29148_cast = slice_by_index(begin = var_29148_begin_0, end = var_29148_end_0, end_mask = var_29148_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29148_cast")];
            tensor<int32, [4]> var_29152_begin_0 = const()[name = tensor<string, []>("op_29152_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_29152_end_0 = const()[name = tensor<string, []>("op_29152_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_29152_end_mask_0 = const()[name = tensor<string, []>("op_29152_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29152_cast = slice_by_index(begin = var_29152_begin_0, end = var_29152_end_0, end_mask = var_29152_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29152_cast")];
            tensor<int32, [4]> var_29156_begin_0 = const()[name = tensor<string, []>("op_29156_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_29156_end_0 = const()[name = tensor<string, []>("op_29156_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_29156_end_mask_0 = const()[name = tensor<string, []>("op_29156_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29156_cast = slice_by_index(begin = var_29156_begin_0, end = var_29156_end_0, end_mask = var_29156_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29156_cast")];
            tensor<int32, [4]> var_29160_begin_0 = const()[name = tensor<string, []>("op_29160_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_29160_end_0 = const()[name = tensor<string, []>("op_29160_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_29160_end_mask_0 = const()[name = tensor<string, []>("op_29160_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29160_cast = slice_by_index(begin = var_29160_begin_0, end = var_29160_end_0, end_mask = var_29160_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29160_cast")];
            tensor<int32, [4]> var_29164_begin_0 = const()[name = tensor<string, []>("op_29164_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_29164_end_0 = const()[name = tensor<string, []>("op_29164_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_29164_end_mask_0 = const()[name = tensor<string, []>("op_29164_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29164_cast = slice_by_index(begin = var_29164_begin_0, end = var_29164_end_0, end_mask = var_29164_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29164_cast")];
            tensor<int32, [4]> var_29168_begin_0 = const()[name = tensor<string, []>("op_29168_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_29168_end_0 = const()[name = tensor<string, []>("op_29168_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_29168_end_mask_0 = const()[name = tensor<string, []>("op_29168_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29168_cast = slice_by_index(begin = var_29168_begin_0, end = var_29168_end_0, end_mask = var_29168_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29168_cast")];
            tensor<int32, [4]> var_29172_begin_0 = const()[name = tensor<string, []>("op_29172_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_29172_end_0 = const()[name = tensor<string, []>("op_29172_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_29172_end_mask_0 = const()[name = tensor<string, []>("op_29172_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29172_cast = slice_by_index(begin = var_29172_begin_0, end = var_29172_end_0, end_mask = var_29172_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29172_cast")];
            tensor<int32, [4]> var_29176_begin_0 = const()[name = tensor<string, []>("op_29176_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_29176_end_0 = const()[name = tensor<string, []>("op_29176_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_29176_end_mask_0 = const()[name = tensor<string, []>("op_29176_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29176_cast = slice_by_index(begin = var_29176_begin_0, end = var_29176_end_0, end_mask = var_29176_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29176_cast")];
            tensor<int32, [4]> var_29180_begin_0 = const()[name = tensor<string, []>("op_29180_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_29180_end_0 = const()[name = tensor<string, []>("op_29180_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_29180_end_mask_0 = const()[name = tensor<string, []>("op_29180_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29180_cast = slice_by_index(begin = var_29180_begin_0, end = var_29180_end_0, end_mask = var_29180_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29180_cast")];
            tensor<int32, [4]> var_29184_begin_0 = const()[name = tensor<string, []>("op_29184_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_29184_end_0 = const()[name = tensor<string, []>("op_29184_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_29184_end_mask_0 = const()[name = tensor<string, []>("op_29184_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29184_cast = slice_by_index(begin = var_29184_begin_0, end = var_29184_end_0, end_mask = var_29184_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29184_cast")];
            tensor<int32, [4]> var_29188_begin_0 = const()[name = tensor<string, []>("op_29188_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_29188_end_0 = const()[name = tensor<string, []>("op_29188_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_29188_end_mask_0 = const()[name = tensor<string, []>("op_29188_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29188_cast = slice_by_index(begin = var_29188_begin_0, end = var_29188_end_0, end_mask = var_29188_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29188_cast")];
            tensor<int32, [4]> var_29192_begin_0 = const()[name = tensor<string, []>("op_29192_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_29192_end_0 = const()[name = tensor<string, []>("op_29192_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_29192_end_mask_0 = const()[name = tensor<string, []>("op_29192_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29192_cast = slice_by_index(begin = var_29192_begin_0, end = var_29192_end_0, end_mask = var_29192_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29192_cast")];
            tensor<int32, [4]> var_29196_begin_0 = const()[name = tensor<string, []>("op_29196_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_29196_end_0 = const()[name = tensor<string, []>("op_29196_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_29196_end_mask_0 = const()[name = tensor<string, []>("op_29196_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_29196_cast = slice_by_index(begin = var_29196_begin_0, end = var_29196_end_0, end_mask = var_29196_end_mask_0, x = v_131_cast)[name = tensor<string, []>("op_29196_cast")];
            tensor<string, []> var_29200_equation_0 = const()[name = tensor<string, []>("op_29200_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29200_cast = einsum(equation = var_29200_equation_0, values = (var_29042_cast, var_28959_cast))[name = tensor<string, []>("op_29200_cast")];
            tensor<fp16, []> var_29201_to_fp16 = const()[name = tensor<string, []>("op_29201_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2441_cast = mul(x = var_29200_cast, y = var_29201_to_fp16)[name = tensor<string, []>("aw_2441_cast")];
            tensor<string, []> var_29204_equation_0 = const()[name = tensor<string, []>("op_29204_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29204_cast = einsum(equation = var_29204_equation_0, values = (var_29046_cast, var_28963_cast))[name = tensor<string, []>("op_29204_cast")];
            tensor<fp16, []> var_29205_to_fp16 = const()[name = tensor<string, []>("op_29205_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2443_cast = mul(x = var_29204_cast, y = var_29205_to_fp16)[name = tensor<string, []>("aw_2443_cast")];
            tensor<string, []> var_29208_equation_0 = const()[name = tensor<string, []>("op_29208_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29208_cast = einsum(equation = var_29208_equation_0, values = (var_29050_cast, var_28967_cast))[name = tensor<string, []>("op_29208_cast")];
            tensor<fp16, []> var_29209_to_fp16 = const()[name = tensor<string, []>("op_29209_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2445_cast = mul(x = var_29208_cast, y = var_29209_to_fp16)[name = tensor<string, []>("aw_2445_cast")];
            tensor<string, []> var_29212_equation_0 = const()[name = tensor<string, []>("op_29212_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29212_cast = einsum(equation = var_29212_equation_0, values = (var_29054_cast, var_28971_cast))[name = tensor<string, []>("op_29212_cast")];
            tensor<fp16, []> var_29213_to_fp16 = const()[name = tensor<string, []>("op_29213_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2447_cast = mul(x = var_29212_cast, y = var_29213_to_fp16)[name = tensor<string, []>("aw_2447_cast")];
            tensor<string, []> var_29216_equation_0 = const()[name = tensor<string, []>("op_29216_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29216_cast = einsum(equation = var_29216_equation_0, values = (var_29058_cast, var_28975_cast))[name = tensor<string, []>("op_29216_cast")];
            tensor<fp16, []> var_29217_to_fp16 = const()[name = tensor<string, []>("op_29217_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2449_cast = mul(x = var_29216_cast, y = var_29217_to_fp16)[name = tensor<string, []>("aw_2449_cast")];
            tensor<string, []> var_29220_equation_0 = const()[name = tensor<string, []>("op_29220_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29220_cast = einsum(equation = var_29220_equation_0, values = (var_29062_cast, var_28979_cast))[name = tensor<string, []>("op_29220_cast")];
            tensor<fp16, []> var_29221_to_fp16 = const()[name = tensor<string, []>("op_29221_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2451_cast = mul(x = var_29220_cast, y = var_29221_to_fp16)[name = tensor<string, []>("aw_2451_cast")];
            tensor<string, []> var_29224_equation_0 = const()[name = tensor<string, []>("op_29224_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29224_cast = einsum(equation = var_29224_equation_0, values = (var_29066_cast, var_28983_cast))[name = tensor<string, []>("op_29224_cast")];
            tensor<fp16, []> var_29225_to_fp16 = const()[name = tensor<string, []>("op_29225_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2453_cast = mul(x = var_29224_cast, y = var_29225_to_fp16)[name = tensor<string, []>("aw_2453_cast")];
            tensor<string, []> var_29228_equation_0 = const()[name = tensor<string, []>("op_29228_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29228_cast = einsum(equation = var_29228_equation_0, values = (var_29070_cast, var_28987_cast))[name = tensor<string, []>("op_29228_cast")];
            tensor<fp16, []> var_29229_to_fp16 = const()[name = tensor<string, []>("op_29229_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2455_cast = mul(x = var_29228_cast, y = var_29229_to_fp16)[name = tensor<string, []>("aw_2455_cast")];
            tensor<string, []> var_29232_equation_0 = const()[name = tensor<string, []>("op_29232_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29232_cast = einsum(equation = var_29232_equation_0, values = (var_29074_cast, var_28991_cast))[name = tensor<string, []>("op_29232_cast")];
            tensor<fp16, []> var_29233_to_fp16 = const()[name = tensor<string, []>("op_29233_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2457_cast = mul(x = var_29232_cast, y = var_29233_to_fp16)[name = tensor<string, []>("aw_2457_cast")];
            tensor<string, []> var_29236_equation_0 = const()[name = tensor<string, []>("op_29236_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29236_cast = einsum(equation = var_29236_equation_0, values = (var_29078_cast, var_28995_cast))[name = tensor<string, []>("op_29236_cast")];
            tensor<fp16, []> var_29237_to_fp16 = const()[name = tensor<string, []>("op_29237_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2459_cast = mul(x = var_29236_cast, y = var_29237_to_fp16)[name = tensor<string, []>("aw_2459_cast")];
            tensor<string, []> var_29240_equation_0 = const()[name = tensor<string, []>("op_29240_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29240_cast = einsum(equation = var_29240_equation_0, values = (var_29082_cast, var_28999_cast))[name = tensor<string, []>("op_29240_cast")];
            tensor<fp16, []> var_29241_to_fp16 = const()[name = tensor<string, []>("op_29241_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2461_cast = mul(x = var_29240_cast, y = var_29241_to_fp16)[name = tensor<string, []>("aw_2461_cast")];
            tensor<string, []> var_29244_equation_0 = const()[name = tensor<string, []>("op_29244_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29244_cast = einsum(equation = var_29244_equation_0, values = (var_29086_cast, var_29003_cast))[name = tensor<string, []>("op_29244_cast")];
            tensor<fp16, []> var_29245_to_fp16 = const()[name = tensor<string, []>("op_29245_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2463_cast = mul(x = var_29244_cast, y = var_29245_to_fp16)[name = tensor<string, []>("aw_2463_cast")];
            tensor<string, []> var_29248_equation_0 = const()[name = tensor<string, []>("op_29248_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29248_cast = einsum(equation = var_29248_equation_0, values = (var_29090_cast, var_29007_cast))[name = tensor<string, []>("op_29248_cast")];
            tensor<fp16, []> var_29249_to_fp16 = const()[name = tensor<string, []>("op_29249_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2465_cast = mul(x = var_29248_cast, y = var_29249_to_fp16)[name = tensor<string, []>("aw_2465_cast")];
            tensor<string, []> var_29252_equation_0 = const()[name = tensor<string, []>("op_29252_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29252_cast = einsum(equation = var_29252_equation_0, values = (var_29094_cast, var_29011_cast))[name = tensor<string, []>("op_29252_cast")];
            tensor<fp16, []> var_29253_to_fp16 = const()[name = tensor<string, []>("op_29253_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2467_cast = mul(x = var_29252_cast, y = var_29253_to_fp16)[name = tensor<string, []>("aw_2467_cast")];
            tensor<string, []> var_29256_equation_0 = const()[name = tensor<string, []>("op_29256_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29256_cast = einsum(equation = var_29256_equation_0, values = (var_29098_cast, var_29015_cast))[name = tensor<string, []>("op_29256_cast")];
            tensor<fp16, []> var_29257_to_fp16 = const()[name = tensor<string, []>("op_29257_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2469_cast = mul(x = var_29256_cast, y = var_29257_to_fp16)[name = tensor<string, []>("aw_2469_cast")];
            tensor<string, []> var_29260_equation_0 = const()[name = tensor<string, []>("op_29260_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29260_cast = einsum(equation = var_29260_equation_0, values = (var_29102_cast, var_29019_cast))[name = tensor<string, []>("op_29260_cast")];
            tensor<fp16, []> var_29261_to_fp16 = const()[name = tensor<string, []>("op_29261_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2471_cast = mul(x = var_29260_cast, y = var_29261_to_fp16)[name = tensor<string, []>("aw_2471_cast")];
            tensor<string, []> var_29264_equation_0 = const()[name = tensor<string, []>("op_29264_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29264_cast = einsum(equation = var_29264_equation_0, values = (var_29106_cast, var_29023_cast))[name = tensor<string, []>("op_29264_cast")];
            tensor<fp16, []> var_29265_to_fp16 = const()[name = tensor<string, []>("op_29265_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2473_cast = mul(x = var_29264_cast, y = var_29265_to_fp16)[name = tensor<string, []>("aw_2473_cast")];
            tensor<string, []> var_29268_equation_0 = const()[name = tensor<string, []>("op_29268_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29268_cast = einsum(equation = var_29268_equation_0, values = (var_29110_cast, var_29027_cast))[name = tensor<string, []>("op_29268_cast")];
            tensor<fp16, []> var_29269_to_fp16 = const()[name = tensor<string, []>("op_29269_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2475_cast = mul(x = var_29268_cast, y = var_29269_to_fp16)[name = tensor<string, []>("aw_2475_cast")];
            tensor<string, []> var_29272_equation_0 = const()[name = tensor<string, []>("op_29272_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29272_cast = einsum(equation = var_29272_equation_0, values = (var_29114_cast, var_29031_cast))[name = tensor<string, []>("op_29272_cast")];
            tensor<fp16, []> var_29273_to_fp16 = const()[name = tensor<string, []>("op_29273_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2477_cast = mul(x = var_29272_cast, y = var_29273_to_fp16)[name = tensor<string, []>("aw_2477_cast")];
            tensor<string, []> var_29276_equation_0 = const()[name = tensor<string, []>("op_29276_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_29276_cast = einsum(equation = var_29276_equation_0, values = (var_29118_cast, var_29035_cast))[name = tensor<string, []>("op_29276_cast")];
            tensor<fp16, []> var_29277_to_fp16 = const()[name = tensor<string, []>("op_29277_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2479_cast = mul(x = var_29276_cast, y = var_29277_to_fp16)[name = tensor<string, []>("aw_2479_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29279_cast = softmax(axis = var_21105, x = aw_2441_cast)[name = tensor<string, []>("op_29279_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29280_cast = softmax(axis = var_21105, x = aw_2443_cast)[name = tensor<string, []>("op_29280_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29281_cast = softmax(axis = var_21105, x = aw_2445_cast)[name = tensor<string, []>("op_29281_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29282_cast = softmax(axis = var_21105, x = aw_2447_cast)[name = tensor<string, []>("op_29282_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29283_cast = softmax(axis = var_21105, x = aw_2449_cast)[name = tensor<string, []>("op_29283_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29284_cast = softmax(axis = var_21105, x = aw_2451_cast)[name = tensor<string, []>("op_29284_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29285_cast = softmax(axis = var_21105, x = aw_2453_cast)[name = tensor<string, []>("op_29285_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29286_cast = softmax(axis = var_21105, x = aw_2455_cast)[name = tensor<string, []>("op_29286_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29287_cast = softmax(axis = var_21105, x = aw_2457_cast)[name = tensor<string, []>("op_29287_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29288_cast = softmax(axis = var_21105, x = aw_2459_cast)[name = tensor<string, []>("op_29288_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29289_cast = softmax(axis = var_21105, x = aw_2461_cast)[name = tensor<string, []>("op_29289_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29290_cast = softmax(axis = var_21105, x = aw_2463_cast)[name = tensor<string, []>("op_29290_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29291_cast = softmax(axis = var_21105, x = aw_2465_cast)[name = tensor<string, []>("op_29291_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29292_cast = softmax(axis = var_21105, x = aw_2467_cast)[name = tensor<string, []>("op_29292_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29293_cast = softmax(axis = var_21105, x = aw_2469_cast)[name = tensor<string, []>("op_29293_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29294_cast = softmax(axis = var_21105, x = aw_2471_cast)[name = tensor<string, []>("op_29294_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29295_cast = softmax(axis = var_21105, x = aw_2473_cast)[name = tensor<string, []>("op_29295_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29296_cast = softmax(axis = var_21105, x = aw_2475_cast)[name = tensor<string, []>("op_29296_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29297_cast = softmax(axis = var_21105, x = aw_2477_cast)[name = tensor<string, []>("op_29297_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_29298_cast = softmax(axis = var_21105, x = aw_2479_cast)[name = tensor<string, []>("op_29298_cast")];
            tensor<string, []> var_29300_equation_0 = const()[name = tensor<string, []>("op_29300_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29300_cast = einsum(equation = var_29300_equation_0, values = (var_29120_cast, var_29279_cast))[name = tensor<string, []>("op_29300_cast")];
            tensor<string, []> var_29302_equation_0 = const()[name = tensor<string, []>("op_29302_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29302_cast = einsum(equation = var_29302_equation_0, values = (var_29124_cast, var_29280_cast))[name = tensor<string, []>("op_29302_cast")];
            tensor<string, []> var_29304_equation_0 = const()[name = tensor<string, []>("op_29304_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29304_cast = einsum(equation = var_29304_equation_0, values = (var_29128_cast, var_29281_cast))[name = tensor<string, []>("op_29304_cast")];
            tensor<string, []> var_29306_equation_0 = const()[name = tensor<string, []>("op_29306_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29306_cast = einsum(equation = var_29306_equation_0, values = (var_29132_cast, var_29282_cast))[name = tensor<string, []>("op_29306_cast")];
            tensor<string, []> var_29308_equation_0 = const()[name = tensor<string, []>("op_29308_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29308_cast = einsum(equation = var_29308_equation_0, values = (var_29136_cast, var_29283_cast))[name = tensor<string, []>("op_29308_cast")];
            tensor<string, []> var_29310_equation_0 = const()[name = tensor<string, []>("op_29310_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29310_cast = einsum(equation = var_29310_equation_0, values = (var_29140_cast, var_29284_cast))[name = tensor<string, []>("op_29310_cast")];
            tensor<string, []> var_29312_equation_0 = const()[name = tensor<string, []>("op_29312_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29312_cast = einsum(equation = var_29312_equation_0, values = (var_29144_cast, var_29285_cast))[name = tensor<string, []>("op_29312_cast")];
            tensor<string, []> var_29314_equation_0 = const()[name = tensor<string, []>("op_29314_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29314_cast = einsum(equation = var_29314_equation_0, values = (var_29148_cast, var_29286_cast))[name = tensor<string, []>("op_29314_cast")];
            tensor<string, []> var_29316_equation_0 = const()[name = tensor<string, []>("op_29316_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29316_cast = einsum(equation = var_29316_equation_0, values = (var_29152_cast, var_29287_cast))[name = tensor<string, []>("op_29316_cast")];
            tensor<string, []> var_29318_equation_0 = const()[name = tensor<string, []>("op_29318_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29318_cast = einsum(equation = var_29318_equation_0, values = (var_29156_cast, var_29288_cast))[name = tensor<string, []>("op_29318_cast")];
            tensor<string, []> var_29320_equation_0 = const()[name = tensor<string, []>("op_29320_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29320_cast = einsum(equation = var_29320_equation_0, values = (var_29160_cast, var_29289_cast))[name = tensor<string, []>("op_29320_cast")];
            tensor<string, []> var_29322_equation_0 = const()[name = tensor<string, []>("op_29322_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29322_cast = einsum(equation = var_29322_equation_0, values = (var_29164_cast, var_29290_cast))[name = tensor<string, []>("op_29322_cast")];
            tensor<string, []> var_29324_equation_0 = const()[name = tensor<string, []>("op_29324_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29324_cast = einsum(equation = var_29324_equation_0, values = (var_29168_cast, var_29291_cast))[name = tensor<string, []>("op_29324_cast")];
            tensor<string, []> var_29326_equation_0 = const()[name = tensor<string, []>("op_29326_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29326_cast = einsum(equation = var_29326_equation_0, values = (var_29172_cast, var_29292_cast))[name = tensor<string, []>("op_29326_cast")];
            tensor<string, []> var_29328_equation_0 = const()[name = tensor<string, []>("op_29328_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29328_cast = einsum(equation = var_29328_equation_0, values = (var_29176_cast, var_29293_cast))[name = tensor<string, []>("op_29328_cast")];
            tensor<string, []> var_29330_equation_0 = const()[name = tensor<string, []>("op_29330_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29330_cast = einsum(equation = var_29330_equation_0, values = (var_29180_cast, var_29294_cast))[name = tensor<string, []>("op_29330_cast")];
            tensor<string, []> var_29332_equation_0 = const()[name = tensor<string, []>("op_29332_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29332_cast = einsum(equation = var_29332_equation_0, values = (var_29184_cast, var_29295_cast))[name = tensor<string, []>("op_29332_cast")];
            tensor<string, []> var_29334_equation_0 = const()[name = tensor<string, []>("op_29334_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29334_cast = einsum(equation = var_29334_equation_0, values = (var_29188_cast, var_29296_cast))[name = tensor<string, []>("op_29334_cast")];
            tensor<string, []> var_29336_equation_0 = const()[name = tensor<string, []>("op_29336_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29336_cast = einsum(equation = var_29336_equation_0, values = (var_29192_cast, var_29297_cast))[name = tensor<string, []>("op_29336_cast")];
            tensor<string, []> var_29338_equation_0 = const()[name = tensor<string, []>("op_29338_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29338_cast = einsum(equation = var_29338_equation_0, values = (var_29196_cast, var_29298_cast))[name = tensor<string, []>("op_29338_cast")];
            tensor<bool, []> input_395_interleave_0 = const()[name = tensor<string, []>("input_395_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_395_cast = concat(axis = var_21105, interleave = input_395_interleave_0, values = (var_29300_cast, var_29302_cast, var_29304_cast, var_29306_cast, var_29308_cast, var_29310_cast, var_29312_cast, var_29314_cast, var_29316_cast, var_29318_cast, var_29320_cast, var_29322_cast, var_29324_cast, var_29326_cast, var_29328_cast, var_29330_cast, var_29332_cast, var_29334_cast, var_29336_cast, var_29338_cast))[name = tensor<string, []>("input_395_cast")];
            tensor<int32, [2]> var_29344 = const()[name = tensor<string, []>("op_29344"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_29346 = const()[name = tensor<string, []>("op_29346"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_29348_pad_type_0 = const()[name = tensor<string, []>("op_29348_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_29348_pad_0 = const()[name = tensor<string, []>("op_29348_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2309096064)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2312372928)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_29348_cast = conv(bias = mid_block_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16, dilations = var_29346, groups = var_21105, pad = var_29348_pad_0, pad_type = var_29348_pad_type_0, strides = var_29344, weight = mid_block_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16, x = input_395_cast)[name = tensor<string, []>("op_29348_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_197_cast = add(x = var_29348_cast, y = inputs_195_cast)[name = tensor<string, []>("inputs_197_cast")];
            tensor<int32, [1]> var_29352 = const()[name = tensor<string, []>("op_29352"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_197_cast = reduce_mean(axes = var_29352, keep_dims = var_21100, x = inputs_197_cast)[name = tensor<string, []>("channels_mean_197_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_197_cast = sub(x = inputs_197_cast, y = channels_mean_197_cast)[name = tensor<string, []>("zero_mean_197_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_197_cast = mul(x = zero_mean_197_cast, y = zero_mean_197_cast)[name = tensor<string, []>("zero_mean_sq_197_cast")];
            tensor<int32, [1]> var_29356 = const()[name = tensor<string, []>("op_29356"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_29357_cast = reduce_mean(axes = var_29356, keep_dims = var_21100, x = zero_mean_sq_197_cast)[name = tensor<string, []>("op_29357_cast")];
            tensor<fp16, []> var_29358_to_fp16 = const()[name = tensor<string, []>("op_29358_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_29359_cast = add(x = var_29357_cast, y = var_29358_to_fp16)[name = tensor<string, []>("op_29359_cast")];
            tensor<fp16, []> denom_197_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_197_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_197_cast = rsqrt(epsilon = denom_197_epsilon_0_to_fp16, x = var_29359_cast)[name = tensor<string, []>("denom_197_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_197_cast = mul(x = zero_mean_197_cast, y = denom_197_cast)[name = tensor<string, []>("out_197_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_29363_to_fp16 = const()[name = tensor<string, []>("op_29363_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2312375552)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_29364_cast = add(x = out_197_cast, y = var_29363_to_fp16)[name = tensor<string, []>("op_29364_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_29366_to_fp16 = const()[name = tensor<string, []>("op_29366_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2312378176)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_397_cast = mul(x = var_29364_cast, y = var_29366_to_fp16)[name = tensor<string, []>("input_397_cast")];
            tensor<int32, [2]> var_29374 = const()[name = tensor<string, []>("op_29374"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_29376 = const()[name = tensor<string, []>("op_29376"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_29378_pad_type_0 = const()[name = tensor<string, []>("op_29378_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_29378_pad_0 = const()[name = tensor<string, []>("op_29378_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2312380800)))];
            tensor<fp16, [10240]> mid_block_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2338595264)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_29378_cast = conv(bias = mid_block_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16, dilations = var_29376, groups = var_21105, pad = var_29378_pad_0, pad_type = var_29378_pad_type_0, strides = var_29374, weight = mid_block_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16, x = input_397_cast)[name = tensor<string, []>("op_29378_cast")];
            tensor<int32, [2]> var_29379_split_sizes_0 = const()[name = tensor<string, []>("op_29379_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_29379_axis_0 = const()[name = tensor<string, []>("op_29379_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_29379_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_29379_cast_1 = split(axis = var_29379_axis_0, split_sizes = var_29379_split_sizes_0, x = var_29378_cast)[name = tensor<string, []>("op_29379_cast")];
            tensor<string, []> var_29381_mode_0 = const()[name = tensor<string, []>("op_29381_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_29381_cast = gelu(mode = var_29381_mode_0, x = var_29379_cast_1)[name = tensor<string, []>("op_29381_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_399_cast = mul(x = var_29379_cast_0, y = var_29381_cast)[name = tensor<string, []>("input_399_cast")];
            tensor<int32, [2]> var_29385 = const()[name = tensor<string, []>("op_29385"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_29387 = const()[name = tensor<string, []>("op_29387"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_29389_pad_type_0 = const()[name = tensor<string, []>("op_29389_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_29389_pad_0 = const()[name = tensor<string, []>("op_29389_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> mid_block_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2338615808)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2351723072)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_29389_cast = conv(bias = mid_block_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16, dilations = var_29387, groups = var_21105, pad = var_29389_pad_0, pad_type = var_29389_pad_type_0, strides = var_29385, weight = mid_block_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16, x = input_399_cast)[name = tensor<string, []>("op_29389_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_199_cast = add(x = var_29389_cast, y = inputs_197_cast)[name = tensor<string, []>("inputs_199_cast")];
            tensor<int32, [1]> var_29399 = const()[name = tensor<string, []>("op_29399"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_199_cast = reduce_mean(axes = var_29399, keep_dims = var_21100, x = inputs_199_cast)[name = tensor<string, []>("channels_mean_199_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_199_cast = sub(x = inputs_199_cast, y = channels_mean_199_cast)[name = tensor<string, []>("zero_mean_199_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_199_cast = mul(x = zero_mean_199_cast, y = zero_mean_199_cast)[name = tensor<string, []>("zero_mean_sq_199_cast")];
            tensor<int32, [1]> var_29403 = const()[name = tensor<string, []>("op_29403"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_29404_cast = reduce_mean(axes = var_29403, keep_dims = var_21100, x = zero_mean_sq_199_cast)[name = tensor<string, []>("op_29404_cast")];
            tensor<fp16, []> var_29405_to_fp16 = const()[name = tensor<string, []>("op_29405_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_29406_cast = add(x = var_29404_cast, y = var_29405_to_fp16)[name = tensor<string, []>("op_29406_cast")];
            tensor<fp16, []> denom_199_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_199_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_199_cast = rsqrt(epsilon = denom_199_epsilon_0_to_fp16, x = var_29406_cast)[name = tensor<string, []>("denom_199_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_199_cast = mul(x = zero_mean_199_cast, y = denom_199_cast)[name = tensor<string, []>("out_199_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_29410_to_fp16 = const()[name = tensor<string, []>("op_29410_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2351725696)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_29411_cast = add(x = out_199_cast, y = var_29410_to_fp16)[name = tensor<string, []>("op_29411_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_29413_to_fp16 = const()[name = tensor<string, []>("op_29413_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2351728320)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_263_cast = mul(x = var_29411_cast, y = var_29413_to_fp16)[name = tensor<string, []>("hidden_states_263_cast")];
            tensor<int32, [2]> var_29420 = const()[name = tensor<string, []>("op_29420"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_29422 = const()[name = tensor<string, []>("op_29422"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_133_pad_type_0 = const()[name = tensor<string, []>("q_133_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_133_pad_0 = const()[name = tensor<string, []>("q_133_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2351730944)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_133_cast = conv(dilations = var_29422, groups = var_21105, pad = q_133_pad_0, pad_type = q_133_pad_type_0, strides = var_29420, weight = mid_block_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16, x = hidden_states_263_cast)[name = tensor<string, []>("q_133_cast")];
            tensor<int32, [2]> var_29426 = const()[name = tensor<string, []>("op_29426"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_29428 = const()[name = tensor<string, []>("op_29428"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_265_pad_type_0 = const()[name = tensor<string, []>("k_265_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_265_pad_0 = const()[name = tensor<string, []>("k_265_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2355007808)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_265_cast = conv(dilations = var_29428, groups = var_21105, pad = k_265_pad_0, pad_type = k_265_pad_type_0, strides = var_29426, weight = mid_block_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16, x = hidden_states_263_cast)[name = tensor<string, []>("k_265_cast")];
            tensor<int32, [2]> var_29432 = const()[name = tensor<string, []>("op_29432"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_29434 = const()[name = tensor<string, []>("op_29434"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_133_pad_type_0 = const()[name = tensor<string, []>("v_133_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_133_pad_0 = const()[name = tensor<string, []>("v_133_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2358284672)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_133_cast = conv(dilations = var_29434, groups = var_21105, pad = v_133_pad_0, pad_type = v_133_pad_type_0, strides = var_29432, weight = mid_block_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16, x = hidden_states_263_cast)[name = tensor<string, []>("v_133_cast")];
            tensor<int32, [4]> var_29438_begin_0 = const()[name = tensor<string, []>("op_29438_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_29438_end_0 = const()[name = tensor<string, []>("op_29438_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_29438_end_mask_0 = const()[name = tensor<string, []>("op_29438_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29438_cast = slice_by_index(begin = var_29438_begin_0, end = var_29438_end_0, end_mask = var_29438_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29438_cast")];
            tensor<int32, [4]> var_29442_begin_0 = const()[name = tensor<string, []>("op_29442_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_29442_end_0 = const()[name = tensor<string, []>("op_29442_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_29442_end_mask_0 = const()[name = tensor<string, []>("op_29442_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29442_cast = slice_by_index(begin = var_29442_begin_0, end = var_29442_end_0, end_mask = var_29442_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29442_cast")];
            tensor<int32, [4]> var_29446_begin_0 = const()[name = tensor<string, []>("op_29446_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_29446_end_0 = const()[name = tensor<string, []>("op_29446_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_29446_end_mask_0 = const()[name = tensor<string, []>("op_29446_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29446_cast = slice_by_index(begin = var_29446_begin_0, end = var_29446_end_0, end_mask = var_29446_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29446_cast")];
            tensor<int32, [4]> var_29450_begin_0 = const()[name = tensor<string, []>("op_29450_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_29450_end_0 = const()[name = tensor<string, []>("op_29450_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_29450_end_mask_0 = const()[name = tensor<string, []>("op_29450_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29450_cast = slice_by_index(begin = var_29450_begin_0, end = var_29450_end_0, end_mask = var_29450_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29450_cast")];
            tensor<int32, [4]> var_29454_begin_0 = const()[name = tensor<string, []>("op_29454_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_29454_end_0 = const()[name = tensor<string, []>("op_29454_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_29454_end_mask_0 = const()[name = tensor<string, []>("op_29454_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29454_cast = slice_by_index(begin = var_29454_begin_0, end = var_29454_end_0, end_mask = var_29454_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29454_cast")];
            tensor<int32, [4]> var_29458_begin_0 = const()[name = tensor<string, []>("op_29458_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_29458_end_0 = const()[name = tensor<string, []>("op_29458_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_29458_end_mask_0 = const()[name = tensor<string, []>("op_29458_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29458_cast = slice_by_index(begin = var_29458_begin_0, end = var_29458_end_0, end_mask = var_29458_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29458_cast")];
            tensor<int32, [4]> var_29462_begin_0 = const()[name = tensor<string, []>("op_29462_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_29462_end_0 = const()[name = tensor<string, []>("op_29462_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_29462_end_mask_0 = const()[name = tensor<string, []>("op_29462_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29462_cast = slice_by_index(begin = var_29462_begin_0, end = var_29462_end_0, end_mask = var_29462_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29462_cast")];
            tensor<int32, [4]> var_29466_begin_0 = const()[name = tensor<string, []>("op_29466_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_29466_end_0 = const()[name = tensor<string, []>("op_29466_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_29466_end_mask_0 = const()[name = tensor<string, []>("op_29466_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29466_cast = slice_by_index(begin = var_29466_begin_0, end = var_29466_end_0, end_mask = var_29466_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29466_cast")];
            tensor<int32, [4]> var_29470_begin_0 = const()[name = tensor<string, []>("op_29470_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_29470_end_0 = const()[name = tensor<string, []>("op_29470_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_29470_end_mask_0 = const()[name = tensor<string, []>("op_29470_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29470_cast = slice_by_index(begin = var_29470_begin_0, end = var_29470_end_0, end_mask = var_29470_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29470_cast")];
            tensor<int32, [4]> var_29474_begin_0 = const()[name = tensor<string, []>("op_29474_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_29474_end_0 = const()[name = tensor<string, []>("op_29474_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_29474_end_mask_0 = const()[name = tensor<string, []>("op_29474_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29474_cast = slice_by_index(begin = var_29474_begin_0, end = var_29474_end_0, end_mask = var_29474_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29474_cast")];
            tensor<int32, [4]> var_29478_begin_0 = const()[name = tensor<string, []>("op_29478_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_29478_end_0 = const()[name = tensor<string, []>("op_29478_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_29478_end_mask_0 = const()[name = tensor<string, []>("op_29478_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29478_cast = slice_by_index(begin = var_29478_begin_0, end = var_29478_end_0, end_mask = var_29478_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29478_cast")];
            tensor<int32, [4]> var_29482_begin_0 = const()[name = tensor<string, []>("op_29482_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_29482_end_0 = const()[name = tensor<string, []>("op_29482_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_29482_end_mask_0 = const()[name = tensor<string, []>("op_29482_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29482_cast = slice_by_index(begin = var_29482_begin_0, end = var_29482_end_0, end_mask = var_29482_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29482_cast")];
            tensor<int32, [4]> var_29486_begin_0 = const()[name = tensor<string, []>("op_29486_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_29486_end_0 = const()[name = tensor<string, []>("op_29486_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_29486_end_mask_0 = const()[name = tensor<string, []>("op_29486_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29486_cast = slice_by_index(begin = var_29486_begin_0, end = var_29486_end_0, end_mask = var_29486_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29486_cast")];
            tensor<int32, [4]> var_29490_begin_0 = const()[name = tensor<string, []>("op_29490_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_29490_end_0 = const()[name = tensor<string, []>("op_29490_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_29490_end_mask_0 = const()[name = tensor<string, []>("op_29490_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29490_cast = slice_by_index(begin = var_29490_begin_0, end = var_29490_end_0, end_mask = var_29490_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29490_cast")];
            tensor<int32, [4]> var_29494_begin_0 = const()[name = tensor<string, []>("op_29494_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_29494_end_0 = const()[name = tensor<string, []>("op_29494_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_29494_end_mask_0 = const()[name = tensor<string, []>("op_29494_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29494_cast = slice_by_index(begin = var_29494_begin_0, end = var_29494_end_0, end_mask = var_29494_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29494_cast")];
            tensor<int32, [4]> var_29498_begin_0 = const()[name = tensor<string, []>("op_29498_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_29498_end_0 = const()[name = tensor<string, []>("op_29498_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_29498_end_mask_0 = const()[name = tensor<string, []>("op_29498_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29498_cast = slice_by_index(begin = var_29498_begin_0, end = var_29498_end_0, end_mask = var_29498_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29498_cast")];
            tensor<int32, [4]> var_29502_begin_0 = const()[name = tensor<string, []>("op_29502_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_29502_end_0 = const()[name = tensor<string, []>("op_29502_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_29502_end_mask_0 = const()[name = tensor<string, []>("op_29502_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29502_cast = slice_by_index(begin = var_29502_begin_0, end = var_29502_end_0, end_mask = var_29502_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29502_cast")];
            tensor<int32, [4]> var_29506_begin_0 = const()[name = tensor<string, []>("op_29506_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_29506_end_0 = const()[name = tensor<string, []>("op_29506_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_29506_end_mask_0 = const()[name = tensor<string, []>("op_29506_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29506_cast = slice_by_index(begin = var_29506_begin_0, end = var_29506_end_0, end_mask = var_29506_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29506_cast")];
            tensor<int32, [4]> var_29510_begin_0 = const()[name = tensor<string, []>("op_29510_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_29510_end_0 = const()[name = tensor<string, []>("op_29510_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_29510_end_mask_0 = const()[name = tensor<string, []>("op_29510_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29510_cast = slice_by_index(begin = var_29510_begin_0, end = var_29510_end_0, end_mask = var_29510_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29510_cast")];
            tensor<int32, [4]> var_29514_begin_0 = const()[name = tensor<string, []>("op_29514_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_29514_end_0 = const()[name = tensor<string, []>("op_29514_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_29514_end_mask_0 = const()[name = tensor<string, []>("op_29514_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29514_cast = slice_by_index(begin = var_29514_begin_0, end = var_29514_end_0, end_mask = var_29514_end_mask_0, x = q_133_cast)[name = tensor<string, []>("op_29514_cast")];
            tensor<int32, [4]> k_267_perm_0 = const()[name = tensor<string, []>("k_267_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_29521_begin_0 = const()[name = tensor<string, []>("op_29521_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_29521_end_0 = const()[name = tensor<string, []>("op_29521_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_29521_end_mask_0 = const()[name = tensor<string, []>("op_29521_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_73 = transpose(perm = k_267_perm_0, x = k_265_cast)[name = tensor<string, []>("transpose_73")];
            tensor<fp16, [2, 1024, 1, 64]> var_29521_cast = slice_by_index(begin = var_29521_begin_0, end = var_29521_end_0, end_mask = var_29521_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29521_cast")];
            tensor<int32, [4]> var_29525_begin_0 = const()[name = tensor<string, []>("op_29525_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_29525_end_0 = const()[name = tensor<string, []>("op_29525_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_29525_end_mask_0 = const()[name = tensor<string, []>("op_29525_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_29525_cast = slice_by_index(begin = var_29525_begin_0, end = var_29525_end_0, end_mask = var_29525_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29525_cast")];
            tensor<int32, [4]> var_29529_begin_0 = const()[name = tensor<string, []>("op_29529_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_29529_end_0 = const()[name = tensor<string, []>("op_29529_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_29529_end_mask_0 = const()[name = tensor<string, []>("op_29529_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_29529_cast = slice_by_index(begin = var_29529_begin_0, end = var_29529_end_0, end_mask = var_29529_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29529_cast")];
            tensor<int32, [4]> var_29533_begin_0 = const()[name = tensor<string, []>("op_29533_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_29533_end_0 = const()[name = tensor<string, []>("op_29533_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_29533_end_mask_0 = const()[name = tensor<string, []>("op_29533_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_29533_cast = slice_by_index(begin = var_29533_begin_0, end = var_29533_end_0, end_mask = var_29533_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29533_cast")];
            tensor<int32, [4]> var_29537_begin_0 = const()[name = tensor<string, []>("op_29537_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_29537_end_0 = const()[name = tensor<string, []>("op_29537_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_29537_end_mask_0 = const()[name = tensor<string, []>("op_29537_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_29537_cast = slice_by_index(begin = var_29537_begin_0, end = var_29537_end_0, end_mask = var_29537_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29537_cast")];
            tensor<int32, [4]> var_29541_begin_0 = const()[name = tensor<string, []>("op_29541_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_29541_end_0 = const()[name = tensor<string, []>("op_29541_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_29541_end_mask_0 = const()[name = tensor<string, []>("op_29541_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_29541_cast = slice_by_index(begin = var_29541_begin_0, end = var_29541_end_0, end_mask = var_29541_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29541_cast")];
            tensor<int32, [4]> var_29545_begin_0 = const()[name = tensor<string, []>("op_29545_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_29545_end_0 = const()[name = tensor<string, []>("op_29545_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_29545_end_mask_0 = const()[name = tensor<string, []>("op_29545_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_29545_cast = slice_by_index(begin = var_29545_begin_0, end = var_29545_end_0, end_mask = var_29545_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29545_cast")];
            tensor<int32, [4]> var_29549_begin_0 = const()[name = tensor<string, []>("op_29549_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_29549_end_0 = const()[name = tensor<string, []>("op_29549_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_29549_end_mask_0 = const()[name = tensor<string, []>("op_29549_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_29549_cast = slice_by_index(begin = var_29549_begin_0, end = var_29549_end_0, end_mask = var_29549_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29549_cast")];
            tensor<int32, [4]> var_29553_begin_0 = const()[name = tensor<string, []>("op_29553_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_29553_end_0 = const()[name = tensor<string, []>("op_29553_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_29553_end_mask_0 = const()[name = tensor<string, []>("op_29553_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_29553_cast = slice_by_index(begin = var_29553_begin_0, end = var_29553_end_0, end_mask = var_29553_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29553_cast")];
            tensor<int32, [4]> var_29557_begin_0 = const()[name = tensor<string, []>("op_29557_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_29557_end_0 = const()[name = tensor<string, []>("op_29557_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_29557_end_mask_0 = const()[name = tensor<string, []>("op_29557_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_29557_cast = slice_by_index(begin = var_29557_begin_0, end = var_29557_end_0, end_mask = var_29557_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29557_cast")];
            tensor<int32, [4]> var_29561_begin_0 = const()[name = tensor<string, []>("op_29561_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_29561_end_0 = const()[name = tensor<string, []>("op_29561_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_29561_end_mask_0 = const()[name = tensor<string, []>("op_29561_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_29561_cast = slice_by_index(begin = var_29561_begin_0, end = var_29561_end_0, end_mask = var_29561_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29561_cast")];
            tensor<int32, [4]> var_29565_begin_0 = const()[name = tensor<string, []>("op_29565_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_29565_end_0 = const()[name = tensor<string, []>("op_29565_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_29565_end_mask_0 = const()[name = tensor<string, []>("op_29565_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_29565_cast = slice_by_index(begin = var_29565_begin_0, end = var_29565_end_0, end_mask = var_29565_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29565_cast")];
            tensor<int32, [4]> var_29569_begin_0 = const()[name = tensor<string, []>("op_29569_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_29569_end_0 = const()[name = tensor<string, []>("op_29569_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_29569_end_mask_0 = const()[name = tensor<string, []>("op_29569_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_29569_cast = slice_by_index(begin = var_29569_begin_0, end = var_29569_end_0, end_mask = var_29569_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29569_cast")];
            tensor<int32, [4]> var_29573_begin_0 = const()[name = tensor<string, []>("op_29573_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_29573_end_0 = const()[name = tensor<string, []>("op_29573_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_29573_end_mask_0 = const()[name = tensor<string, []>("op_29573_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_29573_cast = slice_by_index(begin = var_29573_begin_0, end = var_29573_end_0, end_mask = var_29573_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29573_cast")];
            tensor<int32, [4]> var_29577_begin_0 = const()[name = tensor<string, []>("op_29577_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_29577_end_0 = const()[name = tensor<string, []>("op_29577_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_29577_end_mask_0 = const()[name = tensor<string, []>("op_29577_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_29577_cast = slice_by_index(begin = var_29577_begin_0, end = var_29577_end_0, end_mask = var_29577_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29577_cast")];
            tensor<int32, [4]> var_29581_begin_0 = const()[name = tensor<string, []>("op_29581_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_29581_end_0 = const()[name = tensor<string, []>("op_29581_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_29581_end_mask_0 = const()[name = tensor<string, []>("op_29581_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_29581_cast = slice_by_index(begin = var_29581_begin_0, end = var_29581_end_0, end_mask = var_29581_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29581_cast")];
            tensor<int32, [4]> var_29585_begin_0 = const()[name = tensor<string, []>("op_29585_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_29585_end_0 = const()[name = tensor<string, []>("op_29585_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_29585_end_mask_0 = const()[name = tensor<string, []>("op_29585_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_29585_cast = slice_by_index(begin = var_29585_begin_0, end = var_29585_end_0, end_mask = var_29585_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29585_cast")];
            tensor<int32, [4]> var_29589_begin_0 = const()[name = tensor<string, []>("op_29589_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_29589_end_0 = const()[name = tensor<string, []>("op_29589_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_29589_end_mask_0 = const()[name = tensor<string, []>("op_29589_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_29589_cast = slice_by_index(begin = var_29589_begin_0, end = var_29589_end_0, end_mask = var_29589_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29589_cast")];
            tensor<int32, [4]> var_29593_begin_0 = const()[name = tensor<string, []>("op_29593_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_29593_end_0 = const()[name = tensor<string, []>("op_29593_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_29593_end_mask_0 = const()[name = tensor<string, []>("op_29593_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_29593_cast = slice_by_index(begin = var_29593_begin_0, end = var_29593_end_0, end_mask = var_29593_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29593_cast")];
            tensor<int32, [4]> var_29597_begin_0 = const()[name = tensor<string, []>("op_29597_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_29597_end_0 = const()[name = tensor<string, []>("op_29597_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_29597_end_mask_0 = const()[name = tensor<string, []>("op_29597_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_29597_cast = slice_by_index(begin = var_29597_begin_0, end = var_29597_end_0, end_mask = var_29597_end_mask_0, x = transpose_73)[name = tensor<string, []>("op_29597_cast")];
            tensor<int32, [4]> var_29599_begin_0 = const()[name = tensor<string, []>("op_29599_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_29599_end_0 = const()[name = tensor<string, []>("op_29599_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_29599_end_mask_0 = const()[name = tensor<string, []>("op_29599_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29599_cast = slice_by_index(begin = var_29599_begin_0, end = var_29599_end_0, end_mask = var_29599_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29599_cast")];
            tensor<int32, [4]> var_29603_begin_0 = const()[name = tensor<string, []>("op_29603_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_29603_end_0 = const()[name = tensor<string, []>("op_29603_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_29603_end_mask_0 = const()[name = tensor<string, []>("op_29603_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29603_cast = slice_by_index(begin = var_29603_begin_0, end = var_29603_end_0, end_mask = var_29603_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29603_cast")];
            tensor<int32, [4]> var_29607_begin_0 = const()[name = tensor<string, []>("op_29607_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_29607_end_0 = const()[name = tensor<string, []>("op_29607_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_29607_end_mask_0 = const()[name = tensor<string, []>("op_29607_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29607_cast = slice_by_index(begin = var_29607_begin_0, end = var_29607_end_0, end_mask = var_29607_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29607_cast")];
            tensor<int32, [4]> var_29611_begin_0 = const()[name = tensor<string, []>("op_29611_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_29611_end_0 = const()[name = tensor<string, []>("op_29611_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_29611_end_mask_0 = const()[name = tensor<string, []>("op_29611_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29611_cast = slice_by_index(begin = var_29611_begin_0, end = var_29611_end_0, end_mask = var_29611_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29611_cast")];
            tensor<int32, [4]> var_29615_begin_0 = const()[name = tensor<string, []>("op_29615_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_29615_end_0 = const()[name = tensor<string, []>("op_29615_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_29615_end_mask_0 = const()[name = tensor<string, []>("op_29615_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29615_cast = slice_by_index(begin = var_29615_begin_0, end = var_29615_end_0, end_mask = var_29615_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29615_cast")];
            tensor<int32, [4]> var_29619_begin_0 = const()[name = tensor<string, []>("op_29619_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_29619_end_0 = const()[name = tensor<string, []>("op_29619_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_29619_end_mask_0 = const()[name = tensor<string, []>("op_29619_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29619_cast = slice_by_index(begin = var_29619_begin_0, end = var_29619_end_0, end_mask = var_29619_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29619_cast")];
            tensor<int32, [4]> var_29623_begin_0 = const()[name = tensor<string, []>("op_29623_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_29623_end_0 = const()[name = tensor<string, []>("op_29623_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_29623_end_mask_0 = const()[name = tensor<string, []>("op_29623_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29623_cast = slice_by_index(begin = var_29623_begin_0, end = var_29623_end_0, end_mask = var_29623_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29623_cast")];
            tensor<int32, [4]> var_29627_begin_0 = const()[name = tensor<string, []>("op_29627_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_29627_end_0 = const()[name = tensor<string, []>("op_29627_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_29627_end_mask_0 = const()[name = tensor<string, []>("op_29627_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29627_cast = slice_by_index(begin = var_29627_begin_0, end = var_29627_end_0, end_mask = var_29627_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29627_cast")];
            tensor<int32, [4]> var_29631_begin_0 = const()[name = tensor<string, []>("op_29631_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_29631_end_0 = const()[name = tensor<string, []>("op_29631_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_29631_end_mask_0 = const()[name = tensor<string, []>("op_29631_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29631_cast = slice_by_index(begin = var_29631_begin_0, end = var_29631_end_0, end_mask = var_29631_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29631_cast")];
            tensor<int32, [4]> var_29635_begin_0 = const()[name = tensor<string, []>("op_29635_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_29635_end_0 = const()[name = tensor<string, []>("op_29635_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_29635_end_mask_0 = const()[name = tensor<string, []>("op_29635_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29635_cast = slice_by_index(begin = var_29635_begin_0, end = var_29635_end_0, end_mask = var_29635_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29635_cast")];
            tensor<int32, [4]> var_29639_begin_0 = const()[name = tensor<string, []>("op_29639_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_29639_end_0 = const()[name = tensor<string, []>("op_29639_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_29639_end_mask_0 = const()[name = tensor<string, []>("op_29639_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29639_cast = slice_by_index(begin = var_29639_begin_0, end = var_29639_end_0, end_mask = var_29639_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29639_cast")];
            tensor<int32, [4]> var_29643_begin_0 = const()[name = tensor<string, []>("op_29643_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_29643_end_0 = const()[name = tensor<string, []>("op_29643_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_29643_end_mask_0 = const()[name = tensor<string, []>("op_29643_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29643_cast = slice_by_index(begin = var_29643_begin_0, end = var_29643_end_0, end_mask = var_29643_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29643_cast")];
            tensor<int32, [4]> var_29647_begin_0 = const()[name = tensor<string, []>("op_29647_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_29647_end_0 = const()[name = tensor<string, []>("op_29647_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_29647_end_mask_0 = const()[name = tensor<string, []>("op_29647_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29647_cast = slice_by_index(begin = var_29647_begin_0, end = var_29647_end_0, end_mask = var_29647_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29647_cast")];
            tensor<int32, [4]> var_29651_begin_0 = const()[name = tensor<string, []>("op_29651_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_29651_end_0 = const()[name = tensor<string, []>("op_29651_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_29651_end_mask_0 = const()[name = tensor<string, []>("op_29651_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29651_cast = slice_by_index(begin = var_29651_begin_0, end = var_29651_end_0, end_mask = var_29651_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29651_cast")];
            tensor<int32, [4]> var_29655_begin_0 = const()[name = tensor<string, []>("op_29655_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_29655_end_0 = const()[name = tensor<string, []>("op_29655_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_29655_end_mask_0 = const()[name = tensor<string, []>("op_29655_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29655_cast = slice_by_index(begin = var_29655_begin_0, end = var_29655_end_0, end_mask = var_29655_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29655_cast")];
            tensor<int32, [4]> var_29659_begin_0 = const()[name = tensor<string, []>("op_29659_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_29659_end_0 = const()[name = tensor<string, []>("op_29659_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_29659_end_mask_0 = const()[name = tensor<string, []>("op_29659_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29659_cast = slice_by_index(begin = var_29659_begin_0, end = var_29659_end_0, end_mask = var_29659_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29659_cast")];
            tensor<int32, [4]> var_29663_begin_0 = const()[name = tensor<string, []>("op_29663_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_29663_end_0 = const()[name = tensor<string, []>("op_29663_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_29663_end_mask_0 = const()[name = tensor<string, []>("op_29663_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29663_cast = slice_by_index(begin = var_29663_begin_0, end = var_29663_end_0, end_mask = var_29663_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29663_cast")];
            tensor<int32, [4]> var_29667_begin_0 = const()[name = tensor<string, []>("op_29667_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_29667_end_0 = const()[name = tensor<string, []>("op_29667_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_29667_end_mask_0 = const()[name = tensor<string, []>("op_29667_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29667_cast = slice_by_index(begin = var_29667_begin_0, end = var_29667_end_0, end_mask = var_29667_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29667_cast")];
            tensor<int32, [4]> var_29671_begin_0 = const()[name = tensor<string, []>("op_29671_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_29671_end_0 = const()[name = tensor<string, []>("op_29671_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_29671_end_mask_0 = const()[name = tensor<string, []>("op_29671_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29671_cast = slice_by_index(begin = var_29671_begin_0, end = var_29671_end_0, end_mask = var_29671_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29671_cast")];
            tensor<int32, [4]> var_29675_begin_0 = const()[name = tensor<string, []>("op_29675_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_29675_end_0 = const()[name = tensor<string, []>("op_29675_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_29675_end_mask_0 = const()[name = tensor<string, []>("op_29675_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29675_cast = slice_by_index(begin = var_29675_begin_0, end = var_29675_end_0, end_mask = var_29675_end_mask_0, x = v_133_cast)[name = tensor<string, []>("op_29675_cast")];
            tensor<string, []> var_29679_equation_0 = const()[name = tensor<string, []>("op_29679_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29679_cast = einsum(equation = var_29679_equation_0, values = (var_29521_cast, var_29438_cast))[name = tensor<string, []>("op_29679_cast")];
            tensor<fp16, []> var_29680_to_fp16 = const()[name = tensor<string, []>("op_29680_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2481_cast = mul(x = var_29679_cast, y = var_29680_to_fp16)[name = tensor<string, []>("aw_2481_cast")];
            tensor<string, []> var_29683_equation_0 = const()[name = tensor<string, []>("op_29683_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29683_cast = einsum(equation = var_29683_equation_0, values = (var_29525_cast, var_29442_cast))[name = tensor<string, []>("op_29683_cast")];
            tensor<fp16, []> var_29684_to_fp16 = const()[name = tensor<string, []>("op_29684_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2483_cast = mul(x = var_29683_cast, y = var_29684_to_fp16)[name = tensor<string, []>("aw_2483_cast")];
            tensor<string, []> var_29687_equation_0 = const()[name = tensor<string, []>("op_29687_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29687_cast = einsum(equation = var_29687_equation_0, values = (var_29529_cast, var_29446_cast))[name = tensor<string, []>("op_29687_cast")];
            tensor<fp16, []> var_29688_to_fp16 = const()[name = tensor<string, []>("op_29688_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2485_cast = mul(x = var_29687_cast, y = var_29688_to_fp16)[name = tensor<string, []>("aw_2485_cast")];
            tensor<string, []> var_29691_equation_0 = const()[name = tensor<string, []>("op_29691_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29691_cast = einsum(equation = var_29691_equation_0, values = (var_29533_cast, var_29450_cast))[name = tensor<string, []>("op_29691_cast")];
            tensor<fp16, []> var_29692_to_fp16 = const()[name = tensor<string, []>("op_29692_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2487_cast = mul(x = var_29691_cast, y = var_29692_to_fp16)[name = tensor<string, []>("aw_2487_cast")];
            tensor<string, []> var_29695_equation_0 = const()[name = tensor<string, []>("op_29695_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29695_cast = einsum(equation = var_29695_equation_0, values = (var_29537_cast, var_29454_cast))[name = tensor<string, []>("op_29695_cast")];
            tensor<fp16, []> var_29696_to_fp16 = const()[name = tensor<string, []>("op_29696_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2489_cast = mul(x = var_29695_cast, y = var_29696_to_fp16)[name = tensor<string, []>("aw_2489_cast")];
            tensor<string, []> var_29699_equation_0 = const()[name = tensor<string, []>("op_29699_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29699_cast = einsum(equation = var_29699_equation_0, values = (var_29541_cast, var_29458_cast))[name = tensor<string, []>("op_29699_cast")];
            tensor<fp16, []> var_29700_to_fp16 = const()[name = tensor<string, []>("op_29700_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2491_cast = mul(x = var_29699_cast, y = var_29700_to_fp16)[name = tensor<string, []>("aw_2491_cast")];
            tensor<string, []> var_29703_equation_0 = const()[name = tensor<string, []>("op_29703_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29703_cast = einsum(equation = var_29703_equation_0, values = (var_29545_cast, var_29462_cast))[name = tensor<string, []>("op_29703_cast")];
            tensor<fp16, []> var_29704_to_fp16 = const()[name = tensor<string, []>("op_29704_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2493_cast = mul(x = var_29703_cast, y = var_29704_to_fp16)[name = tensor<string, []>("aw_2493_cast")];
            tensor<string, []> var_29707_equation_0 = const()[name = tensor<string, []>("op_29707_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29707_cast = einsum(equation = var_29707_equation_0, values = (var_29549_cast, var_29466_cast))[name = tensor<string, []>("op_29707_cast")];
            tensor<fp16, []> var_29708_to_fp16 = const()[name = tensor<string, []>("op_29708_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2495_cast = mul(x = var_29707_cast, y = var_29708_to_fp16)[name = tensor<string, []>("aw_2495_cast")];
            tensor<string, []> var_29711_equation_0 = const()[name = tensor<string, []>("op_29711_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29711_cast = einsum(equation = var_29711_equation_0, values = (var_29553_cast, var_29470_cast))[name = tensor<string, []>("op_29711_cast")];
            tensor<fp16, []> var_29712_to_fp16 = const()[name = tensor<string, []>("op_29712_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2497_cast = mul(x = var_29711_cast, y = var_29712_to_fp16)[name = tensor<string, []>("aw_2497_cast")];
            tensor<string, []> var_29715_equation_0 = const()[name = tensor<string, []>("op_29715_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29715_cast = einsum(equation = var_29715_equation_0, values = (var_29557_cast, var_29474_cast))[name = tensor<string, []>("op_29715_cast")];
            tensor<fp16, []> var_29716_to_fp16 = const()[name = tensor<string, []>("op_29716_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2499_cast = mul(x = var_29715_cast, y = var_29716_to_fp16)[name = tensor<string, []>("aw_2499_cast")];
            tensor<string, []> var_29719_equation_0 = const()[name = tensor<string, []>("op_29719_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29719_cast = einsum(equation = var_29719_equation_0, values = (var_29561_cast, var_29478_cast))[name = tensor<string, []>("op_29719_cast")];
            tensor<fp16, []> var_29720_to_fp16 = const()[name = tensor<string, []>("op_29720_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2501_cast = mul(x = var_29719_cast, y = var_29720_to_fp16)[name = tensor<string, []>("aw_2501_cast")];
            tensor<string, []> var_29723_equation_0 = const()[name = tensor<string, []>("op_29723_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29723_cast = einsum(equation = var_29723_equation_0, values = (var_29565_cast, var_29482_cast))[name = tensor<string, []>("op_29723_cast")];
            tensor<fp16, []> var_29724_to_fp16 = const()[name = tensor<string, []>("op_29724_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2503_cast = mul(x = var_29723_cast, y = var_29724_to_fp16)[name = tensor<string, []>("aw_2503_cast")];
            tensor<string, []> var_29727_equation_0 = const()[name = tensor<string, []>("op_29727_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29727_cast = einsum(equation = var_29727_equation_0, values = (var_29569_cast, var_29486_cast))[name = tensor<string, []>("op_29727_cast")];
            tensor<fp16, []> var_29728_to_fp16 = const()[name = tensor<string, []>("op_29728_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2505_cast = mul(x = var_29727_cast, y = var_29728_to_fp16)[name = tensor<string, []>("aw_2505_cast")];
            tensor<string, []> var_29731_equation_0 = const()[name = tensor<string, []>("op_29731_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29731_cast = einsum(equation = var_29731_equation_0, values = (var_29573_cast, var_29490_cast))[name = tensor<string, []>("op_29731_cast")];
            tensor<fp16, []> var_29732_to_fp16 = const()[name = tensor<string, []>("op_29732_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2507_cast = mul(x = var_29731_cast, y = var_29732_to_fp16)[name = tensor<string, []>("aw_2507_cast")];
            tensor<string, []> var_29735_equation_0 = const()[name = tensor<string, []>("op_29735_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29735_cast = einsum(equation = var_29735_equation_0, values = (var_29577_cast, var_29494_cast))[name = tensor<string, []>("op_29735_cast")];
            tensor<fp16, []> var_29736_to_fp16 = const()[name = tensor<string, []>("op_29736_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2509_cast = mul(x = var_29735_cast, y = var_29736_to_fp16)[name = tensor<string, []>("aw_2509_cast")];
            tensor<string, []> var_29739_equation_0 = const()[name = tensor<string, []>("op_29739_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29739_cast = einsum(equation = var_29739_equation_0, values = (var_29581_cast, var_29498_cast))[name = tensor<string, []>("op_29739_cast")];
            tensor<fp16, []> var_29740_to_fp16 = const()[name = tensor<string, []>("op_29740_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2511_cast = mul(x = var_29739_cast, y = var_29740_to_fp16)[name = tensor<string, []>("aw_2511_cast")];
            tensor<string, []> var_29743_equation_0 = const()[name = tensor<string, []>("op_29743_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29743_cast = einsum(equation = var_29743_equation_0, values = (var_29585_cast, var_29502_cast))[name = tensor<string, []>("op_29743_cast")];
            tensor<fp16, []> var_29744_to_fp16 = const()[name = tensor<string, []>("op_29744_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2513_cast = mul(x = var_29743_cast, y = var_29744_to_fp16)[name = tensor<string, []>("aw_2513_cast")];
            tensor<string, []> var_29747_equation_0 = const()[name = tensor<string, []>("op_29747_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29747_cast = einsum(equation = var_29747_equation_0, values = (var_29589_cast, var_29506_cast))[name = tensor<string, []>("op_29747_cast")];
            tensor<fp16, []> var_29748_to_fp16 = const()[name = tensor<string, []>("op_29748_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2515_cast = mul(x = var_29747_cast, y = var_29748_to_fp16)[name = tensor<string, []>("aw_2515_cast")];
            tensor<string, []> var_29751_equation_0 = const()[name = tensor<string, []>("op_29751_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29751_cast = einsum(equation = var_29751_equation_0, values = (var_29593_cast, var_29510_cast))[name = tensor<string, []>("op_29751_cast")];
            tensor<fp16, []> var_29752_to_fp16 = const()[name = tensor<string, []>("op_29752_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2517_cast = mul(x = var_29751_cast, y = var_29752_to_fp16)[name = tensor<string, []>("aw_2517_cast")];
            tensor<string, []> var_29755_equation_0 = const()[name = tensor<string, []>("op_29755_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29755_cast = einsum(equation = var_29755_equation_0, values = (var_29597_cast, var_29514_cast))[name = tensor<string, []>("op_29755_cast")];
            tensor<fp16, []> var_29756_to_fp16 = const()[name = tensor<string, []>("op_29756_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2519_cast = mul(x = var_29755_cast, y = var_29756_to_fp16)[name = tensor<string, []>("aw_2519_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29758_cast = softmax(axis = var_21105, x = aw_2481_cast)[name = tensor<string, []>("op_29758_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29759_cast = softmax(axis = var_21105, x = aw_2483_cast)[name = tensor<string, []>("op_29759_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29760_cast = softmax(axis = var_21105, x = aw_2485_cast)[name = tensor<string, []>("op_29760_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29761_cast = softmax(axis = var_21105, x = aw_2487_cast)[name = tensor<string, []>("op_29761_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29762_cast = softmax(axis = var_21105, x = aw_2489_cast)[name = tensor<string, []>("op_29762_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29763_cast = softmax(axis = var_21105, x = aw_2491_cast)[name = tensor<string, []>("op_29763_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29764_cast = softmax(axis = var_21105, x = aw_2493_cast)[name = tensor<string, []>("op_29764_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29765_cast = softmax(axis = var_21105, x = aw_2495_cast)[name = tensor<string, []>("op_29765_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29766_cast = softmax(axis = var_21105, x = aw_2497_cast)[name = tensor<string, []>("op_29766_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29767_cast = softmax(axis = var_21105, x = aw_2499_cast)[name = tensor<string, []>("op_29767_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29768_cast = softmax(axis = var_21105, x = aw_2501_cast)[name = tensor<string, []>("op_29768_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29769_cast = softmax(axis = var_21105, x = aw_2503_cast)[name = tensor<string, []>("op_29769_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29770_cast = softmax(axis = var_21105, x = aw_2505_cast)[name = tensor<string, []>("op_29770_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29771_cast = softmax(axis = var_21105, x = aw_2507_cast)[name = tensor<string, []>("op_29771_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29772_cast = softmax(axis = var_21105, x = aw_2509_cast)[name = tensor<string, []>("op_29772_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29773_cast = softmax(axis = var_21105, x = aw_2511_cast)[name = tensor<string, []>("op_29773_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29774_cast = softmax(axis = var_21105, x = aw_2513_cast)[name = tensor<string, []>("op_29774_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29775_cast = softmax(axis = var_21105, x = aw_2515_cast)[name = tensor<string, []>("op_29775_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29776_cast = softmax(axis = var_21105, x = aw_2517_cast)[name = tensor<string, []>("op_29776_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_29777_cast = softmax(axis = var_21105, x = aw_2519_cast)[name = tensor<string, []>("op_29777_cast")];
            tensor<string, []> var_29779_equation_0 = const()[name = tensor<string, []>("op_29779_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29779_cast = einsum(equation = var_29779_equation_0, values = (var_29599_cast, var_29758_cast))[name = tensor<string, []>("op_29779_cast")];
            tensor<string, []> var_29781_equation_0 = const()[name = tensor<string, []>("op_29781_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29781_cast = einsum(equation = var_29781_equation_0, values = (var_29603_cast, var_29759_cast))[name = tensor<string, []>("op_29781_cast")];
            tensor<string, []> var_29783_equation_0 = const()[name = tensor<string, []>("op_29783_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29783_cast = einsum(equation = var_29783_equation_0, values = (var_29607_cast, var_29760_cast))[name = tensor<string, []>("op_29783_cast")];
            tensor<string, []> var_29785_equation_0 = const()[name = tensor<string, []>("op_29785_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29785_cast = einsum(equation = var_29785_equation_0, values = (var_29611_cast, var_29761_cast))[name = tensor<string, []>("op_29785_cast")];
            tensor<string, []> var_29787_equation_0 = const()[name = tensor<string, []>("op_29787_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29787_cast = einsum(equation = var_29787_equation_0, values = (var_29615_cast, var_29762_cast))[name = tensor<string, []>("op_29787_cast")];
            tensor<string, []> var_29789_equation_0 = const()[name = tensor<string, []>("op_29789_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29789_cast = einsum(equation = var_29789_equation_0, values = (var_29619_cast, var_29763_cast))[name = tensor<string, []>("op_29789_cast")];
            tensor<string, []> var_29791_equation_0 = const()[name = tensor<string, []>("op_29791_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29791_cast = einsum(equation = var_29791_equation_0, values = (var_29623_cast, var_29764_cast))[name = tensor<string, []>("op_29791_cast")];
            tensor<string, []> var_29793_equation_0 = const()[name = tensor<string, []>("op_29793_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29793_cast = einsum(equation = var_29793_equation_0, values = (var_29627_cast, var_29765_cast))[name = tensor<string, []>("op_29793_cast")];
            tensor<string, []> var_29795_equation_0 = const()[name = tensor<string, []>("op_29795_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29795_cast = einsum(equation = var_29795_equation_0, values = (var_29631_cast, var_29766_cast))[name = tensor<string, []>("op_29795_cast")];
            tensor<string, []> var_29797_equation_0 = const()[name = tensor<string, []>("op_29797_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29797_cast = einsum(equation = var_29797_equation_0, values = (var_29635_cast, var_29767_cast))[name = tensor<string, []>("op_29797_cast")];
            tensor<string, []> var_29799_equation_0 = const()[name = tensor<string, []>("op_29799_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29799_cast = einsum(equation = var_29799_equation_0, values = (var_29639_cast, var_29768_cast))[name = tensor<string, []>("op_29799_cast")];
            tensor<string, []> var_29801_equation_0 = const()[name = tensor<string, []>("op_29801_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29801_cast = einsum(equation = var_29801_equation_0, values = (var_29643_cast, var_29769_cast))[name = tensor<string, []>("op_29801_cast")];
            tensor<string, []> var_29803_equation_0 = const()[name = tensor<string, []>("op_29803_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29803_cast = einsum(equation = var_29803_equation_0, values = (var_29647_cast, var_29770_cast))[name = tensor<string, []>("op_29803_cast")];
            tensor<string, []> var_29805_equation_0 = const()[name = tensor<string, []>("op_29805_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29805_cast = einsum(equation = var_29805_equation_0, values = (var_29651_cast, var_29771_cast))[name = tensor<string, []>("op_29805_cast")];
            tensor<string, []> var_29807_equation_0 = const()[name = tensor<string, []>("op_29807_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29807_cast = einsum(equation = var_29807_equation_0, values = (var_29655_cast, var_29772_cast))[name = tensor<string, []>("op_29807_cast")];
            tensor<string, []> var_29809_equation_0 = const()[name = tensor<string, []>("op_29809_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29809_cast = einsum(equation = var_29809_equation_0, values = (var_29659_cast, var_29773_cast))[name = tensor<string, []>("op_29809_cast")];
            tensor<string, []> var_29811_equation_0 = const()[name = tensor<string, []>("op_29811_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29811_cast = einsum(equation = var_29811_equation_0, values = (var_29663_cast, var_29774_cast))[name = tensor<string, []>("op_29811_cast")];
            tensor<string, []> var_29813_equation_0 = const()[name = tensor<string, []>("op_29813_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29813_cast = einsum(equation = var_29813_equation_0, values = (var_29667_cast, var_29775_cast))[name = tensor<string, []>("op_29813_cast")];
            tensor<string, []> var_29815_equation_0 = const()[name = tensor<string, []>("op_29815_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29815_cast = einsum(equation = var_29815_equation_0, values = (var_29671_cast, var_29776_cast))[name = tensor<string, []>("op_29815_cast")];
            tensor<string, []> var_29817_equation_0 = const()[name = tensor<string, []>("op_29817_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_29817_cast = einsum(equation = var_29817_equation_0, values = (var_29675_cast, var_29777_cast))[name = tensor<string, []>("op_29817_cast")];
            tensor<bool, []> input_401_interleave_0 = const()[name = tensor<string, []>("input_401_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_401_cast = concat(axis = var_21105, interleave = input_401_interleave_0, values = (var_29779_cast, var_29781_cast, var_29783_cast, var_29785_cast, var_29787_cast, var_29789_cast, var_29791_cast, var_29793_cast, var_29795_cast, var_29797_cast, var_29799_cast, var_29801_cast, var_29803_cast, var_29805_cast, var_29807_cast, var_29809_cast, var_29811_cast, var_29813_cast, var_29815_cast, var_29817_cast))[name = tensor<string, []>("input_401_cast")];
            tensor<int32, [2]> var_29823 = const()[name = tensor<string, []>("op_29823"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_29825 = const()[name = tensor<string, []>("op_29825"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_29827_pad_type_0 = const()[name = tensor<string, []>("op_29827_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_29827_pad_0 = const()[name = tensor<string, []>("op_29827_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2361561536)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2364838400)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_29827_cast = conv(bias = mid_block_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16, dilations = var_29825, groups = var_21105, pad = var_29827_pad_0, pad_type = var_29827_pad_type_0, strides = var_29823, weight = mid_block_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16, x = input_401_cast)[name = tensor<string, []>("op_29827_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_201_cast = add(x = var_29827_cast, y = inputs_199_cast)[name = tensor<string, []>("inputs_201_cast")];
            tensor<int32, [1]> var_29831 = const()[name = tensor<string, []>("op_29831"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_201_cast = reduce_mean(axes = var_29831, keep_dims = var_21100, x = inputs_201_cast)[name = tensor<string, []>("channels_mean_201_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_201_cast = sub(x = inputs_201_cast, y = channels_mean_201_cast)[name = tensor<string, []>("zero_mean_201_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_201_cast = mul(x = zero_mean_201_cast, y = zero_mean_201_cast)[name = tensor<string, []>("zero_mean_sq_201_cast")];
            tensor<int32, [1]> var_29835 = const()[name = tensor<string, []>("op_29835"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_29836_cast = reduce_mean(axes = var_29835, keep_dims = var_21100, x = zero_mean_sq_201_cast)[name = tensor<string, []>("op_29836_cast")];
            tensor<fp16, []> var_29837_to_fp16 = const()[name = tensor<string, []>("op_29837_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_29838_cast = add(x = var_29836_cast, y = var_29837_to_fp16)[name = tensor<string, []>("op_29838_cast")];
            tensor<fp16, []> denom_201_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_201_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_201_cast = rsqrt(epsilon = denom_201_epsilon_0_to_fp16, x = var_29838_cast)[name = tensor<string, []>("denom_201_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_201_cast = mul(x = zero_mean_201_cast, y = denom_201_cast)[name = tensor<string, []>("out_201_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_29842_to_fp16 = const()[name = tensor<string, []>("op_29842_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2364841024)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_29843_cast = add(x = out_201_cast, y = var_29842_to_fp16)[name = tensor<string, []>("op_29843_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_29845_to_fp16 = const()[name = tensor<string, []>("op_29845_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2364843648)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_265_cast = mul(x = var_29843_cast, y = var_29845_to_fp16)[name = tensor<string, []>("hidden_states_265_cast")];
            tensor<int32, [2]> var_29852 = const()[name = tensor<string, []>("op_29852"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_29854 = const()[name = tensor<string, []>("op_29854"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_135_pad_type_0 = const()[name = tensor<string, []>("q_135_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_135_pad_0 = const()[name = tensor<string, []>("q_135_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2364846272)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_135_cast = conv(dilations = var_29854, groups = var_21105, pad = q_135_pad_0, pad_type = q_135_pad_type_0, strides = var_29852, weight = mid_block_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16, x = hidden_states_265_cast)[name = tensor<string, []>("q_135_cast")];
            tensor<int32, [2]> var_29858 = const()[name = tensor<string, []>("op_29858"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_29860 = const()[name = tensor<string, []>("op_29860"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_269_pad_type_0 = const()[name = tensor<string, []>("k_269_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_269_pad_0 = const()[name = tensor<string, []>("k_269_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2368123136)))];
            tensor<fp16, [2, 1280, 1, 77]> k_269_cast = conv(dilations = var_29860, groups = var_21105, pad = k_269_pad_0, pad_type = k_269_pad_type_0, strides = var_29858, weight = mid_block_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_269_cast")];
            tensor<int32, [2]> var_29864 = const()[name = tensor<string, []>("op_29864"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_29866 = const()[name = tensor<string, []>("op_29866"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_135_pad_type_0 = const()[name = tensor<string, []>("v_135_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_135_pad_0 = const()[name = tensor<string, []>("v_135_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> mid_block_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2373366080)))];
            tensor<fp16, [2, 1280, 1, 77]> v_135_cast = conv(dilations = var_29866, groups = var_21105, pad = v_135_pad_0, pad_type = v_135_pad_type_0, strides = var_29864, weight = mid_block_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_135_cast")];
            tensor<int32, [4]> var_29870_begin_0 = const()[name = tensor<string, []>("op_29870_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_29870_end_0 = const()[name = tensor<string, []>("op_29870_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_29870_end_mask_0 = const()[name = tensor<string, []>("op_29870_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29870_cast = slice_by_index(begin = var_29870_begin_0, end = var_29870_end_0, end_mask = var_29870_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29870_cast")];
            tensor<int32, [4]> var_29874_begin_0 = const()[name = tensor<string, []>("op_29874_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_29874_end_0 = const()[name = tensor<string, []>("op_29874_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_29874_end_mask_0 = const()[name = tensor<string, []>("op_29874_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29874_cast = slice_by_index(begin = var_29874_begin_0, end = var_29874_end_0, end_mask = var_29874_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29874_cast")];
            tensor<int32, [4]> var_29878_begin_0 = const()[name = tensor<string, []>("op_29878_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_29878_end_0 = const()[name = tensor<string, []>("op_29878_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_29878_end_mask_0 = const()[name = tensor<string, []>("op_29878_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29878_cast = slice_by_index(begin = var_29878_begin_0, end = var_29878_end_0, end_mask = var_29878_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29878_cast")];
            tensor<int32, [4]> var_29882_begin_0 = const()[name = tensor<string, []>("op_29882_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_29882_end_0 = const()[name = tensor<string, []>("op_29882_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_29882_end_mask_0 = const()[name = tensor<string, []>("op_29882_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29882_cast = slice_by_index(begin = var_29882_begin_0, end = var_29882_end_0, end_mask = var_29882_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29882_cast")];
            tensor<int32, [4]> var_29886_begin_0 = const()[name = tensor<string, []>("op_29886_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_29886_end_0 = const()[name = tensor<string, []>("op_29886_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_29886_end_mask_0 = const()[name = tensor<string, []>("op_29886_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29886_cast = slice_by_index(begin = var_29886_begin_0, end = var_29886_end_0, end_mask = var_29886_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29886_cast")];
            tensor<int32, [4]> var_29890_begin_0 = const()[name = tensor<string, []>("op_29890_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_29890_end_0 = const()[name = tensor<string, []>("op_29890_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_29890_end_mask_0 = const()[name = tensor<string, []>("op_29890_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29890_cast = slice_by_index(begin = var_29890_begin_0, end = var_29890_end_0, end_mask = var_29890_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29890_cast")];
            tensor<int32, [4]> var_29894_begin_0 = const()[name = tensor<string, []>("op_29894_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_29894_end_0 = const()[name = tensor<string, []>("op_29894_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_29894_end_mask_0 = const()[name = tensor<string, []>("op_29894_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29894_cast = slice_by_index(begin = var_29894_begin_0, end = var_29894_end_0, end_mask = var_29894_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29894_cast")];
            tensor<int32, [4]> var_29898_begin_0 = const()[name = tensor<string, []>("op_29898_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_29898_end_0 = const()[name = tensor<string, []>("op_29898_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_29898_end_mask_0 = const()[name = tensor<string, []>("op_29898_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29898_cast = slice_by_index(begin = var_29898_begin_0, end = var_29898_end_0, end_mask = var_29898_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29898_cast")];
            tensor<int32, [4]> var_29902_begin_0 = const()[name = tensor<string, []>("op_29902_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_29902_end_0 = const()[name = tensor<string, []>("op_29902_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_29902_end_mask_0 = const()[name = tensor<string, []>("op_29902_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29902_cast = slice_by_index(begin = var_29902_begin_0, end = var_29902_end_0, end_mask = var_29902_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29902_cast")];
            tensor<int32, [4]> var_29906_begin_0 = const()[name = tensor<string, []>("op_29906_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_29906_end_0 = const()[name = tensor<string, []>("op_29906_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_29906_end_mask_0 = const()[name = tensor<string, []>("op_29906_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29906_cast = slice_by_index(begin = var_29906_begin_0, end = var_29906_end_0, end_mask = var_29906_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29906_cast")];
            tensor<int32, [4]> var_29910_begin_0 = const()[name = tensor<string, []>("op_29910_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_29910_end_0 = const()[name = tensor<string, []>("op_29910_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_29910_end_mask_0 = const()[name = tensor<string, []>("op_29910_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29910_cast = slice_by_index(begin = var_29910_begin_0, end = var_29910_end_0, end_mask = var_29910_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29910_cast")];
            tensor<int32, [4]> var_29914_begin_0 = const()[name = tensor<string, []>("op_29914_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_29914_end_0 = const()[name = tensor<string, []>("op_29914_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_29914_end_mask_0 = const()[name = tensor<string, []>("op_29914_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29914_cast = slice_by_index(begin = var_29914_begin_0, end = var_29914_end_0, end_mask = var_29914_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29914_cast")];
            tensor<int32, [4]> var_29918_begin_0 = const()[name = tensor<string, []>("op_29918_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_29918_end_0 = const()[name = tensor<string, []>("op_29918_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_29918_end_mask_0 = const()[name = tensor<string, []>("op_29918_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29918_cast = slice_by_index(begin = var_29918_begin_0, end = var_29918_end_0, end_mask = var_29918_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29918_cast")];
            tensor<int32, [4]> var_29922_begin_0 = const()[name = tensor<string, []>("op_29922_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_29922_end_0 = const()[name = tensor<string, []>("op_29922_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_29922_end_mask_0 = const()[name = tensor<string, []>("op_29922_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29922_cast = slice_by_index(begin = var_29922_begin_0, end = var_29922_end_0, end_mask = var_29922_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29922_cast")];
            tensor<int32, [4]> var_29926_begin_0 = const()[name = tensor<string, []>("op_29926_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_29926_end_0 = const()[name = tensor<string, []>("op_29926_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_29926_end_mask_0 = const()[name = tensor<string, []>("op_29926_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29926_cast = slice_by_index(begin = var_29926_begin_0, end = var_29926_end_0, end_mask = var_29926_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29926_cast")];
            tensor<int32, [4]> var_29930_begin_0 = const()[name = tensor<string, []>("op_29930_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_29930_end_0 = const()[name = tensor<string, []>("op_29930_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_29930_end_mask_0 = const()[name = tensor<string, []>("op_29930_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29930_cast = slice_by_index(begin = var_29930_begin_0, end = var_29930_end_0, end_mask = var_29930_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29930_cast")];
            tensor<int32, [4]> var_29934_begin_0 = const()[name = tensor<string, []>("op_29934_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_29934_end_0 = const()[name = tensor<string, []>("op_29934_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_29934_end_mask_0 = const()[name = tensor<string, []>("op_29934_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29934_cast = slice_by_index(begin = var_29934_begin_0, end = var_29934_end_0, end_mask = var_29934_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29934_cast")];
            tensor<int32, [4]> var_29938_begin_0 = const()[name = tensor<string, []>("op_29938_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_29938_end_0 = const()[name = tensor<string, []>("op_29938_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_29938_end_mask_0 = const()[name = tensor<string, []>("op_29938_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29938_cast = slice_by_index(begin = var_29938_begin_0, end = var_29938_end_0, end_mask = var_29938_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29938_cast")];
            tensor<int32, [4]> var_29942_begin_0 = const()[name = tensor<string, []>("op_29942_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_29942_end_0 = const()[name = tensor<string, []>("op_29942_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_29942_end_mask_0 = const()[name = tensor<string, []>("op_29942_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29942_cast = slice_by_index(begin = var_29942_begin_0, end = var_29942_end_0, end_mask = var_29942_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29942_cast")];
            tensor<int32, [4]> var_29946_begin_0 = const()[name = tensor<string, []>("op_29946_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_29946_end_0 = const()[name = tensor<string, []>("op_29946_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_29946_end_mask_0 = const()[name = tensor<string, []>("op_29946_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_29946_cast = slice_by_index(begin = var_29946_begin_0, end = var_29946_end_0, end_mask = var_29946_end_mask_0, x = q_135_cast)[name = tensor<string, []>("op_29946_cast")];
            tensor<int32, [4]> k_271_perm_0 = const()[name = tensor<string, []>("k_271_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_29953_begin_0 = const()[name = tensor<string, []>("op_29953_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_29953_end_0 = const()[name = tensor<string, []>("op_29953_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_29953_end_mask_0 = const()[name = tensor<string, []>("op_29953_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_72 = transpose(perm = k_271_perm_0, x = k_269_cast)[name = tensor<string, []>("transpose_72")];
            tensor<fp16, [2, 77, 1, 64]> var_29953_cast = slice_by_index(begin = var_29953_begin_0, end = var_29953_end_0, end_mask = var_29953_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_29953_cast")];
            tensor<int32, [4]> var_29957_begin_0 = const()[name = tensor<string, []>("op_29957_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_29957_end_0 = const()[name = tensor<string, []>("op_29957_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_29957_end_mask_0 = const()[name = tensor<string, []>("op_29957_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29957_cast = slice_by_index(begin = var_29957_begin_0, end = var_29957_end_0, end_mask = var_29957_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_29957_cast")];
            tensor<int32, [4]> var_29961_begin_0 = const()[name = tensor<string, []>("op_29961_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_29961_end_0 = const()[name = tensor<string, []>("op_29961_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_29961_end_mask_0 = const()[name = tensor<string, []>("op_29961_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29961_cast = slice_by_index(begin = var_29961_begin_0, end = var_29961_end_0, end_mask = var_29961_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_29961_cast")];
            tensor<int32, [4]> var_29965_begin_0 = const()[name = tensor<string, []>("op_29965_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_29965_end_0 = const()[name = tensor<string, []>("op_29965_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_29965_end_mask_0 = const()[name = tensor<string, []>("op_29965_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29965_cast = slice_by_index(begin = var_29965_begin_0, end = var_29965_end_0, end_mask = var_29965_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_29965_cast")];
            tensor<int32, [4]> var_29969_begin_0 = const()[name = tensor<string, []>("op_29969_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_29969_end_0 = const()[name = tensor<string, []>("op_29969_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_29969_end_mask_0 = const()[name = tensor<string, []>("op_29969_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29969_cast = slice_by_index(begin = var_29969_begin_0, end = var_29969_end_0, end_mask = var_29969_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_29969_cast")];
            tensor<int32, [4]> var_29973_begin_0 = const()[name = tensor<string, []>("op_29973_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_29973_end_0 = const()[name = tensor<string, []>("op_29973_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_29973_end_mask_0 = const()[name = tensor<string, []>("op_29973_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29973_cast = slice_by_index(begin = var_29973_begin_0, end = var_29973_end_0, end_mask = var_29973_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_29973_cast")];
            tensor<int32, [4]> var_29977_begin_0 = const()[name = tensor<string, []>("op_29977_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_29977_end_0 = const()[name = tensor<string, []>("op_29977_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_29977_end_mask_0 = const()[name = tensor<string, []>("op_29977_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29977_cast = slice_by_index(begin = var_29977_begin_0, end = var_29977_end_0, end_mask = var_29977_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_29977_cast")];
            tensor<int32, [4]> var_29981_begin_0 = const()[name = tensor<string, []>("op_29981_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_29981_end_0 = const()[name = tensor<string, []>("op_29981_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_29981_end_mask_0 = const()[name = tensor<string, []>("op_29981_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29981_cast = slice_by_index(begin = var_29981_begin_0, end = var_29981_end_0, end_mask = var_29981_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_29981_cast")];
            tensor<int32, [4]> var_29985_begin_0 = const()[name = tensor<string, []>("op_29985_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_29985_end_0 = const()[name = tensor<string, []>("op_29985_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_29985_end_mask_0 = const()[name = tensor<string, []>("op_29985_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29985_cast = slice_by_index(begin = var_29985_begin_0, end = var_29985_end_0, end_mask = var_29985_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_29985_cast")];
            tensor<int32, [4]> var_29989_begin_0 = const()[name = tensor<string, []>("op_29989_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_29989_end_0 = const()[name = tensor<string, []>("op_29989_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_29989_end_mask_0 = const()[name = tensor<string, []>("op_29989_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29989_cast = slice_by_index(begin = var_29989_begin_0, end = var_29989_end_0, end_mask = var_29989_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_29989_cast")];
            tensor<int32, [4]> var_29993_begin_0 = const()[name = tensor<string, []>("op_29993_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_29993_end_0 = const()[name = tensor<string, []>("op_29993_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_29993_end_mask_0 = const()[name = tensor<string, []>("op_29993_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29993_cast = slice_by_index(begin = var_29993_begin_0, end = var_29993_end_0, end_mask = var_29993_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_29993_cast")];
            tensor<int32, [4]> var_29997_begin_0 = const()[name = tensor<string, []>("op_29997_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_29997_end_0 = const()[name = tensor<string, []>("op_29997_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_29997_end_mask_0 = const()[name = tensor<string, []>("op_29997_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_29997_cast = slice_by_index(begin = var_29997_begin_0, end = var_29997_end_0, end_mask = var_29997_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_29997_cast")];
            tensor<int32, [4]> var_30001_begin_0 = const()[name = tensor<string, []>("op_30001_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_30001_end_0 = const()[name = tensor<string, []>("op_30001_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_30001_end_mask_0 = const()[name = tensor<string, []>("op_30001_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_30001_cast = slice_by_index(begin = var_30001_begin_0, end = var_30001_end_0, end_mask = var_30001_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_30001_cast")];
            tensor<int32, [4]> var_30005_begin_0 = const()[name = tensor<string, []>("op_30005_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_30005_end_0 = const()[name = tensor<string, []>("op_30005_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_30005_end_mask_0 = const()[name = tensor<string, []>("op_30005_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_30005_cast = slice_by_index(begin = var_30005_begin_0, end = var_30005_end_0, end_mask = var_30005_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_30005_cast")];
            tensor<int32, [4]> var_30009_begin_0 = const()[name = tensor<string, []>("op_30009_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_30009_end_0 = const()[name = tensor<string, []>("op_30009_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_30009_end_mask_0 = const()[name = tensor<string, []>("op_30009_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_30009_cast = slice_by_index(begin = var_30009_begin_0, end = var_30009_end_0, end_mask = var_30009_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_30009_cast")];
            tensor<int32, [4]> var_30013_begin_0 = const()[name = tensor<string, []>("op_30013_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_30013_end_0 = const()[name = tensor<string, []>("op_30013_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_30013_end_mask_0 = const()[name = tensor<string, []>("op_30013_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_30013_cast = slice_by_index(begin = var_30013_begin_0, end = var_30013_end_0, end_mask = var_30013_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_30013_cast")];
            tensor<int32, [4]> var_30017_begin_0 = const()[name = tensor<string, []>("op_30017_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_30017_end_0 = const()[name = tensor<string, []>("op_30017_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_30017_end_mask_0 = const()[name = tensor<string, []>("op_30017_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_30017_cast = slice_by_index(begin = var_30017_begin_0, end = var_30017_end_0, end_mask = var_30017_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_30017_cast")];
            tensor<int32, [4]> var_30021_begin_0 = const()[name = tensor<string, []>("op_30021_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_30021_end_0 = const()[name = tensor<string, []>("op_30021_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_30021_end_mask_0 = const()[name = tensor<string, []>("op_30021_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_30021_cast = slice_by_index(begin = var_30021_begin_0, end = var_30021_end_0, end_mask = var_30021_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_30021_cast")];
            tensor<int32, [4]> var_30025_begin_0 = const()[name = tensor<string, []>("op_30025_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_30025_end_0 = const()[name = tensor<string, []>("op_30025_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_30025_end_mask_0 = const()[name = tensor<string, []>("op_30025_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_30025_cast = slice_by_index(begin = var_30025_begin_0, end = var_30025_end_0, end_mask = var_30025_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_30025_cast")];
            tensor<int32, [4]> var_30029_begin_0 = const()[name = tensor<string, []>("op_30029_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_30029_end_0 = const()[name = tensor<string, []>("op_30029_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_30029_end_mask_0 = const()[name = tensor<string, []>("op_30029_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_30029_cast = slice_by_index(begin = var_30029_begin_0, end = var_30029_end_0, end_mask = var_30029_end_mask_0, x = transpose_72)[name = tensor<string, []>("op_30029_cast")];
            tensor<int32, [4]> var_30031_begin_0 = const()[name = tensor<string, []>("op_30031_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_30031_end_0 = const()[name = tensor<string, []>("op_30031_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_30031_end_mask_0 = const()[name = tensor<string, []>("op_30031_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30031_cast = slice_by_index(begin = var_30031_begin_0, end = var_30031_end_0, end_mask = var_30031_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30031_cast")];
            tensor<int32, [4]> var_30035_begin_0 = const()[name = tensor<string, []>("op_30035_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_30035_end_0 = const()[name = tensor<string, []>("op_30035_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_30035_end_mask_0 = const()[name = tensor<string, []>("op_30035_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30035_cast = slice_by_index(begin = var_30035_begin_0, end = var_30035_end_0, end_mask = var_30035_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30035_cast")];
            tensor<int32, [4]> var_30039_begin_0 = const()[name = tensor<string, []>("op_30039_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_30039_end_0 = const()[name = tensor<string, []>("op_30039_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_30039_end_mask_0 = const()[name = tensor<string, []>("op_30039_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30039_cast = slice_by_index(begin = var_30039_begin_0, end = var_30039_end_0, end_mask = var_30039_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30039_cast")];
            tensor<int32, [4]> var_30043_begin_0 = const()[name = tensor<string, []>("op_30043_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_30043_end_0 = const()[name = tensor<string, []>("op_30043_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_30043_end_mask_0 = const()[name = tensor<string, []>("op_30043_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30043_cast = slice_by_index(begin = var_30043_begin_0, end = var_30043_end_0, end_mask = var_30043_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30043_cast")];
            tensor<int32, [4]> var_30047_begin_0 = const()[name = tensor<string, []>("op_30047_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_30047_end_0 = const()[name = tensor<string, []>("op_30047_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_30047_end_mask_0 = const()[name = tensor<string, []>("op_30047_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30047_cast = slice_by_index(begin = var_30047_begin_0, end = var_30047_end_0, end_mask = var_30047_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30047_cast")];
            tensor<int32, [4]> var_30051_begin_0 = const()[name = tensor<string, []>("op_30051_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_30051_end_0 = const()[name = tensor<string, []>("op_30051_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_30051_end_mask_0 = const()[name = tensor<string, []>("op_30051_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30051_cast = slice_by_index(begin = var_30051_begin_0, end = var_30051_end_0, end_mask = var_30051_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30051_cast")];
            tensor<int32, [4]> var_30055_begin_0 = const()[name = tensor<string, []>("op_30055_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_30055_end_0 = const()[name = tensor<string, []>("op_30055_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_30055_end_mask_0 = const()[name = tensor<string, []>("op_30055_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30055_cast = slice_by_index(begin = var_30055_begin_0, end = var_30055_end_0, end_mask = var_30055_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30055_cast")];
            tensor<int32, [4]> var_30059_begin_0 = const()[name = tensor<string, []>("op_30059_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_30059_end_0 = const()[name = tensor<string, []>("op_30059_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_30059_end_mask_0 = const()[name = tensor<string, []>("op_30059_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30059_cast = slice_by_index(begin = var_30059_begin_0, end = var_30059_end_0, end_mask = var_30059_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30059_cast")];
            tensor<int32, [4]> var_30063_begin_0 = const()[name = tensor<string, []>("op_30063_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_30063_end_0 = const()[name = tensor<string, []>("op_30063_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_30063_end_mask_0 = const()[name = tensor<string, []>("op_30063_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30063_cast = slice_by_index(begin = var_30063_begin_0, end = var_30063_end_0, end_mask = var_30063_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30063_cast")];
            tensor<int32, [4]> var_30067_begin_0 = const()[name = tensor<string, []>("op_30067_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_30067_end_0 = const()[name = tensor<string, []>("op_30067_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_30067_end_mask_0 = const()[name = tensor<string, []>("op_30067_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30067_cast = slice_by_index(begin = var_30067_begin_0, end = var_30067_end_0, end_mask = var_30067_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30067_cast")];
            tensor<int32, [4]> var_30071_begin_0 = const()[name = tensor<string, []>("op_30071_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_30071_end_0 = const()[name = tensor<string, []>("op_30071_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_30071_end_mask_0 = const()[name = tensor<string, []>("op_30071_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30071_cast = slice_by_index(begin = var_30071_begin_0, end = var_30071_end_0, end_mask = var_30071_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30071_cast")];
            tensor<int32, [4]> var_30075_begin_0 = const()[name = tensor<string, []>("op_30075_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_30075_end_0 = const()[name = tensor<string, []>("op_30075_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_30075_end_mask_0 = const()[name = tensor<string, []>("op_30075_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30075_cast = slice_by_index(begin = var_30075_begin_0, end = var_30075_end_0, end_mask = var_30075_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30075_cast")];
            tensor<int32, [4]> var_30079_begin_0 = const()[name = tensor<string, []>("op_30079_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_30079_end_0 = const()[name = tensor<string, []>("op_30079_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_30079_end_mask_0 = const()[name = tensor<string, []>("op_30079_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30079_cast = slice_by_index(begin = var_30079_begin_0, end = var_30079_end_0, end_mask = var_30079_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30079_cast")];
            tensor<int32, [4]> var_30083_begin_0 = const()[name = tensor<string, []>("op_30083_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_30083_end_0 = const()[name = tensor<string, []>("op_30083_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_30083_end_mask_0 = const()[name = tensor<string, []>("op_30083_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30083_cast = slice_by_index(begin = var_30083_begin_0, end = var_30083_end_0, end_mask = var_30083_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30083_cast")];
            tensor<int32, [4]> var_30087_begin_0 = const()[name = tensor<string, []>("op_30087_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_30087_end_0 = const()[name = tensor<string, []>("op_30087_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_30087_end_mask_0 = const()[name = tensor<string, []>("op_30087_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30087_cast = slice_by_index(begin = var_30087_begin_0, end = var_30087_end_0, end_mask = var_30087_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30087_cast")];
            tensor<int32, [4]> var_30091_begin_0 = const()[name = tensor<string, []>("op_30091_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_30091_end_0 = const()[name = tensor<string, []>("op_30091_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_30091_end_mask_0 = const()[name = tensor<string, []>("op_30091_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30091_cast = slice_by_index(begin = var_30091_begin_0, end = var_30091_end_0, end_mask = var_30091_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30091_cast")];
            tensor<int32, [4]> var_30095_begin_0 = const()[name = tensor<string, []>("op_30095_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_30095_end_0 = const()[name = tensor<string, []>("op_30095_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_30095_end_mask_0 = const()[name = tensor<string, []>("op_30095_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30095_cast = slice_by_index(begin = var_30095_begin_0, end = var_30095_end_0, end_mask = var_30095_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30095_cast")];
            tensor<int32, [4]> var_30099_begin_0 = const()[name = tensor<string, []>("op_30099_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_30099_end_0 = const()[name = tensor<string, []>("op_30099_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_30099_end_mask_0 = const()[name = tensor<string, []>("op_30099_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30099_cast = slice_by_index(begin = var_30099_begin_0, end = var_30099_end_0, end_mask = var_30099_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30099_cast")];
            tensor<int32, [4]> var_30103_begin_0 = const()[name = tensor<string, []>("op_30103_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_30103_end_0 = const()[name = tensor<string, []>("op_30103_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_30103_end_mask_0 = const()[name = tensor<string, []>("op_30103_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30103_cast = slice_by_index(begin = var_30103_begin_0, end = var_30103_end_0, end_mask = var_30103_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30103_cast")];
            tensor<int32, [4]> var_30107_begin_0 = const()[name = tensor<string, []>("op_30107_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_30107_end_0 = const()[name = tensor<string, []>("op_30107_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_30107_end_mask_0 = const()[name = tensor<string, []>("op_30107_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_30107_cast = slice_by_index(begin = var_30107_begin_0, end = var_30107_end_0, end_mask = var_30107_end_mask_0, x = v_135_cast)[name = tensor<string, []>("op_30107_cast")];
            tensor<string, []> var_30111_equation_0 = const()[name = tensor<string, []>("op_30111_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30111_cast = einsum(equation = var_30111_equation_0, values = (var_29953_cast, var_29870_cast))[name = tensor<string, []>("op_30111_cast")];
            tensor<fp16, []> var_30112_to_fp16 = const()[name = tensor<string, []>("op_30112_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2521_cast = mul(x = var_30111_cast, y = var_30112_to_fp16)[name = tensor<string, []>("aw_2521_cast")];
            tensor<string, []> var_30115_equation_0 = const()[name = tensor<string, []>("op_30115_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30115_cast = einsum(equation = var_30115_equation_0, values = (var_29957_cast, var_29874_cast))[name = tensor<string, []>("op_30115_cast")];
            tensor<fp16, []> var_30116_to_fp16 = const()[name = tensor<string, []>("op_30116_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2523_cast = mul(x = var_30115_cast, y = var_30116_to_fp16)[name = tensor<string, []>("aw_2523_cast")];
            tensor<string, []> var_30119_equation_0 = const()[name = tensor<string, []>("op_30119_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30119_cast = einsum(equation = var_30119_equation_0, values = (var_29961_cast, var_29878_cast))[name = tensor<string, []>("op_30119_cast")];
            tensor<fp16, []> var_30120_to_fp16 = const()[name = tensor<string, []>("op_30120_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2525_cast = mul(x = var_30119_cast, y = var_30120_to_fp16)[name = tensor<string, []>("aw_2525_cast")];
            tensor<string, []> var_30123_equation_0 = const()[name = tensor<string, []>("op_30123_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30123_cast = einsum(equation = var_30123_equation_0, values = (var_29965_cast, var_29882_cast))[name = tensor<string, []>("op_30123_cast")];
            tensor<fp16, []> var_30124_to_fp16 = const()[name = tensor<string, []>("op_30124_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2527_cast = mul(x = var_30123_cast, y = var_30124_to_fp16)[name = tensor<string, []>("aw_2527_cast")];
            tensor<string, []> var_30127_equation_0 = const()[name = tensor<string, []>("op_30127_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30127_cast = einsum(equation = var_30127_equation_0, values = (var_29969_cast, var_29886_cast))[name = tensor<string, []>("op_30127_cast")];
            tensor<fp16, []> var_30128_to_fp16 = const()[name = tensor<string, []>("op_30128_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2529_cast = mul(x = var_30127_cast, y = var_30128_to_fp16)[name = tensor<string, []>("aw_2529_cast")];
            tensor<string, []> var_30131_equation_0 = const()[name = tensor<string, []>("op_30131_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30131_cast = einsum(equation = var_30131_equation_0, values = (var_29973_cast, var_29890_cast))[name = tensor<string, []>("op_30131_cast")];
            tensor<fp16, []> var_30132_to_fp16 = const()[name = tensor<string, []>("op_30132_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2531_cast = mul(x = var_30131_cast, y = var_30132_to_fp16)[name = tensor<string, []>("aw_2531_cast")];
            tensor<string, []> var_30135_equation_0 = const()[name = tensor<string, []>("op_30135_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30135_cast = einsum(equation = var_30135_equation_0, values = (var_29977_cast, var_29894_cast))[name = tensor<string, []>("op_30135_cast")];
            tensor<fp16, []> var_30136_to_fp16 = const()[name = tensor<string, []>("op_30136_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2533_cast = mul(x = var_30135_cast, y = var_30136_to_fp16)[name = tensor<string, []>("aw_2533_cast")];
            tensor<string, []> var_30139_equation_0 = const()[name = tensor<string, []>("op_30139_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30139_cast = einsum(equation = var_30139_equation_0, values = (var_29981_cast, var_29898_cast))[name = tensor<string, []>("op_30139_cast")];
            tensor<fp16, []> var_30140_to_fp16 = const()[name = tensor<string, []>("op_30140_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2535_cast = mul(x = var_30139_cast, y = var_30140_to_fp16)[name = tensor<string, []>("aw_2535_cast")];
            tensor<string, []> var_30143_equation_0 = const()[name = tensor<string, []>("op_30143_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30143_cast = einsum(equation = var_30143_equation_0, values = (var_29985_cast, var_29902_cast))[name = tensor<string, []>("op_30143_cast")];
            tensor<fp16, []> var_30144_to_fp16 = const()[name = tensor<string, []>("op_30144_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2537_cast = mul(x = var_30143_cast, y = var_30144_to_fp16)[name = tensor<string, []>("aw_2537_cast")];
            tensor<string, []> var_30147_equation_0 = const()[name = tensor<string, []>("op_30147_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30147_cast = einsum(equation = var_30147_equation_0, values = (var_29989_cast, var_29906_cast))[name = tensor<string, []>("op_30147_cast")];
            tensor<fp16, []> var_30148_to_fp16 = const()[name = tensor<string, []>("op_30148_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2539_cast = mul(x = var_30147_cast, y = var_30148_to_fp16)[name = tensor<string, []>("aw_2539_cast")];
            tensor<string, []> var_30151_equation_0 = const()[name = tensor<string, []>("op_30151_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30151_cast = einsum(equation = var_30151_equation_0, values = (var_29993_cast, var_29910_cast))[name = tensor<string, []>("op_30151_cast")];
            tensor<fp16, []> var_30152_to_fp16 = const()[name = tensor<string, []>("op_30152_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2541_cast = mul(x = var_30151_cast, y = var_30152_to_fp16)[name = tensor<string, []>("aw_2541_cast")];
            tensor<string, []> var_30155_equation_0 = const()[name = tensor<string, []>("op_30155_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30155_cast = einsum(equation = var_30155_equation_0, values = (var_29997_cast, var_29914_cast))[name = tensor<string, []>("op_30155_cast")];
            tensor<fp16, []> var_30156_to_fp16 = const()[name = tensor<string, []>("op_30156_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2543_cast = mul(x = var_30155_cast, y = var_30156_to_fp16)[name = tensor<string, []>("aw_2543_cast")];
            tensor<string, []> var_30159_equation_0 = const()[name = tensor<string, []>("op_30159_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30159_cast = einsum(equation = var_30159_equation_0, values = (var_30001_cast, var_29918_cast))[name = tensor<string, []>("op_30159_cast")];
            tensor<fp16, []> var_30160_to_fp16 = const()[name = tensor<string, []>("op_30160_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2545_cast = mul(x = var_30159_cast, y = var_30160_to_fp16)[name = tensor<string, []>("aw_2545_cast")];
            tensor<string, []> var_30163_equation_0 = const()[name = tensor<string, []>("op_30163_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30163_cast = einsum(equation = var_30163_equation_0, values = (var_30005_cast, var_29922_cast))[name = tensor<string, []>("op_30163_cast")];
            tensor<fp16, []> var_30164_to_fp16 = const()[name = tensor<string, []>("op_30164_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2547_cast = mul(x = var_30163_cast, y = var_30164_to_fp16)[name = tensor<string, []>("aw_2547_cast")];
            tensor<string, []> var_30167_equation_0 = const()[name = tensor<string, []>("op_30167_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30167_cast = einsum(equation = var_30167_equation_0, values = (var_30009_cast, var_29926_cast))[name = tensor<string, []>("op_30167_cast")];
            tensor<fp16, []> var_30168_to_fp16 = const()[name = tensor<string, []>("op_30168_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2549_cast = mul(x = var_30167_cast, y = var_30168_to_fp16)[name = tensor<string, []>("aw_2549_cast")];
            tensor<string, []> var_30171_equation_0 = const()[name = tensor<string, []>("op_30171_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30171_cast = einsum(equation = var_30171_equation_0, values = (var_30013_cast, var_29930_cast))[name = tensor<string, []>("op_30171_cast")];
            tensor<fp16, []> var_30172_to_fp16 = const()[name = tensor<string, []>("op_30172_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2551_cast = mul(x = var_30171_cast, y = var_30172_to_fp16)[name = tensor<string, []>("aw_2551_cast")];
            tensor<string, []> var_30175_equation_0 = const()[name = tensor<string, []>("op_30175_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30175_cast = einsum(equation = var_30175_equation_0, values = (var_30017_cast, var_29934_cast))[name = tensor<string, []>("op_30175_cast")];
            tensor<fp16, []> var_30176_to_fp16 = const()[name = tensor<string, []>("op_30176_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2553_cast = mul(x = var_30175_cast, y = var_30176_to_fp16)[name = tensor<string, []>("aw_2553_cast")];
            tensor<string, []> var_30179_equation_0 = const()[name = tensor<string, []>("op_30179_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30179_cast = einsum(equation = var_30179_equation_0, values = (var_30021_cast, var_29938_cast))[name = tensor<string, []>("op_30179_cast")];
            tensor<fp16, []> var_30180_to_fp16 = const()[name = tensor<string, []>("op_30180_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2555_cast = mul(x = var_30179_cast, y = var_30180_to_fp16)[name = tensor<string, []>("aw_2555_cast")];
            tensor<string, []> var_30183_equation_0 = const()[name = tensor<string, []>("op_30183_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30183_cast = einsum(equation = var_30183_equation_0, values = (var_30025_cast, var_29942_cast))[name = tensor<string, []>("op_30183_cast")];
            tensor<fp16, []> var_30184_to_fp16 = const()[name = tensor<string, []>("op_30184_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2557_cast = mul(x = var_30183_cast, y = var_30184_to_fp16)[name = tensor<string, []>("aw_2557_cast")];
            tensor<string, []> var_30187_equation_0 = const()[name = tensor<string, []>("op_30187_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_30187_cast = einsum(equation = var_30187_equation_0, values = (var_30029_cast, var_29946_cast))[name = tensor<string, []>("op_30187_cast")];
            tensor<fp16, []> var_30188_to_fp16 = const()[name = tensor<string, []>("op_30188_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2559_cast = mul(x = var_30187_cast, y = var_30188_to_fp16)[name = tensor<string, []>("aw_2559_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30190_cast = softmax(axis = var_21105, x = aw_2521_cast)[name = tensor<string, []>("op_30190_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30191_cast = softmax(axis = var_21105, x = aw_2523_cast)[name = tensor<string, []>("op_30191_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30192_cast = softmax(axis = var_21105, x = aw_2525_cast)[name = tensor<string, []>("op_30192_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30193_cast = softmax(axis = var_21105, x = aw_2527_cast)[name = tensor<string, []>("op_30193_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30194_cast = softmax(axis = var_21105, x = aw_2529_cast)[name = tensor<string, []>("op_30194_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30195_cast = softmax(axis = var_21105, x = aw_2531_cast)[name = tensor<string, []>("op_30195_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30196_cast = softmax(axis = var_21105, x = aw_2533_cast)[name = tensor<string, []>("op_30196_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30197_cast = softmax(axis = var_21105, x = aw_2535_cast)[name = tensor<string, []>("op_30197_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30198_cast = softmax(axis = var_21105, x = aw_2537_cast)[name = tensor<string, []>("op_30198_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30199_cast = softmax(axis = var_21105, x = aw_2539_cast)[name = tensor<string, []>("op_30199_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30200_cast = softmax(axis = var_21105, x = aw_2541_cast)[name = tensor<string, []>("op_30200_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30201_cast = softmax(axis = var_21105, x = aw_2543_cast)[name = tensor<string, []>("op_30201_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30202_cast = softmax(axis = var_21105, x = aw_2545_cast)[name = tensor<string, []>("op_30202_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30203_cast = softmax(axis = var_21105, x = aw_2547_cast)[name = tensor<string, []>("op_30203_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30204_cast = softmax(axis = var_21105, x = aw_2549_cast)[name = tensor<string, []>("op_30204_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30205_cast = softmax(axis = var_21105, x = aw_2551_cast)[name = tensor<string, []>("op_30205_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30206_cast = softmax(axis = var_21105, x = aw_2553_cast)[name = tensor<string, []>("op_30206_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30207_cast = softmax(axis = var_21105, x = aw_2555_cast)[name = tensor<string, []>("op_30207_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30208_cast = softmax(axis = var_21105, x = aw_2557_cast)[name = tensor<string, []>("op_30208_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_30209_cast = softmax(axis = var_21105, x = aw_2559_cast)[name = tensor<string, []>("op_30209_cast")];
            tensor<string, []> var_30211_equation_0 = const()[name = tensor<string, []>("op_30211_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30211_cast = einsum(equation = var_30211_equation_0, values = (var_30031_cast, var_30190_cast))[name = tensor<string, []>("op_30211_cast")];
            tensor<string, []> var_30213_equation_0 = const()[name = tensor<string, []>("op_30213_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30213_cast = einsum(equation = var_30213_equation_0, values = (var_30035_cast, var_30191_cast))[name = tensor<string, []>("op_30213_cast")];
            tensor<string, []> var_30215_equation_0 = const()[name = tensor<string, []>("op_30215_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30215_cast = einsum(equation = var_30215_equation_0, values = (var_30039_cast, var_30192_cast))[name = tensor<string, []>("op_30215_cast")];
            tensor<string, []> var_30217_equation_0 = const()[name = tensor<string, []>("op_30217_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30217_cast = einsum(equation = var_30217_equation_0, values = (var_30043_cast, var_30193_cast))[name = tensor<string, []>("op_30217_cast")];
            tensor<string, []> var_30219_equation_0 = const()[name = tensor<string, []>("op_30219_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30219_cast = einsum(equation = var_30219_equation_0, values = (var_30047_cast, var_30194_cast))[name = tensor<string, []>("op_30219_cast")];
            tensor<string, []> var_30221_equation_0 = const()[name = tensor<string, []>("op_30221_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30221_cast = einsum(equation = var_30221_equation_0, values = (var_30051_cast, var_30195_cast))[name = tensor<string, []>("op_30221_cast")];
            tensor<string, []> var_30223_equation_0 = const()[name = tensor<string, []>("op_30223_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30223_cast = einsum(equation = var_30223_equation_0, values = (var_30055_cast, var_30196_cast))[name = tensor<string, []>("op_30223_cast")];
            tensor<string, []> var_30225_equation_0 = const()[name = tensor<string, []>("op_30225_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30225_cast = einsum(equation = var_30225_equation_0, values = (var_30059_cast, var_30197_cast))[name = tensor<string, []>("op_30225_cast")];
            tensor<string, []> var_30227_equation_0 = const()[name = tensor<string, []>("op_30227_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30227_cast = einsum(equation = var_30227_equation_0, values = (var_30063_cast, var_30198_cast))[name = tensor<string, []>("op_30227_cast")];
            tensor<string, []> var_30229_equation_0 = const()[name = tensor<string, []>("op_30229_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30229_cast = einsum(equation = var_30229_equation_0, values = (var_30067_cast, var_30199_cast))[name = tensor<string, []>("op_30229_cast")];
            tensor<string, []> var_30231_equation_0 = const()[name = tensor<string, []>("op_30231_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30231_cast = einsum(equation = var_30231_equation_0, values = (var_30071_cast, var_30200_cast))[name = tensor<string, []>("op_30231_cast")];
            tensor<string, []> var_30233_equation_0 = const()[name = tensor<string, []>("op_30233_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30233_cast = einsum(equation = var_30233_equation_0, values = (var_30075_cast, var_30201_cast))[name = tensor<string, []>("op_30233_cast")];
            tensor<string, []> var_30235_equation_0 = const()[name = tensor<string, []>("op_30235_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30235_cast = einsum(equation = var_30235_equation_0, values = (var_30079_cast, var_30202_cast))[name = tensor<string, []>("op_30235_cast")];
            tensor<string, []> var_30237_equation_0 = const()[name = tensor<string, []>("op_30237_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30237_cast = einsum(equation = var_30237_equation_0, values = (var_30083_cast, var_30203_cast))[name = tensor<string, []>("op_30237_cast")];
            tensor<string, []> var_30239_equation_0 = const()[name = tensor<string, []>("op_30239_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30239_cast = einsum(equation = var_30239_equation_0, values = (var_30087_cast, var_30204_cast))[name = tensor<string, []>("op_30239_cast")];
            tensor<string, []> var_30241_equation_0 = const()[name = tensor<string, []>("op_30241_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30241_cast = einsum(equation = var_30241_equation_0, values = (var_30091_cast, var_30205_cast))[name = tensor<string, []>("op_30241_cast")];
            tensor<string, []> var_30243_equation_0 = const()[name = tensor<string, []>("op_30243_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30243_cast = einsum(equation = var_30243_equation_0, values = (var_30095_cast, var_30206_cast))[name = tensor<string, []>("op_30243_cast")];
            tensor<string, []> var_30245_equation_0 = const()[name = tensor<string, []>("op_30245_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30245_cast = einsum(equation = var_30245_equation_0, values = (var_30099_cast, var_30207_cast))[name = tensor<string, []>("op_30245_cast")];
            tensor<string, []> var_30247_equation_0 = const()[name = tensor<string, []>("op_30247_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30247_cast = einsum(equation = var_30247_equation_0, values = (var_30103_cast, var_30208_cast))[name = tensor<string, []>("op_30247_cast")];
            tensor<string, []> var_30249_equation_0 = const()[name = tensor<string, []>("op_30249_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30249_cast = einsum(equation = var_30249_equation_0, values = (var_30107_cast, var_30209_cast))[name = tensor<string, []>("op_30249_cast")];
            tensor<bool, []> input_403_interleave_0 = const()[name = tensor<string, []>("input_403_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_403_cast = concat(axis = var_21105, interleave = input_403_interleave_0, values = (var_30211_cast, var_30213_cast, var_30215_cast, var_30217_cast, var_30219_cast, var_30221_cast, var_30223_cast, var_30225_cast, var_30227_cast, var_30229_cast, var_30231_cast, var_30233_cast, var_30235_cast, var_30237_cast, var_30239_cast, var_30241_cast, var_30243_cast, var_30245_cast, var_30247_cast, var_30249_cast))[name = tensor<string, []>("input_403_cast")];
            tensor<int32, [2]> var_30255 = const()[name = tensor<string, []>("op_30255"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_30257 = const()[name = tensor<string, []>("op_30257"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_30259_pad_type_0 = const()[name = tensor<string, []>("op_30259_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_30259_pad_0 = const()[name = tensor<string, []>("op_30259_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2378609024)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2381885888)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_30259_cast = conv(bias = mid_block_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16, dilations = var_30257, groups = var_21105, pad = var_30259_pad_0, pad_type = var_30259_pad_type_0, strides = var_30255, weight = mid_block_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16, x = input_403_cast)[name = tensor<string, []>("op_30259_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_203_cast = add(x = var_30259_cast, y = inputs_201_cast)[name = tensor<string, []>("inputs_203_cast")];
            tensor<int32, [1]> var_30263 = const()[name = tensor<string, []>("op_30263"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_203_cast = reduce_mean(axes = var_30263, keep_dims = var_21100, x = inputs_203_cast)[name = tensor<string, []>("channels_mean_203_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_203_cast = sub(x = inputs_203_cast, y = channels_mean_203_cast)[name = tensor<string, []>("zero_mean_203_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_203_cast = mul(x = zero_mean_203_cast, y = zero_mean_203_cast)[name = tensor<string, []>("zero_mean_sq_203_cast")];
            tensor<int32, [1]> var_30267 = const()[name = tensor<string, []>("op_30267"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_30268_cast = reduce_mean(axes = var_30267, keep_dims = var_21100, x = zero_mean_sq_203_cast)[name = tensor<string, []>("op_30268_cast")];
            tensor<fp16, []> var_30269_to_fp16 = const()[name = tensor<string, []>("op_30269_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_30270_cast = add(x = var_30268_cast, y = var_30269_to_fp16)[name = tensor<string, []>("op_30270_cast")];
            tensor<fp16, []> denom_203_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_203_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_203_cast = rsqrt(epsilon = denom_203_epsilon_0_to_fp16, x = var_30270_cast)[name = tensor<string, []>("denom_203_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_203_cast = mul(x = zero_mean_203_cast, y = denom_203_cast)[name = tensor<string, []>("out_203_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_30274_to_fp16 = const()[name = tensor<string, []>("op_30274_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2381888512)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_30275_cast = add(x = out_203_cast, y = var_30274_to_fp16)[name = tensor<string, []>("op_30275_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_30277_to_fp16 = const()[name = tensor<string, []>("op_30277_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2381891136)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_405_cast = mul(x = var_30275_cast, y = var_30277_to_fp16)[name = tensor<string, []>("input_405_cast")];
            tensor<int32, [2]> var_30285 = const()[name = tensor<string, []>("op_30285"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_30287 = const()[name = tensor<string, []>("op_30287"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_30289_pad_type_0 = const()[name = tensor<string, []>("op_30289_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_30289_pad_0 = const()[name = tensor<string, []>("op_30289_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2381893760)))];
            tensor<fp16, [10240]> mid_block_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2408108224)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_30289_cast = conv(bias = mid_block_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16, dilations = var_30287, groups = var_21105, pad = var_30289_pad_0, pad_type = var_30289_pad_type_0, strides = var_30285, weight = mid_block_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16, x = input_405_cast)[name = tensor<string, []>("op_30289_cast")];
            tensor<int32, [2]> var_30290_split_sizes_0 = const()[name = tensor<string, []>("op_30290_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_30290_axis_0 = const()[name = tensor<string, []>("op_30290_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_30290_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_30290_cast_1 = split(axis = var_30290_axis_0, split_sizes = var_30290_split_sizes_0, x = var_30289_cast)[name = tensor<string, []>("op_30290_cast")];
            tensor<string, []> var_30292_mode_0 = const()[name = tensor<string, []>("op_30292_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_30292_cast = gelu(mode = var_30292_mode_0, x = var_30290_cast_1)[name = tensor<string, []>("op_30292_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_407_cast = mul(x = var_30290_cast_0, y = var_30292_cast)[name = tensor<string, []>("input_407_cast")];
            tensor<int32, [2]> var_30296 = const()[name = tensor<string, []>("op_30296"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_30298 = const()[name = tensor<string, []>("op_30298"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_30300_pad_type_0 = const()[name = tensor<string, []>("op_30300_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_30300_pad_0 = const()[name = tensor<string, []>("op_30300_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> mid_block_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2408128768)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2421236032)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_30300_cast = conv(bias = mid_block_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16, dilations = var_30298, groups = var_21105, pad = var_30300_pad_0, pad_type = var_30300_pad_type_0, strides = var_30296, weight = mid_block_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16, x = input_407_cast)[name = tensor<string, []>("op_30300_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_269_cast = add(x = var_30300_cast, y = inputs_203_cast)[name = tensor<string, []>("hidden_states_269_cast")];
            tensor<int32, [4]> var_30302 = const()[name = tensor<string, []>("op_30302"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> input_409_cast = reshape(shape = var_30302, x = hidden_states_269_cast)[name = tensor<string, []>("input_409_cast")];
            tensor<int32, [2]> var_30306 = const()[name = tensor<string, []>("op_30306"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_30308 = const()[name = tensor<string, []>("op_30308"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_271_pad_type_0 = const()[name = tensor<string, []>("hidden_states_271_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_271_pad_0 = const()[name = tensor<string, []>("hidden_states_271_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_proj_out_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2421238656)))];
            tensor<fp16, [1280]> mid_block_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_proj_out_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2424515520)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_271_cast = conv(bias = mid_block_attentions_0_proj_out_bias_to_fp16, dilations = var_30308, groups = var_21105, pad = hidden_states_271_pad_0, pad_type = hidden_states_271_pad_type_0, strides = var_30306, weight = mid_block_attentions_0_proj_out_weight_to_fp16, x = input_409_cast)[name = tensor<string, []>("hidden_states_271_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_411_cast = add(x = hidden_states_271_cast, y = hidden_states_205_cast)[name = tensor<string, []>("input_411_cast")];
            tensor<int32, [5]> reshape_76_shape_0 = const()[name = tensor<string, []>("reshape_76_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_76_cast = reshape(shape = reshape_76_shape_0, x = input_411_cast)[name = tensor<string, []>("reshape_76_cast")];
            tensor<int32, [3]> reduce_mean_57_axes_0 = const()[name = tensor<string, []>("reduce_mean_57_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_57_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_57_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_57_cast = reduce_mean(axes = reduce_mean_57_axes_0, keep_dims = reduce_mean_57_keep_dims_0, x = reshape_76_cast)[name = tensor<string, []>("reduce_mean_57_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_38_cast = sub(x = reshape_76_cast, y = reduce_mean_57_cast)[name = tensor<string, []>("sub_38_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_19_cast = square(x = sub_38_cast)[name = tensor<string, []>("square_19_cast")];
            tensor<int32, [3]> reduce_mean_59_axes_0 = const()[name = tensor<string, []>("reduce_mean_59_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_59_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_59_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_59_cast = reduce_mean(axes = reduce_mean_59_axes_0, keep_dims = reduce_mean_59_keep_dims_0, x = square_19_cast)[name = tensor<string, []>("reduce_mean_59_cast")];
            tensor<fp16, []> add_38_y_0_to_fp16 = const()[name = tensor<string, []>("add_38_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_38_cast = add(x = reduce_mean_59_cast, y = add_38_y_0_to_fp16)[name = tensor<string, []>("add_38_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_19_cast = sqrt(x = add_38_cast)[name = tensor<string, []>("sqrt_19_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_19_cast = real_div(x = sub_38_cast, y = sqrt_19_cast)[name = tensor<string, []>("real_div_19_cast")];
            tensor<int32, [4]> reshape_77_shape_0 = const()[name = tensor<string, []>("reshape_77_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_77_cast = reshape(shape = reshape_77_shape_0, x = real_div_19_cast)[name = tensor<string, []>("reshape_77_cast")];
            tensor<fp16, [1280]> add_39_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_39_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2424518144)))];
            tensor<fp16, [1280]> add_39_beta_0_to_fp16 = const()[name = tensor<string, []>("add_39_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2424520768)))];
            tensor<fp16, []> add_39_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_39_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 32, 32]> add_39_cast = batch_norm(beta = add_39_beta_0_to_fp16, epsilon = add_39_epsilon_0_to_fp16, gamma = add_39_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_77_cast)[name = tensor<string, []>("add_39_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_415_cast = silu(x = add_39_cast)[name = tensor<string, []>("input_415_cast")];
            tensor<int32, [2]> var_30323 = const()[name = tensor<string, []>("op_30323"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_30325 = const()[name = tensor<string, []>("op_30325"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_273_pad_type_0 = const()[name = tensor<string, []>("hidden_states_273_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_273_pad_0 = const()[name = tensor<string, []>("hidden_states_273_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> mid_block_resnets_1_conv1_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_1_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2424523392)))];
            tensor<fp16, [1280]> mid_block_resnets_1_conv1_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_1_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2454014656)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_273_cast = conv(bias = mid_block_resnets_1_conv1_bias_to_fp16, dilations = var_30325, groups = var_21105, pad = hidden_states_273_pad_0, pad_type = hidden_states_273_pad_type_0, strides = var_30323, weight = mid_block_resnets_1_conv1_weight_to_fp16, x = input_415_cast)[name = tensor<string, []>("hidden_states_273_cast")];
            tensor<int32, [2]> var_30331 = const()[name = tensor<string, []>("op_30331"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_30333 = const()[name = tensor<string, []>("op_30333"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_15_pad_type_0 = const()[name = tensor<string, []>("temb_15_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_15_pad_0 = const()[name = tensor<string, []>("temb_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2454017280)))];
            tensor<fp16, [1280]> mid_block_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2457294144)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_15_cast = conv(bias = mid_block_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_30333, groups = var_21105, pad = temb_15_pad_0, pad_type = temb_15_pad_type_0, strides = var_30331, weight = mid_block_resnets_1_time_emb_proj_weight_to_fp16, x = input_23_cast)[name = tensor<string, []>("temb_15_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_419_cast = add(x = hidden_states_273_cast, y = temb_15_cast)[name = tensor<string, []>("input_419_cast")];
            tensor<int32, [5]> reshape_80_shape_0 = const()[name = tensor<string, []>("reshape_80_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_80_cast = reshape(shape = reshape_80_shape_0, x = input_419_cast)[name = tensor<string, []>("reshape_80_cast")];
            tensor<int32, [3]> reduce_mean_60_axes_0 = const()[name = tensor<string, []>("reduce_mean_60_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_60_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_60_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_60_cast = reduce_mean(axes = reduce_mean_60_axes_0, keep_dims = reduce_mean_60_keep_dims_0, x = reshape_80_cast)[name = tensor<string, []>("reduce_mean_60_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_40_cast = sub(x = reshape_80_cast, y = reduce_mean_60_cast)[name = tensor<string, []>("sub_40_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_20_cast = square(x = sub_40_cast)[name = tensor<string, []>("square_20_cast")];
            tensor<int32, [3]> reduce_mean_62_axes_0 = const()[name = tensor<string, []>("reduce_mean_62_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_62_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_62_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_62_cast = reduce_mean(axes = reduce_mean_62_axes_0, keep_dims = reduce_mean_62_keep_dims_0, x = square_20_cast)[name = tensor<string, []>("reduce_mean_62_cast")];
            tensor<fp16, []> add_40_y_0_to_fp16 = const()[name = tensor<string, []>("add_40_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_40_cast = add(x = reduce_mean_62_cast, y = add_40_y_0_to_fp16)[name = tensor<string, []>("add_40_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_20_cast = sqrt(x = add_40_cast)[name = tensor<string, []>("sqrt_20_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_20_cast = real_div(x = sub_40_cast, y = sqrt_20_cast)[name = tensor<string, []>("real_div_20_cast")];
            tensor<int32, [4]> reshape_81_shape_0 = const()[name = tensor<string, []>("reshape_81_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_81_cast = reshape(shape = reshape_81_shape_0, x = real_div_20_cast)[name = tensor<string, []>("reshape_81_cast")];
            tensor<fp16, [1280]> add_41_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_41_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2457296768)))];
            tensor<fp16, [1280]> add_41_beta_0_to_fp16 = const()[name = tensor<string, []>("add_41_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2457299392)))];
            tensor<fp16, []> add_41_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_41_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 32, 32]> add_41_cast = batch_norm(beta = add_41_beta_0_to_fp16, epsilon = add_41_epsilon_0_to_fp16, gamma = add_41_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_81_cast)[name = tensor<string, []>("add_41_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_423_cast = silu(x = add_41_cast)[name = tensor<string, []>("input_423_cast")];
            tensor<int32, [2]> var_30343 = const()[name = tensor<string, []>("op_30343"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_30345 = const()[name = tensor<string, []>("op_30345"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_275_pad_type_0 = const()[name = tensor<string, []>("hidden_states_275_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_275_pad_0 = const()[name = tensor<string, []>("hidden_states_275_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> mid_block_resnets_1_conv2_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_1_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2457302016)))];
            tensor<fp16, [1280]> mid_block_resnets_1_conv2_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_1_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2486793280)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_275_cast = conv(bias = mid_block_resnets_1_conv2_bias_to_fp16, dilations = var_30345, groups = var_21105, pad = hidden_states_275_pad_0, pad_type = hidden_states_275_pad_type_0, strides = var_30343, weight = mid_block_resnets_1_conv2_weight_to_fp16, x = input_423_cast)[name = tensor<string, []>("hidden_states_275_cast")];
            tensor<fp16, [2, 1280, 32, 32]> sample_cast = add(x = input_411_cast, y = hidden_states_275_cast)[name = tensor<string, []>("sample_cast")];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_277_cast = add(x = sample_cast, y = additional_residual_9)[name = tensor<string, []>("hidden_states_277_cast")];
            tensor<bool, []> var_30380 = const()[name = tensor<string, []>("op_30380"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_30385 = const()[name = tensor<string, []>("op_30385"), val = tensor<int32, []>(1)];
            tensor<bool, []> input_425_interleave_0 = const()[name = tensor<string, []>("input_425_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 2560, 32, 32]> input_425_cast = concat(axis = var_30385, interleave = input_425_interleave_0, values = (hidden_states_277_cast, res_hidden_states_1_cast))[name = tensor<string, []>("input_425_cast")];
            tensor<int32, [5]> reshape_84_shape_0 = const()[name = tensor<string, []>("reshape_84_shape_0"), val = tensor<int32, [5]>([2, 32, 80, 32, 32])];
            tensor<fp16, [2, 32, 80, 32, 32]> reshape_84_cast = reshape(shape = reshape_84_shape_0, x = input_425_cast)[name = tensor<string, []>("reshape_84_cast")];
            tensor<int32, [3]> reduce_mean_63_axes_0 = const()[name = tensor<string, []>("reduce_mean_63_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_63_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_63_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_63_cast = reduce_mean(axes = reduce_mean_63_axes_0, keep_dims = reduce_mean_63_keep_dims_0, x = reshape_84_cast)[name = tensor<string, []>("reduce_mean_63_cast")];
            tensor<fp16, [2, 32, 80, 32, 32]> sub_42_cast = sub(x = reshape_84_cast, y = reduce_mean_63_cast)[name = tensor<string, []>("sub_42_cast")];
            tensor<fp16, [2, 32, 80, 32, 32]> square_21_cast = square(x = sub_42_cast)[name = tensor<string, []>("square_21_cast")];
            tensor<int32, [3]> reduce_mean_65_axes_0 = const()[name = tensor<string, []>("reduce_mean_65_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_65_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_65_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_65_cast = reduce_mean(axes = reduce_mean_65_axes_0, keep_dims = reduce_mean_65_keep_dims_0, x = square_21_cast)[name = tensor<string, []>("reduce_mean_65_cast")];
            tensor<fp16, []> add_42_y_0_to_fp16 = const()[name = tensor<string, []>("add_42_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_42_cast = add(x = reduce_mean_65_cast, y = add_42_y_0_to_fp16)[name = tensor<string, []>("add_42_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_21_cast = sqrt(x = add_42_cast)[name = tensor<string, []>("sqrt_21_cast")];
            tensor<fp16, [2, 32, 80, 32, 32]> real_div_21_cast = real_div(x = sub_42_cast, y = sqrt_21_cast)[name = tensor<string, []>("real_div_21_cast")];
            tensor<int32, [4]> reshape_85_shape_0 = const()[name = tensor<string, []>("reshape_85_shape_0"), val = tensor<int32, [4]>([2, 2560, 32, 32])];
            tensor<fp16, [2, 2560, 32, 32]> reshape_85_cast = reshape(shape = reshape_85_shape_0, x = real_div_21_cast)[name = tensor<string, []>("reshape_85_cast")];
            tensor<fp16, [2560]> add_43_mean_0_to_fp16 = const()[name = tensor<string, []>("add_43_mean_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2486795904)))];
            tensor<fp16, [2560]> add_43_variance_0_to_fp16 = const()[name = tensor<string, []>("add_43_variance_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2486801088)))];
            tensor<fp16, [2560]> add_43_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_43_gamma_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2486806272)))];
            tensor<fp16, [2560]> add_43_beta_0_to_fp16 = const()[name = tensor<string, []>("add_43_beta_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2486811456)))];
            tensor<fp16, []> add_43_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_43_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 2560, 32, 32]> add_43_cast = batch_norm(beta = add_43_beta_0_to_fp16, epsilon = add_43_epsilon_0_to_fp16, gamma = add_43_gamma_0_to_fp16, mean = add_43_mean_0_to_fp16, variance = add_43_variance_0_to_fp16, x = reshape_85_cast)[name = tensor<string, []>("add_43_cast")];
            tensor<fp16, [2, 2560, 32, 32]> input_429_cast = silu(x = add_43_cast)[name = tensor<string, []>("input_429_cast")];
            tensor<int32, [2]> var_30414 = const()[name = tensor<string, []>("op_30414"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_30416 = const()[name = tensor<string, []>("op_30416"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_279_pad_type_0 = const()[name = tensor<string, []>("hidden_states_279_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_279_pad_0 = const()[name = tensor<string, []>("hidden_states_279_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 2560, 3, 3]> up_blocks_0_resnets_0_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 2560, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2486816640)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_0_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2545799104)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_279_cast = conv(bias = up_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = var_30416, groups = var_30385, pad = hidden_states_279_pad_0, pad_type = hidden_states_279_pad_type_0, strides = var_30414, weight = up_blocks_0_resnets_0_conv1_weight_to_fp16, x = input_429_cast)[name = tensor<string, []>("hidden_states_279_cast")];
            tensor<int32, [2]> var_30422 = const()[name = tensor<string, []>("op_30422"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_30424 = const()[name = tensor<string, []>("op_30424"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_17_pad_type_0 = const()[name = tensor<string, []>("temb_17_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_17_pad_0 = const()[name = tensor<string, []>("temb_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2545801728)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2549078592)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_17_cast = conv(bias = up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_30424, groups = var_30385, pad = temb_17_pad_0, pad_type = temb_17_pad_type_0, strides = var_30422, weight = up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16, x = input_23_cast)[name = tensor<string, []>("temb_17_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_433_cast = add(x = hidden_states_279_cast, y = temb_17_cast)[name = tensor<string, []>("input_433_cast")];
            tensor<int32, [5]> reshape_88_shape_0 = const()[name = tensor<string, []>("reshape_88_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_88_cast = reshape(shape = reshape_88_shape_0, x = input_433_cast)[name = tensor<string, []>("reshape_88_cast")];
            tensor<int32, [3]> reduce_mean_66_axes_0 = const()[name = tensor<string, []>("reduce_mean_66_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_66_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_66_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_66_cast = reduce_mean(axes = reduce_mean_66_axes_0, keep_dims = reduce_mean_66_keep_dims_0, x = reshape_88_cast)[name = tensor<string, []>("reduce_mean_66_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_44_cast = sub(x = reshape_88_cast, y = reduce_mean_66_cast)[name = tensor<string, []>("sub_44_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_22_cast = square(x = sub_44_cast)[name = tensor<string, []>("square_22_cast")];
            tensor<int32, [3]> reduce_mean_68_axes_0 = const()[name = tensor<string, []>("reduce_mean_68_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_68_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_68_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_68_cast = reduce_mean(axes = reduce_mean_68_axes_0, keep_dims = reduce_mean_68_keep_dims_0, x = square_22_cast)[name = tensor<string, []>("reduce_mean_68_cast")];
            tensor<fp16, []> add_44_y_0_to_fp16 = const()[name = tensor<string, []>("add_44_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_44_cast = add(x = reduce_mean_68_cast, y = add_44_y_0_to_fp16)[name = tensor<string, []>("add_44_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_22_cast = sqrt(x = add_44_cast)[name = tensor<string, []>("sqrt_22_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_22_cast = real_div(x = sub_44_cast, y = sqrt_22_cast)[name = tensor<string, []>("real_div_22_cast")];
            tensor<int32, [4]> reshape_89_shape_0 = const()[name = tensor<string, []>("reshape_89_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_89_cast = reshape(shape = reshape_89_shape_0, x = real_div_22_cast)[name = tensor<string, []>("reshape_89_cast")];
            tensor<fp16, [1280]> add_45_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_45_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2549081216)))];
            tensor<fp16, [1280]> add_45_beta_0_to_fp16 = const()[name = tensor<string, []>("add_45_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2549083840)))];
            tensor<fp16, []> add_45_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_45_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 32, 32]> add_45_cast = batch_norm(beta = add_45_beta_0_to_fp16, epsilon = add_45_epsilon_0_to_fp16, gamma = add_45_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_89_cast)[name = tensor<string, []>("add_45_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_437_cast = silu(x = add_45_cast)[name = tensor<string, []>("input_437_cast")];
            tensor<int32, [2]> var_30434 = const()[name = tensor<string, []>("op_30434"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_30436 = const()[name = tensor<string, []>("op_30436"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_281_pad_type_0 = const()[name = tensor<string, []>("hidden_states_281_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_281_pad_0 = const()[name = tensor<string, []>("hidden_states_281_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> up_blocks_0_resnets_0_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2549086464)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_0_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2578577728)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_281_cast = conv(bias = up_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = var_30436, groups = var_30385, pad = hidden_states_281_pad_0, pad_type = hidden_states_281_pad_type_0, strides = var_30434, weight = up_blocks_0_resnets_0_conv2_weight_to_fp16, x = input_437_cast)[name = tensor<string, []>("hidden_states_281_cast")];
            tensor<int32, [2]> var_30441 = const()[name = tensor<string, []>("op_30441"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_30443 = const()[name = tensor<string, []>("op_30443"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_5_pad_type_0 = const()[name = tensor<string, []>("x_5_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_5_pad_0 = const()[name = tensor<string, []>("x_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2560, 1, 1]> up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [1280, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2578580352)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2585134016)))];
            tensor<fp16, [2, 1280, 32, 32]> x_5_cast = conv(bias = up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_30443, groups = var_30385, pad = x_5_pad_0, pad_type = x_5_pad_type_0, strides = var_30441, weight = up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16, x = input_425_cast)[name = tensor<string, []>("x_5_cast")];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_283_cast = add(x = x_5_cast, y = hidden_states_281_cast)[name = tensor<string, []>("hidden_states_283_cast")];
            tensor<int32, [5]> reshape_92_shape_0 = const()[name = tensor<string, []>("reshape_92_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_92_cast = reshape(shape = reshape_92_shape_0, x = hidden_states_283_cast)[name = tensor<string, []>("reshape_92_cast")];
            tensor<int32, [3]> reduce_mean_69_axes_0 = const()[name = tensor<string, []>("reduce_mean_69_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_69_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_69_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_69_cast = reduce_mean(axes = reduce_mean_69_axes_0, keep_dims = reduce_mean_69_keep_dims_0, x = reshape_92_cast)[name = tensor<string, []>("reduce_mean_69_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_46_cast = sub(x = reshape_92_cast, y = reduce_mean_69_cast)[name = tensor<string, []>("sub_46_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_23_cast = square(x = sub_46_cast)[name = tensor<string, []>("square_23_cast")];
            tensor<int32, [3]> reduce_mean_71_axes_0 = const()[name = tensor<string, []>("reduce_mean_71_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_71_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_71_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_71_cast = reduce_mean(axes = reduce_mean_71_axes_0, keep_dims = reduce_mean_71_keep_dims_0, x = square_23_cast)[name = tensor<string, []>("reduce_mean_71_cast")];
            tensor<fp16, []> add_46_y_0_to_fp16 = const()[name = tensor<string, []>("add_46_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_46_cast = add(x = reduce_mean_71_cast, y = add_46_y_0_to_fp16)[name = tensor<string, []>("add_46_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_23_cast = sqrt(x = add_46_cast)[name = tensor<string, []>("sqrt_23_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_23_cast = real_div(x = sub_46_cast, y = sqrt_23_cast)[name = tensor<string, []>("real_div_23_cast")];
            tensor<int32, [4]> reshape_93_shape_0 = const()[name = tensor<string, []>("reshape_93_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_93_cast = reshape(shape = reshape_93_shape_0, x = real_div_23_cast)[name = tensor<string, []>("reshape_93_cast")];
            tensor<fp16, [1280]> add_47_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_47_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2585136640)))];
            tensor<fp16, [1280]> add_47_beta_0_to_fp16 = const()[name = tensor<string, []>("add_47_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2585139264)))];
            tensor<fp16, []> add_47_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_47_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 32, 32]> add_47_cast = batch_norm(beta = add_47_beta_0_to_fp16, epsilon = add_47_epsilon_0_to_fp16, gamma = add_47_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_93_cast)[name = tensor<string, []>("add_47_cast")];
            tensor<int32, [2]> var_30481 = const()[name = tensor<string, []>("op_30481"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_30483 = const()[name = tensor<string, []>("op_30483"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_285_pad_type_0 = const()[name = tensor<string, []>("hidden_states_285_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_285_pad_0 = const()[name = tensor<string, []>("hidden_states_285_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_proj_in_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2585141888)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_proj_in_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2588418752)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_285_cast = conv(bias = up_blocks_0_attentions_0_proj_in_bias_to_fp16, dilations = var_30483, groups = var_30385, pad = hidden_states_285_pad_0, pad_type = hidden_states_285_pad_type_0, strides = var_30481, weight = up_blocks_0_attentions_0_proj_in_weight_to_fp16, x = add_47_cast)[name = tensor<string, []>("hidden_states_285_cast")];
            tensor<int32, [4]> var_30488 = const()[name = tensor<string, []>("op_30488"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_205_cast = reshape(shape = var_30488, x = hidden_states_285_cast)[name = tensor<string, []>("inputs_205_cast")];
            tensor<int32, [1]> var_30498 = const()[name = tensor<string, []>("op_30498"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_205_cast = reduce_mean(axes = var_30498, keep_dims = var_30380, x = inputs_205_cast)[name = tensor<string, []>("channels_mean_205_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_205_cast = sub(x = inputs_205_cast, y = channels_mean_205_cast)[name = tensor<string, []>("zero_mean_205_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_205_cast = mul(x = zero_mean_205_cast, y = zero_mean_205_cast)[name = tensor<string, []>("zero_mean_sq_205_cast")];
            tensor<int32, [1]> var_30502 = const()[name = tensor<string, []>("op_30502"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_30503_cast = reduce_mean(axes = var_30502, keep_dims = var_30380, x = zero_mean_sq_205_cast)[name = tensor<string, []>("op_30503_cast")];
            tensor<fp16, []> var_30504_to_fp16 = const()[name = tensor<string, []>("op_30504_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_30505_cast = add(x = var_30503_cast, y = var_30504_to_fp16)[name = tensor<string, []>("op_30505_cast")];
            tensor<fp16, []> denom_205_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_205_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_205_cast = rsqrt(epsilon = denom_205_epsilon_0_to_fp16, x = var_30505_cast)[name = tensor<string, []>("denom_205_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_205_cast = mul(x = zero_mean_205_cast, y = denom_205_cast)[name = tensor<string, []>("out_205_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_30509_to_fp16 = const()[name = tensor<string, []>("op_30509_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2588421376)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_30510_cast = add(x = out_205_cast, y = var_30509_to_fp16)[name = tensor<string, []>("op_30510_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_30512_to_fp16 = const()[name = tensor<string, []>("op_30512_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2588424000)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_287_cast = mul(x = var_30510_cast, y = var_30512_to_fp16)[name = tensor<string, []>("hidden_states_287_cast")];
            tensor<int32, [2]> var_30519 = const()[name = tensor<string, []>("op_30519"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_30521 = const()[name = tensor<string, []>("op_30521"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_137_pad_type_0 = const()[name = tensor<string, []>("q_137_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_137_pad_0 = const()[name = tensor<string, []>("q_137_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2588426624)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_137_cast = conv(dilations = var_30521, groups = var_30385, pad = q_137_pad_0, pad_type = q_137_pad_type_0, strides = var_30519, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_287_cast)[name = tensor<string, []>("q_137_cast")];
            tensor<int32, [2]> var_30525 = const()[name = tensor<string, []>("op_30525"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_30527 = const()[name = tensor<string, []>("op_30527"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_273_pad_type_0 = const()[name = tensor<string, []>("k_273_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_273_pad_0 = const()[name = tensor<string, []>("k_273_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2591703488)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_273_cast = conv(dilations = var_30527, groups = var_30385, pad = k_273_pad_0, pad_type = k_273_pad_type_0, strides = var_30525, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_287_cast)[name = tensor<string, []>("k_273_cast")];
            tensor<int32, [2]> var_30531 = const()[name = tensor<string, []>("op_30531"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_30533 = const()[name = tensor<string, []>("op_30533"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_137_pad_type_0 = const()[name = tensor<string, []>("v_137_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_137_pad_0 = const()[name = tensor<string, []>("v_137_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2594980352)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_137_cast = conv(dilations = var_30533, groups = var_30385, pad = v_137_pad_0, pad_type = v_137_pad_type_0, strides = var_30531, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_287_cast)[name = tensor<string, []>("v_137_cast")];
            tensor<int32, [4]> var_30537_begin_0 = const()[name = tensor<string, []>("op_30537_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_30537_end_0 = const()[name = tensor<string, []>("op_30537_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_30537_end_mask_0 = const()[name = tensor<string, []>("op_30537_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30537_cast = slice_by_index(begin = var_30537_begin_0, end = var_30537_end_0, end_mask = var_30537_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30537_cast")];
            tensor<int32, [4]> var_30541_begin_0 = const()[name = tensor<string, []>("op_30541_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_30541_end_0 = const()[name = tensor<string, []>("op_30541_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_30541_end_mask_0 = const()[name = tensor<string, []>("op_30541_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30541_cast = slice_by_index(begin = var_30541_begin_0, end = var_30541_end_0, end_mask = var_30541_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30541_cast")];
            tensor<int32, [4]> var_30545_begin_0 = const()[name = tensor<string, []>("op_30545_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_30545_end_0 = const()[name = tensor<string, []>("op_30545_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_30545_end_mask_0 = const()[name = tensor<string, []>("op_30545_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30545_cast = slice_by_index(begin = var_30545_begin_0, end = var_30545_end_0, end_mask = var_30545_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30545_cast")];
            tensor<int32, [4]> var_30549_begin_0 = const()[name = tensor<string, []>("op_30549_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_30549_end_0 = const()[name = tensor<string, []>("op_30549_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_30549_end_mask_0 = const()[name = tensor<string, []>("op_30549_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30549_cast = slice_by_index(begin = var_30549_begin_0, end = var_30549_end_0, end_mask = var_30549_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30549_cast")];
            tensor<int32, [4]> var_30553_begin_0 = const()[name = tensor<string, []>("op_30553_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_30553_end_0 = const()[name = tensor<string, []>("op_30553_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_30553_end_mask_0 = const()[name = tensor<string, []>("op_30553_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30553_cast = slice_by_index(begin = var_30553_begin_0, end = var_30553_end_0, end_mask = var_30553_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30553_cast")];
            tensor<int32, [4]> var_30557_begin_0 = const()[name = tensor<string, []>("op_30557_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_30557_end_0 = const()[name = tensor<string, []>("op_30557_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_30557_end_mask_0 = const()[name = tensor<string, []>("op_30557_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30557_cast = slice_by_index(begin = var_30557_begin_0, end = var_30557_end_0, end_mask = var_30557_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30557_cast")];
            tensor<int32, [4]> var_30561_begin_0 = const()[name = tensor<string, []>("op_30561_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_30561_end_0 = const()[name = tensor<string, []>("op_30561_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_30561_end_mask_0 = const()[name = tensor<string, []>("op_30561_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30561_cast = slice_by_index(begin = var_30561_begin_0, end = var_30561_end_0, end_mask = var_30561_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30561_cast")];
            tensor<int32, [4]> var_30565_begin_0 = const()[name = tensor<string, []>("op_30565_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_30565_end_0 = const()[name = tensor<string, []>("op_30565_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_30565_end_mask_0 = const()[name = tensor<string, []>("op_30565_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30565_cast = slice_by_index(begin = var_30565_begin_0, end = var_30565_end_0, end_mask = var_30565_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30565_cast")];
            tensor<int32, [4]> var_30569_begin_0 = const()[name = tensor<string, []>("op_30569_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_30569_end_0 = const()[name = tensor<string, []>("op_30569_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_30569_end_mask_0 = const()[name = tensor<string, []>("op_30569_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30569_cast = slice_by_index(begin = var_30569_begin_0, end = var_30569_end_0, end_mask = var_30569_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30569_cast")];
            tensor<int32, [4]> var_30573_begin_0 = const()[name = tensor<string, []>("op_30573_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_30573_end_0 = const()[name = tensor<string, []>("op_30573_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_30573_end_mask_0 = const()[name = tensor<string, []>("op_30573_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30573_cast = slice_by_index(begin = var_30573_begin_0, end = var_30573_end_0, end_mask = var_30573_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30573_cast")];
            tensor<int32, [4]> var_30577_begin_0 = const()[name = tensor<string, []>("op_30577_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_30577_end_0 = const()[name = tensor<string, []>("op_30577_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_30577_end_mask_0 = const()[name = tensor<string, []>("op_30577_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30577_cast = slice_by_index(begin = var_30577_begin_0, end = var_30577_end_0, end_mask = var_30577_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30577_cast")];
            tensor<int32, [4]> var_30581_begin_0 = const()[name = tensor<string, []>("op_30581_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_30581_end_0 = const()[name = tensor<string, []>("op_30581_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_30581_end_mask_0 = const()[name = tensor<string, []>("op_30581_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30581_cast = slice_by_index(begin = var_30581_begin_0, end = var_30581_end_0, end_mask = var_30581_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30581_cast")];
            tensor<int32, [4]> var_30585_begin_0 = const()[name = tensor<string, []>("op_30585_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_30585_end_0 = const()[name = tensor<string, []>("op_30585_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_30585_end_mask_0 = const()[name = tensor<string, []>("op_30585_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30585_cast = slice_by_index(begin = var_30585_begin_0, end = var_30585_end_0, end_mask = var_30585_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30585_cast")];
            tensor<int32, [4]> var_30589_begin_0 = const()[name = tensor<string, []>("op_30589_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_30589_end_0 = const()[name = tensor<string, []>("op_30589_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_30589_end_mask_0 = const()[name = tensor<string, []>("op_30589_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30589_cast = slice_by_index(begin = var_30589_begin_0, end = var_30589_end_0, end_mask = var_30589_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30589_cast")];
            tensor<int32, [4]> var_30593_begin_0 = const()[name = tensor<string, []>("op_30593_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_30593_end_0 = const()[name = tensor<string, []>("op_30593_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_30593_end_mask_0 = const()[name = tensor<string, []>("op_30593_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30593_cast = slice_by_index(begin = var_30593_begin_0, end = var_30593_end_0, end_mask = var_30593_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30593_cast")];
            tensor<int32, [4]> var_30597_begin_0 = const()[name = tensor<string, []>("op_30597_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_30597_end_0 = const()[name = tensor<string, []>("op_30597_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_30597_end_mask_0 = const()[name = tensor<string, []>("op_30597_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30597_cast = slice_by_index(begin = var_30597_begin_0, end = var_30597_end_0, end_mask = var_30597_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30597_cast")];
            tensor<int32, [4]> var_30601_begin_0 = const()[name = tensor<string, []>("op_30601_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_30601_end_0 = const()[name = tensor<string, []>("op_30601_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_30601_end_mask_0 = const()[name = tensor<string, []>("op_30601_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30601_cast = slice_by_index(begin = var_30601_begin_0, end = var_30601_end_0, end_mask = var_30601_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30601_cast")];
            tensor<int32, [4]> var_30605_begin_0 = const()[name = tensor<string, []>("op_30605_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_30605_end_0 = const()[name = tensor<string, []>("op_30605_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_30605_end_mask_0 = const()[name = tensor<string, []>("op_30605_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30605_cast = slice_by_index(begin = var_30605_begin_0, end = var_30605_end_0, end_mask = var_30605_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30605_cast")];
            tensor<int32, [4]> var_30609_begin_0 = const()[name = tensor<string, []>("op_30609_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_30609_end_0 = const()[name = tensor<string, []>("op_30609_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_30609_end_mask_0 = const()[name = tensor<string, []>("op_30609_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30609_cast = slice_by_index(begin = var_30609_begin_0, end = var_30609_end_0, end_mask = var_30609_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30609_cast")];
            tensor<int32, [4]> var_30613_begin_0 = const()[name = tensor<string, []>("op_30613_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_30613_end_0 = const()[name = tensor<string, []>("op_30613_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_30613_end_mask_0 = const()[name = tensor<string, []>("op_30613_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30613_cast = slice_by_index(begin = var_30613_begin_0, end = var_30613_end_0, end_mask = var_30613_end_mask_0, x = q_137_cast)[name = tensor<string, []>("op_30613_cast")];
            tensor<int32, [4]> k_275_perm_0 = const()[name = tensor<string, []>("k_275_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_30620_begin_0 = const()[name = tensor<string, []>("op_30620_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_30620_end_0 = const()[name = tensor<string, []>("op_30620_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_30620_end_mask_0 = const()[name = tensor<string, []>("op_30620_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_71 = transpose(perm = k_275_perm_0, x = k_273_cast)[name = tensor<string, []>("transpose_71")];
            tensor<fp16, [2, 1024, 1, 64]> var_30620_cast = slice_by_index(begin = var_30620_begin_0, end = var_30620_end_0, end_mask = var_30620_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30620_cast")];
            tensor<int32, [4]> var_30624_begin_0 = const()[name = tensor<string, []>("op_30624_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_30624_end_0 = const()[name = tensor<string, []>("op_30624_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_30624_end_mask_0 = const()[name = tensor<string, []>("op_30624_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30624_cast = slice_by_index(begin = var_30624_begin_0, end = var_30624_end_0, end_mask = var_30624_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30624_cast")];
            tensor<int32, [4]> var_30628_begin_0 = const()[name = tensor<string, []>("op_30628_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_30628_end_0 = const()[name = tensor<string, []>("op_30628_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_30628_end_mask_0 = const()[name = tensor<string, []>("op_30628_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30628_cast = slice_by_index(begin = var_30628_begin_0, end = var_30628_end_0, end_mask = var_30628_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30628_cast")];
            tensor<int32, [4]> var_30632_begin_0 = const()[name = tensor<string, []>("op_30632_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_30632_end_0 = const()[name = tensor<string, []>("op_30632_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_30632_end_mask_0 = const()[name = tensor<string, []>("op_30632_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30632_cast = slice_by_index(begin = var_30632_begin_0, end = var_30632_end_0, end_mask = var_30632_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30632_cast")];
            tensor<int32, [4]> var_30636_begin_0 = const()[name = tensor<string, []>("op_30636_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_30636_end_0 = const()[name = tensor<string, []>("op_30636_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_30636_end_mask_0 = const()[name = tensor<string, []>("op_30636_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30636_cast = slice_by_index(begin = var_30636_begin_0, end = var_30636_end_0, end_mask = var_30636_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30636_cast")];
            tensor<int32, [4]> var_30640_begin_0 = const()[name = tensor<string, []>("op_30640_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_30640_end_0 = const()[name = tensor<string, []>("op_30640_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_30640_end_mask_0 = const()[name = tensor<string, []>("op_30640_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30640_cast = slice_by_index(begin = var_30640_begin_0, end = var_30640_end_0, end_mask = var_30640_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30640_cast")];
            tensor<int32, [4]> var_30644_begin_0 = const()[name = tensor<string, []>("op_30644_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_30644_end_0 = const()[name = tensor<string, []>("op_30644_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_30644_end_mask_0 = const()[name = tensor<string, []>("op_30644_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30644_cast = slice_by_index(begin = var_30644_begin_0, end = var_30644_end_0, end_mask = var_30644_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30644_cast")];
            tensor<int32, [4]> var_30648_begin_0 = const()[name = tensor<string, []>("op_30648_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_30648_end_0 = const()[name = tensor<string, []>("op_30648_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_30648_end_mask_0 = const()[name = tensor<string, []>("op_30648_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30648_cast = slice_by_index(begin = var_30648_begin_0, end = var_30648_end_0, end_mask = var_30648_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30648_cast")];
            tensor<int32, [4]> var_30652_begin_0 = const()[name = tensor<string, []>("op_30652_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_30652_end_0 = const()[name = tensor<string, []>("op_30652_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_30652_end_mask_0 = const()[name = tensor<string, []>("op_30652_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30652_cast = slice_by_index(begin = var_30652_begin_0, end = var_30652_end_0, end_mask = var_30652_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30652_cast")];
            tensor<int32, [4]> var_30656_begin_0 = const()[name = tensor<string, []>("op_30656_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_30656_end_0 = const()[name = tensor<string, []>("op_30656_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_30656_end_mask_0 = const()[name = tensor<string, []>("op_30656_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30656_cast = slice_by_index(begin = var_30656_begin_0, end = var_30656_end_0, end_mask = var_30656_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30656_cast")];
            tensor<int32, [4]> var_30660_begin_0 = const()[name = tensor<string, []>("op_30660_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_30660_end_0 = const()[name = tensor<string, []>("op_30660_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_30660_end_mask_0 = const()[name = tensor<string, []>("op_30660_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30660_cast = slice_by_index(begin = var_30660_begin_0, end = var_30660_end_0, end_mask = var_30660_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30660_cast")];
            tensor<int32, [4]> var_30664_begin_0 = const()[name = tensor<string, []>("op_30664_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_30664_end_0 = const()[name = tensor<string, []>("op_30664_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_30664_end_mask_0 = const()[name = tensor<string, []>("op_30664_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30664_cast = slice_by_index(begin = var_30664_begin_0, end = var_30664_end_0, end_mask = var_30664_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30664_cast")];
            tensor<int32, [4]> var_30668_begin_0 = const()[name = tensor<string, []>("op_30668_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_30668_end_0 = const()[name = tensor<string, []>("op_30668_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_30668_end_mask_0 = const()[name = tensor<string, []>("op_30668_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30668_cast = slice_by_index(begin = var_30668_begin_0, end = var_30668_end_0, end_mask = var_30668_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30668_cast")];
            tensor<int32, [4]> var_30672_begin_0 = const()[name = tensor<string, []>("op_30672_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_30672_end_0 = const()[name = tensor<string, []>("op_30672_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_30672_end_mask_0 = const()[name = tensor<string, []>("op_30672_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30672_cast = slice_by_index(begin = var_30672_begin_0, end = var_30672_end_0, end_mask = var_30672_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30672_cast")];
            tensor<int32, [4]> var_30676_begin_0 = const()[name = tensor<string, []>("op_30676_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_30676_end_0 = const()[name = tensor<string, []>("op_30676_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_30676_end_mask_0 = const()[name = tensor<string, []>("op_30676_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30676_cast = slice_by_index(begin = var_30676_begin_0, end = var_30676_end_0, end_mask = var_30676_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30676_cast")];
            tensor<int32, [4]> var_30680_begin_0 = const()[name = tensor<string, []>("op_30680_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_30680_end_0 = const()[name = tensor<string, []>("op_30680_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_30680_end_mask_0 = const()[name = tensor<string, []>("op_30680_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30680_cast = slice_by_index(begin = var_30680_begin_0, end = var_30680_end_0, end_mask = var_30680_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30680_cast")];
            tensor<int32, [4]> var_30684_begin_0 = const()[name = tensor<string, []>("op_30684_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_30684_end_0 = const()[name = tensor<string, []>("op_30684_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_30684_end_mask_0 = const()[name = tensor<string, []>("op_30684_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30684_cast = slice_by_index(begin = var_30684_begin_0, end = var_30684_end_0, end_mask = var_30684_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30684_cast")];
            tensor<int32, [4]> var_30688_begin_0 = const()[name = tensor<string, []>("op_30688_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_30688_end_0 = const()[name = tensor<string, []>("op_30688_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_30688_end_mask_0 = const()[name = tensor<string, []>("op_30688_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30688_cast = slice_by_index(begin = var_30688_begin_0, end = var_30688_end_0, end_mask = var_30688_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30688_cast")];
            tensor<int32, [4]> var_30692_begin_0 = const()[name = tensor<string, []>("op_30692_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_30692_end_0 = const()[name = tensor<string, []>("op_30692_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_30692_end_mask_0 = const()[name = tensor<string, []>("op_30692_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30692_cast = slice_by_index(begin = var_30692_begin_0, end = var_30692_end_0, end_mask = var_30692_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30692_cast")];
            tensor<int32, [4]> var_30696_begin_0 = const()[name = tensor<string, []>("op_30696_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_30696_end_0 = const()[name = tensor<string, []>("op_30696_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_30696_end_mask_0 = const()[name = tensor<string, []>("op_30696_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_30696_cast = slice_by_index(begin = var_30696_begin_0, end = var_30696_end_0, end_mask = var_30696_end_mask_0, x = transpose_71)[name = tensor<string, []>("op_30696_cast")];
            tensor<int32, [4]> var_30698_begin_0 = const()[name = tensor<string, []>("op_30698_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_30698_end_0 = const()[name = tensor<string, []>("op_30698_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_30698_end_mask_0 = const()[name = tensor<string, []>("op_30698_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30698_cast = slice_by_index(begin = var_30698_begin_0, end = var_30698_end_0, end_mask = var_30698_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30698_cast")];
            tensor<int32, [4]> var_30702_begin_0 = const()[name = tensor<string, []>("op_30702_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_30702_end_0 = const()[name = tensor<string, []>("op_30702_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_30702_end_mask_0 = const()[name = tensor<string, []>("op_30702_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30702_cast = slice_by_index(begin = var_30702_begin_0, end = var_30702_end_0, end_mask = var_30702_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30702_cast")];
            tensor<int32, [4]> var_30706_begin_0 = const()[name = tensor<string, []>("op_30706_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_30706_end_0 = const()[name = tensor<string, []>("op_30706_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_30706_end_mask_0 = const()[name = tensor<string, []>("op_30706_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30706_cast = slice_by_index(begin = var_30706_begin_0, end = var_30706_end_0, end_mask = var_30706_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30706_cast")];
            tensor<int32, [4]> var_30710_begin_0 = const()[name = tensor<string, []>("op_30710_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_30710_end_0 = const()[name = tensor<string, []>("op_30710_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_30710_end_mask_0 = const()[name = tensor<string, []>("op_30710_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30710_cast = slice_by_index(begin = var_30710_begin_0, end = var_30710_end_0, end_mask = var_30710_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30710_cast")];
            tensor<int32, [4]> var_30714_begin_0 = const()[name = tensor<string, []>("op_30714_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_30714_end_0 = const()[name = tensor<string, []>("op_30714_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_30714_end_mask_0 = const()[name = tensor<string, []>("op_30714_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30714_cast = slice_by_index(begin = var_30714_begin_0, end = var_30714_end_0, end_mask = var_30714_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30714_cast")];
            tensor<int32, [4]> var_30718_begin_0 = const()[name = tensor<string, []>("op_30718_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_30718_end_0 = const()[name = tensor<string, []>("op_30718_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_30718_end_mask_0 = const()[name = tensor<string, []>("op_30718_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30718_cast = slice_by_index(begin = var_30718_begin_0, end = var_30718_end_0, end_mask = var_30718_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30718_cast")];
            tensor<int32, [4]> var_30722_begin_0 = const()[name = tensor<string, []>("op_30722_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_30722_end_0 = const()[name = tensor<string, []>("op_30722_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_30722_end_mask_0 = const()[name = tensor<string, []>("op_30722_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30722_cast = slice_by_index(begin = var_30722_begin_0, end = var_30722_end_0, end_mask = var_30722_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30722_cast")];
            tensor<int32, [4]> var_30726_begin_0 = const()[name = tensor<string, []>("op_30726_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_30726_end_0 = const()[name = tensor<string, []>("op_30726_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_30726_end_mask_0 = const()[name = tensor<string, []>("op_30726_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30726_cast = slice_by_index(begin = var_30726_begin_0, end = var_30726_end_0, end_mask = var_30726_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30726_cast")];
            tensor<int32, [4]> var_30730_begin_0 = const()[name = tensor<string, []>("op_30730_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_30730_end_0 = const()[name = tensor<string, []>("op_30730_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_30730_end_mask_0 = const()[name = tensor<string, []>("op_30730_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30730_cast = slice_by_index(begin = var_30730_begin_0, end = var_30730_end_0, end_mask = var_30730_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30730_cast")];
            tensor<int32, [4]> var_30734_begin_0 = const()[name = tensor<string, []>("op_30734_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_30734_end_0 = const()[name = tensor<string, []>("op_30734_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_30734_end_mask_0 = const()[name = tensor<string, []>("op_30734_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30734_cast = slice_by_index(begin = var_30734_begin_0, end = var_30734_end_0, end_mask = var_30734_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30734_cast")];
            tensor<int32, [4]> var_30738_begin_0 = const()[name = tensor<string, []>("op_30738_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_30738_end_0 = const()[name = tensor<string, []>("op_30738_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_30738_end_mask_0 = const()[name = tensor<string, []>("op_30738_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30738_cast = slice_by_index(begin = var_30738_begin_0, end = var_30738_end_0, end_mask = var_30738_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30738_cast")];
            tensor<int32, [4]> var_30742_begin_0 = const()[name = tensor<string, []>("op_30742_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_30742_end_0 = const()[name = tensor<string, []>("op_30742_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_30742_end_mask_0 = const()[name = tensor<string, []>("op_30742_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30742_cast = slice_by_index(begin = var_30742_begin_0, end = var_30742_end_0, end_mask = var_30742_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30742_cast")];
            tensor<int32, [4]> var_30746_begin_0 = const()[name = tensor<string, []>("op_30746_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_30746_end_0 = const()[name = tensor<string, []>("op_30746_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_30746_end_mask_0 = const()[name = tensor<string, []>("op_30746_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30746_cast = slice_by_index(begin = var_30746_begin_0, end = var_30746_end_0, end_mask = var_30746_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30746_cast")];
            tensor<int32, [4]> var_30750_begin_0 = const()[name = tensor<string, []>("op_30750_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_30750_end_0 = const()[name = tensor<string, []>("op_30750_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_30750_end_mask_0 = const()[name = tensor<string, []>("op_30750_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30750_cast = slice_by_index(begin = var_30750_begin_0, end = var_30750_end_0, end_mask = var_30750_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30750_cast")];
            tensor<int32, [4]> var_30754_begin_0 = const()[name = tensor<string, []>("op_30754_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_30754_end_0 = const()[name = tensor<string, []>("op_30754_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_30754_end_mask_0 = const()[name = tensor<string, []>("op_30754_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30754_cast = slice_by_index(begin = var_30754_begin_0, end = var_30754_end_0, end_mask = var_30754_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30754_cast")];
            tensor<int32, [4]> var_30758_begin_0 = const()[name = tensor<string, []>("op_30758_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_30758_end_0 = const()[name = tensor<string, []>("op_30758_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_30758_end_mask_0 = const()[name = tensor<string, []>("op_30758_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30758_cast = slice_by_index(begin = var_30758_begin_0, end = var_30758_end_0, end_mask = var_30758_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30758_cast")];
            tensor<int32, [4]> var_30762_begin_0 = const()[name = tensor<string, []>("op_30762_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_30762_end_0 = const()[name = tensor<string, []>("op_30762_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_30762_end_mask_0 = const()[name = tensor<string, []>("op_30762_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30762_cast = slice_by_index(begin = var_30762_begin_0, end = var_30762_end_0, end_mask = var_30762_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30762_cast")];
            tensor<int32, [4]> var_30766_begin_0 = const()[name = tensor<string, []>("op_30766_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_30766_end_0 = const()[name = tensor<string, []>("op_30766_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_30766_end_mask_0 = const()[name = tensor<string, []>("op_30766_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30766_cast = slice_by_index(begin = var_30766_begin_0, end = var_30766_end_0, end_mask = var_30766_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30766_cast")];
            tensor<int32, [4]> var_30770_begin_0 = const()[name = tensor<string, []>("op_30770_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_30770_end_0 = const()[name = tensor<string, []>("op_30770_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_30770_end_mask_0 = const()[name = tensor<string, []>("op_30770_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30770_cast = slice_by_index(begin = var_30770_begin_0, end = var_30770_end_0, end_mask = var_30770_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30770_cast")];
            tensor<int32, [4]> var_30774_begin_0 = const()[name = tensor<string, []>("op_30774_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_30774_end_0 = const()[name = tensor<string, []>("op_30774_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_30774_end_mask_0 = const()[name = tensor<string, []>("op_30774_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30774_cast = slice_by_index(begin = var_30774_begin_0, end = var_30774_end_0, end_mask = var_30774_end_mask_0, x = v_137_cast)[name = tensor<string, []>("op_30774_cast")];
            tensor<string, []> var_30778_equation_0 = const()[name = tensor<string, []>("op_30778_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30778_cast = einsum(equation = var_30778_equation_0, values = (var_30620_cast, var_30537_cast))[name = tensor<string, []>("op_30778_cast")];
            tensor<fp16, []> var_30779_to_fp16 = const()[name = tensor<string, []>("op_30779_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2561_cast = mul(x = var_30778_cast, y = var_30779_to_fp16)[name = tensor<string, []>("aw_2561_cast")];
            tensor<string, []> var_30782_equation_0 = const()[name = tensor<string, []>("op_30782_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30782_cast = einsum(equation = var_30782_equation_0, values = (var_30624_cast, var_30541_cast))[name = tensor<string, []>("op_30782_cast")];
            tensor<fp16, []> var_30783_to_fp16 = const()[name = tensor<string, []>("op_30783_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2563_cast = mul(x = var_30782_cast, y = var_30783_to_fp16)[name = tensor<string, []>("aw_2563_cast")];
            tensor<string, []> var_30786_equation_0 = const()[name = tensor<string, []>("op_30786_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30786_cast = einsum(equation = var_30786_equation_0, values = (var_30628_cast, var_30545_cast))[name = tensor<string, []>("op_30786_cast")];
            tensor<fp16, []> var_30787_to_fp16 = const()[name = tensor<string, []>("op_30787_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2565_cast = mul(x = var_30786_cast, y = var_30787_to_fp16)[name = tensor<string, []>("aw_2565_cast")];
            tensor<string, []> var_30790_equation_0 = const()[name = tensor<string, []>("op_30790_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30790_cast = einsum(equation = var_30790_equation_0, values = (var_30632_cast, var_30549_cast))[name = tensor<string, []>("op_30790_cast")];
            tensor<fp16, []> var_30791_to_fp16 = const()[name = tensor<string, []>("op_30791_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2567_cast = mul(x = var_30790_cast, y = var_30791_to_fp16)[name = tensor<string, []>("aw_2567_cast")];
            tensor<string, []> var_30794_equation_0 = const()[name = tensor<string, []>("op_30794_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30794_cast = einsum(equation = var_30794_equation_0, values = (var_30636_cast, var_30553_cast))[name = tensor<string, []>("op_30794_cast")];
            tensor<fp16, []> var_30795_to_fp16 = const()[name = tensor<string, []>("op_30795_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2569_cast = mul(x = var_30794_cast, y = var_30795_to_fp16)[name = tensor<string, []>("aw_2569_cast")];
            tensor<string, []> var_30798_equation_0 = const()[name = tensor<string, []>("op_30798_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30798_cast = einsum(equation = var_30798_equation_0, values = (var_30640_cast, var_30557_cast))[name = tensor<string, []>("op_30798_cast")];
            tensor<fp16, []> var_30799_to_fp16 = const()[name = tensor<string, []>("op_30799_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2571_cast = mul(x = var_30798_cast, y = var_30799_to_fp16)[name = tensor<string, []>("aw_2571_cast")];
            tensor<string, []> var_30802_equation_0 = const()[name = tensor<string, []>("op_30802_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30802_cast = einsum(equation = var_30802_equation_0, values = (var_30644_cast, var_30561_cast))[name = tensor<string, []>("op_30802_cast")];
            tensor<fp16, []> var_30803_to_fp16 = const()[name = tensor<string, []>("op_30803_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2573_cast = mul(x = var_30802_cast, y = var_30803_to_fp16)[name = tensor<string, []>("aw_2573_cast")];
            tensor<string, []> var_30806_equation_0 = const()[name = tensor<string, []>("op_30806_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30806_cast = einsum(equation = var_30806_equation_0, values = (var_30648_cast, var_30565_cast))[name = tensor<string, []>("op_30806_cast")];
            tensor<fp16, []> var_30807_to_fp16 = const()[name = tensor<string, []>("op_30807_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2575_cast = mul(x = var_30806_cast, y = var_30807_to_fp16)[name = tensor<string, []>("aw_2575_cast")];
            tensor<string, []> var_30810_equation_0 = const()[name = tensor<string, []>("op_30810_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30810_cast = einsum(equation = var_30810_equation_0, values = (var_30652_cast, var_30569_cast))[name = tensor<string, []>("op_30810_cast")];
            tensor<fp16, []> var_30811_to_fp16 = const()[name = tensor<string, []>("op_30811_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2577_cast = mul(x = var_30810_cast, y = var_30811_to_fp16)[name = tensor<string, []>("aw_2577_cast")];
            tensor<string, []> var_30814_equation_0 = const()[name = tensor<string, []>("op_30814_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30814_cast = einsum(equation = var_30814_equation_0, values = (var_30656_cast, var_30573_cast))[name = tensor<string, []>("op_30814_cast")];
            tensor<fp16, []> var_30815_to_fp16 = const()[name = tensor<string, []>("op_30815_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2579_cast = mul(x = var_30814_cast, y = var_30815_to_fp16)[name = tensor<string, []>("aw_2579_cast")];
            tensor<string, []> var_30818_equation_0 = const()[name = tensor<string, []>("op_30818_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30818_cast = einsum(equation = var_30818_equation_0, values = (var_30660_cast, var_30577_cast))[name = tensor<string, []>("op_30818_cast")];
            tensor<fp16, []> var_30819_to_fp16 = const()[name = tensor<string, []>("op_30819_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2581_cast = mul(x = var_30818_cast, y = var_30819_to_fp16)[name = tensor<string, []>("aw_2581_cast")];
            tensor<string, []> var_30822_equation_0 = const()[name = tensor<string, []>("op_30822_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30822_cast = einsum(equation = var_30822_equation_0, values = (var_30664_cast, var_30581_cast))[name = tensor<string, []>("op_30822_cast")];
            tensor<fp16, []> var_30823_to_fp16 = const()[name = tensor<string, []>("op_30823_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2583_cast = mul(x = var_30822_cast, y = var_30823_to_fp16)[name = tensor<string, []>("aw_2583_cast")];
            tensor<string, []> var_30826_equation_0 = const()[name = tensor<string, []>("op_30826_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30826_cast = einsum(equation = var_30826_equation_0, values = (var_30668_cast, var_30585_cast))[name = tensor<string, []>("op_30826_cast")];
            tensor<fp16, []> var_30827_to_fp16 = const()[name = tensor<string, []>("op_30827_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2585_cast = mul(x = var_30826_cast, y = var_30827_to_fp16)[name = tensor<string, []>("aw_2585_cast")];
            tensor<string, []> var_30830_equation_0 = const()[name = tensor<string, []>("op_30830_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30830_cast = einsum(equation = var_30830_equation_0, values = (var_30672_cast, var_30589_cast))[name = tensor<string, []>("op_30830_cast")];
            tensor<fp16, []> var_30831_to_fp16 = const()[name = tensor<string, []>("op_30831_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2587_cast = mul(x = var_30830_cast, y = var_30831_to_fp16)[name = tensor<string, []>("aw_2587_cast")];
            tensor<string, []> var_30834_equation_0 = const()[name = tensor<string, []>("op_30834_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30834_cast = einsum(equation = var_30834_equation_0, values = (var_30676_cast, var_30593_cast))[name = tensor<string, []>("op_30834_cast")];
            tensor<fp16, []> var_30835_to_fp16 = const()[name = tensor<string, []>("op_30835_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2589_cast = mul(x = var_30834_cast, y = var_30835_to_fp16)[name = tensor<string, []>("aw_2589_cast")];
            tensor<string, []> var_30838_equation_0 = const()[name = tensor<string, []>("op_30838_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30838_cast = einsum(equation = var_30838_equation_0, values = (var_30680_cast, var_30597_cast))[name = tensor<string, []>("op_30838_cast")];
            tensor<fp16, []> var_30839_to_fp16 = const()[name = tensor<string, []>("op_30839_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2591_cast = mul(x = var_30838_cast, y = var_30839_to_fp16)[name = tensor<string, []>("aw_2591_cast")];
            tensor<string, []> var_30842_equation_0 = const()[name = tensor<string, []>("op_30842_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30842_cast = einsum(equation = var_30842_equation_0, values = (var_30684_cast, var_30601_cast))[name = tensor<string, []>("op_30842_cast")];
            tensor<fp16, []> var_30843_to_fp16 = const()[name = tensor<string, []>("op_30843_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2593_cast = mul(x = var_30842_cast, y = var_30843_to_fp16)[name = tensor<string, []>("aw_2593_cast")];
            tensor<string, []> var_30846_equation_0 = const()[name = tensor<string, []>("op_30846_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30846_cast = einsum(equation = var_30846_equation_0, values = (var_30688_cast, var_30605_cast))[name = tensor<string, []>("op_30846_cast")];
            tensor<fp16, []> var_30847_to_fp16 = const()[name = tensor<string, []>("op_30847_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2595_cast = mul(x = var_30846_cast, y = var_30847_to_fp16)[name = tensor<string, []>("aw_2595_cast")];
            tensor<string, []> var_30850_equation_0 = const()[name = tensor<string, []>("op_30850_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30850_cast = einsum(equation = var_30850_equation_0, values = (var_30692_cast, var_30609_cast))[name = tensor<string, []>("op_30850_cast")];
            tensor<fp16, []> var_30851_to_fp16 = const()[name = tensor<string, []>("op_30851_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2597_cast = mul(x = var_30850_cast, y = var_30851_to_fp16)[name = tensor<string, []>("aw_2597_cast")];
            tensor<string, []> var_30854_equation_0 = const()[name = tensor<string, []>("op_30854_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30854_cast = einsum(equation = var_30854_equation_0, values = (var_30696_cast, var_30613_cast))[name = tensor<string, []>("op_30854_cast")];
            tensor<fp16, []> var_30855_to_fp16 = const()[name = tensor<string, []>("op_30855_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2599_cast = mul(x = var_30854_cast, y = var_30855_to_fp16)[name = tensor<string, []>("aw_2599_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30857_cast = softmax(axis = var_30385, x = aw_2561_cast)[name = tensor<string, []>("op_30857_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30858_cast = softmax(axis = var_30385, x = aw_2563_cast)[name = tensor<string, []>("op_30858_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30859_cast = softmax(axis = var_30385, x = aw_2565_cast)[name = tensor<string, []>("op_30859_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30860_cast = softmax(axis = var_30385, x = aw_2567_cast)[name = tensor<string, []>("op_30860_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30861_cast = softmax(axis = var_30385, x = aw_2569_cast)[name = tensor<string, []>("op_30861_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30862_cast = softmax(axis = var_30385, x = aw_2571_cast)[name = tensor<string, []>("op_30862_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30863_cast = softmax(axis = var_30385, x = aw_2573_cast)[name = tensor<string, []>("op_30863_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30864_cast = softmax(axis = var_30385, x = aw_2575_cast)[name = tensor<string, []>("op_30864_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30865_cast = softmax(axis = var_30385, x = aw_2577_cast)[name = tensor<string, []>("op_30865_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30866_cast = softmax(axis = var_30385, x = aw_2579_cast)[name = tensor<string, []>("op_30866_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30867_cast = softmax(axis = var_30385, x = aw_2581_cast)[name = tensor<string, []>("op_30867_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30868_cast = softmax(axis = var_30385, x = aw_2583_cast)[name = tensor<string, []>("op_30868_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30869_cast = softmax(axis = var_30385, x = aw_2585_cast)[name = tensor<string, []>("op_30869_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30870_cast = softmax(axis = var_30385, x = aw_2587_cast)[name = tensor<string, []>("op_30870_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30871_cast = softmax(axis = var_30385, x = aw_2589_cast)[name = tensor<string, []>("op_30871_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30872_cast = softmax(axis = var_30385, x = aw_2591_cast)[name = tensor<string, []>("op_30872_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30873_cast = softmax(axis = var_30385, x = aw_2593_cast)[name = tensor<string, []>("op_30873_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30874_cast = softmax(axis = var_30385, x = aw_2595_cast)[name = tensor<string, []>("op_30874_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30875_cast = softmax(axis = var_30385, x = aw_2597_cast)[name = tensor<string, []>("op_30875_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_30876_cast = softmax(axis = var_30385, x = aw_2599_cast)[name = tensor<string, []>("op_30876_cast")];
            tensor<string, []> var_30878_equation_0 = const()[name = tensor<string, []>("op_30878_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30878_cast = einsum(equation = var_30878_equation_0, values = (var_30698_cast, var_30857_cast))[name = tensor<string, []>("op_30878_cast")];
            tensor<string, []> var_30880_equation_0 = const()[name = tensor<string, []>("op_30880_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30880_cast = einsum(equation = var_30880_equation_0, values = (var_30702_cast, var_30858_cast))[name = tensor<string, []>("op_30880_cast")];
            tensor<string, []> var_30882_equation_0 = const()[name = tensor<string, []>("op_30882_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30882_cast = einsum(equation = var_30882_equation_0, values = (var_30706_cast, var_30859_cast))[name = tensor<string, []>("op_30882_cast")];
            tensor<string, []> var_30884_equation_0 = const()[name = tensor<string, []>("op_30884_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30884_cast = einsum(equation = var_30884_equation_0, values = (var_30710_cast, var_30860_cast))[name = tensor<string, []>("op_30884_cast")];
            tensor<string, []> var_30886_equation_0 = const()[name = tensor<string, []>("op_30886_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30886_cast = einsum(equation = var_30886_equation_0, values = (var_30714_cast, var_30861_cast))[name = tensor<string, []>("op_30886_cast")];
            tensor<string, []> var_30888_equation_0 = const()[name = tensor<string, []>("op_30888_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30888_cast = einsum(equation = var_30888_equation_0, values = (var_30718_cast, var_30862_cast))[name = tensor<string, []>("op_30888_cast")];
            tensor<string, []> var_30890_equation_0 = const()[name = tensor<string, []>("op_30890_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30890_cast = einsum(equation = var_30890_equation_0, values = (var_30722_cast, var_30863_cast))[name = tensor<string, []>("op_30890_cast")];
            tensor<string, []> var_30892_equation_0 = const()[name = tensor<string, []>("op_30892_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30892_cast = einsum(equation = var_30892_equation_0, values = (var_30726_cast, var_30864_cast))[name = tensor<string, []>("op_30892_cast")];
            tensor<string, []> var_30894_equation_0 = const()[name = tensor<string, []>("op_30894_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30894_cast = einsum(equation = var_30894_equation_0, values = (var_30730_cast, var_30865_cast))[name = tensor<string, []>("op_30894_cast")];
            tensor<string, []> var_30896_equation_0 = const()[name = tensor<string, []>("op_30896_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30896_cast = einsum(equation = var_30896_equation_0, values = (var_30734_cast, var_30866_cast))[name = tensor<string, []>("op_30896_cast")];
            tensor<string, []> var_30898_equation_0 = const()[name = tensor<string, []>("op_30898_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30898_cast = einsum(equation = var_30898_equation_0, values = (var_30738_cast, var_30867_cast))[name = tensor<string, []>("op_30898_cast")];
            tensor<string, []> var_30900_equation_0 = const()[name = tensor<string, []>("op_30900_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30900_cast = einsum(equation = var_30900_equation_0, values = (var_30742_cast, var_30868_cast))[name = tensor<string, []>("op_30900_cast")];
            tensor<string, []> var_30902_equation_0 = const()[name = tensor<string, []>("op_30902_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30902_cast = einsum(equation = var_30902_equation_0, values = (var_30746_cast, var_30869_cast))[name = tensor<string, []>("op_30902_cast")];
            tensor<string, []> var_30904_equation_0 = const()[name = tensor<string, []>("op_30904_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30904_cast = einsum(equation = var_30904_equation_0, values = (var_30750_cast, var_30870_cast))[name = tensor<string, []>("op_30904_cast")];
            tensor<string, []> var_30906_equation_0 = const()[name = tensor<string, []>("op_30906_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30906_cast = einsum(equation = var_30906_equation_0, values = (var_30754_cast, var_30871_cast))[name = tensor<string, []>("op_30906_cast")];
            tensor<string, []> var_30908_equation_0 = const()[name = tensor<string, []>("op_30908_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30908_cast = einsum(equation = var_30908_equation_0, values = (var_30758_cast, var_30872_cast))[name = tensor<string, []>("op_30908_cast")];
            tensor<string, []> var_30910_equation_0 = const()[name = tensor<string, []>("op_30910_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30910_cast = einsum(equation = var_30910_equation_0, values = (var_30762_cast, var_30873_cast))[name = tensor<string, []>("op_30910_cast")];
            tensor<string, []> var_30912_equation_0 = const()[name = tensor<string, []>("op_30912_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30912_cast = einsum(equation = var_30912_equation_0, values = (var_30766_cast, var_30874_cast))[name = tensor<string, []>("op_30912_cast")];
            tensor<string, []> var_30914_equation_0 = const()[name = tensor<string, []>("op_30914_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30914_cast = einsum(equation = var_30914_equation_0, values = (var_30770_cast, var_30875_cast))[name = tensor<string, []>("op_30914_cast")];
            tensor<string, []> var_30916_equation_0 = const()[name = tensor<string, []>("op_30916_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_30916_cast = einsum(equation = var_30916_equation_0, values = (var_30774_cast, var_30876_cast))[name = tensor<string, []>("op_30916_cast")];
            tensor<bool, []> input_441_interleave_0 = const()[name = tensor<string, []>("input_441_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_441_cast = concat(axis = var_30385, interleave = input_441_interleave_0, values = (var_30878_cast, var_30880_cast, var_30882_cast, var_30884_cast, var_30886_cast, var_30888_cast, var_30890_cast, var_30892_cast, var_30894_cast, var_30896_cast, var_30898_cast, var_30900_cast, var_30902_cast, var_30904_cast, var_30906_cast, var_30908_cast, var_30910_cast, var_30912_cast, var_30914_cast, var_30916_cast))[name = tensor<string, []>("input_441_cast")];
            tensor<int32, [2]> var_30922 = const()[name = tensor<string, []>("op_30922"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_30924 = const()[name = tensor<string, []>("op_30924"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_30926_pad_type_0 = const()[name = tensor<string, []>("op_30926_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_30926_pad_0 = const()[name = tensor<string, []>("op_30926_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2598257216)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2601534080)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_30926_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_30924, groups = var_30385, pad = var_30926_pad_0, pad_type = var_30926_pad_type_0, strides = var_30922, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_441_cast)[name = tensor<string, []>("op_30926_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_207_cast = add(x = var_30926_cast, y = inputs_205_cast)[name = tensor<string, []>("inputs_207_cast")];
            tensor<int32, [1]> var_30930 = const()[name = tensor<string, []>("op_30930"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_207_cast = reduce_mean(axes = var_30930, keep_dims = var_30380, x = inputs_207_cast)[name = tensor<string, []>("channels_mean_207_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_207_cast = sub(x = inputs_207_cast, y = channels_mean_207_cast)[name = tensor<string, []>("zero_mean_207_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_207_cast = mul(x = zero_mean_207_cast, y = zero_mean_207_cast)[name = tensor<string, []>("zero_mean_sq_207_cast")];
            tensor<int32, [1]> var_30934 = const()[name = tensor<string, []>("op_30934"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_30935_cast = reduce_mean(axes = var_30934, keep_dims = var_30380, x = zero_mean_sq_207_cast)[name = tensor<string, []>("op_30935_cast")];
            tensor<fp16, []> var_30936_to_fp16 = const()[name = tensor<string, []>("op_30936_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_30937_cast = add(x = var_30935_cast, y = var_30936_to_fp16)[name = tensor<string, []>("op_30937_cast")];
            tensor<fp16, []> denom_207_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_207_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_207_cast = rsqrt(epsilon = denom_207_epsilon_0_to_fp16, x = var_30937_cast)[name = tensor<string, []>("denom_207_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_207_cast = mul(x = zero_mean_207_cast, y = denom_207_cast)[name = tensor<string, []>("out_207_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_30941_to_fp16 = const()[name = tensor<string, []>("op_30941_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2601536704)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_30942_cast = add(x = out_207_cast, y = var_30941_to_fp16)[name = tensor<string, []>("op_30942_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_30944_to_fp16 = const()[name = tensor<string, []>("op_30944_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2601539328)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_289_cast = mul(x = var_30942_cast, y = var_30944_to_fp16)[name = tensor<string, []>("hidden_states_289_cast")];
            tensor<int32, [2]> var_30951 = const()[name = tensor<string, []>("op_30951"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_30953 = const()[name = tensor<string, []>("op_30953"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_139_pad_type_0 = const()[name = tensor<string, []>("q_139_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_139_pad_0 = const()[name = tensor<string, []>("q_139_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2601541952)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_139_cast = conv(dilations = var_30953, groups = var_30385, pad = q_139_pad_0, pad_type = q_139_pad_type_0, strides = var_30951, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_289_cast)[name = tensor<string, []>("q_139_cast")];
            tensor<int32, [2]> var_30957 = const()[name = tensor<string, []>("op_30957"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_30959 = const()[name = tensor<string, []>("op_30959"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_277_pad_type_0 = const()[name = tensor<string, []>("k_277_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_277_pad_0 = const()[name = tensor<string, []>("k_277_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2604818816)))];
            tensor<fp16, [2, 1280, 1, 77]> k_277_cast = conv(dilations = var_30959, groups = var_30385, pad = k_277_pad_0, pad_type = k_277_pad_type_0, strides = var_30957, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_277_cast")];
            tensor<int32, [2]> var_30963 = const()[name = tensor<string, []>("op_30963"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_30965 = const()[name = tensor<string, []>("op_30965"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_139_pad_type_0 = const()[name = tensor<string, []>("v_139_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_139_pad_0 = const()[name = tensor<string, []>("v_139_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2610061760)))];
            tensor<fp16, [2, 1280, 1, 77]> v_139_cast = conv(dilations = var_30965, groups = var_30385, pad = v_139_pad_0, pad_type = v_139_pad_type_0, strides = var_30963, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_139_cast")];
            tensor<int32, [4]> var_30969_begin_0 = const()[name = tensor<string, []>("op_30969_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_30969_end_0 = const()[name = tensor<string, []>("op_30969_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_30969_end_mask_0 = const()[name = tensor<string, []>("op_30969_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30969_cast = slice_by_index(begin = var_30969_begin_0, end = var_30969_end_0, end_mask = var_30969_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_30969_cast")];
            tensor<int32, [4]> var_30973_begin_0 = const()[name = tensor<string, []>("op_30973_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_30973_end_0 = const()[name = tensor<string, []>("op_30973_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_30973_end_mask_0 = const()[name = tensor<string, []>("op_30973_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30973_cast = slice_by_index(begin = var_30973_begin_0, end = var_30973_end_0, end_mask = var_30973_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_30973_cast")];
            tensor<int32, [4]> var_30977_begin_0 = const()[name = tensor<string, []>("op_30977_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_30977_end_0 = const()[name = tensor<string, []>("op_30977_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_30977_end_mask_0 = const()[name = tensor<string, []>("op_30977_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30977_cast = slice_by_index(begin = var_30977_begin_0, end = var_30977_end_0, end_mask = var_30977_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_30977_cast")];
            tensor<int32, [4]> var_30981_begin_0 = const()[name = tensor<string, []>("op_30981_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_30981_end_0 = const()[name = tensor<string, []>("op_30981_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_30981_end_mask_0 = const()[name = tensor<string, []>("op_30981_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30981_cast = slice_by_index(begin = var_30981_begin_0, end = var_30981_end_0, end_mask = var_30981_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_30981_cast")];
            tensor<int32, [4]> var_30985_begin_0 = const()[name = tensor<string, []>("op_30985_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_30985_end_0 = const()[name = tensor<string, []>("op_30985_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_30985_end_mask_0 = const()[name = tensor<string, []>("op_30985_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30985_cast = slice_by_index(begin = var_30985_begin_0, end = var_30985_end_0, end_mask = var_30985_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_30985_cast")];
            tensor<int32, [4]> var_30989_begin_0 = const()[name = tensor<string, []>("op_30989_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_30989_end_0 = const()[name = tensor<string, []>("op_30989_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_30989_end_mask_0 = const()[name = tensor<string, []>("op_30989_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30989_cast = slice_by_index(begin = var_30989_begin_0, end = var_30989_end_0, end_mask = var_30989_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_30989_cast")];
            tensor<int32, [4]> var_30993_begin_0 = const()[name = tensor<string, []>("op_30993_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_30993_end_0 = const()[name = tensor<string, []>("op_30993_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_30993_end_mask_0 = const()[name = tensor<string, []>("op_30993_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30993_cast = slice_by_index(begin = var_30993_begin_0, end = var_30993_end_0, end_mask = var_30993_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_30993_cast")];
            tensor<int32, [4]> var_30997_begin_0 = const()[name = tensor<string, []>("op_30997_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_30997_end_0 = const()[name = tensor<string, []>("op_30997_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_30997_end_mask_0 = const()[name = tensor<string, []>("op_30997_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_30997_cast = slice_by_index(begin = var_30997_begin_0, end = var_30997_end_0, end_mask = var_30997_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_30997_cast")];
            tensor<int32, [4]> var_31001_begin_0 = const()[name = tensor<string, []>("op_31001_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_31001_end_0 = const()[name = tensor<string, []>("op_31001_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_31001_end_mask_0 = const()[name = tensor<string, []>("op_31001_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31001_cast = slice_by_index(begin = var_31001_begin_0, end = var_31001_end_0, end_mask = var_31001_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_31001_cast")];
            tensor<int32, [4]> var_31005_begin_0 = const()[name = tensor<string, []>("op_31005_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_31005_end_0 = const()[name = tensor<string, []>("op_31005_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_31005_end_mask_0 = const()[name = tensor<string, []>("op_31005_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31005_cast = slice_by_index(begin = var_31005_begin_0, end = var_31005_end_0, end_mask = var_31005_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_31005_cast")];
            tensor<int32, [4]> var_31009_begin_0 = const()[name = tensor<string, []>("op_31009_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_31009_end_0 = const()[name = tensor<string, []>("op_31009_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_31009_end_mask_0 = const()[name = tensor<string, []>("op_31009_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31009_cast = slice_by_index(begin = var_31009_begin_0, end = var_31009_end_0, end_mask = var_31009_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_31009_cast")];
            tensor<int32, [4]> var_31013_begin_0 = const()[name = tensor<string, []>("op_31013_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_31013_end_0 = const()[name = tensor<string, []>("op_31013_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_31013_end_mask_0 = const()[name = tensor<string, []>("op_31013_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31013_cast = slice_by_index(begin = var_31013_begin_0, end = var_31013_end_0, end_mask = var_31013_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_31013_cast")];
            tensor<int32, [4]> var_31017_begin_0 = const()[name = tensor<string, []>("op_31017_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_31017_end_0 = const()[name = tensor<string, []>("op_31017_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_31017_end_mask_0 = const()[name = tensor<string, []>("op_31017_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31017_cast = slice_by_index(begin = var_31017_begin_0, end = var_31017_end_0, end_mask = var_31017_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_31017_cast")];
            tensor<int32, [4]> var_31021_begin_0 = const()[name = tensor<string, []>("op_31021_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_31021_end_0 = const()[name = tensor<string, []>("op_31021_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_31021_end_mask_0 = const()[name = tensor<string, []>("op_31021_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31021_cast = slice_by_index(begin = var_31021_begin_0, end = var_31021_end_0, end_mask = var_31021_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_31021_cast")];
            tensor<int32, [4]> var_31025_begin_0 = const()[name = tensor<string, []>("op_31025_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_31025_end_0 = const()[name = tensor<string, []>("op_31025_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_31025_end_mask_0 = const()[name = tensor<string, []>("op_31025_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31025_cast = slice_by_index(begin = var_31025_begin_0, end = var_31025_end_0, end_mask = var_31025_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_31025_cast")];
            tensor<int32, [4]> var_31029_begin_0 = const()[name = tensor<string, []>("op_31029_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_31029_end_0 = const()[name = tensor<string, []>("op_31029_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_31029_end_mask_0 = const()[name = tensor<string, []>("op_31029_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31029_cast = slice_by_index(begin = var_31029_begin_0, end = var_31029_end_0, end_mask = var_31029_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_31029_cast")];
            tensor<int32, [4]> var_31033_begin_0 = const()[name = tensor<string, []>("op_31033_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_31033_end_0 = const()[name = tensor<string, []>("op_31033_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_31033_end_mask_0 = const()[name = tensor<string, []>("op_31033_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31033_cast = slice_by_index(begin = var_31033_begin_0, end = var_31033_end_0, end_mask = var_31033_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_31033_cast")];
            tensor<int32, [4]> var_31037_begin_0 = const()[name = tensor<string, []>("op_31037_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_31037_end_0 = const()[name = tensor<string, []>("op_31037_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_31037_end_mask_0 = const()[name = tensor<string, []>("op_31037_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31037_cast = slice_by_index(begin = var_31037_begin_0, end = var_31037_end_0, end_mask = var_31037_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_31037_cast")];
            tensor<int32, [4]> var_31041_begin_0 = const()[name = tensor<string, []>("op_31041_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_31041_end_0 = const()[name = tensor<string, []>("op_31041_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_31041_end_mask_0 = const()[name = tensor<string, []>("op_31041_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31041_cast = slice_by_index(begin = var_31041_begin_0, end = var_31041_end_0, end_mask = var_31041_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_31041_cast")];
            tensor<int32, [4]> var_31045_begin_0 = const()[name = tensor<string, []>("op_31045_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_31045_end_0 = const()[name = tensor<string, []>("op_31045_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_31045_end_mask_0 = const()[name = tensor<string, []>("op_31045_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31045_cast = slice_by_index(begin = var_31045_begin_0, end = var_31045_end_0, end_mask = var_31045_end_mask_0, x = q_139_cast)[name = tensor<string, []>("op_31045_cast")];
            tensor<int32, [4]> k_279_perm_0 = const()[name = tensor<string, []>("k_279_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_31052_begin_0 = const()[name = tensor<string, []>("op_31052_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_31052_end_0 = const()[name = tensor<string, []>("op_31052_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_31052_end_mask_0 = const()[name = tensor<string, []>("op_31052_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_70 = transpose(perm = k_279_perm_0, x = k_277_cast)[name = tensor<string, []>("transpose_70")];
            tensor<fp16, [2, 77, 1, 64]> var_31052_cast = slice_by_index(begin = var_31052_begin_0, end = var_31052_end_0, end_mask = var_31052_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31052_cast")];
            tensor<int32, [4]> var_31056_begin_0 = const()[name = tensor<string, []>("op_31056_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_31056_end_0 = const()[name = tensor<string, []>("op_31056_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_31056_end_mask_0 = const()[name = tensor<string, []>("op_31056_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31056_cast = slice_by_index(begin = var_31056_begin_0, end = var_31056_end_0, end_mask = var_31056_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31056_cast")];
            tensor<int32, [4]> var_31060_begin_0 = const()[name = tensor<string, []>("op_31060_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_31060_end_0 = const()[name = tensor<string, []>("op_31060_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_31060_end_mask_0 = const()[name = tensor<string, []>("op_31060_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31060_cast = slice_by_index(begin = var_31060_begin_0, end = var_31060_end_0, end_mask = var_31060_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31060_cast")];
            tensor<int32, [4]> var_31064_begin_0 = const()[name = tensor<string, []>("op_31064_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_31064_end_0 = const()[name = tensor<string, []>("op_31064_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_31064_end_mask_0 = const()[name = tensor<string, []>("op_31064_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31064_cast = slice_by_index(begin = var_31064_begin_0, end = var_31064_end_0, end_mask = var_31064_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31064_cast")];
            tensor<int32, [4]> var_31068_begin_0 = const()[name = tensor<string, []>("op_31068_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_31068_end_0 = const()[name = tensor<string, []>("op_31068_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_31068_end_mask_0 = const()[name = tensor<string, []>("op_31068_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31068_cast = slice_by_index(begin = var_31068_begin_0, end = var_31068_end_0, end_mask = var_31068_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31068_cast")];
            tensor<int32, [4]> var_31072_begin_0 = const()[name = tensor<string, []>("op_31072_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_31072_end_0 = const()[name = tensor<string, []>("op_31072_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_31072_end_mask_0 = const()[name = tensor<string, []>("op_31072_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31072_cast = slice_by_index(begin = var_31072_begin_0, end = var_31072_end_0, end_mask = var_31072_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31072_cast")];
            tensor<int32, [4]> var_31076_begin_0 = const()[name = tensor<string, []>("op_31076_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_31076_end_0 = const()[name = tensor<string, []>("op_31076_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_31076_end_mask_0 = const()[name = tensor<string, []>("op_31076_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31076_cast = slice_by_index(begin = var_31076_begin_0, end = var_31076_end_0, end_mask = var_31076_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31076_cast")];
            tensor<int32, [4]> var_31080_begin_0 = const()[name = tensor<string, []>("op_31080_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_31080_end_0 = const()[name = tensor<string, []>("op_31080_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_31080_end_mask_0 = const()[name = tensor<string, []>("op_31080_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31080_cast = slice_by_index(begin = var_31080_begin_0, end = var_31080_end_0, end_mask = var_31080_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31080_cast")];
            tensor<int32, [4]> var_31084_begin_0 = const()[name = tensor<string, []>("op_31084_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_31084_end_0 = const()[name = tensor<string, []>("op_31084_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_31084_end_mask_0 = const()[name = tensor<string, []>("op_31084_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31084_cast = slice_by_index(begin = var_31084_begin_0, end = var_31084_end_0, end_mask = var_31084_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31084_cast")];
            tensor<int32, [4]> var_31088_begin_0 = const()[name = tensor<string, []>("op_31088_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_31088_end_0 = const()[name = tensor<string, []>("op_31088_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_31088_end_mask_0 = const()[name = tensor<string, []>("op_31088_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31088_cast = slice_by_index(begin = var_31088_begin_0, end = var_31088_end_0, end_mask = var_31088_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31088_cast")];
            tensor<int32, [4]> var_31092_begin_0 = const()[name = tensor<string, []>("op_31092_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_31092_end_0 = const()[name = tensor<string, []>("op_31092_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_31092_end_mask_0 = const()[name = tensor<string, []>("op_31092_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31092_cast = slice_by_index(begin = var_31092_begin_0, end = var_31092_end_0, end_mask = var_31092_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31092_cast")];
            tensor<int32, [4]> var_31096_begin_0 = const()[name = tensor<string, []>("op_31096_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_31096_end_0 = const()[name = tensor<string, []>("op_31096_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_31096_end_mask_0 = const()[name = tensor<string, []>("op_31096_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31096_cast = slice_by_index(begin = var_31096_begin_0, end = var_31096_end_0, end_mask = var_31096_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31096_cast")];
            tensor<int32, [4]> var_31100_begin_0 = const()[name = tensor<string, []>("op_31100_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_31100_end_0 = const()[name = tensor<string, []>("op_31100_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_31100_end_mask_0 = const()[name = tensor<string, []>("op_31100_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31100_cast = slice_by_index(begin = var_31100_begin_0, end = var_31100_end_0, end_mask = var_31100_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31100_cast")];
            tensor<int32, [4]> var_31104_begin_0 = const()[name = tensor<string, []>("op_31104_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_31104_end_0 = const()[name = tensor<string, []>("op_31104_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_31104_end_mask_0 = const()[name = tensor<string, []>("op_31104_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31104_cast = slice_by_index(begin = var_31104_begin_0, end = var_31104_end_0, end_mask = var_31104_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31104_cast")];
            tensor<int32, [4]> var_31108_begin_0 = const()[name = tensor<string, []>("op_31108_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_31108_end_0 = const()[name = tensor<string, []>("op_31108_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_31108_end_mask_0 = const()[name = tensor<string, []>("op_31108_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31108_cast = slice_by_index(begin = var_31108_begin_0, end = var_31108_end_0, end_mask = var_31108_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31108_cast")];
            tensor<int32, [4]> var_31112_begin_0 = const()[name = tensor<string, []>("op_31112_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_31112_end_0 = const()[name = tensor<string, []>("op_31112_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_31112_end_mask_0 = const()[name = tensor<string, []>("op_31112_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31112_cast = slice_by_index(begin = var_31112_begin_0, end = var_31112_end_0, end_mask = var_31112_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31112_cast")];
            tensor<int32, [4]> var_31116_begin_0 = const()[name = tensor<string, []>("op_31116_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_31116_end_0 = const()[name = tensor<string, []>("op_31116_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_31116_end_mask_0 = const()[name = tensor<string, []>("op_31116_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31116_cast = slice_by_index(begin = var_31116_begin_0, end = var_31116_end_0, end_mask = var_31116_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31116_cast")];
            tensor<int32, [4]> var_31120_begin_0 = const()[name = tensor<string, []>("op_31120_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_31120_end_0 = const()[name = tensor<string, []>("op_31120_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_31120_end_mask_0 = const()[name = tensor<string, []>("op_31120_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31120_cast = slice_by_index(begin = var_31120_begin_0, end = var_31120_end_0, end_mask = var_31120_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31120_cast")];
            tensor<int32, [4]> var_31124_begin_0 = const()[name = tensor<string, []>("op_31124_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_31124_end_0 = const()[name = tensor<string, []>("op_31124_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_31124_end_mask_0 = const()[name = tensor<string, []>("op_31124_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31124_cast = slice_by_index(begin = var_31124_begin_0, end = var_31124_end_0, end_mask = var_31124_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31124_cast")];
            tensor<int32, [4]> var_31128_begin_0 = const()[name = tensor<string, []>("op_31128_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_31128_end_0 = const()[name = tensor<string, []>("op_31128_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_31128_end_mask_0 = const()[name = tensor<string, []>("op_31128_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31128_cast = slice_by_index(begin = var_31128_begin_0, end = var_31128_end_0, end_mask = var_31128_end_mask_0, x = transpose_70)[name = tensor<string, []>("op_31128_cast")];
            tensor<int32, [4]> var_31130_begin_0 = const()[name = tensor<string, []>("op_31130_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_31130_end_0 = const()[name = tensor<string, []>("op_31130_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_31130_end_mask_0 = const()[name = tensor<string, []>("op_31130_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31130_cast = slice_by_index(begin = var_31130_begin_0, end = var_31130_end_0, end_mask = var_31130_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31130_cast")];
            tensor<int32, [4]> var_31134_begin_0 = const()[name = tensor<string, []>("op_31134_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_31134_end_0 = const()[name = tensor<string, []>("op_31134_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_31134_end_mask_0 = const()[name = tensor<string, []>("op_31134_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31134_cast = slice_by_index(begin = var_31134_begin_0, end = var_31134_end_0, end_mask = var_31134_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31134_cast")];
            tensor<int32, [4]> var_31138_begin_0 = const()[name = tensor<string, []>("op_31138_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_31138_end_0 = const()[name = tensor<string, []>("op_31138_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_31138_end_mask_0 = const()[name = tensor<string, []>("op_31138_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31138_cast = slice_by_index(begin = var_31138_begin_0, end = var_31138_end_0, end_mask = var_31138_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31138_cast")];
            tensor<int32, [4]> var_31142_begin_0 = const()[name = tensor<string, []>("op_31142_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_31142_end_0 = const()[name = tensor<string, []>("op_31142_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_31142_end_mask_0 = const()[name = tensor<string, []>("op_31142_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31142_cast = slice_by_index(begin = var_31142_begin_0, end = var_31142_end_0, end_mask = var_31142_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31142_cast")];
            tensor<int32, [4]> var_31146_begin_0 = const()[name = tensor<string, []>("op_31146_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_31146_end_0 = const()[name = tensor<string, []>("op_31146_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_31146_end_mask_0 = const()[name = tensor<string, []>("op_31146_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31146_cast = slice_by_index(begin = var_31146_begin_0, end = var_31146_end_0, end_mask = var_31146_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31146_cast")];
            tensor<int32, [4]> var_31150_begin_0 = const()[name = tensor<string, []>("op_31150_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_31150_end_0 = const()[name = tensor<string, []>("op_31150_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_31150_end_mask_0 = const()[name = tensor<string, []>("op_31150_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31150_cast = slice_by_index(begin = var_31150_begin_0, end = var_31150_end_0, end_mask = var_31150_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31150_cast")];
            tensor<int32, [4]> var_31154_begin_0 = const()[name = tensor<string, []>("op_31154_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_31154_end_0 = const()[name = tensor<string, []>("op_31154_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_31154_end_mask_0 = const()[name = tensor<string, []>("op_31154_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31154_cast = slice_by_index(begin = var_31154_begin_0, end = var_31154_end_0, end_mask = var_31154_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31154_cast")];
            tensor<int32, [4]> var_31158_begin_0 = const()[name = tensor<string, []>("op_31158_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_31158_end_0 = const()[name = tensor<string, []>("op_31158_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_31158_end_mask_0 = const()[name = tensor<string, []>("op_31158_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31158_cast = slice_by_index(begin = var_31158_begin_0, end = var_31158_end_0, end_mask = var_31158_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31158_cast")];
            tensor<int32, [4]> var_31162_begin_0 = const()[name = tensor<string, []>("op_31162_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_31162_end_0 = const()[name = tensor<string, []>("op_31162_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_31162_end_mask_0 = const()[name = tensor<string, []>("op_31162_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31162_cast = slice_by_index(begin = var_31162_begin_0, end = var_31162_end_0, end_mask = var_31162_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31162_cast")];
            tensor<int32, [4]> var_31166_begin_0 = const()[name = tensor<string, []>("op_31166_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_31166_end_0 = const()[name = tensor<string, []>("op_31166_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_31166_end_mask_0 = const()[name = tensor<string, []>("op_31166_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31166_cast = slice_by_index(begin = var_31166_begin_0, end = var_31166_end_0, end_mask = var_31166_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31166_cast")];
            tensor<int32, [4]> var_31170_begin_0 = const()[name = tensor<string, []>("op_31170_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_31170_end_0 = const()[name = tensor<string, []>("op_31170_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_31170_end_mask_0 = const()[name = tensor<string, []>("op_31170_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31170_cast = slice_by_index(begin = var_31170_begin_0, end = var_31170_end_0, end_mask = var_31170_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31170_cast")];
            tensor<int32, [4]> var_31174_begin_0 = const()[name = tensor<string, []>("op_31174_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_31174_end_0 = const()[name = tensor<string, []>("op_31174_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_31174_end_mask_0 = const()[name = tensor<string, []>("op_31174_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31174_cast = slice_by_index(begin = var_31174_begin_0, end = var_31174_end_0, end_mask = var_31174_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31174_cast")];
            tensor<int32, [4]> var_31178_begin_0 = const()[name = tensor<string, []>("op_31178_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_31178_end_0 = const()[name = tensor<string, []>("op_31178_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_31178_end_mask_0 = const()[name = tensor<string, []>("op_31178_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31178_cast = slice_by_index(begin = var_31178_begin_0, end = var_31178_end_0, end_mask = var_31178_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31178_cast")];
            tensor<int32, [4]> var_31182_begin_0 = const()[name = tensor<string, []>("op_31182_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_31182_end_0 = const()[name = tensor<string, []>("op_31182_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_31182_end_mask_0 = const()[name = tensor<string, []>("op_31182_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31182_cast = slice_by_index(begin = var_31182_begin_0, end = var_31182_end_0, end_mask = var_31182_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31182_cast")];
            tensor<int32, [4]> var_31186_begin_0 = const()[name = tensor<string, []>("op_31186_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_31186_end_0 = const()[name = tensor<string, []>("op_31186_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_31186_end_mask_0 = const()[name = tensor<string, []>("op_31186_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31186_cast = slice_by_index(begin = var_31186_begin_0, end = var_31186_end_0, end_mask = var_31186_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31186_cast")];
            tensor<int32, [4]> var_31190_begin_0 = const()[name = tensor<string, []>("op_31190_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_31190_end_0 = const()[name = tensor<string, []>("op_31190_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_31190_end_mask_0 = const()[name = tensor<string, []>("op_31190_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31190_cast = slice_by_index(begin = var_31190_begin_0, end = var_31190_end_0, end_mask = var_31190_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31190_cast")];
            tensor<int32, [4]> var_31194_begin_0 = const()[name = tensor<string, []>("op_31194_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_31194_end_0 = const()[name = tensor<string, []>("op_31194_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_31194_end_mask_0 = const()[name = tensor<string, []>("op_31194_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31194_cast = slice_by_index(begin = var_31194_begin_0, end = var_31194_end_0, end_mask = var_31194_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31194_cast")];
            tensor<int32, [4]> var_31198_begin_0 = const()[name = tensor<string, []>("op_31198_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_31198_end_0 = const()[name = tensor<string, []>("op_31198_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_31198_end_mask_0 = const()[name = tensor<string, []>("op_31198_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31198_cast = slice_by_index(begin = var_31198_begin_0, end = var_31198_end_0, end_mask = var_31198_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31198_cast")];
            tensor<int32, [4]> var_31202_begin_0 = const()[name = tensor<string, []>("op_31202_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_31202_end_0 = const()[name = tensor<string, []>("op_31202_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_31202_end_mask_0 = const()[name = tensor<string, []>("op_31202_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31202_cast = slice_by_index(begin = var_31202_begin_0, end = var_31202_end_0, end_mask = var_31202_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31202_cast")];
            tensor<int32, [4]> var_31206_begin_0 = const()[name = tensor<string, []>("op_31206_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_31206_end_0 = const()[name = tensor<string, []>("op_31206_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_31206_end_mask_0 = const()[name = tensor<string, []>("op_31206_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_31206_cast = slice_by_index(begin = var_31206_begin_0, end = var_31206_end_0, end_mask = var_31206_end_mask_0, x = v_139_cast)[name = tensor<string, []>("op_31206_cast")];
            tensor<string, []> var_31210_equation_0 = const()[name = tensor<string, []>("op_31210_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31210_cast = einsum(equation = var_31210_equation_0, values = (var_31052_cast, var_30969_cast))[name = tensor<string, []>("op_31210_cast")];
            tensor<fp16, []> var_31211_to_fp16 = const()[name = tensor<string, []>("op_31211_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2601_cast = mul(x = var_31210_cast, y = var_31211_to_fp16)[name = tensor<string, []>("aw_2601_cast")];
            tensor<string, []> var_31214_equation_0 = const()[name = tensor<string, []>("op_31214_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31214_cast = einsum(equation = var_31214_equation_0, values = (var_31056_cast, var_30973_cast))[name = tensor<string, []>("op_31214_cast")];
            tensor<fp16, []> var_31215_to_fp16 = const()[name = tensor<string, []>("op_31215_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2603_cast = mul(x = var_31214_cast, y = var_31215_to_fp16)[name = tensor<string, []>("aw_2603_cast")];
            tensor<string, []> var_31218_equation_0 = const()[name = tensor<string, []>("op_31218_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31218_cast = einsum(equation = var_31218_equation_0, values = (var_31060_cast, var_30977_cast))[name = tensor<string, []>("op_31218_cast")];
            tensor<fp16, []> var_31219_to_fp16 = const()[name = tensor<string, []>("op_31219_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2605_cast = mul(x = var_31218_cast, y = var_31219_to_fp16)[name = tensor<string, []>("aw_2605_cast")];
            tensor<string, []> var_31222_equation_0 = const()[name = tensor<string, []>("op_31222_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31222_cast = einsum(equation = var_31222_equation_0, values = (var_31064_cast, var_30981_cast))[name = tensor<string, []>("op_31222_cast")];
            tensor<fp16, []> var_31223_to_fp16 = const()[name = tensor<string, []>("op_31223_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2607_cast = mul(x = var_31222_cast, y = var_31223_to_fp16)[name = tensor<string, []>("aw_2607_cast")];
            tensor<string, []> var_31226_equation_0 = const()[name = tensor<string, []>("op_31226_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31226_cast = einsum(equation = var_31226_equation_0, values = (var_31068_cast, var_30985_cast))[name = tensor<string, []>("op_31226_cast")];
            tensor<fp16, []> var_31227_to_fp16 = const()[name = tensor<string, []>("op_31227_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2609_cast = mul(x = var_31226_cast, y = var_31227_to_fp16)[name = tensor<string, []>("aw_2609_cast")];
            tensor<string, []> var_31230_equation_0 = const()[name = tensor<string, []>("op_31230_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31230_cast = einsum(equation = var_31230_equation_0, values = (var_31072_cast, var_30989_cast))[name = tensor<string, []>("op_31230_cast")];
            tensor<fp16, []> var_31231_to_fp16 = const()[name = tensor<string, []>("op_31231_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2611_cast = mul(x = var_31230_cast, y = var_31231_to_fp16)[name = tensor<string, []>("aw_2611_cast")];
            tensor<string, []> var_31234_equation_0 = const()[name = tensor<string, []>("op_31234_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31234_cast = einsum(equation = var_31234_equation_0, values = (var_31076_cast, var_30993_cast))[name = tensor<string, []>("op_31234_cast")];
            tensor<fp16, []> var_31235_to_fp16 = const()[name = tensor<string, []>("op_31235_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2613_cast = mul(x = var_31234_cast, y = var_31235_to_fp16)[name = tensor<string, []>("aw_2613_cast")];
            tensor<string, []> var_31238_equation_0 = const()[name = tensor<string, []>("op_31238_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31238_cast = einsum(equation = var_31238_equation_0, values = (var_31080_cast, var_30997_cast))[name = tensor<string, []>("op_31238_cast")];
            tensor<fp16, []> var_31239_to_fp16 = const()[name = tensor<string, []>("op_31239_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2615_cast = mul(x = var_31238_cast, y = var_31239_to_fp16)[name = tensor<string, []>("aw_2615_cast")];
            tensor<string, []> var_31242_equation_0 = const()[name = tensor<string, []>("op_31242_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31242_cast = einsum(equation = var_31242_equation_0, values = (var_31084_cast, var_31001_cast))[name = tensor<string, []>("op_31242_cast")];
            tensor<fp16, []> var_31243_to_fp16 = const()[name = tensor<string, []>("op_31243_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2617_cast = mul(x = var_31242_cast, y = var_31243_to_fp16)[name = tensor<string, []>("aw_2617_cast")];
            tensor<string, []> var_31246_equation_0 = const()[name = tensor<string, []>("op_31246_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31246_cast = einsum(equation = var_31246_equation_0, values = (var_31088_cast, var_31005_cast))[name = tensor<string, []>("op_31246_cast")];
            tensor<fp16, []> var_31247_to_fp16 = const()[name = tensor<string, []>("op_31247_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2619_cast = mul(x = var_31246_cast, y = var_31247_to_fp16)[name = tensor<string, []>("aw_2619_cast")];
            tensor<string, []> var_31250_equation_0 = const()[name = tensor<string, []>("op_31250_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31250_cast = einsum(equation = var_31250_equation_0, values = (var_31092_cast, var_31009_cast))[name = tensor<string, []>("op_31250_cast")];
            tensor<fp16, []> var_31251_to_fp16 = const()[name = tensor<string, []>("op_31251_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2621_cast = mul(x = var_31250_cast, y = var_31251_to_fp16)[name = tensor<string, []>("aw_2621_cast")];
            tensor<string, []> var_31254_equation_0 = const()[name = tensor<string, []>("op_31254_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31254_cast = einsum(equation = var_31254_equation_0, values = (var_31096_cast, var_31013_cast))[name = tensor<string, []>("op_31254_cast")];
            tensor<fp16, []> var_31255_to_fp16 = const()[name = tensor<string, []>("op_31255_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2623_cast = mul(x = var_31254_cast, y = var_31255_to_fp16)[name = tensor<string, []>("aw_2623_cast")];
            tensor<string, []> var_31258_equation_0 = const()[name = tensor<string, []>("op_31258_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31258_cast = einsum(equation = var_31258_equation_0, values = (var_31100_cast, var_31017_cast))[name = tensor<string, []>("op_31258_cast")];
            tensor<fp16, []> var_31259_to_fp16 = const()[name = tensor<string, []>("op_31259_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2625_cast = mul(x = var_31258_cast, y = var_31259_to_fp16)[name = tensor<string, []>("aw_2625_cast")];
            tensor<string, []> var_31262_equation_0 = const()[name = tensor<string, []>("op_31262_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31262_cast = einsum(equation = var_31262_equation_0, values = (var_31104_cast, var_31021_cast))[name = tensor<string, []>("op_31262_cast")];
            tensor<fp16, []> var_31263_to_fp16 = const()[name = tensor<string, []>("op_31263_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2627_cast = mul(x = var_31262_cast, y = var_31263_to_fp16)[name = tensor<string, []>("aw_2627_cast")];
            tensor<string, []> var_31266_equation_0 = const()[name = tensor<string, []>("op_31266_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31266_cast = einsum(equation = var_31266_equation_0, values = (var_31108_cast, var_31025_cast))[name = tensor<string, []>("op_31266_cast")];
            tensor<fp16, []> var_31267_to_fp16 = const()[name = tensor<string, []>("op_31267_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2629_cast = mul(x = var_31266_cast, y = var_31267_to_fp16)[name = tensor<string, []>("aw_2629_cast")];
            tensor<string, []> var_31270_equation_0 = const()[name = tensor<string, []>("op_31270_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31270_cast = einsum(equation = var_31270_equation_0, values = (var_31112_cast, var_31029_cast))[name = tensor<string, []>("op_31270_cast")];
            tensor<fp16, []> var_31271_to_fp16 = const()[name = tensor<string, []>("op_31271_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2631_cast = mul(x = var_31270_cast, y = var_31271_to_fp16)[name = tensor<string, []>("aw_2631_cast")];
            tensor<string, []> var_31274_equation_0 = const()[name = tensor<string, []>("op_31274_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31274_cast = einsum(equation = var_31274_equation_0, values = (var_31116_cast, var_31033_cast))[name = tensor<string, []>("op_31274_cast")];
            tensor<fp16, []> var_31275_to_fp16 = const()[name = tensor<string, []>("op_31275_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2633_cast = mul(x = var_31274_cast, y = var_31275_to_fp16)[name = tensor<string, []>("aw_2633_cast")];
            tensor<string, []> var_31278_equation_0 = const()[name = tensor<string, []>("op_31278_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31278_cast = einsum(equation = var_31278_equation_0, values = (var_31120_cast, var_31037_cast))[name = tensor<string, []>("op_31278_cast")];
            tensor<fp16, []> var_31279_to_fp16 = const()[name = tensor<string, []>("op_31279_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2635_cast = mul(x = var_31278_cast, y = var_31279_to_fp16)[name = tensor<string, []>("aw_2635_cast")];
            tensor<string, []> var_31282_equation_0 = const()[name = tensor<string, []>("op_31282_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31282_cast = einsum(equation = var_31282_equation_0, values = (var_31124_cast, var_31041_cast))[name = tensor<string, []>("op_31282_cast")];
            tensor<fp16, []> var_31283_to_fp16 = const()[name = tensor<string, []>("op_31283_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2637_cast = mul(x = var_31282_cast, y = var_31283_to_fp16)[name = tensor<string, []>("aw_2637_cast")];
            tensor<string, []> var_31286_equation_0 = const()[name = tensor<string, []>("op_31286_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_31286_cast = einsum(equation = var_31286_equation_0, values = (var_31128_cast, var_31045_cast))[name = tensor<string, []>("op_31286_cast")];
            tensor<fp16, []> var_31287_to_fp16 = const()[name = tensor<string, []>("op_31287_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2639_cast = mul(x = var_31286_cast, y = var_31287_to_fp16)[name = tensor<string, []>("aw_2639_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31289_cast = softmax(axis = var_30385, x = aw_2601_cast)[name = tensor<string, []>("op_31289_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31290_cast = softmax(axis = var_30385, x = aw_2603_cast)[name = tensor<string, []>("op_31290_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31291_cast = softmax(axis = var_30385, x = aw_2605_cast)[name = tensor<string, []>("op_31291_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31292_cast = softmax(axis = var_30385, x = aw_2607_cast)[name = tensor<string, []>("op_31292_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31293_cast = softmax(axis = var_30385, x = aw_2609_cast)[name = tensor<string, []>("op_31293_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31294_cast = softmax(axis = var_30385, x = aw_2611_cast)[name = tensor<string, []>("op_31294_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31295_cast = softmax(axis = var_30385, x = aw_2613_cast)[name = tensor<string, []>("op_31295_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31296_cast = softmax(axis = var_30385, x = aw_2615_cast)[name = tensor<string, []>("op_31296_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31297_cast = softmax(axis = var_30385, x = aw_2617_cast)[name = tensor<string, []>("op_31297_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31298_cast = softmax(axis = var_30385, x = aw_2619_cast)[name = tensor<string, []>("op_31298_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31299_cast = softmax(axis = var_30385, x = aw_2621_cast)[name = tensor<string, []>("op_31299_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31300_cast = softmax(axis = var_30385, x = aw_2623_cast)[name = tensor<string, []>("op_31300_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31301_cast = softmax(axis = var_30385, x = aw_2625_cast)[name = tensor<string, []>("op_31301_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31302_cast = softmax(axis = var_30385, x = aw_2627_cast)[name = tensor<string, []>("op_31302_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31303_cast = softmax(axis = var_30385, x = aw_2629_cast)[name = tensor<string, []>("op_31303_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31304_cast = softmax(axis = var_30385, x = aw_2631_cast)[name = tensor<string, []>("op_31304_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31305_cast = softmax(axis = var_30385, x = aw_2633_cast)[name = tensor<string, []>("op_31305_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31306_cast = softmax(axis = var_30385, x = aw_2635_cast)[name = tensor<string, []>("op_31306_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31307_cast = softmax(axis = var_30385, x = aw_2637_cast)[name = tensor<string, []>("op_31307_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_31308_cast = softmax(axis = var_30385, x = aw_2639_cast)[name = tensor<string, []>("op_31308_cast")];
            tensor<string, []> var_31310_equation_0 = const()[name = tensor<string, []>("op_31310_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31310_cast = einsum(equation = var_31310_equation_0, values = (var_31130_cast, var_31289_cast))[name = tensor<string, []>("op_31310_cast")];
            tensor<string, []> var_31312_equation_0 = const()[name = tensor<string, []>("op_31312_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31312_cast = einsum(equation = var_31312_equation_0, values = (var_31134_cast, var_31290_cast))[name = tensor<string, []>("op_31312_cast")];
            tensor<string, []> var_31314_equation_0 = const()[name = tensor<string, []>("op_31314_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31314_cast = einsum(equation = var_31314_equation_0, values = (var_31138_cast, var_31291_cast))[name = tensor<string, []>("op_31314_cast")];
            tensor<string, []> var_31316_equation_0 = const()[name = tensor<string, []>("op_31316_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31316_cast = einsum(equation = var_31316_equation_0, values = (var_31142_cast, var_31292_cast))[name = tensor<string, []>("op_31316_cast")];
            tensor<string, []> var_31318_equation_0 = const()[name = tensor<string, []>("op_31318_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31318_cast = einsum(equation = var_31318_equation_0, values = (var_31146_cast, var_31293_cast))[name = tensor<string, []>("op_31318_cast")];
            tensor<string, []> var_31320_equation_0 = const()[name = tensor<string, []>("op_31320_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31320_cast = einsum(equation = var_31320_equation_0, values = (var_31150_cast, var_31294_cast))[name = tensor<string, []>("op_31320_cast")];
            tensor<string, []> var_31322_equation_0 = const()[name = tensor<string, []>("op_31322_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31322_cast = einsum(equation = var_31322_equation_0, values = (var_31154_cast, var_31295_cast))[name = tensor<string, []>("op_31322_cast")];
            tensor<string, []> var_31324_equation_0 = const()[name = tensor<string, []>("op_31324_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31324_cast = einsum(equation = var_31324_equation_0, values = (var_31158_cast, var_31296_cast))[name = tensor<string, []>("op_31324_cast")];
            tensor<string, []> var_31326_equation_0 = const()[name = tensor<string, []>("op_31326_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31326_cast = einsum(equation = var_31326_equation_0, values = (var_31162_cast, var_31297_cast))[name = tensor<string, []>("op_31326_cast")];
            tensor<string, []> var_31328_equation_0 = const()[name = tensor<string, []>("op_31328_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31328_cast = einsum(equation = var_31328_equation_0, values = (var_31166_cast, var_31298_cast))[name = tensor<string, []>("op_31328_cast")];
            tensor<string, []> var_31330_equation_0 = const()[name = tensor<string, []>("op_31330_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31330_cast = einsum(equation = var_31330_equation_0, values = (var_31170_cast, var_31299_cast))[name = tensor<string, []>("op_31330_cast")];
            tensor<string, []> var_31332_equation_0 = const()[name = tensor<string, []>("op_31332_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31332_cast = einsum(equation = var_31332_equation_0, values = (var_31174_cast, var_31300_cast))[name = tensor<string, []>("op_31332_cast")];
            tensor<string, []> var_31334_equation_0 = const()[name = tensor<string, []>("op_31334_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31334_cast = einsum(equation = var_31334_equation_0, values = (var_31178_cast, var_31301_cast))[name = tensor<string, []>("op_31334_cast")];
            tensor<string, []> var_31336_equation_0 = const()[name = tensor<string, []>("op_31336_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31336_cast = einsum(equation = var_31336_equation_0, values = (var_31182_cast, var_31302_cast))[name = tensor<string, []>("op_31336_cast")];
            tensor<string, []> var_31338_equation_0 = const()[name = tensor<string, []>("op_31338_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31338_cast = einsum(equation = var_31338_equation_0, values = (var_31186_cast, var_31303_cast))[name = tensor<string, []>("op_31338_cast")];
            tensor<string, []> var_31340_equation_0 = const()[name = tensor<string, []>("op_31340_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31340_cast = einsum(equation = var_31340_equation_0, values = (var_31190_cast, var_31304_cast))[name = tensor<string, []>("op_31340_cast")];
            tensor<string, []> var_31342_equation_0 = const()[name = tensor<string, []>("op_31342_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31342_cast = einsum(equation = var_31342_equation_0, values = (var_31194_cast, var_31305_cast))[name = tensor<string, []>("op_31342_cast")];
            tensor<string, []> var_31344_equation_0 = const()[name = tensor<string, []>("op_31344_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31344_cast = einsum(equation = var_31344_equation_0, values = (var_31198_cast, var_31306_cast))[name = tensor<string, []>("op_31344_cast")];
            tensor<string, []> var_31346_equation_0 = const()[name = tensor<string, []>("op_31346_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31346_cast = einsum(equation = var_31346_equation_0, values = (var_31202_cast, var_31307_cast))[name = tensor<string, []>("op_31346_cast")];
            tensor<string, []> var_31348_equation_0 = const()[name = tensor<string, []>("op_31348_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31348_cast = einsum(equation = var_31348_equation_0, values = (var_31206_cast, var_31308_cast))[name = tensor<string, []>("op_31348_cast")];
            tensor<bool, []> input_443_interleave_0 = const()[name = tensor<string, []>("input_443_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_443_cast = concat(axis = var_30385, interleave = input_443_interleave_0, values = (var_31310_cast, var_31312_cast, var_31314_cast, var_31316_cast, var_31318_cast, var_31320_cast, var_31322_cast, var_31324_cast, var_31326_cast, var_31328_cast, var_31330_cast, var_31332_cast, var_31334_cast, var_31336_cast, var_31338_cast, var_31340_cast, var_31342_cast, var_31344_cast, var_31346_cast, var_31348_cast))[name = tensor<string, []>("input_443_cast")];
            tensor<int32, [2]> var_31354 = const()[name = tensor<string, []>("op_31354"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_31356 = const()[name = tensor<string, []>("op_31356"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_31358_pad_type_0 = const()[name = tensor<string, []>("op_31358_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_31358_pad_0 = const()[name = tensor<string, []>("op_31358_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2615304704)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2618581568)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_31358_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_31356, groups = var_30385, pad = var_31358_pad_0, pad_type = var_31358_pad_type_0, strides = var_31354, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_443_cast)[name = tensor<string, []>("op_31358_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_209_cast = add(x = var_31358_cast, y = inputs_207_cast)[name = tensor<string, []>("inputs_209_cast")];
            tensor<int32, [1]> var_31362 = const()[name = tensor<string, []>("op_31362"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_209_cast = reduce_mean(axes = var_31362, keep_dims = var_30380, x = inputs_209_cast)[name = tensor<string, []>("channels_mean_209_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_209_cast = sub(x = inputs_209_cast, y = channels_mean_209_cast)[name = tensor<string, []>("zero_mean_209_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_209_cast = mul(x = zero_mean_209_cast, y = zero_mean_209_cast)[name = tensor<string, []>("zero_mean_sq_209_cast")];
            tensor<int32, [1]> var_31366 = const()[name = tensor<string, []>("op_31366"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_31367_cast = reduce_mean(axes = var_31366, keep_dims = var_30380, x = zero_mean_sq_209_cast)[name = tensor<string, []>("op_31367_cast")];
            tensor<fp16, []> var_31368_to_fp16 = const()[name = tensor<string, []>("op_31368_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_31369_cast = add(x = var_31367_cast, y = var_31368_to_fp16)[name = tensor<string, []>("op_31369_cast")];
            tensor<fp16, []> denom_209_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_209_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_209_cast = rsqrt(epsilon = denom_209_epsilon_0_to_fp16, x = var_31369_cast)[name = tensor<string, []>("denom_209_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_209_cast = mul(x = zero_mean_209_cast, y = denom_209_cast)[name = tensor<string, []>("out_209_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_31373_to_fp16 = const()[name = tensor<string, []>("op_31373_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2618584192)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_31374_cast = add(x = out_209_cast, y = var_31373_to_fp16)[name = tensor<string, []>("op_31374_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_31376_to_fp16 = const()[name = tensor<string, []>("op_31376_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2618586816)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_445_cast = mul(x = var_31374_cast, y = var_31376_to_fp16)[name = tensor<string, []>("input_445_cast")];
            tensor<int32, [2]> var_31384 = const()[name = tensor<string, []>("op_31384"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_31386 = const()[name = tensor<string, []>("op_31386"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_31388_pad_type_0 = const()[name = tensor<string, []>("op_31388_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_31388_pad_0 = const()[name = tensor<string, []>("op_31388_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2618589440)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2644803904)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_31388_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_31386, groups = var_30385, pad = var_31388_pad_0, pad_type = var_31388_pad_type_0, strides = var_31384, weight = up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_445_cast)[name = tensor<string, []>("op_31388_cast")];
            tensor<int32, [2]> var_31389_split_sizes_0 = const()[name = tensor<string, []>("op_31389_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_31389_axis_0 = const()[name = tensor<string, []>("op_31389_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_31389_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_31389_cast_1 = split(axis = var_31389_axis_0, split_sizes = var_31389_split_sizes_0, x = var_31388_cast)[name = tensor<string, []>("op_31389_cast")];
            tensor<string, []> var_31391_mode_0 = const()[name = tensor<string, []>("op_31391_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_31391_cast = gelu(mode = var_31391_mode_0, x = var_31389_cast_1)[name = tensor<string, []>("op_31391_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_447_cast = mul(x = var_31389_cast_0, y = var_31391_cast)[name = tensor<string, []>("input_447_cast")];
            tensor<int32, [2]> var_31395 = const()[name = tensor<string, []>("op_31395"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_31397 = const()[name = tensor<string, []>("op_31397"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_31399_pad_type_0 = const()[name = tensor<string, []>("op_31399_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_31399_pad_0 = const()[name = tensor<string, []>("op_31399_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2644824448)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2657931712)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_31399_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_31397, groups = var_30385, pad = var_31399_pad_0, pad_type = var_31399_pad_type_0, strides = var_31395, weight = up_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_447_cast)[name = tensor<string, []>("op_31399_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_211_cast = add(x = var_31399_cast, y = inputs_209_cast)[name = tensor<string, []>("inputs_211_cast")];
            tensor<int32, [1]> var_31409 = const()[name = tensor<string, []>("op_31409"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_211_cast = reduce_mean(axes = var_31409, keep_dims = var_30380, x = inputs_211_cast)[name = tensor<string, []>("channels_mean_211_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_211_cast = sub(x = inputs_211_cast, y = channels_mean_211_cast)[name = tensor<string, []>("zero_mean_211_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_211_cast = mul(x = zero_mean_211_cast, y = zero_mean_211_cast)[name = tensor<string, []>("zero_mean_sq_211_cast")];
            tensor<int32, [1]> var_31413 = const()[name = tensor<string, []>("op_31413"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_31414_cast = reduce_mean(axes = var_31413, keep_dims = var_30380, x = zero_mean_sq_211_cast)[name = tensor<string, []>("op_31414_cast")];
            tensor<fp16, []> var_31415_to_fp16 = const()[name = tensor<string, []>("op_31415_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_31416_cast = add(x = var_31414_cast, y = var_31415_to_fp16)[name = tensor<string, []>("op_31416_cast")];
            tensor<fp16, []> denom_211_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_211_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_211_cast = rsqrt(epsilon = denom_211_epsilon_0_to_fp16, x = var_31416_cast)[name = tensor<string, []>("denom_211_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_211_cast = mul(x = zero_mean_211_cast, y = denom_211_cast)[name = tensor<string, []>("out_211_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_31420_to_fp16 = const()[name = tensor<string, []>("op_31420_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2657934336)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_31421_cast = add(x = out_211_cast, y = var_31420_to_fp16)[name = tensor<string, []>("op_31421_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_31423_to_fp16 = const()[name = tensor<string, []>("op_31423_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2657936960)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_293_cast = mul(x = var_31421_cast, y = var_31423_to_fp16)[name = tensor<string, []>("hidden_states_293_cast")];
            tensor<int32, [2]> var_31430 = const()[name = tensor<string, []>("op_31430"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_31432 = const()[name = tensor<string, []>("op_31432"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_141_pad_type_0 = const()[name = tensor<string, []>("q_141_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_141_pad_0 = const()[name = tensor<string, []>("q_141_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2657939584)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_141_cast = conv(dilations = var_31432, groups = var_30385, pad = q_141_pad_0, pad_type = q_141_pad_type_0, strides = var_31430, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_293_cast)[name = tensor<string, []>("q_141_cast")];
            tensor<int32, [2]> var_31436 = const()[name = tensor<string, []>("op_31436"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_31438 = const()[name = tensor<string, []>("op_31438"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_281_pad_type_0 = const()[name = tensor<string, []>("k_281_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_281_pad_0 = const()[name = tensor<string, []>("k_281_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2661216448)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_281_cast = conv(dilations = var_31438, groups = var_30385, pad = k_281_pad_0, pad_type = k_281_pad_type_0, strides = var_31436, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_293_cast)[name = tensor<string, []>("k_281_cast")];
            tensor<int32, [2]> var_31442 = const()[name = tensor<string, []>("op_31442"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_31444 = const()[name = tensor<string, []>("op_31444"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_141_pad_type_0 = const()[name = tensor<string, []>("v_141_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_141_pad_0 = const()[name = tensor<string, []>("v_141_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2664493312)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_141_cast = conv(dilations = var_31444, groups = var_30385, pad = v_141_pad_0, pad_type = v_141_pad_type_0, strides = var_31442, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_293_cast)[name = tensor<string, []>("v_141_cast")];
            tensor<int32, [4]> var_31448_begin_0 = const()[name = tensor<string, []>("op_31448_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_31448_end_0 = const()[name = tensor<string, []>("op_31448_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_31448_end_mask_0 = const()[name = tensor<string, []>("op_31448_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31448_cast = slice_by_index(begin = var_31448_begin_0, end = var_31448_end_0, end_mask = var_31448_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31448_cast")];
            tensor<int32, [4]> var_31452_begin_0 = const()[name = tensor<string, []>("op_31452_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_31452_end_0 = const()[name = tensor<string, []>("op_31452_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_31452_end_mask_0 = const()[name = tensor<string, []>("op_31452_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31452_cast = slice_by_index(begin = var_31452_begin_0, end = var_31452_end_0, end_mask = var_31452_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31452_cast")];
            tensor<int32, [4]> var_31456_begin_0 = const()[name = tensor<string, []>("op_31456_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_31456_end_0 = const()[name = tensor<string, []>("op_31456_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_31456_end_mask_0 = const()[name = tensor<string, []>("op_31456_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31456_cast = slice_by_index(begin = var_31456_begin_0, end = var_31456_end_0, end_mask = var_31456_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31456_cast")];
            tensor<int32, [4]> var_31460_begin_0 = const()[name = tensor<string, []>("op_31460_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_31460_end_0 = const()[name = tensor<string, []>("op_31460_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_31460_end_mask_0 = const()[name = tensor<string, []>("op_31460_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31460_cast = slice_by_index(begin = var_31460_begin_0, end = var_31460_end_0, end_mask = var_31460_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31460_cast")];
            tensor<int32, [4]> var_31464_begin_0 = const()[name = tensor<string, []>("op_31464_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_31464_end_0 = const()[name = tensor<string, []>("op_31464_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_31464_end_mask_0 = const()[name = tensor<string, []>("op_31464_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31464_cast = slice_by_index(begin = var_31464_begin_0, end = var_31464_end_0, end_mask = var_31464_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31464_cast")];
            tensor<int32, [4]> var_31468_begin_0 = const()[name = tensor<string, []>("op_31468_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_31468_end_0 = const()[name = tensor<string, []>("op_31468_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_31468_end_mask_0 = const()[name = tensor<string, []>("op_31468_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31468_cast = slice_by_index(begin = var_31468_begin_0, end = var_31468_end_0, end_mask = var_31468_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31468_cast")];
            tensor<int32, [4]> var_31472_begin_0 = const()[name = tensor<string, []>("op_31472_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_31472_end_0 = const()[name = tensor<string, []>("op_31472_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_31472_end_mask_0 = const()[name = tensor<string, []>("op_31472_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31472_cast = slice_by_index(begin = var_31472_begin_0, end = var_31472_end_0, end_mask = var_31472_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31472_cast")];
            tensor<int32, [4]> var_31476_begin_0 = const()[name = tensor<string, []>("op_31476_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_31476_end_0 = const()[name = tensor<string, []>("op_31476_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_31476_end_mask_0 = const()[name = tensor<string, []>("op_31476_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31476_cast = slice_by_index(begin = var_31476_begin_0, end = var_31476_end_0, end_mask = var_31476_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31476_cast")];
            tensor<int32, [4]> var_31480_begin_0 = const()[name = tensor<string, []>("op_31480_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_31480_end_0 = const()[name = tensor<string, []>("op_31480_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_31480_end_mask_0 = const()[name = tensor<string, []>("op_31480_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31480_cast = slice_by_index(begin = var_31480_begin_0, end = var_31480_end_0, end_mask = var_31480_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31480_cast")];
            tensor<int32, [4]> var_31484_begin_0 = const()[name = tensor<string, []>("op_31484_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_31484_end_0 = const()[name = tensor<string, []>("op_31484_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_31484_end_mask_0 = const()[name = tensor<string, []>("op_31484_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31484_cast = slice_by_index(begin = var_31484_begin_0, end = var_31484_end_0, end_mask = var_31484_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31484_cast")];
            tensor<int32, [4]> var_31488_begin_0 = const()[name = tensor<string, []>("op_31488_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_31488_end_0 = const()[name = tensor<string, []>("op_31488_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_31488_end_mask_0 = const()[name = tensor<string, []>("op_31488_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31488_cast = slice_by_index(begin = var_31488_begin_0, end = var_31488_end_0, end_mask = var_31488_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31488_cast")];
            tensor<int32, [4]> var_31492_begin_0 = const()[name = tensor<string, []>("op_31492_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_31492_end_0 = const()[name = tensor<string, []>("op_31492_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_31492_end_mask_0 = const()[name = tensor<string, []>("op_31492_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31492_cast = slice_by_index(begin = var_31492_begin_0, end = var_31492_end_0, end_mask = var_31492_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31492_cast")];
            tensor<int32, [4]> var_31496_begin_0 = const()[name = tensor<string, []>("op_31496_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_31496_end_0 = const()[name = tensor<string, []>("op_31496_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_31496_end_mask_0 = const()[name = tensor<string, []>("op_31496_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31496_cast = slice_by_index(begin = var_31496_begin_0, end = var_31496_end_0, end_mask = var_31496_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31496_cast")];
            tensor<int32, [4]> var_31500_begin_0 = const()[name = tensor<string, []>("op_31500_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_31500_end_0 = const()[name = tensor<string, []>("op_31500_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_31500_end_mask_0 = const()[name = tensor<string, []>("op_31500_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31500_cast = slice_by_index(begin = var_31500_begin_0, end = var_31500_end_0, end_mask = var_31500_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31500_cast")];
            tensor<int32, [4]> var_31504_begin_0 = const()[name = tensor<string, []>("op_31504_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_31504_end_0 = const()[name = tensor<string, []>("op_31504_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_31504_end_mask_0 = const()[name = tensor<string, []>("op_31504_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31504_cast = slice_by_index(begin = var_31504_begin_0, end = var_31504_end_0, end_mask = var_31504_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31504_cast")];
            tensor<int32, [4]> var_31508_begin_0 = const()[name = tensor<string, []>("op_31508_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_31508_end_0 = const()[name = tensor<string, []>("op_31508_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_31508_end_mask_0 = const()[name = tensor<string, []>("op_31508_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31508_cast = slice_by_index(begin = var_31508_begin_0, end = var_31508_end_0, end_mask = var_31508_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31508_cast")];
            tensor<int32, [4]> var_31512_begin_0 = const()[name = tensor<string, []>("op_31512_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_31512_end_0 = const()[name = tensor<string, []>("op_31512_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_31512_end_mask_0 = const()[name = tensor<string, []>("op_31512_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31512_cast = slice_by_index(begin = var_31512_begin_0, end = var_31512_end_0, end_mask = var_31512_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31512_cast")];
            tensor<int32, [4]> var_31516_begin_0 = const()[name = tensor<string, []>("op_31516_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_31516_end_0 = const()[name = tensor<string, []>("op_31516_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_31516_end_mask_0 = const()[name = tensor<string, []>("op_31516_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31516_cast = slice_by_index(begin = var_31516_begin_0, end = var_31516_end_0, end_mask = var_31516_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31516_cast")];
            tensor<int32, [4]> var_31520_begin_0 = const()[name = tensor<string, []>("op_31520_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_31520_end_0 = const()[name = tensor<string, []>("op_31520_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_31520_end_mask_0 = const()[name = tensor<string, []>("op_31520_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31520_cast = slice_by_index(begin = var_31520_begin_0, end = var_31520_end_0, end_mask = var_31520_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31520_cast")];
            tensor<int32, [4]> var_31524_begin_0 = const()[name = tensor<string, []>("op_31524_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_31524_end_0 = const()[name = tensor<string, []>("op_31524_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_31524_end_mask_0 = const()[name = tensor<string, []>("op_31524_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31524_cast = slice_by_index(begin = var_31524_begin_0, end = var_31524_end_0, end_mask = var_31524_end_mask_0, x = q_141_cast)[name = tensor<string, []>("op_31524_cast")];
            tensor<int32, [4]> k_283_perm_0 = const()[name = tensor<string, []>("k_283_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_31531_begin_0 = const()[name = tensor<string, []>("op_31531_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_31531_end_0 = const()[name = tensor<string, []>("op_31531_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_31531_end_mask_0 = const()[name = tensor<string, []>("op_31531_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_69 = transpose(perm = k_283_perm_0, x = k_281_cast)[name = tensor<string, []>("transpose_69")];
            tensor<fp16, [2, 1024, 1, 64]> var_31531_cast = slice_by_index(begin = var_31531_begin_0, end = var_31531_end_0, end_mask = var_31531_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31531_cast")];
            tensor<int32, [4]> var_31535_begin_0 = const()[name = tensor<string, []>("op_31535_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_31535_end_0 = const()[name = tensor<string, []>("op_31535_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_31535_end_mask_0 = const()[name = tensor<string, []>("op_31535_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31535_cast = slice_by_index(begin = var_31535_begin_0, end = var_31535_end_0, end_mask = var_31535_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31535_cast")];
            tensor<int32, [4]> var_31539_begin_0 = const()[name = tensor<string, []>("op_31539_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_31539_end_0 = const()[name = tensor<string, []>("op_31539_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_31539_end_mask_0 = const()[name = tensor<string, []>("op_31539_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31539_cast = slice_by_index(begin = var_31539_begin_0, end = var_31539_end_0, end_mask = var_31539_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31539_cast")];
            tensor<int32, [4]> var_31543_begin_0 = const()[name = tensor<string, []>("op_31543_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_31543_end_0 = const()[name = tensor<string, []>("op_31543_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_31543_end_mask_0 = const()[name = tensor<string, []>("op_31543_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31543_cast = slice_by_index(begin = var_31543_begin_0, end = var_31543_end_0, end_mask = var_31543_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31543_cast")];
            tensor<int32, [4]> var_31547_begin_0 = const()[name = tensor<string, []>("op_31547_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_31547_end_0 = const()[name = tensor<string, []>("op_31547_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_31547_end_mask_0 = const()[name = tensor<string, []>("op_31547_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31547_cast = slice_by_index(begin = var_31547_begin_0, end = var_31547_end_0, end_mask = var_31547_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31547_cast")];
            tensor<int32, [4]> var_31551_begin_0 = const()[name = tensor<string, []>("op_31551_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_31551_end_0 = const()[name = tensor<string, []>("op_31551_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_31551_end_mask_0 = const()[name = tensor<string, []>("op_31551_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31551_cast = slice_by_index(begin = var_31551_begin_0, end = var_31551_end_0, end_mask = var_31551_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31551_cast")];
            tensor<int32, [4]> var_31555_begin_0 = const()[name = tensor<string, []>("op_31555_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_31555_end_0 = const()[name = tensor<string, []>("op_31555_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_31555_end_mask_0 = const()[name = tensor<string, []>("op_31555_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31555_cast = slice_by_index(begin = var_31555_begin_0, end = var_31555_end_0, end_mask = var_31555_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31555_cast")];
            tensor<int32, [4]> var_31559_begin_0 = const()[name = tensor<string, []>("op_31559_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_31559_end_0 = const()[name = tensor<string, []>("op_31559_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_31559_end_mask_0 = const()[name = tensor<string, []>("op_31559_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31559_cast = slice_by_index(begin = var_31559_begin_0, end = var_31559_end_0, end_mask = var_31559_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31559_cast")];
            tensor<int32, [4]> var_31563_begin_0 = const()[name = tensor<string, []>("op_31563_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_31563_end_0 = const()[name = tensor<string, []>("op_31563_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_31563_end_mask_0 = const()[name = tensor<string, []>("op_31563_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31563_cast = slice_by_index(begin = var_31563_begin_0, end = var_31563_end_0, end_mask = var_31563_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31563_cast")];
            tensor<int32, [4]> var_31567_begin_0 = const()[name = tensor<string, []>("op_31567_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_31567_end_0 = const()[name = tensor<string, []>("op_31567_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_31567_end_mask_0 = const()[name = tensor<string, []>("op_31567_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31567_cast = slice_by_index(begin = var_31567_begin_0, end = var_31567_end_0, end_mask = var_31567_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31567_cast")];
            tensor<int32, [4]> var_31571_begin_0 = const()[name = tensor<string, []>("op_31571_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_31571_end_0 = const()[name = tensor<string, []>("op_31571_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_31571_end_mask_0 = const()[name = tensor<string, []>("op_31571_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31571_cast = slice_by_index(begin = var_31571_begin_0, end = var_31571_end_0, end_mask = var_31571_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31571_cast")];
            tensor<int32, [4]> var_31575_begin_0 = const()[name = tensor<string, []>("op_31575_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_31575_end_0 = const()[name = tensor<string, []>("op_31575_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_31575_end_mask_0 = const()[name = tensor<string, []>("op_31575_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31575_cast = slice_by_index(begin = var_31575_begin_0, end = var_31575_end_0, end_mask = var_31575_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31575_cast")];
            tensor<int32, [4]> var_31579_begin_0 = const()[name = tensor<string, []>("op_31579_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_31579_end_0 = const()[name = tensor<string, []>("op_31579_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_31579_end_mask_0 = const()[name = tensor<string, []>("op_31579_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31579_cast = slice_by_index(begin = var_31579_begin_0, end = var_31579_end_0, end_mask = var_31579_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31579_cast")];
            tensor<int32, [4]> var_31583_begin_0 = const()[name = tensor<string, []>("op_31583_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_31583_end_0 = const()[name = tensor<string, []>("op_31583_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_31583_end_mask_0 = const()[name = tensor<string, []>("op_31583_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31583_cast = slice_by_index(begin = var_31583_begin_0, end = var_31583_end_0, end_mask = var_31583_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31583_cast")];
            tensor<int32, [4]> var_31587_begin_0 = const()[name = tensor<string, []>("op_31587_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_31587_end_0 = const()[name = tensor<string, []>("op_31587_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_31587_end_mask_0 = const()[name = tensor<string, []>("op_31587_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31587_cast = slice_by_index(begin = var_31587_begin_0, end = var_31587_end_0, end_mask = var_31587_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31587_cast")];
            tensor<int32, [4]> var_31591_begin_0 = const()[name = tensor<string, []>("op_31591_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_31591_end_0 = const()[name = tensor<string, []>("op_31591_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_31591_end_mask_0 = const()[name = tensor<string, []>("op_31591_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31591_cast = slice_by_index(begin = var_31591_begin_0, end = var_31591_end_0, end_mask = var_31591_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31591_cast")];
            tensor<int32, [4]> var_31595_begin_0 = const()[name = tensor<string, []>("op_31595_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_31595_end_0 = const()[name = tensor<string, []>("op_31595_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_31595_end_mask_0 = const()[name = tensor<string, []>("op_31595_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31595_cast = slice_by_index(begin = var_31595_begin_0, end = var_31595_end_0, end_mask = var_31595_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31595_cast")];
            tensor<int32, [4]> var_31599_begin_0 = const()[name = tensor<string, []>("op_31599_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_31599_end_0 = const()[name = tensor<string, []>("op_31599_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_31599_end_mask_0 = const()[name = tensor<string, []>("op_31599_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31599_cast = slice_by_index(begin = var_31599_begin_0, end = var_31599_end_0, end_mask = var_31599_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31599_cast")];
            tensor<int32, [4]> var_31603_begin_0 = const()[name = tensor<string, []>("op_31603_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_31603_end_0 = const()[name = tensor<string, []>("op_31603_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_31603_end_mask_0 = const()[name = tensor<string, []>("op_31603_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31603_cast = slice_by_index(begin = var_31603_begin_0, end = var_31603_end_0, end_mask = var_31603_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31603_cast")];
            tensor<int32, [4]> var_31607_begin_0 = const()[name = tensor<string, []>("op_31607_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_31607_end_0 = const()[name = tensor<string, []>("op_31607_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_31607_end_mask_0 = const()[name = tensor<string, []>("op_31607_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_31607_cast = slice_by_index(begin = var_31607_begin_0, end = var_31607_end_0, end_mask = var_31607_end_mask_0, x = transpose_69)[name = tensor<string, []>("op_31607_cast")];
            tensor<int32, [4]> var_31609_begin_0 = const()[name = tensor<string, []>("op_31609_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_31609_end_0 = const()[name = tensor<string, []>("op_31609_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_31609_end_mask_0 = const()[name = tensor<string, []>("op_31609_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31609_cast = slice_by_index(begin = var_31609_begin_0, end = var_31609_end_0, end_mask = var_31609_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31609_cast")];
            tensor<int32, [4]> var_31613_begin_0 = const()[name = tensor<string, []>("op_31613_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_31613_end_0 = const()[name = tensor<string, []>("op_31613_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_31613_end_mask_0 = const()[name = tensor<string, []>("op_31613_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31613_cast = slice_by_index(begin = var_31613_begin_0, end = var_31613_end_0, end_mask = var_31613_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31613_cast")];
            tensor<int32, [4]> var_31617_begin_0 = const()[name = tensor<string, []>("op_31617_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_31617_end_0 = const()[name = tensor<string, []>("op_31617_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_31617_end_mask_0 = const()[name = tensor<string, []>("op_31617_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31617_cast = slice_by_index(begin = var_31617_begin_0, end = var_31617_end_0, end_mask = var_31617_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31617_cast")];
            tensor<int32, [4]> var_31621_begin_0 = const()[name = tensor<string, []>("op_31621_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_31621_end_0 = const()[name = tensor<string, []>("op_31621_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_31621_end_mask_0 = const()[name = tensor<string, []>("op_31621_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31621_cast = slice_by_index(begin = var_31621_begin_0, end = var_31621_end_0, end_mask = var_31621_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31621_cast")];
            tensor<int32, [4]> var_31625_begin_0 = const()[name = tensor<string, []>("op_31625_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_31625_end_0 = const()[name = tensor<string, []>("op_31625_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_31625_end_mask_0 = const()[name = tensor<string, []>("op_31625_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31625_cast = slice_by_index(begin = var_31625_begin_0, end = var_31625_end_0, end_mask = var_31625_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31625_cast")];
            tensor<int32, [4]> var_31629_begin_0 = const()[name = tensor<string, []>("op_31629_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_31629_end_0 = const()[name = tensor<string, []>("op_31629_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_31629_end_mask_0 = const()[name = tensor<string, []>("op_31629_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31629_cast = slice_by_index(begin = var_31629_begin_0, end = var_31629_end_0, end_mask = var_31629_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31629_cast")];
            tensor<int32, [4]> var_31633_begin_0 = const()[name = tensor<string, []>("op_31633_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_31633_end_0 = const()[name = tensor<string, []>("op_31633_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_31633_end_mask_0 = const()[name = tensor<string, []>("op_31633_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31633_cast = slice_by_index(begin = var_31633_begin_0, end = var_31633_end_0, end_mask = var_31633_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31633_cast")];
            tensor<int32, [4]> var_31637_begin_0 = const()[name = tensor<string, []>("op_31637_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_31637_end_0 = const()[name = tensor<string, []>("op_31637_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_31637_end_mask_0 = const()[name = tensor<string, []>("op_31637_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31637_cast = slice_by_index(begin = var_31637_begin_0, end = var_31637_end_0, end_mask = var_31637_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31637_cast")];
            tensor<int32, [4]> var_31641_begin_0 = const()[name = tensor<string, []>("op_31641_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_31641_end_0 = const()[name = tensor<string, []>("op_31641_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_31641_end_mask_0 = const()[name = tensor<string, []>("op_31641_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31641_cast = slice_by_index(begin = var_31641_begin_0, end = var_31641_end_0, end_mask = var_31641_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31641_cast")];
            tensor<int32, [4]> var_31645_begin_0 = const()[name = tensor<string, []>("op_31645_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_31645_end_0 = const()[name = tensor<string, []>("op_31645_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_31645_end_mask_0 = const()[name = tensor<string, []>("op_31645_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31645_cast = slice_by_index(begin = var_31645_begin_0, end = var_31645_end_0, end_mask = var_31645_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31645_cast")];
            tensor<int32, [4]> var_31649_begin_0 = const()[name = tensor<string, []>("op_31649_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_31649_end_0 = const()[name = tensor<string, []>("op_31649_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_31649_end_mask_0 = const()[name = tensor<string, []>("op_31649_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31649_cast = slice_by_index(begin = var_31649_begin_0, end = var_31649_end_0, end_mask = var_31649_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31649_cast")];
            tensor<int32, [4]> var_31653_begin_0 = const()[name = tensor<string, []>("op_31653_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_31653_end_0 = const()[name = tensor<string, []>("op_31653_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_31653_end_mask_0 = const()[name = tensor<string, []>("op_31653_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31653_cast = slice_by_index(begin = var_31653_begin_0, end = var_31653_end_0, end_mask = var_31653_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31653_cast")];
            tensor<int32, [4]> var_31657_begin_0 = const()[name = tensor<string, []>("op_31657_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_31657_end_0 = const()[name = tensor<string, []>("op_31657_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_31657_end_mask_0 = const()[name = tensor<string, []>("op_31657_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31657_cast = slice_by_index(begin = var_31657_begin_0, end = var_31657_end_0, end_mask = var_31657_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31657_cast")];
            tensor<int32, [4]> var_31661_begin_0 = const()[name = tensor<string, []>("op_31661_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_31661_end_0 = const()[name = tensor<string, []>("op_31661_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_31661_end_mask_0 = const()[name = tensor<string, []>("op_31661_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31661_cast = slice_by_index(begin = var_31661_begin_0, end = var_31661_end_0, end_mask = var_31661_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31661_cast")];
            tensor<int32, [4]> var_31665_begin_0 = const()[name = tensor<string, []>("op_31665_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_31665_end_0 = const()[name = tensor<string, []>("op_31665_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_31665_end_mask_0 = const()[name = tensor<string, []>("op_31665_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31665_cast = slice_by_index(begin = var_31665_begin_0, end = var_31665_end_0, end_mask = var_31665_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31665_cast")];
            tensor<int32, [4]> var_31669_begin_0 = const()[name = tensor<string, []>("op_31669_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_31669_end_0 = const()[name = tensor<string, []>("op_31669_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_31669_end_mask_0 = const()[name = tensor<string, []>("op_31669_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31669_cast = slice_by_index(begin = var_31669_begin_0, end = var_31669_end_0, end_mask = var_31669_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31669_cast")];
            tensor<int32, [4]> var_31673_begin_0 = const()[name = tensor<string, []>("op_31673_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_31673_end_0 = const()[name = tensor<string, []>("op_31673_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_31673_end_mask_0 = const()[name = tensor<string, []>("op_31673_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31673_cast = slice_by_index(begin = var_31673_begin_0, end = var_31673_end_0, end_mask = var_31673_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31673_cast")];
            tensor<int32, [4]> var_31677_begin_0 = const()[name = tensor<string, []>("op_31677_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_31677_end_0 = const()[name = tensor<string, []>("op_31677_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_31677_end_mask_0 = const()[name = tensor<string, []>("op_31677_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31677_cast = slice_by_index(begin = var_31677_begin_0, end = var_31677_end_0, end_mask = var_31677_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31677_cast")];
            tensor<int32, [4]> var_31681_begin_0 = const()[name = tensor<string, []>("op_31681_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_31681_end_0 = const()[name = tensor<string, []>("op_31681_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_31681_end_mask_0 = const()[name = tensor<string, []>("op_31681_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31681_cast = slice_by_index(begin = var_31681_begin_0, end = var_31681_end_0, end_mask = var_31681_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31681_cast")];
            tensor<int32, [4]> var_31685_begin_0 = const()[name = tensor<string, []>("op_31685_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_31685_end_0 = const()[name = tensor<string, []>("op_31685_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_31685_end_mask_0 = const()[name = tensor<string, []>("op_31685_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31685_cast = slice_by_index(begin = var_31685_begin_0, end = var_31685_end_0, end_mask = var_31685_end_mask_0, x = v_141_cast)[name = tensor<string, []>("op_31685_cast")];
            tensor<string, []> var_31689_equation_0 = const()[name = tensor<string, []>("op_31689_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31689_cast = einsum(equation = var_31689_equation_0, values = (var_31531_cast, var_31448_cast))[name = tensor<string, []>("op_31689_cast")];
            tensor<fp16, []> var_31690_to_fp16 = const()[name = tensor<string, []>("op_31690_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2641_cast = mul(x = var_31689_cast, y = var_31690_to_fp16)[name = tensor<string, []>("aw_2641_cast")];
            tensor<string, []> var_31693_equation_0 = const()[name = tensor<string, []>("op_31693_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31693_cast = einsum(equation = var_31693_equation_0, values = (var_31535_cast, var_31452_cast))[name = tensor<string, []>("op_31693_cast")];
            tensor<fp16, []> var_31694_to_fp16 = const()[name = tensor<string, []>("op_31694_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2643_cast = mul(x = var_31693_cast, y = var_31694_to_fp16)[name = tensor<string, []>("aw_2643_cast")];
            tensor<string, []> var_31697_equation_0 = const()[name = tensor<string, []>("op_31697_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31697_cast = einsum(equation = var_31697_equation_0, values = (var_31539_cast, var_31456_cast))[name = tensor<string, []>("op_31697_cast")];
            tensor<fp16, []> var_31698_to_fp16 = const()[name = tensor<string, []>("op_31698_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2645_cast = mul(x = var_31697_cast, y = var_31698_to_fp16)[name = tensor<string, []>("aw_2645_cast")];
            tensor<string, []> var_31701_equation_0 = const()[name = tensor<string, []>("op_31701_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31701_cast = einsum(equation = var_31701_equation_0, values = (var_31543_cast, var_31460_cast))[name = tensor<string, []>("op_31701_cast")];
            tensor<fp16, []> var_31702_to_fp16 = const()[name = tensor<string, []>("op_31702_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2647_cast = mul(x = var_31701_cast, y = var_31702_to_fp16)[name = tensor<string, []>("aw_2647_cast")];
            tensor<string, []> var_31705_equation_0 = const()[name = tensor<string, []>("op_31705_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31705_cast = einsum(equation = var_31705_equation_0, values = (var_31547_cast, var_31464_cast))[name = tensor<string, []>("op_31705_cast")];
            tensor<fp16, []> var_31706_to_fp16 = const()[name = tensor<string, []>("op_31706_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2649_cast = mul(x = var_31705_cast, y = var_31706_to_fp16)[name = tensor<string, []>("aw_2649_cast")];
            tensor<string, []> var_31709_equation_0 = const()[name = tensor<string, []>("op_31709_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31709_cast = einsum(equation = var_31709_equation_0, values = (var_31551_cast, var_31468_cast))[name = tensor<string, []>("op_31709_cast")];
            tensor<fp16, []> var_31710_to_fp16 = const()[name = tensor<string, []>("op_31710_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2651_cast = mul(x = var_31709_cast, y = var_31710_to_fp16)[name = tensor<string, []>("aw_2651_cast")];
            tensor<string, []> var_31713_equation_0 = const()[name = tensor<string, []>("op_31713_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31713_cast = einsum(equation = var_31713_equation_0, values = (var_31555_cast, var_31472_cast))[name = tensor<string, []>("op_31713_cast")];
            tensor<fp16, []> var_31714_to_fp16 = const()[name = tensor<string, []>("op_31714_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2653_cast = mul(x = var_31713_cast, y = var_31714_to_fp16)[name = tensor<string, []>("aw_2653_cast")];
            tensor<string, []> var_31717_equation_0 = const()[name = tensor<string, []>("op_31717_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31717_cast = einsum(equation = var_31717_equation_0, values = (var_31559_cast, var_31476_cast))[name = tensor<string, []>("op_31717_cast")];
            tensor<fp16, []> var_31718_to_fp16 = const()[name = tensor<string, []>("op_31718_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2655_cast = mul(x = var_31717_cast, y = var_31718_to_fp16)[name = tensor<string, []>("aw_2655_cast")];
            tensor<string, []> var_31721_equation_0 = const()[name = tensor<string, []>("op_31721_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31721_cast = einsum(equation = var_31721_equation_0, values = (var_31563_cast, var_31480_cast))[name = tensor<string, []>("op_31721_cast")];
            tensor<fp16, []> var_31722_to_fp16 = const()[name = tensor<string, []>("op_31722_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2657_cast = mul(x = var_31721_cast, y = var_31722_to_fp16)[name = tensor<string, []>("aw_2657_cast")];
            tensor<string, []> var_31725_equation_0 = const()[name = tensor<string, []>("op_31725_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31725_cast = einsum(equation = var_31725_equation_0, values = (var_31567_cast, var_31484_cast))[name = tensor<string, []>("op_31725_cast")];
            tensor<fp16, []> var_31726_to_fp16 = const()[name = tensor<string, []>("op_31726_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2659_cast = mul(x = var_31725_cast, y = var_31726_to_fp16)[name = tensor<string, []>("aw_2659_cast")];
            tensor<string, []> var_31729_equation_0 = const()[name = tensor<string, []>("op_31729_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31729_cast = einsum(equation = var_31729_equation_0, values = (var_31571_cast, var_31488_cast))[name = tensor<string, []>("op_31729_cast")];
            tensor<fp16, []> var_31730_to_fp16 = const()[name = tensor<string, []>("op_31730_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2661_cast = mul(x = var_31729_cast, y = var_31730_to_fp16)[name = tensor<string, []>("aw_2661_cast")];
            tensor<string, []> var_31733_equation_0 = const()[name = tensor<string, []>("op_31733_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31733_cast = einsum(equation = var_31733_equation_0, values = (var_31575_cast, var_31492_cast))[name = tensor<string, []>("op_31733_cast")];
            tensor<fp16, []> var_31734_to_fp16 = const()[name = tensor<string, []>("op_31734_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2663_cast = mul(x = var_31733_cast, y = var_31734_to_fp16)[name = tensor<string, []>("aw_2663_cast")];
            tensor<string, []> var_31737_equation_0 = const()[name = tensor<string, []>("op_31737_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31737_cast = einsum(equation = var_31737_equation_0, values = (var_31579_cast, var_31496_cast))[name = tensor<string, []>("op_31737_cast")];
            tensor<fp16, []> var_31738_to_fp16 = const()[name = tensor<string, []>("op_31738_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2665_cast = mul(x = var_31737_cast, y = var_31738_to_fp16)[name = tensor<string, []>("aw_2665_cast")];
            tensor<string, []> var_31741_equation_0 = const()[name = tensor<string, []>("op_31741_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31741_cast = einsum(equation = var_31741_equation_0, values = (var_31583_cast, var_31500_cast))[name = tensor<string, []>("op_31741_cast")];
            tensor<fp16, []> var_31742_to_fp16 = const()[name = tensor<string, []>("op_31742_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2667_cast = mul(x = var_31741_cast, y = var_31742_to_fp16)[name = tensor<string, []>("aw_2667_cast")];
            tensor<string, []> var_31745_equation_0 = const()[name = tensor<string, []>("op_31745_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31745_cast = einsum(equation = var_31745_equation_0, values = (var_31587_cast, var_31504_cast))[name = tensor<string, []>("op_31745_cast")];
            tensor<fp16, []> var_31746_to_fp16 = const()[name = tensor<string, []>("op_31746_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2669_cast = mul(x = var_31745_cast, y = var_31746_to_fp16)[name = tensor<string, []>("aw_2669_cast")];
            tensor<string, []> var_31749_equation_0 = const()[name = tensor<string, []>("op_31749_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31749_cast = einsum(equation = var_31749_equation_0, values = (var_31591_cast, var_31508_cast))[name = tensor<string, []>("op_31749_cast")];
            tensor<fp16, []> var_31750_to_fp16 = const()[name = tensor<string, []>("op_31750_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2671_cast = mul(x = var_31749_cast, y = var_31750_to_fp16)[name = tensor<string, []>("aw_2671_cast")];
            tensor<string, []> var_31753_equation_0 = const()[name = tensor<string, []>("op_31753_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31753_cast = einsum(equation = var_31753_equation_0, values = (var_31595_cast, var_31512_cast))[name = tensor<string, []>("op_31753_cast")];
            tensor<fp16, []> var_31754_to_fp16 = const()[name = tensor<string, []>("op_31754_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2673_cast = mul(x = var_31753_cast, y = var_31754_to_fp16)[name = tensor<string, []>("aw_2673_cast")];
            tensor<string, []> var_31757_equation_0 = const()[name = tensor<string, []>("op_31757_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31757_cast = einsum(equation = var_31757_equation_0, values = (var_31599_cast, var_31516_cast))[name = tensor<string, []>("op_31757_cast")];
            tensor<fp16, []> var_31758_to_fp16 = const()[name = tensor<string, []>("op_31758_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2675_cast = mul(x = var_31757_cast, y = var_31758_to_fp16)[name = tensor<string, []>("aw_2675_cast")];
            tensor<string, []> var_31761_equation_0 = const()[name = tensor<string, []>("op_31761_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31761_cast = einsum(equation = var_31761_equation_0, values = (var_31603_cast, var_31520_cast))[name = tensor<string, []>("op_31761_cast")];
            tensor<fp16, []> var_31762_to_fp16 = const()[name = tensor<string, []>("op_31762_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2677_cast = mul(x = var_31761_cast, y = var_31762_to_fp16)[name = tensor<string, []>("aw_2677_cast")];
            tensor<string, []> var_31765_equation_0 = const()[name = tensor<string, []>("op_31765_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31765_cast = einsum(equation = var_31765_equation_0, values = (var_31607_cast, var_31524_cast))[name = tensor<string, []>("op_31765_cast")];
            tensor<fp16, []> var_31766_to_fp16 = const()[name = tensor<string, []>("op_31766_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2679_cast = mul(x = var_31765_cast, y = var_31766_to_fp16)[name = tensor<string, []>("aw_2679_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31768_cast = softmax(axis = var_30385, x = aw_2641_cast)[name = tensor<string, []>("op_31768_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31769_cast = softmax(axis = var_30385, x = aw_2643_cast)[name = tensor<string, []>("op_31769_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31770_cast = softmax(axis = var_30385, x = aw_2645_cast)[name = tensor<string, []>("op_31770_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31771_cast = softmax(axis = var_30385, x = aw_2647_cast)[name = tensor<string, []>("op_31771_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31772_cast = softmax(axis = var_30385, x = aw_2649_cast)[name = tensor<string, []>("op_31772_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31773_cast = softmax(axis = var_30385, x = aw_2651_cast)[name = tensor<string, []>("op_31773_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31774_cast = softmax(axis = var_30385, x = aw_2653_cast)[name = tensor<string, []>("op_31774_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31775_cast = softmax(axis = var_30385, x = aw_2655_cast)[name = tensor<string, []>("op_31775_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31776_cast = softmax(axis = var_30385, x = aw_2657_cast)[name = tensor<string, []>("op_31776_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31777_cast = softmax(axis = var_30385, x = aw_2659_cast)[name = tensor<string, []>("op_31777_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31778_cast = softmax(axis = var_30385, x = aw_2661_cast)[name = tensor<string, []>("op_31778_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31779_cast = softmax(axis = var_30385, x = aw_2663_cast)[name = tensor<string, []>("op_31779_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31780_cast = softmax(axis = var_30385, x = aw_2665_cast)[name = tensor<string, []>("op_31780_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31781_cast = softmax(axis = var_30385, x = aw_2667_cast)[name = tensor<string, []>("op_31781_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31782_cast = softmax(axis = var_30385, x = aw_2669_cast)[name = tensor<string, []>("op_31782_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31783_cast = softmax(axis = var_30385, x = aw_2671_cast)[name = tensor<string, []>("op_31783_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31784_cast = softmax(axis = var_30385, x = aw_2673_cast)[name = tensor<string, []>("op_31784_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31785_cast = softmax(axis = var_30385, x = aw_2675_cast)[name = tensor<string, []>("op_31785_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31786_cast = softmax(axis = var_30385, x = aw_2677_cast)[name = tensor<string, []>("op_31786_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_31787_cast = softmax(axis = var_30385, x = aw_2679_cast)[name = tensor<string, []>("op_31787_cast")];
            tensor<string, []> var_31789_equation_0 = const()[name = tensor<string, []>("op_31789_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31789_cast = einsum(equation = var_31789_equation_0, values = (var_31609_cast, var_31768_cast))[name = tensor<string, []>("op_31789_cast")];
            tensor<string, []> var_31791_equation_0 = const()[name = tensor<string, []>("op_31791_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31791_cast = einsum(equation = var_31791_equation_0, values = (var_31613_cast, var_31769_cast))[name = tensor<string, []>("op_31791_cast")];
            tensor<string, []> var_31793_equation_0 = const()[name = tensor<string, []>("op_31793_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31793_cast = einsum(equation = var_31793_equation_0, values = (var_31617_cast, var_31770_cast))[name = tensor<string, []>("op_31793_cast")];
            tensor<string, []> var_31795_equation_0 = const()[name = tensor<string, []>("op_31795_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31795_cast = einsum(equation = var_31795_equation_0, values = (var_31621_cast, var_31771_cast))[name = tensor<string, []>("op_31795_cast")];
            tensor<string, []> var_31797_equation_0 = const()[name = tensor<string, []>("op_31797_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31797_cast = einsum(equation = var_31797_equation_0, values = (var_31625_cast, var_31772_cast))[name = tensor<string, []>("op_31797_cast")];
            tensor<string, []> var_31799_equation_0 = const()[name = tensor<string, []>("op_31799_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31799_cast = einsum(equation = var_31799_equation_0, values = (var_31629_cast, var_31773_cast))[name = tensor<string, []>("op_31799_cast")];
            tensor<string, []> var_31801_equation_0 = const()[name = tensor<string, []>("op_31801_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31801_cast = einsum(equation = var_31801_equation_0, values = (var_31633_cast, var_31774_cast))[name = tensor<string, []>("op_31801_cast")];
            tensor<string, []> var_31803_equation_0 = const()[name = tensor<string, []>("op_31803_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31803_cast = einsum(equation = var_31803_equation_0, values = (var_31637_cast, var_31775_cast))[name = tensor<string, []>("op_31803_cast")];
            tensor<string, []> var_31805_equation_0 = const()[name = tensor<string, []>("op_31805_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31805_cast = einsum(equation = var_31805_equation_0, values = (var_31641_cast, var_31776_cast))[name = tensor<string, []>("op_31805_cast")];
            tensor<string, []> var_31807_equation_0 = const()[name = tensor<string, []>("op_31807_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31807_cast = einsum(equation = var_31807_equation_0, values = (var_31645_cast, var_31777_cast))[name = tensor<string, []>("op_31807_cast")];
            tensor<string, []> var_31809_equation_0 = const()[name = tensor<string, []>("op_31809_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31809_cast = einsum(equation = var_31809_equation_0, values = (var_31649_cast, var_31778_cast))[name = tensor<string, []>("op_31809_cast")];
            tensor<string, []> var_31811_equation_0 = const()[name = tensor<string, []>("op_31811_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31811_cast = einsum(equation = var_31811_equation_0, values = (var_31653_cast, var_31779_cast))[name = tensor<string, []>("op_31811_cast")];
            tensor<string, []> var_31813_equation_0 = const()[name = tensor<string, []>("op_31813_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31813_cast = einsum(equation = var_31813_equation_0, values = (var_31657_cast, var_31780_cast))[name = tensor<string, []>("op_31813_cast")];
            tensor<string, []> var_31815_equation_0 = const()[name = tensor<string, []>("op_31815_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31815_cast = einsum(equation = var_31815_equation_0, values = (var_31661_cast, var_31781_cast))[name = tensor<string, []>("op_31815_cast")];
            tensor<string, []> var_31817_equation_0 = const()[name = tensor<string, []>("op_31817_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31817_cast = einsum(equation = var_31817_equation_0, values = (var_31665_cast, var_31782_cast))[name = tensor<string, []>("op_31817_cast")];
            tensor<string, []> var_31819_equation_0 = const()[name = tensor<string, []>("op_31819_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31819_cast = einsum(equation = var_31819_equation_0, values = (var_31669_cast, var_31783_cast))[name = tensor<string, []>("op_31819_cast")];
            tensor<string, []> var_31821_equation_0 = const()[name = tensor<string, []>("op_31821_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31821_cast = einsum(equation = var_31821_equation_0, values = (var_31673_cast, var_31784_cast))[name = tensor<string, []>("op_31821_cast")];
            tensor<string, []> var_31823_equation_0 = const()[name = tensor<string, []>("op_31823_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31823_cast = einsum(equation = var_31823_equation_0, values = (var_31677_cast, var_31785_cast))[name = tensor<string, []>("op_31823_cast")];
            tensor<string, []> var_31825_equation_0 = const()[name = tensor<string, []>("op_31825_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31825_cast = einsum(equation = var_31825_equation_0, values = (var_31681_cast, var_31786_cast))[name = tensor<string, []>("op_31825_cast")];
            tensor<string, []> var_31827_equation_0 = const()[name = tensor<string, []>("op_31827_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_31827_cast = einsum(equation = var_31827_equation_0, values = (var_31685_cast, var_31787_cast))[name = tensor<string, []>("op_31827_cast")];
            tensor<bool, []> input_449_interleave_0 = const()[name = tensor<string, []>("input_449_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_449_cast = concat(axis = var_30385, interleave = input_449_interleave_0, values = (var_31789_cast, var_31791_cast, var_31793_cast, var_31795_cast, var_31797_cast, var_31799_cast, var_31801_cast, var_31803_cast, var_31805_cast, var_31807_cast, var_31809_cast, var_31811_cast, var_31813_cast, var_31815_cast, var_31817_cast, var_31819_cast, var_31821_cast, var_31823_cast, var_31825_cast, var_31827_cast))[name = tensor<string, []>("input_449_cast")];
            tensor<int32, [2]> var_31833 = const()[name = tensor<string, []>("op_31833"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_31835 = const()[name = tensor<string, []>("op_31835"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_31837_pad_type_0 = const()[name = tensor<string, []>("op_31837_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_31837_pad_0 = const()[name = tensor<string, []>("op_31837_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2667770176)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2671047040)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_31837_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_31835, groups = var_30385, pad = var_31837_pad_0, pad_type = var_31837_pad_type_0, strides = var_31833, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_449_cast)[name = tensor<string, []>("op_31837_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_213_cast = add(x = var_31837_cast, y = inputs_211_cast)[name = tensor<string, []>("inputs_213_cast")];
            tensor<int32, [1]> var_31841 = const()[name = tensor<string, []>("op_31841"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_213_cast = reduce_mean(axes = var_31841, keep_dims = var_30380, x = inputs_213_cast)[name = tensor<string, []>("channels_mean_213_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_213_cast = sub(x = inputs_213_cast, y = channels_mean_213_cast)[name = tensor<string, []>("zero_mean_213_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_213_cast = mul(x = zero_mean_213_cast, y = zero_mean_213_cast)[name = tensor<string, []>("zero_mean_sq_213_cast")];
            tensor<int32, [1]> var_31845 = const()[name = tensor<string, []>("op_31845"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_31846_cast = reduce_mean(axes = var_31845, keep_dims = var_30380, x = zero_mean_sq_213_cast)[name = tensor<string, []>("op_31846_cast")];
            tensor<fp16, []> var_31847_to_fp16 = const()[name = tensor<string, []>("op_31847_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_31848_cast = add(x = var_31846_cast, y = var_31847_to_fp16)[name = tensor<string, []>("op_31848_cast")];
            tensor<fp16, []> denom_213_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_213_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_213_cast = rsqrt(epsilon = denom_213_epsilon_0_to_fp16, x = var_31848_cast)[name = tensor<string, []>("denom_213_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_213_cast = mul(x = zero_mean_213_cast, y = denom_213_cast)[name = tensor<string, []>("out_213_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_31852_to_fp16 = const()[name = tensor<string, []>("op_31852_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2671049664)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_31853_cast = add(x = out_213_cast, y = var_31852_to_fp16)[name = tensor<string, []>("op_31853_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_31855_to_fp16 = const()[name = tensor<string, []>("op_31855_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2671052288)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_295_cast = mul(x = var_31853_cast, y = var_31855_to_fp16)[name = tensor<string, []>("hidden_states_295_cast")];
            tensor<int32, [2]> var_31862 = const()[name = tensor<string, []>("op_31862"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_31864 = const()[name = tensor<string, []>("op_31864"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_143_pad_type_0 = const()[name = tensor<string, []>("q_143_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_143_pad_0 = const()[name = tensor<string, []>("q_143_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2671054912)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_143_cast = conv(dilations = var_31864, groups = var_30385, pad = q_143_pad_0, pad_type = q_143_pad_type_0, strides = var_31862, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_295_cast)[name = tensor<string, []>("q_143_cast")];
            tensor<int32, [2]> var_31868 = const()[name = tensor<string, []>("op_31868"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_31870 = const()[name = tensor<string, []>("op_31870"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_285_pad_type_0 = const()[name = tensor<string, []>("k_285_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_285_pad_0 = const()[name = tensor<string, []>("k_285_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2674331776)))];
            tensor<fp16, [2, 1280, 1, 77]> k_285_cast = conv(dilations = var_31870, groups = var_30385, pad = k_285_pad_0, pad_type = k_285_pad_type_0, strides = var_31868, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_285_cast")];
            tensor<int32, [2]> var_31874 = const()[name = tensor<string, []>("op_31874"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_31876 = const()[name = tensor<string, []>("op_31876"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_143_pad_type_0 = const()[name = tensor<string, []>("v_143_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_143_pad_0 = const()[name = tensor<string, []>("v_143_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2679574720)))];
            tensor<fp16, [2, 1280, 1, 77]> v_143_cast = conv(dilations = var_31876, groups = var_30385, pad = v_143_pad_0, pad_type = v_143_pad_type_0, strides = var_31874, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_143_cast")];
            tensor<int32, [4]> var_31880_begin_0 = const()[name = tensor<string, []>("op_31880_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_31880_end_0 = const()[name = tensor<string, []>("op_31880_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_31880_end_mask_0 = const()[name = tensor<string, []>("op_31880_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31880_cast = slice_by_index(begin = var_31880_begin_0, end = var_31880_end_0, end_mask = var_31880_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31880_cast")];
            tensor<int32, [4]> var_31884_begin_0 = const()[name = tensor<string, []>("op_31884_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_31884_end_0 = const()[name = tensor<string, []>("op_31884_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_31884_end_mask_0 = const()[name = tensor<string, []>("op_31884_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31884_cast = slice_by_index(begin = var_31884_begin_0, end = var_31884_end_0, end_mask = var_31884_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31884_cast")];
            tensor<int32, [4]> var_31888_begin_0 = const()[name = tensor<string, []>("op_31888_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_31888_end_0 = const()[name = tensor<string, []>("op_31888_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_31888_end_mask_0 = const()[name = tensor<string, []>("op_31888_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31888_cast = slice_by_index(begin = var_31888_begin_0, end = var_31888_end_0, end_mask = var_31888_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31888_cast")];
            tensor<int32, [4]> var_31892_begin_0 = const()[name = tensor<string, []>("op_31892_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_31892_end_0 = const()[name = tensor<string, []>("op_31892_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_31892_end_mask_0 = const()[name = tensor<string, []>("op_31892_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31892_cast = slice_by_index(begin = var_31892_begin_0, end = var_31892_end_0, end_mask = var_31892_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31892_cast")];
            tensor<int32, [4]> var_31896_begin_0 = const()[name = tensor<string, []>("op_31896_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_31896_end_0 = const()[name = tensor<string, []>("op_31896_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_31896_end_mask_0 = const()[name = tensor<string, []>("op_31896_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31896_cast = slice_by_index(begin = var_31896_begin_0, end = var_31896_end_0, end_mask = var_31896_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31896_cast")];
            tensor<int32, [4]> var_31900_begin_0 = const()[name = tensor<string, []>("op_31900_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_31900_end_0 = const()[name = tensor<string, []>("op_31900_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_31900_end_mask_0 = const()[name = tensor<string, []>("op_31900_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31900_cast = slice_by_index(begin = var_31900_begin_0, end = var_31900_end_0, end_mask = var_31900_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31900_cast")];
            tensor<int32, [4]> var_31904_begin_0 = const()[name = tensor<string, []>("op_31904_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_31904_end_0 = const()[name = tensor<string, []>("op_31904_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_31904_end_mask_0 = const()[name = tensor<string, []>("op_31904_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31904_cast = slice_by_index(begin = var_31904_begin_0, end = var_31904_end_0, end_mask = var_31904_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31904_cast")];
            tensor<int32, [4]> var_31908_begin_0 = const()[name = tensor<string, []>("op_31908_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_31908_end_0 = const()[name = tensor<string, []>("op_31908_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_31908_end_mask_0 = const()[name = tensor<string, []>("op_31908_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31908_cast = slice_by_index(begin = var_31908_begin_0, end = var_31908_end_0, end_mask = var_31908_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31908_cast")];
            tensor<int32, [4]> var_31912_begin_0 = const()[name = tensor<string, []>("op_31912_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_31912_end_0 = const()[name = tensor<string, []>("op_31912_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_31912_end_mask_0 = const()[name = tensor<string, []>("op_31912_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31912_cast = slice_by_index(begin = var_31912_begin_0, end = var_31912_end_0, end_mask = var_31912_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31912_cast")];
            tensor<int32, [4]> var_31916_begin_0 = const()[name = tensor<string, []>("op_31916_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_31916_end_0 = const()[name = tensor<string, []>("op_31916_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_31916_end_mask_0 = const()[name = tensor<string, []>("op_31916_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31916_cast = slice_by_index(begin = var_31916_begin_0, end = var_31916_end_0, end_mask = var_31916_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31916_cast")];
            tensor<int32, [4]> var_31920_begin_0 = const()[name = tensor<string, []>("op_31920_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_31920_end_0 = const()[name = tensor<string, []>("op_31920_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_31920_end_mask_0 = const()[name = tensor<string, []>("op_31920_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31920_cast = slice_by_index(begin = var_31920_begin_0, end = var_31920_end_0, end_mask = var_31920_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31920_cast")];
            tensor<int32, [4]> var_31924_begin_0 = const()[name = tensor<string, []>("op_31924_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_31924_end_0 = const()[name = tensor<string, []>("op_31924_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_31924_end_mask_0 = const()[name = tensor<string, []>("op_31924_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31924_cast = slice_by_index(begin = var_31924_begin_0, end = var_31924_end_0, end_mask = var_31924_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31924_cast")];
            tensor<int32, [4]> var_31928_begin_0 = const()[name = tensor<string, []>("op_31928_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_31928_end_0 = const()[name = tensor<string, []>("op_31928_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_31928_end_mask_0 = const()[name = tensor<string, []>("op_31928_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31928_cast = slice_by_index(begin = var_31928_begin_0, end = var_31928_end_0, end_mask = var_31928_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31928_cast")];
            tensor<int32, [4]> var_31932_begin_0 = const()[name = tensor<string, []>("op_31932_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_31932_end_0 = const()[name = tensor<string, []>("op_31932_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_31932_end_mask_0 = const()[name = tensor<string, []>("op_31932_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31932_cast = slice_by_index(begin = var_31932_begin_0, end = var_31932_end_0, end_mask = var_31932_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31932_cast")];
            tensor<int32, [4]> var_31936_begin_0 = const()[name = tensor<string, []>("op_31936_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_31936_end_0 = const()[name = tensor<string, []>("op_31936_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_31936_end_mask_0 = const()[name = tensor<string, []>("op_31936_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31936_cast = slice_by_index(begin = var_31936_begin_0, end = var_31936_end_0, end_mask = var_31936_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31936_cast")];
            tensor<int32, [4]> var_31940_begin_0 = const()[name = tensor<string, []>("op_31940_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_31940_end_0 = const()[name = tensor<string, []>("op_31940_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_31940_end_mask_0 = const()[name = tensor<string, []>("op_31940_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31940_cast = slice_by_index(begin = var_31940_begin_0, end = var_31940_end_0, end_mask = var_31940_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31940_cast")];
            tensor<int32, [4]> var_31944_begin_0 = const()[name = tensor<string, []>("op_31944_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_31944_end_0 = const()[name = tensor<string, []>("op_31944_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_31944_end_mask_0 = const()[name = tensor<string, []>("op_31944_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31944_cast = slice_by_index(begin = var_31944_begin_0, end = var_31944_end_0, end_mask = var_31944_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31944_cast")];
            tensor<int32, [4]> var_31948_begin_0 = const()[name = tensor<string, []>("op_31948_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_31948_end_0 = const()[name = tensor<string, []>("op_31948_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_31948_end_mask_0 = const()[name = tensor<string, []>("op_31948_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31948_cast = slice_by_index(begin = var_31948_begin_0, end = var_31948_end_0, end_mask = var_31948_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31948_cast")];
            tensor<int32, [4]> var_31952_begin_0 = const()[name = tensor<string, []>("op_31952_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_31952_end_0 = const()[name = tensor<string, []>("op_31952_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_31952_end_mask_0 = const()[name = tensor<string, []>("op_31952_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31952_cast = slice_by_index(begin = var_31952_begin_0, end = var_31952_end_0, end_mask = var_31952_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31952_cast")];
            tensor<int32, [4]> var_31956_begin_0 = const()[name = tensor<string, []>("op_31956_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_31956_end_0 = const()[name = tensor<string, []>("op_31956_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_31956_end_mask_0 = const()[name = tensor<string, []>("op_31956_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_31956_cast = slice_by_index(begin = var_31956_begin_0, end = var_31956_end_0, end_mask = var_31956_end_mask_0, x = q_143_cast)[name = tensor<string, []>("op_31956_cast")];
            tensor<int32, [4]> k_287_perm_0 = const()[name = tensor<string, []>("k_287_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_31963_begin_0 = const()[name = tensor<string, []>("op_31963_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_31963_end_0 = const()[name = tensor<string, []>("op_31963_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_31963_end_mask_0 = const()[name = tensor<string, []>("op_31963_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_68 = transpose(perm = k_287_perm_0, x = k_285_cast)[name = tensor<string, []>("transpose_68")];
            tensor<fp16, [2, 77, 1, 64]> var_31963_cast = slice_by_index(begin = var_31963_begin_0, end = var_31963_end_0, end_mask = var_31963_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_31963_cast")];
            tensor<int32, [4]> var_31967_begin_0 = const()[name = tensor<string, []>("op_31967_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_31967_end_0 = const()[name = tensor<string, []>("op_31967_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_31967_end_mask_0 = const()[name = tensor<string, []>("op_31967_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31967_cast = slice_by_index(begin = var_31967_begin_0, end = var_31967_end_0, end_mask = var_31967_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_31967_cast")];
            tensor<int32, [4]> var_31971_begin_0 = const()[name = tensor<string, []>("op_31971_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_31971_end_0 = const()[name = tensor<string, []>("op_31971_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_31971_end_mask_0 = const()[name = tensor<string, []>("op_31971_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31971_cast = slice_by_index(begin = var_31971_begin_0, end = var_31971_end_0, end_mask = var_31971_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_31971_cast")];
            tensor<int32, [4]> var_31975_begin_0 = const()[name = tensor<string, []>("op_31975_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_31975_end_0 = const()[name = tensor<string, []>("op_31975_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_31975_end_mask_0 = const()[name = tensor<string, []>("op_31975_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31975_cast = slice_by_index(begin = var_31975_begin_0, end = var_31975_end_0, end_mask = var_31975_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_31975_cast")];
            tensor<int32, [4]> var_31979_begin_0 = const()[name = tensor<string, []>("op_31979_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_31979_end_0 = const()[name = tensor<string, []>("op_31979_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_31979_end_mask_0 = const()[name = tensor<string, []>("op_31979_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31979_cast = slice_by_index(begin = var_31979_begin_0, end = var_31979_end_0, end_mask = var_31979_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_31979_cast")];
            tensor<int32, [4]> var_31983_begin_0 = const()[name = tensor<string, []>("op_31983_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_31983_end_0 = const()[name = tensor<string, []>("op_31983_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_31983_end_mask_0 = const()[name = tensor<string, []>("op_31983_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31983_cast = slice_by_index(begin = var_31983_begin_0, end = var_31983_end_0, end_mask = var_31983_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_31983_cast")];
            tensor<int32, [4]> var_31987_begin_0 = const()[name = tensor<string, []>("op_31987_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_31987_end_0 = const()[name = tensor<string, []>("op_31987_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_31987_end_mask_0 = const()[name = tensor<string, []>("op_31987_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31987_cast = slice_by_index(begin = var_31987_begin_0, end = var_31987_end_0, end_mask = var_31987_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_31987_cast")];
            tensor<int32, [4]> var_31991_begin_0 = const()[name = tensor<string, []>("op_31991_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_31991_end_0 = const()[name = tensor<string, []>("op_31991_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_31991_end_mask_0 = const()[name = tensor<string, []>("op_31991_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31991_cast = slice_by_index(begin = var_31991_begin_0, end = var_31991_end_0, end_mask = var_31991_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_31991_cast")];
            tensor<int32, [4]> var_31995_begin_0 = const()[name = tensor<string, []>("op_31995_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_31995_end_0 = const()[name = tensor<string, []>("op_31995_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_31995_end_mask_0 = const()[name = tensor<string, []>("op_31995_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31995_cast = slice_by_index(begin = var_31995_begin_0, end = var_31995_end_0, end_mask = var_31995_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_31995_cast")];
            tensor<int32, [4]> var_31999_begin_0 = const()[name = tensor<string, []>("op_31999_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_31999_end_0 = const()[name = tensor<string, []>("op_31999_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_31999_end_mask_0 = const()[name = tensor<string, []>("op_31999_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_31999_cast = slice_by_index(begin = var_31999_begin_0, end = var_31999_end_0, end_mask = var_31999_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_31999_cast")];
            tensor<int32, [4]> var_32003_begin_0 = const()[name = tensor<string, []>("op_32003_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_32003_end_0 = const()[name = tensor<string, []>("op_32003_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_32003_end_mask_0 = const()[name = tensor<string, []>("op_32003_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32003_cast = slice_by_index(begin = var_32003_begin_0, end = var_32003_end_0, end_mask = var_32003_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_32003_cast")];
            tensor<int32, [4]> var_32007_begin_0 = const()[name = tensor<string, []>("op_32007_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_32007_end_0 = const()[name = tensor<string, []>("op_32007_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_32007_end_mask_0 = const()[name = tensor<string, []>("op_32007_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32007_cast = slice_by_index(begin = var_32007_begin_0, end = var_32007_end_0, end_mask = var_32007_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_32007_cast")];
            tensor<int32, [4]> var_32011_begin_0 = const()[name = tensor<string, []>("op_32011_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_32011_end_0 = const()[name = tensor<string, []>("op_32011_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_32011_end_mask_0 = const()[name = tensor<string, []>("op_32011_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32011_cast = slice_by_index(begin = var_32011_begin_0, end = var_32011_end_0, end_mask = var_32011_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_32011_cast")];
            tensor<int32, [4]> var_32015_begin_0 = const()[name = tensor<string, []>("op_32015_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_32015_end_0 = const()[name = tensor<string, []>("op_32015_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_32015_end_mask_0 = const()[name = tensor<string, []>("op_32015_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32015_cast = slice_by_index(begin = var_32015_begin_0, end = var_32015_end_0, end_mask = var_32015_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_32015_cast")];
            tensor<int32, [4]> var_32019_begin_0 = const()[name = tensor<string, []>("op_32019_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_32019_end_0 = const()[name = tensor<string, []>("op_32019_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_32019_end_mask_0 = const()[name = tensor<string, []>("op_32019_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32019_cast = slice_by_index(begin = var_32019_begin_0, end = var_32019_end_0, end_mask = var_32019_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_32019_cast")];
            tensor<int32, [4]> var_32023_begin_0 = const()[name = tensor<string, []>("op_32023_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_32023_end_0 = const()[name = tensor<string, []>("op_32023_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_32023_end_mask_0 = const()[name = tensor<string, []>("op_32023_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32023_cast = slice_by_index(begin = var_32023_begin_0, end = var_32023_end_0, end_mask = var_32023_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_32023_cast")];
            tensor<int32, [4]> var_32027_begin_0 = const()[name = tensor<string, []>("op_32027_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_32027_end_0 = const()[name = tensor<string, []>("op_32027_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_32027_end_mask_0 = const()[name = tensor<string, []>("op_32027_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32027_cast = slice_by_index(begin = var_32027_begin_0, end = var_32027_end_0, end_mask = var_32027_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_32027_cast")];
            tensor<int32, [4]> var_32031_begin_0 = const()[name = tensor<string, []>("op_32031_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_32031_end_0 = const()[name = tensor<string, []>("op_32031_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_32031_end_mask_0 = const()[name = tensor<string, []>("op_32031_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32031_cast = slice_by_index(begin = var_32031_begin_0, end = var_32031_end_0, end_mask = var_32031_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_32031_cast")];
            tensor<int32, [4]> var_32035_begin_0 = const()[name = tensor<string, []>("op_32035_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_32035_end_0 = const()[name = tensor<string, []>("op_32035_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_32035_end_mask_0 = const()[name = tensor<string, []>("op_32035_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32035_cast = slice_by_index(begin = var_32035_begin_0, end = var_32035_end_0, end_mask = var_32035_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_32035_cast")];
            tensor<int32, [4]> var_32039_begin_0 = const()[name = tensor<string, []>("op_32039_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_32039_end_0 = const()[name = tensor<string, []>("op_32039_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_32039_end_mask_0 = const()[name = tensor<string, []>("op_32039_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32039_cast = slice_by_index(begin = var_32039_begin_0, end = var_32039_end_0, end_mask = var_32039_end_mask_0, x = transpose_68)[name = tensor<string, []>("op_32039_cast")];
            tensor<int32, [4]> var_32041_begin_0 = const()[name = tensor<string, []>("op_32041_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_32041_end_0 = const()[name = tensor<string, []>("op_32041_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_32041_end_mask_0 = const()[name = tensor<string, []>("op_32041_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32041_cast = slice_by_index(begin = var_32041_begin_0, end = var_32041_end_0, end_mask = var_32041_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32041_cast")];
            tensor<int32, [4]> var_32045_begin_0 = const()[name = tensor<string, []>("op_32045_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_32045_end_0 = const()[name = tensor<string, []>("op_32045_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_32045_end_mask_0 = const()[name = tensor<string, []>("op_32045_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32045_cast = slice_by_index(begin = var_32045_begin_0, end = var_32045_end_0, end_mask = var_32045_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32045_cast")];
            tensor<int32, [4]> var_32049_begin_0 = const()[name = tensor<string, []>("op_32049_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_32049_end_0 = const()[name = tensor<string, []>("op_32049_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_32049_end_mask_0 = const()[name = tensor<string, []>("op_32049_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32049_cast = slice_by_index(begin = var_32049_begin_0, end = var_32049_end_0, end_mask = var_32049_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32049_cast")];
            tensor<int32, [4]> var_32053_begin_0 = const()[name = tensor<string, []>("op_32053_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_32053_end_0 = const()[name = tensor<string, []>("op_32053_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_32053_end_mask_0 = const()[name = tensor<string, []>("op_32053_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32053_cast = slice_by_index(begin = var_32053_begin_0, end = var_32053_end_0, end_mask = var_32053_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32053_cast")];
            tensor<int32, [4]> var_32057_begin_0 = const()[name = tensor<string, []>("op_32057_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_32057_end_0 = const()[name = tensor<string, []>("op_32057_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_32057_end_mask_0 = const()[name = tensor<string, []>("op_32057_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32057_cast = slice_by_index(begin = var_32057_begin_0, end = var_32057_end_0, end_mask = var_32057_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32057_cast")];
            tensor<int32, [4]> var_32061_begin_0 = const()[name = tensor<string, []>("op_32061_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_32061_end_0 = const()[name = tensor<string, []>("op_32061_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_32061_end_mask_0 = const()[name = tensor<string, []>("op_32061_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32061_cast = slice_by_index(begin = var_32061_begin_0, end = var_32061_end_0, end_mask = var_32061_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32061_cast")];
            tensor<int32, [4]> var_32065_begin_0 = const()[name = tensor<string, []>("op_32065_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_32065_end_0 = const()[name = tensor<string, []>("op_32065_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_32065_end_mask_0 = const()[name = tensor<string, []>("op_32065_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32065_cast = slice_by_index(begin = var_32065_begin_0, end = var_32065_end_0, end_mask = var_32065_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32065_cast")];
            tensor<int32, [4]> var_32069_begin_0 = const()[name = tensor<string, []>("op_32069_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_32069_end_0 = const()[name = tensor<string, []>("op_32069_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_32069_end_mask_0 = const()[name = tensor<string, []>("op_32069_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32069_cast = slice_by_index(begin = var_32069_begin_0, end = var_32069_end_0, end_mask = var_32069_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32069_cast")];
            tensor<int32, [4]> var_32073_begin_0 = const()[name = tensor<string, []>("op_32073_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_32073_end_0 = const()[name = tensor<string, []>("op_32073_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_32073_end_mask_0 = const()[name = tensor<string, []>("op_32073_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32073_cast = slice_by_index(begin = var_32073_begin_0, end = var_32073_end_0, end_mask = var_32073_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32073_cast")];
            tensor<int32, [4]> var_32077_begin_0 = const()[name = tensor<string, []>("op_32077_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_32077_end_0 = const()[name = tensor<string, []>("op_32077_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_32077_end_mask_0 = const()[name = tensor<string, []>("op_32077_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32077_cast = slice_by_index(begin = var_32077_begin_0, end = var_32077_end_0, end_mask = var_32077_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32077_cast")];
            tensor<int32, [4]> var_32081_begin_0 = const()[name = tensor<string, []>("op_32081_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_32081_end_0 = const()[name = tensor<string, []>("op_32081_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_32081_end_mask_0 = const()[name = tensor<string, []>("op_32081_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32081_cast = slice_by_index(begin = var_32081_begin_0, end = var_32081_end_0, end_mask = var_32081_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32081_cast")];
            tensor<int32, [4]> var_32085_begin_0 = const()[name = tensor<string, []>("op_32085_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_32085_end_0 = const()[name = tensor<string, []>("op_32085_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_32085_end_mask_0 = const()[name = tensor<string, []>("op_32085_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32085_cast = slice_by_index(begin = var_32085_begin_0, end = var_32085_end_0, end_mask = var_32085_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32085_cast")];
            tensor<int32, [4]> var_32089_begin_0 = const()[name = tensor<string, []>("op_32089_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_32089_end_0 = const()[name = tensor<string, []>("op_32089_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_32089_end_mask_0 = const()[name = tensor<string, []>("op_32089_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32089_cast = slice_by_index(begin = var_32089_begin_0, end = var_32089_end_0, end_mask = var_32089_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32089_cast")];
            tensor<int32, [4]> var_32093_begin_0 = const()[name = tensor<string, []>("op_32093_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_32093_end_0 = const()[name = tensor<string, []>("op_32093_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_32093_end_mask_0 = const()[name = tensor<string, []>("op_32093_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32093_cast = slice_by_index(begin = var_32093_begin_0, end = var_32093_end_0, end_mask = var_32093_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32093_cast")];
            tensor<int32, [4]> var_32097_begin_0 = const()[name = tensor<string, []>("op_32097_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_32097_end_0 = const()[name = tensor<string, []>("op_32097_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_32097_end_mask_0 = const()[name = tensor<string, []>("op_32097_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32097_cast = slice_by_index(begin = var_32097_begin_0, end = var_32097_end_0, end_mask = var_32097_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32097_cast")];
            tensor<int32, [4]> var_32101_begin_0 = const()[name = tensor<string, []>("op_32101_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_32101_end_0 = const()[name = tensor<string, []>("op_32101_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_32101_end_mask_0 = const()[name = tensor<string, []>("op_32101_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32101_cast = slice_by_index(begin = var_32101_begin_0, end = var_32101_end_0, end_mask = var_32101_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32101_cast")];
            tensor<int32, [4]> var_32105_begin_0 = const()[name = tensor<string, []>("op_32105_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_32105_end_0 = const()[name = tensor<string, []>("op_32105_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_32105_end_mask_0 = const()[name = tensor<string, []>("op_32105_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32105_cast = slice_by_index(begin = var_32105_begin_0, end = var_32105_end_0, end_mask = var_32105_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32105_cast")];
            tensor<int32, [4]> var_32109_begin_0 = const()[name = tensor<string, []>("op_32109_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_32109_end_0 = const()[name = tensor<string, []>("op_32109_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_32109_end_mask_0 = const()[name = tensor<string, []>("op_32109_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32109_cast = slice_by_index(begin = var_32109_begin_0, end = var_32109_end_0, end_mask = var_32109_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32109_cast")];
            tensor<int32, [4]> var_32113_begin_0 = const()[name = tensor<string, []>("op_32113_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_32113_end_0 = const()[name = tensor<string, []>("op_32113_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_32113_end_mask_0 = const()[name = tensor<string, []>("op_32113_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32113_cast = slice_by_index(begin = var_32113_begin_0, end = var_32113_end_0, end_mask = var_32113_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32113_cast")];
            tensor<int32, [4]> var_32117_begin_0 = const()[name = tensor<string, []>("op_32117_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_32117_end_0 = const()[name = tensor<string, []>("op_32117_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_32117_end_mask_0 = const()[name = tensor<string, []>("op_32117_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32117_cast = slice_by_index(begin = var_32117_begin_0, end = var_32117_end_0, end_mask = var_32117_end_mask_0, x = v_143_cast)[name = tensor<string, []>("op_32117_cast")];
            tensor<string, []> var_32121_equation_0 = const()[name = tensor<string, []>("op_32121_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32121_cast = einsum(equation = var_32121_equation_0, values = (var_31963_cast, var_31880_cast))[name = tensor<string, []>("op_32121_cast")];
            tensor<fp16, []> var_32122_to_fp16 = const()[name = tensor<string, []>("op_32122_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2681_cast = mul(x = var_32121_cast, y = var_32122_to_fp16)[name = tensor<string, []>("aw_2681_cast")];
            tensor<string, []> var_32125_equation_0 = const()[name = tensor<string, []>("op_32125_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32125_cast = einsum(equation = var_32125_equation_0, values = (var_31967_cast, var_31884_cast))[name = tensor<string, []>("op_32125_cast")];
            tensor<fp16, []> var_32126_to_fp16 = const()[name = tensor<string, []>("op_32126_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2683_cast = mul(x = var_32125_cast, y = var_32126_to_fp16)[name = tensor<string, []>("aw_2683_cast")];
            tensor<string, []> var_32129_equation_0 = const()[name = tensor<string, []>("op_32129_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32129_cast = einsum(equation = var_32129_equation_0, values = (var_31971_cast, var_31888_cast))[name = tensor<string, []>("op_32129_cast")];
            tensor<fp16, []> var_32130_to_fp16 = const()[name = tensor<string, []>("op_32130_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2685_cast = mul(x = var_32129_cast, y = var_32130_to_fp16)[name = tensor<string, []>("aw_2685_cast")];
            tensor<string, []> var_32133_equation_0 = const()[name = tensor<string, []>("op_32133_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32133_cast = einsum(equation = var_32133_equation_0, values = (var_31975_cast, var_31892_cast))[name = tensor<string, []>("op_32133_cast")];
            tensor<fp16, []> var_32134_to_fp16 = const()[name = tensor<string, []>("op_32134_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2687_cast = mul(x = var_32133_cast, y = var_32134_to_fp16)[name = tensor<string, []>("aw_2687_cast")];
            tensor<string, []> var_32137_equation_0 = const()[name = tensor<string, []>("op_32137_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32137_cast = einsum(equation = var_32137_equation_0, values = (var_31979_cast, var_31896_cast))[name = tensor<string, []>("op_32137_cast")];
            tensor<fp16, []> var_32138_to_fp16 = const()[name = tensor<string, []>("op_32138_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2689_cast = mul(x = var_32137_cast, y = var_32138_to_fp16)[name = tensor<string, []>("aw_2689_cast")];
            tensor<string, []> var_32141_equation_0 = const()[name = tensor<string, []>("op_32141_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32141_cast = einsum(equation = var_32141_equation_0, values = (var_31983_cast, var_31900_cast))[name = tensor<string, []>("op_32141_cast")];
            tensor<fp16, []> var_32142_to_fp16 = const()[name = tensor<string, []>("op_32142_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2691_cast = mul(x = var_32141_cast, y = var_32142_to_fp16)[name = tensor<string, []>("aw_2691_cast")];
            tensor<string, []> var_32145_equation_0 = const()[name = tensor<string, []>("op_32145_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32145_cast = einsum(equation = var_32145_equation_0, values = (var_31987_cast, var_31904_cast))[name = tensor<string, []>("op_32145_cast")];
            tensor<fp16, []> var_32146_to_fp16 = const()[name = tensor<string, []>("op_32146_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2693_cast = mul(x = var_32145_cast, y = var_32146_to_fp16)[name = tensor<string, []>("aw_2693_cast")];
            tensor<string, []> var_32149_equation_0 = const()[name = tensor<string, []>("op_32149_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32149_cast = einsum(equation = var_32149_equation_0, values = (var_31991_cast, var_31908_cast))[name = tensor<string, []>("op_32149_cast")];
            tensor<fp16, []> var_32150_to_fp16 = const()[name = tensor<string, []>("op_32150_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2695_cast = mul(x = var_32149_cast, y = var_32150_to_fp16)[name = tensor<string, []>("aw_2695_cast")];
            tensor<string, []> var_32153_equation_0 = const()[name = tensor<string, []>("op_32153_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32153_cast = einsum(equation = var_32153_equation_0, values = (var_31995_cast, var_31912_cast))[name = tensor<string, []>("op_32153_cast")];
            tensor<fp16, []> var_32154_to_fp16 = const()[name = tensor<string, []>("op_32154_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2697_cast = mul(x = var_32153_cast, y = var_32154_to_fp16)[name = tensor<string, []>("aw_2697_cast")];
            tensor<string, []> var_32157_equation_0 = const()[name = tensor<string, []>("op_32157_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32157_cast = einsum(equation = var_32157_equation_0, values = (var_31999_cast, var_31916_cast))[name = tensor<string, []>("op_32157_cast")];
            tensor<fp16, []> var_32158_to_fp16 = const()[name = tensor<string, []>("op_32158_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2699_cast = mul(x = var_32157_cast, y = var_32158_to_fp16)[name = tensor<string, []>("aw_2699_cast")];
            tensor<string, []> var_32161_equation_0 = const()[name = tensor<string, []>("op_32161_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32161_cast = einsum(equation = var_32161_equation_0, values = (var_32003_cast, var_31920_cast))[name = tensor<string, []>("op_32161_cast")];
            tensor<fp16, []> var_32162_to_fp16 = const()[name = tensor<string, []>("op_32162_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2701_cast = mul(x = var_32161_cast, y = var_32162_to_fp16)[name = tensor<string, []>("aw_2701_cast")];
            tensor<string, []> var_32165_equation_0 = const()[name = tensor<string, []>("op_32165_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32165_cast = einsum(equation = var_32165_equation_0, values = (var_32007_cast, var_31924_cast))[name = tensor<string, []>("op_32165_cast")];
            tensor<fp16, []> var_32166_to_fp16 = const()[name = tensor<string, []>("op_32166_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2703_cast = mul(x = var_32165_cast, y = var_32166_to_fp16)[name = tensor<string, []>("aw_2703_cast")];
            tensor<string, []> var_32169_equation_0 = const()[name = tensor<string, []>("op_32169_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32169_cast = einsum(equation = var_32169_equation_0, values = (var_32011_cast, var_31928_cast))[name = tensor<string, []>("op_32169_cast")];
            tensor<fp16, []> var_32170_to_fp16 = const()[name = tensor<string, []>("op_32170_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2705_cast = mul(x = var_32169_cast, y = var_32170_to_fp16)[name = tensor<string, []>("aw_2705_cast")];
            tensor<string, []> var_32173_equation_0 = const()[name = tensor<string, []>("op_32173_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32173_cast = einsum(equation = var_32173_equation_0, values = (var_32015_cast, var_31932_cast))[name = tensor<string, []>("op_32173_cast")];
            tensor<fp16, []> var_32174_to_fp16 = const()[name = tensor<string, []>("op_32174_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2707_cast = mul(x = var_32173_cast, y = var_32174_to_fp16)[name = tensor<string, []>("aw_2707_cast")];
            tensor<string, []> var_32177_equation_0 = const()[name = tensor<string, []>("op_32177_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32177_cast = einsum(equation = var_32177_equation_0, values = (var_32019_cast, var_31936_cast))[name = tensor<string, []>("op_32177_cast")];
            tensor<fp16, []> var_32178_to_fp16 = const()[name = tensor<string, []>("op_32178_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2709_cast = mul(x = var_32177_cast, y = var_32178_to_fp16)[name = tensor<string, []>("aw_2709_cast")];
            tensor<string, []> var_32181_equation_0 = const()[name = tensor<string, []>("op_32181_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32181_cast = einsum(equation = var_32181_equation_0, values = (var_32023_cast, var_31940_cast))[name = tensor<string, []>("op_32181_cast")];
            tensor<fp16, []> var_32182_to_fp16 = const()[name = tensor<string, []>("op_32182_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2711_cast = mul(x = var_32181_cast, y = var_32182_to_fp16)[name = tensor<string, []>("aw_2711_cast")];
            tensor<string, []> var_32185_equation_0 = const()[name = tensor<string, []>("op_32185_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32185_cast = einsum(equation = var_32185_equation_0, values = (var_32027_cast, var_31944_cast))[name = tensor<string, []>("op_32185_cast")];
            tensor<fp16, []> var_32186_to_fp16 = const()[name = tensor<string, []>("op_32186_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2713_cast = mul(x = var_32185_cast, y = var_32186_to_fp16)[name = tensor<string, []>("aw_2713_cast")];
            tensor<string, []> var_32189_equation_0 = const()[name = tensor<string, []>("op_32189_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32189_cast = einsum(equation = var_32189_equation_0, values = (var_32031_cast, var_31948_cast))[name = tensor<string, []>("op_32189_cast")];
            tensor<fp16, []> var_32190_to_fp16 = const()[name = tensor<string, []>("op_32190_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2715_cast = mul(x = var_32189_cast, y = var_32190_to_fp16)[name = tensor<string, []>("aw_2715_cast")];
            tensor<string, []> var_32193_equation_0 = const()[name = tensor<string, []>("op_32193_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32193_cast = einsum(equation = var_32193_equation_0, values = (var_32035_cast, var_31952_cast))[name = tensor<string, []>("op_32193_cast")];
            tensor<fp16, []> var_32194_to_fp16 = const()[name = tensor<string, []>("op_32194_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2717_cast = mul(x = var_32193_cast, y = var_32194_to_fp16)[name = tensor<string, []>("aw_2717_cast")];
            tensor<string, []> var_32197_equation_0 = const()[name = tensor<string, []>("op_32197_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_32197_cast = einsum(equation = var_32197_equation_0, values = (var_32039_cast, var_31956_cast))[name = tensor<string, []>("op_32197_cast")];
            tensor<fp16, []> var_32198_to_fp16 = const()[name = tensor<string, []>("op_32198_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2719_cast = mul(x = var_32197_cast, y = var_32198_to_fp16)[name = tensor<string, []>("aw_2719_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32200_cast = softmax(axis = var_30385, x = aw_2681_cast)[name = tensor<string, []>("op_32200_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32201_cast = softmax(axis = var_30385, x = aw_2683_cast)[name = tensor<string, []>("op_32201_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32202_cast = softmax(axis = var_30385, x = aw_2685_cast)[name = tensor<string, []>("op_32202_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32203_cast = softmax(axis = var_30385, x = aw_2687_cast)[name = tensor<string, []>("op_32203_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32204_cast = softmax(axis = var_30385, x = aw_2689_cast)[name = tensor<string, []>("op_32204_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32205_cast = softmax(axis = var_30385, x = aw_2691_cast)[name = tensor<string, []>("op_32205_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32206_cast = softmax(axis = var_30385, x = aw_2693_cast)[name = tensor<string, []>("op_32206_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32207_cast = softmax(axis = var_30385, x = aw_2695_cast)[name = tensor<string, []>("op_32207_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32208_cast = softmax(axis = var_30385, x = aw_2697_cast)[name = tensor<string, []>("op_32208_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32209_cast = softmax(axis = var_30385, x = aw_2699_cast)[name = tensor<string, []>("op_32209_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32210_cast = softmax(axis = var_30385, x = aw_2701_cast)[name = tensor<string, []>("op_32210_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32211_cast = softmax(axis = var_30385, x = aw_2703_cast)[name = tensor<string, []>("op_32211_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32212_cast = softmax(axis = var_30385, x = aw_2705_cast)[name = tensor<string, []>("op_32212_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32213_cast = softmax(axis = var_30385, x = aw_2707_cast)[name = tensor<string, []>("op_32213_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32214_cast = softmax(axis = var_30385, x = aw_2709_cast)[name = tensor<string, []>("op_32214_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32215_cast = softmax(axis = var_30385, x = aw_2711_cast)[name = tensor<string, []>("op_32215_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32216_cast = softmax(axis = var_30385, x = aw_2713_cast)[name = tensor<string, []>("op_32216_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32217_cast = softmax(axis = var_30385, x = aw_2715_cast)[name = tensor<string, []>("op_32217_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32218_cast = softmax(axis = var_30385, x = aw_2717_cast)[name = tensor<string, []>("op_32218_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_32219_cast = softmax(axis = var_30385, x = aw_2719_cast)[name = tensor<string, []>("op_32219_cast")];
            tensor<string, []> var_32221_equation_0 = const()[name = tensor<string, []>("op_32221_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32221_cast = einsum(equation = var_32221_equation_0, values = (var_32041_cast, var_32200_cast))[name = tensor<string, []>("op_32221_cast")];
            tensor<string, []> var_32223_equation_0 = const()[name = tensor<string, []>("op_32223_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32223_cast = einsum(equation = var_32223_equation_0, values = (var_32045_cast, var_32201_cast))[name = tensor<string, []>("op_32223_cast")];
            tensor<string, []> var_32225_equation_0 = const()[name = tensor<string, []>("op_32225_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32225_cast = einsum(equation = var_32225_equation_0, values = (var_32049_cast, var_32202_cast))[name = tensor<string, []>("op_32225_cast")];
            tensor<string, []> var_32227_equation_0 = const()[name = tensor<string, []>("op_32227_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32227_cast = einsum(equation = var_32227_equation_0, values = (var_32053_cast, var_32203_cast))[name = tensor<string, []>("op_32227_cast")];
            tensor<string, []> var_32229_equation_0 = const()[name = tensor<string, []>("op_32229_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32229_cast = einsum(equation = var_32229_equation_0, values = (var_32057_cast, var_32204_cast))[name = tensor<string, []>("op_32229_cast")];
            tensor<string, []> var_32231_equation_0 = const()[name = tensor<string, []>("op_32231_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32231_cast = einsum(equation = var_32231_equation_0, values = (var_32061_cast, var_32205_cast))[name = tensor<string, []>("op_32231_cast")];
            tensor<string, []> var_32233_equation_0 = const()[name = tensor<string, []>("op_32233_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32233_cast = einsum(equation = var_32233_equation_0, values = (var_32065_cast, var_32206_cast))[name = tensor<string, []>("op_32233_cast")];
            tensor<string, []> var_32235_equation_0 = const()[name = tensor<string, []>("op_32235_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32235_cast = einsum(equation = var_32235_equation_0, values = (var_32069_cast, var_32207_cast))[name = tensor<string, []>("op_32235_cast")];
            tensor<string, []> var_32237_equation_0 = const()[name = tensor<string, []>("op_32237_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32237_cast = einsum(equation = var_32237_equation_0, values = (var_32073_cast, var_32208_cast))[name = tensor<string, []>("op_32237_cast")];
            tensor<string, []> var_32239_equation_0 = const()[name = tensor<string, []>("op_32239_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32239_cast = einsum(equation = var_32239_equation_0, values = (var_32077_cast, var_32209_cast))[name = tensor<string, []>("op_32239_cast")];
            tensor<string, []> var_32241_equation_0 = const()[name = tensor<string, []>("op_32241_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32241_cast = einsum(equation = var_32241_equation_0, values = (var_32081_cast, var_32210_cast))[name = tensor<string, []>("op_32241_cast")];
            tensor<string, []> var_32243_equation_0 = const()[name = tensor<string, []>("op_32243_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32243_cast = einsum(equation = var_32243_equation_0, values = (var_32085_cast, var_32211_cast))[name = tensor<string, []>("op_32243_cast")];
            tensor<string, []> var_32245_equation_0 = const()[name = tensor<string, []>("op_32245_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32245_cast = einsum(equation = var_32245_equation_0, values = (var_32089_cast, var_32212_cast))[name = tensor<string, []>("op_32245_cast")];
            tensor<string, []> var_32247_equation_0 = const()[name = tensor<string, []>("op_32247_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32247_cast = einsum(equation = var_32247_equation_0, values = (var_32093_cast, var_32213_cast))[name = tensor<string, []>("op_32247_cast")];
            tensor<string, []> var_32249_equation_0 = const()[name = tensor<string, []>("op_32249_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32249_cast = einsum(equation = var_32249_equation_0, values = (var_32097_cast, var_32214_cast))[name = tensor<string, []>("op_32249_cast")];
            tensor<string, []> var_32251_equation_0 = const()[name = tensor<string, []>("op_32251_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32251_cast = einsum(equation = var_32251_equation_0, values = (var_32101_cast, var_32215_cast))[name = tensor<string, []>("op_32251_cast")];
            tensor<string, []> var_32253_equation_0 = const()[name = tensor<string, []>("op_32253_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32253_cast = einsum(equation = var_32253_equation_0, values = (var_32105_cast, var_32216_cast))[name = tensor<string, []>("op_32253_cast")];
            tensor<string, []> var_32255_equation_0 = const()[name = tensor<string, []>("op_32255_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32255_cast = einsum(equation = var_32255_equation_0, values = (var_32109_cast, var_32217_cast))[name = tensor<string, []>("op_32255_cast")];
            tensor<string, []> var_32257_equation_0 = const()[name = tensor<string, []>("op_32257_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32257_cast = einsum(equation = var_32257_equation_0, values = (var_32113_cast, var_32218_cast))[name = tensor<string, []>("op_32257_cast")];
            tensor<string, []> var_32259_equation_0 = const()[name = tensor<string, []>("op_32259_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32259_cast = einsum(equation = var_32259_equation_0, values = (var_32117_cast, var_32219_cast))[name = tensor<string, []>("op_32259_cast")];
            tensor<bool, []> input_451_interleave_0 = const()[name = tensor<string, []>("input_451_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_451_cast = concat(axis = var_30385, interleave = input_451_interleave_0, values = (var_32221_cast, var_32223_cast, var_32225_cast, var_32227_cast, var_32229_cast, var_32231_cast, var_32233_cast, var_32235_cast, var_32237_cast, var_32239_cast, var_32241_cast, var_32243_cast, var_32245_cast, var_32247_cast, var_32249_cast, var_32251_cast, var_32253_cast, var_32255_cast, var_32257_cast, var_32259_cast))[name = tensor<string, []>("input_451_cast")];
            tensor<int32, [2]> var_32265 = const()[name = tensor<string, []>("op_32265"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_32267 = const()[name = tensor<string, []>("op_32267"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_32269_pad_type_0 = const()[name = tensor<string, []>("op_32269_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_32269_pad_0 = const()[name = tensor<string, []>("op_32269_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2684817664)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2688094528)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_32269_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_32267, groups = var_30385, pad = var_32269_pad_0, pad_type = var_32269_pad_type_0, strides = var_32265, weight = up_blocks_0_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_451_cast)[name = tensor<string, []>("op_32269_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_215_cast = add(x = var_32269_cast, y = inputs_213_cast)[name = tensor<string, []>("inputs_215_cast")];
            tensor<int32, [1]> var_32273 = const()[name = tensor<string, []>("op_32273"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_215_cast = reduce_mean(axes = var_32273, keep_dims = var_30380, x = inputs_215_cast)[name = tensor<string, []>("channels_mean_215_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_215_cast = sub(x = inputs_215_cast, y = channels_mean_215_cast)[name = tensor<string, []>("zero_mean_215_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_215_cast = mul(x = zero_mean_215_cast, y = zero_mean_215_cast)[name = tensor<string, []>("zero_mean_sq_215_cast")];
            tensor<int32, [1]> var_32277 = const()[name = tensor<string, []>("op_32277"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_32278_cast = reduce_mean(axes = var_32277, keep_dims = var_30380, x = zero_mean_sq_215_cast)[name = tensor<string, []>("op_32278_cast")];
            tensor<fp16, []> var_32279_to_fp16 = const()[name = tensor<string, []>("op_32279_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_32280_cast = add(x = var_32278_cast, y = var_32279_to_fp16)[name = tensor<string, []>("op_32280_cast")];
            tensor<fp16, []> denom_215_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_215_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_215_cast = rsqrt(epsilon = denom_215_epsilon_0_to_fp16, x = var_32280_cast)[name = tensor<string, []>("denom_215_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_215_cast = mul(x = zero_mean_215_cast, y = denom_215_cast)[name = tensor<string, []>("out_215_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_32284_to_fp16 = const()[name = tensor<string, []>("op_32284_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2688097152)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_32285_cast = add(x = out_215_cast, y = var_32284_to_fp16)[name = tensor<string, []>("op_32285_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_32287_to_fp16 = const()[name = tensor<string, []>("op_32287_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2688099776)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_453_cast = mul(x = var_32285_cast, y = var_32287_to_fp16)[name = tensor<string, []>("input_453_cast")];
            tensor<int32, [2]> var_32295 = const()[name = tensor<string, []>("op_32295"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_32297 = const()[name = tensor<string, []>("op_32297"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_32299_pad_type_0 = const()[name = tensor<string, []>("op_32299_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_32299_pad_0 = const()[name = tensor<string, []>("op_32299_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2688102400)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2714316864)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_32299_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_32297, groups = var_30385, pad = var_32299_pad_0, pad_type = var_32299_pad_type_0, strides = var_32295, weight = up_blocks_0_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_453_cast)[name = tensor<string, []>("op_32299_cast")];
            tensor<int32, [2]> var_32300_split_sizes_0 = const()[name = tensor<string, []>("op_32300_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_32300_axis_0 = const()[name = tensor<string, []>("op_32300_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_32300_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_32300_cast_1 = split(axis = var_32300_axis_0, split_sizes = var_32300_split_sizes_0, x = var_32299_cast)[name = tensor<string, []>("op_32300_cast")];
            tensor<string, []> var_32302_mode_0 = const()[name = tensor<string, []>("op_32302_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_32302_cast = gelu(mode = var_32302_mode_0, x = var_32300_cast_1)[name = tensor<string, []>("op_32302_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_455_cast = mul(x = var_32300_cast_0, y = var_32302_cast)[name = tensor<string, []>("input_455_cast")];
            tensor<int32, [2]> var_32306 = const()[name = tensor<string, []>("op_32306"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_32308 = const()[name = tensor<string, []>("op_32308"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_32310_pad_type_0 = const()[name = tensor<string, []>("op_32310_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_32310_pad_0 = const()[name = tensor<string, []>("op_32310_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2714337408)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2727444672)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_32310_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_32308, groups = var_30385, pad = var_32310_pad_0, pad_type = var_32310_pad_type_0, strides = var_32306, weight = up_blocks_0_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_455_cast)[name = tensor<string, []>("op_32310_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_217_cast = add(x = var_32310_cast, y = inputs_215_cast)[name = tensor<string, []>("inputs_217_cast")];
            tensor<int32, [1]> var_32320 = const()[name = tensor<string, []>("op_32320"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_217_cast = reduce_mean(axes = var_32320, keep_dims = var_30380, x = inputs_217_cast)[name = tensor<string, []>("channels_mean_217_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_217_cast = sub(x = inputs_217_cast, y = channels_mean_217_cast)[name = tensor<string, []>("zero_mean_217_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_217_cast = mul(x = zero_mean_217_cast, y = zero_mean_217_cast)[name = tensor<string, []>("zero_mean_sq_217_cast")];
            tensor<int32, [1]> var_32324 = const()[name = tensor<string, []>("op_32324"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_32325_cast = reduce_mean(axes = var_32324, keep_dims = var_30380, x = zero_mean_sq_217_cast)[name = tensor<string, []>("op_32325_cast")];
            tensor<fp16, []> var_32326_to_fp16 = const()[name = tensor<string, []>("op_32326_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_32327_cast = add(x = var_32325_cast, y = var_32326_to_fp16)[name = tensor<string, []>("op_32327_cast")];
            tensor<fp16, []> denom_217_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_217_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_217_cast = rsqrt(epsilon = denom_217_epsilon_0_to_fp16, x = var_32327_cast)[name = tensor<string, []>("denom_217_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_217_cast = mul(x = zero_mean_217_cast, y = denom_217_cast)[name = tensor<string, []>("out_217_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_32331_to_fp16 = const()[name = tensor<string, []>("op_32331_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2727447296)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_32332_cast = add(x = out_217_cast, y = var_32331_to_fp16)[name = tensor<string, []>("op_32332_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_32334_to_fp16 = const()[name = tensor<string, []>("op_32334_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2727449920)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_299_cast = mul(x = var_32332_cast, y = var_32334_to_fp16)[name = tensor<string, []>("hidden_states_299_cast")];
            tensor<int32, [2]> var_32341 = const()[name = tensor<string, []>("op_32341"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_32343 = const()[name = tensor<string, []>("op_32343"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_145_pad_type_0 = const()[name = tensor<string, []>("q_145_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_145_pad_0 = const()[name = tensor<string, []>("q_145_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2727452544)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_145_cast = conv(dilations = var_32343, groups = var_30385, pad = q_145_pad_0, pad_type = q_145_pad_type_0, strides = var_32341, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_q_weight_to_fp16, x = hidden_states_299_cast)[name = tensor<string, []>("q_145_cast")];
            tensor<int32, [2]> var_32347 = const()[name = tensor<string, []>("op_32347"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_32349 = const()[name = tensor<string, []>("op_32349"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_289_pad_type_0 = const()[name = tensor<string, []>("k_289_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_289_pad_0 = const()[name = tensor<string, []>("k_289_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2730729408)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_289_cast = conv(dilations = var_32349, groups = var_30385, pad = k_289_pad_0, pad_type = k_289_pad_type_0, strides = var_32347, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_k_weight_to_fp16, x = hidden_states_299_cast)[name = tensor<string, []>("k_289_cast")];
            tensor<int32, [2]> var_32353 = const()[name = tensor<string, []>("op_32353"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_32355 = const()[name = tensor<string, []>("op_32355"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_145_pad_type_0 = const()[name = tensor<string, []>("v_145_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_145_pad_0 = const()[name = tensor<string, []>("v_145_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2734006272)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_145_cast = conv(dilations = var_32355, groups = var_30385, pad = v_145_pad_0, pad_type = v_145_pad_type_0, strides = var_32353, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_v_weight_to_fp16, x = hidden_states_299_cast)[name = tensor<string, []>("v_145_cast")];
            tensor<int32, [4]> var_32359_begin_0 = const()[name = tensor<string, []>("op_32359_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_32359_end_0 = const()[name = tensor<string, []>("op_32359_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_32359_end_mask_0 = const()[name = tensor<string, []>("op_32359_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32359_cast = slice_by_index(begin = var_32359_begin_0, end = var_32359_end_0, end_mask = var_32359_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32359_cast")];
            tensor<int32, [4]> var_32363_begin_0 = const()[name = tensor<string, []>("op_32363_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_32363_end_0 = const()[name = tensor<string, []>("op_32363_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_32363_end_mask_0 = const()[name = tensor<string, []>("op_32363_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32363_cast = slice_by_index(begin = var_32363_begin_0, end = var_32363_end_0, end_mask = var_32363_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32363_cast")];
            tensor<int32, [4]> var_32367_begin_0 = const()[name = tensor<string, []>("op_32367_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_32367_end_0 = const()[name = tensor<string, []>("op_32367_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_32367_end_mask_0 = const()[name = tensor<string, []>("op_32367_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32367_cast = slice_by_index(begin = var_32367_begin_0, end = var_32367_end_0, end_mask = var_32367_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32367_cast")];
            tensor<int32, [4]> var_32371_begin_0 = const()[name = tensor<string, []>("op_32371_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_32371_end_0 = const()[name = tensor<string, []>("op_32371_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_32371_end_mask_0 = const()[name = tensor<string, []>("op_32371_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32371_cast = slice_by_index(begin = var_32371_begin_0, end = var_32371_end_0, end_mask = var_32371_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32371_cast")];
            tensor<int32, [4]> var_32375_begin_0 = const()[name = tensor<string, []>("op_32375_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_32375_end_0 = const()[name = tensor<string, []>("op_32375_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_32375_end_mask_0 = const()[name = tensor<string, []>("op_32375_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32375_cast = slice_by_index(begin = var_32375_begin_0, end = var_32375_end_0, end_mask = var_32375_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32375_cast")];
            tensor<int32, [4]> var_32379_begin_0 = const()[name = tensor<string, []>("op_32379_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_32379_end_0 = const()[name = tensor<string, []>("op_32379_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_32379_end_mask_0 = const()[name = tensor<string, []>("op_32379_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32379_cast = slice_by_index(begin = var_32379_begin_0, end = var_32379_end_0, end_mask = var_32379_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32379_cast")];
            tensor<int32, [4]> var_32383_begin_0 = const()[name = tensor<string, []>("op_32383_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_32383_end_0 = const()[name = tensor<string, []>("op_32383_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_32383_end_mask_0 = const()[name = tensor<string, []>("op_32383_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32383_cast = slice_by_index(begin = var_32383_begin_0, end = var_32383_end_0, end_mask = var_32383_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32383_cast")];
            tensor<int32, [4]> var_32387_begin_0 = const()[name = tensor<string, []>("op_32387_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_32387_end_0 = const()[name = tensor<string, []>("op_32387_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_32387_end_mask_0 = const()[name = tensor<string, []>("op_32387_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32387_cast = slice_by_index(begin = var_32387_begin_0, end = var_32387_end_0, end_mask = var_32387_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32387_cast")];
            tensor<int32, [4]> var_32391_begin_0 = const()[name = tensor<string, []>("op_32391_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_32391_end_0 = const()[name = tensor<string, []>("op_32391_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_32391_end_mask_0 = const()[name = tensor<string, []>("op_32391_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32391_cast = slice_by_index(begin = var_32391_begin_0, end = var_32391_end_0, end_mask = var_32391_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32391_cast")];
            tensor<int32, [4]> var_32395_begin_0 = const()[name = tensor<string, []>("op_32395_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_32395_end_0 = const()[name = tensor<string, []>("op_32395_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_32395_end_mask_0 = const()[name = tensor<string, []>("op_32395_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32395_cast = slice_by_index(begin = var_32395_begin_0, end = var_32395_end_0, end_mask = var_32395_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32395_cast")];
            tensor<int32, [4]> var_32399_begin_0 = const()[name = tensor<string, []>("op_32399_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_32399_end_0 = const()[name = tensor<string, []>("op_32399_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_32399_end_mask_0 = const()[name = tensor<string, []>("op_32399_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32399_cast = slice_by_index(begin = var_32399_begin_0, end = var_32399_end_0, end_mask = var_32399_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32399_cast")];
            tensor<int32, [4]> var_32403_begin_0 = const()[name = tensor<string, []>("op_32403_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_32403_end_0 = const()[name = tensor<string, []>("op_32403_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_32403_end_mask_0 = const()[name = tensor<string, []>("op_32403_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32403_cast = slice_by_index(begin = var_32403_begin_0, end = var_32403_end_0, end_mask = var_32403_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32403_cast")];
            tensor<int32, [4]> var_32407_begin_0 = const()[name = tensor<string, []>("op_32407_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_32407_end_0 = const()[name = tensor<string, []>("op_32407_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_32407_end_mask_0 = const()[name = tensor<string, []>("op_32407_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32407_cast = slice_by_index(begin = var_32407_begin_0, end = var_32407_end_0, end_mask = var_32407_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32407_cast")];
            tensor<int32, [4]> var_32411_begin_0 = const()[name = tensor<string, []>("op_32411_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_32411_end_0 = const()[name = tensor<string, []>("op_32411_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_32411_end_mask_0 = const()[name = tensor<string, []>("op_32411_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32411_cast = slice_by_index(begin = var_32411_begin_0, end = var_32411_end_0, end_mask = var_32411_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32411_cast")];
            tensor<int32, [4]> var_32415_begin_0 = const()[name = tensor<string, []>("op_32415_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_32415_end_0 = const()[name = tensor<string, []>("op_32415_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_32415_end_mask_0 = const()[name = tensor<string, []>("op_32415_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32415_cast = slice_by_index(begin = var_32415_begin_0, end = var_32415_end_0, end_mask = var_32415_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32415_cast")];
            tensor<int32, [4]> var_32419_begin_0 = const()[name = tensor<string, []>("op_32419_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_32419_end_0 = const()[name = tensor<string, []>("op_32419_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_32419_end_mask_0 = const()[name = tensor<string, []>("op_32419_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32419_cast = slice_by_index(begin = var_32419_begin_0, end = var_32419_end_0, end_mask = var_32419_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32419_cast")];
            tensor<int32, [4]> var_32423_begin_0 = const()[name = tensor<string, []>("op_32423_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_32423_end_0 = const()[name = tensor<string, []>("op_32423_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_32423_end_mask_0 = const()[name = tensor<string, []>("op_32423_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32423_cast = slice_by_index(begin = var_32423_begin_0, end = var_32423_end_0, end_mask = var_32423_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32423_cast")];
            tensor<int32, [4]> var_32427_begin_0 = const()[name = tensor<string, []>("op_32427_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_32427_end_0 = const()[name = tensor<string, []>("op_32427_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_32427_end_mask_0 = const()[name = tensor<string, []>("op_32427_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32427_cast = slice_by_index(begin = var_32427_begin_0, end = var_32427_end_0, end_mask = var_32427_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32427_cast")];
            tensor<int32, [4]> var_32431_begin_0 = const()[name = tensor<string, []>("op_32431_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_32431_end_0 = const()[name = tensor<string, []>("op_32431_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_32431_end_mask_0 = const()[name = tensor<string, []>("op_32431_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32431_cast = slice_by_index(begin = var_32431_begin_0, end = var_32431_end_0, end_mask = var_32431_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32431_cast")];
            tensor<int32, [4]> var_32435_begin_0 = const()[name = tensor<string, []>("op_32435_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_32435_end_0 = const()[name = tensor<string, []>("op_32435_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_32435_end_mask_0 = const()[name = tensor<string, []>("op_32435_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32435_cast = slice_by_index(begin = var_32435_begin_0, end = var_32435_end_0, end_mask = var_32435_end_mask_0, x = q_145_cast)[name = tensor<string, []>("op_32435_cast")];
            tensor<int32, [4]> k_291_perm_0 = const()[name = tensor<string, []>("k_291_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_32442_begin_0 = const()[name = tensor<string, []>("op_32442_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_32442_end_0 = const()[name = tensor<string, []>("op_32442_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_32442_end_mask_0 = const()[name = tensor<string, []>("op_32442_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_67 = transpose(perm = k_291_perm_0, x = k_289_cast)[name = tensor<string, []>("transpose_67")];
            tensor<fp16, [2, 1024, 1, 64]> var_32442_cast = slice_by_index(begin = var_32442_begin_0, end = var_32442_end_0, end_mask = var_32442_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32442_cast")];
            tensor<int32, [4]> var_32446_begin_0 = const()[name = tensor<string, []>("op_32446_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_32446_end_0 = const()[name = tensor<string, []>("op_32446_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_32446_end_mask_0 = const()[name = tensor<string, []>("op_32446_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32446_cast = slice_by_index(begin = var_32446_begin_0, end = var_32446_end_0, end_mask = var_32446_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32446_cast")];
            tensor<int32, [4]> var_32450_begin_0 = const()[name = tensor<string, []>("op_32450_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_32450_end_0 = const()[name = tensor<string, []>("op_32450_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_32450_end_mask_0 = const()[name = tensor<string, []>("op_32450_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32450_cast = slice_by_index(begin = var_32450_begin_0, end = var_32450_end_0, end_mask = var_32450_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32450_cast")];
            tensor<int32, [4]> var_32454_begin_0 = const()[name = tensor<string, []>("op_32454_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_32454_end_0 = const()[name = tensor<string, []>("op_32454_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_32454_end_mask_0 = const()[name = tensor<string, []>("op_32454_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32454_cast = slice_by_index(begin = var_32454_begin_0, end = var_32454_end_0, end_mask = var_32454_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32454_cast")];
            tensor<int32, [4]> var_32458_begin_0 = const()[name = tensor<string, []>("op_32458_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_32458_end_0 = const()[name = tensor<string, []>("op_32458_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_32458_end_mask_0 = const()[name = tensor<string, []>("op_32458_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32458_cast = slice_by_index(begin = var_32458_begin_0, end = var_32458_end_0, end_mask = var_32458_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32458_cast")];
            tensor<int32, [4]> var_32462_begin_0 = const()[name = tensor<string, []>("op_32462_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_32462_end_0 = const()[name = tensor<string, []>("op_32462_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_32462_end_mask_0 = const()[name = tensor<string, []>("op_32462_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32462_cast = slice_by_index(begin = var_32462_begin_0, end = var_32462_end_0, end_mask = var_32462_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32462_cast")];
            tensor<int32, [4]> var_32466_begin_0 = const()[name = tensor<string, []>("op_32466_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_32466_end_0 = const()[name = tensor<string, []>("op_32466_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_32466_end_mask_0 = const()[name = tensor<string, []>("op_32466_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32466_cast = slice_by_index(begin = var_32466_begin_0, end = var_32466_end_0, end_mask = var_32466_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32466_cast")];
            tensor<int32, [4]> var_32470_begin_0 = const()[name = tensor<string, []>("op_32470_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_32470_end_0 = const()[name = tensor<string, []>("op_32470_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_32470_end_mask_0 = const()[name = tensor<string, []>("op_32470_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32470_cast = slice_by_index(begin = var_32470_begin_0, end = var_32470_end_0, end_mask = var_32470_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32470_cast")];
            tensor<int32, [4]> var_32474_begin_0 = const()[name = tensor<string, []>("op_32474_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_32474_end_0 = const()[name = tensor<string, []>("op_32474_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_32474_end_mask_0 = const()[name = tensor<string, []>("op_32474_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32474_cast = slice_by_index(begin = var_32474_begin_0, end = var_32474_end_0, end_mask = var_32474_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32474_cast")];
            tensor<int32, [4]> var_32478_begin_0 = const()[name = tensor<string, []>("op_32478_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_32478_end_0 = const()[name = tensor<string, []>("op_32478_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_32478_end_mask_0 = const()[name = tensor<string, []>("op_32478_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32478_cast = slice_by_index(begin = var_32478_begin_0, end = var_32478_end_0, end_mask = var_32478_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32478_cast")];
            tensor<int32, [4]> var_32482_begin_0 = const()[name = tensor<string, []>("op_32482_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_32482_end_0 = const()[name = tensor<string, []>("op_32482_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_32482_end_mask_0 = const()[name = tensor<string, []>("op_32482_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32482_cast = slice_by_index(begin = var_32482_begin_0, end = var_32482_end_0, end_mask = var_32482_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32482_cast")];
            tensor<int32, [4]> var_32486_begin_0 = const()[name = tensor<string, []>("op_32486_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_32486_end_0 = const()[name = tensor<string, []>("op_32486_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_32486_end_mask_0 = const()[name = tensor<string, []>("op_32486_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32486_cast = slice_by_index(begin = var_32486_begin_0, end = var_32486_end_0, end_mask = var_32486_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32486_cast")];
            tensor<int32, [4]> var_32490_begin_0 = const()[name = tensor<string, []>("op_32490_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_32490_end_0 = const()[name = tensor<string, []>("op_32490_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_32490_end_mask_0 = const()[name = tensor<string, []>("op_32490_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32490_cast = slice_by_index(begin = var_32490_begin_0, end = var_32490_end_0, end_mask = var_32490_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32490_cast")];
            tensor<int32, [4]> var_32494_begin_0 = const()[name = tensor<string, []>("op_32494_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_32494_end_0 = const()[name = tensor<string, []>("op_32494_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_32494_end_mask_0 = const()[name = tensor<string, []>("op_32494_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32494_cast = slice_by_index(begin = var_32494_begin_0, end = var_32494_end_0, end_mask = var_32494_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32494_cast")];
            tensor<int32, [4]> var_32498_begin_0 = const()[name = tensor<string, []>("op_32498_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_32498_end_0 = const()[name = tensor<string, []>("op_32498_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_32498_end_mask_0 = const()[name = tensor<string, []>("op_32498_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32498_cast = slice_by_index(begin = var_32498_begin_0, end = var_32498_end_0, end_mask = var_32498_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32498_cast")];
            tensor<int32, [4]> var_32502_begin_0 = const()[name = tensor<string, []>("op_32502_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_32502_end_0 = const()[name = tensor<string, []>("op_32502_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_32502_end_mask_0 = const()[name = tensor<string, []>("op_32502_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32502_cast = slice_by_index(begin = var_32502_begin_0, end = var_32502_end_0, end_mask = var_32502_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32502_cast")];
            tensor<int32, [4]> var_32506_begin_0 = const()[name = tensor<string, []>("op_32506_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_32506_end_0 = const()[name = tensor<string, []>("op_32506_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_32506_end_mask_0 = const()[name = tensor<string, []>("op_32506_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32506_cast = slice_by_index(begin = var_32506_begin_0, end = var_32506_end_0, end_mask = var_32506_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32506_cast")];
            tensor<int32, [4]> var_32510_begin_0 = const()[name = tensor<string, []>("op_32510_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_32510_end_0 = const()[name = tensor<string, []>("op_32510_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_32510_end_mask_0 = const()[name = tensor<string, []>("op_32510_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32510_cast = slice_by_index(begin = var_32510_begin_0, end = var_32510_end_0, end_mask = var_32510_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32510_cast")];
            tensor<int32, [4]> var_32514_begin_0 = const()[name = tensor<string, []>("op_32514_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_32514_end_0 = const()[name = tensor<string, []>("op_32514_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_32514_end_mask_0 = const()[name = tensor<string, []>("op_32514_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32514_cast = slice_by_index(begin = var_32514_begin_0, end = var_32514_end_0, end_mask = var_32514_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32514_cast")];
            tensor<int32, [4]> var_32518_begin_0 = const()[name = tensor<string, []>("op_32518_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_32518_end_0 = const()[name = tensor<string, []>("op_32518_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_32518_end_mask_0 = const()[name = tensor<string, []>("op_32518_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_32518_cast = slice_by_index(begin = var_32518_begin_0, end = var_32518_end_0, end_mask = var_32518_end_mask_0, x = transpose_67)[name = tensor<string, []>("op_32518_cast")];
            tensor<int32, [4]> var_32520_begin_0 = const()[name = tensor<string, []>("op_32520_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_32520_end_0 = const()[name = tensor<string, []>("op_32520_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_32520_end_mask_0 = const()[name = tensor<string, []>("op_32520_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32520_cast = slice_by_index(begin = var_32520_begin_0, end = var_32520_end_0, end_mask = var_32520_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32520_cast")];
            tensor<int32, [4]> var_32524_begin_0 = const()[name = tensor<string, []>("op_32524_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_32524_end_0 = const()[name = tensor<string, []>("op_32524_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_32524_end_mask_0 = const()[name = tensor<string, []>("op_32524_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32524_cast = slice_by_index(begin = var_32524_begin_0, end = var_32524_end_0, end_mask = var_32524_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32524_cast")];
            tensor<int32, [4]> var_32528_begin_0 = const()[name = tensor<string, []>("op_32528_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_32528_end_0 = const()[name = tensor<string, []>("op_32528_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_32528_end_mask_0 = const()[name = tensor<string, []>("op_32528_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32528_cast = slice_by_index(begin = var_32528_begin_0, end = var_32528_end_0, end_mask = var_32528_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32528_cast")];
            tensor<int32, [4]> var_32532_begin_0 = const()[name = tensor<string, []>("op_32532_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_32532_end_0 = const()[name = tensor<string, []>("op_32532_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_32532_end_mask_0 = const()[name = tensor<string, []>("op_32532_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32532_cast = slice_by_index(begin = var_32532_begin_0, end = var_32532_end_0, end_mask = var_32532_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32532_cast")];
            tensor<int32, [4]> var_32536_begin_0 = const()[name = tensor<string, []>("op_32536_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_32536_end_0 = const()[name = tensor<string, []>("op_32536_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_32536_end_mask_0 = const()[name = tensor<string, []>("op_32536_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32536_cast = slice_by_index(begin = var_32536_begin_0, end = var_32536_end_0, end_mask = var_32536_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32536_cast")];
            tensor<int32, [4]> var_32540_begin_0 = const()[name = tensor<string, []>("op_32540_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_32540_end_0 = const()[name = tensor<string, []>("op_32540_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_32540_end_mask_0 = const()[name = tensor<string, []>("op_32540_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32540_cast = slice_by_index(begin = var_32540_begin_0, end = var_32540_end_0, end_mask = var_32540_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32540_cast")];
            tensor<int32, [4]> var_32544_begin_0 = const()[name = tensor<string, []>("op_32544_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_32544_end_0 = const()[name = tensor<string, []>("op_32544_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_32544_end_mask_0 = const()[name = tensor<string, []>("op_32544_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32544_cast = slice_by_index(begin = var_32544_begin_0, end = var_32544_end_0, end_mask = var_32544_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32544_cast")];
            tensor<int32, [4]> var_32548_begin_0 = const()[name = tensor<string, []>("op_32548_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_32548_end_0 = const()[name = tensor<string, []>("op_32548_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_32548_end_mask_0 = const()[name = tensor<string, []>("op_32548_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32548_cast = slice_by_index(begin = var_32548_begin_0, end = var_32548_end_0, end_mask = var_32548_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32548_cast")];
            tensor<int32, [4]> var_32552_begin_0 = const()[name = tensor<string, []>("op_32552_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_32552_end_0 = const()[name = tensor<string, []>("op_32552_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_32552_end_mask_0 = const()[name = tensor<string, []>("op_32552_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32552_cast = slice_by_index(begin = var_32552_begin_0, end = var_32552_end_0, end_mask = var_32552_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32552_cast")];
            tensor<int32, [4]> var_32556_begin_0 = const()[name = tensor<string, []>("op_32556_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_32556_end_0 = const()[name = tensor<string, []>("op_32556_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_32556_end_mask_0 = const()[name = tensor<string, []>("op_32556_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32556_cast = slice_by_index(begin = var_32556_begin_0, end = var_32556_end_0, end_mask = var_32556_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32556_cast")];
            tensor<int32, [4]> var_32560_begin_0 = const()[name = tensor<string, []>("op_32560_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_32560_end_0 = const()[name = tensor<string, []>("op_32560_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_32560_end_mask_0 = const()[name = tensor<string, []>("op_32560_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32560_cast = slice_by_index(begin = var_32560_begin_0, end = var_32560_end_0, end_mask = var_32560_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32560_cast")];
            tensor<int32, [4]> var_32564_begin_0 = const()[name = tensor<string, []>("op_32564_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_32564_end_0 = const()[name = tensor<string, []>("op_32564_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_32564_end_mask_0 = const()[name = tensor<string, []>("op_32564_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32564_cast = slice_by_index(begin = var_32564_begin_0, end = var_32564_end_0, end_mask = var_32564_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32564_cast")];
            tensor<int32, [4]> var_32568_begin_0 = const()[name = tensor<string, []>("op_32568_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_32568_end_0 = const()[name = tensor<string, []>("op_32568_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_32568_end_mask_0 = const()[name = tensor<string, []>("op_32568_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32568_cast = slice_by_index(begin = var_32568_begin_0, end = var_32568_end_0, end_mask = var_32568_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32568_cast")];
            tensor<int32, [4]> var_32572_begin_0 = const()[name = tensor<string, []>("op_32572_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_32572_end_0 = const()[name = tensor<string, []>("op_32572_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_32572_end_mask_0 = const()[name = tensor<string, []>("op_32572_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32572_cast = slice_by_index(begin = var_32572_begin_0, end = var_32572_end_0, end_mask = var_32572_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32572_cast")];
            tensor<int32, [4]> var_32576_begin_0 = const()[name = tensor<string, []>("op_32576_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_32576_end_0 = const()[name = tensor<string, []>("op_32576_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_32576_end_mask_0 = const()[name = tensor<string, []>("op_32576_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32576_cast = slice_by_index(begin = var_32576_begin_0, end = var_32576_end_0, end_mask = var_32576_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32576_cast")];
            tensor<int32, [4]> var_32580_begin_0 = const()[name = tensor<string, []>("op_32580_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_32580_end_0 = const()[name = tensor<string, []>("op_32580_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_32580_end_mask_0 = const()[name = tensor<string, []>("op_32580_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32580_cast = slice_by_index(begin = var_32580_begin_0, end = var_32580_end_0, end_mask = var_32580_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32580_cast")];
            tensor<int32, [4]> var_32584_begin_0 = const()[name = tensor<string, []>("op_32584_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_32584_end_0 = const()[name = tensor<string, []>("op_32584_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_32584_end_mask_0 = const()[name = tensor<string, []>("op_32584_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32584_cast = slice_by_index(begin = var_32584_begin_0, end = var_32584_end_0, end_mask = var_32584_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32584_cast")];
            tensor<int32, [4]> var_32588_begin_0 = const()[name = tensor<string, []>("op_32588_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_32588_end_0 = const()[name = tensor<string, []>("op_32588_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_32588_end_mask_0 = const()[name = tensor<string, []>("op_32588_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32588_cast = slice_by_index(begin = var_32588_begin_0, end = var_32588_end_0, end_mask = var_32588_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32588_cast")];
            tensor<int32, [4]> var_32592_begin_0 = const()[name = tensor<string, []>("op_32592_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_32592_end_0 = const()[name = tensor<string, []>("op_32592_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_32592_end_mask_0 = const()[name = tensor<string, []>("op_32592_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32592_cast = slice_by_index(begin = var_32592_begin_0, end = var_32592_end_0, end_mask = var_32592_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32592_cast")];
            tensor<int32, [4]> var_32596_begin_0 = const()[name = tensor<string, []>("op_32596_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_32596_end_0 = const()[name = tensor<string, []>("op_32596_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_32596_end_mask_0 = const()[name = tensor<string, []>("op_32596_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32596_cast = slice_by_index(begin = var_32596_begin_0, end = var_32596_end_0, end_mask = var_32596_end_mask_0, x = v_145_cast)[name = tensor<string, []>("op_32596_cast")];
            tensor<string, []> var_32600_equation_0 = const()[name = tensor<string, []>("op_32600_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32600_cast = einsum(equation = var_32600_equation_0, values = (var_32442_cast, var_32359_cast))[name = tensor<string, []>("op_32600_cast")];
            tensor<fp16, []> var_32601_to_fp16 = const()[name = tensor<string, []>("op_32601_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2721_cast = mul(x = var_32600_cast, y = var_32601_to_fp16)[name = tensor<string, []>("aw_2721_cast")];
            tensor<string, []> var_32604_equation_0 = const()[name = tensor<string, []>("op_32604_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32604_cast = einsum(equation = var_32604_equation_0, values = (var_32446_cast, var_32363_cast))[name = tensor<string, []>("op_32604_cast")];
            tensor<fp16, []> var_32605_to_fp16 = const()[name = tensor<string, []>("op_32605_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2723_cast = mul(x = var_32604_cast, y = var_32605_to_fp16)[name = tensor<string, []>("aw_2723_cast")];
            tensor<string, []> var_32608_equation_0 = const()[name = tensor<string, []>("op_32608_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32608_cast = einsum(equation = var_32608_equation_0, values = (var_32450_cast, var_32367_cast))[name = tensor<string, []>("op_32608_cast")];
            tensor<fp16, []> var_32609_to_fp16 = const()[name = tensor<string, []>("op_32609_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2725_cast = mul(x = var_32608_cast, y = var_32609_to_fp16)[name = tensor<string, []>("aw_2725_cast")];
            tensor<string, []> var_32612_equation_0 = const()[name = tensor<string, []>("op_32612_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32612_cast = einsum(equation = var_32612_equation_0, values = (var_32454_cast, var_32371_cast))[name = tensor<string, []>("op_32612_cast")];
            tensor<fp16, []> var_32613_to_fp16 = const()[name = tensor<string, []>("op_32613_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2727_cast = mul(x = var_32612_cast, y = var_32613_to_fp16)[name = tensor<string, []>("aw_2727_cast")];
            tensor<string, []> var_32616_equation_0 = const()[name = tensor<string, []>("op_32616_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32616_cast = einsum(equation = var_32616_equation_0, values = (var_32458_cast, var_32375_cast))[name = tensor<string, []>("op_32616_cast")];
            tensor<fp16, []> var_32617_to_fp16 = const()[name = tensor<string, []>("op_32617_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2729_cast = mul(x = var_32616_cast, y = var_32617_to_fp16)[name = tensor<string, []>("aw_2729_cast")];
            tensor<string, []> var_32620_equation_0 = const()[name = tensor<string, []>("op_32620_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32620_cast = einsum(equation = var_32620_equation_0, values = (var_32462_cast, var_32379_cast))[name = tensor<string, []>("op_32620_cast")];
            tensor<fp16, []> var_32621_to_fp16 = const()[name = tensor<string, []>("op_32621_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2731_cast = mul(x = var_32620_cast, y = var_32621_to_fp16)[name = tensor<string, []>("aw_2731_cast")];
            tensor<string, []> var_32624_equation_0 = const()[name = tensor<string, []>("op_32624_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32624_cast = einsum(equation = var_32624_equation_0, values = (var_32466_cast, var_32383_cast))[name = tensor<string, []>("op_32624_cast")];
            tensor<fp16, []> var_32625_to_fp16 = const()[name = tensor<string, []>("op_32625_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2733_cast = mul(x = var_32624_cast, y = var_32625_to_fp16)[name = tensor<string, []>("aw_2733_cast")];
            tensor<string, []> var_32628_equation_0 = const()[name = tensor<string, []>("op_32628_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32628_cast = einsum(equation = var_32628_equation_0, values = (var_32470_cast, var_32387_cast))[name = tensor<string, []>("op_32628_cast")];
            tensor<fp16, []> var_32629_to_fp16 = const()[name = tensor<string, []>("op_32629_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2735_cast = mul(x = var_32628_cast, y = var_32629_to_fp16)[name = tensor<string, []>("aw_2735_cast")];
            tensor<string, []> var_32632_equation_0 = const()[name = tensor<string, []>("op_32632_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32632_cast = einsum(equation = var_32632_equation_0, values = (var_32474_cast, var_32391_cast))[name = tensor<string, []>("op_32632_cast")];
            tensor<fp16, []> var_32633_to_fp16 = const()[name = tensor<string, []>("op_32633_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2737_cast = mul(x = var_32632_cast, y = var_32633_to_fp16)[name = tensor<string, []>("aw_2737_cast")];
            tensor<string, []> var_32636_equation_0 = const()[name = tensor<string, []>("op_32636_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32636_cast = einsum(equation = var_32636_equation_0, values = (var_32478_cast, var_32395_cast))[name = tensor<string, []>("op_32636_cast")];
            tensor<fp16, []> var_32637_to_fp16 = const()[name = tensor<string, []>("op_32637_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2739_cast = mul(x = var_32636_cast, y = var_32637_to_fp16)[name = tensor<string, []>("aw_2739_cast")];
            tensor<string, []> var_32640_equation_0 = const()[name = tensor<string, []>("op_32640_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32640_cast = einsum(equation = var_32640_equation_0, values = (var_32482_cast, var_32399_cast))[name = tensor<string, []>("op_32640_cast")];
            tensor<fp16, []> var_32641_to_fp16 = const()[name = tensor<string, []>("op_32641_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2741_cast = mul(x = var_32640_cast, y = var_32641_to_fp16)[name = tensor<string, []>("aw_2741_cast")];
            tensor<string, []> var_32644_equation_0 = const()[name = tensor<string, []>("op_32644_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32644_cast = einsum(equation = var_32644_equation_0, values = (var_32486_cast, var_32403_cast))[name = tensor<string, []>("op_32644_cast")];
            tensor<fp16, []> var_32645_to_fp16 = const()[name = tensor<string, []>("op_32645_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2743_cast = mul(x = var_32644_cast, y = var_32645_to_fp16)[name = tensor<string, []>("aw_2743_cast")];
            tensor<string, []> var_32648_equation_0 = const()[name = tensor<string, []>("op_32648_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32648_cast = einsum(equation = var_32648_equation_0, values = (var_32490_cast, var_32407_cast))[name = tensor<string, []>("op_32648_cast")];
            tensor<fp16, []> var_32649_to_fp16 = const()[name = tensor<string, []>("op_32649_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2745_cast = mul(x = var_32648_cast, y = var_32649_to_fp16)[name = tensor<string, []>("aw_2745_cast")];
            tensor<string, []> var_32652_equation_0 = const()[name = tensor<string, []>("op_32652_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32652_cast = einsum(equation = var_32652_equation_0, values = (var_32494_cast, var_32411_cast))[name = tensor<string, []>("op_32652_cast")];
            tensor<fp16, []> var_32653_to_fp16 = const()[name = tensor<string, []>("op_32653_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2747_cast = mul(x = var_32652_cast, y = var_32653_to_fp16)[name = tensor<string, []>("aw_2747_cast")];
            tensor<string, []> var_32656_equation_0 = const()[name = tensor<string, []>("op_32656_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32656_cast = einsum(equation = var_32656_equation_0, values = (var_32498_cast, var_32415_cast))[name = tensor<string, []>("op_32656_cast")];
            tensor<fp16, []> var_32657_to_fp16 = const()[name = tensor<string, []>("op_32657_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2749_cast = mul(x = var_32656_cast, y = var_32657_to_fp16)[name = tensor<string, []>("aw_2749_cast")];
            tensor<string, []> var_32660_equation_0 = const()[name = tensor<string, []>("op_32660_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32660_cast = einsum(equation = var_32660_equation_0, values = (var_32502_cast, var_32419_cast))[name = tensor<string, []>("op_32660_cast")];
            tensor<fp16, []> var_32661_to_fp16 = const()[name = tensor<string, []>("op_32661_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2751_cast = mul(x = var_32660_cast, y = var_32661_to_fp16)[name = tensor<string, []>("aw_2751_cast")];
            tensor<string, []> var_32664_equation_0 = const()[name = tensor<string, []>("op_32664_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32664_cast = einsum(equation = var_32664_equation_0, values = (var_32506_cast, var_32423_cast))[name = tensor<string, []>("op_32664_cast")];
            tensor<fp16, []> var_32665_to_fp16 = const()[name = tensor<string, []>("op_32665_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2753_cast = mul(x = var_32664_cast, y = var_32665_to_fp16)[name = tensor<string, []>("aw_2753_cast")];
            tensor<string, []> var_32668_equation_0 = const()[name = tensor<string, []>("op_32668_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32668_cast = einsum(equation = var_32668_equation_0, values = (var_32510_cast, var_32427_cast))[name = tensor<string, []>("op_32668_cast")];
            tensor<fp16, []> var_32669_to_fp16 = const()[name = tensor<string, []>("op_32669_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2755_cast = mul(x = var_32668_cast, y = var_32669_to_fp16)[name = tensor<string, []>("aw_2755_cast")];
            tensor<string, []> var_32672_equation_0 = const()[name = tensor<string, []>("op_32672_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32672_cast = einsum(equation = var_32672_equation_0, values = (var_32514_cast, var_32431_cast))[name = tensor<string, []>("op_32672_cast")];
            tensor<fp16, []> var_32673_to_fp16 = const()[name = tensor<string, []>("op_32673_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2757_cast = mul(x = var_32672_cast, y = var_32673_to_fp16)[name = tensor<string, []>("aw_2757_cast")];
            tensor<string, []> var_32676_equation_0 = const()[name = tensor<string, []>("op_32676_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32676_cast = einsum(equation = var_32676_equation_0, values = (var_32518_cast, var_32435_cast))[name = tensor<string, []>("op_32676_cast")];
            tensor<fp16, []> var_32677_to_fp16 = const()[name = tensor<string, []>("op_32677_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2759_cast = mul(x = var_32676_cast, y = var_32677_to_fp16)[name = tensor<string, []>("aw_2759_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32679_cast = softmax(axis = var_30385, x = aw_2721_cast)[name = tensor<string, []>("op_32679_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32680_cast = softmax(axis = var_30385, x = aw_2723_cast)[name = tensor<string, []>("op_32680_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32681_cast = softmax(axis = var_30385, x = aw_2725_cast)[name = tensor<string, []>("op_32681_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32682_cast = softmax(axis = var_30385, x = aw_2727_cast)[name = tensor<string, []>("op_32682_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32683_cast = softmax(axis = var_30385, x = aw_2729_cast)[name = tensor<string, []>("op_32683_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32684_cast = softmax(axis = var_30385, x = aw_2731_cast)[name = tensor<string, []>("op_32684_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32685_cast = softmax(axis = var_30385, x = aw_2733_cast)[name = tensor<string, []>("op_32685_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32686_cast = softmax(axis = var_30385, x = aw_2735_cast)[name = tensor<string, []>("op_32686_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32687_cast = softmax(axis = var_30385, x = aw_2737_cast)[name = tensor<string, []>("op_32687_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32688_cast = softmax(axis = var_30385, x = aw_2739_cast)[name = tensor<string, []>("op_32688_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32689_cast = softmax(axis = var_30385, x = aw_2741_cast)[name = tensor<string, []>("op_32689_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32690_cast = softmax(axis = var_30385, x = aw_2743_cast)[name = tensor<string, []>("op_32690_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32691_cast = softmax(axis = var_30385, x = aw_2745_cast)[name = tensor<string, []>("op_32691_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32692_cast = softmax(axis = var_30385, x = aw_2747_cast)[name = tensor<string, []>("op_32692_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32693_cast = softmax(axis = var_30385, x = aw_2749_cast)[name = tensor<string, []>("op_32693_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32694_cast = softmax(axis = var_30385, x = aw_2751_cast)[name = tensor<string, []>("op_32694_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32695_cast = softmax(axis = var_30385, x = aw_2753_cast)[name = tensor<string, []>("op_32695_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32696_cast = softmax(axis = var_30385, x = aw_2755_cast)[name = tensor<string, []>("op_32696_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32697_cast = softmax(axis = var_30385, x = aw_2757_cast)[name = tensor<string, []>("op_32697_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_32698_cast = softmax(axis = var_30385, x = aw_2759_cast)[name = tensor<string, []>("op_32698_cast")];
            tensor<string, []> var_32700_equation_0 = const()[name = tensor<string, []>("op_32700_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32700_cast = einsum(equation = var_32700_equation_0, values = (var_32520_cast, var_32679_cast))[name = tensor<string, []>("op_32700_cast")];
            tensor<string, []> var_32702_equation_0 = const()[name = tensor<string, []>("op_32702_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32702_cast = einsum(equation = var_32702_equation_0, values = (var_32524_cast, var_32680_cast))[name = tensor<string, []>("op_32702_cast")];
            tensor<string, []> var_32704_equation_0 = const()[name = tensor<string, []>("op_32704_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32704_cast = einsum(equation = var_32704_equation_0, values = (var_32528_cast, var_32681_cast))[name = tensor<string, []>("op_32704_cast")];
            tensor<string, []> var_32706_equation_0 = const()[name = tensor<string, []>("op_32706_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32706_cast = einsum(equation = var_32706_equation_0, values = (var_32532_cast, var_32682_cast))[name = tensor<string, []>("op_32706_cast")];
            tensor<string, []> var_32708_equation_0 = const()[name = tensor<string, []>("op_32708_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32708_cast = einsum(equation = var_32708_equation_0, values = (var_32536_cast, var_32683_cast))[name = tensor<string, []>("op_32708_cast")];
            tensor<string, []> var_32710_equation_0 = const()[name = tensor<string, []>("op_32710_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32710_cast = einsum(equation = var_32710_equation_0, values = (var_32540_cast, var_32684_cast))[name = tensor<string, []>("op_32710_cast")];
            tensor<string, []> var_32712_equation_0 = const()[name = tensor<string, []>("op_32712_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32712_cast = einsum(equation = var_32712_equation_0, values = (var_32544_cast, var_32685_cast))[name = tensor<string, []>("op_32712_cast")];
            tensor<string, []> var_32714_equation_0 = const()[name = tensor<string, []>("op_32714_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32714_cast = einsum(equation = var_32714_equation_0, values = (var_32548_cast, var_32686_cast))[name = tensor<string, []>("op_32714_cast")];
            tensor<string, []> var_32716_equation_0 = const()[name = tensor<string, []>("op_32716_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32716_cast = einsum(equation = var_32716_equation_0, values = (var_32552_cast, var_32687_cast))[name = tensor<string, []>("op_32716_cast")];
            tensor<string, []> var_32718_equation_0 = const()[name = tensor<string, []>("op_32718_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32718_cast = einsum(equation = var_32718_equation_0, values = (var_32556_cast, var_32688_cast))[name = tensor<string, []>("op_32718_cast")];
            tensor<string, []> var_32720_equation_0 = const()[name = tensor<string, []>("op_32720_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32720_cast = einsum(equation = var_32720_equation_0, values = (var_32560_cast, var_32689_cast))[name = tensor<string, []>("op_32720_cast")];
            tensor<string, []> var_32722_equation_0 = const()[name = tensor<string, []>("op_32722_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32722_cast = einsum(equation = var_32722_equation_0, values = (var_32564_cast, var_32690_cast))[name = tensor<string, []>("op_32722_cast")];
            tensor<string, []> var_32724_equation_0 = const()[name = tensor<string, []>("op_32724_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32724_cast = einsum(equation = var_32724_equation_0, values = (var_32568_cast, var_32691_cast))[name = tensor<string, []>("op_32724_cast")];
            tensor<string, []> var_32726_equation_0 = const()[name = tensor<string, []>("op_32726_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32726_cast = einsum(equation = var_32726_equation_0, values = (var_32572_cast, var_32692_cast))[name = tensor<string, []>("op_32726_cast")];
            tensor<string, []> var_32728_equation_0 = const()[name = tensor<string, []>("op_32728_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32728_cast = einsum(equation = var_32728_equation_0, values = (var_32576_cast, var_32693_cast))[name = tensor<string, []>("op_32728_cast")];
            tensor<string, []> var_32730_equation_0 = const()[name = tensor<string, []>("op_32730_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32730_cast = einsum(equation = var_32730_equation_0, values = (var_32580_cast, var_32694_cast))[name = tensor<string, []>("op_32730_cast")];
            tensor<string, []> var_32732_equation_0 = const()[name = tensor<string, []>("op_32732_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32732_cast = einsum(equation = var_32732_equation_0, values = (var_32584_cast, var_32695_cast))[name = tensor<string, []>("op_32732_cast")];
            tensor<string, []> var_32734_equation_0 = const()[name = tensor<string, []>("op_32734_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32734_cast = einsum(equation = var_32734_equation_0, values = (var_32588_cast, var_32696_cast))[name = tensor<string, []>("op_32734_cast")];
            tensor<string, []> var_32736_equation_0 = const()[name = tensor<string, []>("op_32736_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32736_cast = einsum(equation = var_32736_equation_0, values = (var_32592_cast, var_32697_cast))[name = tensor<string, []>("op_32736_cast")];
            tensor<string, []> var_32738_equation_0 = const()[name = tensor<string, []>("op_32738_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_32738_cast = einsum(equation = var_32738_equation_0, values = (var_32596_cast, var_32698_cast))[name = tensor<string, []>("op_32738_cast")];
            tensor<bool, []> input_457_interleave_0 = const()[name = tensor<string, []>("input_457_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_457_cast = concat(axis = var_30385, interleave = input_457_interleave_0, values = (var_32700_cast, var_32702_cast, var_32704_cast, var_32706_cast, var_32708_cast, var_32710_cast, var_32712_cast, var_32714_cast, var_32716_cast, var_32718_cast, var_32720_cast, var_32722_cast, var_32724_cast, var_32726_cast, var_32728_cast, var_32730_cast, var_32732_cast, var_32734_cast, var_32736_cast, var_32738_cast))[name = tensor<string, []>("input_457_cast")];
            tensor<int32, [2]> var_32744 = const()[name = tensor<string, []>("op_32744"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_32746 = const()[name = tensor<string, []>("op_32746"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_32748_pad_type_0 = const()[name = tensor<string, []>("op_32748_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_32748_pad_0 = const()[name = tensor<string, []>("op_32748_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2737283136)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2740560000)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_32748_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_out_0_bias_to_fp16, dilations = var_32746, groups = var_30385, pad = var_32748_pad_0, pad_type = var_32748_pad_type_0, strides = var_32744, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn1_to_out_0_weight_to_fp16, x = input_457_cast)[name = tensor<string, []>("op_32748_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_219_cast = add(x = var_32748_cast, y = inputs_217_cast)[name = tensor<string, []>("inputs_219_cast")];
            tensor<int32, [1]> var_32752 = const()[name = tensor<string, []>("op_32752"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_219_cast = reduce_mean(axes = var_32752, keep_dims = var_30380, x = inputs_219_cast)[name = tensor<string, []>("channels_mean_219_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_219_cast = sub(x = inputs_219_cast, y = channels_mean_219_cast)[name = tensor<string, []>("zero_mean_219_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_219_cast = mul(x = zero_mean_219_cast, y = zero_mean_219_cast)[name = tensor<string, []>("zero_mean_sq_219_cast")];
            tensor<int32, [1]> var_32756 = const()[name = tensor<string, []>("op_32756"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_32757_cast = reduce_mean(axes = var_32756, keep_dims = var_30380, x = zero_mean_sq_219_cast)[name = tensor<string, []>("op_32757_cast")];
            tensor<fp16, []> var_32758_to_fp16 = const()[name = tensor<string, []>("op_32758_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_32759_cast = add(x = var_32757_cast, y = var_32758_to_fp16)[name = tensor<string, []>("op_32759_cast")];
            tensor<fp16, []> denom_219_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_219_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_219_cast = rsqrt(epsilon = denom_219_epsilon_0_to_fp16, x = var_32759_cast)[name = tensor<string, []>("denom_219_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_219_cast = mul(x = zero_mean_219_cast, y = denom_219_cast)[name = tensor<string, []>("out_219_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_32763_to_fp16 = const()[name = tensor<string, []>("op_32763_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2740562624)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_32764_cast = add(x = out_219_cast, y = var_32763_to_fp16)[name = tensor<string, []>("op_32764_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_32766_to_fp16 = const()[name = tensor<string, []>("op_32766_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2740565248)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_301_cast = mul(x = var_32764_cast, y = var_32766_to_fp16)[name = tensor<string, []>("hidden_states_301_cast")];
            tensor<int32, [2]> var_32773 = const()[name = tensor<string, []>("op_32773"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_32775 = const()[name = tensor<string, []>("op_32775"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_147_pad_type_0 = const()[name = tensor<string, []>("q_147_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_147_pad_0 = const()[name = tensor<string, []>("q_147_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2740567872)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_147_cast = conv(dilations = var_32775, groups = var_30385, pad = q_147_pad_0, pad_type = q_147_pad_type_0, strides = var_32773, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_q_weight_to_fp16, x = hidden_states_301_cast)[name = tensor<string, []>("q_147_cast")];
            tensor<int32, [2]> var_32779 = const()[name = tensor<string, []>("op_32779"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_32781 = const()[name = tensor<string, []>("op_32781"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_293_pad_type_0 = const()[name = tensor<string, []>("k_293_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_293_pad_0 = const()[name = tensor<string, []>("k_293_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2743844736)))];
            tensor<fp16, [2, 1280, 1, 77]> k_293_cast = conv(dilations = var_32781, groups = var_30385, pad = k_293_pad_0, pad_type = k_293_pad_type_0, strides = var_32779, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_293_cast")];
            tensor<int32, [2]> var_32785 = const()[name = tensor<string, []>("op_32785"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_32787 = const()[name = tensor<string, []>("op_32787"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_147_pad_type_0 = const()[name = tensor<string, []>("v_147_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_147_pad_0 = const()[name = tensor<string, []>("v_147_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2749087680)))];
            tensor<fp16, [2, 1280, 1, 77]> v_147_cast = conv(dilations = var_32787, groups = var_30385, pad = v_147_pad_0, pad_type = v_147_pad_type_0, strides = var_32785, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_147_cast")];
            tensor<int32, [4]> var_32791_begin_0 = const()[name = tensor<string, []>("op_32791_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_32791_end_0 = const()[name = tensor<string, []>("op_32791_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_32791_end_mask_0 = const()[name = tensor<string, []>("op_32791_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32791_cast = slice_by_index(begin = var_32791_begin_0, end = var_32791_end_0, end_mask = var_32791_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32791_cast")];
            tensor<int32, [4]> var_32795_begin_0 = const()[name = tensor<string, []>("op_32795_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_32795_end_0 = const()[name = tensor<string, []>("op_32795_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_32795_end_mask_0 = const()[name = tensor<string, []>("op_32795_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32795_cast = slice_by_index(begin = var_32795_begin_0, end = var_32795_end_0, end_mask = var_32795_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32795_cast")];
            tensor<int32, [4]> var_32799_begin_0 = const()[name = tensor<string, []>("op_32799_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_32799_end_0 = const()[name = tensor<string, []>("op_32799_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_32799_end_mask_0 = const()[name = tensor<string, []>("op_32799_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32799_cast = slice_by_index(begin = var_32799_begin_0, end = var_32799_end_0, end_mask = var_32799_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32799_cast")];
            tensor<int32, [4]> var_32803_begin_0 = const()[name = tensor<string, []>("op_32803_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_32803_end_0 = const()[name = tensor<string, []>("op_32803_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_32803_end_mask_0 = const()[name = tensor<string, []>("op_32803_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32803_cast = slice_by_index(begin = var_32803_begin_0, end = var_32803_end_0, end_mask = var_32803_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32803_cast")];
            tensor<int32, [4]> var_32807_begin_0 = const()[name = tensor<string, []>("op_32807_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_32807_end_0 = const()[name = tensor<string, []>("op_32807_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_32807_end_mask_0 = const()[name = tensor<string, []>("op_32807_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32807_cast = slice_by_index(begin = var_32807_begin_0, end = var_32807_end_0, end_mask = var_32807_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32807_cast")];
            tensor<int32, [4]> var_32811_begin_0 = const()[name = tensor<string, []>("op_32811_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_32811_end_0 = const()[name = tensor<string, []>("op_32811_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_32811_end_mask_0 = const()[name = tensor<string, []>("op_32811_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32811_cast = slice_by_index(begin = var_32811_begin_0, end = var_32811_end_0, end_mask = var_32811_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32811_cast")];
            tensor<int32, [4]> var_32815_begin_0 = const()[name = tensor<string, []>("op_32815_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_32815_end_0 = const()[name = tensor<string, []>("op_32815_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_32815_end_mask_0 = const()[name = tensor<string, []>("op_32815_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32815_cast = slice_by_index(begin = var_32815_begin_0, end = var_32815_end_0, end_mask = var_32815_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32815_cast")];
            tensor<int32, [4]> var_32819_begin_0 = const()[name = tensor<string, []>("op_32819_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_32819_end_0 = const()[name = tensor<string, []>("op_32819_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_32819_end_mask_0 = const()[name = tensor<string, []>("op_32819_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32819_cast = slice_by_index(begin = var_32819_begin_0, end = var_32819_end_0, end_mask = var_32819_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32819_cast")];
            tensor<int32, [4]> var_32823_begin_0 = const()[name = tensor<string, []>("op_32823_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_32823_end_0 = const()[name = tensor<string, []>("op_32823_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_32823_end_mask_0 = const()[name = tensor<string, []>("op_32823_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32823_cast = slice_by_index(begin = var_32823_begin_0, end = var_32823_end_0, end_mask = var_32823_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32823_cast")];
            tensor<int32, [4]> var_32827_begin_0 = const()[name = tensor<string, []>("op_32827_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_32827_end_0 = const()[name = tensor<string, []>("op_32827_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_32827_end_mask_0 = const()[name = tensor<string, []>("op_32827_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32827_cast = slice_by_index(begin = var_32827_begin_0, end = var_32827_end_0, end_mask = var_32827_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32827_cast")];
            tensor<int32, [4]> var_32831_begin_0 = const()[name = tensor<string, []>("op_32831_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_32831_end_0 = const()[name = tensor<string, []>("op_32831_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_32831_end_mask_0 = const()[name = tensor<string, []>("op_32831_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32831_cast = slice_by_index(begin = var_32831_begin_0, end = var_32831_end_0, end_mask = var_32831_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32831_cast")];
            tensor<int32, [4]> var_32835_begin_0 = const()[name = tensor<string, []>("op_32835_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_32835_end_0 = const()[name = tensor<string, []>("op_32835_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_32835_end_mask_0 = const()[name = tensor<string, []>("op_32835_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32835_cast = slice_by_index(begin = var_32835_begin_0, end = var_32835_end_0, end_mask = var_32835_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32835_cast")];
            tensor<int32, [4]> var_32839_begin_0 = const()[name = tensor<string, []>("op_32839_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_32839_end_0 = const()[name = tensor<string, []>("op_32839_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_32839_end_mask_0 = const()[name = tensor<string, []>("op_32839_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32839_cast = slice_by_index(begin = var_32839_begin_0, end = var_32839_end_0, end_mask = var_32839_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32839_cast")];
            tensor<int32, [4]> var_32843_begin_0 = const()[name = tensor<string, []>("op_32843_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_32843_end_0 = const()[name = tensor<string, []>("op_32843_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_32843_end_mask_0 = const()[name = tensor<string, []>("op_32843_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32843_cast = slice_by_index(begin = var_32843_begin_0, end = var_32843_end_0, end_mask = var_32843_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32843_cast")];
            tensor<int32, [4]> var_32847_begin_0 = const()[name = tensor<string, []>("op_32847_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_32847_end_0 = const()[name = tensor<string, []>("op_32847_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_32847_end_mask_0 = const()[name = tensor<string, []>("op_32847_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32847_cast = slice_by_index(begin = var_32847_begin_0, end = var_32847_end_0, end_mask = var_32847_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32847_cast")];
            tensor<int32, [4]> var_32851_begin_0 = const()[name = tensor<string, []>("op_32851_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_32851_end_0 = const()[name = tensor<string, []>("op_32851_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_32851_end_mask_0 = const()[name = tensor<string, []>("op_32851_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32851_cast = slice_by_index(begin = var_32851_begin_0, end = var_32851_end_0, end_mask = var_32851_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32851_cast")];
            tensor<int32, [4]> var_32855_begin_0 = const()[name = tensor<string, []>("op_32855_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_32855_end_0 = const()[name = tensor<string, []>("op_32855_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_32855_end_mask_0 = const()[name = tensor<string, []>("op_32855_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32855_cast = slice_by_index(begin = var_32855_begin_0, end = var_32855_end_0, end_mask = var_32855_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32855_cast")];
            tensor<int32, [4]> var_32859_begin_0 = const()[name = tensor<string, []>("op_32859_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_32859_end_0 = const()[name = tensor<string, []>("op_32859_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_32859_end_mask_0 = const()[name = tensor<string, []>("op_32859_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32859_cast = slice_by_index(begin = var_32859_begin_0, end = var_32859_end_0, end_mask = var_32859_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32859_cast")];
            tensor<int32, [4]> var_32863_begin_0 = const()[name = tensor<string, []>("op_32863_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_32863_end_0 = const()[name = tensor<string, []>("op_32863_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_32863_end_mask_0 = const()[name = tensor<string, []>("op_32863_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32863_cast = slice_by_index(begin = var_32863_begin_0, end = var_32863_end_0, end_mask = var_32863_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32863_cast")];
            tensor<int32, [4]> var_32867_begin_0 = const()[name = tensor<string, []>("op_32867_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_32867_end_0 = const()[name = tensor<string, []>("op_32867_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_32867_end_mask_0 = const()[name = tensor<string, []>("op_32867_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_32867_cast = slice_by_index(begin = var_32867_begin_0, end = var_32867_end_0, end_mask = var_32867_end_mask_0, x = q_147_cast)[name = tensor<string, []>("op_32867_cast")];
            tensor<int32, [4]> k_295_perm_0 = const()[name = tensor<string, []>("k_295_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_32874_begin_0 = const()[name = tensor<string, []>("op_32874_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_32874_end_0 = const()[name = tensor<string, []>("op_32874_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_32874_end_mask_0 = const()[name = tensor<string, []>("op_32874_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_66 = transpose(perm = k_295_perm_0, x = k_293_cast)[name = tensor<string, []>("transpose_66")];
            tensor<fp16, [2, 77, 1, 64]> var_32874_cast = slice_by_index(begin = var_32874_begin_0, end = var_32874_end_0, end_mask = var_32874_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32874_cast")];
            tensor<int32, [4]> var_32878_begin_0 = const()[name = tensor<string, []>("op_32878_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_32878_end_0 = const()[name = tensor<string, []>("op_32878_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_32878_end_mask_0 = const()[name = tensor<string, []>("op_32878_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32878_cast = slice_by_index(begin = var_32878_begin_0, end = var_32878_end_0, end_mask = var_32878_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32878_cast")];
            tensor<int32, [4]> var_32882_begin_0 = const()[name = tensor<string, []>("op_32882_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_32882_end_0 = const()[name = tensor<string, []>("op_32882_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_32882_end_mask_0 = const()[name = tensor<string, []>("op_32882_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32882_cast = slice_by_index(begin = var_32882_begin_0, end = var_32882_end_0, end_mask = var_32882_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32882_cast")];
            tensor<int32, [4]> var_32886_begin_0 = const()[name = tensor<string, []>("op_32886_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_32886_end_0 = const()[name = tensor<string, []>("op_32886_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_32886_end_mask_0 = const()[name = tensor<string, []>("op_32886_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32886_cast = slice_by_index(begin = var_32886_begin_0, end = var_32886_end_0, end_mask = var_32886_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32886_cast")];
            tensor<int32, [4]> var_32890_begin_0 = const()[name = tensor<string, []>("op_32890_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_32890_end_0 = const()[name = tensor<string, []>("op_32890_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_32890_end_mask_0 = const()[name = tensor<string, []>("op_32890_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32890_cast = slice_by_index(begin = var_32890_begin_0, end = var_32890_end_0, end_mask = var_32890_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32890_cast")];
            tensor<int32, [4]> var_32894_begin_0 = const()[name = tensor<string, []>("op_32894_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_32894_end_0 = const()[name = tensor<string, []>("op_32894_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_32894_end_mask_0 = const()[name = tensor<string, []>("op_32894_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32894_cast = slice_by_index(begin = var_32894_begin_0, end = var_32894_end_0, end_mask = var_32894_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32894_cast")];
            tensor<int32, [4]> var_32898_begin_0 = const()[name = tensor<string, []>("op_32898_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_32898_end_0 = const()[name = tensor<string, []>("op_32898_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_32898_end_mask_0 = const()[name = tensor<string, []>("op_32898_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32898_cast = slice_by_index(begin = var_32898_begin_0, end = var_32898_end_0, end_mask = var_32898_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32898_cast")];
            tensor<int32, [4]> var_32902_begin_0 = const()[name = tensor<string, []>("op_32902_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_32902_end_0 = const()[name = tensor<string, []>("op_32902_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_32902_end_mask_0 = const()[name = tensor<string, []>("op_32902_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32902_cast = slice_by_index(begin = var_32902_begin_0, end = var_32902_end_0, end_mask = var_32902_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32902_cast")];
            tensor<int32, [4]> var_32906_begin_0 = const()[name = tensor<string, []>("op_32906_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_32906_end_0 = const()[name = tensor<string, []>("op_32906_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_32906_end_mask_0 = const()[name = tensor<string, []>("op_32906_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32906_cast = slice_by_index(begin = var_32906_begin_0, end = var_32906_end_0, end_mask = var_32906_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32906_cast")];
            tensor<int32, [4]> var_32910_begin_0 = const()[name = tensor<string, []>("op_32910_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_32910_end_0 = const()[name = tensor<string, []>("op_32910_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_32910_end_mask_0 = const()[name = tensor<string, []>("op_32910_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32910_cast = slice_by_index(begin = var_32910_begin_0, end = var_32910_end_0, end_mask = var_32910_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32910_cast")];
            tensor<int32, [4]> var_32914_begin_0 = const()[name = tensor<string, []>("op_32914_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_32914_end_0 = const()[name = tensor<string, []>("op_32914_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_32914_end_mask_0 = const()[name = tensor<string, []>("op_32914_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32914_cast = slice_by_index(begin = var_32914_begin_0, end = var_32914_end_0, end_mask = var_32914_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32914_cast")];
            tensor<int32, [4]> var_32918_begin_0 = const()[name = tensor<string, []>("op_32918_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_32918_end_0 = const()[name = tensor<string, []>("op_32918_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_32918_end_mask_0 = const()[name = tensor<string, []>("op_32918_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32918_cast = slice_by_index(begin = var_32918_begin_0, end = var_32918_end_0, end_mask = var_32918_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32918_cast")];
            tensor<int32, [4]> var_32922_begin_0 = const()[name = tensor<string, []>("op_32922_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_32922_end_0 = const()[name = tensor<string, []>("op_32922_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_32922_end_mask_0 = const()[name = tensor<string, []>("op_32922_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32922_cast = slice_by_index(begin = var_32922_begin_0, end = var_32922_end_0, end_mask = var_32922_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32922_cast")];
            tensor<int32, [4]> var_32926_begin_0 = const()[name = tensor<string, []>("op_32926_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_32926_end_0 = const()[name = tensor<string, []>("op_32926_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_32926_end_mask_0 = const()[name = tensor<string, []>("op_32926_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32926_cast = slice_by_index(begin = var_32926_begin_0, end = var_32926_end_0, end_mask = var_32926_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32926_cast")];
            tensor<int32, [4]> var_32930_begin_0 = const()[name = tensor<string, []>("op_32930_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_32930_end_0 = const()[name = tensor<string, []>("op_32930_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_32930_end_mask_0 = const()[name = tensor<string, []>("op_32930_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32930_cast = slice_by_index(begin = var_32930_begin_0, end = var_32930_end_0, end_mask = var_32930_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32930_cast")];
            tensor<int32, [4]> var_32934_begin_0 = const()[name = tensor<string, []>("op_32934_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_32934_end_0 = const()[name = tensor<string, []>("op_32934_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_32934_end_mask_0 = const()[name = tensor<string, []>("op_32934_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32934_cast = slice_by_index(begin = var_32934_begin_0, end = var_32934_end_0, end_mask = var_32934_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32934_cast")];
            tensor<int32, [4]> var_32938_begin_0 = const()[name = tensor<string, []>("op_32938_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_32938_end_0 = const()[name = tensor<string, []>("op_32938_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_32938_end_mask_0 = const()[name = tensor<string, []>("op_32938_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32938_cast = slice_by_index(begin = var_32938_begin_0, end = var_32938_end_0, end_mask = var_32938_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32938_cast")];
            tensor<int32, [4]> var_32942_begin_0 = const()[name = tensor<string, []>("op_32942_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_32942_end_0 = const()[name = tensor<string, []>("op_32942_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_32942_end_mask_0 = const()[name = tensor<string, []>("op_32942_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32942_cast = slice_by_index(begin = var_32942_begin_0, end = var_32942_end_0, end_mask = var_32942_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32942_cast")];
            tensor<int32, [4]> var_32946_begin_0 = const()[name = tensor<string, []>("op_32946_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_32946_end_0 = const()[name = tensor<string, []>("op_32946_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_32946_end_mask_0 = const()[name = tensor<string, []>("op_32946_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32946_cast = slice_by_index(begin = var_32946_begin_0, end = var_32946_end_0, end_mask = var_32946_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32946_cast")];
            tensor<int32, [4]> var_32950_begin_0 = const()[name = tensor<string, []>("op_32950_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_32950_end_0 = const()[name = tensor<string, []>("op_32950_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_32950_end_mask_0 = const()[name = tensor<string, []>("op_32950_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_32950_cast = slice_by_index(begin = var_32950_begin_0, end = var_32950_end_0, end_mask = var_32950_end_mask_0, x = transpose_66)[name = tensor<string, []>("op_32950_cast")];
            tensor<int32, [4]> var_32952_begin_0 = const()[name = tensor<string, []>("op_32952_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_32952_end_0 = const()[name = tensor<string, []>("op_32952_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_32952_end_mask_0 = const()[name = tensor<string, []>("op_32952_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32952_cast = slice_by_index(begin = var_32952_begin_0, end = var_32952_end_0, end_mask = var_32952_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_32952_cast")];
            tensor<int32, [4]> var_32956_begin_0 = const()[name = tensor<string, []>("op_32956_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_32956_end_0 = const()[name = tensor<string, []>("op_32956_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_32956_end_mask_0 = const()[name = tensor<string, []>("op_32956_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32956_cast = slice_by_index(begin = var_32956_begin_0, end = var_32956_end_0, end_mask = var_32956_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_32956_cast")];
            tensor<int32, [4]> var_32960_begin_0 = const()[name = tensor<string, []>("op_32960_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_32960_end_0 = const()[name = tensor<string, []>("op_32960_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_32960_end_mask_0 = const()[name = tensor<string, []>("op_32960_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32960_cast = slice_by_index(begin = var_32960_begin_0, end = var_32960_end_0, end_mask = var_32960_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_32960_cast")];
            tensor<int32, [4]> var_32964_begin_0 = const()[name = tensor<string, []>("op_32964_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_32964_end_0 = const()[name = tensor<string, []>("op_32964_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_32964_end_mask_0 = const()[name = tensor<string, []>("op_32964_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32964_cast = slice_by_index(begin = var_32964_begin_0, end = var_32964_end_0, end_mask = var_32964_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_32964_cast")];
            tensor<int32, [4]> var_32968_begin_0 = const()[name = tensor<string, []>("op_32968_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_32968_end_0 = const()[name = tensor<string, []>("op_32968_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_32968_end_mask_0 = const()[name = tensor<string, []>("op_32968_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32968_cast = slice_by_index(begin = var_32968_begin_0, end = var_32968_end_0, end_mask = var_32968_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_32968_cast")];
            tensor<int32, [4]> var_32972_begin_0 = const()[name = tensor<string, []>("op_32972_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_32972_end_0 = const()[name = tensor<string, []>("op_32972_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_32972_end_mask_0 = const()[name = tensor<string, []>("op_32972_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32972_cast = slice_by_index(begin = var_32972_begin_0, end = var_32972_end_0, end_mask = var_32972_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_32972_cast")];
            tensor<int32, [4]> var_32976_begin_0 = const()[name = tensor<string, []>("op_32976_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_32976_end_0 = const()[name = tensor<string, []>("op_32976_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_32976_end_mask_0 = const()[name = tensor<string, []>("op_32976_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32976_cast = slice_by_index(begin = var_32976_begin_0, end = var_32976_end_0, end_mask = var_32976_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_32976_cast")];
            tensor<int32, [4]> var_32980_begin_0 = const()[name = tensor<string, []>("op_32980_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_32980_end_0 = const()[name = tensor<string, []>("op_32980_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_32980_end_mask_0 = const()[name = tensor<string, []>("op_32980_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32980_cast = slice_by_index(begin = var_32980_begin_0, end = var_32980_end_0, end_mask = var_32980_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_32980_cast")];
            tensor<int32, [4]> var_32984_begin_0 = const()[name = tensor<string, []>("op_32984_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_32984_end_0 = const()[name = tensor<string, []>("op_32984_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_32984_end_mask_0 = const()[name = tensor<string, []>("op_32984_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32984_cast = slice_by_index(begin = var_32984_begin_0, end = var_32984_end_0, end_mask = var_32984_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_32984_cast")];
            tensor<int32, [4]> var_32988_begin_0 = const()[name = tensor<string, []>("op_32988_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_32988_end_0 = const()[name = tensor<string, []>("op_32988_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_32988_end_mask_0 = const()[name = tensor<string, []>("op_32988_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32988_cast = slice_by_index(begin = var_32988_begin_0, end = var_32988_end_0, end_mask = var_32988_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_32988_cast")];
            tensor<int32, [4]> var_32992_begin_0 = const()[name = tensor<string, []>("op_32992_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_32992_end_0 = const()[name = tensor<string, []>("op_32992_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_32992_end_mask_0 = const()[name = tensor<string, []>("op_32992_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32992_cast = slice_by_index(begin = var_32992_begin_0, end = var_32992_end_0, end_mask = var_32992_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_32992_cast")];
            tensor<int32, [4]> var_32996_begin_0 = const()[name = tensor<string, []>("op_32996_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_32996_end_0 = const()[name = tensor<string, []>("op_32996_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_32996_end_mask_0 = const()[name = tensor<string, []>("op_32996_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_32996_cast = slice_by_index(begin = var_32996_begin_0, end = var_32996_end_0, end_mask = var_32996_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_32996_cast")];
            tensor<int32, [4]> var_33000_begin_0 = const()[name = tensor<string, []>("op_33000_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_33000_end_0 = const()[name = tensor<string, []>("op_33000_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_33000_end_mask_0 = const()[name = tensor<string, []>("op_33000_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33000_cast = slice_by_index(begin = var_33000_begin_0, end = var_33000_end_0, end_mask = var_33000_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_33000_cast")];
            tensor<int32, [4]> var_33004_begin_0 = const()[name = tensor<string, []>("op_33004_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_33004_end_0 = const()[name = tensor<string, []>("op_33004_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_33004_end_mask_0 = const()[name = tensor<string, []>("op_33004_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33004_cast = slice_by_index(begin = var_33004_begin_0, end = var_33004_end_0, end_mask = var_33004_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_33004_cast")];
            tensor<int32, [4]> var_33008_begin_0 = const()[name = tensor<string, []>("op_33008_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_33008_end_0 = const()[name = tensor<string, []>("op_33008_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_33008_end_mask_0 = const()[name = tensor<string, []>("op_33008_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33008_cast = slice_by_index(begin = var_33008_begin_0, end = var_33008_end_0, end_mask = var_33008_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_33008_cast")];
            tensor<int32, [4]> var_33012_begin_0 = const()[name = tensor<string, []>("op_33012_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_33012_end_0 = const()[name = tensor<string, []>("op_33012_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_33012_end_mask_0 = const()[name = tensor<string, []>("op_33012_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33012_cast = slice_by_index(begin = var_33012_begin_0, end = var_33012_end_0, end_mask = var_33012_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_33012_cast")];
            tensor<int32, [4]> var_33016_begin_0 = const()[name = tensor<string, []>("op_33016_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_33016_end_0 = const()[name = tensor<string, []>("op_33016_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_33016_end_mask_0 = const()[name = tensor<string, []>("op_33016_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33016_cast = slice_by_index(begin = var_33016_begin_0, end = var_33016_end_0, end_mask = var_33016_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_33016_cast")];
            tensor<int32, [4]> var_33020_begin_0 = const()[name = tensor<string, []>("op_33020_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_33020_end_0 = const()[name = tensor<string, []>("op_33020_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_33020_end_mask_0 = const()[name = tensor<string, []>("op_33020_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33020_cast = slice_by_index(begin = var_33020_begin_0, end = var_33020_end_0, end_mask = var_33020_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_33020_cast")];
            tensor<int32, [4]> var_33024_begin_0 = const()[name = tensor<string, []>("op_33024_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_33024_end_0 = const()[name = tensor<string, []>("op_33024_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_33024_end_mask_0 = const()[name = tensor<string, []>("op_33024_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33024_cast = slice_by_index(begin = var_33024_begin_0, end = var_33024_end_0, end_mask = var_33024_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_33024_cast")];
            tensor<int32, [4]> var_33028_begin_0 = const()[name = tensor<string, []>("op_33028_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_33028_end_0 = const()[name = tensor<string, []>("op_33028_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_33028_end_mask_0 = const()[name = tensor<string, []>("op_33028_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33028_cast = slice_by_index(begin = var_33028_begin_0, end = var_33028_end_0, end_mask = var_33028_end_mask_0, x = v_147_cast)[name = tensor<string, []>("op_33028_cast")];
            tensor<string, []> var_33032_equation_0 = const()[name = tensor<string, []>("op_33032_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33032_cast = einsum(equation = var_33032_equation_0, values = (var_32874_cast, var_32791_cast))[name = tensor<string, []>("op_33032_cast")];
            tensor<fp16, []> var_33033_to_fp16 = const()[name = tensor<string, []>("op_33033_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2761_cast = mul(x = var_33032_cast, y = var_33033_to_fp16)[name = tensor<string, []>("aw_2761_cast")];
            tensor<string, []> var_33036_equation_0 = const()[name = tensor<string, []>("op_33036_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33036_cast = einsum(equation = var_33036_equation_0, values = (var_32878_cast, var_32795_cast))[name = tensor<string, []>("op_33036_cast")];
            tensor<fp16, []> var_33037_to_fp16 = const()[name = tensor<string, []>("op_33037_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2763_cast = mul(x = var_33036_cast, y = var_33037_to_fp16)[name = tensor<string, []>("aw_2763_cast")];
            tensor<string, []> var_33040_equation_0 = const()[name = tensor<string, []>("op_33040_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33040_cast = einsum(equation = var_33040_equation_0, values = (var_32882_cast, var_32799_cast))[name = tensor<string, []>("op_33040_cast")];
            tensor<fp16, []> var_33041_to_fp16 = const()[name = tensor<string, []>("op_33041_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2765_cast = mul(x = var_33040_cast, y = var_33041_to_fp16)[name = tensor<string, []>("aw_2765_cast")];
            tensor<string, []> var_33044_equation_0 = const()[name = tensor<string, []>("op_33044_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33044_cast = einsum(equation = var_33044_equation_0, values = (var_32886_cast, var_32803_cast))[name = tensor<string, []>("op_33044_cast")];
            tensor<fp16, []> var_33045_to_fp16 = const()[name = tensor<string, []>("op_33045_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2767_cast = mul(x = var_33044_cast, y = var_33045_to_fp16)[name = tensor<string, []>("aw_2767_cast")];
            tensor<string, []> var_33048_equation_0 = const()[name = tensor<string, []>("op_33048_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33048_cast = einsum(equation = var_33048_equation_0, values = (var_32890_cast, var_32807_cast))[name = tensor<string, []>("op_33048_cast")];
            tensor<fp16, []> var_33049_to_fp16 = const()[name = tensor<string, []>("op_33049_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2769_cast = mul(x = var_33048_cast, y = var_33049_to_fp16)[name = tensor<string, []>("aw_2769_cast")];
            tensor<string, []> var_33052_equation_0 = const()[name = tensor<string, []>("op_33052_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33052_cast = einsum(equation = var_33052_equation_0, values = (var_32894_cast, var_32811_cast))[name = tensor<string, []>("op_33052_cast")];
            tensor<fp16, []> var_33053_to_fp16 = const()[name = tensor<string, []>("op_33053_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2771_cast = mul(x = var_33052_cast, y = var_33053_to_fp16)[name = tensor<string, []>("aw_2771_cast")];
            tensor<string, []> var_33056_equation_0 = const()[name = tensor<string, []>("op_33056_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33056_cast = einsum(equation = var_33056_equation_0, values = (var_32898_cast, var_32815_cast))[name = tensor<string, []>("op_33056_cast")];
            tensor<fp16, []> var_33057_to_fp16 = const()[name = tensor<string, []>("op_33057_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2773_cast = mul(x = var_33056_cast, y = var_33057_to_fp16)[name = tensor<string, []>("aw_2773_cast")];
            tensor<string, []> var_33060_equation_0 = const()[name = tensor<string, []>("op_33060_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33060_cast = einsum(equation = var_33060_equation_0, values = (var_32902_cast, var_32819_cast))[name = tensor<string, []>("op_33060_cast")];
            tensor<fp16, []> var_33061_to_fp16 = const()[name = tensor<string, []>("op_33061_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2775_cast = mul(x = var_33060_cast, y = var_33061_to_fp16)[name = tensor<string, []>("aw_2775_cast")];
            tensor<string, []> var_33064_equation_0 = const()[name = tensor<string, []>("op_33064_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33064_cast = einsum(equation = var_33064_equation_0, values = (var_32906_cast, var_32823_cast))[name = tensor<string, []>("op_33064_cast")];
            tensor<fp16, []> var_33065_to_fp16 = const()[name = tensor<string, []>("op_33065_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2777_cast = mul(x = var_33064_cast, y = var_33065_to_fp16)[name = tensor<string, []>("aw_2777_cast")];
            tensor<string, []> var_33068_equation_0 = const()[name = tensor<string, []>("op_33068_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33068_cast = einsum(equation = var_33068_equation_0, values = (var_32910_cast, var_32827_cast))[name = tensor<string, []>("op_33068_cast")];
            tensor<fp16, []> var_33069_to_fp16 = const()[name = tensor<string, []>("op_33069_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2779_cast = mul(x = var_33068_cast, y = var_33069_to_fp16)[name = tensor<string, []>("aw_2779_cast")];
            tensor<string, []> var_33072_equation_0 = const()[name = tensor<string, []>("op_33072_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33072_cast = einsum(equation = var_33072_equation_0, values = (var_32914_cast, var_32831_cast))[name = tensor<string, []>("op_33072_cast")];
            tensor<fp16, []> var_33073_to_fp16 = const()[name = tensor<string, []>("op_33073_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2781_cast = mul(x = var_33072_cast, y = var_33073_to_fp16)[name = tensor<string, []>("aw_2781_cast")];
            tensor<string, []> var_33076_equation_0 = const()[name = tensor<string, []>("op_33076_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33076_cast = einsum(equation = var_33076_equation_0, values = (var_32918_cast, var_32835_cast))[name = tensor<string, []>("op_33076_cast")];
            tensor<fp16, []> var_33077_to_fp16 = const()[name = tensor<string, []>("op_33077_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2783_cast = mul(x = var_33076_cast, y = var_33077_to_fp16)[name = tensor<string, []>("aw_2783_cast")];
            tensor<string, []> var_33080_equation_0 = const()[name = tensor<string, []>("op_33080_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33080_cast = einsum(equation = var_33080_equation_0, values = (var_32922_cast, var_32839_cast))[name = tensor<string, []>("op_33080_cast")];
            tensor<fp16, []> var_33081_to_fp16 = const()[name = tensor<string, []>("op_33081_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2785_cast = mul(x = var_33080_cast, y = var_33081_to_fp16)[name = tensor<string, []>("aw_2785_cast")];
            tensor<string, []> var_33084_equation_0 = const()[name = tensor<string, []>("op_33084_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33084_cast = einsum(equation = var_33084_equation_0, values = (var_32926_cast, var_32843_cast))[name = tensor<string, []>("op_33084_cast")];
            tensor<fp16, []> var_33085_to_fp16 = const()[name = tensor<string, []>("op_33085_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2787_cast = mul(x = var_33084_cast, y = var_33085_to_fp16)[name = tensor<string, []>("aw_2787_cast")];
            tensor<string, []> var_33088_equation_0 = const()[name = tensor<string, []>("op_33088_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33088_cast = einsum(equation = var_33088_equation_0, values = (var_32930_cast, var_32847_cast))[name = tensor<string, []>("op_33088_cast")];
            tensor<fp16, []> var_33089_to_fp16 = const()[name = tensor<string, []>("op_33089_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2789_cast = mul(x = var_33088_cast, y = var_33089_to_fp16)[name = tensor<string, []>("aw_2789_cast")];
            tensor<string, []> var_33092_equation_0 = const()[name = tensor<string, []>("op_33092_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33092_cast = einsum(equation = var_33092_equation_0, values = (var_32934_cast, var_32851_cast))[name = tensor<string, []>("op_33092_cast")];
            tensor<fp16, []> var_33093_to_fp16 = const()[name = tensor<string, []>("op_33093_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2791_cast = mul(x = var_33092_cast, y = var_33093_to_fp16)[name = tensor<string, []>("aw_2791_cast")];
            tensor<string, []> var_33096_equation_0 = const()[name = tensor<string, []>("op_33096_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33096_cast = einsum(equation = var_33096_equation_0, values = (var_32938_cast, var_32855_cast))[name = tensor<string, []>("op_33096_cast")];
            tensor<fp16, []> var_33097_to_fp16 = const()[name = tensor<string, []>("op_33097_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2793_cast = mul(x = var_33096_cast, y = var_33097_to_fp16)[name = tensor<string, []>("aw_2793_cast")];
            tensor<string, []> var_33100_equation_0 = const()[name = tensor<string, []>("op_33100_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33100_cast = einsum(equation = var_33100_equation_0, values = (var_32942_cast, var_32859_cast))[name = tensor<string, []>("op_33100_cast")];
            tensor<fp16, []> var_33101_to_fp16 = const()[name = tensor<string, []>("op_33101_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2795_cast = mul(x = var_33100_cast, y = var_33101_to_fp16)[name = tensor<string, []>("aw_2795_cast")];
            tensor<string, []> var_33104_equation_0 = const()[name = tensor<string, []>("op_33104_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33104_cast = einsum(equation = var_33104_equation_0, values = (var_32946_cast, var_32863_cast))[name = tensor<string, []>("op_33104_cast")];
            tensor<fp16, []> var_33105_to_fp16 = const()[name = tensor<string, []>("op_33105_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2797_cast = mul(x = var_33104_cast, y = var_33105_to_fp16)[name = tensor<string, []>("aw_2797_cast")];
            tensor<string, []> var_33108_equation_0 = const()[name = tensor<string, []>("op_33108_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33108_cast = einsum(equation = var_33108_equation_0, values = (var_32950_cast, var_32867_cast))[name = tensor<string, []>("op_33108_cast")];
            tensor<fp16, []> var_33109_to_fp16 = const()[name = tensor<string, []>("op_33109_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2799_cast = mul(x = var_33108_cast, y = var_33109_to_fp16)[name = tensor<string, []>("aw_2799_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33111_cast = softmax(axis = var_30385, x = aw_2761_cast)[name = tensor<string, []>("op_33111_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33112_cast = softmax(axis = var_30385, x = aw_2763_cast)[name = tensor<string, []>("op_33112_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33113_cast = softmax(axis = var_30385, x = aw_2765_cast)[name = tensor<string, []>("op_33113_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33114_cast = softmax(axis = var_30385, x = aw_2767_cast)[name = tensor<string, []>("op_33114_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33115_cast = softmax(axis = var_30385, x = aw_2769_cast)[name = tensor<string, []>("op_33115_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33116_cast = softmax(axis = var_30385, x = aw_2771_cast)[name = tensor<string, []>("op_33116_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33117_cast = softmax(axis = var_30385, x = aw_2773_cast)[name = tensor<string, []>("op_33117_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33118_cast = softmax(axis = var_30385, x = aw_2775_cast)[name = tensor<string, []>("op_33118_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33119_cast = softmax(axis = var_30385, x = aw_2777_cast)[name = tensor<string, []>("op_33119_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33120_cast = softmax(axis = var_30385, x = aw_2779_cast)[name = tensor<string, []>("op_33120_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33121_cast = softmax(axis = var_30385, x = aw_2781_cast)[name = tensor<string, []>("op_33121_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33122_cast = softmax(axis = var_30385, x = aw_2783_cast)[name = tensor<string, []>("op_33122_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33123_cast = softmax(axis = var_30385, x = aw_2785_cast)[name = tensor<string, []>("op_33123_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33124_cast = softmax(axis = var_30385, x = aw_2787_cast)[name = tensor<string, []>("op_33124_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33125_cast = softmax(axis = var_30385, x = aw_2789_cast)[name = tensor<string, []>("op_33125_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33126_cast = softmax(axis = var_30385, x = aw_2791_cast)[name = tensor<string, []>("op_33126_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33127_cast = softmax(axis = var_30385, x = aw_2793_cast)[name = tensor<string, []>("op_33127_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33128_cast = softmax(axis = var_30385, x = aw_2795_cast)[name = tensor<string, []>("op_33128_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33129_cast = softmax(axis = var_30385, x = aw_2797_cast)[name = tensor<string, []>("op_33129_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_33130_cast = softmax(axis = var_30385, x = aw_2799_cast)[name = tensor<string, []>("op_33130_cast")];
            tensor<string, []> var_33132_equation_0 = const()[name = tensor<string, []>("op_33132_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33132_cast = einsum(equation = var_33132_equation_0, values = (var_32952_cast, var_33111_cast))[name = tensor<string, []>("op_33132_cast")];
            tensor<string, []> var_33134_equation_0 = const()[name = tensor<string, []>("op_33134_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33134_cast = einsum(equation = var_33134_equation_0, values = (var_32956_cast, var_33112_cast))[name = tensor<string, []>("op_33134_cast")];
            tensor<string, []> var_33136_equation_0 = const()[name = tensor<string, []>("op_33136_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33136_cast = einsum(equation = var_33136_equation_0, values = (var_32960_cast, var_33113_cast))[name = tensor<string, []>("op_33136_cast")];
            tensor<string, []> var_33138_equation_0 = const()[name = tensor<string, []>("op_33138_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33138_cast = einsum(equation = var_33138_equation_0, values = (var_32964_cast, var_33114_cast))[name = tensor<string, []>("op_33138_cast")];
            tensor<string, []> var_33140_equation_0 = const()[name = tensor<string, []>("op_33140_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33140_cast = einsum(equation = var_33140_equation_0, values = (var_32968_cast, var_33115_cast))[name = tensor<string, []>("op_33140_cast")];
            tensor<string, []> var_33142_equation_0 = const()[name = tensor<string, []>("op_33142_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33142_cast = einsum(equation = var_33142_equation_0, values = (var_32972_cast, var_33116_cast))[name = tensor<string, []>("op_33142_cast")];
            tensor<string, []> var_33144_equation_0 = const()[name = tensor<string, []>("op_33144_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33144_cast = einsum(equation = var_33144_equation_0, values = (var_32976_cast, var_33117_cast))[name = tensor<string, []>("op_33144_cast")];
            tensor<string, []> var_33146_equation_0 = const()[name = tensor<string, []>("op_33146_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33146_cast = einsum(equation = var_33146_equation_0, values = (var_32980_cast, var_33118_cast))[name = tensor<string, []>("op_33146_cast")];
            tensor<string, []> var_33148_equation_0 = const()[name = tensor<string, []>("op_33148_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33148_cast = einsum(equation = var_33148_equation_0, values = (var_32984_cast, var_33119_cast))[name = tensor<string, []>("op_33148_cast")];
            tensor<string, []> var_33150_equation_0 = const()[name = tensor<string, []>("op_33150_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33150_cast = einsum(equation = var_33150_equation_0, values = (var_32988_cast, var_33120_cast))[name = tensor<string, []>("op_33150_cast")];
            tensor<string, []> var_33152_equation_0 = const()[name = tensor<string, []>("op_33152_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33152_cast = einsum(equation = var_33152_equation_0, values = (var_32992_cast, var_33121_cast))[name = tensor<string, []>("op_33152_cast")];
            tensor<string, []> var_33154_equation_0 = const()[name = tensor<string, []>("op_33154_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33154_cast = einsum(equation = var_33154_equation_0, values = (var_32996_cast, var_33122_cast))[name = tensor<string, []>("op_33154_cast")];
            tensor<string, []> var_33156_equation_0 = const()[name = tensor<string, []>("op_33156_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33156_cast = einsum(equation = var_33156_equation_0, values = (var_33000_cast, var_33123_cast))[name = tensor<string, []>("op_33156_cast")];
            tensor<string, []> var_33158_equation_0 = const()[name = tensor<string, []>("op_33158_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33158_cast = einsum(equation = var_33158_equation_0, values = (var_33004_cast, var_33124_cast))[name = tensor<string, []>("op_33158_cast")];
            tensor<string, []> var_33160_equation_0 = const()[name = tensor<string, []>("op_33160_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33160_cast = einsum(equation = var_33160_equation_0, values = (var_33008_cast, var_33125_cast))[name = tensor<string, []>("op_33160_cast")];
            tensor<string, []> var_33162_equation_0 = const()[name = tensor<string, []>("op_33162_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33162_cast = einsum(equation = var_33162_equation_0, values = (var_33012_cast, var_33126_cast))[name = tensor<string, []>("op_33162_cast")];
            tensor<string, []> var_33164_equation_0 = const()[name = tensor<string, []>("op_33164_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33164_cast = einsum(equation = var_33164_equation_0, values = (var_33016_cast, var_33127_cast))[name = tensor<string, []>("op_33164_cast")];
            tensor<string, []> var_33166_equation_0 = const()[name = tensor<string, []>("op_33166_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33166_cast = einsum(equation = var_33166_equation_0, values = (var_33020_cast, var_33128_cast))[name = tensor<string, []>("op_33166_cast")];
            tensor<string, []> var_33168_equation_0 = const()[name = tensor<string, []>("op_33168_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33168_cast = einsum(equation = var_33168_equation_0, values = (var_33024_cast, var_33129_cast))[name = tensor<string, []>("op_33168_cast")];
            tensor<string, []> var_33170_equation_0 = const()[name = tensor<string, []>("op_33170_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33170_cast = einsum(equation = var_33170_equation_0, values = (var_33028_cast, var_33130_cast))[name = tensor<string, []>("op_33170_cast")];
            tensor<bool, []> input_459_interleave_0 = const()[name = tensor<string, []>("input_459_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_459_cast = concat(axis = var_30385, interleave = input_459_interleave_0, values = (var_33132_cast, var_33134_cast, var_33136_cast, var_33138_cast, var_33140_cast, var_33142_cast, var_33144_cast, var_33146_cast, var_33148_cast, var_33150_cast, var_33152_cast, var_33154_cast, var_33156_cast, var_33158_cast, var_33160_cast, var_33162_cast, var_33164_cast, var_33166_cast, var_33168_cast, var_33170_cast))[name = tensor<string, []>("input_459_cast")];
            tensor<int32, [2]> var_33176 = const()[name = tensor<string, []>("op_33176"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_33178 = const()[name = tensor<string, []>("op_33178"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_33180_pad_type_0 = const()[name = tensor<string, []>("op_33180_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_33180_pad_0 = const()[name = tensor<string, []>("op_33180_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2754330624)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2757607488)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_33180_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_out_0_bias_to_fp16, dilations = var_33178, groups = var_30385, pad = var_33180_pad_0, pad_type = var_33180_pad_type_0, strides = var_33176, weight = up_blocks_0_attentions_0_transformer_blocks_2_attn2_to_out_0_weight_to_fp16, x = input_459_cast)[name = tensor<string, []>("op_33180_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_221_cast = add(x = var_33180_cast, y = inputs_219_cast)[name = tensor<string, []>("inputs_221_cast")];
            tensor<int32, [1]> var_33184 = const()[name = tensor<string, []>("op_33184"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_221_cast = reduce_mean(axes = var_33184, keep_dims = var_30380, x = inputs_221_cast)[name = tensor<string, []>("channels_mean_221_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_221_cast = sub(x = inputs_221_cast, y = channels_mean_221_cast)[name = tensor<string, []>("zero_mean_221_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_221_cast = mul(x = zero_mean_221_cast, y = zero_mean_221_cast)[name = tensor<string, []>("zero_mean_sq_221_cast")];
            tensor<int32, [1]> var_33188 = const()[name = tensor<string, []>("op_33188"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_33189_cast = reduce_mean(axes = var_33188, keep_dims = var_30380, x = zero_mean_sq_221_cast)[name = tensor<string, []>("op_33189_cast")];
            tensor<fp16, []> var_33190_to_fp16 = const()[name = tensor<string, []>("op_33190_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_33191_cast = add(x = var_33189_cast, y = var_33190_to_fp16)[name = tensor<string, []>("op_33191_cast")];
            tensor<fp16, []> denom_221_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_221_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_221_cast = rsqrt(epsilon = denom_221_epsilon_0_to_fp16, x = var_33191_cast)[name = tensor<string, []>("denom_221_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_221_cast = mul(x = zero_mean_221_cast, y = denom_221_cast)[name = tensor<string, []>("out_221_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_33195_to_fp16 = const()[name = tensor<string, []>("op_33195_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2757610112)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_33196_cast = add(x = out_221_cast, y = var_33195_to_fp16)[name = tensor<string, []>("op_33196_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_33198_to_fp16 = const()[name = tensor<string, []>("op_33198_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2757612736)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_461_cast = mul(x = var_33196_cast, y = var_33198_to_fp16)[name = tensor<string, []>("input_461_cast")];
            tensor<int32, [2]> var_33206 = const()[name = tensor<string, []>("op_33206"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_33208 = const()[name = tensor<string, []>("op_33208"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_33210_pad_type_0 = const()[name = tensor<string, []>("op_33210_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_33210_pad_0 = const()[name = tensor<string, []>("op_33210_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2757615360)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2783829824)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_33210_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_2_ff_net_0_proj_bias_to_fp16, dilations = var_33208, groups = var_30385, pad = var_33210_pad_0, pad_type = var_33210_pad_type_0, strides = var_33206, weight = up_blocks_0_attentions_0_transformer_blocks_2_ff_net_0_proj_weight_to_fp16, x = input_461_cast)[name = tensor<string, []>("op_33210_cast")];
            tensor<int32, [2]> var_33211_split_sizes_0 = const()[name = tensor<string, []>("op_33211_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_33211_axis_0 = const()[name = tensor<string, []>("op_33211_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_33211_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_33211_cast_1 = split(axis = var_33211_axis_0, split_sizes = var_33211_split_sizes_0, x = var_33210_cast)[name = tensor<string, []>("op_33211_cast")];
            tensor<string, []> var_33213_mode_0 = const()[name = tensor<string, []>("op_33213_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_33213_cast = gelu(mode = var_33213_mode_0, x = var_33211_cast_1)[name = tensor<string, []>("op_33213_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_463_cast = mul(x = var_33211_cast_0, y = var_33213_cast)[name = tensor<string, []>("input_463_cast")];
            tensor<int32, [2]> var_33217 = const()[name = tensor<string, []>("op_33217"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_33219 = const()[name = tensor<string, []>("op_33219"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_33221_pad_type_0 = const()[name = tensor<string, []>("op_33221_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_33221_pad_0 = const()[name = tensor<string, []>("op_33221_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2783850368)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2796957632)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_33221_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_2_ff_net_2_bias_to_fp16, dilations = var_33219, groups = var_30385, pad = var_33221_pad_0, pad_type = var_33221_pad_type_0, strides = var_33217, weight = up_blocks_0_attentions_0_transformer_blocks_2_ff_net_2_weight_to_fp16, x = input_463_cast)[name = tensor<string, []>("op_33221_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_223_cast = add(x = var_33221_cast, y = inputs_221_cast)[name = tensor<string, []>("inputs_223_cast")];
            tensor<int32, [1]> var_33231 = const()[name = tensor<string, []>("op_33231"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_223_cast = reduce_mean(axes = var_33231, keep_dims = var_30380, x = inputs_223_cast)[name = tensor<string, []>("channels_mean_223_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_223_cast = sub(x = inputs_223_cast, y = channels_mean_223_cast)[name = tensor<string, []>("zero_mean_223_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_223_cast = mul(x = zero_mean_223_cast, y = zero_mean_223_cast)[name = tensor<string, []>("zero_mean_sq_223_cast")];
            tensor<int32, [1]> var_33235 = const()[name = tensor<string, []>("op_33235"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_33236_cast = reduce_mean(axes = var_33235, keep_dims = var_30380, x = zero_mean_sq_223_cast)[name = tensor<string, []>("op_33236_cast")];
            tensor<fp16, []> var_33237_to_fp16 = const()[name = tensor<string, []>("op_33237_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_33238_cast = add(x = var_33236_cast, y = var_33237_to_fp16)[name = tensor<string, []>("op_33238_cast")];
            tensor<fp16, []> denom_223_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_223_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_223_cast = rsqrt(epsilon = denom_223_epsilon_0_to_fp16, x = var_33238_cast)[name = tensor<string, []>("denom_223_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_223_cast = mul(x = zero_mean_223_cast, y = denom_223_cast)[name = tensor<string, []>("out_223_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_33242_to_fp16 = const()[name = tensor<string, []>("op_33242_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2796960256)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_33243_cast = add(x = out_223_cast, y = var_33242_to_fp16)[name = tensor<string, []>("op_33243_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_33245_to_fp16 = const()[name = tensor<string, []>("op_33245_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2796962880)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_305_cast = mul(x = var_33243_cast, y = var_33245_to_fp16)[name = tensor<string, []>("hidden_states_305_cast")];
            tensor<int32, [2]> var_33252 = const()[name = tensor<string, []>("op_33252"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_33254 = const()[name = tensor<string, []>("op_33254"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_149_pad_type_0 = const()[name = tensor<string, []>("q_149_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_149_pad_0 = const()[name = tensor<string, []>("q_149_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2796965504)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_149_cast = conv(dilations = var_33254, groups = var_30385, pad = q_149_pad_0, pad_type = q_149_pad_type_0, strides = var_33252, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_q_weight_to_fp16, x = hidden_states_305_cast)[name = tensor<string, []>("q_149_cast")];
            tensor<int32, [2]> var_33258 = const()[name = tensor<string, []>("op_33258"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_33260 = const()[name = tensor<string, []>("op_33260"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_297_pad_type_0 = const()[name = tensor<string, []>("k_297_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_297_pad_0 = const()[name = tensor<string, []>("k_297_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2800242368)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_297_cast = conv(dilations = var_33260, groups = var_30385, pad = k_297_pad_0, pad_type = k_297_pad_type_0, strides = var_33258, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_k_weight_to_fp16, x = hidden_states_305_cast)[name = tensor<string, []>("k_297_cast")];
            tensor<int32, [2]> var_33264 = const()[name = tensor<string, []>("op_33264"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_33266 = const()[name = tensor<string, []>("op_33266"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_149_pad_type_0 = const()[name = tensor<string, []>("v_149_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_149_pad_0 = const()[name = tensor<string, []>("v_149_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2803519232)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_149_cast = conv(dilations = var_33266, groups = var_30385, pad = v_149_pad_0, pad_type = v_149_pad_type_0, strides = var_33264, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_v_weight_to_fp16, x = hidden_states_305_cast)[name = tensor<string, []>("v_149_cast")];
            tensor<int32, [4]> var_33270_begin_0 = const()[name = tensor<string, []>("op_33270_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_33270_end_0 = const()[name = tensor<string, []>("op_33270_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_33270_end_mask_0 = const()[name = tensor<string, []>("op_33270_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33270_cast = slice_by_index(begin = var_33270_begin_0, end = var_33270_end_0, end_mask = var_33270_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33270_cast")];
            tensor<int32, [4]> var_33274_begin_0 = const()[name = tensor<string, []>("op_33274_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_33274_end_0 = const()[name = tensor<string, []>("op_33274_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_33274_end_mask_0 = const()[name = tensor<string, []>("op_33274_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33274_cast = slice_by_index(begin = var_33274_begin_0, end = var_33274_end_0, end_mask = var_33274_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33274_cast")];
            tensor<int32, [4]> var_33278_begin_0 = const()[name = tensor<string, []>("op_33278_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_33278_end_0 = const()[name = tensor<string, []>("op_33278_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_33278_end_mask_0 = const()[name = tensor<string, []>("op_33278_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33278_cast = slice_by_index(begin = var_33278_begin_0, end = var_33278_end_0, end_mask = var_33278_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33278_cast")];
            tensor<int32, [4]> var_33282_begin_0 = const()[name = tensor<string, []>("op_33282_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_33282_end_0 = const()[name = tensor<string, []>("op_33282_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_33282_end_mask_0 = const()[name = tensor<string, []>("op_33282_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33282_cast = slice_by_index(begin = var_33282_begin_0, end = var_33282_end_0, end_mask = var_33282_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33282_cast")];
            tensor<int32, [4]> var_33286_begin_0 = const()[name = tensor<string, []>("op_33286_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_33286_end_0 = const()[name = tensor<string, []>("op_33286_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_33286_end_mask_0 = const()[name = tensor<string, []>("op_33286_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33286_cast = slice_by_index(begin = var_33286_begin_0, end = var_33286_end_0, end_mask = var_33286_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33286_cast")];
            tensor<int32, [4]> var_33290_begin_0 = const()[name = tensor<string, []>("op_33290_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_33290_end_0 = const()[name = tensor<string, []>("op_33290_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_33290_end_mask_0 = const()[name = tensor<string, []>("op_33290_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33290_cast = slice_by_index(begin = var_33290_begin_0, end = var_33290_end_0, end_mask = var_33290_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33290_cast")];
            tensor<int32, [4]> var_33294_begin_0 = const()[name = tensor<string, []>("op_33294_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_33294_end_0 = const()[name = tensor<string, []>("op_33294_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_33294_end_mask_0 = const()[name = tensor<string, []>("op_33294_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33294_cast = slice_by_index(begin = var_33294_begin_0, end = var_33294_end_0, end_mask = var_33294_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33294_cast")];
            tensor<int32, [4]> var_33298_begin_0 = const()[name = tensor<string, []>("op_33298_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_33298_end_0 = const()[name = tensor<string, []>("op_33298_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_33298_end_mask_0 = const()[name = tensor<string, []>("op_33298_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33298_cast = slice_by_index(begin = var_33298_begin_0, end = var_33298_end_0, end_mask = var_33298_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33298_cast")];
            tensor<int32, [4]> var_33302_begin_0 = const()[name = tensor<string, []>("op_33302_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_33302_end_0 = const()[name = tensor<string, []>("op_33302_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_33302_end_mask_0 = const()[name = tensor<string, []>("op_33302_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33302_cast = slice_by_index(begin = var_33302_begin_0, end = var_33302_end_0, end_mask = var_33302_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33302_cast")];
            tensor<int32, [4]> var_33306_begin_0 = const()[name = tensor<string, []>("op_33306_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_33306_end_0 = const()[name = tensor<string, []>("op_33306_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_33306_end_mask_0 = const()[name = tensor<string, []>("op_33306_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33306_cast = slice_by_index(begin = var_33306_begin_0, end = var_33306_end_0, end_mask = var_33306_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33306_cast")];
            tensor<int32, [4]> var_33310_begin_0 = const()[name = tensor<string, []>("op_33310_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_33310_end_0 = const()[name = tensor<string, []>("op_33310_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_33310_end_mask_0 = const()[name = tensor<string, []>("op_33310_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33310_cast = slice_by_index(begin = var_33310_begin_0, end = var_33310_end_0, end_mask = var_33310_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33310_cast")];
            tensor<int32, [4]> var_33314_begin_0 = const()[name = tensor<string, []>("op_33314_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_33314_end_0 = const()[name = tensor<string, []>("op_33314_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_33314_end_mask_0 = const()[name = tensor<string, []>("op_33314_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33314_cast = slice_by_index(begin = var_33314_begin_0, end = var_33314_end_0, end_mask = var_33314_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33314_cast")];
            tensor<int32, [4]> var_33318_begin_0 = const()[name = tensor<string, []>("op_33318_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_33318_end_0 = const()[name = tensor<string, []>("op_33318_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_33318_end_mask_0 = const()[name = tensor<string, []>("op_33318_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33318_cast = slice_by_index(begin = var_33318_begin_0, end = var_33318_end_0, end_mask = var_33318_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33318_cast")];
            tensor<int32, [4]> var_33322_begin_0 = const()[name = tensor<string, []>("op_33322_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_33322_end_0 = const()[name = tensor<string, []>("op_33322_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_33322_end_mask_0 = const()[name = tensor<string, []>("op_33322_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33322_cast = slice_by_index(begin = var_33322_begin_0, end = var_33322_end_0, end_mask = var_33322_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33322_cast")];
            tensor<int32, [4]> var_33326_begin_0 = const()[name = tensor<string, []>("op_33326_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_33326_end_0 = const()[name = tensor<string, []>("op_33326_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_33326_end_mask_0 = const()[name = tensor<string, []>("op_33326_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33326_cast = slice_by_index(begin = var_33326_begin_0, end = var_33326_end_0, end_mask = var_33326_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33326_cast")];
            tensor<int32, [4]> var_33330_begin_0 = const()[name = tensor<string, []>("op_33330_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_33330_end_0 = const()[name = tensor<string, []>("op_33330_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_33330_end_mask_0 = const()[name = tensor<string, []>("op_33330_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33330_cast = slice_by_index(begin = var_33330_begin_0, end = var_33330_end_0, end_mask = var_33330_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33330_cast")];
            tensor<int32, [4]> var_33334_begin_0 = const()[name = tensor<string, []>("op_33334_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_33334_end_0 = const()[name = tensor<string, []>("op_33334_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_33334_end_mask_0 = const()[name = tensor<string, []>("op_33334_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33334_cast = slice_by_index(begin = var_33334_begin_0, end = var_33334_end_0, end_mask = var_33334_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33334_cast")];
            tensor<int32, [4]> var_33338_begin_0 = const()[name = tensor<string, []>("op_33338_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_33338_end_0 = const()[name = tensor<string, []>("op_33338_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_33338_end_mask_0 = const()[name = tensor<string, []>("op_33338_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33338_cast = slice_by_index(begin = var_33338_begin_0, end = var_33338_end_0, end_mask = var_33338_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33338_cast")];
            tensor<int32, [4]> var_33342_begin_0 = const()[name = tensor<string, []>("op_33342_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_33342_end_0 = const()[name = tensor<string, []>("op_33342_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_33342_end_mask_0 = const()[name = tensor<string, []>("op_33342_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33342_cast = slice_by_index(begin = var_33342_begin_0, end = var_33342_end_0, end_mask = var_33342_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33342_cast")];
            tensor<int32, [4]> var_33346_begin_0 = const()[name = tensor<string, []>("op_33346_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_33346_end_0 = const()[name = tensor<string, []>("op_33346_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_33346_end_mask_0 = const()[name = tensor<string, []>("op_33346_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33346_cast = slice_by_index(begin = var_33346_begin_0, end = var_33346_end_0, end_mask = var_33346_end_mask_0, x = q_149_cast)[name = tensor<string, []>("op_33346_cast")];
            tensor<int32, [4]> k_299_perm_0 = const()[name = tensor<string, []>("k_299_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_33353_begin_0 = const()[name = tensor<string, []>("op_33353_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_33353_end_0 = const()[name = tensor<string, []>("op_33353_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_33353_end_mask_0 = const()[name = tensor<string, []>("op_33353_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_65 = transpose(perm = k_299_perm_0, x = k_297_cast)[name = tensor<string, []>("transpose_65")];
            tensor<fp16, [2, 1024, 1, 64]> var_33353_cast = slice_by_index(begin = var_33353_begin_0, end = var_33353_end_0, end_mask = var_33353_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33353_cast")];
            tensor<int32, [4]> var_33357_begin_0 = const()[name = tensor<string, []>("op_33357_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_33357_end_0 = const()[name = tensor<string, []>("op_33357_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_33357_end_mask_0 = const()[name = tensor<string, []>("op_33357_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33357_cast = slice_by_index(begin = var_33357_begin_0, end = var_33357_end_0, end_mask = var_33357_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33357_cast")];
            tensor<int32, [4]> var_33361_begin_0 = const()[name = tensor<string, []>("op_33361_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_33361_end_0 = const()[name = tensor<string, []>("op_33361_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_33361_end_mask_0 = const()[name = tensor<string, []>("op_33361_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33361_cast = slice_by_index(begin = var_33361_begin_0, end = var_33361_end_0, end_mask = var_33361_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33361_cast")];
            tensor<int32, [4]> var_33365_begin_0 = const()[name = tensor<string, []>("op_33365_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_33365_end_0 = const()[name = tensor<string, []>("op_33365_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_33365_end_mask_0 = const()[name = tensor<string, []>("op_33365_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33365_cast = slice_by_index(begin = var_33365_begin_0, end = var_33365_end_0, end_mask = var_33365_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33365_cast")];
            tensor<int32, [4]> var_33369_begin_0 = const()[name = tensor<string, []>("op_33369_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_33369_end_0 = const()[name = tensor<string, []>("op_33369_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_33369_end_mask_0 = const()[name = tensor<string, []>("op_33369_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33369_cast = slice_by_index(begin = var_33369_begin_0, end = var_33369_end_0, end_mask = var_33369_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33369_cast")];
            tensor<int32, [4]> var_33373_begin_0 = const()[name = tensor<string, []>("op_33373_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_33373_end_0 = const()[name = tensor<string, []>("op_33373_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_33373_end_mask_0 = const()[name = tensor<string, []>("op_33373_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33373_cast = slice_by_index(begin = var_33373_begin_0, end = var_33373_end_0, end_mask = var_33373_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33373_cast")];
            tensor<int32, [4]> var_33377_begin_0 = const()[name = tensor<string, []>("op_33377_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_33377_end_0 = const()[name = tensor<string, []>("op_33377_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_33377_end_mask_0 = const()[name = tensor<string, []>("op_33377_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33377_cast = slice_by_index(begin = var_33377_begin_0, end = var_33377_end_0, end_mask = var_33377_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33377_cast")];
            tensor<int32, [4]> var_33381_begin_0 = const()[name = tensor<string, []>("op_33381_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_33381_end_0 = const()[name = tensor<string, []>("op_33381_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_33381_end_mask_0 = const()[name = tensor<string, []>("op_33381_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33381_cast = slice_by_index(begin = var_33381_begin_0, end = var_33381_end_0, end_mask = var_33381_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33381_cast")];
            tensor<int32, [4]> var_33385_begin_0 = const()[name = tensor<string, []>("op_33385_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_33385_end_0 = const()[name = tensor<string, []>("op_33385_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_33385_end_mask_0 = const()[name = tensor<string, []>("op_33385_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33385_cast = slice_by_index(begin = var_33385_begin_0, end = var_33385_end_0, end_mask = var_33385_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33385_cast")];
            tensor<int32, [4]> var_33389_begin_0 = const()[name = tensor<string, []>("op_33389_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_33389_end_0 = const()[name = tensor<string, []>("op_33389_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_33389_end_mask_0 = const()[name = tensor<string, []>("op_33389_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33389_cast = slice_by_index(begin = var_33389_begin_0, end = var_33389_end_0, end_mask = var_33389_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33389_cast")];
            tensor<int32, [4]> var_33393_begin_0 = const()[name = tensor<string, []>("op_33393_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_33393_end_0 = const()[name = tensor<string, []>("op_33393_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_33393_end_mask_0 = const()[name = tensor<string, []>("op_33393_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33393_cast = slice_by_index(begin = var_33393_begin_0, end = var_33393_end_0, end_mask = var_33393_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33393_cast")];
            tensor<int32, [4]> var_33397_begin_0 = const()[name = tensor<string, []>("op_33397_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_33397_end_0 = const()[name = tensor<string, []>("op_33397_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_33397_end_mask_0 = const()[name = tensor<string, []>("op_33397_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33397_cast = slice_by_index(begin = var_33397_begin_0, end = var_33397_end_0, end_mask = var_33397_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33397_cast")];
            tensor<int32, [4]> var_33401_begin_0 = const()[name = tensor<string, []>("op_33401_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_33401_end_0 = const()[name = tensor<string, []>("op_33401_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_33401_end_mask_0 = const()[name = tensor<string, []>("op_33401_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33401_cast = slice_by_index(begin = var_33401_begin_0, end = var_33401_end_0, end_mask = var_33401_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33401_cast")];
            tensor<int32, [4]> var_33405_begin_0 = const()[name = tensor<string, []>("op_33405_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_33405_end_0 = const()[name = tensor<string, []>("op_33405_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_33405_end_mask_0 = const()[name = tensor<string, []>("op_33405_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33405_cast = slice_by_index(begin = var_33405_begin_0, end = var_33405_end_0, end_mask = var_33405_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33405_cast")];
            tensor<int32, [4]> var_33409_begin_0 = const()[name = tensor<string, []>("op_33409_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_33409_end_0 = const()[name = tensor<string, []>("op_33409_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_33409_end_mask_0 = const()[name = tensor<string, []>("op_33409_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33409_cast = slice_by_index(begin = var_33409_begin_0, end = var_33409_end_0, end_mask = var_33409_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33409_cast")];
            tensor<int32, [4]> var_33413_begin_0 = const()[name = tensor<string, []>("op_33413_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_33413_end_0 = const()[name = tensor<string, []>("op_33413_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_33413_end_mask_0 = const()[name = tensor<string, []>("op_33413_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33413_cast = slice_by_index(begin = var_33413_begin_0, end = var_33413_end_0, end_mask = var_33413_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33413_cast")];
            tensor<int32, [4]> var_33417_begin_0 = const()[name = tensor<string, []>("op_33417_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_33417_end_0 = const()[name = tensor<string, []>("op_33417_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_33417_end_mask_0 = const()[name = tensor<string, []>("op_33417_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33417_cast = slice_by_index(begin = var_33417_begin_0, end = var_33417_end_0, end_mask = var_33417_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33417_cast")];
            tensor<int32, [4]> var_33421_begin_0 = const()[name = tensor<string, []>("op_33421_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_33421_end_0 = const()[name = tensor<string, []>("op_33421_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_33421_end_mask_0 = const()[name = tensor<string, []>("op_33421_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33421_cast = slice_by_index(begin = var_33421_begin_0, end = var_33421_end_0, end_mask = var_33421_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33421_cast")];
            tensor<int32, [4]> var_33425_begin_0 = const()[name = tensor<string, []>("op_33425_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_33425_end_0 = const()[name = tensor<string, []>("op_33425_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_33425_end_mask_0 = const()[name = tensor<string, []>("op_33425_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33425_cast = slice_by_index(begin = var_33425_begin_0, end = var_33425_end_0, end_mask = var_33425_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33425_cast")];
            tensor<int32, [4]> var_33429_begin_0 = const()[name = tensor<string, []>("op_33429_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_33429_end_0 = const()[name = tensor<string, []>("op_33429_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_33429_end_mask_0 = const()[name = tensor<string, []>("op_33429_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_33429_cast = slice_by_index(begin = var_33429_begin_0, end = var_33429_end_0, end_mask = var_33429_end_mask_0, x = transpose_65)[name = tensor<string, []>("op_33429_cast")];
            tensor<int32, [4]> var_33431_begin_0 = const()[name = tensor<string, []>("op_33431_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_33431_end_0 = const()[name = tensor<string, []>("op_33431_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_33431_end_mask_0 = const()[name = tensor<string, []>("op_33431_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33431_cast = slice_by_index(begin = var_33431_begin_0, end = var_33431_end_0, end_mask = var_33431_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33431_cast")];
            tensor<int32, [4]> var_33435_begin_0 = const()[name = tensor<string, []>("op_33435_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_33435_end_0 = const()[name = tensor<string, []>("op_33435_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_33435_end_mask_0 = const()[name = tensor<string, []>("op_33435_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33435_cast = slice_by_index(begin = var_33435_begin_0, end = var_33435_end_0, end_mask = var_33435_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33435_cast")];
            tensor<int32, [4]> var_33439_begin_0 = const()[name = tensor<string, []>("op_33439_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_33439_end_0 = const()[name = tensor<string, []>("op_33439_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_33439_end_mask_0 = const()[name = tensor<string, []>("op_33439_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33439_cast = slice_by_index(begin = var_33439_begin_0, end = var_33439_end_0, end_mask = var_33439_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33439_cast")];
            tensor<int32, [4]> var_33443_begin_0 = const()[name = tensor<string, []>("op_33443_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_33443_end_0 = const()[name = tensor<string, []>("op_33443_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_33443_end_mask_0 = const()[name = tensor<string, []>("op_33443_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33443_cast = slice_by_index(begin = var_33443_begin_0, end = var_33443_end_0, end_mask = var_33443_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33443_cast")];
            tensor<int32, [4]> var_33447_begin_0 = const()[name = tensor<string, []>("op_33447_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_33447_end_0 = const()[name = tensor<string, []>("op_33447_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_33447_end_mask_0 = const()[name = tensor<string, []>("op_33447_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33447_cast = slice_by_index(begin = var_33447_begin_0, end = var_33447_end_0, end_mask = var_33447_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33447_cast")];
            tensor<int32, [4]> var_33451_begin_0 = const()[name = tensor<string, []>("op_33451_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_33451_end_0 = const()[name = tensor<string, []>("op_33451_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_33451_end_mask_0 = const()[name = tensor<string, []>("op_33451_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33451_cast = slice_by_index(begin = var_33451_begin_0, end = var_33451_end_0, end_mask = var_33451_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33451_cast")];
            tensor<int32, [4]> var_33455_begin_0 = const()[name = tensor<string, []>("op_33455_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_33455_end_0 = const()[name = tensor<string, []>("op_33455_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_33455_end_mask_0 = const()[name = tensor<string, []>("op_33455_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33455_cast = slice_by_index(begin = var_33455_begin_0, end = var_33455_end_0, end_mask = var_33455_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33455_cast")];
            tensor<int32, [4]> var_33459_begin_0 = const()[name = tensor<string, []>("op_33459_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_33459_end_0 = const()[name = tensor<string, []>("op_33459_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_33459_end_mask_0 = const()[name = tensor<string, []>("op_33459_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33459_cast = slice_by_index(begin = var_33459_begin_0, end = var_33459_end_0, end_mask = var_33459_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33459_cast")];
            tensor<int32, [4]> var_33463_begin_0 = const()[name = tensor<string, []>("op_33463_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_33463_end_0 = const()[name = tensor<string, []>("op_33463_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_33463_end_mask_0 = const()[name = tensor<string, []>("op_33463_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33463_cast = slice_by_index(begin = var_33463_begin_0, end = var_33463_end_0, end_mask = var_33463_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33463_cast")];
            tensor<int32, [4]> var_33467_begin_0 = const()[name = tensor<string, []>("op_33467_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_33467_end_0 = const()[name = tensor<string, []>("op_33467_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_33467_end_mask_0 = const()[name = tensor<string, []>("op_33467_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33467_cast = slice_by_index(begin = var_33467_begin_0, end = var_33467_end_0, end_mask = var_33467_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33467_cast")];
            tensor<int32, [4]> var_33471_begin_0 = const()[name = tensor<string, []>("op_33471_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_33471_end_0 = const()[name = tensor<string, []>("op_33471_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_33471_end_mask_0 = const()[name = tensor<string, []>("op_33471_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33471_cast = slice_by_index(begin = var_33471_begin_0, end = var_33471_end_0, end_mask = var_33471_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33471_cast")];
            tensor<int32, [4]> var_33475_begin_0 = const()[name = tensor<string, []>("op_33475_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_33475_end_0 = const()[name = tensor<string, []>("op_33475_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_33475_end_mask_0 = const()[name = tensor<string, []>("op_33475_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33475_cast = slice_by_index(begin = var_33475_begin_0, end = var_33475_end_0, end_mask = var_33475_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33475_cast")];
            tensor<int32, [4]> var_33479_begin_0 = const()[name = tensor<string, []>("op_33479_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_33479_end_0 = const()[name = tensor<string, []>("op_33479_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_33479_end_mask_0 = const()[name = tensor<string, []>("op_33479_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33479_cast = slice_by_index(begin = var_33479_begin_0, end = var_33479_end_0, end_mask = var_33479_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33479_cast")];
            tensor<int32, [4]> var_33483_begin_0 = const()[name = tensor<string, []>("op_33483_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_33483_end_0 = const()[name = tensor<string, []>("op_33483_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_33483_end_mask_0 = const()[name = tensor<string, []>("op_33483_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33483_cast = slice_by_index(begin = var_33483_begin_0, end = var_33483_end_0, end_mask = var_33483_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33483_cast")];
            tensor<int32, [4]> var_33487_begin_0 = const()[name = tensor<string, []>("op_33487_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_33487_end_0 = const()[name = tensor<string, []>("op_33487_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_33487_end_mask_0 = const()[name = tensor<string, []>("op_33487_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33487_cast = slice_by_index(begin = var_33487_begin_0, end = var_33487_end_0, end_mask = var_33487_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33487_cast")];
            tensor<int32, [4]> var_33491_begin_0 = const()[name = tensor<string, []>("op_33491_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_33491_end_0 = const()[name = tensor<string, []>("op_33491_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_33491_end_mask_0 = const()[name = tensor<string, []>("op_33491_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33491_cast = slice_by_index(begin = var_33491_begin_0, end = var_33491_end_0, end_mask = var_33491_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33491_cast")];
            tensor<int32, [4]> var_33495_begin_0 = const()[name = tensor<string, []>("op_33495_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_33495_end_0 = const()[name = tensor<string, []>("op_33495_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_33495_end_mask_0 = const()[name = tensor<string, []>("op_33495_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33495_cast = slice_by_index(begin = var_33495_begin_0, end = var_33495_end_0, end_mask = var_33495_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33495_cast")];
            tensor<int32, [4]> var_33499_begin_0 = const()[name = tensor<string, []>("op_33499_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_33499_end_0 = const()[name = tensor<string, []>("op_33499_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_33499_end_mask_0 = const()[name = tensor<string, []>("op_33499_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33499_cast = slice_by_index(begin = var_33499_begin_0, end = var_33499_end_0, end_mask = var_33499_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33499_cast")];
            tensor<int32, [4]> var_33503_begin_0 = const()[name = tensor<string, []>("op_33503_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_33503_end_0 = const()[name = tensor<string, []>("op_33503_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_33503_end_mask_0 = const()[name = tensor<string, []>("op_33503_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33503_cast = slice_by_index(begin = var_33503_begin_0, end = var_33503_end_0, end_mask = var_33503_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33503_cast")];
            tensor<int32, [4]> var_33507_begin_0 = const()[name = tensor<string, []>("op_33507_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_33507_end_0 = const()[name = tensor<string, []>("op_33507_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_33507_end_mask_0 = const()[name = tensor<string, []>("op_33507_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33507_cast = slice_by_index(begin = var_33507_begin_0, end = var_33507_end_0, end_mask = var_33507_end_mask_0, x = v_149_cast)[name = tensor<string, []>("op_33507_cast")];
            tensor<string, []> var_33511_equation_0 = const()[name = tensor<string, []>("op_33511_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33511_cast = einsum(equation = var_33511_equation_0, values = (var_33353_cast, var_33270_cast))[name = tensor<string, []>("op_33511_cast")];
            tensor<fp16, []> var_33512_to_fp16 = const()[name = tensor<string, []>("op_33512_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2801_cast = mul(x = var_33511_cast, y = var_33512_to_fp16)[name = tensor<string, []>("aw_2801_cast")];
            tensor<string, []> var_33515_equation_0 = const()[name = tensor<string, []>("op_33515_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33515_cast = einsum(equation = var_33515_equation_0, values = (var_33357_cast, var_33274_cast))[name = tensor<string, []>("op_33515_cast")];
            tensor<fp16, []> var_33516_to_fp16 = const()[name = tensor<string, []>("op_33516_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2803_cast = mul(x = var_33515_cast, y = var_33516_to_fp16)[name = tensor<string, []>("aw_2803_cast")];
            tensor<string, []> var_33519_equation_0 = const()[name = tensor<string, []>("op_33519_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33519_cast = einsum(equation = var_33519_equation_0, values = (var_33361_cast, var_33278_cast))[name = tensor<string, []>("op_33519_cast")];
            tensor<fp16, []> var_33520_to_fp16 = const()[name = tensor<string, []>("op_33520_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2805_cast = mul(x = var_33519_cast, y = var_33520_to_fp16)[name = tensor<string, []>("aw_2805_cast")];
            tensor<string, []> var_33523_equation_0 = const()[name = tensor<string, []>("op_33523_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33523_cast = einsum(equation = var_33523_equation_0, values = (var_33365_cast, var_33282_cast))[name = tensor<string, []>("op_33523_cast")];
            tensor<fp16, []> var_33524_to_fp16 = const()[name = tensor<string, []>("op_33524_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2807_cast = mul(x = var_33523_cast, y = var_33524_to_fp16)[name = tensor<string, []>("aw_2807_cast")];
            tensor<string, []> var_33527_equation_0 = const()[name = tensor<string, []>("op_33527_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33527_cast = einsum(equation = var_33527_equation_0, values = (var_33369_cast, var_33286_cast))[name = tensor<string, []>("op_33527_cast")];
            tensor<fp16, []> var_33528_to_fp16 = const()[name = tensor<string, []>("op_33528_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2809_cast = mul(x = var_33527_cast, y = var_33528_to_fp16)[name = tensor<string, []>("aw_2809_cast")];
            tensor<string, []> var_33531_equation_0 = const()[name = tensor<string, []>("op_33531_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33531_cast = einsum(equation = var_33531_equation_0, values = (var_33373_cast, var_33290_cast))[name = tensor<string, []>("op_33531_cast")];
            tensor<fp16, []> var_33532_to_fp16 = const()[name = tensor<string, []>("op_33532_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2811_cast = mul(x = var_33531_cast, y = var_33532_to_fp16)[name = tensor<string, []>("aw_2811_cast")];
            tensor<string, []> var_33535_equation_0 = const()[name = tensor<string, []>("op_33535_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33535_cast = einsum(equation = var_33535_equation_0, values = (var_33377_cast, var_33294_cast))[name = tensor<string, []>("op_33535_cast")];
            tensor<fp16, []> var_33536_to_fp16 = const()[name = tensor<string, []>("op_33536_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2813_cast = mul(x = var_33535_cast, y = var_33536_to_fp16)[name = tensor<string, []>("aw_2813_cast")];
            tensor<string, []> var_33539_equation_0 = const()[name = tensor<string, []>("op_33539_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33539_cast = einsum(equation = var_33539_equation_0, values = (var_33381_cast, var_33298_cast))[name = tensor<string, []>("op_33539_cast")];
            tensor<fp16, []> var_33540_to_fp16 = const()[name = tensor<string, []>("op_33540_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2815_cast = mul(x = var_33539_cast, y = var_33540_to_fp16)[name = tensor<string, []>("aw_2815_cast")];
            tensor<string, []> var_33543_equation_0 = const()[name = tensor<string, []>("op_33543_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33543_cast = einsum(equation = var_33543_equation_0, values = (var_33385_cast, var_33302_cast))[name = tensor<string, []>("op_33543_cast")];
            tensor<fp16, []> var_33544_to_fp16 = const()[name = tensor<string, []>("op_33544_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2817_cast = mul(x = var_33543_cast, y = var_33544_to_fp16)[name = tensor<string, []>("aw_2817_cast")];
            tensor<string, []> var_33547_equation_0 = const()[name = tensor<string, []>("op_33547_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33547_cast = einsum(equation = var_33547_equation_0, values = (var_33389_cast, var_33306_cast))[name = tensor<string, []>("op_33547_cast")];
            tensor<fp16, []> var_33548_to_fp16 = const()[name = tensor<string, []>("op_33548_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2819_cast = mul(x = var_33547_cast, y = var_33548_to_fp16)[name = tensor<string, []>("aw_2819_cast")];
            tensor<string, []> var_33551_equation_0 = const()[name = tensor<string, []>("op_33551_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33551_cast = einsum(equation = var_33551_equation_0, values = (var_33393_cast, var_33310_cast))[name = tensor<string, []>("op_33551_cast")];
            tensor<fp16, []> var_33552_to_fp16 = const()[name = tensor<string, []>("op_33552_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2821_cast = mul(x = var_33551_cast, y = var_33552_to_fp16)[name = tensor<string, []>("aw_2821_cast")];
            tensor<string, []> var_33555_equation_0 = const()[name = tensor<string, []>("op_33555_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33555_cast = einsum(equation = var_33555_equation_0, values = (var_33397_cast, var_33314_cast))[name = tensor<string, []>("op_33555_cast")];
            tensor<fp16, []> var_33556_to_fp16 = const()[name = tensor<string, []>("op_33556_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2823_cast = mul(x = var_33555_cast, y = var_33556_to_fp16)[name = tensor<string, []>("aw_2823_cast")];
            tensor<string, []> var_33559_equation_0 = const()[name = tensor<string, []>("op_33559_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33559_cast = einsum(equation = var_33559_equation_0, values = (var_33401_cast, var_33318_cast))[name = tensor<string, []>("op_33559_cast")];
            tensor<fp16, []> var_33560_to_fp16 = const()[name = tensor<string, []>("op_33560_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2825_cast = mul(x = var_33559_cast, y = var_33560_to_fp16)[name = tensor<string, []>("aw_2825_cast")];
            tensor<string, []> var_33563_equation_0 = const()[name = tensor<string, []>("op_33563_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33563_cast = einsum(equation = var_33563_equation_0, values = (var_33405_cast, var_33322_cast))[name = tensor<string, []>("op_33563_cast")];
            tensor<fp16, []> var_33564_to_fp16 = const()[name = tensor<string, []>("op_33564_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2827_cast = mul(x = var_33563_cast, y = var_33564_to_fp16)[name = tensor<string, []>("aw_2827_cast")];
            tensor<string, []> var_33567_equation_0 = const()[name = tensor<string, []>("op_33567_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33567_cast = einsum(equation = var_33567_equation_0, values = (var_33409_cast, var_33326_cast))[name = tensor<string, []>("op_33567_cast")];
            tensor<fp16, []> var_33568_to_fp16 = const()[name = tensor<string, []>("op_33568_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2829_cast = mul(x = var_33567_cast, y = var_33568_to_fp16)[name = tensor<string, []>("aw_2829_cast")];
            tensor<string, []> var_33571_equation_0 = const()[name = tensor<string, []>("op_33571_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33571_cast = einsum(equation = var_33571_equation_0, values = (var_33413_cast, var_33330_cast))[name = tensor<string, []>("op_33571_cast")];
            tensor<fp16, []> var_33572_to_fp16 = const()[name = tensor<string, []>("op_33572_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2831_cast = mul(x = var_33571_cast, y = var_33572_to_fp16)[name = tensor<string, []>("aw_2831_cast")];
            tensor<string, []> var_33575_equation_0 = const()[name = tensor<string, []>("op_33575_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33575_cast = einsum(equation = var_33575_equation_0, values = (var_33417_cast, var_33334_cast))[name = tensor<string, []>("op_33575_cast")];
            tensor<fp16, []> var_33576_to_fp16 = const()[name = tensor<string, []>("op_33576_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2833_cast = mul(x = var_33575_cast, y = var_33576_to_fp16)[name = tensor<string, []>("aw_2833_cast")];
            tensor<string, []> var_33579_equation_0 = const()[name = tensor<string, []>("op_33579_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33579_cast = einsum(equation = var_33579_equation_0, values = (var_33421_cast, var_33338_cast))[name = tensor<string, []>("op_33579_cast")];
            tensor<fp16, []> var_33580_to_fp16 = const()[name = tensor<string, []>("op_33580_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2835_cast = mul(x = var_33579_cast, y = var_33580_to_fp16)[name = tensor<string, []>("aw_2835_cast")];
            tensor<string, []> var_33583_equation_0 = const()[name = tensor<string, []>("op_33583_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33583_cast = einsum(equation = var_33583_equation_0, values = (var_33425_cast, var_33342_cast))[name = tensor<string, []>("op_33583_cast")];
            tensor<fp16, []> var_33584_to_fp16 = const()[name = tensor<string, []>("op_33584_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2837_cast = mul(x = var_33583_cast, y = var_33584_to_fp16)[name = tensor<string, []>("aw_2837_cast")];
            tensor<string, []> var_33587_equation_0 = const()[name = tensor<string, []>("op_33587_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33587_cast = einsum(equation = var_33587_equation_0, values = (var_33429_cast, var_33346_cast))[name = tensor<string, []>("op_33587_cast")];
            tensor<fp16, []> var_33588_to_fp16 = const()[name = tensor<string, []>("op_33588_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2839_cast = mul(x = var_33587_cast, y = var_33588_to_fp16)[name = tensor<string, []>("aw_2839_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33590_cast = softmax(axis = var_30385, x = aw_2801_cast)[name = tensor<string, []>("op_33590_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33591_cast = softmax(axis = var_30385, x = aw_2803_cast)[name = tensor<string, []>("op_33591_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33592_cast = softmax(axis = var_30385, x = aw_2805_cast)[name = tensor<string, []>("op_33592_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33593_cast = softmax(axis = var_30385, x = aw_2807_cast)[name = tensor<string, []>("op_33593_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33594_cast = softmax(axis = var_30385, x = aw_2809_cast)[name = tensor<string, []>("op_33594_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33595_cast = softmax(axis = var_30385, x = aw_2811_cast)[name = tensor<string, []>("op_33595_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33596_cast = softmax(axis = var_30385, x = aw_2813_cast)[name = tensor<string, []>("op_33596_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33597_cast = softmax(axis = var_30385, x = aw_2815_cast)[name = tensor<string, []>("op_33597_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33598_cast = softmax(axis = var_30385, x = aw_2817_cast)[name = tensor<string, []>("op_33598_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33599_cast = softmax(axis = var_30385, x = aw_2819_cast)[name = tensor<string, []>("op_33599_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33600_cast = softmax(axis = var_30385, x = aw_2821_cast)[name = tensor<string, []>("op_33600_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33601_cast = softmax(axis = var_30385, x = aw_2823_cast)[name = tensor<string, []>("op_33601_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33602_cast = softmax(axis = var_30385, x = aw_2825_cast)[name = tensor<string, []>("op_33602_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33603_cast = softmax(axis = var_30385, x = aw_2827_cast)[name = tensor<string, []>("op_33603_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33604_cast = softmax(axis = var_30385, x = aw_2829_cast)[name = tensor<string, []>("op_33604_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33605_cast = softmax(axis = var_30385, x = aw_2831_cast)[name = tensor<string, []>("op_33605_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33606_cast = softmax(axis = var_30385, x = aw_2833_cast)[name = tensor<string, []>("op_33606_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33607_cast = softmax(axis = var_30385, x = aw_2835_cast)[name = tensor<string, []>("op_33607_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33608_cast = softmax(axis = var_30385, x = aw_2837_cast)[name = tensor<string, []>("op_33608_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_33609_cast = softmax(axis = var_30385, x = aw_2839_cast)[name = tensor<string, []>("op_33609_cast")];
            tensor<string, []> var_33611_equation_0 = const()[name = tensor<string, []>("op_33611_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33611_cast = einsum(equation = var_33611_equation_0, values = (var_33431_cast, var_33590_cast))[name = tensor<string, []>("op_33611_cast")];
            tensor<string, []> var_33613_equation_0 = const()[name = tensor<string, []>("op_33613_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33613_cast = einsum(equation = var_33613_equation_0, values = (var_33435_cast, var_33591_cast))[name = tensor<string, []>("op_33613_cast")];
            tensor<string, []> var_33615_equation_0 = const()[name = tensor<string, []>("op_33615_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33615_cast = einsum(equation = var_33615_equation_0, values = (var_33439_cast, var_33592_cast))[name = tensor<string, []>("op_33615_cast")];
            tensor<string, []> var_33617_equation_0 = const()[name = tensor<string, []>("op_33617_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33617_cast = einsum(equation = var_33617_equation_0, values = (var_33443_cast, var_33593_cast))[name = tensor<string, []>("op_33617_cast")];
            tensor<string, []> var_33619_equation_0 = const()[name = tensor<string, []>("op_33619_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33619_cast = einsum(equation = var_33619_equation_0, values = (var_33447_cast, var_33594_cast))[name = tensor<string, []>("op_33619_cast")];
            tensor<string, []> var_33621_equation_0 = const()[name = tensor<string, []>("op_33621_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33621_cast = einsum(equation = var_33621_equation_0, values = (var_33451_cast, var_33595_cast))[name = tensor<string, []>("op_33621_cast")];
            tensor<string, []> var_33623_equation_0 = const()[name = tensor<string, []>("op_33623_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33623_cast = einsum(equation = var_33623_equation_0, values = (var_33455_cast, var_33596_cast))[name = tensor<string, []>("op_33623_cast")];
            tensor<string, []> var_33625_equation_0 = const()[name = tensor<string, []>("op_33625_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33625_cast = einsum(equation = var_33625_equation_0, values = (var_33459_cast, var_33597_cast))[name = tensor<string, []>("op_33625_cast")];
            tensor<string, []> var_33627_equation_0 = const()[name = tensor<string, []>("op_33627_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33627_cast = einsum(equation = var_33627_equation_0, values = (var_33463_cast, var_33598_cast))[name = tensor<string, []>("op_33627_cast")];
            tensor<string, []> var_33629_equation_0 = const()[name = tensor<string, []>("op_33629_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33629_cast = einsum(equation = var_33629_equation_0, values = (var_33467_cast, var_33599_cast))[name = tensor<string, []>("op_33629_cast")];
            tensor<string, []> var_33631_equation_0 = const()[name = tensor<string, []>("op_33631_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33631_cast = einsum(equation = var_33631_equation_0, values = (var_33471_cast, var_33600_cast))[name = tensor<string, []>("op_33631_cast")];
            tensor<string, []> var_33633_equation_0 = const()[name = tensor<string, []>("op_33633_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33633_cast = einsum(equation = var_33633_equation_0, values = (var_33475_cast, var_33601_cast))[name = tensor<string, []>("op_33633_cast")];
            tensor<string, []> var_33635_equation_0 = const()[name = tensor<string, []>("op_33635_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33635_cast = einsum(equation = var_33635_equation_0, values = (var_33479_cast, var_33602_cast))[name = tensor<string, []>("op_33635_cast")];
            tensor<string, []> var_33637_equation_0 = const()[name = tensor<string, []>("op_33637_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33637_cast = einsum(equation = var_33637_equation_0, values = (var_33483_cast, var_33603_cast))[name = tensor<string, []>("op_33637_cast")];
            tensor<string, []> var_33639_equation_0 = const()[name = tensor<string, []>("op_33639_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33639_cast = einsum(equation = var_33639_equation_0, values = (var_33487_cast, var_33604_cast))[name = tensor<string, []>("op_33639_cast")];
            tensor<string, []> var_33641_equation_0 = const()[name = tensor<string, []>("op_33641_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33641_cast = einsum(equation = var_33641_equation_0, values = (var_33491_cast, var_33605_cast))[name = tensor<string, []>("op_33641_cast")];
            tensor<string, []> var_33643_equation_0 = const()[name = tensor<string, []>("op_33643_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33643_cast = einsum(equation = var_33643_equation_0, values = (var_33495_cast, var_33606_cast))[name = tensor<string, []>("op_33643_cast")];
            tensor<string, []> var_33645_equation_0 = const()[name = tensor<string, []>("op_33645_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33645_cast = einsum(equation = var_33645_equation_0, values = (var_33499_cast, var_33607_cast))[name = tensor<string, []>("op_33645_cast")];
            tensor<string, []> var_33647_equation_0 = const()[name = tensor<string, []>("op_33647_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33647_cast = einsum(equation = var_33647_equation_0, values = (var_33503_cast, var_33608_cast))[name = tensor<string, []>("op_33647_cast")];
            tensor<string, []> var_33649_equation_0 = const()[name = tensor<string, []>("op_33649_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_33649_cast = einsum(equation = var_33649_equation_0, values = (var_33507_cast, var_33609_cast))[name = tensor<string, []>("op_33649_cast")];
            tensor<bool, []> input_465_interleave_0 = const()[name = tensor<string, []>("input_465_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_465_cast = concat(axis = var_30385, interleave = input_465_interleave_0, values = (var_33611_cast, var_33613_cast, var_33615_cast, var_33617_cast, var_33619_cast, var_33621_cast, var_33623_cast, var_33625_cast, var_33627_cast, var_33629_cast, var_33631_cast, var_33633_cast, var_33635_cast, var_33637_cast, var_33639_cast, var_33641_cast, var_33643_cast, var_33645_cast, var_33647_cast, var_33649_cast))[name = tensor<string, []>("input_465_cast")];
            tensor<int32, [2]> var_33655 = const()[name = tensor<string, []>("op_33655"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_33657 = const()[name = tensor<string, []>("op_33657"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_33659_pad_type_0 = const()[name = tensor<string, []>("op_33659_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_33659_pad_0 = const()[name = tensor<string, []>("op_33659_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2806796096)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2810072960)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_33659_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_out_0_bias_to_fp16, dilations = var_33657, groups = var_30385, pad = var_33659_pad_0, pad_type = var_33659_pad_type_0, strides = var_33655, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn1_to_out_0_weight_to_fp16, x = input_465_cast)[name = tensor<string, []>("op_33659_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_225_cast = add(x = var_33659_cast, y = inputs_223_cast)[name = tensor<string, []>("inputs_225_cast")];
            tensor<int32, [1]> var_33663 = const()[name = tensor<string, []>("op_33663"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_225_cast = reduce_mean(axes = var_33663, keep_dims = var_30380, x = inputs_225_cast)[name = tensor<string, []>("channels_mean_225_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_225_cast = sub(x = inputs_225_cast, y = channels_mean_225_cast)[name = tensor<string, []>("zero_mean_225_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_225_cast = mul(x = zero_mean_225_cast, y = zero_mean_225_cast)[name = tensor<string, []>("zero_mean_sq_225_cast")];
            tensor<int32, [1]> var_33667 = const()[name = tensor<string, []>("op_33667"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_33668_cast = reduce_mean(axes = var_33667, keep_dims = var_30380, x = zero_mean_sq_225_cast)[name = tensor<string, []>("op_33668_cast")];
            tensor<fp16, []> var_33669_to_fp16 = const()[name = tensor<string, []>("op_33669_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_33670_cast = add(x = var_33668_cast, y = var_33669_to_fp16)[name = tensor<string, []>("op_33670_cast")];
            tensor<fp16, []> denom_225_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_225_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_225_cast = rsqrt(epsilon = denom_225_epsilon_0_to_fp16, x = var_33670_cast)[name = tensor<string, []>("denom_225_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_225_cast = mul(x = zero_mean_225_cast, y = denom_225_cast)[name = tensor<string, []>("out_225_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_33674_to_fp16 = const()[name = tensor<string, []>("op_33674_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2810075584)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_33675_cast = add(x = out_225_cast, y = var_33674_to_fp16)[name = tensor<string, []>("op_33675_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_33677_to_fp16 = const()[name = tensor<string, []>("op_33677_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2810078208)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_307_cast = mul(x = var_33675_cast, y = var_33677_to_fp16)[name = tensor<string, []>("hidden_states_307_cast")];
            tensor<int32, [2]> var_33684 = const()[name = tensor<string, []>("op_33684"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_33686 = const()[name = tensor<string, []>("op_33686"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_151_pad_type_0 = const()[name = tensor<string, []>("q_151_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_151_pad_0 = const()[name = tensor<string, []>("q_151_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2810080832)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_151_cast = conv(dilations = var_33686, groups = var_30385, pad = q_151_pad_0, pad_type = q_151_pad_type_0, strides = var_33684, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_q_weight_to_fp16, x = hidden_states_307_cast)[name = tensor<string, []>("q_151_cast")];
            tensor<int32, [2]> var_33690 = const()[name = tensor<string, []>("op_33690"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_33692 = const()[name = tensor<string, []>("op_33692"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_301_pad_type_0 = const()[name = tensor<string, []>("k_301_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_301_pad_0 = const()[name = tensor<string, []>("k_301_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2813357696)))];
            tensor<fp16, [2, 1280, 1, 77]> k_301_cast = conv(dilations = var_33692, groups = var_30385, pad = k_301_pad_0, pad_type = k_301_pad_type_0, strides = var_33690, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_301_cast")];
            tensor<int32, [2]> var_33696 = const()[name = tensor<string, []>("op_33696"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_33698 = const()[name = tensor<string, []>("op_33698"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_151_pad_type_0 = const()[name = tensor<string, []>("v_151_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_151_pad_0 = const()[name = tensor<string, []>("v_151_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2818600640)))];
            tensor<fp16, [2, 1280, 1, 77]> v_151_cast = conv(dilations = var_33698, groups = var_30385, pad = v_151_pad_0, pad_type = v_151_pad_type_0, strides = var_33696, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_151_cast")];
            tensor<int32, [4]> var_33702_begin_0 = const()[name = tensor<string, []>("op_33702_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_33702_end_0 = const()[name = tensor<string, []>("op_33702_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_33702_end_mask_0 = const()[name = tensor<string, []>("op_33702_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33702_cast = slice_by_index(begin = var_33702_begin_0, end = var_33702_end_0, end_mask = var_33702_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33702_cast")];
            tensor<int32, [4]> var_33706_begin_0 = const()[name = tensor<string, []>("op_33706_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_33706_end_0 = const()[name = tensor<string, []>("op_33706_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_33706_end_mask_0 = const()[name = tensor<string, []>("op_33706_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33706_cast = slice_by_index(begin = var_33706_begin_0, end = var_33706_end_0, end_mask = var_33706_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33706_cast")];
            tensor<int32, [4]> var_33710_begin_0 = const()[name = tensor<string, []>("op_33710_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_33710_end_0 = const()[name = tensor<string, []>("op_33710_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_33710_end_mask_0 = const()[name = tensor<string, []>("op_33710_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33710_cast = slice_by_index(begin = var_33710_begin_0, end = var_33710_end_0, end_mask = var_33710_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33710_cast")];
            tensor<int32, [4]> var_33714_begin_0 = const()[name = tensor<string, []>("op_33714_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_33714_end_0 = const()[name = tensor<string, []>("op_33714_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_33714_end_mask_0 = const()[name = tensor<string, []>("op_33714_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33714_cast = slice_by_index(begin = var_33714_begin_0, end = var_33714_end_0, end_mask = var_33714_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33714_cast")];
            tensor<int32, [4]> var_33718_begin_0 = const()[name = tensor<string, []>("op_33718_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_33718_end_0 = const()[name = tensor<string, []>("op_33718_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_33718_end_mask_0 = const()[name = tensor<string, []>("op_33718_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33718_cast = slice_by_index(begin = var_33718_begin_0, end = var_33718_end_0, end_mask = var_33718_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33718_cast")];
            tensor<int32, [4]> var_33722_begin_0 = const()[name = tensor<string, []>("op_33722_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_33722_end_0 = const()[name = tensor<string, []>("op_33722_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_33722_end_mask_0 = const()[name = tensor<string, []>("op_33722_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33722_cast = slice_by_index(begin = var_33722_begin_0, end = var_33722_end_0, end_mask = var_33722_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33722_cast")];
            tensor<int32, [4]> var_33726_begin_0 = const()[name = tensor<string, []>("op_33726_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_33726_end_0 = const()[name = tensor<string, []>("op_33726_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_33726_end_mask_0 = const()[name = tensor<string, []>("op_33726_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33726_cast = slice_by_index(begin = var_33726_begin_0, end = var_33726_end_0, end_mask = var_33726_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33726_cast")];
            tensor<int32, [4]> var_33730_begin_0 = const()[name = tensor<string, []>("op_33730_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_33730_end_0 = const()[name = tensor<string, []>("op_33730_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_33730_end_mask_0 = const()[name = tensor<string, []>("op_33730_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33730_cast = slice_by_index(begin = var_33730_begin_0, end = var_33730_end_0, end_mask = var_33730_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33730_cast")];
            tensor<int32, [4]> var_33734_begin_0 = const()[name = tensor<string, []>("op_33734_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_33734_end_0 = const()[name = tensor<string, []>("op_33734_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_33734_end_mask_0 = const()[name = tensor<string, []>("op_33734_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33734_cast = slice_by_index(begin = var_33734_begin_0, end = var_33734_end_0, end_mask = var_33734_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33734_cast")];
            tensor<int32, [4]> var_33738_begin_0 = const()[name = tensor<string, []>("op_33738_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_33738_end_0 = const()[name = tensor<string, []>("op_33738_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_33738_end_mask_0 = const()[name = tensor<string, []>("op_33738_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33738_cast = slice_by_index(begin = var_33738_begin_0, end = var_33738_end_0, end_mask = var_33738_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33738_cast")];
            tensor<int32, [4]> var_33742_begin_0 = const()[name = tensor<string, []>("op_33742_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_33742_end_0 = const()[name = tensor<string, []>("op_33742_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_33742_end_mask_0 = const()[name = tensor<string, []>("op_33742_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33742_cast = slice_by_index(begin = var_33742_begin_0, end = var_33742_end_0, end_mask = var_33742_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33742_cast")];
            tensor<int32, [4]> var_33746_begin_0 = const()[name = tensor<string, []>("op_33746_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_33746_end_0 = const()[name = tensor<string, []>("op_33746_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_33746_end_mask_0 = const()[name = tensor<string, []>("op_33746_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33746_cast = slice_by_index(begin = var_33746_begin_0, end = var_33746_end_0, end_mask = var_33746_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33746_cast")];
            tensor<int32, [4]> var_33750_begin_0 = const()[name = tensor<string, []>("op_33750_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_33750_end_0 = const()[name = tensor<string, []>("op_33750_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_33750_end_mask_0 = const()[name = tensor<string, []>("op_33750_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33750_cast = slice_by_index(begin = var_33750_begin_0, end = var_33750_end_0, end_mask = var_33750_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33750_cast")];
            tensor<int32, [4]> var_33754_begin_0 = const()[name = tensor<string, []>("op_33754_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_33754_end_0 = const()[name = tensor<string, []>("op_33754_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_33754_end_mask_0 = const()[name = tensor<string, []>("op_33754_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33754_cast = slice_by_index(begin = var_33754_begin_0, end = var_33754_end_0, end_mask = var_33754_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33754_cast")];
            tensor<int32, [4]> var_33758_begin_0 = const()[name = tensor<string, []>("op_33758_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_33758_end_0 = const()[name = tensor<string, []>("op_33758_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_33758_end_mask_0 = const()[name = tensor<string, []>("op_33758_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33758_cast = slice_by_index(begin = var_33758_begin_0, end = var_33758_end_0, end_mask = var_33758_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33758_cast")];
            tensor<int32, [4]> var_33762_begin_0 = const()[name = tensor<string, []>("op_33762_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_33762_end_0 = const()[name = tensor<string, []>("op_33762_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_33762_end_mask_0 = const()[name = tensor<string, []>("op_33762_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33762_cast = slice_by_index(begin = var_33762_begin_0, end = var_33762_end_0, end_mask = var_33762_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33762_cast")];
            tensor<int32, [4]> var_33766_begin_0 = const()[name = tensor<string, []>("op_33766_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_33766_end_0 = const()[name = tensor<string, []>("op_33766_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_33766_end_mask_0 = const()[name = tensor<string, []>("op_33766_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33766_cast = slice_by_index(begin = var_33766_begin_0, end = var_33766_end_0, end_mask = var_33766_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33766_cast")];
            tensor<int32, [4]> var_33770_begin_0 = const()[name = tensor<string, []>("op_33770_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_33770_end_0 = const()[name = tensor<string, []>("op_33770_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_33770_end_mask_0 = const()[name = tensor<string, []>("op_33770_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33770_cast = slice_by_index(begin = var_33770_begin_0, end = var_33770_end_0, end_mask = var_33770_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33770_cast")];
            tensor<int32, [4]> var_33774_begin_0 = const()[name = tensor<string, []>("op_33774_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_33774_end_0 = const()[name = tensor<string, []>("op_33774_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_33774_end_mask_0 = const()[name = tensor<string, []>("op_33774_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33774_cast = slice_by_index(begin = var_33774_begin_0, end = var_33774_end_0, end_mask = var_33774_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33774_cast")];
            tensor<int32, [4]> var_33778_begin_0 = const()[name = tensor<string, []>("op_33778_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_33778_end_0 = const()[name = tensor<string, []>("op_33778_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_33778_end_mask_0 = const()[name = tensor<string, []>("op_33778_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_33778_cast = slice_by_index(begin = var_33778_begin_0, end = var_33778_end_0, end_mask = var_33778_end_mask_0, x = q_151_cast)[name = tensor<string, []>("op_33778_cast")];
            tensor<int32, [4]> k_303_perm_0 = const()[name = tensor<string, []>("k_303_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_33785_begin_0 = const()[name = tensor<string, []>("op_33785_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_33785_end_0 = const()[name = tensor<string, []>("op_33785_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_33785_end_mask_0 = const()[name = tensor<string, []>("op_33785_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_64 = transpose(perm = k_303_perm_0, x = k_301_cast)[name = tensor<string, []>("transpose_64")];
            tensor<fp16, [2, 77, 1, 64]> var_33785_cast = slice_by_index(begin = var_33785_begin_0, end = var_33785_end_0, end_mask = var_33785_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33785_cast")];
            tensor<int32, [4]> var_33789_begin_0 = const()[name = tensor<string, []>("op_33789_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_33789_end_0 = const()[name = tensor<string, []>("op_33789_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_33789_end_mask_0 = const()[name = tensor<string, []>("op_33789_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33789_cast = slice_by_index(begin = var_33789_begin_0, end = var_33789_end_0, end_mask = var_33789_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33789_cast")];
            tensor<int32, [4]> var_33793_begin_0 = const()[name = tensor<string, []>("op_33793_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_33793_end_0 = const()[name = tensor<string, []>("op_33793_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_33793_end_mask_0 = const()[name = tensor<string, []>("op_33793_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33793_cast = slice_by_index(begin = var_33793_begin_0, end = var_33793_end_0, end_mask = var_33793_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33793_cast")];
            tensor<int32, [4]> var_33797_begin_0 = const()[name = tensor<string, []>("op_33797_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_33797_end_0 = const()[name = tensor<string, []>("op_33797_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_33797_end_mask_0 = const()[name = tensor<string, []>("op_33797_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33797_cast = slice_by_index(begin = var_33797_begin_0, end = var_33797_end_0, end_mask = var_33797_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33797_cast")];
            tensor<int32, [4]> var_33801_begin_0 = const()[name = tensor<string, []>("op_33801_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_33801_end_0 = const()[name = tensor<string, []>("op_33801_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_33801_end_mask_0 = const()[name = tensor<string, []>("op_33801_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33801_cast = slice_by_index(begin = var_33801_begin_0, end = var_33801_end_0, end_mask = var_33801_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33801_cast")];
            tensor<int32, [4]> var_33805_begin_0 = const()[name = tensor<string, []>("op_33805_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_33805_end_0 = const()[name = tensor<string, []>("op_33805_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_33805_end_mask_0 = const()[name = tensor<string, []>("op_33805_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33805_cast = slice_by_index(begin = var_33805_begin_0, end = var_33805_end_0, end_mask = var_33805_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33805_cast")];
            tensor<int32, [4]> var_33809_begin_0 = const()[name = tensor<string, []>("op_33809_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_33809_end_0 = const()[name = tensor<string, []>("op_33809_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_33809_end_mask_0 = const()[name = tensor<string, []>("op_33809_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33809_cast = slice_by_index(begin = var_33809_begin_0, end = var_33809_end_0, end_mask = var_33809_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33809_cast")];
            tensor<int32, [4]> var_33813_begin_0 = const()[name = tensor<string, []>("op_33813_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_33813_end_0 = const()[name = tensor<string, []>("op_33813_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_33813_end_mask_0 = const()[name = tensor<string, []>("op_33813_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33813_cast = slice_by_index(begin = var_33813_begin_0, end = var_33813_end_0, end_mask = var_33813_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33813_cast")];
            tensor<int32, [4]> var_33817_begin_0 = const()[name = tensor<string, []>("op_33817_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_33817_end_0 = const()[name = tensor<string, []>("op_33817_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_33817_end_mask_0 = const()[name = tensor<string, []>("op_33817_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33817_cast = slice_by_index(begin = var_33817_begin_0, end = var_33817_end_0, end_mask = var_33817_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33817_cast")];
            tensor<int32, [4]> var_33821_begin_0 = const()[name = tensor<string, []>("op_33821_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_33821_end_0 = const()[name = tensor<string, []>("op_33821_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_33821_end_mask_0 = const()[name = tensor<string, []>("op_33821_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33821_cast = slice_by_index(begin = var_33821_begin_0, end = var_33821_end_0, end_mask = var_33821_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33821_cast")];
            tensor<int32, [4]> var_33825_begin_0 = const()[name = tensor<string, []>("op_33825_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_33825_end_0 = const()[name = tensor<string, []>("op_33825_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_33825_end_mask_0 = const()[name = tensor<string, []>("op_33825_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33825_cast = slice_by_index(begin = var_33825_begin_0, end = var_33825_end_0, end_mask = var_33825_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33825_cast")];
            tensor<int32, [4]> var_33829_begin_0 = const()[name = tensor<string, []>("op_33829_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_33829_end_0 = const()[name = tensor<string, []>("op_33829_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_33829_end_mask_0 = const()[name = tensor<string, []>("op_33829_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33829_cast = slice_by_index(begin = var_33829_begin_0, end = var_33829_end_0, end_mask = var_33829_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33829_cast")];
            tensor<int32, [4]> var_33833_begin_0 = const()[name = tensor<string, []>("op_33833_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_33833_end_0 = const()[name = tensor<string, []>("op_33833_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_33833_end_mask_0 = const()[name = tensor<string, []>("op_33833_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33833_cast = slice_by_index(begin = var_33833_begin_0, end = var_33833_end_0, end_mask = var_33833_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33833_cast")];
            tensor<int32, [4]> var_33837_begin_0 = const()[name = tensor<string, []>("op_33837_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_33837_end_0 = const()[name = tensor<string, []>("op_33837_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_33837_end_mask_0 = const()[name = tensor<string, []>("op_33837_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33837_cast = slice_by_index(begin = var_33837_begin_0, end = var_33837_end_0, end_mask = var_33837_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33837_cast")];
            tensor<int32, [4]> var_33841_begin_0 = const()[name = tensor<string, []>("op_33841_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_33841_end_0 = const()[name = tensor<string, []>("op_33841_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_33841_end_mask_0 = const()[name = tensor<string, []>("op_33841_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33841_cast = slice_by_index(begin = var_33841_begin_0, end = var_33841_end_0, end_mask = var_33841_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33841_cast")];
            tensor<int32, [4]> var_33845_begin_0 = const()[name = tensor<string, []>("op_33845_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_33845_end_0 = const()[name = tensor<string, []>("op_33845_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_33845_end_mask_0 = const()[name = tensor<string, []>("op_33845_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33845_cast = slice_by_index(begin = var_33845_begin_0, end = var_33845_end_0, end_mask = var_33845_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33845_cast")];
            tensor<int32, [4]> var_33849_begin_0 = const()[name = tensor<string, []>("op_33849_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_33849_end_0 = const()[name = tensor<string, []>("op_33849_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_33849_end_mask_0 = const()[name = tensor<string, []>("op_33849_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33849_cast = slice_by_index(begin = var_33849_begin_0, end = var_33849_end_0, end_mask = var_33849_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33849_cast")];
            tensor<int32, [4]> var_33853_begin_0 = const()[name = tensor<string, []>("op_33853_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_33853_end_0 = const()[name = tensor<string, []>("op_33853_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_33853_end_mask_0 = const()[name = tensor<string, []>("op_33853_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33853_cast = slice_by_index(begin = var_33853_begin_0, end = var_33853_end_0, end_mask = var_33853_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33853_cast")];
            tensor<int32, [4]> var_33857_begin_0 = const()[name = tensor<string, []>("op_33857_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_33857_end_0 = const()[name = tensor<string, []>("op_33857_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_33857_end_mask_0 = const()[name = tensor<string, []>("op_33857_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33857_cast = slice_by_index(begin = var_33857_begin_0, end = var_33857_end_0, end_mask = var_33857_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33857_cast")];
            tensor<int32, [4]> var_33861_begin_0 = const()[name = tensor<string, []>("op_33861_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_33861_end_0 = const()[name = tensor<string, []>("op_33861_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_33861_end_mask_0 = const()[name = tensor<string, []>("op_33861_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_33861_cast = slice_by_index(begin = var_33861_begin_0, end = var_33861_end_0, end_mask = var_33861_end_mask_0, x = transpose_64)[name = tensor<string, []>("op_33861_cast")];
            tensor<int32, [4]> var_33863_begin_0 = const()[name = tensor<string, []>("op_33863_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_33863_end_0 = const()[name = tensor<string, []>("op_33863_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_33863_end_mask_0 = const()[name = tensor<string, []>("op_33863_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33863_cast = slice_by_index(begin = var_33863_begin_0, end = var_33863_end_0, end_mask = var_33863_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33863_cast")];
            tensor<int32, [4]> var_33867_begin_0 = const()[name = tensor<string, []>("op_33867_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_33867_end_0 = const()[name = tensor<string, []>("op_33867_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_33867_end_mask_0 = const()[name = tensor<string, []>("op_33867_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33867_cast = slice_by_index(begin = var_33867_begin_0, end = var_33867_end_0, end_mask = var_33867_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33867_cast")];
            tensor<int32, [4]> var_33871_begin_0 = const()[name = tensor<string, []>("op_33871_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_33871_end_0 = const()[name = tensor<string, []>("op_33871_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_33871_end_mask_0 = const()[name = tensor<string, []>("op_33871_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33871_cast = slice_by_index(begin = var_33871_begin_0, end = var_33871_end_0, end_mask = var_33871_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33871_cast")];
            tensor<int32, [4]> var_33875_begin_0 = const()[name = tensor<string, []>("op_33875_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_33875_end_0 = const()[name = tensor<string, []>("op_33875_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_33875_end_mask_0 = const()[name = tensor<string, []>("op_33875_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33875_cast = slice_by_index(begin = var_33875_begin_0, end = var_33875_end_0, end_mask = var_33875_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33875_cast")];
            tensor<int32, [4]> var_33879_begin_0 = const()[name = tensor<string, []>("op_33879_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_33879_end_0 = const()[name = tensor<string, []>("op_33879_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_33879_end_mask_0 = const()[name = tensor<string, []>("op_33879_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33879_cast = slice_by_index(begin = var_33879_begin_0, end = var_33879_end_0, end_mask = var_33879_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33879_cast")];
            tensor<int32, [4]> var_33883_begin_0 = const()[name = tensor<string, []>("op_33883_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_33883_end_0 = const()[name = tensor<string, []>("op_33883_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_33883_end_mask_0 = const()[name = tensor<string, []>("op_33883_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33883_cast = slice_by_index(begin = var_33883_begin_0, end = var_33883_end_0, end_mask = var_33883_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33883_cast")];
            tensor<int32, [4]> var_33887_begin_0 = const()[name = tensor<string, []>("op_33887_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_33887_end_0 = const()[name = tensor<string, []>("op_33887_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_33887_end_mask_0 = const()[name = tensor<string, []>("op_33887_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33887_cast = slice_by_index(begin = var_33887_begin_0, end = var_33887_end_0, end_mask = var_33887_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33887_cast")];
            tensor<int32, [4]> var_33891_begin_0 = const()[name = tensor<string, []>("op_33891_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_33891_end_0 = const()[name = tensor<string, []>("op_33891_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_33891_end_mask_0 = const()[name = tensor<string, []>("op_33891_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33891_cast = slice_by_index(begin = var_33891_begin_0, end = var_33891_end_0, end_mask = var_33891_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33891_cast")];
            tensor<int32, [4]> var_33895_begin_0 = const()[name = tensor<string, []>("op_33895_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_33895_end_0 = const()[name = tensor<string, []>("op_33895_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_33895_end_mask_0 = const()[name = tensor<string, []>("op_33895_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33895_cast = slice_by_index(begin = var_33895_begin_0, end = var_33895_end_0, end_mask = var_33895_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33895_cast")];
            tensor<int32, [4]> var_33899_begin_0 = const()[name = tensor<string, []>("op_33899_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_33899_end_0 = const()[name = tensor<string, []>("op_33899_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_33899_end_mask_0 = const()[name = tensor<string, []>("op_33899_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33899_cast = slice_by_index(begin = var_33899_begin_0, end = var_33899_end_0, end_mask = var_33899_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33899_cast")];
            tensor<int32, [4]> var_33903_begin_0 = const()[name = tensor<string, []>("op_33903_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_33903_end_0 = const()[name = tensor<string, []>("op_33903_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_33903_end_mask_0 = const()[name = tensor<string, []>("op_33903_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33903_cast = slice_by_index(begin = var_33903_begin_0, end = var_33903_end_0, end_mask = var_33903_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33903_cast")];
            tensor<int32, [4]> var_33907_begin_0 = const()[name = tensor<string, []>("op_33907_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_33907_end_0 = const()[name = tensor<string, []>("op_33907_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_33907_end_mask_0 = const()[name = tensor<string, []>("op_33907_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33907_cast = slice_by_index(begin = var_33907_begin_0, end = var_33907_end_0, end_mask = var_33907_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33907_cast")];
            tensor<int32, [4]> var_33911_begin_0 = const()[name = tensor<string, []>("op_33911_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_33911_end_0 = const()[name = tensor<string, []>("op_33911_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_33911_end_mask_0 = const()[name = tensor<string, []>("op_33911_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33911_cast = slice_by_index(begin = var_33911_begin_0, end = var_33911_end_0, end_mask = var_33911_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33911_cast")];
            tensor<int32, [4]> var_33915_begin_0 = const()[name = tensor<string, []>("op_33915_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_33915_end_0 = const()[name = tensor<string, []>("op_33915_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_33915_end_mask_0 = const()[name = tensor<string, []>("op_33915_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33915_cast = slice_by_index(begin = var_33915_begin_0, end = var_33915_end_0, end_mask = var_33915_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33915_cast")];
            tensor<int32, [4]> var_33919_begin_0 = const()[name = tensor<string, []>("op_33919_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_33919_end_0 = const()[name = tensor<string, []>("op_33919_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_33919_end_mask_0 = const()[name = tensor<string, []>("op_33919_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33919_cast = slice_by_index(begin = var_33919_begin_0, end = var_33919_end_0, end_mask = var_33919_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33919_cast")];
            tensor<int32, [4]> var_33923_begin_0 = const()[name = tensor<string, []>("op_33923_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_33923_end_0 = const()[name = tensor<string, []>("op_33923_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_33923_end_mask_0 = const()[name = tensor<string, []>("op_33923_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33923_cast = slice_by_index(begin = var_33923_begin_0, end = var_33923_end_0, end_mask = var_33923_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33923_cast")];
            tensor<int32, [4]> var_33927_begin_0 = const()[name = tensor<string, []>("op_33927_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_33927_end_0 = const()[name = tensor<string, []>("op_33927_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_33927_end_mask_0 = const()[name = tensor<string, []>("op_33927_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33927_cast = slice_by_index(begin = var_33927_begin_0, end = var_33927_end_0, end_mask = var_33927_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33927_cast")];
            tensor<int32, [4]> var_33931_begin_0 = const()[name = tensor<string, []>("op_33931_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_33931_end_0 = const()[name = tensor<string, []>("op_33931_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_33931_end_mask_0 = const()[name = tensor<string, []>("op_33931_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33931_cast = slice_by_index(begin = var_33931_begin_0, end = var_33931_end_0, end_mask = var_33931_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33931_cast")];
            tensor<int32, [4]> var_33935_begin_0 = const()[name = tensor<string, []>("op_33935_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_33935_end_0 = const()[name = tensor<string, []>("op_33935_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_33935_end_mask_0 = const()[name = tensor<string, []>("op_33935_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33935_cast = slice_by_index(begin = var_33935_begin_0, end = var_33935_end_0, end_mask = var_33935_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33935_cast")];
            tensor<int32, [4]> var_33939_begin_0 = const()[name = tensor<string, []>("op_33939_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_33939_end_0 = const()[name = tensor<string, []>("op_33939_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_33939_end_mask_0 = const()[name = tensor<string, []>("op_33939_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_33939_cast = slice_by_index(begin = var_33939_begin_0, end = var_33939_end_0, end_mask = var_33939_end_mask_0, x = v_151_cast)[name = tensor<string, []>("op_33939_cast")];
            tensor<string, []> var_33943_equation_0 = const()[name = tensor<string, []>("op_33943_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33943_cast = einsum(equation = var_33943_equation_0, values = (var_33785_cast, var_33702_cast))[name = tensor<string, []>("op_33943_cast")];
            tensor<fp16, []> var_33944_to_fp16 = const()[name = tensor<string, []>("op_33944_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2841_cast = mul(x = var_33943_cast, y = var_33944_to_fp16)[name = tensor<string, []>("aw_2841_cast")];
            tensor<string, []> var_33947_equation_0 = const()[name = tensor<string, []>("op_33947_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33947_cast = einsum(equation = var_33947_equation_0, values = (var_33789_cast, var_33706_cast))[name = tensor<string, []>("op_33947_cast")];
            tensor<fp16, []> var_33948_to_fp16 = const()[name = tensor<string, []>("op_33948_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2843_cast = mul(x = var_33947_cast, y = var_33948_to_fp16)[name = tensor<string, []>("aw_2843_cast")];
            tensor<string, []> var_33951_equation_0 = const()[name = tensor<string, []>("op_33951_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33951_cast = einsum(equation = var_33951_equation_0, values = (var_33793_cast, var_33710_cast))[name = tensor<string, []>("op_33951_cast")];
            tensor<fp16, []> var_33952_to_fp16 = const()[name = tensor<string, []>("op_33952_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2845_cast = mul(x = var_33951_cast, y = var_33952_to_fp16)[name = tensor<string, []>("aw_2845_cast")];
            tensor<string, []> var_33955_equation_0 = const()[name = tensor<string, []>("op_33955_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33955_cast = einsum(equation = var_33955_equation_0, values = (var_33797_cast, var_33714_cast))[name = tensor<string, []>("op_33955_cast")];
            tensor<fp16, []> var_33956_to_fp16 = const()[name = tensor<string, []>("op_33956_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2847_cast = mul(x = var_33955_cast, y = var_33956_to_fp16)[name = tensor<string, []>("aw_2847_cast")];
            tensor<string, []> var_33959_equation_0 = const()[name = tensor<string, []>("op_33959_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33959_cast = einsum(equation = var_33959_equation_0, values = (var_33801_cast, var_33718_cast))[name = tensor<string, []>("op_33959_cast")];
            tensor<fp16, []> var_33960_to_fp16 = const()[name = tensor<string, []>("op_33960_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2849_cast = mul(x = var_33959_cast, y = var_33960_to_fp16)[name = tensor<string, []>("aw_2849_cast")];
            tensor<string, []> var_33963_equation_0 = const()[name = tensor<string, []>("op_33963_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33963_cast = einsum(equation = var_33963_equation_0, values = (var_33805_cast, var_33722_cast))[name = tensor<string, []>("op_33963_cast")];
            tensor<fp16, []> var_33964_to_fp16 = const()[name = tensor<string, []>("op_33964_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2851_cast = mul(x = var_33963_cast, y = var_33964_to_fp16)[name = tensor<string, []>("aw_2851_cast")];
            tensor<string, []> var_33967_equation_0 = const()[name = tensor<string, []>("op_33967_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33967_cast = einsum(equation = var_33967_equation_0, values = (var_33809_cast, var_33726_cast))[name = tensor<string, []>("op_33967_cast")];
            tensor<fp16, []> var_33968_to_fp16 = const()[name = tensor<string, []>("op_33968_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2853_cast = mul(x = var_33967_cast, y = var_33968_to_fp16)[name = tensor<string, []>("aw_2853_cast")];
            tensor<string, []> var_33971_equation_0 = const()[name = tensor<string, []>("op_33971_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33971_cast = einsum(equation = var_33971_equation_0, values = (var_33813_cast, var_33730_cast))[name = tensor<string, []>("op_33971_cast")];
            tensor<fp16, []> var_33972_to_fp16 = const()[name = tensor<string, []>("op_33972_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2855_cast = mul(x = var_33971_cast, y = var_33972_to_fp16)[name = tensor<string, []>("aw_2855_cast")];
            tensor<string, []> var_33975_equation_0 = const()[name = tensor<string, []>("op_33975_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33975_cast = einsum(equation = var_33975_equation_0, values = (var_33817_cast, var_33734_cast))[name = tensor<string, []>("op_33975_cast")];
            tensor<fp16, []> var_33976_to_fp16 = const()[name = tensor<string, []>("op_33976_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2857_cast = mul(x = var_33975_cast, y = var_33976_to_fp16)[name = tensor<string, []>("aw_2857_cast")];
            tensor<string, []> var_33979_equation_0 = const()[name = tensor<string, []>("op_33979_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33979_cast = einsum(equation = var_33979_equation_0, values = (var_33821_cast, var_33738_cast))[name = tensor<string, []>("op_33979_cast")];
            tensor<fp16, []> var_33980_to_fp16 = const()[name = tensor<string, []>("op_33980_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2859_cast = mul(x = var_33979_cast, y = var_33980_to_fp16)[name = tensor<string, []>("aw_2859_cast")];
            tensor<string, []> var_33983_equation_0 = const()[name = tensor<string, []>("op_33983_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33983_cast = einsum(equation = var_33983_equation_0, values = (var_33825_cast, var_33742_cast))[name = tensor<string, []>("op_33983_cast")];
            tensor<fp16, []> var_33984_to_fp16 = const()[name = tensor<string, []>("op_33984_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2861_cast = mul(x = var_33983_cast, y = var_33984_to_fp16)[name = tensor<string, []>("aw_2861_cast")];
            tensor<string, []> var_33987_equation_0 = const()[name = tensor<string, []>("op_33987_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33987_cast = einsum(equation = var_33987_equation_0, values = (var_33829_cast, var_33746_cast))[name = tensor<string, []>("op_33987_cast")];
            tensor<fp16, []> var_33988_to_fp16 = const()[name = tensor<string, []>("op_33988_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2863_cast = mul(x = var_33987_cast, y = var_33988_to_fp16)[name = tensor<string, []>("aw_2863_cast")];
            tensor<string, []> var_33991_equation_0 = const()[name = tensor<string, []>("op_33991_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33991_cast = einsum(equation = var_33991_equation_0, values = (var_33833_cast, var_33750_cast))[name = tensor<string, []>("op_33991_cast")];
            tensor<fp16, []> var_33992_to_fp16 = const()[name = tensor<string, []>("op_33992_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2865_cast = mul(x = var_33991_cast, y = var_33992_to_fp16)[name = tensor<string, []>("aw_2865_cast")];
            tensor<string, []> var_33995_equation_0 = const()[name = tensor<string, []>("op_33995_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33995_cast = einsum(equation = var_33995_equation_0, values = (var_33837_cast, var_33754_cast))[name = tensor<string, []>("op_33995_cast")];
            tensor<fp16, []> var_33996_to_fp16 = const()[name = tensor<string, []>("op_33996_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2867_cast = mul(x = var_33995_cast, y = var_33996_to_fp16)[name = tensor<string, []>("aw_2867_cast")];
            tensor<string, []> var_33999_equation_0 = const()[name = tensor<string, []>("op_33999_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_33999_cast = einsum(equation = var_33999_equation_0, values = (var_33841_cast, var_33758_cast))[name = tensor<string, []>("op_33999_cast")];
            tensor<fp16, []> var_34000_to_fp16 = const()[name = tensor<string, []>("op_34000_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2869_cast = mul(x = var_33999_cast, y = var_34000_to_fp16)[name = tensor<string, []>("aw_2869_cast")];
            tensor<string, []> var_34003_equation_0 = const()[name = tensor<string, []>("op_34003_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34003_cast = einsum(equation = var_34003_equation_0, values = (var_33845_cast, var_33762_cast))[name = tensor<string, []>("op_34003_cast")];
            tensor<fp16, []> var_34004_to_fp16 = const()[name = tensor<string, []>("op_34004_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2871_cast = mul(x = var_34003_cast, y = var_34004_to_fp16)[name = tensor<string, []>("aw_2871_cast")];
            tensor<string, []> var_34007_equation_0 = const()[name = tensor<string, []>("op_34007_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34007_cast = einsum(equation = var_34007_equation_0, values = (var_33849_cast, var_33766_cast))[name = tensor<string, []>("op_34007_cast")];
            tensor<fp16, []> var_34008_to_fp16 = const()[name = tensor<string, []>("op_34008_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2873_cast = mul(x = var_34007_cast, y = var_34008_to_fp16)[name = tensor<string, []>("aw_2873_cast")];
            tensor<string, []> var_34011_equation_0 = const()[name = tensor<string, []>("op_34011_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34011_cast = einsum(equation = var_34011_equation_0, values = (var_33853_cast, var_33770_cast))[name = tensor<string, []>("op_34011_cast")];
            tensor<fp16, []> var_34012_to_fp16 = const()[name = tensor<string, []>("op_34012_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2875_cast = mul(x = var_34011_cast, y = var_34012_to_fp16)[name = tensor<string, []>("aw_2875_cast")];
            tensor<string, []> var_34015_equation_0 = const()[name = tensor<string, []>("op_34015_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34015_cast = einsum(equation = var_34015_equation_0, values = (var_33857_cast, var_33774_cast))[name = tensor<string, []>("op_34015_cast")];
            tensor<fp16, []> var_34016_to_fp16 = const()[name = tensor<string, []>("op_34016_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2877_cast = mul(x = var_34015_cast, y = var_34016_to_fp16)[name = tensor<string, []>("aw_2877_cast")];
            tensor<string, []> var_34019_equation_0 = const()[name = tensor<string, []>("op_34019_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34019_cast = einsum(equation = var_34019_equation_0, values = (var_33861_cast, var_33778_cast))[name = tensor<string, []>("op_34019_cast")];
            tensor<fp16, []> var_34020_to_fp16 = const()[name = tensor<string, []>("op_34020_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2879_cast = mul(x = var_34019_cast, y = var_34020_to_fp16)[name = tensor<string, []>("aw_2879_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34022_cast = softmax(axis = var_30385, x = aw_2841_cast)[name = tensor<string, []>("op_34022_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34023_cast = softmax(axis = var_30385, x = aw_2843_cast)[name = tensor<string, []>("op_34023_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34024_cast = softmax(axis = var_30385, x = aw_2845_cast)[name = tensor<string, []>("op_34024_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34025_cast = softmax(axis = var_30385, x = aw_2847_cast)[name = tensor<string, []>("op_34025_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34026_cast = softmax(axis = var_30385, x = aw_2849_cast)[name = tensor<string, []>("op_34026_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34027_cast = softmax(axis = var_30385, x = aw_2851_cast)[name = tensor<string, []>("op_34027_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34028_cast = softmax(axis = var_30385, x = aw_2853_cast)[name = tensor<string, []>("op_34028_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34029_cast = softmax(axis = var_30385, x = aw_2855_cast)[name = tensor<string, []>("op_34029_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34030_cast = softmax(axis = var_30385, x = aw_2857_cast)[name = tensor<string, []>("op_34030_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34031_cast = softmax(axis = var_30385, x = aw_2859_cast)[name = tensor<string, []>("op_34031_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34032_cast = softmax(axis = var_30385, x = aw_2861_cast)[name = tensor<string, []>("op_34032_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34033_cast = softmax(axis = var_30385, x = aw_2863_cast)[name = tensor<string, []>("op_34033_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34034_cast = softmax(axis = var_30385, x = aw_2865_cast)[name = tensor<string, []>("op_34034_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34035_cast = softmax(axis = var_30385, x = aw_2867_cast)[name = tensor<string, []>("op_34035_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34036_cast = softmax(axis = var_30385, x = aw_2869_cast)[name = tensor<string, []>("op_34036_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34037_cast = softmax(axis = var_30385, x = aw_2871_cast)[name = tensor<string, []>("op_34037_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34038_cast = softmax(axis = var_30385, x = aw_2873_cast)[name = tensor<string, []>("op_34038_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34039_cast = softmax(axis = var_30385, x = aw_2875_cast)[name = tensor<string, []>("op_34039_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34040_cast = softmax(axis = var_30385, x = aw_2877_cast)[name = tensor<string, []>("op_34040_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34041_cast = softmax(axis = var_30385, x = aw_2879_cast)[name = tensor<string, []>("op_34041_cast")];
            tensor<string, []> var_34043_equation_0 = const()[name = tensor<string, []>("op_34043_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34043_cast = einsum(equation = var_34043_equation_0, values = (var_33863_cast, var_34022_cast))[name = tensor<string, []>("op_34043_cast")];
            tensor<string, []> var_34045_equation_0 = const()[name = tensor<string, []>("op_34045_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34045_cast = einsum(equation = var_34045_equation_0, values = (var_33867_cast, var_34023_cast))[name = tensor<string, []>("op_34045_cast")];
            tensor<string, []> var_34047_equation_0 = const()[name = tensor<string, []>("op_34047_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34047_cast = einsum(equation = var_34047_equation_0, values = (var_33871_cast, var_34024_cast))[name = tensor<string, []>("op_34047_cast")];
            tensor<string, []> var_34049_equation_0 = const()[name = tensor<string, []>("op_34049_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34049_cast = einsum(equation = var_34049_equation_0, values = (var_33875_cast, var_34025_cast))[name = tensor<string, []>("op_34049_cast")];
            tensor<string, []> var_34051_equation_0 = const()[name = tensor<string, []>("op_34051_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34051_cast = einsum(equation = var_34051_equation_0, values = (var_33879_cast, var_34026_cast))[name = tensor<string, []>("op_34051_cast")];
            tensor<string, []> var_34053_equation_0 = const()[name = tensor<string, []>("op_34053_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34053_cast = einsum(equation = var_34053_equation_0, values = (var_33883_cast, var_34027_cast))[name = tensor<string, []>("op_34053_cast")];
            tensor<string, []> var_34055_equation_0 = const()[name = tensor<string, []>("op_34055_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34055_cast = einsum(equation = var_34055_equation_0, values = (var_33887_cast, var_34028_cast))[name = tensor<string, []>("op_34055_cast")];
            tensor<string, []> var_34057_equation_0 = const()[name = tensor<string, []>("op_34057_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34057_cast = einsum(equation = var_34057_equation_0, values = (var_33891_cast, var_34029_cast))[name = tensor<string, []>("op_34057_cast")];
            tensor<string, []> var_34059_equation_0 = const()[name = tensor<string, []>("op_34059_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34059_cast = einsum(equation = var_34059_equation_0, values = (var_33895_cast, var_34030_cast))[name = tensor<string, []>("op_34059_cast")];
            tensor<string, []> var_34061_equation_0 = const()[name = tensor<string, []>("op_34061_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34061_cast = einsum(equation = var_34061_equation_0, values = (var_33899_cast, var_34031_cast))[name = tensor<string, []>("op_34061_cast")];
            tensor<string, []> var_34063_equation_0 = const()[name = tensor<string, []>("op_34063_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34063_cast = einsum(equation = var_34063_equation_0, values = (var_33903_cast, var_34032_cast))[name = tensor<string, []>("op_34063_cast")];
            tensor<string, []> var_34065_equation_0 = const()[name = tensor<string, []>("op_34065_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34065_cast = einsum(equation = var_34065_equation_0, values = (var_33907_cast, var_34033_cast))[name = tensor<string, []>("op_34065_cast")];
            tensor<string, []> var_34067_equation_0 = const()[name = tensor<string, []>("op_34067_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34067_cast = einsum(equation = var_34067_equation_0, values = (var_33911_cast, var_34034_cast))[name = tensor<string, []>("op_34067_cast")];
            tensor<string, []> var_34069_equation_0 = const()[name = tensor<string, []>("op_34069_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34069_cast = einsum(equation = var_34069_equation_0, values = (var_33915_cast, var_34035_cast))[name = tensor<string, []>("op_34069_cast")];
            tensor<string, []> var_34071_equation_0 = const()[name = tensor<string, []>("op_34071_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34071_cast = einsum(equation = var_34071_equation_0, values = (var_33919_cast, var_34036_cast))[name = tensor<string, []>("op_34071_cast")];
            tensor<string, []> var_34073_equation_0 = const()[name = tensor<string, []>("op_34073_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34073_cast = einsum(equation = var_34073_equation_0, values = (var_33923_cast, var_34037_cast))[name = tensor<string, []>("op_34073_cast")];
            tensor<string, []> var_34075_equation_0 = const()[name = tensor<string, []>("op_34075_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34075_cast = einsum(equation = var_34075_equation_0, values = (var_33927_cast, var_34038_cast))[name = tensor<string, []>("op_34075_cast")];
            tensor<string, []> var_34077_equation_0 = const()[name = tensor<string, []>("op_34077_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34077_cast = einsum(equation = var_34077_equation_0, values = (var_33931_cast, var_34039_cast))[name = tensor<string, []>("op_34077_cast")];
            tensor<string, []> var_34079_equation_0 = const()[name = tensor<string, []>("op_34079_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34079_cast = einsum(equation = var_34079_equation_0, values = (var_33935_cast, var_34040_cast))[name = tensor<string, []>("op_34079_cast")];
            tensor<string, []> var_34081_equation_0 = const()[name = tensor<string, []>("op_34081_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34081_cast = einsum(equation = var_34081_equation_0, values = (var_33939_cast, var_34041_cast))[name = tensor<string, []>("op_34081_cast")];
            tensor<bool, []> input_467_interleave_0 = const()[name = tensor<string, []>("input_467_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_467_cast = concat(axis = var_30385, interleave = input_467_interleave_0, values = (var_34043_cast, var_34045_cast, var_34047_cast, var_34049_cast, var_34051_cast, var_34053_cast, var_34055_cast, var_34057_cast, var_34059_cast, var_34061_cast, var_34063_cast, var_34065_cast, var_34067_cast, var_34069_cast, var_34071_cast, var_34073_cast, var_34075_cast, var_34077_cast, var_34079_cast, var_34081_cast))[name = tensor<string, []>("input_467_cast")];
            tensor<int32, [2]> var_34087 = const()[name = tensor<string, []>("op_34087"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_34089 = const()[name = tensor<string, []>("op_34089"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_34091_pad_type_0 = const()[name = tensor<string, []>("op_34091_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_34091_pad_0 = const()[name = tensor<string, []>("op_34091_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2823843584)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2827120448)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_34091_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_out_0_bias_to_fp16, dilations = var_34089, groups = var_30385, pad = var_34091_pad_0, pad_type = var_34091_pad_type_0, strides = var_34087, weight = up_blocks_0_attentions_0_transformer_blocks_3_attn2_to_out_0_weight_to_fp16, x = input_467_cast)[name = tensor<string, []>("op_34091_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_227_cast = add(x = var_34091_cast, y = inputs_225_cast)[name = tensor<string, []>("inputs_227_cast")];
            tensor<int32, [1]> var_34095 = const()[name = tensor<string, []>("op_34095"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_227_cast = reduce_mean(axes = var_34095, keep_dims = var_30380, x = inputs_227_cast)[name = tensor<string, []>("channels_mean_227_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_227_cast = sub(x = inputs_227_cast, y = channels_mean_227_cast)[name = tensor<string, []>("zero_mean_227_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_227_cast = mul(x = zero_mean_227_cast, y = zero_mean_227_cast)[name = tensor<string, []>("zero_mean_sq_227_cast")];
            tensor<int32, [1]> var_34099 = const()[name = tensor<string, []>("op_34099"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_34100_cast = reduce_mean(axes = var_34099, keep_dims = var_30380, x = zero_mean_sq_227_cast)[name = tensor<string, []>("op_34100_cast")];
            tensor<fp16, []> var_34101_to_fp16 = const()[name = tensor<string, []>("op_34101_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_34102_cast = add(x = var_34100_cast, y = var_34101_to_fp16)[name = tensor<string, []>("op_34102_cast")];
            tensor<fp16, []> denom_227_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_227_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_227_cast = rsqrt(epsilon = denom_227_epsilon_0_to_fp16, x = var_34102_cast)[name = tensor<string, []>("denom_227_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_227_cast = mul(x = zero_mean_227_cast, y = denom_227_cast)[name = tensor<string, []>("out_227_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_34106_to_fp16 = const()[name = tensor<string, []>("op_34106_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2827123072)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_34107_cast = add(x = out_227_cast, y = var_34106_to_fp16)[name = tensor<string, []>("op_34107_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_34109_to_fp16 = const()[name = tensor<string, []>("op_34109_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2827125696)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_469_cast = mul(x = var_34107_cast, y = var_34109_to_fp16)[name = tensor<string, []>("input_469_cast")];
            tensor<int32, [2]> var_34117 = const()[name = tensor<string, []>("op_34117"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_34119 = const()[name = tensor<string, []>("op_34119"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_34121_pad_type_0 = const()[name = tensor<string, []>("op_34121_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_34121_pad_0 = const()[name = tensor<string, []>("op_34121_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2827128320)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2853342784)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_34121_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_3_ff_net_0_proj_bias_to_fp16, dilations = var_34119, groups = var_30385, pad = var_34121_pad_0, pad_type = var_34121_pad_type_0, strides = var_34117, weight = up_blocks_0_attentions_0_transformer_blocks_3_ff_net_0_proj_weight_to_fp16, x = input_469_cast)[name = tensor<string, []>("op_34121_cast")];
            tensor<int32, [2]> var_34122_split_sizes_0 = const()[name = tensor<string, []>("op_34122_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_34122_axis_0 = const()[name = tensor<string, []>("op_34122_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_34122_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_34122_cast_1 = split(axis = var_34122_axis_0, split_sizes = var_34122_split_sizes_0, x = var_34121_cast)[name = tensor<string, []>("op_34122_cast")];
            tensor<string, []> var_34124_mode_0 = const()[name = tensor<string, []>("op_34124_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_34124_cast = gelu(mode = var_34124_mode_0, x = var_34122_cast_1)[name = tensor<string, []>("op_34124_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_471_cast = mul(x = var_34122_cast_0, y = var_34124_cast)[name = tensor<string, []>("input_471_cast")];
            tensor<int32, [2]> var_34128 = const()[name = tensor<string, []>("op_34128"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_34130 = const()[name = tensor<string, []>("op_34130"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_34132_pad_type_0 = const()[name = tensor<string, []>("op_34132_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_34132_pad_0 = const()[name = tensor<string, []>("op_34132_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2853363328)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2866470592)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_34132_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_3_ff_net_2_bias_to_fp16, dilations = var_34130, groups = var_30385, pad = var_34132_pad_0, pad_type = var_34132_pad_type_0, strides = var_34128, weight = up_blocks_0_attentions_0_transformer_blocks_3_ff_net_2_weight_to_fp16, x = input_471_cast)[name = tensor<string, []>("op_34132_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_229_cast = add(x = var_34132_cast, y = inputs_227_cast)[name = tensor<string, []>("inputs_229_cast")];
            tensor<int32, [1]> var_34142 = const()[name = tensor<string, []>("op_34142"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_229_cast = reduce_mean(axes = var_34142, keep_dims = var_30380, x = inputs_229_cast)[name = tensor<string, []>("channels_mean_229_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_229_cast = sub(x = inputs_229_cast, y = channels_mean_229_cast)[name = tensor<string, []>("zero_mean_229_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_229_cast = mul(x = zero_mean_229_cast, y = zero_mean_229_cast)[name = tensor<string, []>("zero_mean_sq_229_cast")];
            tensor<int32, [1]> var_34146 = const()[name = tensor<string, []>("op_34146"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_34147_cast = reduce_mean(axes = var_34146, keep_dims = var_30380, x = zero_mean_sq_229_cast)[name = tensor<string, []>("op_34147_cast")];
            tensor<fp16, []> var_34148_to_fp16 = const()[name = tensor<string, []>("op_34148_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_34149_cast = add(x = var_34147_cast, y = var_34148_to_fp16)[name = tensor<string, []>("op_34149_cast")];
            tensor<fp16, []> denom_229_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_229_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_229_cast = rsqrt(epsilon = denom_229_epsilon_0_to_fp16, x = var_34149_cast)[name = tensor<string, []>("denom_229_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_229_cast = mul(x = zero_mean_229_cast, y = denom_229_cast)[name = tensor<string, []>("out_229_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_34153_to_fp16 = const()[name = tensor<string, []>("op_34153_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2866473216)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_34154_cast = add(x = out_229_cast, y = var_34153_to_fp16)[name = tensor<string, []>("op_34154_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_34156_to_fp16 = const()[name = tensor<string, []>("op_34156_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2866475840)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_311_cast = mul(x = var_34154_cast, y = var_34156_to_fp16)[name = tensor<string, []>("hidden_states_311_cast")];
            tensor<int32, [2]> var_34163 = const()[name = tensor<string, []>("op_34163"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_34165 = const()[name = tensor<string, []>("op_34165"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_153_pad_type_0 = const()[name = tensor<string, []>("q_153_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_153_pad_0 = const()[name = tensor<string, []>("q_153_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2866478464)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_153_cast = conv(dilations = var_34165, groups = var_30385, pad = q_153_pad_0, pad_type = q_153_pad_type_0, strides = var_34163, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_q_weight_to_fp16, x = hidden_states_311_cast)[name = tensor<string, []>("q_153_cast")];
            tensor<int32, [2]> var_34169 = const()[name = tensor<string, []>("op_34169"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_34171 = const()[name = tensor<string, []>("op_34171"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_305_pad_type_0 = const()[name = tensor<string, []>("k_305_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_305_pad_0 = const()[name = tensor<string, []>("k_305_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2869755328)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_305_cast = conv(dilations = var_34171, groups = var_30385, pad = k_305_pad_0, pad_type = k_305_pad_type_0, strides = var_34169, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_k_weight_to_fp16, x = hidden_states_311_cast)[name = tensor<string, []>("k_305_cast")];
            tensor<int32, [2]> var_34175 = const()[name = tensor<string, []>("op_34175"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_34177 = const()[name = tensor<string, []>("op_34177"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_153_pad_type_0 = const()[name = tensor<string, []>("v_153_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_153_pad_0 = const()[name = tensor<string, []>("v_153_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2873032192)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_153_cast = conv(dilations = var_34177, groups = var_30385, pad = v_153_pad_0, pad_type = v_153_pad_type_0, strides = var_34175, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_v_weight_to_fp16, x = hidden_states_311_cast)[name = tensor<string, []>("v_153_cast")];
            tensor<int32, [4]> var_34181_begin_0 = const()[name = tensor<string, []>("op_34181_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_34181_end_0 = const()[name = tensor<string, []>("op_34181_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_34181_end_mask_0 = const()[name = tensor<string, []>("op_34181_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34181_cast = slice_by_index(begin = var_34181_begin_0, end = var_34181_end_0, end_mask = var_34181_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34181_cast")];
            tensor<int32, [4]> var_34185_begin_0 = const()[name = tensor<string, []>("op_34185_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_34185_end_0 = const()[name = tensor<string, []>("op_34185_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_34185_end_mask_0 = const()[name = tensor<string, []>("op_34185_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34185_cast = slice_by_index(begin = var_34185_begin_0, end = var_34185_end_0, end_mask = var_34185_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34185_cast")];
            tensor<int32, [4]> var_34189_begin_0 = const()[name = tensor<string, []>("op_34189_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_34189_end_0 = const()[name = tensor<string, []>("op_34189_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_34189_end_mask_0 = const()[name = tensor<string, []>("op_34189_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34189_cast = slice_by_index(begin = var_34189_begin_0, end = var_34189_end_0, end_mask = var_34189_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34189_cast")];
            tensor<int32, [4]> var_34193_begin_0 = const()[name = tensor<string, []>("op_34193_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_34193_end_0 = const()[name = tensor<string, []>("op_34193_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_34193_end_mask_0 = const()[name = tensor<string, []>("op_34193_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34193_cast = slice_by_index(begin = var_34193_begin_0, end = var_34193_end_0, end_mask = var_34193_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34193_cast")];
            tensor<int32, [4]> var_34197_begin_0 = const()[name = tensor<string, []>("op_34197_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_34197_end_0 = const()[name = tensor<string, []>("op_34197_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_34197_end_mask_0 = const()[name = tensor<string, []>("op_34197_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34197_cast = slice_by_index(begin = var_34197_begin_0, end = var_34197_end_0, end_mask = var_34197_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34197_cast")];
            tensor<int32, [4]> var_34201_begin_0 = const()[name = tensor<string, []>("op_34201_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_34201_end_0 = const()[name = tensor<string, []>("op_34201_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_34201_end_mask_0 = const()[name = tensor<string, []>("op_34201_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34201_cast = slice_by_index(begin = var_34201_begin_0, end = var_34201_end_0, end_mask = var_34201_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34201_cast")];
            tensor<int32, [4]> var_34205_begin_0 = const()[name = tensor<string, []>("op_34205_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_34205_end_0 = const()[name = tensor<string, []>("op_34205_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_34205_end_mask_0 = const()[name = tensor<string, []>("op_34205_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34205_cast = slice_by_index(begin = var_34205_begin_0, end = var_34205_end_0, end_mask = var_34205_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34205_cast")];
            tensor<int32, [4]> var_34209_begin_0 = const()[name = tensor<string, []>("op_34209_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_34209_end_0 = const()[name = tensor<string, []>("op_34209_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_34209_end_mask_0 = const()[name = tensor<string, []>("op_34209_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34209_cast = slice_by_index(begin = var_34209_begin_0, end = var_34209_end_0, end_mask = var_34209_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34209_cast")];
            tensor<int32, [4]> var_34213_begin_0 = const()[name = tensor<string, []>("op_34213_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_34213_end_0 = const()[name = tensor<string, []>("op_34213_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_34213_end_mask_0 = const()[name = tensor<string, []>("op_34213_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34213_cast = slice_by_index(begin = var_34213_begin_0, end = var_34213_end_0, end_mask = var_34213_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34213_cast")];
            tensor<int32, [4]> var_34217_begin_0 = const()[name = tensor<string, []>("op_34217_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_34217_end_0 = const()[name = tensor<string, []>("op_34217_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_34217_end_mask_0 = const()[name = tensor<string, []>("op_34217_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34217_cast = slice_by_index(begin = var_34217_begin_0, end = var_34217_end_0, end_mask = var_34217_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34217_cast")];
            tensor<int32, [4]> var_34221_begin_0 = const()[name = tensor<string, []>("op_34221_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_34221_end_0 = const()[name = tensor<string, []>("op_34221_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_34221_end_mask_0 = const()[name = tensor<string, []>("op_34221_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34221_cast = slice_by_index(begin = var_34221_begin_0, end = var_34221_end_0, end_mask = var_34221_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34221_cast")];
            tensor<int32, [4]> var_34225_begin_0 = const()[name = tensor<string, []>("op_34225_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_34225_end_0 = const()[name = tensor<string, []>("op_34225_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_34225_end_mask_0 = const()[name = tensor<string, []>("op_34225_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34225_cast = slice_by_index(begin = var_34225_begin_0, end = var_34225_end_0, end_mask = var_34225_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34225_cast")];
            tensor<int32, [4]> var_34229_begin_0 = const()[name = tensor<string, []>("op_34229_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_34229_end_0 = const()[name = tensor<string, []>("op_34229_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_34229_end_mask_0 = const()[name = tensor<string, []>("op_34229_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34229_cast = slice_by_index(begin = var_34229_begin_0, end = var_34229_end_0, end_mask = var_34229_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34229_cast")];
            tensor<int32, [4]> var_34233_begin_0 = const()[name = tensor<string, []>("op_34233_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_34233_end_0 = const()[name = tensor<string, []>("op_34233_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_34233_end_mask_0 = const()[name = tensor<string, []>("op_34233_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34233_cast = slice_by_index(begin = var_34233_begin_0, end = var_34233_end_0, end_mask = var_34233_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34233_cast")];
            tensor<int32, [4]> var_34237_begin_0 = const()[name = tensor<string, []>("op_34237_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_34237_end_0 = const()[name = tensor<string, []>("op_34237_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_34237_end_mask_0 = const()[name = tensor<string, []>("op_34237_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34237_cast = slice_by_index(begin = var_34237_begin_0, end = var_34237_end_0, end_mask = var_34237_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34237_cast")];
            tensor<int32, [4]> var_34241_begin_0 = const()[name = tensor<string, []>("op_34241_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_34241_end_0 = const()[name = tensor<string, []>("op_34241_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_34241_end_mask_0 = const()[name = tensor<string, []>("op_34241_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34241_cast = slice_by_index(begin = var_34241_begin_0, end = var_34241_end_0, end_mask = var_34241_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34241_cast")];
            tensor<int32, [4]> var_34245_begin_0 = const()[name = tensor<string, []>("op_34245_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_34245_end_0 = const()[name = tensor<string, []>("op_34245_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_34245_end_mask_0 = const()[name = tensor<string, []>("op_34245_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34245_cast = slice_by_index(begin = var_34245_begin_0, end = var_34245_end_0, end_mask = var_34245_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34245_cast")];
            tensor<int32, [4]> var_34249_begin_0 = const()[name = tensor<string, []>("op_34249_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_34249_end_0 = const()[name = tensor<string, []>("op_34249_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_34249_end_mask_0 = const()[name = tensor<string, []>("op_34249_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34249_cast = slice_by_index(begin = var_34249_begin_0, end = var_34249_end_0, end_mask = var_34249_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34249_cast")];
            tensor<int32, [4]> var_34253_begin_0 = const()[name = tensor<string, []>("op_34253_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_34253_end_0 = const()[name = tensor<string, []>("op_34253_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_34253_end_mask_0 = const()[name = tensor<string, []>("op_34253_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34253_cast = slice_by_index(begin = var_34253_begin_0, end = var_34253_end_0, end_mask = var_34253_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34253_cast")];
            tensor<int32, [4]> var_34257_begin_0 = const()[name = tensor<string, []>("op_34257_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_34257_end_0 = const()[name = tensor<string, []>("op_34257_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_34257_end_mask_0 = const()[name = tensor<string, []>("op_34257_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34257_cast = slice_by_index(begin = var_34257_begin_0, end = var_34257_end_0, end_mask = var_34257_end_mask_0, x = q_153_cast)[name = tensor<string, []>("op_34257_cast")];
            tensor<int32, [4]> k_307_perm_0 = const()[name = tensor<string, []>("k_307_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_34264_begin_0 = const()[name = tensor<string, []>("op_34264_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_34264_end_0 = const()[name = tensor<string, []>("op_34264_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_34264_end_mask_0 = const()[name = tensor<string, []>("op_34264_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_63 = transpose(perm = k_307_perm_0, x = k_305_cast)[name = tensor<string, []>("transpose_63")];
            tensor<fp16, [2, 1024, 1, 64]> var_34264_cast = slice_by_index(begin = var_34264_begin_0, end = var_34264_end_0, end_mask = var_34264_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34264_cast")];
            tensor<int32, [4]> var_34268_begin_0 = const()[name = tensor<string, []>("op_34268_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_34268_end_0 = const()[name = tensor<string, []>("op_34268_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_34268_end_mask_0 = const()[name = tensor<string, []>("op_34268_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34268_cast = slice_by_index(begin = var_34268_begin_0, end = var_34268_end_0, end_mask = var_34268_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34268_cast")];
            tensor<int32, [4]> var_34272_begin_0 = const()[name = tensor<string, []>("op_34272_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_34272_end_0 = const()[name = tensor<string, []>("op_34272_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_34272_end_mask_0 = const()[name = tensor<string, []>("op_34272_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34272_cast = slice_by_index(begin = var_34272_begin_0, end = var_34272_end_0, end_mask = var_34272_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34272_cast")];
            tensor<int32, [4]> var_34276_begin_0 = const()[name = tensor<string, []>("op_34276_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_34276_end_0 = const()[name = tensor<string, []>("op_34276_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_34276_end_mask_0 = const()[name = tensor<string, []>("op_34276_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34276_cast = slice_by_index(begin = var_34276_begin_0, end = var_34276_end_0, end_mask = var_34276_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34276_cast")];
            tensor<int32, [4]> var_34280_begin_0 = const()[name = tensor<string, []>("op_34280_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_34280_end_0 = const()[name = tensor<string, []>("op_34280_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_34280_end_mask_0 = const()[name = tensor<string, []>("op_34280_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34280_cast = slice_by_index(begin = var_34280_begin_0, end = var_34280_end_0, end_mask = var_34280_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34280_cast")];
            tensor<int32, [4]> var_34284_begin_0 = const()[name = tensor<string, []>("op_34284_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_34284_end_0 = const()[name = tensor<string, []>("op_34284_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_34284_end_mask_0 = const()[name = tensor<string, []>("op_34284_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34284_cast = slice_by_index(begin = var_34284_begin_0, end = var_34284_end_0, end_mask = var_34284_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34284_cast")];
            tensor<int32, [4]> var_34288_begin_0 = const()[name = tensor<string, []>("op_34288_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_34288_end_0 = const()[name = tensor<string, []>("op_34288_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_34288_end_mask_0 = const()[name = tensor<string, []>("op_34288_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34288_cast = slice_by_index(begin = var_34288_begin_0, end = var_34288_end_0, end_mask = var_34288_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34288_cast")];
            tensor<int32, [4]> var_34292_begin_0 = const()[name = tensor<string, []>("op_34292_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_34292_end_0 = const()[name = tensor<string, []>("op_34292_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_34292_end_mask_0 = const()[name = tensor<string, []>("op_34292_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34292_cast = slice_by_index(begin = var_34292_begin_0, end = var_34292_end_0, end_mask = var_34292_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34292_cast")];
            tensor<int32, [4]> var_34296_begin_0 = const()[name = tensor<string, []>("op_34296_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_34296_end_0 = const()[name = tensor<string, []>("op_34296_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_34296_end_mask_0 = const()[name = tensor<string, []>("op_34296_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34296_cast = slice_by_index(begin = var_34296_begin_0, end = var_34296_end_0, end_mask = var_34296_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34296_cast")];
            tensor<int32, [4]> var_34300_begin_0 = const()[name = tensor<string, []>("op_34300_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_34300_end_0 = const()[name = tensor<string, []>("op_34300_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_34300_end_mask_0 = const()[name = tensor<string, []>("op_34300_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34300_cast = slice_by_index(begin = var_34300_begin_0, end = var_34300_end_0, end_mask = var_34300_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34300_cast")];
            tensor<int32, [4]> var_34304_begin_0 = const()[name = tensor<string, []>("op_34304_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_34304_end_0 = const()[name = tensor<string, []>("op_34304_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_34304_end_mask_0 = const()[name = tensor<string, []>("op_34304_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34304_cast = slice_by_index(begin = var_34304_begin_0, end = var_34304_end_0, end_mask = var_34304_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34304_cast")];
            tensor<int32, [4]> var_34308_begin_0 = const()[name = tensor<string, []>("op_34308_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_34308_end_0 = const()[name = tensor<string, []>("op_34308_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_34308_end_mask_0 = const()[name = tensor<string, []>("op_34308_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34308_cast = slice_by_index(begin = var_34308_begin_0, end = var_34308_end_0, end_mask = var_34308_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34308_cast")];
            tensor<int32, [4]> var_34312_begin_0 = const()[name = tensor<string, []>("op_34312_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_34312_end_0 = const()[name = tensor<string, []>("op_34312_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_34312_end_mask_0 = const()[name = tensor<string, []>("op_34312_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34312_cast = slice_by_index(begin = var_34312_begin_0, end = var_34312_end_0, end_mask = var_34312_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34312_cast")];
            tensor<int32, [4]> var_34316_begin_0 = const()[name = tensor<string, []>("op_34316_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_34316_end_0 = const()[name = tensor<string, []>("op_34316_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_34316_end_mask_0 = const()[name = tensor<string, []>("op_34316_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34316_cast = slice_by_index(begin = var_34316_begin_0, end = var_34316_end_0, end_mask = var_34316_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34316_cast")];
            tensor<int32, [4]> var_34320_begin_0 = const()[name = tensor<string, []>("op_34320_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_34320_end_0 = const()[name = tensor<string, []>("op_34320_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_34320_end_mask_0 = const()[name = tensor<string, []>("op_34320_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34320_cast = slice_by_index(begin = var_34320_begin_0, end = var_34320_end_0, end_mask = var_34320_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34320_cast")];
            tensor<int32, [4]> var_34324_begin_0 = const()[name = tensor<string, []>("op_34324_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_34324_end_0 = const()[name = tensor<string, []>("op_34324_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_34324_end_mask_0 = const()[name = tensor<string, []>("op_34324_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34324_cast = slice_by_index(begin = var_34324_begin_0, end = var_34324_end_0, end_mask = var_34324_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34324_cast")];
            tensor<int32, [4]> var_34328_begin_0 = const()[name = tensor<string, []>("op_34328_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_34328_end_0 = const()[name = tensor<string, []>("op_34328_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_34328_end_mask_0 = const()[name = tensor<string, []>("op_34328_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34328_cast = slice_by_index(begin = var_34328_begin_0, end = var_34328_end_0, end_mask = var_34328_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34328_cast")];
            tensor<int32, [4]> var_34332_begin_0 = const()[name = tensor<string, []>("op_34332_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_34332_end_0 = const()[name = tensor<string, []>("op_34332_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_34332_end_mask_0 = const()[name = tensor<string, []>("op_34332_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34332_cast = slice_by_index(begin = var_34332_begin_0, end = var_34332_end_0, end_mask = var_34332_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34332_cast")];
            tensor<int32, [4]> var_34336_begin_0 = const()[name = tensor<string, []>("op_34336_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_34336_end_0 = const()[name = tensor<string, []>("op_34336_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_34336_end_mask_0 = const()[name = tensor<string, []>("op_34336_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34336_cast = slice_by_index(begin = var_34336_begin_0, end = var_34336_end_0, end_mask = var_34336_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34336_cast")];
            tensor<int32, [4]> var_34340_begin_0 = const()[name = tensor<string, []>("op_34340_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_34340_end_0 = const()[name = tensor<string, []>("op_34340_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_34340_end_mask_0 = const()[name = tensor<string, []>("op_34340_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_34340_cast = slice_by_index(begin = var_34340_begin_0, end = var_34340_end_0, end_mask = var_34340_end_mask_0, x = transpose_63)[name = tensor<string, []>("op_34340_cast")];
            tensor<int32, [4]> var_34342_begin_0 = const()[name = tensor<string, []>("op_34342_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_34342_end_0 = const()[name = tensor<string, []>("op_34342_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_34342_end_mask_0 = const()[name = tensor<string, []>("op_34342_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34342_cast = slice_by_index(begin = var_34342_begin_0, end = var_34342_end_0, end_mask = var_34342_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34342_cast")];
            tensor<int32, [4]> var_34346_begin_0 = const()[name = tensor<string, []>("op_34346_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_34346_end_0 = const()[name = tensor<string, []>("op_34346_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_34346_end_mask_0 = const()[name = tensor<string, []>("op_34346_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34346_cast = slice_by_index(begin = var_34346_begin_0, end = var_34346_end_0, end_mask = var_34346_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34346_cast")];
            tensor<int32, [4]> var_34350_begin_0 = const()[name = tensor<string, []>("op_34350_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_34350_end_0 = const()[name = tensor<string, []>("op_34350_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_34350_end_mask_0 = const()[name = tensor<string, []>("op_34350_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34350_cast = slice_by_index(begin = var_34350_begin_0, end = var_34350_end_0, end_mask = var_34350_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34350_cast")];
            tensor<int32, [4]> var_34354_begin_0 = const()[name = tensor<string, []>("op_34354_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_34354_end_0 = const()[name = tensor<string, []>("op_34354_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_34354_end_mask_0 = const()[name = tensor<string, []>("op_34354_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34354_cast = slice_by_index(begin = var_34354_begin_0, end = var_34354_end_0, end_mask = var_34354_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34354_cast")];
            tensor<int32, [4]> var_34358_begin_0 = const()[name = tensor<string, []>("op_34358_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_34358_end_0 = const()[name = tensor<string, []>("op_34358_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_34358_end_mask_0 = const()[name = tensor<string, []>("op_34358_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34358_cast = slice_by_index(begin = var_34358_begin_0, end = var_34358_end_0, end_mask = var_34358_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34358_cast")];
            tensor<int32, [4]> var_34362_begin_0 = const()[name = tensor<string, []>("op_34362_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_34362_end_0 = const()[name = tensor<string, []>("op_34362_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_34362_end_mask_0 = const()[name = tensor<string, []>("op_34362_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34362_cast = slice_by_index(begin = var_34362_begin_0, end = var_34362_end_0, end_mask = var_34362_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34362_cast")];
            tensor<int32, [4]> var_34366_begin_0 = const()[name = tensor<string, []>("op_34366_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_34366_end_0 = const()[name = tensor<string, []>("op_34366_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_34366_end_mask_0 = const()[name = tensor<string, []>("op_34366_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34366_cast = slice_by_index(begin = var_34366_begin_0, end = var_34366_end_0, end_mask = var_34366_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34366_cast")];
            tensor<int32, [4]> var_34370_begin_0 = const()[name = tensor<string, []>("op_34370_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_34370_end_0 = const()[name = tensor<string, []>("op_34370_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_34370_end_mask_0 = const()[name = tensor<string, []>("op_34370_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34370_cast = slice_by_index(begin = var_34370_begin_0, end = var_34370_end_0, end_mask = var_34370_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34370_cast")];
            tensor<int32, [4]> var_34374_begin_0 = const()[name = tensor<string, []>("op_34374_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_34374_end_0 = const()[name = tensor<string, []>("op_34374_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_34374_end_mask_0 = const()[name = tensor<string, []>("op_34374_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34374_cast = slice_by_index(begin = var_34374_begin_0, end = var_34374_end_0, end_mask = var_34374_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34374_cast")];
            tensor<int32, [4]> var_34378_begin_0 = const()[name = tensor<string, []>("op_34378_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_34378_end_0 = const()[name = tensor<string, []>("op_34378_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_34378_end_mask_0 = const()[name = tensor<string, []>("op_34378_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34378_cast = slice_by_index(begin = var_34378_begin_0, end = var_34378_end_0, end_mask = var_34378_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34378_cast")];
            tensor<int32, [4]> var_34382_begin_0 = const()[name = tensor<string, []>("op_34382_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_34382_end_0 = const()[name = tensor<string, []>("op_34382_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_34382_end_mask_0 = const()[name = tensor<string, []>("op_34382_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34382_cast = slice_by_index(begin = var_34382_begin_0, end = var_34382_end_0, end_mask = var_34382_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34382_cast")];
            tensor<int32, [4]> var_34386_begin_0 = const()[name = tensor<string, []>("op_34386_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_34386_end_0 = const()[name = tensor<string, []>("op_34386_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_34386_end_mask_0 = const()[name = tensor<string, []>("op_34386_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34386_cast = slice_by_index(begin = var_34386_begin_0, end = var_34386_end_0, end_mask = var_34386_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34386_cast")];
            tensor<int32, [4]> var_34390_begin_0 = const()[name = tensor<string, []>("op_34390_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_34390_end_0 = const()[name = tensor<string, []>("op_34390_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_34390_end_mask_0 = const()[name = tensor<string, []>("op_34390_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34390_cast = slice_by_index(begin = var_34390_begin_0, end = var_34390_end_0, end_mask = var_34390_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34390_cast")];
            tensor<int32, [4]> var_34394_begin_0 = const()[name = tensor<string, []>("op_34394_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_34394_end_0 = const()[name = tensor<string, []>("op_34394_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_34394_end_mask_0 = const()[name = tensor<string, []>("op_34394_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34394_cast = slice_by_index(begin = var_34394_begin_0, end = var_34394_end_0, end_mask = var_34394_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34394_cast")];
            tensor<int32, [4]> var_34398_begin_0 = const()[name = tensor<string, []>("op_34398_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_34398_end_0 = const()[name = tensor<string, []>("op_34398_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_34398_end_mask_0 = const()[name = tensor<string, []>("op_34398_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34398_cast = slice_by_index(begin = var_34398_begin_0, end = var_34398_end_0, end_mask = var_34398_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34398_cast")];
            tensor<int32, [4]> var_34402_begin_0 = const()[name = tensor<string, []>("op_34402_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_34402_end_0 = const()[name = tensor<string, []>("op_34402_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_34402_end_mask_0 = const()[name = tensor<string, []>("op_34402_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34402_cast = slice_by_index(begin = var_34402_begin_0, end = var_34402_end_0, end_mask = var_34402_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34402_cast")];
            tensor<int32, [4]> var_34406_begin_0 = const()[name = tensor<string, []>("op_34406_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_34406_end_0 = const()[name = tensor<string, []>("op_34406_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_34406_end_mask_0 = const()[name = tensor<string, []>("op_34406_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34406_cast = slice_by_index(begin = var_34406_begin_0, end = var_34406_end_0, end_mask = var_34406_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34406_cast")];
            tensor<int32, [4]> var_34410_begin_0 = const()[name = tensor<string, []>("op_34410_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_34410_end_0 = const()[name = tensor<string, []>("op_34410_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_34410_end_mask_0 = const()[name = tensor<string, []>("op_34410_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34410_cast = slice_by_index(begin = var_34410_begin_0, end = var_34410_end_0, end_mask = var_34410_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34410_cast")];
            tensor<int32, [4]> var_34414_begin_0 = const()[name = tensor<string, []>("op_34414_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_34414_end_0 = const()[name = tensor<string, []>("op_34414_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_34414_end_mask_0 = const()[name = tensor<string, []>("op_34414_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34414_cast = slice_by_index(begin = var_34414_begin_0, end = var_34414_end_0, end_mask = var_34414_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34414_cast")];
            tensor<int32, [4]> var_34418_begin_0 = const()[name = tensor<string, []>("op_34418_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_34418_end_0 = const()[name = tensor<string, []>("op_34418_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_34418_end_mask_0 = const()[name = tensor<string, []>("op_34418_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34418_cast = slice_by_index(begin = var_34418_begin_0, end = var_34418_end_0, end_mask = var_34418_end_mask_0, x = v_153_cast)[name = tensor<string, []>("op_34418_cast")];
            tensor<string, []> var_34422_equation_0 = const()[name = tensor<string, []>("op_34422_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34422_cast = einsum(equation = var_34422_equation_0, values = (var_34264_cast, var_34181_cast))[name = tensor<string, []>("op_34422_cast")];
            tensor<fp16, []> var_34423_to_fp16 = const()[name = tensor<string, []>("op_34423_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2881_cast = mul(x = var_34422_cast, y = var_34423_to_fp16)[name = tensor<string, []>("aw_2881_cast")];
            tensor<string, []> var_34426_equation_0 = const()[name = tensor<string, []>("op_34426_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34426_cast = einsum(equation = var_34426_equation_0, values = (var_34268_cast, var_34185_cast))[name = tensor<string, []>("op_34426_cast")];
            tensor<fp16, []> var_34427_to_fp16 = const()[name = tensor<string, []>("op_34427_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2883_cast = mul(x = var_34426_cast, y = var_34427_to_fp16)[name = tensor<string, []>("aw_2883_cast")];
            tensor<string, []> var_34430_equation_0 = const()[name = tensor<string, []>("op_34430_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34430_cast = einsum(equation = var_34430_equation_0, values = (var_34272_cast, var_34189_cast))[name = tensor<string, []>("op_34430_cast")];
            tensor<fp16, []> var_34431_to_fp16 = const()[name = tensor<string, []>("op_34431_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2885_cast = mul(x = var_34430_cast, y = var_34431_to_fp16)[name = tensor<string, []>("aw_2885_cast")];
            tensor<string, []> var_34434_equation_0 = const()[name = tensor<string, []>("op_34434_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34434_cast = einsum(equation = var_34434_equation_0, values = (var_34276_cast, var_34193_cast))[name = tensor<string, []>("op_34434_cast")];
            tensor<fp16, []> var_34435_to_fp16 = const()[name = tensor<string, []>("op_34435_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2887_cast = mul(x = var_34434_cast, y = var_34435_to_fp16)[name = tensor<string, []>("aw_2887_cast")];
            tensor<string, []> var_34438_equation_0 = const()[name = tensor<string, []>("op_34438_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34438_cast = einsum(equation = var_34438_equation_0, values = (var_34280_cast, var_34197_cast))[name = tensor<string, []>("op_34438_cast")];
            tensor<fp16, []> var_34439_to_fp16 = const()[name = tensor<string, []>("op_34439_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2889_cast = mul(x = var_34438_cast, y = var_34439_to_fp16)[name = tensor<string, []>("aw_2889_cast")];
            tensor<string, []> var_34442_equation_0 = const()[name = tensor<string, []>("op_34442_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34442_cast = einsum(equation = var_34442_equation_0, values = (var_34284_cast, var_34201_cast))[name = tensor<string, []>("op_34442_cast")];
            tensor<fp16, []> var_34443_to_fp16 = const()[name = tensor<string, []>("op_34443_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2891_cast = mul(x = var_34442_cast, y = var_34443_to_fp16)[name = tensor<string, []>("aw_2891_cast")];
            tensor<string, []> var_34446_equation_0 = const()[name = tensor<string, []>("op_34446_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34446_cast = einsum(equation = var_34446_equation_0, values = (var_34288_cast, var_34205_cast))[name = tensor<string, []>("op_34446_cast")];
            tensor<fp16, []> var_34447_to_fp16 = const()[name = tensor<string, []>("op_34447_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2893_cast = mul(x = var_34446_cast, y = var_34447_to_fp16)[name = tensor<string, []>("aw_2893_cast")];
            tensor<string, []> var_34450_equation_0 = const()[name = tensor<string, []>("op_34450_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34450_cast = einsum(equation = var_34450_equation_0, values = (var_34292_cast, var_34209_cast))[name = tensor<string, []>("op_34450_cast")];
            tensor<fp16, []> var_34451_to_fp16 = const()[name = tensor<string, []>("op_34451_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2895_cast = mul(x = var_34450_cast, y = var_34451_to_fp16)[name = tensor<string, []>("aw_2895_cast")];
            tensor<string, []> var_34454_equation_0 = const()[name = tensor<string, []>("op_34454_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34454_cast = einsum(equation = var_34454_equation_0, values = (var_34296_cast, var_34213_cast))[name = tensor<string, []>("op_34454_cast")];
            tensor<fp16, []> var_34455_to_fp16 = const()[name = tensor<string, []>("op_34455_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2897_cast = mul(x = var_34454_cast, y = var_34455_to_fp16)[name = tensor<string, []>("aw_2897_cast")];
            tensor<string, []> var_34458_equation_0 = const()[name = tensor<string, []>("op_34458_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34458_cast = einsum(equation = var_34458_equation_0, values = (var_34300_cast, var_34217_cast))[name = tensor<string, []>("op_34458_cast")];
            tensor<fp16, []> var_34459_to_fp16 = const()[name = tensor<string, []>("op_34459_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2899_cast = mul(x = var_34458_cast, y = var_34459_to_fp16)[name = tensor<string, []>("aw_2899_cast")];
            tensor<string, []> var_34462_equation_0 = const()[name = tensor<string, []>("op_34462_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34462_cast = einsum(equation = var_34462_equation_0, values = (var_34304_cast, var_34221_cast))[name = tensor<string, []>("op_34462_cast")];
            tensor<fp16, []> var_34463_to_fp16 = const()[name = tensor<string, []>("op_34463_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2901_cast = mul(x = var_34462_cast, y = var_34463_to_fp16)[name = tensor<string, []>("aw_2901_cast")];
            tensor<string, []> var_34466_equation_0 = const()[name = tensor<string, []>("op_34466_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34466_cast = einsum(equation = var_34466_equation_0, values = (var_34308_cast, var_34225_cast))[name = tensor<string, []>("op_34466_cast")];
            tensor<fp16, []> var_34467_to_fp16 = const()[name = tensor<string, []>("op_34467_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2903_cast = mul(x = var_34466_cast, y = var_34467_to_fp16)[name = tensor<string, []>("aw_2903_cast")];
            tensor<string, []> var_34470_equation_0 = const()[name = tensor<string, []>("op_34470_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34470_cast = einsum(equation = var_34470_equation_0, values = (var_34312_cast, var_34229_cast))[name = tensor<string, []>("op_34470_cast")];
            tensor<fp16, []> var_34471_to_fp16 = const()[name = tensor<string, []>("op_34471_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2905_cast = mul(x = var_34470_cast, y = var_34471_to_fp16)[name = tensor<string, []>("aw_2905_cast")];
            tensor<string, []> var_34474_equation_0 = const()[name = tensor<string, []>("op_34474_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34474_cast = einsum(equation = var_34474_equation_0, values = (var_34316_cast, var_34233_cast))[name = tensor<string, []>("op_34474_cast")];
            tensor<fp16, []> var_34475_to_fp16 = const()[name = tensor<string, []>("op_34475_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2907_cast = mul(x = var_34474_cast, y = var_34475_to_fp16)[name = tensor<string, []>("aw_2907_cast")];
            tensor<string, []> var_34478_equation_0 = const()[name = tensor<string, []>("op_34478_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34478_cast = einsum(equation = var_34478_equation_0, values = (var_34320_cast, var_34237_cast))[name = tensor<string, []>("op_34478_cast")];
            tensor<fp16, []> var_34479_to_fp16 = const()[name = tensor<string, []>("op_34479_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2909_cast = mul(x = var_34478_cast, y = var_34479_to_fp16)[name = tensor<string, []>("aw_2909_cast")];
            tensor<string, []> var_34482_equation_0 = const()[name = tensor<string, []>("op_34482_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34482_cast = einsum(equation = var_34482_equation_0, values = (var_34324_cast, var_34241_cast))[name = tensor<string, []>("op_34482_cast")];
            tensor<fp16, []> var_34483_to_fp16 = const()[name = tensor<string, []>("op_34483_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2911_cast = mul(x = var_34482_cast, y = var_34483_to_fp16)[name = tensor<string, []>("aw_2911_cast")];
            tensor<string, []> var_34486_equation_0 = const()[name = tensor<string, []>("op_34486_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34486_cast = einsum(equation = var_34486_equation_0, values = (var_34328_cast, var_34245_cast))[name = tensor<string, []>("op_34486_cast")];
            tensor<fp16, []> var_34487_to_fp16 = const()[name = tensor<string, []>("op_34487_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2913_cast = mul(x = var_34486_cast, y = var_34487_to_fp16)[name = tensor<string, []>("aw_2913_cast")];
            tensor<string, []> var_34490_equation_0 = const()[name = tensor<string, []>("op_34490_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34490_cast = einsum(equation = var_34490_equation_0, values = (var_34332_cast, var_34249_cast))[name = tensor<string, []>("op_34490_cast")];
            tensor<fp16, []> var_34491_to_fp16 = const()[name = tensor<string, []>("op_34491_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2915_cast = mul(x = var_34490_cast, y = var_34491_to_fp16)[name = tensor<string, []>("aw_2915_cast")];
            tensor<string, []> var_34494_equation_0 = const()[name = tensor<string, []>("op_34494_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34494_cast = einsum(equation = var_34494_equation_0, values = (var_34336_cast, var_34253_cast))[name = tensor<string, []>("op_34494_cast")];
            tensor<fp16, []> var_34495_to_fp16 = const()[name = tensor<string, []>("op_34495_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2917_cast = mul(x = var_34494_cast, y = var_34495_to_fp16)[name = tensor<string, []>("aw_2917_cast")];
            tensor<string, []> var_34498_equation_0 = const()[name = tensor<string, []>("op_34498_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34498_cast = einsum(equation = var_34498_equation_0, values = (var_34340_cast, var_34257_cast))[name = tensor<string, []>("op_34498_cast")];
            tensor<fp16, []> var_34499_to_fp16 = const()[name = tensor<string, []>("op_34499_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2919_cast = mul(x = var_34498_cast, y = var_34499_to_fp16)[name = tensor<string, []>("aw_2919_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34501_cast = softmax(axis = var_30385, x = aw_2881_cast)[name = tensor<string, []>("op_34501_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34502_cast = softmax(axis = var_30385, x = aw_2883_cast)[name = tensor<string, []>("op_34502_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34503_cast = softmax(axis = var_30385, x = aw_2885_cast)[name = tensor<string, []>("op_34503_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34504_cast = softmax(axis = var_30385, x = aw_2887_cast)[name = tensor<string, []>("op_34504_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34505_cast = softmax(axis = var_30385, x = aw_2889_cast)[name = tensor<string, []>("op_34505_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34506_cast = softmax(axis = var_30385, x = aw_2891_cast)[name = tensor<string, []>("op_34506_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34507_cast = softmax(axis = var_30385, x = aw_2893_cast)[name = tensor<string, []>("op_34507_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34508_cast = softmax(axis = var_30385, x = aw_2895_cast)[name = tensor<string, []>("op_34508_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34509_cast = softmax(axis = var_30385, x = aw_2897_cast)[name = tensor<string, []>("op_34509_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34510_cast = softmax(axis = var_30385, x = aw_2899_cast)[name = tensor<string, []>("op_34510_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34511_cast = softmax(axis = var_30385, x = aw_2901_cast)[name = tensor<string, []>("op_34511_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34512_cast = softmax(axis = var_30385, x = aw_2903_cast)[name = tensor<string, []>("op_34512_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34513_cast = softmax(axis = var_30385, x = aw_2905_cast)[name = tensor<string, []>("op_34513_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34514_cast = softmax(axis = var_30385, x = aw_2907_cast)[name = tensor<string, []>("op_34514_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34515_cast = softmax(axis = var_30385, x = aw_2909_cast)[name = tensor<string, []>("op_34515_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34516_cast = softmax(axis = var_30385, x = aw_2911_cast)[name = tensor<string, []>("op_34516_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34517_cast = softmax(axis = var_30385, x = aw_2913_cast)[name = tensor<string, []>("op_34517_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34518_cast = softmax(axis = var_30385, x = aw_2915_cast)[name = tensor<string, []>("op_34518_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34519_cast = softmax(axis = var_30385, x = aw_2917_cast)[name = tensor<string, []>("op_34519_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_34520_cast = softmax(axis = var_30385, x = aw_2919_cast)[name = tensor<string, []>("op_34520_cast")];
            tensor<string, []> var_34522_equation_0 = const()[name = tensor<string, []>("op_34522_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34522_cast = einsum(equation = var_34522_equation_0, values = (var_34342_cast, var_34501_cast))[name = tensor<string, []>("op_34522_cast")];
            tensor<string, []> var_34524_equation_0 = const()[name = tensor<string, []>("op_34524_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34524_cast = einsum(equation = var_34524_equation_0, values = (var_34346_cast, var_34502_cast))[name = tensor<string, []>("op_34524_cast")];
            tensor<string, []> var_34526_equation_0 = const()[name = tensor<string, []>("op_34526_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34526_cast = einsum(equation = var_34526_equation_0, values = (var_34350_cast, var_34503_cast))[name = tensor<string, []>("op_34526_cast")];
            tensor<string, []> var_34528_equation_0 = const()[name = tensor<string, []>("op_34528_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34528_cast = einsum(equation = var_34528_equation_0, values = (var_34354_cast, var_34504_cast))[name = tensor<string, []>("op_34528_cast")];
            tensor<string, []> var_34530_equation_0 = const()[name = tensor<string, []>("op_34530_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34530_cast = einsum(equation = var_34530_equation_0, values = (var_34358_cast, var_34505_cast))[name = tensor<string, []>("op_34530_cast")];
            tensor<string, []> var_34532_equation_0 = const()[name = tensor<string, []>("op_34532_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34532_cast = einsum(equation = var_34532_equation_0, values = (var_34362_cast, var_34506_cast))[name = tensor<string, []>("op_34532_cast")];
            tensor<string, []> var_34534_equation_0 = const()[name = tensor<string, []>("op_34534_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34534_cast = einsum(equation = var_34534_equation_0, values = (var_34366_cast, var_34507_cast))[name = tensor<string, []>("op_34534_cast")];
            tensor<string, []> var_34536_equation_0 = const()[name = tensor<string, []>("op_34536_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34536_cast = einsum(equation = var_34536_equation_0, values = (var_34370_cast, var_34508_cast))[name = tensor<string, []>("op_34536_cast")];
            tensor<string, []> var_34538_equation_0 = const()[name = tensor<string, []>("op_34538_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34538_cast = einsum(equation = var_34538_equation_0, values = (var_34374_cast, var_34509_cast))[name = tensor<string, []>("op_34538_cast")];
            tensor<string, []> var_34540_equation_0 = const()[name = tensor<string, []>("op_34540_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34540_cast = einsum(equation = var_34540_equation_0, values = (var_34378_cast, var_34510_cast))[name = tensor<string, []>("op_34540_cast")];
            tensor<string, []> var_34542_equation_0 = const()[name = tensor<string, []>("op_34542_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34542_cast = einsum(equation = var_34542_equation_0, values = (var_34382_cast, var_34511_cast))[name = tensor<string, []>("op_34542_cast")];
            tensor<string, []> var_34544_equation_0 = const()[name = tensor<string, []>("op_34544_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34544_cast = einsum(equation = var_34544_equation_0, values = (var_34386_cast, var_34512_cast))[name = tensor<string, []>("op_34544_cast")];
            tensor<string, []> var_34546_equation_0 = const()[name = tensor<string, []>("op_34546_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34546_cast = einsum(equation = var_34546_equation_0, values = (var_34390_cast, var_34513_cast))[name = tensor<string, []>("op_34546_cast")];
            tensor<string, []> var_34548_equation_0 = const()[name = tensor<string, []>("op_34548_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34548_cast = einsum(equation = var_34548_equation_0, values = (var_34394_cast, var_34514_cast))[name = tensor<string, []>("op_34548_cast")];
            tensor<string, []> var_34550_equation_0 = const()[name = tensor<string, []>("op_34550_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34550_cast = einsum(equation = var_34550_equation_0, values = (var_34398_cast, var_34515_cast))[name = tensor<string, []>("op_34550_cast")];
            tensor<string, []> var_34552_equation_0 = const()[name = tensor<string, []>("op_34552_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34552_cast = einsum(equation = var_34552_equation_0, values = (var_34402_cast, var_34516_cast))[name = tensor<string, []>("op_34552_cast")];
            tensor<string, []> var_34554_equation_0 = const()[name = tensor<string, []>("op_34554_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34554_cast = einsum(equation = var_34554_equation_0, values = (var_34406_cast, var_34517_cast))[name = tensor<string, []>("op_34554_cast")];
            tensor<string, []> var_34556_equation_0 = const()[name = tensor<string, []>("op_34556_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34556_cast = einsum(equation = var_34556_equation_0, values = (var_34410_cast, var_34518_cast))[name = tensor<string, []>("op_34556_cast")];
            tensor<string, []> var_34558_equation_0 = const()[name = tensor<string, []>("op_34558_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34558_cast = einsum(equation = var_34558_equation_0, values = (var_34414_cast, var_34519_cast))[name = tensor<string, []>("op_34558_cast")];
            tensor<string, []> var_34560_equation_0 = const()[name = tensor<string, []>("op_34560_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34560_cast = einsum(equation = var_34560_equation_0, values = (var_34418_cast, var_34520_cast))[name = tensor<string, []>("op_34560_cast")];
            tensor<bool, []> input_473_interleave_0 = const()[name = tensor<string, []>("input_473_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_473_cast = concat(axis = var_30385, interleave = input_473_interleave_0, values = (var_34522_cast, var_34524_cast, var_34526_cast, var_34528_cast, var_34530_cast, var_34532_cast, var_34534_cast, var_34536_cast, var_34538_cast, var_34540_cast, var_34542_cast, var_34544_cast, var_34546_cast, var_34548_cast, var_34550_cast, var_34552_cast, var_34554_cast, var_34556_cast, var_34558_cast, var_34560_cast))[name = tensor<string, []>("input_473_cast")];
            tensor<int32, [2]> var_34566 = const()[name = tensor<string, []>("op_34566"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_34568 = const()[name = tensor<string, []>("op_34568"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_34570_pad_type_0 = const()[name = tensor<string, []>("op_34570_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_34570_pad_0 = const()[name = tensor<string, []>("op_34570_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2876309056)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2879585920)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_34570_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_out_0_bias_to_fp16, dilations = var_34568, groups = var_30385, pad = var_34570_pad_0, pad_type = var_34570_pad_type_0, strides = var_34566, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn1_to_out_0_weight_to_fp16, x = input_473_cast)[name = tensor<string, []>("op_34570_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_231_cast = add(x = var_34570_cast, y = inputs_229_cast)[name = tensor<string, []>("inputs_231_cast")];
            tensor<int32, [1]> var_34574 = const()[name = tensor<string, []>("op_34574"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_231_cast = reduce_mean(axes = var_34574, keep_dims = var_30380, x = inputs_231_cast)[name = tensor<string, []>("channels_mean_231_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_231_cast = sub(x = inputs_231_cast, y = channels_mean_231_cast)[name = tensor<string, []>("zero_mean_231_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_231_cast = mul(x = zero_mean_231_cast, y = zero_mean_231_cast)[name = tensor<string, []>("zero_mean_sq_231_cast")];
            tensor<int32, [1]> var_34578 = const()[name = tensor<string, []>("op_34578"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_34579_cast = reduce_mean(axes = var_34578, keep_dims = var_30380, x = zero_mean_sq_231_cast)[name = tensor<string, []>("op_34579_cast")];
            tensor<fp16, []> var_34580_to_fp16 = const()[name = tensor<string, []>("op_34580_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_34581_cast = add(x = var_34579_cast, y = var_34580_to_fp16)[name = tensor<string, []>("op_34581_cast")];
            tensor<fp16, []> denom_231_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_231_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_231_cast = rsqrt(epsilon = denom_231_epsilon_0_to_fp16, x = var_34581_cast)[name = tensor<string, []>("denom_231_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_231_cast = mul(x = zero_mean_231_cast, y = denom_231_cast)[name = tensor<string, []>("out_231_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_34585_to_fp16 = const()[name = tensor<string, []>("op_34585_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2879588544)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_34586_cast = add(x = out_231_cast, y = var_34585_to_fp16)[name = tensor<string, []>("op_34586_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_34588_to_fp16 = const()[name = tensor<string, []>("op_34588_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2879591168)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_313_cast = mul(x = var_34586_cast, y = var_34588_to_fp16)[name = tensor<string, []>("hidden_states_313_cast")];
            tensor<int32, [2]> var_34595 = const()[name = tensor<string, []>("op_34595"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_34597 = const()[name = tensor<string, []>("op_34597"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_155_pad_type_0 = const()[name = tensor<string, []>("q_155_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_155_pad_0 = const()[name = tensor<string, []>("q_155_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2879593792)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_155_cast = conv(dilations = var_34597, groups = var_30385, pad = q_155_pad_0, pad_type = q_155_pad_type_0, strides = var_34595, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_q_weight_to_fp16, x = hidden_states_313_cast)[name = tensor<string, []>("q_155_cast")];
            tensor<int32, [2]> var_34601 = const()[name = tensor<string, []>("op_34601"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_34603 = const()[name = tensor<string, []>("op_34603"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_309_pad_type_0 = const()[name = tensor<string, []>("k_309_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_309_pad_0 = const()[name = tensor<string, []>("k_309_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2882870656)))];
            tensor<fp16, [2, 1280, 1, 77]> k_309_cast = conv(dilations = var_34603, groups = var_30385, pad = k_309_pad_0, pad_type = k_309_pad_type_0, strides = var_34601, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_309_cast")];
            tensor<int32, [2]> var_34607 = const()[name = tensor<string, []>("op_34607"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_34609 = const()[name = tensor<string, []>("op_34609"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_155_pad_type_0 = const()[name = tensor<string, []>("v_155_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_155_pad_0 = const()[name = tensor<string, []>("v_155_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2888113600)))];
            tensor<fp16, [2, 1280, 1, 77]> v_155_cast = conv(dilations = var_34609, groups = var_30385, pad = v_155_pad_0, pad_type = v_155_pad_type_0, strides = var_34607, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_155_cast")];
            tensor<int32, [4]> var_34613_begin_0 = const()[name = tensor<string, []>("op_34613_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_34613_end_0 = const()[name = tensor<string, []>("op_34613_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_34613_end_mask_0 = const()[name = tensor<string, []>("op_34613_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34613_cast = slice_by_index(begin = var_34613_begin_0, end = var_34613_end_0, end_mask = var_34613_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34613_cast")];
            tensor<int32, [4]> var_34617_begin_0 = const()[name = tensor<string, []>("op_34617_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_34617_end_0 = const()[name = tensor<string, []>("op_34617_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_34617_end_mask_0 = const()[name = tensor<string, []>("op_34617_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34617_cast = slice_by_index(begin = var_34617_begin_0, end = var_34617_end_0, end_mask = var_34617_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34617_cast")];
            tensor<int32, [4]> var_34621_begin_0 = const()[name = tensor<string, []>("op_34621_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_34621_end_0 = const()[name = tensor<string, []>("op_34621_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_34621_end_mask_0 = const()[name = tensor<string, []>("op_34621_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34621_cast = slice_by_index(begin = var_34621_begin_0, end = var_34621_end_0, end_mask = var_34621_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34621_cast")];
            tensor<int32, [4]> var_34625_begin_0 = const()[name = tensor<string, []>("op_34625_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_34625_end_0 = const()[name = tensor<string, []>("op_34625_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_34625_end_mask_0 = const()[name = tensor<string, []>("op_34625_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34625_cast = slice_by_index(begin = var_34625_begin_0, end = var_34625_end_0, end_mask = var_34625_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34625_cast")];
            tensor<int32, [4]> var_34629_begin_0 = const()[name = tensor<string, []>("op_34629_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_34629_end_0 = const()[name = tensor<string, []>("op_34629_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_34629_end_mask_0 = const()[name = tensor<string, []>("op_34629_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34629_cast = slice_by_index(begin = var_34629_begin_0, end = var_34629_end_0, end_mask = var_34629_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34629_cast")];
            tensor<int32, [4]> var_34633_begin_0 = const()[name = tensor<string, []>("op_34633_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_34633_end_0 = const()[name = tensor<string, []>("op_34633_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_34633_end_mask_0 = const()[name = tensor<string, []>("op_34633_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34633_cast = slice_by_index(begin = var_34633_begin_0, end = var_34633_end_0, end_mask = var_34633_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34633_cast")];
            tensor<int32, [4]> var_34637_begin_0 = const()[name = tensor<string, []>("op_34637_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_34637_end_0 = const()[name = tensor<string, []>("op_34637_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_34637_end_mask_0 = const()[name = tensor<string, []>("op_34637_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34637_cast = slice_by_index(begin = var_34637_begin_0, end = var_34637_end_0, end_mask = var_34637_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34637_cast")];
            tensor<int32, [4]> var_34641_begin_0 = const()[name = tensor<string, []>("op_34641_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_34641_end_0 = const()[name = tensor<string, []>("op_34641_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_34641_end_mask_0 = const()[name = tensor<string, []>("op_34641_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34641_cast = slice_by_index(begin = var_34641_begin_0, end = var_34641_end_0, end_mask = var_34641_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34641_cast")];
            tensor<int32, [4]> var_34645_begin_0 = const()[name = tensor<string, []>("op_34645_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_34645_end_0 = const()[name = tensor<string, []>("op_34645_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_34645_end_mask_0 = const()[name = tensor<string, []>("op_34645_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34645_cast = slice_by_index(begin = var_34645_begin_0, end = var_34645_end_0, end_mask = var_34645_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34645_cast")];
            tensor<int32, [4]> var_34649_begin_0 = const()[name = tensor<string, []>("op_34649_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_34649_end_0 = const()[name = tensor<string, []>("op_34649_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_34649_end_mask_0 = const()[name = tensor<string, []>("op_34649_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34649_cast = slice_by_index(begin = var_34649_begin_0, end = var_34649_end_0, end_mask = var_34649_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34649_cast")];
            tensor<int32, [4]> var_34653_begin_0 = const()[name = tensor<string, []>("op_34653_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_34653_end_0 = const()[name = tensor<string, []>("op_34653_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_34653_end_mask_0 = const()[name = tensor<string, []>("op_34653_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34653_cast = slice_by_index(begin = var_34653_begin_0, end = var_34653_end_0, end_mask = var_34653_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34653_cast")];
            tensor<int32, [4]> var_34657_begin_0 = const()[name = tensor<string, []>("op_34657_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_34657_end_0 = const()[name = tensor<string, []>("op_34657_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_34657_end_mask_0 = const()[name = tensor<string, []>("op_34657_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34657_cast = slice_by_index(begin = var_34657_begin_0, end = var_34657_end_0, end_mask = var_34657_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34657_cast")];
            tensor<int32, [4]> var_34661_begin_0 = const()[name = tensor<string, []>("op_34661_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_34661_end_0 = const()[name = tensor<string, []>("op_34661_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_34661_end_mask_0 = const()[name = tensor<string, []>("op_34661_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34661_cast = slice_by_index(begin = var_34661_begin_0, end = var_34661_end_0, end_mask = var_34661_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34661_cast")];
            tensor<int32, [4]> var_34665_begin_0 = const()[name = tensor<string, []>("op_34665_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_34665_end_0 = const()[name = tensor<string, []>("op_34665_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_34665_end_mask_0 = const()[name = tensor<string, []>("op_34665_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34665_cast = slice_by_index(begin = var_34665_begin_0, end = var_34665_end_0, end_mask = var_34665_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34665_cast")];
            tensor<int32, [4]> var_34669_begin_0 = const()[name = tensor<string, []>("op_34669_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_34669_end_0 = const()[name = tensor<string, []>("op_34669_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_34669_end_mask_0 = const()[name = tensor<string, []>("op_34669_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34669_cast = slice_by_index(begin = var_34669_begin_0, end = var_34669_end_0, end_mask = var_34669_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34669_cast")];
            tensor<int32, [4]> var_34673_begin_0 = const()[name = tensor<string, []>("op_34673_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_34673_end_0 = const()[name = tensor<string, []>("op_34673_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_34673_end_mask_0 = const()[name = tensor<string, []>("op_34673_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34673_cast = slice_by_index(begin = var_34673_begin_0, end = var_34673_end_0, end_mask = var_34673_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34673_cast")];
            tensor<int32, [4]> var_34677_begin_0 = const()[name = tensor<string, []>("op_34677_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_34677_end_0 = const()[name = tensor<string, []>("op_34677_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_34677_end_mask_0 = const()[name = tensor<string, []>("op_34677_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34677_cast = slice_by_index(begin = var_34677_begin_0, end = var_34677_end_0, end_mask = var_34677_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34677_cast")];
            tensor<int32, [4]> var_34681_begin_0 = const()[name = tensor<string, []>("op_34681_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_34681_end_0 = const()[name = tensor<string, []>("op_34681_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_34681_end_mask_0 = const()[name = tensor<string, []>("op_34681_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34681_cast = slice_by_index(begin = var_34681_begin_0, end = var_34681_end_0, end_mask = var_34681_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34681_cast")];
            tensor<int32, [4]> var_34685_begin_0 = const()[name = tensor<string, []>("op_34685_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_34685_end_0 = const()[name = tensor<string, []>("op_34685_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_34685_end_mask_0 = const()[name = tensor<string, []>("op_34685_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34685_cast = slice_by_index(begin = var_34685_begin_0, end = var_34685_end_0, end_mask = var_34685_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34685_cast")];
            tensor<int32, [4]> var_34689_begin_0 = const()[name = tensor<string, []>("op_34689_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_34689_end_0 = const()[name = tensor<string, []>("op_34689_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_34689_end_mask_0 = const()[name = tensor<string, []>("op_34689_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_34689_cast = slice_by_index(begin = var_34689_begin_0, end = var_34689_end_0, end_mask = var_34689_end_mask_0, x = q_155_cast)[name = tensor<string, []>("op_34689_cast")];
            tensor<int32, [4]> k_311_perm_0 = const()[name = tensor<string, []>("k_311_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_34696_begin_0 = const()[name = tensor<string, []>("op_34696_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_34696_end_0 = const()[name = tensor<string, []>("op_34696_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_34696_end_mask_0 = const()[name = tensor<string, []>("op_34696_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_62 = transpose(perm = k_311_perm_0, x = k_309_cast)[name = tensor<string, []>("transpose_62")];
            tensor<fp16, [2, 77, 1, 64]> var_34696_cast = slice_by_index(begin = var_34696_begin_0, end = var_34696_end_0, end_mask = var_34696_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34696_cast")];
            tensor<int32, [4]> var_34700_begin_0 = const()[name = tensor<string, []>("op_34700_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_34700_end_0 = const()[name = tensor<string, []>("op_34700_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_34700_end_mask_0 = const()[name = tensor<string, []>("op_34700_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34700_cast = slice_by_index(begin = var_34700_begin_0, end = var_34700_end_0, end_mask = var_34700_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34700_cast")];
            tensor<int32, [4]> var_34704_begin_0 = const()[name = tensor<string, []>("op_34704_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_34704_end_0 = const()[name = tensor<string, []>("op_34704_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_34704_end_mask_0 = const()[name = tensor<string, []>("op_34704_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34704_cast = slice_by_index(begin = var_34704_begin_0, end = var_34704_end_0, end_mask = var_34704_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34704_cast")];
            tensor<int32, [4]> var_34708_begin_0 = const()[name = tensor<string, []>("op_34708_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_34708_end_0 = const()[name = tensor<string, []>("op_34708_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_34708_end_mask_0 = const()[name = tensor<string, []>("op_34708_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34708_cast = slice_by_index(begin = var_34708_begin_0, end = var_34708_end_0, end_mask = var_34708_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34708_cast")];
            tensor<int32, [4]> var_34712_begin_0 = const()[name = tensor<string, []>("op_34712_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_34712_end_0 = const()[name = tensor<string, []>("op_34712_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_34712_end_mask_0 = const()[name = tensor<string, []>("op_34712_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34712_cast = slice_by_index(begin = var_34712_begin_0, end = var_34712_end_0, end_mask = var_34712_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34712_cast")];
            tensor<int32, [4]> var_34716_begin_0 = const()[name = tensor<string, []>("op_34716_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_34716_end_0 = const()[name = tensor<string, []>("op_34716_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_34716_end_mask_0 = const()[name = tensor<string, []>("op_34716_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34716_cast = slice_by_index(begin = var_34716_begin_0, end = var_34716_end_0, end_mask = var_34716_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34716_cast")];
            tensor<int32, [4]> var_34720_begin_0 = const()[name = tensor<string, []>("op_34720_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_34720_end_0 = const()[name = tensor<string, []>("op_34720_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_34720_end_mask_0 = const()[name = tensor<string, []>("op_34720_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34720_cast = slice_by_index(begin = var_34720_begin_0, end = var_34720_end_0, end_mask = var_34720_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34720_cast")];
            tensor<int32, [4]> var_34724_begin_0 = const()[name = tensor<string, []>("op_34724_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_34724_end_0 = const()[name = tensor<string, []>("op_34724_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_34724_end_mask_0 = const()[name = tensor<string, []>("op_34724_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34724_cast = slice_by_index(begin = var_34724_begin_0, end = var_34724_end_0, end_mask = var_34724_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34724_cast")];
            tensor<int32, [4]> var_34728_begin_0 = const()[name = tensor<string, []>("op_34728_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_34728_end_0 = const()[name = tensor<string, []>("op_34728_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_34728_end_mask_0 = const()[name = tensor<string, []>("op_34728_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34728_cast = slice_by_index(begin = var_34728_begin_0, end = var_34728_end_0, end_mask = var_34728_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34728_cast")];
            tensor<int32, [4]> var_34732_begin_0 = const()[name = tensor<string, []>("op_34732_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_34732_end_0 = const()[name = tensor<string, []>("op_34732_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_34732_end_mask_0 = const()[name = tensor<string, []>("op_34732_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34732_cast = slice_by_index(begin = var_34732_begin_0, end = var_34732_end_0, end_mask = var_34732_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34732_cast")];
            tensor<int32, [4]> var_34736_begin_0 = const()[name = tensor<string, []>("op_34736_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_34736_end_0 = const()[name = tensor<string, []>("op_34736_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_34736_end_mask_0 = const()[name = tensor<string, []>("op_34736_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34736_cast = slice_by_index(begin = var_34736_begin_0, end = var_34736_end_0, end_mask = var_34736_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34736_cast")];
            tensor<int32, [4]> var_34740_begin_0 = const()[name = tensor<string, []>("op_34740_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_34740_end_0 = const()[name = tensor<string, []>("op_34740_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_34740_end_mask_0 = const()[name = tensor<string, []>("op_34740_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34740_cast = slice_by_index(begin = var_34740_begin_0, end = var_34740_end_0, end_mask = var_34740_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34740_cast")];
            tensor<int32, [4]> var_34744_begin_0 = const()[name = tensor<string, []>("op_34744_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_34744_end_0 = const()[name = tensor<string, []>("op_34744_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_34744_end_mask_0 = const()[name = tensor<string, []>("op_34744_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34744_cast = slice_by_index(begin = var_34744_begin_0, end = var_34744_end_0, end_mask = var_34744_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34744_cast")];
            tensor<int32, [4]> var_34748_begin_0 = const()[name = tensor<string, []>("op_34748_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_34748_end_0 = const()[name = tensor<string, []>("op_34748_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_34748_end_mask_0 = const()[name = tensor<string, []>("op_34748_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34748_cast = slice_by_index(begin = var_34748_begin_0, end = var_34748_end_0, end_mask = var_34748_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34748_cast")];
            tensor<int32, [4]> var_34752_begin_0 = const()[name = tensor<string, []>("op_34752_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_34752_end_0 = const()[name = tensor<string, []>("op_34752_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_34752_end_mask_0 = const()[name = tensor<string, []>("op_34752_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34752_cast = slice_by_index(begin = var_34752_begin_0, end = var_34752_end_0, end_mask = var_34752_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34752_cast")];
            tensor<int32, [4]> var_34756_begin_0 = const()[name = tensor<string, []>("op_34756_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_34756_end_0 = const()[name = tensor<string, []>("op_34756_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_34756_end_mask_0 = const()[name = tensor<string, []>("op_34756_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34756_cast = slice_by_index(begin = var_34756_begin_0, end = var_34756_end_0, end_mask = var_34756_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34756_cast")];
            tensor<int32, [4]> var_34760_begin_0 = const()[name = tensor<string, []>("op_34760_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_34760_end_0 = const()[name = tensor<string, []>("op_34760_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_34760_end_mask_0 = const()[name = tensor<string, []>("op_34760_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34760_cast = slice_by_index(begin = var_34760_begin_0, end = var_34760_end_0, end_mask = var_34760_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34760_cast")];
            tensor<int32, [4]> var_34764_begin_0 = const()[name = tensor<string, []>("op_34764_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_34764_end_0 = const()[name = tensor<string, []>("op_34764_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_34764_end_mask_0 = const()[name = tensor<string, []>("op_34764_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34764_cast = slice_by_index(begin = var_34764_begin_0, end = var_34764_end_0, end_mask = var_34764_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34764_cast")];
            tensor<int32, [4]> var_34768_begin_0 = const()[name = tensor<string, []>("op_34768_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_34768_end_0 = const()[name = tensor<string, []>("op_34768_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_34768_end_mask_0 = const()[name = tensor<string, []>("op_34768_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34768_cast = slice_by_index(begin = var_34768_begin_0, end = var_34768_end_0, end_mask = var_34768_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34768_cast")];
            tensor<int32, [4]> var_34772_begin_0 = const()[name = tensor<string, []>("op_34772_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_34772_end_0 = const()[name = tensor<string, []>("op_34772_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_34772_end_mask_0 = const()[name = tensor<string, []>("op_34772_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_34772_cast = slice_by_index(begin = var_34772_begin_0, end = var_34772_end_0, end_mask = var_34772_end_mask_0, x = transpose_62)[name = tensor<string, []>("op_34772_cast")];
            tensor<int32, [4]> var_34774_begin_0 = const()[name = tensor<string, []>("op_34774_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_34774_end_0 = const()[name = tensor<string, []>("op_34774_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_34774_end_mask_0 = const()[name = tensor<string, []>("op_34774_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34774_cast = slice_by_index(begin = var_34774_begin_0, end = var_34774_end_0, end_mask = var_34774_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34774_cast")];
            tensor<int32, [4]> var_34778_begin_0 = const()[name = tensor<string, []>("op_34778_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_34778_end_0 = const()[name = tensor<string, []>("op_34778_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_34778_end_mask_0 = const()[name = tensor<string, []>("op_34778_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34778_cast = slice_by_index(begin = var_34778_begin_0, end = var_34778_end_0, end_mask = var_34778_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34778_cast")];
            tensor<int32, [4]> var_34782_begin_0 = const()[name = tensor<string, []>("op_34782_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_34782_end_0 = const()[name = tensor<string, []>("op_34782_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_34782_end_mask_0 = const()[name = tensor<string, []>("op_34782_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34782_cast = slice_by_index(begin = var_34782_begin_0, end = var_34782_end_0, end_mask = var_34782_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34782_cast")];
            tensor<int32, [4]> var_34786_begin_0 = const()[name = tensor<string, []>("op_34786_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_34786_end_0 = const()[name = tensor<string, []>("op_34786_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_34786_end_mask_0 = const()[name = tensor<string, []>("op_34786_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34786_cast = slice_by_index(begin = var_34786_begin_0, end = var_34786_end_0, end_mask = var_34786_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34786_cast")];
            tensor<int32, [4]> var_34790_begin_0 = const()[name = tensor<string, []>("op_34790_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_34790_end_0 = const()[name = tensor<string, []>("op_34790_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_34790_end_mask_0 = const()[name = tensor<string, []>("op_34790_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34790_cast = slice_by_index(begin = var_34790_begin_0, end = var_34790_end_0, end_mask = var_34790_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34790_cast")];
            tensor<int32, [4]> var_34794_begin_0 = const()[name = tensor<string, []>("op_34794_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_34794_end_0 = const()[name = tensor<string, []>("op_34794_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_34794_end_mask_0 = const()[name = tensor<string, []>("op_34794_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34794_cast = slice_by_index(begin = var_34794_begin_0, end = var_34794_end_0, end_mask = var_34794_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34794_cast")];
            tensor<int32, [4]> var_34798_begin_0 = const()[name = tensor<string, []>("op_34798_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_34798_end_0 = const()[name = tensor<string, []>("op_34798_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_34798_end_mask_0 = const()[name = tensor<string, []>("op_34798_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34798_cast = slice_by_index(begin = var_34798_begin_0, end = var_34798_end_0, end_mask = var_34798_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34798_cast")];
            tensor<int32, [4]> var_34802_begin_0 = const()[name = tensor<string, []>("op_34802_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_34802_end_0 = const()[name = tensor<string, []>("op_34802_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_34802_end_mask_0 = const()[name = tensor<string, []>("op_34802_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34802_cast = slice_by_index(begin = var_34802_begin_0, end = var_34802_end_0, end_mask = var_34802_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34802_cast")];
            tensor<int32, [4]> var_34806_begin_0 = const()[name = tensor<string, []>("op_34806_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_34806_end_0 = const()[name = tensor<string, []>("op_34806_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_34806_end_mask_0 = const()[name = tensor<string, []>("op_34806_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34806_cast = slice_by_index(begin = var_34806_begin_0, end = var_34806_end_0, end_mask = var_34806_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34806_cast")];
            tensor<int32, [4]> var_34810_begin_0 = const()[name = tensor<string, []>("op_34810_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_34810_end_0 = const()[name = tensor<string, []>("op_34810_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_34810_end_mask_0 = const()[name = tensor<string, []>("op_34810_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34810_cast = slice_by_index(begin = var_34810_begin_0, end = var_34810_end_0, end_mask = var_34810_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34810_cast")];
            tensor<int32, [4]> var_34814_begin_0 = const()[name = tensor<string, []>("op_34814_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_34814_end_0 = const()[name = tensor<string, []>("op_34814_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_34814_end_mask_0 = const()[name = tensor<string, []>("op_34814_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34814_cast = slice_by_index(begin = var_34814_begin_0, end = var_34814_end_0, end_mask = var_34814_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34814_cast")];
            tensor<int32, [4]> var_34818_begin_0 = const()[name = tensor<string, []>("op_34818_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_34818_end_0 = const()[name = tensor<string, []>("op_34818_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_34818_end_mask_0 = const()[name = tensor<string, []>("op_34818_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34818_cast = slice_by_index(begin = var_34818_begin_0, end = var_34818_end_0, end_mask = var_34818_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34818_cast")];
            tensor<int32, [4]> var_34822_begin_0 = const()[name = tensor<string, []>("op_34822_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_34822_end_0 = const()[name = tensor<string, []>("op_34822_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_34822_end_mask_0 = const()[name = tensor<string, []>("op_34822_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34822_cast = slice_by_index(begin = var_34822_begin_0, end = var_34822_end_0, end_mask = var_34822_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34822_cast")];
            tensor<int32, [4]> var_34826_begin_0 = const()[name = tensor<string, []>("op_34826_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_34826_end_0 = const()[name = tensor<string, []>("op_34826_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_34826_end_mask_0 = const()[name = tensor<string, []>("op_34826_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34826_cast = slice_by_index(begin = var_34826_begin_0, end = var_34826_end_0, end_mask = var_34826_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34826_cast")];
            tensor<int32, [4]> var_34830_begin_0 = const()[name = tensor<string, []>("op_34830_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_34830_end_0 = const()[name = tensor<string, []>("op_34830_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_34830_end_mask_0 = const()[name = tensor<string, []>("op_34830_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34830_cast = slice_by_index(begin = var_34830_begin_0, end = var_34830_end_0, end_mask = var_34830_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34830_cast")];
            tensor<int32, [4]> var_34834_begin_0 = const()[name = tensor<string, []>("op_34834_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_34834_end_0 = const()[name = tensor<string, []>("op_34834_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_34834_end_mask_0 = const()[name = tensor<string, []>("op_34834_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34834_cast = slice_by_index(begin = var_34834_begin_0, end = var_34834_end_0, end_mask = var_34834_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34834_cast")];
            tensor<int32, [4]> var_34838_begin_0 = const()[name = tensor<string, []>("op_34838_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_34838_end_0 = const()[name = tensor<string, []>("op_34838_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_34838_end_mask_0 = const()[name = tensor<string, []>("op_34838_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34838_cast = slice_by_index(begin = var_34838_begin_0, end = var_34838_end_0, end_mask = var_34838_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34838_cast")];
            tensor<int32, [4]> var_34842_begin_0 = const()[name = tensor<string, []>("op_34842_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_34842_end_0 = const()[name = tensor<string, []>("op_34842_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_34842_end_mask_0 = const()[name = tensor<string, []>("op_34842_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34842_cast = slice_by_index(begin = var_34842_begin_0, end = var_34842_end_0, end_mask = var_34842_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34842_cast")];
            tensor<int32, [4]> var_34846_begin_0 = const()[name = tensor<string, []>("op_34846_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_34846_end_0 = const()[name = tensor<string, []>("op_34846_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_34846_end_mask_0 = const()[name = tensor<string, []>("op_34846_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34846_cast = slice_by_index(begin = var_34846_begin_0, end = var_34846_end_0, end_mask = var_34846_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34846_cast")];
            tensor<int32, [4]> var_34850_begin_0 = const()[name = tensor<string, []>("op_34850_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_34850_end_0 = const()[name = tensor<string, []>("op_34850_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_34850_end_mask_0 = const()[name = tensor<string, []>("op_34850_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_34850_cast = slice_by_index(begin = var_34850_begin_0, end = var_34850_end_0, end_mask = var_34850_end_mask_0, x = v_155_cast)[name = tensor<string, []>("op_34850_cast")];
            tensor<string, []> var_34854_equation_0 = const()[name = tensor<string, []>("op_34854_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34854_cast = einsum(equation = var_34854_equation_0, values = (var_34696_cast, var_34613_cast))[name = tensor<string, []>("op_34854_cast")];
            tensor<fp16, []> var_34855_to_fp16 = const()[name = tensor<string, []>("op_34855_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2921_cast = mul(x = var_34854_cast, y = var_34855_to_fp16)[name = tensor<string, []>("aw_2921_cast")];
            tensor<string, []> var_34858_equation_0 = const()[name = tensor<string, []>("op_34858_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34858_cast = einsum(equation = var_34858_equation_0, values = (var_34700_cast, var_34617_cast))[name = tensor<string, []>("op_34858_cast")];
            tensor<fp16, []> var_34859_to_fp16 = const()[name = tensor<string, []>("op_34859_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2923_cast = mul(x = var_34858_cast, y = var_34859_to_fp16)[name = tensor<string, []>("aw_2923_cast")];
            tensor<string, []> var_34862_equation_0 = const()[name = tensor<string, []>("op_34862_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34862_cast = einsum(equation = var_34862_equation_0, values = (var_34704_cast, var_34621_cast))[name = tensor<string, []>("op_34862_cast")];
            tensor<fp16, []> var_34863_to_fp16 = const()[name = tensor<string, []>("op_34863_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2925_cast = mul(x = var_34862_cast, y = var_34863_to_fp16)[name = tensor<string, []>("aw_2925_cast")];
            tensor<string, []> var_34866_equation_0 = const()[name = tensor<string, []>("op_34866_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34866_cast = einsum(equation = var_34866_equation_0, values = (var_34708_cast, var_34625_cast))[name = tensor<string, []>("op_34866_cast")];
            tensor<fp16, []> var_34867_to_fp16 = const()[name = tensor<string, []>("op_34867_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2927_cast = mul(x = var_34866_cast, y = var_34867_to_fp16)[name = tensor<string, []>("aw_2927_cast")];
            tensor<string, []> var_34870_equation_0 = const()[name = tensor<string, []>("op_34870_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34870_cast = einsum(equation = var_34870_equation_0, values = (var_34712_cast, var_34629_cast))[name = tensor<string, []>("op_34870_cast")];
            tensor<fp16, []> var_34871_to_fp16 = const()[name = tensor<string, []>("op_34871_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2929_cast = mul(x = var_34870_cast, y = var_34871_to_fp16)[name = tensor<string, []>("aw_2929_cast")];
            tensor<string, []> var_34874_equation_0 = const()[name = tensor<string, []>("op_34874_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34874_cast = einsum(equation = var_34874_equation_0, values = (var_34716_cast, var_34633_cast))[name = tensor<string, []>("op_34874_cast")];
            tensor<fp16, []> var_34875_to_fp16 = const()[name = tensor<string, []>("op_34875_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2931_cast = mul(x = var_34874_cast, y = var_34875_to_fp16)[name = tensor<string, []>("aw_2931_cast")];
            tensor<string, []> var_34878_equation_0 = const()[name = tensor<string, []>("op_34878_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34878_cast = einsum(equation = var_34878_equation_0, values = (var_34720_cast, var_34637_cast))[name = tensor<string, []>("op_34878_cast")];
            tensor<fp16, []> var_34879_to_fp16 = const()[name = tensor<string, []>("op_34879_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2933_cast = mul(x = var_34878_cast, y = var_34879_to_fp16)[name = tensor<string, []>("aw_2933_cast")];
            tensor<string, []> var_34882_equation_0 = const()[name = tensor<string, []>("op_34882_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34882_cast = einsum(equation = var_34882_equation_0, values = (var_34724_cast, var_34641_cast))[name = tensor<string, []>("op_34882_cast")];
            tensor<fp16, []> var_34883_to_fp16 = const()[name = tensor<string, []>("op_34883_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2935_cast = mul(x = var_34882_cast, y = var_34883_to_fp16)[name = tensor<string, []>("aw_2935_cast")];
            tensor<string, []> var_34886_equation_0 = const()[name = tensor<string, []>("op_34886_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34886_cast = einsum(equation = var_34886_equation_0, values = (var_34728_cast, var_34645_cast))[name = tensor<string, []>("op_34886_cast")];
            tensor<fp16, []> var_34887_to_fp16 = const()[name = tensor<string, []>("op_34887_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2937_cast = mul(x = var_34886_cast, y = var_34887_to_fp16)[name = tensor<string, []>("aw_2937_cast")];
            tensor<string, []> var_34890_equation_0 = const()[name = tensor<string, []>("op_34890_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34890_cast = einsum(equation = var_34890_equation_0, values = (var_34732_cast, var_34649_cast))[name = tensor<string, []>("op_34890_cast")];
            tensor<fp16, []> var_34891_to_fp16 = const()[name = tensor<string, []>("op_34891_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2939_cast = mul(x = var_34890_cast, y = var_34891_to_fp16)[name = tensor<string, []>("aw_2939_cast")];
            tensor<string, []> var_34894_equation_0 = const()[name = tensor<string, []>("op_34894_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34894_cast = einsum(equation = var_34894_equation_0, values = (var_34736_cast, var_34653_cast))[name = tensor<string, []>("op_34894_cast")];
            tensor<fp16, []> var_34895_to_fp16 = const()[name = tensor<string, []>("op_34895_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2941_cast = mul(x = var_34894_cast, y = var_34895_to_fp16)[name = tensor<string, []>("aw_2941_cast")];
            tensor<string, []> var_34898_equation_0 = const()[name = tensor<string, []>("op_34898_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34898_cast = einsum(equation = var_34898_equation_0, values = (var_34740_cast, var_34657_cast))[name = tensor<string, []>("op_34898_cast")];
            tensor<fp16, []> var_34899_to_fp16 = const()[name = tensor<string, []>("op_34899_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2943_cast = mul(x = var_34898_cast, y = var_34899_to_fp16)[name = tensor<string, []>("aw_2943_cast")];
            tensor<string, []> var_34902_equation_0 = const()[name = tensor<string, []>("op_34902_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34902_cast = einsum(equation = var_34902_equation_0, values = (var_34744_cast, var_34661_cast))[name = tensor<string, []>("op_34902_cast")];
            tensor<fp16, []> var_34903_to_fp16 = const()[name = tensor<string, []>("op_34903_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2945_cast = mul(x = var_34902_cast, y = var_34903_to_fp16)[name = tensor<string, []>("aw_2945_cast")];
            tensor<string, []> var_34906_equation_0 = const()[name = tensor<string, []>("op_34906_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34906_cast = einsum(equation = var_34906_equation_0, values = (var_34748_cast, var_34665_cast))[name = tensor<string, []>("op_34906_cast")];
            tensor<fp16, []> var_34907_to_fp16 = const()[name = tensor<string, []>("op_34907_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2947_cast = mul(x = var_34906_cast, y = var_34907_to_fp16)[name = tensor<string, []>("aw_2947_cast")];
            tensor<string, []> var_34910_equation_0 = const()[name = tensor<string, []>("op_34910_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34910_cast = einsum(equation = var_34910_equation_0, values = (var_34752_cast, var_34669_cast))[name = tensor<string, []>("op_34910_cast")];
            tensor<fp16, []> var_34911_to_fp16 = const()[name = tensor<string, []>("op_34911_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2949_cast = mul(x = var_34910_cast, y = var_34911_to_fp16)[name = tensor<string, []>("aw_2949_cast")];
            tensor<string, []> var_34914_equation_0 = const()[name = tensor<string, []>("op_34914_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34914_cast = einsum(equation = var_34914_equation_0, values = (var_34756_cast, var_34673_cast))[name = tensor<string, []>("op_34914_cast")];
            tensor<fp16, []> var_34915_to_fp16 = const()[name = tensor<string, []>("op_34915_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2951_cast = mul(x = var_34914_cast, y = var_34915_to_fp16)[name = tensor<string, []>("aw_2951_cast")];
            tensor<string, []> var_34918_equation_0 = const()[name = tensor<string, []>("op_34918_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34918_cast = einsum(equation = var_34918_equation_0, values = (var_34760_cast, var_34677_cast))[name = tensor<string, []>("op_34918_cast")];
            tensor<fp16, []> var_34919_to_fp16 = const()[name = tensor<string, []>("op_34919_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2953_cast = mul(x = var_34918_cast, y = var_34919_to_fp16)[name = tensor<string, []>("aw_2953_cast")];
            tensor<string, []> var_34922_equation_0 = const()[name = tensor<string, []>("op_34922_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34922_cast = einsum(equation = var_34922_equation_0, values = (var_34764_cast, var_34681_cast))[name = tensor<string, []>("op_34922_cast")];
            tensor<fp16, []> var_34923_to_fp16 = const()[name = tensor<string, []>("op_34923_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2955_cast = mul(x = var_34922_cast, y = var_34923_to_fp16)[name = tensor<string, []>("aw_2955_cast")];
            tensor<string, []> var_34926_equation_0 = const()[name = tensor<string, []>("op_34926_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34926_cast = einsum(equation = var_34926_equation_0, values = (var_34768_cast, var_34685_cast))[name = tensor<string, []>("op_34926_cast")];
            tensor<fp16, []> var_34927_to_fp16 = const()[name = tensor<string, []>("op_34927_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2957_cast = mul(x = var_34926_cast, y = var_34927_to_fp16)[name = tensor<string, []>("aw_2957_cast")];
            tensor<string, []> var_34930_equation_0 = const()[name = tensor<string, []>("op_34930_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_34930_cast = einsum(equation = var_34930_equation_0, values = (var_34772_cast, var_34689_cast))[name = tensor<string, []>("op_34930_cast")];
            tensor<fp16, []> var_34931_to_fp16 = const()[name = tensor<string, []>("op_34931_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_2959_cast = mul(x = var_34930_cast, y = var_34931_to_fp16)[name = tensor<string, []>("aw_2959_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34933_cast = softmax(axis = var_30385, x = aw_2921_cast)[name = tensor<string, []>("op_34933_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34934_cast = softmax(axis = var_30385, x = aw_2923_cast)[name = tensor<string, []>("op_34934_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34935_cast = softmax(axis = var_30385, x = aw_2925_cast)[name = tensor<string, []>("op_34935_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34936_cast = softmax(axis = var_30385, x = aw_2927_cast)[name = tensor<string, []>("op_34936_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34937_cast = softmax(axis = var_30385, x = aw_2929_cast)[name = tensor<string, []>("op_34937_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34938_cast = softmax(axis = var_30385, x = aw_2931_cast)[name = tensor<string, []>("op_34938_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34939_cast = softmax(axis = var_30385, x = aw_2933_cast)[name = tensor<string, []>("op_34939_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34940_cast = softmax(axis = var_30385, x = aw_2935_cast)[name = tensor<string, []>("op_34940_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34941_cast = softmax(axis = var_30385, x = aw_2937_cast)[name = tensor<string, []>("op_34941_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34942_cast = softmax(axis = var_30385, x = aw_2939_cast)[name = tensor<string, []>("op_34942_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34943_cast = softmax(axis = var_30385, x = aw_2941_cast)[name = tensor<string, []>("op_34943_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34944_cast = softmax(axis = var_30385, x = aw_2943_cast)[name = tensor<string, []>("op_34944_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34945_cast = softmax(axis = var_30385, x = aw_2945_cast)[name = tensor<string, []>("op_34945_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34946_cast = softmax(axis = var_30385, x = aw_2947_cast)[name = tensor<string, []>("op_34946_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34947_cast = softmax(axis = var_30385, x = aw_2949_cast)[name = tensor<string, []>("op_34947_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34948_cast = softmax(axis = var_30385, x = aw_2951_cast)[name = tensor<string, []>("op_34948_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34949_cast = softmax(axis = var_30385, x = aw_2953_cast)[name = tensor<string, []>("op_34949_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34950_cast = softmax(axis = var_30385, x = aw_2955_cast)[name = tensor<string, []>("op_34950_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34951_cast = softmax(axis = var_30385, x = aw_2957_cast)[name = tensor<string, []>("op_34951_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_34952_cast = softmax(axis = var_30385, x = aw_2959_cast)[name = tensor<string, []>("op_34952_cast")];
            tensor<string, []> var_34954_equation_0 = const()[name = tensor<string, []>("op_34954_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34954_cast = einsum(equation = var_34954_equation_0, values = (var_34774_cast, var_34933_cast))[name = tensor<string, []>("op_34954_cast")];
            tensor<string, []> var_34956_equation_0 = const()[name = tensor<string, []>("op_34956_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34956_cast = einsum(equation = var_34956_equation_0, values = (var_34778_cast, var_34934_cast))[name = tensor<string, []>("op_34956_cast")];
            tensor<string, []> var_34958_equation_0 = const()[name = tensor<string, []>("op_34958_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34958_cast = einsum(equation = var_34958_equation_0, values = (var_34782_cast, var_34935_cast))[name = tensor<string, []>("op_34958_cast")];
            tensor<string, []> var_34960_equation_0 = const()[name = tensor<string, []>("op_34960_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34960_cast = einsum(equation = var_34960_equation_0, values = (var_34786_cast, var_34936_cast))[name = tensor<string, []>("op_34960_cast")];
            tensor<string, []> var_34962_equation_0 = const()[name = tensor<string, []>("op_34962_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34962_cast = einsum(equation = var_34962_equation_0, values = (var_34790_cast, var_34937_cast))[name = tensor<string, []>("op_34962_cast")];
            tensor<string, []> var_34964_equation_0 = const()[name = tensor<string, []>("op_34964_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34964_cast = einsum(equation = var_34964_equation_0, values = (var_34794_cast, var_34938_cast))[name = tensor<string, []>("op_34964_cast")];
            tensor<string, []> var_34966_equation_0 = const()[name = tensor<string, []>("op_34966_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34966_cast = einsum(equation = var_34966_equation_0, values = (var_34798_cast, var_34939_cast))[name = tensor<string, []>("op_34966_cast")];
            tensor<string, []> var_34968_equation_0 = const()[name = tensor<string, []>("op_34968_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34968_cast = einsum(equation = var_34968_equation_0, values = (var_34802_cast, var_34940_cast))[name = tensor<string, []>("op_34968_cast")];
            tensor<string, []> var_34970_equation_0 = const()[name = tensor<string, []>("op_34970_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34970_cast = einsum(equation = var_34970_equation_0, values = (var_34806_cast, var_34941_cast))[name = tensor<string, []>("op_34970_cast")];
            tensor<string, []> var_34972_equation_0 = const()[name = tensor<string, []>("op_34972_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34972_cast = einsum(equation = var_34972_equation_0, values = (var_34810_cast, var_34942_cast))[name = tensor<string, []>("op_34972_cast")];
            tensor<string, []> var_34974_equation_0 = const()[name = tensor<string, []>("op_34974_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34974_cast = einsum(equation = var_34974_equation_0, values = (var_34814_cast, var_34943_cast))[name = tensor<string, []>("op_34974_cast")];
            tensor<string, []> var_34976_equation_0 = const()[name = tensor<string, []>("op_34976_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34976_cast = einsum(equation = var_34976_equation_0, values = (var_34818_cast, var_34944_cast))[name = tensor<string, []>("op_34976_cast")];
            tensor<string, []> var_34978_equation_0 = const()[name = tensor<string, []>("op_34978_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34978_cast = einsum(equation = var_34978_equation_0, values = (var_34822_cast, var_34945_cast))[name = tensor<string, []>("op_34978_cast")];
            tensor<string, []> var_34980_equation_0 = const()[name = tensor<string, []>("op_34980_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34980_cast = einsum(equation = var_34980_equation_0, values = (var_34826_cast, var_34946_cast))[name = tensor<string, []>("op_34980_cast")];
            tensor<string, []> var_34982_equation_0 = const()[name = tensor<string, []>("op_34982_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34982_cast = einsum(equation = var_34982_equation_0, values = (var_34830_cast, var_34947_cast))[name = tensor<string, []>("op_34982_cast")];
            tensor<string, []> var_34984_equation_0 = const()[name = tensor<string, []>("op_34984_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34984_cast = einsum(equation = var_34984_equation_0, values = (var_34834_cast, var_34948_cast))[name = tensor<string, []>("op_34984_cast")];
            tensor<string, []> var_34986_equation_0 = const()[name = tensor<string, []>("op_34986_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34986_cast = einsum(equation = var_34986_equation_0, values = (var_34838_cast, var_34949_cast))[name = tensor<string, []>("op_34986_cast")];
            tensor<string, []> var_34988_equation_0 = const()[name = tensor<string, []>("op_34988_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34988_cast = einsum(equation = var_34988_equation_0, values = (var_34842_cast, var_34950_cast))[name = tensor<string, []>("op_34988_cast")];
            tensor<string, []> var_34990_equation_0 = const()[name = tensor<string, []>("op_34990_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34990_cast = einsum(equation = var_34990_equation_0, values = (var_34846_cast, var_34951_cast))[name = tensor<string, []>("op_34990_cast")];
            tensor<string, []> var_34992_equation_0 = const()[name = tensor<string, []>("op_34992_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_34992_cast = einsum(equation = var_34992_equation_0, values = (var_34850_cast, var_34952_cast))[name = tensor<string, []>("op_34992_cast")];
            tensor<bool, []> input_475_interleave_0 = const()[name = tensor<string, []>("input_475_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_475_cast = concat(axis = var_30385, interleave = input_475_interleave_0, values = (var_34954_cast, var_34956_cast, var_34958_cast, var_34960_cast, var_34962_cast, var_34964_cast, var_34966_cast, var_34968_cast, var_34970_cast, var_34972_cast, var_34974_cast, var_34976_cast, var_34978_cast, var_34980_cast, var_34982_cast, var_34984_cast, var_34986_cast, var_34988_cast, var_34990_cast, var_34992_cast))[name = tensor<string, []>("input_475_cast")];
            tensor<int32, [2]> var_34998 = const()[name = tensor<string, []>("op_34998"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_35000 = const()[name = tensor<string, []>("op_35000"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_35002_pad_type_0 = const()[name = tensor<string, []>("op_35002_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_35002_pad_0 = const()[name = tensor<string, []>("op_35002_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2893356544)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2896633408)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_35002_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_out_0_bias_to_fp16, dilations = var_35000, groups = var_30385, pad = var_35002_pad_0, pad_type = var_35002_pad_type_0, strides = var_34998, weight = up_blocks_0_attentions_0_transformer_blocks_4_attn2_to_out_0_weight_to_fp16, x = input_475_cast)[name = tensor<string, []>("op_35002_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_233_cast = add(x = var_35002_cast, y = inputs_231_cast)[name = tensor<string, []>("inputs_233_cast")];
            tensor<int32, [1]> var_35006 = const()[name = tensor<string, []>("op_35006"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_233_cast = reduce_mean(axes = var_35006, keep_dims = var_30380, x = inputs_233_cast)[name = tensor<string, []>("channels_mean_233_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_233_cast = sub(x = inputs_233_cast, y = channels_mean_233_cast)[name = tensor<string, []>("zero_mean_233_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_233_cast = mul(x = zero_mean_233_cast, y = zero_mean_233_cast)[name = tensor<string, []>("zero_mean_sq_233_cast")];
            tensor<int32, [1]> var_35010 = const()[name = tensor<string, []>("op_35010"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_35011_cast = reduce_mean(axes = var_35010, keep_dims = var_30380, x = zero_mean_sq_233_cast)[name = tensor<string, []>("op_35011_cast")];
            tensor<fp16, []> var_35012_to_fp16 = const()[name = tensor<string, []>("op_35012_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_35013_cast = add(x = var_35011_cast, y = var_35012_to_fp16)[name = tensor<string, []>("op_35013_cast")];
            tensor<fp16, []> denom_233_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_233_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_233_cast = rsqrt(epsilon = denom_233_epsilon_0_to_fp16, x = var_35013_cast)[name = tensor<string, []>("denom_233_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_233_cast = mul(x = zero_mean_233_cast, y = denom_233_cast)[name = tensor<string, []>("out_233_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_35017_to_fp16 = const()[name = tensor<string, []>("op_35017_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2896636032)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_35018_cast = add(x = out_233_cast, y = var_35017_to_fp16)[name = tensor<string, []>("op_35018_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_35020_to_fp16 = const()[name = tensor<string, []>("op_35020_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2896638656)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_477_cast = mul(x = var_35018_cast, y = var_35020_to_fp16)[name = tensor<string, []>("input_477_cast")];
            tensor<int32, [2]> var_35028 = const()[name = tensor<string, []>("op_35028"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_35030 = const()[name = tensor<string, []>("op_35030"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_35032_pad_type_0 = const()[name = tensor<string, []>("op_35032_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_35032_pad_0 = const()[name = tensor<string, []>("op_35032_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2896641280)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2922855744)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_35032_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_4_ff_net_0_proj_bias_to_fp16, dilations = var_35030, groups = var_30385, pad = var_35032_pad_0, pad_type = var_35032_pad_type_0, strides = var_35028, weight = up_blocks_0_attentions_0_transformer_blocks_4_ff_net_0_proj_weight_to_fp16, x = input_477_cast)[name = tensor<string, []>("op_35032_cast")];
            tensor<int32, [2]> var_35033_split_sizes_0 = const()[name = tensor<string, []>("op_35033_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_35033_axis_0 = const()[name = tensor<string, []>("op_35033_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_35033_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_35033_cast_1 = split(axis = var_35033_axis_0, split_sizes = var_35033_split_sizes_0, x = var_35032_cast)[name = tensor<string, []>("op_35033_cast")];
            tensor<string, []> var_35035_mode_0 = const()[name = tensor<string, []>("op_35035_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_35035_cast = gelu(mode = var_35035_mode_0, x = var_35033_cast_1)[name = tensor<string, []>("op_35035_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_479_cast = mul(x = var_35033_cast_0, y = var_35035_cast)[name = tensor<string, []>("input_479_cast")];
            tensor<int32, [2]> var_35039 = const()[name = tensor<string, []>("op_35039"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_35041 = const()[name = tensor<string, []>("op_35041"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_35043_pad_type_0 = const()[name = tensor<string, []>("op_35043_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_35043_pad_0 = const()[name = tensor<string, []>("op_35043_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2922876288)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2935983552)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_35043_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_4_ff_net_2_bias_to_fp16, dilations = var_35041, groups = var_30385, pad = var_35043_pad_0, pad_type = var_35043_pad_type_0, strides = var_35039, weight = up_blocks_0_attentions_0_transformer_blocks_4_ff_net_2_weight_to_fp16, x = input_479_cast)[name = tensor<string, []>("op_35043_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_235_cast = add(x = var_35043_cast, y = inputs_233_cast)[name = tensor<string, []>("inputs_235_cast")];
            tensor<int32, [1]> var_35053 = const()[name = tensor<string, []>("op_35053"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_235_cast = reduce_mean(axes = var_35053, keep_dims = var_30380, x = inputs_235_cast)[name = tensor<string, []>("channels_mean_235_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_235_cast = sub(x = inputs_235_cast, y = channels_mean_235_cast)[name = tensor<string, []>("zero_mean_235_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_235_cast = mul(x = zero_mean_235_cast, y = zero_mean_235_cast)[name = tensor<string, []>("zero_mean_sq_235_cast")];
            tensor<int32, [1]> var_35057 = const()[name = tensor<string, []>("op_35057"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_35058_cast = reduce_mean(axes = var_35057, keep_dims = var_30380, x = zero_mean_sq_235_cast)[name = tensor<string, []>("op_35058_cast")];
            tensor<fp16, []> var_35059_to_fp16 = const()[name = tensor<string, []>("op_35059_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_35060_cast = add(x = var_35058_cast, y = var_35059_to_fp16)[name = tensor<string, []>("op_35060_cast")];
            tensor<fp16, []> denom_235_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_235_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_235_cast = rsqrt(epsilon = denom_235_epsilon_0_to_fp16, x = var_35060_cast)[name = tensor<string, []>("denom_235_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_235_cast = mul(x = zero_mean_235_cast, y = denom_235_cast)[name = tensor<string, []>("out_235_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_35064_to_fp16 = const()[name = tensor<string, []>("op_35064_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2935986176)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_35065_cast = add(x = out_235_cast, y = var_35064_to_fp16)[name = tensor<string, []>("op_35065_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_35067_to_fp16 = const()[name = tensor<string, []>("op_35067_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2935988800)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_317_cast = mul(x = var_35065_cast, y = var_35067_to_fp16)[name = tensor<string, []>("hidden_states_317_cast")];
            tensor<int32, [2]> var_35074 = const()[name = tensor<string, []>("op_35074"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_35076 = const()[name = tensor<string, []>("op_35076"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_157_pad_type_0 = const()[name = tensor<string, []>("q_157_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_157_pad_0 = const()[name = tensor<string, []>("q_157_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2935991424)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_157_cast = conv(dilations = var_35076, groups = var_30385, pad = q_157_pad_0, pad_type = q_157_pad_type_0, strides = var_35074, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_q_weight_to_fp16, x = hidden_states_317_cast)[name = tensor<string, []>("q_157_cast")];
            tensor<int32, [2]> var_35080 = const()[name = tensor<string, []>("op_35080"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_35082 = const()[name = tensor<string, []>("op_35082"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_313_pad_type_0 = const()[name = tensor<string, []>("k_313_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_313_pad_0 = const()[name = tensor<string, []>("k_313_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2939268288)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_313_cast = conv(dilations = var_35082, groups = var_30385, pad = k_313_pad_0, pad_type = k_313_pad_type_0, strides = var_35080, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_k_weight_to_fp16, x = hidden_states_317_cast)[name = tensor<string, []>("k_313_cast")];
            tensor<int32, [2]> var_35086 = const()[name = tensor<string, []>("op_35086"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_35088 = const()[name = tensor<string, []>("op_35088"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_157_pad_type_0 = const()[name = tensor<string, []>("v_157_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_157_pad_0 = const()[name = tensor<string, []>("v_157_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2942545152)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_157_cast = conv(dilations = var_35088, groups = var_30385, pad = v_157_pad_0, pad_type = v_157_pad_type_0, strides = var_35086, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_v_weight_to_fp16, x = hidden_states_317_cast)[name = tensor<string, []>("v_157_cast")];
            tensor<int32, [4]> var_35092_begin_0 = const()[name = tensor<string, []>("op_35092_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_35092_end_0 = const()[name = tensor<string, []>("op_35092_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_35092_end_mask_0 = const()[name = tensor<string, []>("op_35092_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35092_cast = slice_by_index(begin = var_35092_begin_0, end = var_35092_end_0, end_mask = var_35092_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35092_cast")];
            tensor<int32, [4]> var_35096_begin_0 = const()[name = tensor<string, []>("op_35096_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_35096_end_0 = const()[name = tensor<string, []>("op_35096_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_35096_end_mask_0 = const()[name = tensor<string, []>("op_35096_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35096_cast = slice_by_index(begin = var_35096_begin_0, end = var_35096_end_0, end_mask = var_35096_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35096_cast")];
            tensor<int32, [4]> var_35100_begin_0 = const()[name = tensor<string, []>("op_35100_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_35100_end_0 = const()[name = tensor<string, []>("op_35100_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_35100_end_mask_0 = const()[name = tensor<string, []>("op_35100_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35100_cast = slice_by_index(begin = var_35100_begin_0, end = var_35100_end_0, end_mask = var_35100_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35100_cast")];
            tensor<int32, [4]> var_35104_begin_0 = const()[name = tensor<string, []>("op_35104_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_35104_end_0 = const()[name = tensor<string, []>("op_35104_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_35104_end_mask_0 = const()[name = tensor<string, []>("op_35104_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35104_cast = slice_by_index(begin = var_35104_begin_0, end = var_35104_end_0, end_mask = var_35104_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35104_cast")];
            tensor<int32, [4]> var_35108_begin_0 = const()[name = tensor<string, []>("op_35108_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_35108_end_0 = const()[name = tensor<string, []>("op_35108_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_35108_end_mask_0 = const()[name = tensor<string, []>("op_35108_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35108_cast = slice_by_index(begin = var_35108_begin_0, end = var_35108_end_0, end_mask = var_35108_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35108_cast")];
            tensor<int32, [4]> var_35112_begin_0 = const()[name = tensor<string, []>("op_35112_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_35112_end_0 = const()[name = tensor<string, []>("op_35112_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_35112_end_mask_0 = const()[name = tensor<string, []>("op_35112_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35112_cast = slice_by_index(begin = var_35112_begin_0, end = var_35112_end_0, end_mask = var_35112_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35112_cast")];
            tensor<int32, [4]> var_35116_begin_0 = const()[name = tensor<string, []>("op_35116_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_35116_end_0 = const()[name = tensor<string, []>("op_35116_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_35116_end_mask_0 = const()[name = tensor<string, []>("op_35116_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35116_cast = slice_by_index(begin = var_35116_begin_0, end = var_35116_end_0, end_mask = var_35116_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35116_cast")];
            tensor<int32, [4]> var_35120_begin_0 = const()[name = tensor<string, []>("op_35120_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_35120_end_0 = const()[name = tensor<string, []>("op_35120_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_35120_end_mask_0 = const()[name = tensor<string, []>("op_35120_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35120_cast = slice_by_index(begin = var_35120_begin_0, end = var_35120_end_0, end_mask = var_35120_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35120_cast")];
            tensor<int32, [4]> var_35124_begin_0 = const()[name = tensor<string, []>("op_35124_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_35124_end_0 = const()[name = tensor<string, []>("op_35124_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_35124_end_mask_0 = const()[name = tensor<string, []>("op_35124_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35124_cast = slice_by_index(begin = var_35124_begin_0, end = var_35124_end_0, end_mask = var_35124_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35124_cast")];
            tensor<int32, [4]> var_35128_begin_0 = const()[name = tensor<string, []>("op_35128_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_35128_end_0 = const()[name = tensor<string, []>("op_35128_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_35128_end_mask_0 = const()[name = tensor<string, []>("op_35128_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35128_cast = slice_by_index(begin = var_35128_begin_0, end = var_35128_end_0, end_mask = var_35128_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35128_cast")];
            tensor<int32, [4]> var_35132_begin_0 = const()[name = tensor<string, []>("op_35132_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_35132_end_0 = const()[name = tensor<string, []>("op_35132_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_35132_end_mask_0 = const()[name = tensor<string, []>("op_35132_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35132_cast = slice_by_index(begin = var_35132_begin_0, end = var_35132_end_0, end_mask = var_35132_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35132_cast")];
            tensor<int32, [4]> var_35136_begin_0 = const()[name = tensor<string, []>("op_35136_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_35136_end_0 = const()[name = tensor<string, []>("op_35136_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_35136_end_mask_0 = const()[name = tensor<string, []>("op_35136_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35136_cast = slice_by_index(begin = var_35136_begin_0, end = var_35136_end_0, end_mask = var_35136_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35136_cast")];
            tensor<int32, [4]> var_35140_begin_0 = const()[name = tensor<string, []>("op_35140_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_35140_end_0 = const()[name = tensor<string, []>("op_35140_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_35140_end_mask_0 = const()[name = tensor<string, []>("op_35140_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35140_cast = slice_by_index(begin = var_35140_begin_0, end = var_35140_end_0, end_mask = var_35140_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35140_cast")];
            tensor<int32, [4]> var_35144_begin_0 = const()[name = tensor<string, []>("op_35144_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_35144_end_0 = const()[name = tensor<string, []>("op_35144_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_35144_end_mask_0 = const()[name = tensor<string, []>("op_35144_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35144_cast = slice_by_index(begin = var_35144_begin_0, end = var_35144_end_0, end_mask = var_35144_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35144_cast")];
            tensor<int32, [4]> var_35148_begin_0 = const()[name = tensor<string, []>("op_35148_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_35148_end_0 = const()[name = tensor<string, []>("op_35148_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_35148_end_mask_0 = const()[name = tensor<string, []>("op_35148_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35148_cast = slice_by_index(begin = var_35148_begin_0, end = var_35148_end_0, end_mask = var_35148_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35148_cast")];
            tensor<int32, [4]> var_35152_begin_0 = const()[name = tensor<string, []>("op_35152_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_35152_end_0 = const()[name = tensor<string, []>("op_35152_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_35152_end_mask_0 = const()[name = tensor<string, []>("op_35152_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35152_cast = slice_by_index(begin = var_35152_begin_0, end = var_35152_end_0, end_mask = var_35152_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35152_cast")];
            tensor<int32, [4]> var_35156_begin_0 = const()[name = tensor<string, []>("op_35156_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_35156_end_0 = const()[name = tensor<string, []>("op_35156_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_35156_end_mask_0 = const()[name = tensor<string, []>("op_35156_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35156_cast = slice_by_index(begin = var_35156_begin_0, end = var_35156_end_0, end_mask = var_35156_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35156_cast")];
            tensor<int32, [4]> var_35160_begin_0 = const()[name = tensor<string, []>("op_35160_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_35160_end_0 = const()[name = tensor<string, []>("op_35160_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_35160_end_mask_0 = const()[name = tensor<string, []>("op_35160_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35160_cast = slice_by_index(begin = var_35160_begin_0, end = var_35160_end_0, end_mask = var_35160_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35160_cast")];
            tensor<int32, [4]> var_35164_begin_0 = const()[name = tensor<string, []>("op_35164_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_35164_end_0 = const()[name = tensor<string, []>("op_35164_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_35164_end_mask_0 = const()[name = tensor<string, []>("op_35164_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35164_cast = slice_by_index(begin = var_35164_begin_0, end = var_35164_end_0, end_mask = var_35164_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35164_cast")];
            tensor<int32, [4]> var_35168_begin_0 = const()[name = tensor<string, []>("op_35168_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_35168_end_0 = const()[name = tensor<string, []>("op_35168_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_35168_end_mask_0 = const()[name = tensor<string, []>("op_35168_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35168_cast = slice_by_index(begin = var_35168_begin_0, end = var_35168_end_0, end_mask = var_35168_end_mask_0, x = q_157_cast)[name = tensor<string, []>("op_35168_cast")];
            tensor<int32, [4]> k_315_perm_0 = const()[name = tensor<string, []>("k_315_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_35175_begin_0 = const()[name = tensor<string, []>("op_35175_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_35175_end_0 = const()[name = tensor<string, []>("op_35175_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_35175_end_mask_0 = const()[name = tensor<string, []>("op_35175_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_61 = transpose(perm = k_315_perm_0, x = k_313_cast)[name = tensor<string, []>("transpose_61")];
            tensor<fp16, [2, 1024, 1, 64]> var_35175_cast = slice_by_index(begin = var_35175_begin_0, end = var_35175_end_0, end_mask = var_35175_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35175_cast")];
            tensor<int32, [4]> var_35179_begin_0 = const()[name = tensor<string, []>("op_35179_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_35179_end_0 = const()[name = tensor<string, []>("op_35179_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_35179_end_mask_0 = const()[name = tensor<string, []>("op_35179_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35179_cast = slice_by_index(begin = var_35179_begin_0, end = var_35179_end_0, end_mask = var_35179_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35179_cast")];
            tensor<int32, [4]> var_35183_begin_0 = const()[name = tensor<string, []>("op_35183_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_35183_end_0 = const()[name = tensor<string, []>("op_35183_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_35183_end_mask_0 = const()[name = tensor<string, []>("op_35183_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35183_cast = slice_by_index(begin = var_35183_begin_0, end = var_35183_end_0, end_mask = var_35183_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35183_cast")];
            tensor<int32, [4]> var_35187_begin_0 = const()[name = tensor<string, []>("op_35187_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_35187_end_0 = const()[name = tensor<string, []>("op_35187_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_35187_end_mask_0 = const()[name = tensor<string, []>("op_35187_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35187_cast = slice_by_index(begin = var_35187_begin_0, end = var_35187_end_0, end_mask = var_35187_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35187_cast")];
            tensor<int32, [4]> var_35191_begin_0 = const()[name = tensor<string, []>("op_35191_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_35191_end_0 = const()[name = tensor<string, []>("op_35191_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_35191_end_mask_0 = const()[name = tensor<string, []>("op_35191_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35191_cast = slice_by_index(begin = var_35191_begin_0, end = var_35191_end_0, end_mask = var_35191_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35191_cast")];
            tensor<int32, [4]> var_35195_begin_0 = const()[name = tensor<string, []>("op_35195_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_35195_end_0 = const()[name = tensor<string, []>("op_35195_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_35195_end_mask_0 = const()[name = tensor<string, []>("op_35195_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35195_cast = slice_by_index(begin = var_35195_begin_0, end = var_35195_end_0, end_mask = var_35195_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35195_cast")];
            tensor<int32, [4]> var_35199_begin_0 = const()[name = tensor<string, []>("op_35199_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_35199_end_0 = const()[name = tensor<string, []>("op_35199_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_35199_end_mask_0 = const()[name = tensor<string, []>("op_35199_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35199_cast = slice_by_index(begin = var_35199_begin_0, end = var_35199_end_0, end_mask = var_35199_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35199_cast")];
            tensor<int32, [4]> var_35203_begin_0 = const()[name = tensor<string, []>("op_35203_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_35203_end_0 = const()[name = tensor<string, []>("op_35203_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_35203_end_mask_0 = const()[name = tensor<string, []>("op_35203_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35203_cast = slice_by_index(begin = var_35203_begin_0, end = var_35203_end_0, end_mask = var_35203_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35203_cast")];
            tensor<int32, [4]> var_35207_begin_0 = const()[name = tensor<string, []>("op_35207_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_35207_end_0 = const()[name = tensor<string, []>("op_35207_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_35207_end_mask_0 = const()[name = tensor<string, []>("op_35207_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35207_cast = slice_by_index(begin = var_35207_begin_0, end = var_35207_end_0, end_mask = var_35207_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35207_cast")];
            tensor<int32, [4]> var_35211_begin_0 = const()[name = tensor<string, []>("op_35211_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_35211_end_0 = const()[name = tensor<string, []>("op_35211_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_35211_end_mask_0 = const()[name = tensor<string, []>("op_35211_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35211_cast = slice_by_index(begin = var_35211_begin_0, end = var_35211_end_0, end_mask = var_35211_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35211_cast")];
            tensor<int32, [4]> var_35215_begin_0 = const()[name = tensor<string, []>("op_35215_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_35215_end_0 = const()[name = tensor<string, []>("op_35215_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_35215_end_mask_0 = const()[name = tensor<string, []>("op_35215_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35215_cast = slice_by_index(begin = var_35215_begin_0, end = var_35215_end_0, end_mask = var_35215_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35215_cast")];
            tensor<int32, [4]> var_35219_begin_0 = const()[name = tensor<string, []>("op_35219_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_35219_end_0 = const()[name = tensor<string, []>("op_35219_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_35219_end_mask_0 = const()[name = tensor<string, []>("op_35219_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35219_cast = slice_by_index(begin = var_35219_begin_0, end = var_35219_end_0, end_mask = var_35219_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35219_cast")];
            tensor<int32, [4]> var_35223_begin_0 = const()[name = tensor<string, []>("op_35223_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_35223_end_0 = const()[name = tensor<string, []>("op_35223_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_35223_end_mask_0 = const()[name = tensor<string, []>("op_35223_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35223_cast = slice_by_index(begin = var_35223_begin_0, end = var_35223_end_0, end_mask = var_35223_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35223_cast")];
            tensor<int32, [4]> var_35227_begin_0 = const()[name = tensor<string, []>("op_35227_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_35227_end_0 = const()[name = tensor<string, []>("op_35227_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_35227_end_mask_0 = const()[name = tensor<string, []>("op_35227_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35227_cast = slice_by_index(begin = var_35227_begin_0, end = var_35227_end_0, end_mask = var_35227_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35227_cast")];
            tensor<int32, [4]> var_35231_begin_0 = const()[name = tensor<string, []>("op_35231_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_35231_end_0 = const()[name = tensor<string, []>("op_35231_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_35231_end_mask_0 = const()[name = tensor<string, []>("op_35231_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35231_cast = slice_by_index(begin = var_35231_begin_0, end = var_35231_end_0, end_mask = var_35231_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35231_cast")];
            tensor<int32, [4]> var_35235_begin_0 = const()[name = tensor<string, []>("op_35235_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_35235_end_0 = const()[name = tensor<string, []>("op_35235_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_35235_end_mask_0 = const()[name = tensor<string, []>("op_35235_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35235_cast = slice_by_index(begin = var_35235_begin_0, end = var_35235_end_0, end_mask = var_35235_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35235_cast")];
            tensor<int32, [4]> var_35239_begin_0 = const()[name = tensor<string, []>("op_35239_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_35239_end_0 = const()[name = tensor<string, []>("op_35239_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_35239_end_mask_0 = const()[name = tensor<string, []>("op_35239_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35239_cast = slice_by_index(begin = var_35239_begin_0, end = var_35239_end_0, end_mask = var_35239_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35239_cast")];
            tensor<int32, [4]> var_35243_begin_0 = const()[name = tensor<string, []>("op_35243_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_35243_end_0 = const()[name = tensor<string, []>("op_35243_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_35243_end_mask_0 = const()[name = tensor<string, []>("op_35243_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35243_cast = slice_by_index(begin = var_35243_begin_0, end = var_35243_end_0, end_mask = var_35243_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35243_cast")];
            tensor<int32, [4]> var_35247_begin_0 = const()[name = tensor<string, []>("op_35247_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_35247_end_0 = const()[name = tensor<string, []>("op_35247_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_35247_end_mask_0 = const()[name = tensor<string, []>("op_35247_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35247_cast = slice_by_index(begin = var_35247_begin_0, end = var_35247_end_0, end_mask = var_35247_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35247_cast")];
            tensor<int32, [4]> var_35251_begin_0 = const()[name = tensor<string, []>("op_35251_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_35251_end_0 = const()[name = tensor<string, []>("op_35251_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_35251_end_mask_0 = const()[name = tensor<string, []>("op_35251_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_35251_cast = slice_by_index(begin = var_35251_begin_0, end = var_35251_end_0, end_mask = var_35251_end_mask_0, x = transpose_61)[name = tensor<string, []>("op_35251_cast")];
            tensor<int32, [4]> var_35253_begin_0 = const()[name = tensor<string, []>("op_35253_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_35253_end_0 = const()[name = tensor<string, []>("op_35253_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_35253_end_mask_0 = const()[name = tensor<string, []>("op_35253_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35253_cast = slice_by_index(begin = var_35253_begin_0, end = var_35253_end_0, end_mask = var_35253_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35253_cast")];
            tensor<int32, [4]> var_35257_begin_0 = const()[name = tensor<string, []>("op_35257_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_35257_end_0 = const()[name = tensor<string, []>("op_35257_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_35257_end_mask_0 = const()[name = tensor<string, []>("op_35257_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35257_cast = slice_by_index(begin = var_35257_begin_0, end = var_35257_end_0, end_mask = var_35257_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35257_cast")];
            tensor<int32, [4]> var_35261_begin_0 = const()[name = tensor<string, []>("op_35261_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_35261_end_0 = const()[name = tensor<string, []>("op_35261_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_35261_end_mask_0 = const()[name = tensor<string, []>("op_35261_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35261_cast = slice_by_index(begin = var_35261_begin_0, end = var_35261_end_0, end_mask = var_35261_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35261_cast")];
            tensor<int32, [4]> var_35265_begin_0 = const()[name = tensor<string, []>("op_35265_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_35265_end_0 = const()[name = tensor<string, []>("op_35265_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_35265_end_mask_0 = const()[name = tensor<string, []>("op_35265_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35265_cast = slice_by_index(begin = var_35265_begin_0, end = var_35265_end_0, end_mask = var_35265_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35265_cast")];
            tensor<int32, [4]> var_35269_begin_0 = const()[name = tensor<string, []>("op_35269_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_35269_end_0 = const()[name = tensor<string, []>("op_35269_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_35269_end_mask_0 = const()[name = tensor<string, []>("op_35269_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35269_cast = slice_by_index(begin = var_35269_begin_0, end = var_35269_end_0, end_mask = var_35269_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35269_cast")];
            tensor<int32, [4]> var_35273_begin_0 = const()[name = tensor<string, []>("op_35273_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_35273_end_0 = const()[name = tensor<string, []>("op_35273_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_35273_end_mask_0 = const()[name = tensor<string, []>("op_35273_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35273_cast = slice_by_index(begin = var_35273_begin_0, end = var_35273_end_0, end_mask = var_35273_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35273_cast")];
            tensor<int32, [4]> var_35277_begin_0 = const()[name = tensor<string, []>("op_35277_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_35277_end_0 = const()[name = tensor<string, []>("op_35277_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_35277_end_mask_0 = const()[name = tensor<string, []>("op_35277_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35277_cast = slice_by_index(begin = var_35277_begin_0, end = var_35277_end_0, end_mask = var_35277_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35277_cast")];
            tensor<int32, [4]> var_35281_begin_0 = const()[name = tensor<string, []>("op_35281_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_35281_end_0 = const()[name = tensor<string, []>("op_35281_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_35281_end_mask_0 = const()[name = tensor<string, []>("op_35281_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35281_cast = slice_by_index(begin = var_35281_begin_0, end = var_35281_end_0, end_mask = var_35281_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35281_cast")];
            tensor<int32, [4]> var_35285_begin_0 = const()[name = tensor<string, []>("op_35285_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_35285_end_0 = const()[name = tensor<string, []>("op_35285_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_35285_end_mask_0 = const()[name = tensor<string, []>("op_35285_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35285_cast = slice_by_index(begin = var_35285_begin_0, end = var_35285_end_0, end_mask = var_35285_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35285_cast")];
            tensor<int32, [4]> var_35289_begin_0 = const()[name = tensor<string, []>("op_35289_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_35289_end_0 = const()[name = tensor<string, []>("op_35289_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_35289_end_mask_0 = const()[name = tensor<string, []>("op_35289_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35289_cast = slice_by_index(begin = var_35289_begin_0, end = var_35289_end_0, end_mask = var_35289_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35289_cast")];
            tensor<int32, [4]> var_35293_begin_0 = const()[name = tensor<string, []>("op_35293_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_35293_end_0 = const()[name = tensor<string, []>("op_35293_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_35293_end_mask_0 = const()[name = tensor<string, []>("op_35293_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35293_cast = slice_by_index(begin = var_35293_begin_0, end = var_35293_end_0, end_mask = var_35293_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35293_cast")];
            tensor<int32, [4]> var_35297_begin_0 = const()[name = tensor<string, []>("op_35297_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_35297_end_0 = const()[name = tensor<string, []>("op_35297_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_35297_end_mask_0 = const()[name = tensor<string, []>("op_35297_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35297_cast = slice_by_index(begin = var_35297_begin_0, end = var_35297_end_0, end_mask = var_35297_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35297_cast")];
            tensor<int32, [4]> var_35301_begin_0 = const()[name = tensor<string, []>("op_35301_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_35301_end_0 = const()[name = tensor<string, []>("op_35301_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_35301_end_mask_0 = const()[name = tensor<string, []>("op_35301_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35301_cast = slice_by_index(begin = var_35301_begin_0, end = var_35301_end_0, end_mask = var_35301_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35301_cast")];
            tensor<int32, [4]> var_35305_begin_0 = const()[name = tensor<string, []>("op_35305_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_35305_end_0 = const()[name = tensor<string, []>("op_35305_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_35305_end_mask_0 = const()[name = tensor<string, []>("op_35305_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35305_cast = slice_by_index(begin = var_35305_begin_0, end = var_35305_end_0, end_mask = var_35305_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35305_cast")];
            tensor<int32, [4]> var_35309_begin_0 = const()[name = tensor<string, []>("op_35309_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_35309_end_0 = const()[name = tensor<string, []>("op_35309_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_35309_end_mask_0 = const()[name = tensor<string, []>("op_35309_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35309_cast = slice_by_index(begin = var_35309_begin_0, end = var_35309_end_0, end_mask = var_35309_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35309_cast")];
            tensor<int32, [4]> var_35313_begin_0 = const()[name = tensor<string, []>("op_35313_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_35313_end_0 = const()[name = tensor<string, []>("op_35313_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_35313_end_mask_0 = const()[name = tensor<string, []>("op_35313_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35313_cast = slice_by_index(begin = var_35313_begin_0, end = var_35313_end_0, end_mask = var_35313_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35313_cast")];
            tensor<int32, [4]> var_35317_begin_0 = const()[name = tensor<string, []>("op_35317_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_35317_end_0 = const()[name = tensor<string, []>("op_35317_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_35317_end_mask_0 = const()[name = tensor<string, []>("op_35317_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35317_cast = slice_by_index(begin = var_35317_begin_0, end = var_35317_end_0, end_mask = var_35317_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35317_cast")];
            tensor<int32, [4]> var_35321_begin_0 = const()[name = tensor<string, []>("op_35321_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_35321_end_0 = const()[name = tensor<string, []>("op_35321_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_35321_end_mask_0 = const()[name = tensor<string, []>("op_35321_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35321_cast = slice_by_index(begin = var_35321_begin_0, end = var_35321_end_0, end_mask = var_35321_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35321_cast")];
            tensor<int32, [4]> var_35325_begin_0 = const()[name = tensor<string, []>("op_35325_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_35325_end_0 = const()[name = tensor<string, []>("op_35325_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_35325_end_mask_0 = const()[name = tensor<string, []>("op_35325_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35325_cast = slice_by_index(begin = var_35325_begin_0, end = var_35325_end_0, end_mask = var_35325_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35325_cast")];
            tensor<int32, [4]> var_35329_begin_0 = const()[name = tensor<string, []>("op_35329_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_35329_end_0 = const()[name = tensor<string, []>("op_35329_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_35329_end_mask_0 = const()[name = tensor<string, []>("op_35329_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35329_cast = slice_by_index(begin = var_35329_begin_0, end = var_35329_end_0, end_mask = var_35329_end_mask_0, x = v_157_cast)[name = tensor<string, []>("op_35329_cast")];
            tensor<string, []> var_35333_equation_0 = const()[name = tensor<string, []>("op_35333_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35333_cast = einsum(equation = var_35333_equation_0, values = (var_35175_cast, var_35092_cast))[name = tensor<string, []>("op_35333_cast")];
            tensor<fp16, []> var_35334_to_fp16 = const()[name = tensor<string, []>("op_35334_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2961_cast = mul(x = var_35333_cast, y = var_35334_to_fp16)[name = tensor<string, []>("aw_2961_cast")];
            tensor<string, []> var_35337_equation_0 = const()[name = tensor<string, []>("op_35337_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35337_cast = einsum(equation = var_35337_equation_0, values = (var_35179_cast, var_35096_cast))[name = tensor<string, []>("op_35337_cast")];
            tensor<fp16, []> var_35338_to_fp16 = const()[name = tensor<string, []>("op_35338_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2963_cast = mul(x = var_35337_cast, y = var_35338_to_fp16)[name = tensor<string, []>("aw_2963_cast")];
            tensor<string, []> var_35341_equation_0 = const()[name = tensor<string, []>("op_35341_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35341_cast = einsum(equation = var_35341_equation_0, values = (var_35183_cast, var_35100_cast))[name = tensor<string, []>("op_35341_cast")];
            tensor<fp16, []> var_35342_to_fp16 = const()[name = tensor<string, []>("op_35342_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2965_cast = mul(x = var_35341_cast, y = var_35342_to_fp16)[name = tensor<string, []>("aw_2965_cast")];
            tensor<string, []> var_35345_equation_0 = const()[name = tensor<string, []>("op_35345_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35345_cast = einsum(equation = var_35345_equation_0, values = (var_35187_cast, var_35104_cast))[name = tensor<string, []>("op_35345_cast")];
            tensor<fp16, []> var_35346_to_fp16 = const()[name = tensor<string, []>("op_35346_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2967_cast = mul(x = var_35345_cast, y = var_35346_to_fp16)[name = tensor<string, []>("aw_2967_cast")];
            tensor<string, []> var_35349_equation_0 = const()[name = tensor<string, []>("op_35349_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35349_cast = einsum(equation = var_35349_equation_0, values = (var_35191_cast, var_35108_cast))[name = tensor<string, []>("op_35349_cast")];
            tensor<fp16, []> var_35350_to_fp16 = const()[name = tensor<string, []>("op_35350_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2969_cast = mul(x = var_35349_cast, y = var_35350_to_fp16)[name = tensor<string, []>("aw_2969_cast")];
            tensor<string, []> var_35353_equation_0 = const()[name = tensor<string, []>("op_35353_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35353_cast = einsum(equation = var_35353_equation_0, values = (var_35195_cast, var_35112_cast))[name = tensor<string, []>("op_35353_cast")];
            tensor<fp16, []> var_35354_to_fp16 = const()[name = tensor<string, []>("op_35354_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2971_cast = mul(x = var_35353_cast, y = var_35354_to_fp16)[name = tensor<string, []>("aw_2971_cast")];
            tensor<string, []> var_35357_equation_0 = const()[name = tensor<string, []>("op_35357_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35357_cast = einsum(equation = var_35357_equation_0, values = (var_35199_cast, var_35116_cast))[name = tensor<string, []>("op_35357_cast")];
            tensor<fp16, []> var_35358_to_fp16 = const()[name = tensor<string, []>("op_35358_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2973_cast = mul(x = var_35357_cast, y = var_35358_to_fp16)[name = tensor<string, []>("aw_2973_cast")];
            tensor<string, []> var_35361_equation_0 = const()[name = tensor<string, []>("op_35361_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35361_cast = einsum(equation = var_35361_equation_0, values = (var_35203_cast, var_35120_cast))[name = tensor<string, []>("op_35361_cast")];
            tensor<fp16, []> var_35362_to_fp16 = const()[name = tensor<string, []>("op_35362_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2975_cast = mul(x = var_35361_cast, y = var_35362_to_fp16)[name = tensor<string, []>("aw_2975_cast")];
            tensor<string, []> var_35365_equation_0 = const()[name = tensor<string, []>("op_35365_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35365_cast = einsum(equation = var_35365_equation_0, values = (var_35207_cast, var_35124_cast))[name = tensor<string, []>("op_35365_cast")];
            tensor<fp16, []> var_35366_to_fp16 = const()[name = tensor<string, []>("op_35366_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2977_cast = mul(x = var_35365_cast, y = var_35366_to_fp16)[name = tensor<string, []>("aw_2977_cast")];
            tensor<string, []> var_35369_equation_0 = const()[name = tensor<string, []>("op_35369_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35369_cast = einsum(equation = var_35369_equation_0, values = (var_35211_cast, var_35128_cast))[name = tensor<string, []>("op_35369_cast")];
            tensor<fp16, []> var_35370_to_fp16 = const()[name = tensor<string, []>("op_35370_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2979_cast = mul(x = var_35369_cast, y = var_35370_to_fp16)[name = tensor<string, []>("aw_2979_cast")];
            tensor<string, []> var_35373_equation_0 = const()[name = tensor<string, []>("op_35373_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35373_cast = einsum(equation = var_35373_equation_0, values = (var_35215_cast, var_35132_cast))[name = tensor<string, []>("op_35373_cast")];
            tensor<fp16, []> var_35374_to_fp16 = const()[name = tensor<string, []>("op_35374_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2981_cast = mul(x = var_35373_cast, y = var_35374_to_fp16)[name = tensor<string, []>("aw_2981_cast")];
            tensor<string, []> var_35377_equation_0 = const()[name = tensor<string, []>("op_35377_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35377_cast = einsum(equation = var_35377_equation_0, values = (var_35219_cast, var_35136_cast))[name = tensor<string, []>("op_35377_cast")];
            tensor<fp16, []> var_35378_to_fp16 = const()[name = tensor<string, []>("op_35378_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2983_cast = mul(x = var_35377_cast, y = var_35378_to_fp16)[name = tensor<string, []>("aw_2983_cast")];
            tensor<string, []> var_35381_equation_0 = const()[name = tensor<string, []>("op_35381_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35381_cast = einsum(equation = var_35381_equation_0, values = (var_35223_cast, var_35140_cast))[name = tensor<string, []>("op_35381_cast")];
            tensor<fp16, []> var_35382_to_fp16 = const()[name = tensor<string, []>("op_35382_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2985_cast = mul(x = var_35381_cast, y = var_35382_to_fp16)[name = tensor<string, []>("aw_2985_cast")];
            tensor<string, []> var_35385_equation_0 = const()[name = tensor<string, []>("op_35385_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35385_cast = einsum(equation = var_35385_equation_0, values = (var_35227_cast, var_35144_cast))[name = tensor<string, []>("op_35385_cast")];
            tensor<fp16, []> var_35386_to_fp16 = const()[name = tensor<string, []>("op_35386_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2987_cast = mul(x = var_35385_cast, y = var_35386_to_fp16)[name = tensor<string, []>("aw_2987_cast")];
            tensor<string, []> var_35389_equation_0 = const()[name = tensor<string, []>("op_35389_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35389_cast = einsum(equation = var_35389_equation_0, values = (var_35231_cast, var_35148_cast))[name = tensor<string, []>("op_35389_cast")];
            tensor<fp16, []> var_35390_to_fp16 = const()[name = tensor<string, []>("op_35390_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2989_cast = mul(x = var_35389_cast, y = var_35390_to_fp16)[name = tensor<string, []>("aw_2989_cast")];
            tensor<string, []> var_35393_equation_0 = const()[name = tensor<string, []>("op_35393_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35393_cast = einsum(equation = var_35393_equation_0, values = (var_35235_cast, var_35152_cast))[name = tensor<string, []>("op_35393_cast")];
            tensor<fp16, []> var_35394_to_fp16 = const()[name = tensor<string, []>("op_35394_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2991_cast = mul(x = var_35393_cast, y = var_35394_to_fp16)[name = tensor<string, []>("aw_2991_cast")];
            tensor<string, []> var_35397_equation_0 = const()[name = tensor<string, []>("op_35397_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35397_cast = einsum(equation = var_35397_equation_0, values = (var_35239_cast, var_35156_cast))[name = tensor<string, []>("op_35397_cast")];
            tensor<fp16, []> var_35398_to_fp16 = const()[name = tensor<string, []>("op_35398_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2993_cast = mul(x = var_35397_cast, y = var_35398_to_fp16)[name = tensor<string, []>("aw_2993_cast")];
            tensor<string, []> var_35401_equation_0 = const()[name = tensor<string, []>("op_35401_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35401_cast = einsum(equation = var_35401_equation_0, values = (var_35243_cast, var_35160_cast))[name = tensor<string, []>("op_35401_cast")];
            tensor<fp16, []> var_35402_to_fp16 = const()[name = tensor<string, []>("op_35402_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2995_cast = mul(x = var_35401_cast, y = var_35402_to_fp16)[name = tensor<string, []>("aw_2995_cast")];
            tensor<string, []> var_35405_equation_0 = const()[name = tensor<string, []>("op_35405_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35405_cast = einsum(equation = var_35405_equation_0, values = (var_35247_cast, var_35164_cast))[name = tensor<string, []>("op_35405_cast")];
            tensor<fp16, []> var_35406_to_fp16 = const()[name = tensor<string, []>("op_35406_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2997_cast = mul(x = var_35405_cast, y = var_35406_to_fp16)[name = tensor<string, []>("aw_2997_cast")];
            tensor<string, []> var_35409_equation_0 = const()[name = tensor<string, []>("op_35409_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35409_cast = einsum(equation = var_35409_equation_0, values = (var_35251_cast, var_35168_cast))[name = tensor<string, []>("op_35409_cast")];
            tensor<fp16, []> var_35410_to_fp16 = const()[name = tensor<string, []>("op_35410_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_2999_cast = mul(x = var_35409_cast, y = var_35410_to_fp16)[name = tensor<string, []>("aw_2999_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35412_cast = softmax(axis = var_30385, x = aw_2961_cast)[name = tensor<string, []>("op_35412_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35413_cast = softmax(axis = var_30385, x = aw_2963_cast)[name = tensor<string, []>("op_35413_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35414_cast = softmax(axis = var_30385, x = aw_2965_cast)[name = tensor<string, []>("op_35414_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35415_cast = softmax(axis = var_30385, x = aw_2967_cast)[name = tensor<string, []>("op_35415_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35416_cast = softmax(axis = var_30385, x = aw_2969_cast)[name = tensor<string, []>("op_35416_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35417_cast = softmax(axis = var_30385, x = aw_2971_cast)[name = tensor<string, []>("op_35417_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35418_cast = softmax(axis = var_30385, x = aw_2973_cast)[name = tensor<string, []>("op_35418_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35419_cast = softmax(axis = var_30385, x = aw_2975_cast)[name = tensor<string, []>("op_35419_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35420_cast = softmax(axis = var_30385, x = aw_2977_cast)[name = tensor<string, []>("op_35420_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35421_cast = softmax(axis = var_30385, x = aw_2979_cast)[name = tensor<string, []>("op_35421_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35422_cast = softmax(axis = var_30385, x = aw_2981_cast)[name = tensor<string, []>("op_35422_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35423_cast = softmax(axis = var_30385, x = aw_2983_cast)[name = tensor<string, []>("op_35423_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35424_cast = softmax(axis = var_30385, x = aw_2985_cast)[name = tensor<string, []>("op_35424_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35425_cast = softmax(axis = var_30385, x = aw_2987_cast)[name = tensor<string, []>("op_35425_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35426_cast = softmax(axis = var_30385, x = aw_2989_cast)[name = tensor<string, []>("op_35426_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35427_cast = softmax(axis = var_30385, x = aw_2991_cast)[name = tensor<string, []>("op_35427_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35428_cast = softmax(axis = var_30385, x = aw_2993_cast)[name = tensor<string, []>("op_35428_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35429_cast = softmax(axis = var_30385, x = aw_2995_cast)[name = tensor<string, []>("op_35429_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35430_cast = softmax(axis = var_30385, x = aw_2997_cast)[name = tensor<string, []>("op_35430_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_35431_cast = softmax(axis = var_30385, x = aw_2999_cast)[name = tensor<string, []>("op_35431_cast")];
            tensor<string, []> var_35433_equation_0 = const()[name = tensor<string, []>("op_35433_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35433_cast = einsum(equation = var_35433_equation_0, values = (var_35253_cast, var_35412_cast))[name = tensor<string, []>("op_35433_cast")];
            tensor<string, []> var_35435_equation_0 = const()[name = tensor<string, []>("op_35435_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35435_cast = einsum(equation = var_35435_equation_0, values = (var_35257_cast, var_35413_cast))[name = tensor<string, []>("op_35435_cast")];
            tensor<string, []> var_35437_equation_0 = const()[name = tensor<string, []>("op_35437_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35437_cast = einsum(equation = var_35437_equation_0, values = (var_35261_cast, var_35414_cast))[name = tensor<string, []>("op_35437_cast")];
            tensor<string, []> var_35439_equation_0 = const()[name = tensor<string, []>("op_35439_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35439_cast = einsum(equation = var_35439_equation_0, values = (var_35265_cast, var_35415_cast))[name = tensor<string, []>("op_35439_cast")];
            tensor<string, []> var_35441_equation_0 = const()[name = tensor<string, []>("op_35441_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35441_cast = einsum(equation = var_35441_equation_0, values = (var_35269_cast, var_35416_cast))[name = tensor<string, []>("op_35441_cast")];
            tensor<string, []> var_35443_equation_0 = const()[name = tensor<string, []>("op_35443_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35443_cast = einsum(equation = var_35443_equation_0, values = (var_35273_cast, var_35417_cast))[name = tensor<string, []>("op_35443_cast")];
            tensor<string, []> var_35445_equation_0 = const()[name = tensor<string, []>("op_35445_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35445_cast = einsum(equation = var_35445_equation_0, values = (var_35277_cast, var_35418_cast))[name = tensor<string, []>("op_35445_cast")];
            tensor<string, []> var_35447_equation_0 = const()[name = tensor<string, []>("op_35447_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35447_cast = einsum(equation = var_35447_equation_0, values = (var_35281_cast, var_35419_cast))[name = tensor<string, []>("op_35447_cast")];
            tensor<string, []> var_35449_equation_0 = const()[name = tensor<string, []>("op_35449_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35449_cast = einsum(equation = var_35449_equation_0, values = (var_35285_cast, var_35420_cast))[name = tensor<string, []>("op_35449_cast")];
            tensor<string, []> var_35451_equation_0 = const()[name = tensor<string, []>("op_35451_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35451_cast = einsum(equation = var_35451_equation_0, values = (var_35289_cast, var_35421_cast))[name = tensor<string, []>("op_35451_cast")];
            tensor<string, []> var_35453_equation_0 = const()[name = tensor<string, []>("op_35453_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35453_cast = einsum(equation = var_35453_equation_0, values = (var_35293_cast, var_35422_cast))[name = tensor<string, []>("op_35453_cast")];
            tensor<string, []> var_35455_equation_0 = const()[name = tensor<string, []>("op_35455_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35455_cast = einsum(equation = var_35455_equation_0, values = (var_35297_cast, var_35423_cast))[name = tensor<string, []>("op_35455_cast")];
            tensor<string, []> var_35457_equation_0 = const()[name = tensor<string, []>("op_35457_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35457_cast = einsum(equation = var_35457_equation_0, values = (var_35301_cast, var_35424_cast))[name = tensor<string, []>("op_35457_cast")];
            tensor<string, []> var_35459_equation_0 = const()[name = tensor<string, []>("op_35459_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35459_cast = einsum(equation = var_35459_equation_0, values = (var_35305_cast, var_35425_cast))[name = tensor<string, []>("op_35459_cast")];
            tensor<string, []> var_35461_equation_0 = const()[name = tensor<string, []>("op_35461_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35461_cast = einsum(equation = var_35461_equation_0, values = (var_35309_cast, var_35426_cast))[name = tensor<string, []>("op_35461_cast")];
            tensor<string, []> var_35463_equation_0 = const()[name = tensor<string, []>("op_35463_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35463_cast = einsum(equation = var_35463_equation_0, values = (var_35313_cast, var_35427_cast))[name = tensor<string, []>("op_35463_cast")];
            tensor<string, []> var_35465_equation_0 = const()[name = tensor<string, []>("op_35465_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35465_cast = einsum(equation = var_35465_equation_0, values = (var_35317_cast, var_35428_cast))[name = tensor<string, []>("op_35465_cast")];
            tensor<string, []> var_35467_equation_0 = const()[name = tensor<string, []>("op_35467_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35467_cast = einsum(equation = var_35467_equation_0, values = (var_35321_cast, var_35429_cast))[name = tensor<string, []>("op_35467_cast")];
            tensor<string, []> var_35469_equation_0 = const()[name = tensor<string, []>("op_35469_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35469_cast = einsum(equation = var_35469_equation_0, values = (var_35325_cast, var_35430_cast))[name = tensor<string, []>("op_35469_cast")];
            tensor<string, []> var_35471_equation_0 = const()[name = tensor<string, []>("op_35471_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35471_cast = einsum(equation = var_35471_equation_0, values = (var_35329_cast, var_35431_cast))[name = tensor<string, []>("op_35471_cast")];
            tensor<bool, []> input_481_interleave_0 = const()[name = tensor<string, []>("input_481_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_481_cast = concat(axis = var_30385, interleave = input_481_interleave_0, values = (var_35433_cast, var_35435_cast, var_35437_cast, var_35439_cast, var_35441_cast, var_35443_cast, var_35445_cast, var_35447_cast, var_35449_cast, var_35451_cast, var_35453_cast, var_35455_cast, var_35457_cast, var_35459_cast, var_35461_cast, var_35463_cast, var_35465_cast, var_35467_cast, var_35469_cast, var_35471_cast))[name = tensor<string, []>("input_481_cast")];
            tensor<int32, [2]> var_35477 = const()[name = tensor<string, []>("op_35477"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_35479 = const()[name = tensor<string, []>("op_35479"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_35481_pad_type_0 = const()[name = tensor<string, []>("op_35481_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_35481_pad_0 = const()[name = tensor<string, []>("op_35481_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2945822016)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2949098880)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_35481_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_out_0_bias_to_fp16, dilations = var_35479, groups = var_30385, pad = var_35481_pad_0, pad_type = var_35481_pad_type_0, strides = var_35477, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn1_to_out_0_weight_to_fp16, x = input_481_cast)[name = tensor<string, []>("op_35481_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_237_cast = add(x = var_35481_cast, y = inputs_235_cast)[name = tensor<string, []>("inputs_237_cast")];
            tensor<int32, [1]> var_35485 = const()[name = tensor<string, []>("op_35485"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_237_cast = reduce_mean(axes = var_35485, keep_dims = var_30380, x = inputs_237_cast)[name = tensor<string, []>("channels_mean_237_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_237_cast = sub(x = inputs_237_cast, y = channels_mean_237_cast)[name = tensor<string, []>("zero_mean_237_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_237_cast = mul(x = zero_mean_237_cast, y = zero_mean_237_cast)[name = tensor<string, []>("zero_mean_sq_237_cast")];
            tensor<int32, [1]> var_35489 = const()[name = tensor<string, []>("op_35489"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_35490_cast = reduce_mean(axes = var_35489, keep_dims = var_30380, x = zero_mean_sq_237_cast)[name = tensor<string, []>("op_35490_cast")];
            tensor<fp16, []> var_35491_to_fp16 = const()[name = tensor<string, []>("op_35491_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_35492_cast = add(x = var_35490_cast, y = var_35491_to_fp16)[name = tensor<string, []>("op_35492_cast")];
            tensor<fp16, []> denom_237_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_237_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_237_cast = rsqrt(epsilon = denom_237_epsilon_0_to_fp16, x = var_35492_cast)[name = tensor<string, []>("denom_237_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_237_cast = mul(x = zero_mean_237_cast, y = denom_237_cast)[name = tensor<string, []>("out_237_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_35496_to_fp16 = const()[name = tensor<string, []>("op_35496_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2949101504)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_35497_cast = add(x = out_237_cast, y = var_35496_to_fp16)[name = tensor<string, []>("op_35497_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_35499_to_fp16 = const()[name = tensor<string, []>("op_35499_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2949104128)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_319_cast = mul(x = var_35497_cast, y = var_35499_to_fp16)[name = tensor<string, []>("hidden_states_319_cast")];
            tensor<int32, [2]> var_35506 = const()[name = tensor<string, []>("op_35506"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_35508 = const()[name = tensor<string, []>("op_35508"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_159_pad_type_0 = const()[name = tensor<string, []>("q_159_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_159_pad_0 = const()[name = tensor<string, []>("q_159_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2949106752)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_159_cast = conv(dilations = var_35508, groups = var_30385, pad = q_159_pad_0, pad_type = q_159_pad_type_0, strides = var_35506, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_q_weight_to_fp16, x = hidden_states_319_cast)[name = tensor<string, []>("q_159_cast")];
            tensor<int32, [2]> var_35512 = const()[name = tensor<string, []>("op_35512"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_35514 = const()[name = tensor<string, []>("op_35514"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_317_pad_type_0 = const()[name = tensor<string, []>("k_317_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_317_pad_0 = const()[name = tensor<string, []>("k_317_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2952383616)))];
            tensor<fp16, [2, 1280, 1, 77]> k_317_cast = conv(dilations = var_35514, groups = var_30385, pad = k_317_pad_0, pad_type = k_317_pad_type_0, strides = var_35512, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_317_cast")];
            tensor<int32, [2]> var_35518 = const()[name = tensor<string, []>("op_35518"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_35520 = const()[name = tensor<string, []>("op_35520"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_159_pad_type_0 = const()[name = tensor<string, []>("v_159_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_159_pad_0 = const()[name = tensor<string, []>("v_159_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2957626560)))];
            tensor<fp16, [2, 1280, 1, 77]> v_159_cast = conv(dilations = var_35520, groups = var_30385, pad = v_159_pad_0, pad_type = v_159_pad_type_0, strides = var_35518, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_159_cast")];
            tensor<int32, [4]> var_35524_begin_0 = const()[name = tensor<string, []>("op_35524_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_35524_end_0 = const()[name = tensor<string, []>("op_35524_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_35524_end_mask_0 = const()[name = tensor<string, []>("op_35524_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35524_cast = slice_by_index(begin = var_35524_begin_0, end = var_35524_end_0, end_mask = var_35524_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35524_cast")];
            tensor<int32, [4]> var_35528_begin_0 = const()[name = tensor<string, []>("op_35528_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_35528_end_0 = const()[name = tensor<string, []>("op_35528_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_35528_end_mask_0 = const()[name = tensor<string, []>("op_35528_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35528_cast = slice_by_index(begin = var_35528_begin_0, end = var_35528_end_0, end_mask = var_35528_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35528_cast")];
            tensor<int32, [4]> var_35532_begin_0 = const()[name = tensor<string, []>("op_35532_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_35532_end_0 = const()[name = tensor<string, []>("op_35532_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_35532_end_mask_0 = const()[name = tensor<string, []>("op_35532_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35532_cast = slice_by_index(begin = var_35532_begin_0, end = var_35532_end_0, end_mask = var_35532_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35532_cast")];
            tensor<int32, [4]> var_35536_begin_0 = const()[name = tensor<string, []>("op_35536_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_35536_end_0 = const()[name = tensor<string, []>("op_35536_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_35536_end_mask_0 = const()[name = tensor<string, []>("op_35536_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35536_cast = slice_by_index(begin = var_35536_begin_0, end = var_35536_end_0, end_mask = var_35536_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35536_cast")];
            tensor<int32, [4]> var_35540_begin_0 = const()[name = tensor<string, []>("op_35540_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_35540_end_0 = const()[name = tensor<string, []>("op_35540_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_35540_end_mask_0 = const()[name = tensor<string, []>("op_35540_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35540_cast = slice_by_index(begin = var_35540_begin_0, end = var_35540_end_0, end_mask = var_35540_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35540_cast")];
            tensor<int32, [4]> var_35544_begin_0 = const()[name = tensor<string, []>("op_35544_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_35544_end_0 = const()[name = tensor<string, []>("op_35544_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_35544_end_mask_0 = const()[name = tensor<string, []>("op_35544_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35544_cast = slice_by_index(begin = var_35544_begin_0, end = var_35544_end_0, end_mask = var_35544_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35544_cast")];
            tensor<int32, [4]> var_35548_begin_0 = const()[name = tensor<string, []>("op_35548_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_35548_end_0 = const()[name = tensor<string, []>("op_35548_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_35548_end_mask_0 = const()[name = tensor<string, []>("op_35548_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35548_cast = slice_by_index(begin = var_35548_begin_0, end = var_35548_end_0, end_mask = var_35548_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35548_cast")];
            tensor<int32, [4]> var_35552_begin_0 = const()[name = tensor<string, []>("op_35552_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_35552_end_0 = const()[name = tensor<string, []>("op_35552_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_35552_end_mask_0 = const()[name = tensor<string, []>("op_35552_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35552_cast = slice_by_index(begin = var_35552_begin_0, end = var_35552_end_0, end_mask = var_35552_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35552_cast")];
            tensor<int32, [4]> var_35556_begin_0 = const()[name = tensor<string, []>("op_35556_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_35556_end_0 = const()[name = tensor<string, []>("op_35556_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_35556_end_mask_0 = const()[name = tensor<string, []>("op_35556_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35556_cast = slice_by_index(begin = var_35556_begin_0, end = var_35556_end_0, end_mask = var_35556_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35556_cast")];
            tensor<int32, [4]> var_35560_begin_0 = const()[name = tensor<string, []>("op_35560_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_35560_end_0 = const()[name = tensor<string, []>("op_35560_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_35560_end_mask_0 = const()[name = tensor<string, []>("op_35560_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35560_cast = slice_by_index(begin = var_35560_begin_0, end = var_35560_end_0, end_mask = var_35560_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35560_cast")];
            tensor<int32, [4]> var_35564_begin_0 = const()[name = tensor<string, []>("op_35564_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_35564_end_0 = const()[name = tensor<string, []>("op_35564_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_35564_end_mask_0 = const()[name = tensor<string, []>("op_35564_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35564_cast = slice_by_index(begin = var_35564_begin_0, end = var_35564_end_0, end_mask = var_35564_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35564_cast")];
            tensor<int32, [4]> var_35568_begin_0 = const()[name = tensor<string, []>("op_35568_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_35568_end_0 = const()[name = tensor<string, []>("op_35568_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_35568_end_mask_0 = const()[name = tensor<string, []>("op_35568_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35568_cast = slice_by_index(begin = var_35568_begin_0, end = var_35568_end_0, end_mask = var_35568_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35568_cast")];
            tensor<int32, [4]> var_35572_begin_0 = const()[name = tensor<string, []>("op_35572_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_35572_end_0 = const()[name = tensor<string, []>("op_35572_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_35572_end_mask_0 = const()[name = tensor<string, []>("op_35572_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35572_cast = slice_by_index(begin = var_35572_begin_0, end = var_35572_end_0, end_mask = var_35572_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35572_cast")];
            tensor<int32, [4]> var_35576_begin_0 = const()[name = tensor<string, []>("op_35576_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_35576_end_0 = const()[name = tensor<string, []>("op_35576_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_35576_end_mask_0 = const()[name = tensor<string, []>("op_35576_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35576_cast = slice_by_index(begin = var_35576_begin_0, end = var_35576_end_0, end_mask = var_35576_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35576_cast")];
            tensor<int32, [4]> var_35580_begin_0 = const()[name = tensor<string, []>("op_35580_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_35580_end_0 = const()[name = tensor<string, []>("op_35580_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_35580_end_mask_0 = const()[name = tensor<string, []>("op_35580_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35580_cast = slice_by_index(begin = var_35580_begin_0, end = var_35580_end_0, end_mask = var_35580_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35580_cast")];
            tensor<int32, [4]> var_35584_begin_0 = const()[name = tensor<string, []>("op_35584_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_35584_end_0 = const()[name = tensor<string, []>("op_35584_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_35584_end_mask_0 = const()[name = tensor<string, []>("op_35584_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35584_cast = slice_by_index(begin = var_35584_begin_0, end = var_35584_end_0, end_mask = var_35584_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35584_cast")];
            tensor<int32, [4]> var_35588_begin_0 = const()[name = tensor<string, []>("op_35588_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_35588_end_0 = const()[name = tensor<string, []>("op_35588_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_35588_end_mask_0 = const()[name = tensor<string, []>("op_35588_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35588_cast = slice_by_index(begin = var_35588_begin_0, end = var_35588_end_0, end_mask = var_35588_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35588_cast")];
            tensor<int32, [4]> var_35592_begin_0 = const()[name = tensor<string, []>("op_35592_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_35592_end_0 = const()[name = tensor<string, []>("op_35592_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_35592_end_mask_0 = const()[name = tensor<string, []>("op_35592_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35592_cast = slice_by_index(begin = var_35592_begin_0, end = var_35592_end_0, end_mask = var_35592_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35592_cast")];
            tensor<int32, [4]> var_35596_begin_0 = const()[name = tensor<string, []>("op_35596_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_35596_end_0 = const()[name = tensor<string, []>("op_35596_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_35596_end_mask_0 = const()[name = tensor<string, []>("op_35596_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35596_cast = slice_by_index(begin = var_35596_begin_0, end = var_35596_end_0, end_mask = var_35596_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35596_cast")];
            tensor<int32, [4]> var_35600_begin_0 = const()[name = tensor<string, []>("op_35600_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_35600_end_0 = const()[name = tensor<string, []>("op_35600_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_35600_end_mask_0 = const()[name = tensor<string, []>("op_35600_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_35600_cast = slice_by_index(begin = var_35600_begin_0, end = var_35600_end_0, end_mask = var_35600_end_mask_0, x = q_159_cast)[name = tensor<string, []>("op_35600_cast")];
            tensor<int32, [4]> k_319_perm_0 = const()[name = tensor<string, []>("k_319_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_35607_begin_0 = const()[name = tensor<string, []>("op_35607_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_35607_end_0 = const()[name = tensor<string, []>("op_35607_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_35607_end_mask_0 = const()[name = tensor<string, []>("op_35607_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_60 = transpose(perm = k_319_perm_0, x = k_317_cast)[name = tensor<string, []>("transpose_60")];
            tensor<fp16, [2, 77, 1, 64]> var_35607_cast = slice_by_index(begin = var_35607_begin_0, end = var_35607_end_0, end_mask = var_35607_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35607_cast")];
            tensor<int32, [4]> var_35611_begin_0 = const()[name = tensor<string, []>("op_35611_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_35611_end_0 = const()[name = tensor<string, []>("op_35611_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_35611_end_mask_0 = const()[name = tensor<string, []>("op_35611_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35611_cast = slice_by_index(begin = var_35611_begin_0, end = var_35611_end_0, end_mask = var_35611_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35611_cast")];
            tensor<int32, [4]> var_35615_begin_0 = const()[name = tensor<string, []>("op_35615_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_35615_end_0 = const()[name = tensor<string, []>("op_35615_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_35615_end_mask_0 = const()[name = tensor<string, []>("op_35615_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35615_cast = slice_by_index(begin = var_35615_begin_0, end = var_35615_end_0, end_mask = var_35615_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35615_cast")];
            tensor<int32, [4]> var_35619_begin_0 = const()[name = tensor<string, []>("op_35619_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_35619_end_0 = const()[name = tensor<string, []>("op_35619_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_35619_end_mask_0 = const()[name = tensor<string, []>("op_35619_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35619_cast = slice_by_index(begin = var_35619_begin_0, end = var_35619_end_0, end_mask = var_35619_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35619_cast")];
            tensor<int32, [4]> var_35623_begin_0 = const()[name = tensor<string, []>("op_35623_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_35623_end_0 = const()[name = tensor<string, []>("op_35623_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_35623_end_mask_0 = const()[name = tensor<string, []>("op_35623_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35623_cast = slice_by_index(begin = var_35623_begin_0, end = var_35623_end_0, end_mask = var_35623_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35623_cast")];
            tensor<int32, [4]> var_35627_begin_0 = const()[name = tensor<string, []>("op_35627_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_35627_end_0 = const()[name = tensor<string, []>("op_35627_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_35627_end_mask_0 = const()[name = tensor<string, []>("op_35627_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35627_cast = slice_by_index(begin = var_35627_begin_0, end = var_35627_end_0, end_mask = var_35627_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35627_cast")];
            tensor<int32, [4]> var_35631_begin_0 = const()[name = tensor<string, []>("op_35631_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_35631_end_0 = const()[name = tensor<string, []>("op_35631_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_35631_end_mask_0 = const()[name = tensor<string, []>("op_35631_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35631_cast = slice_by_index(begin = var_35631_begin_0, end = var_35631_end_0, end_mask = var_35631_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35631_cast")];
            tensor<int32, [4]> var_35635_begin_0 = const()[name = tensor<string, []>("op_35635_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_35635_end_0 = const()[name = tensor<string, []>("op_35635_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_35635_end_mask_0 = const()[name = tensor<string, []>("op_35635_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35635_cast = slice_by_index(begin = var_35635_begin_0, end = var_35635_end_0, end_mask = var_35635_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35635_cast")];
            tensor<int32, [4]> var_35639_begin_0 = const()[name = tensor<string, []>("op_35639_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_35639_end_0 = const()[name = tensor<string, []>("op_35639_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_35639_end_mask_0 = const()[name = tensor<string, []>("op_35639_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35639_cast = slice_by_index(begin = var_35639_begin_0, end = var_35639_end_0, end_mask = var_35639_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35639_cast")];
            tensor<int32, [4]> var_35643_begin_0 = const()[name = tensor<string, []>("op_35643_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_35643_end_0 = const()[name = tensor<string, []>("op_35643_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_35643_end_mask_0 = const()[name = tensor<string, []>("op_35643_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35643_cast = slice_by_index(begin = var_35643_begin_0, end = var_35643_end_0, end_mask = var_35643_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35643_cast")];
            tensor<int32, [4]> var_35647_begin_0 = const()[name = tensor<string, []>("op_35647_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_35647_end_0 = const()[name = tensor<string, []>("op_35647_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_35647_end_mask_0 = const()[name = tensor<string, []>("op_35647_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35647_cast = slice_by_index(begin = var_35647_begin_0, end = var_35647_end_0, end_mask = var_35647_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35647_cast")];
            tensor<int32, [4]> var_35651_begin_0 = const()[name = tensor<string, []>("op_35651_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_35651_end_0 = const()[name = tensor<string, []>("op_35651_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_35651_end_mask_0 = const()[name = tensor<string, []>("op_35651_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35651_cast = slice_by_index(begin = var_35651_begin_0, end = var_35651_end_0, end_mask = var_35651_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35651_cast")];
            tensor<int32, [4]> var_35655_begin_0 = const()[name = tensor<string, []>("op_35655_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_35655_end_0 = const()[name = tensor<string, []>("op_35655_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_35655_end_mask_0 = const()[name = tensor<string, []>("op_35655_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35655_cast = slice_by_index(begin = var_35655_begin_0, end = var_35655_end_0, end_mask = var_35655_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35655_cast")];
            tensor<int32, [4]> var_35659_begin_0 = const()[name = tensor<string, []>("op_35659_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_35659_end_0 = const()[name = tensor<string, []>("op_35659_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_35659_end_mask_0 = const()[name = tensor<string, []>("op_35659_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35659_cast = slice_by_index(begin = var_35659_begin_0, end = var_35659_end_0, end_mask = var_35659_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35659_cast")];
            tensor<int32, [4]> var_35663_begin_0 = const()[name = tensor<string, []>("op_35663_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_35663_end_0 = const()[name = tensor<string, []>("op_35663_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_35663_end_mask_0 = const()[name = tensor<string, []>("op_35663_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35663_cast = slice_by_index(begin = var_35663_begin_0, end = var_35663_end_0, end_mask = var_35663_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35663_cast")];
            tensor<int32, [4]> var_35667_begin_0 = const()[name = tensor<string, []>("op_35667_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_35667_end_0 = const()[name = tensor<string, []>("op_35667_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_35667_end_mask_0 = const()[name = tensor<string, []>("op_35667_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35667_cast = slice_by_index(begin = var_35667_begin_0, end = var_35667_end_0, end_mask = var_35667_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35667_cast")];
            tensor<int32, [4]> var_35671_begin_0 = const()[name = tensor<string, []>("op_35671_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_35671_end_0 = const()[name = tensor<string, []>("op_35671_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_35671_end_mask_0 = const()[name = tensor<string, []>("op_35671_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35671_cast = slice_by_index(begin = var_35671_begin_0, end = var_35671_end_0, end_mask = var_35671_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35671_cast")];
            tensor<int32, [4]> var_35675_begin_0 = const()[name = tensor<string, []>("op_35675_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_35675_end_0 = const()[name = tensor<string, []>("op_35675_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_35675_end_mask_0 = const()[name = tensor<string, []>("op_35675_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35675_cast = slice_by_index(begin = var_35675_begin_0, end = var_35675_end_0, end_mask = var_35675_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35675_cast")];
            tensor<int32, [4]> var_35679_begin_0 = const()[name = tensor<string, []>("op_35679_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_35679_end_0 = const()[name = tensor<string, []>("op_35679_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_35679_end_mask_0 = const()[name = tensor<string, []>("op_35679_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35679_cast = slice_by_index(begin = var_35679_begin_0, end = var_35679_end_0, end_mask = var_35679_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35679_cast")];
            tensor<int32, [4]> var_35683_begin_0 = const()[name = tensor<string, []>("op_35683_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_35683_end_0 = const()[name = tensor<string, []>("op_35683_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_35683_end_mask_0 = const()[name = tensor<string, []>("op_35683_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_35683_cast = slice_by_index(begin = var_35683_begin_0, end = var_35683_end_0, end_mask = var_35683_end_mask_0, x = transpose_60)[name = tensor<string, []>("op_35683_cast")];
            tensor<int32, [4]> var_35685_begin_0 = const()[name = tensor<string, []>("op_35685_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_35685_end_0 = const()[name = tensor<string, []>("op_35685_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_35685_end_mask_0 = const()[name = tensor<string, []>("op_35685_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35685_cast = slice_by_index(begin = var_35685_begin_0, end = var_35685_end_0, end_mask = var_35685_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35685_cast")];
            tensor<int32, [4]> var_35689_begin_0 = const()[name = tensor<string, []>("op_35689_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_35689_end_0 = const()[name = tensor<string, []>("op_35689_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_35689_end_mask_0 = const()[name = tensor<string, []>("op_35689_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35689_cast = slice_by_index(begin = var_35689_begin_0, end = var_35689_end_0, end_mask = var_35689_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35689_cast")];
            tensor<int32, [4]> var_35693_begin_0 = const()[name = tensor<string, []>("op_35693_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_35693_end_0 = const()[name = tensor<string, []>("op_35693_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_35693_end_mask_0 = const()[name = tensor<string, []>("op_35693_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35693_cast = slice_by_index(begin = var_35693_begin_0, end = var_35693_end_0, end_mask = var_35693_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35693_cast")];
            tensor<int32, [4]> var_35697_begin_0 = const()[name = tensor<string, []>("op_35697_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_35697_end_0 = const()[name = tensor<string, []>("op_35697_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_35697_end_mask_0 = const()[name = tensor<string, []>("op_35697_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35697_cast = slice_by_index(begin = var_35697_begin_0, end = var_35697_end_0, end_mask = var_35697_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35697_cast")];
            tensor<int32, [4]> var_35701_begin_0 = const()[name = tensor<string, []>("op_35701_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_35701_end_0 = const()[name = tensor<string, []>("op_35701_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_35701_end_mask_0 = const()[name = tensor<string, []>("op_35701_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35701_cast = slice_by_index(begin = var_35701_begin_0, end = var_35701_end_0, end_mask = var_35701_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35701_cast")];
            tensor<int32, [4]> var_35705_begin_0 = const()[name = tensor<string, []>("op_35705_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_35705_end_0 = const()[name = tensor<string, []>("op_35705_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_35705_end_mask_0 = const()[name = tensor<string, []>("op_35705_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35705_cast = slice_by_index(begin = var_35705_begin_0, end = var_35705_end_0, end_mask = var_35705_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35705_cast")];
            tensor<int32, [4]> var_35709_begin_0 = const()[name = tensor<string, []>("op_35709_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_35709_end_0 = const()[name = tensor<string, []>("op_35709_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_35709_end_mask_0 = const()[name = tensor<string, []>("op_35709_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35709_cast = slice_by_index(begin = var_35709_begin_0, end = var_35709_end_0, end_mask = var_35709_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35709_cast")];
            tensor<int32, [4]> var_35713_begin_0 = const()[name = tensor<string, []>("op_35713_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_35713_end_0 = const()[name = tensor<string, []>("op_35713_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_35713_end_mask_0 = const()[name = tensor<string, []>("op_35713_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35713_cast = slice_by_index(begin = var_35713_begin_0, end = var_35713_end_0, end_mask = var_35713_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35713_cast")];
            tensor<int32, [4]> var_35717_begin_0 = const()[name = tensor<string, []>("op_35717_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_35717_end_0 = const()[name = tensor<string, []>("op_35717_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_35717_end_mask_0 = const()[name = tensor<string, []>("op_35717_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35717_cast = slice_by_index(begin = var_35717_begin_0, end = var_35717_end_0, end_mask = var_35717_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35717_cast")];
            tensor<int32, [4]> var_35721_begin_0 = const()[name = tensor<string, []>("op_35721_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_35721_end_0 = const()[name = tensor<string, []>("op_35721_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_35721_end_mask_0 = const()[name = tensor<string, []>("op_35721_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35721_cast = slice_by_index(begin = var_35721_begin_0, end = var_35721_end_0, end_mask = var_35721_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35721_cast")];
            tensor<int32, [4]> var_35725_begin_0 = const()[name = tensor<string, []>("op_35725_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_35725_end_0 = const()[name = tensor<string, []>("op_35725_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_35725_end_mask_0 = const()[name = tensor<string, []>("op_35725_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35725_cast = slice_by_index(begin = var_35725_begin_0, end = var_35725_end_0, end_mask = var_35725_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35725_cast")];
            tensor<int32, [4]> var_35729_begin_0 = const()[name = tensor<string, []>("op_35729_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_35729_end_0 = const()[name = tensor<string, []>("op_35729_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_35729_end_mask_0 = const()[name = tensor<string, []>("op_35729_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35729_cast = slice_by_index(begin = var_35729_begin_0, end = var_35729_end_0, end_mask = var_35729_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35729_cast")];
            tensor<int32, [4]> var_35733_begin_0 = const()[name = tensor<string, []>("op_35733_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_35733_end_0 = const()[name = tensor<string, []>("op_35733_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_35733_end_mask_0 = const()[name = tensor<string, []>("op_35733_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35733_cast = slice_by_index(begin = var_35733_begin_0, end = var_35733_end_0, end_mask = var_35733_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35733_cast")];
            tensor<int32, [4]> var_35737_begin_0 = const()[name = tensor<string, []>("op_35737_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_35737_end_0 = const()[name = tensor<string, []>("op_35737_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_35737_end_mask_0 = const()[name = tensor<string, []>("op_35737_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35737_cast = slice_by_index(begin = var_35737_begin_0, end = var_35737_end_0, end_mask = var_35737_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35737_cast")];
            tensor<int32, [4]> var_35741_begin_0 = const()[name = tensor<string, []>("op_35741_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_35741_end_0 = const()[name = tensor<string, []>("op_35741_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_35741_end_mask_0 = const()[name = tensor<string, []>("op_35741_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35741_cast = slice_by_index(begin = var_35741_begin_0, end = var_35741_end_0, end_mask = var_35741_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35741_cast")];
            tensor<int32, [4]> var_35745_begin_0 = const()[name = tensor<string, []>("op_35745_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_35745_end_0 = const()[name = tensor<string, []>("op_35745_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_35745_end_mask_0 = const()[name = tensor<string, []>("op_35745_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35745_cast = slice_by_index(begin = var_35745_begin_0, end = var_35745_end_0, end_mask = var_35745_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35745_cast")];
            tensor<int32, [4]> var_35749_begin_0 = const()[name = tensor<string, []>("op_35749_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_35749_end_0 = const()[name = tensor<string, []>("op_35749_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_35749_end_mask_0 = const()[name = tensor<string, []>("op_35749_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35749_cast = slice_by_index(begin = var_35749_begin_0, end = var_35749_end_0, end_mask = var_35749_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35749_cast")];
            tensor<int32, [4]> var_35753_begin_0 = const()[name = tensor<string, []>("op_35753_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_35753_end_0 = const()[name = tensor<string, []>("op_35753_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_35753_end_mask_0 = const()[name = tensor<string, []>("op_35753_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35753_cast = slice_by_index(begin = var_35753_begin_0, end = var_35753_end_0, end_mask = var_35753_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35753_cast")];
            tensor<int32, [4]> var_35757_begin_0 = const()[name = tensor<string, []>("op_35757_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_35757_end_0 = const()[name = tensor<string, []>("op_35757_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_35757_end_mask_0 = const()[name = tensor<string, []>("op_35757_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35757_cast = slice_by_index(begin = var_35757_begin_0, end = var_35757_end_0, end_mask = var_35757_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35757_cast")];
            tensor<int32, [4]> var_35761_begin_0 = const()[name = tensor<string, []>("op_35761_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_35761_end_0 = const()[name = tensor<string, []>("op_35761_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_35761_end_mask_0 = const()[name = tensor<string, []>("op_35761_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_35761_cast = slice_by_index(begin = var_35761_begin_0, end = var_35761_end_0, end_mask = var_35761_end_mask_0, x = v_159_cast)[name = tensor<string, []>("op_35761_cast")];
            tensor<string, []> var_35765_equation_0 = const()[name = tensor<string, []>("op_35765_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35765_cast = einsum(equation = var_35765_equation_0, values = (var_35607_cast, var_35524_cast))[name = tensor<string, []>("op_35765_cast")];
            tensor<fp16, []> var_35766_to_fp16 = const()[name = tensor<string, []>("op_35766_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3001_cast = mul(x = var_35765_cast, y = var_35766_to_fp16)[name = tensor<string, []>("aw_3001_cast")];
            tensor<string, []> var_35769_equation_0 = const()[name = tensor<string, []>("op_35769_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35769_cast = einsum(equation = var_35769_equation_0, values = (var_35611_cast, var_35528_cast))[name = tensor<string, []>("op_35769_cast")];
            tensor<fp16, []> var_35770_to_fp16 = const()[name = tensor<string, []>("op_35770_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3003_cast = mul(x = var_35769_cast, y = var_35770_to_fp16)[name = tensor<string, []>("aw_3003_cast")];
            tensor<string, []> var_35773_equation_0 = const()[name = tensor<string, []>("op_35773_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35773_cast = einsum(equation = var_35773_equation_0, values = (var_35615_cast, var_35532_cast))[name = tensor<string, []>("op_35773_cast")];
            tensor<fp16, []> var_35774_to_fp16 = const()[name = tensor<string, []>("op_35774_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3005_cast = mul(x = var_35773_cast, y = var_35774_to_fp16)[name = tensor<string, []>("aw_3005_cast")];
            tensor<string, []> var_35777_equation_0 = const()[name = tensor<string, []>("op_35777_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35777_cast = einsum(equation = var_35777_equation_0, values = (var_35619_cast, var_35536_cast))[name = tensor<string, []>("op_35777_cast")];
            tensor<fp16, []> var_35778_to_fp16 = const()[name = tensor<string, []>("op_35778_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3007_cast = mul(x = var_35777_cast, y = var_35778_to_fp16)[name = tensor<string, []>("aw_3007_cast")];
            tensor<string, []> var_35781_equation_0 = const()[name = tensor<string, []>("op_35781_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35781_cast = einsum(equation = var_35781_equation_0, values = (var_35623_cast, var_35540_cast))[name = tensor<string, []>("op_35781_cast")];
            tensor<fp16, []> var_35782_to_fp16 = const()[name = tensor<string, []>("op_35782_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3009_cast = mul(x = var_35781_cast, y = var_35782_to_fp16)[name = tensor<string, []>("aw_3009_cast")];
            tensor<string, []> var_35785_equation_0 = const()[name = tensor<string, []>("op_35785_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35785_cast = einsum(equation = var_35785_equation_0, values = (var_35627_cast, var_35544_cast))[name = tensor<string, []>("op_35785_cast")];
            tensor<fp16, []> var_35786_to_fp16 = const()[name = tensor<string, []>("op_35786_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3011_cast = mul(x = var_35785_cast, y = var_35786_to_fp16)[name = tensor<string, []>("aw_3011_cast")];
            tensor<string, []> var_35789_equation_0 = const()[name = tensor<string, []>("op_35789_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35789_cast = einsum(equation = var_35789_equation_0, values = (var_35631_cast, var_35548_cast))[name = tensor<string, []>("op_35789_cast")];
            tensor<fp16, []> var_35790_to_fp16 = const()[name = tensor<string, []>("op_35790_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3013_cast = mul(x = var_35789_cast, y = var_35790_to_fp16)[name = tensor<string, []>("aw_3013_cast")];
            tensor<string, []> var_35793_equation_0 = const()[name = tensor<string, []>("op_35793_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35793_cast = einsum(equation = var_35793_equation_0, values = (var_35635_cast, var_35552_cast))[name = tensor<string, []>("op_35793_cast")];
            tensor<fp16, []> var_35794_to_fp16 = const()[name = tensor<string, []>("op_35794_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3015_cast = mul(x = var_35793_cast, y = var_35794_to_fp16)[name = tensor<string, []>("aw_3015_cast")];
            tensor<string, []> var_35797_equation_0 = const()[name = tensor<string, []>("op_35797_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35797_cast = einsum(equation = var_35797_equation_0, values = (var_35639_cast, var_35556_cast))[name = tensor<string, []>("op_35797_cast")];
            tensor<fp16, []> var_35798_to_fp16 = const()[name = tensor<string, []>("op_35798_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3017_cast = mul(x = var_35797_cast, y = var_35798_to_fp16)[name = tensor<string, []>("aw_3017_cast")];
            tensor<string, []> var_35801_equation_0 = const()[name = tensor<string, []>("op_35801_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35801_cast = einsum(equation = var_35801_equation_0, values = (var_35643_cast, var_35560_cast))[name = tensor<string, []>("op_35801_cast")];
            tensor<fp16, []> var_35802_to_fp16 = const()[name = tensor<string, []>("op_35802_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3019_cast = mul(x = var_35801_cast, y = var_35802_to_fp16)[name = tensor<string, []>("aw_3019_cast")];
            tensor<string, []> var_35805_equation_0 = const()[name = tensor<string, []>("op_35805_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35805_cast = einsum(equation = var_35805_equation_0, values = (var_35647_cast, var_35564_cast))[name = tensor<string, []>("op_35805_cast")];
            tensor<fp16, []> var_35806_to_fp16 = const()[name = tensor<string, []>("op_35806_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3021_cast = mul(x = var_35805_cast, y = var_35806_to_fp16)[name = tensor<string, []>("aw_3021_cast")];
            tensor<string, []> var_35809_equation_0 = const()[name = tensor<string, []>("op_35809_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35809_cast = einsum(equation = var_35809_equation_0, values = (var_35651_cast, var_35568_cast))[name = tensor<string, []>("op_35809_cast")];
            tensor<fp16, []> var_35810_to_fp16 = const()[name = tensor<string, []>("op_35810_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3023_cast = mul(x = var_35809_cast, y = var_35810_to_fp16)[name = tensor<string, []>("aw_3023_cast")];
            tensor<string, []> var_35813_equation_0 = const()[name = tensor<string, []>("op_35813_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35813_cast = einsum(equation = var_35813_equation_0, values = (var_35655_cast, var_35572_cast))[name = tensor<string, []>("op_35813_cast")];
            tensor<fp16, []> var_35814_to_fp16 = const()[name = tensor<string, []>("op_35814_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3025_cast = mul(x = var_35813_cast, y = var_35814_to_fp16)[name = tensor<string, []>("aw_3025_cast")];
            tensor<string, []> var_35817_equation_0 = const()[name = tensor<string, []>("op_35817_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35817_cast = einsum(equation = var_35817_equation_0, values = (var_35659_cast, var_35576_cast))[name = tensor<string, []>("op_35817_cast")];
            tensor<fp16, []> var_35818_to_fp16 = const()[name = tensor<string, []>("op_35818_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3027_cast = mul(x = var_35817_cast, y = var_35818_to_fp16)[name = tensor<string, []>("aw_3027_cast")];
            tensor<string, []> var_35821_equation_0 = const()[name = tensor<string, []>("op_35821_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35821_cast = einsum(equation = var_35821_equation_0, values = (var_35663_cast, var_35580_cast))[name = tensor<string, []>("op_35821_cast")];
            tensor<fp16, []> var_35822_to_fp16 = const()[name = tensor<string, []>("op_35822_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3029_cast = mul(x = var_35821_cast, y = var_35822_to_fp16)[name = tensor<string, []>("aw_3029_cast")];
            tensor<string, []> var_35825_equation_0 = const()[name = tensor<string, []>("op_35825_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35825_cast = einsum(equation = var_35825_equation_0, values = (var_35667_cast, var_35584_cast))[name = tensor<string, []>("op_35825_cast")];
            tensor<fp16, []> var_35826_to_fp16 = const()[name = tensor<string, []>("op_35826_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3031_cast = mul(x = var_35825_cast, y = var_35826_to_fp16)[name = tensor<string, []>("aw_3031_cast")];
            tensor<string, []> var_35829_equation_0 = const()[name = tensor<string, []>("op_35829_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35829_cast = einsum(equation = var_35829_equation_0, values = (var_35671_cast, var_35588_cast))[name = tensor<string, []>("op_35829_cast")];
            tensor<fp16, []> var_35830_to_fp16 = const()[name = tensor<string, []>("op_35830_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3033_cast = mul(x = var_35829_cast, y = var_35830_to_fp16)[name = tensor<string, []>("aw_3033_cast")];
            tensor<string, []> var_35833_equation_0 = const()[name = tensor<string, []>("op_35833_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35833_cast = einsum(equation = var_35833_equation_0, values = (var_35675_cast, var_35592_cast))[name = tensor<string, []>("op_35833_cast")];
            tensor<fp16, []> var_35834_to_fp16 = const()[name = tensor<string, []>("op_35834_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3035_cast = mul(x = var_35833_cast, y = var_35834_to_fp16)[name = tensor<string, []>("aw_3035_cast")];
            tensor<string, []> var_35837_equation_0 = const()[name = tensor<string, []>("op_35837_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35837_cast = einsum(equation = var_35837_equation_0, values = (var_35679_cast, var_35596_cast))[name = tensor<string, []>("op_35837_cast")];
            tensor<fp16, []> var_35838_to_fp16 = const()[name = tensor<string, []>("op_35838_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3037_cast = mul(x = var_35837_cast, y = var_35838_to_fp16)[name = tensor<string, []>("aw_3037_cast")];
            tensor<string, []> var_35841_equation_0 = const()[name = tensor<string, []>("op_35841_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_35841_cast = einsum(equation = var_35841_equation_0, values = (var_35683_cast, var_35600_cast))[name = tensor<string, []>("op_35841_cast")];
            tensor<fp16, []> var_35842_to_fp16 = const()[name = tensor<string, []>("op_35842_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3039_cast = mul(x = var_35841_cast, y = var_35842_to_fp16)[name = tensor<string, []>("aw_3039_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35844_cast = softmax(axis = var_30385, x = aw_3001_cast)[name = tensor<string, []>("op_35844_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35845_cast = softmax(axis = var_30385, x = aw_3003_cast)[name = tensor<string, []>("op_35845_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35846_cast = softmax(axis = var_30385, x = aw_3005_cast)[name = tensor<string, []>("op_35846_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35847_cast = softmax(axis = var_30385, x = aw_3007_cast)[name = tensor<string, []>("op_35847_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35848_cast = softmax(axis = var_30385, x = aw_3009_cast)[name = tensor<string, []>("op_35848_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35849_cast = softmax(axis = var_30385, x = aw_3011_cast)[name = tensor<string, []>("op_35849_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35850_cast = softmax(axis = var_30385, x = aw_3013_cast)[name = tensor<string, []>("op_35850_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35851_cast = softmax(axis = var_30385, x = aw_3015_cast)[name = tensor<string, []>("op_35851_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35852_cast = softmax(axis = var_30385, x = aw_3017_cast)[name = tensor<string, []>("op_35852_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35853_cast = softmax(axis = var_30385, x = aw_3019_cast)[name = tensor<string, []>("op_35853_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35854_cast = softmax(axis = var_30385, x = aw_3021_cast)[name = tensor<string, []>("op_35854_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35855_cast = softmax(axis = var_30385, x = aw_3023_cast)[name = tensor<string, []>("op_35855_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35856_cast = softmax(axis = var_30385, x = aw_3025_cast)[name = tensor<string, []>("op_35856_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35857_cast = softmax(axis = var_30385, x = aw_3027_cast)[name = tensor<string, []>("op_35857_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35858_cast = softmax(axis = var_30385, x = aw_3029_cast)[name = tensor<string, []>("op_35858_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35859_cast = softmax(axis = var_30385, x = aw_3031_cast)[name = tensor<string, []>("op_35859_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35860_cast = softmax(axis = var_30385, x = aw_3033_cast)[name = tensor<string, []>("op_35860_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35861_cast = softmax(axis = var_30385, x = aw_3035_cast)[name = tensor<string, []>("op_35861_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35862_cast = softmax(axis = var_30385, x = aw_3037_cast)[name = tensor<string, []>("op_35862_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_35863_cast = softmax(axis = var_30385, x = aw_3039_cast)[name = tensor<string, []>("op_35863_cast")];
            tensor<string, []> var_35865_equation_0 = const()[name = tensor<string, []>("op_35865_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35865_cast = einsum(equation = var_35865_equation_0, values = (var_35685_cast, var_35844_cast))[name = tensor<string, []>("op_35865_cast")];
            tensor<string, []> var_35867_equation_0 = const()[name = tensor<string, []>("op_35867_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35867_cast = einsum(equation = var_35867_equation_0, values = (var_35689_cast, var_35845_cast))[name = tensor<string, []>("op_35867_cast")];
            tensor<string, []> var_35869_equation_0 = const()[name = tensor<string, []>("op_35869_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35869_cast = einsum(equation = var_35869_equation_0, values = (var_35693_cast, var_35846_cast))[name = tensor<string, []>("op_35869_cast")];
            tensor<string, []> var_35871_equation_0 = const()[name = tensor<string, []>("op_35871_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35871_cast = einsum(equation = var_35871_equation_0, values = (var_35697_cast, var_35847_cast))[name = tensor<string, []>("op_35871_cast")];
            tensor<string, []> var_35873_equation_0 = const()[name = tensor<string, []>("op_35873_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35873_cast = einsum(equation = var_35873_equation_0, values = (var_35701_cast, var_35848_cast))[name = tensor<string, []>("op_35873_cast")];
            tensor<string, []> var_35875_equation_0 = const()[name = tensor<string, []>("op_35875_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35875_cast = einsum(equation = var_35875_equation_0, values = (var_35705_cast, var_35849_cast))[name = tensor<string, []>("op_35875_cast")];
            tensor<string, []> var_35877_equation_0 = const()[name = tensor<string, []>("op_35877_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35877_cast = einsum(equation = var_35877_equation_0, values = (var_35709_cast, var_35850_cast))[name = tensor<string, []>("op_35877_cast")];
            tensor<string, []> var_35879_equation_0 = const()[name = tensor<string, []>("op_35879_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35879_cast = einsum(equation = var_35879_equation_0, values = (var_35713_cast, var_35851_cast))[name = tensor<string, []>("op_35879_cast")];
            tensor<string, []> var_35881_equation_0 = const()[name = tensor<string, []>("op_35881_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35881_cast = einsum(equation = var_35881_equation_0, values = (var_35717_cast, var_35852_cast))[name = tensor<string, []>("op_35881_cast")];
            tensor<string, []> var_35883_equation_0 = const()[name = tensor<string, []>("op_35883_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35883_cast = einsum(equation = var_35883_equation_0, values = (var_35721_cast, var_35853_cast))[name = tensor<string, []>("op_35883_cast")];
            tensor<string, []> var_35885_equation_0 = const()[name = tensor<string, []>("op_35885_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35885_cast = einsum(equation = var_35885_equation_0, values = (var_35725_cast, var_35854_cast))[name = tensor<string, []>("op_35885_cast")];
            tensor<string, []> var_35887_equation_0 = const()[name = tensor<string, []>("op_35887_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35887_cast = einsum(equation = var_35887_equation_0, values = (var_35729_cast, var_35855_cast))[name = tensor<string, []>("op_35887_cast")];
            tensor<string, []> var_35889_equation_0 = const()[name = tensor<string, []>("op_35889_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35889_cast = einsum(equation = var_35889_equation_0, values = (var_35733_cast, var_35856_cast))[name = tensor<string, []>("op_35889_cast")];
            tensor<string, []> var_35891_equation_0 = const()[name = tensor<string, []>("op_35891_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35891_cast = einsum(equation = var_35891_equation_0, values = (var_35737_cast, var_35857_cast))[name = tensor<string, []>("op_35891_cast")];
            tensor<string, []> var_35893_equation_0 = const()[name = tensor<string, []>("op_35893_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35893_cast = einsum(equation = var_35893_equation_0, values = (var_35741_cast, var_35858_cast))[name = tensor<string, []>("op_35893_cast")];
            tensor<string, []> var_35895_equation_0 = const()[name = tensor<string, []>("op_35895_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35895_cast = einsum(equation = var_35895_equation_0, values = (var_35745_cast, var_35859_cast))[name = tensor<string, []>("op_35895_cast")];
            tensor<string, []> var_35897_equation_0 = const()[name = tensor<string, []>("op_35897_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35897_cast = einsum(equation = var_35897_equation_0, values = (var_35749_cast, var_35860_cast))[name = tensor<string, []>("op_35897_cast")];
            tensor<string, []> var_35899_equation_0 = const()[name = tensor<string, []>("op_35899_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35899_cast = einsum(equation = var_35899_equation_0, values = (var_35753_cast, var_35861_cast))[name = tensor<string, []>("op_35899_cast")];
            tensor<string, []> var_35901_equation_0 = const()[name = tensor<string, []>("op_35901_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35901_cast = einsum(equation = var_35901_equation_0, values = (var_35757_cast, var_35862_cast))[name = tensor<string, []>("op_35901_cast")];
            tensor<string, []> var_35903_equation_0 = const()[name = tensor<string, []>("op_35903_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_35903_cast = einsum(equation = var_35903_equation_0, values = (var_35761_cast, var_35863_cast))[name = tensor<string, []>("op_35903_cast")];
            tensor<bool, []> input_483_interleave_0 = const()[name = tensor<string, []>("input_483_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_483_cast = concat(axis = var_30385, interleave = input_483_interleave_0, values = (var_35865_cast, var_35867_cast, var_35869_cast, var_35871_cast, var_35873_cast, var_35875_cast, var_35877_cast, var_35879_cast, var_35881_cast, var_35883_cast, var_35885_cast, var_35887_cast, var_35889_cast, var_35891_cast, var_35893_cast, var_35895_cast, var_35897_cast, var_35899_cast, var_35901_cast, var_35903_cast))[name = tensor<string, []>("input_483_cast")];
            tensor<int32, [2]> var_35909 = const()[name = tensor<string, []>("op_35909"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_35911 = const()[name = tensor<string, []>("op_35911"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_35913_pad_type_0 = const()[name = tensor<string, []>("op_35913_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_35913_pad_0 = const()[name = tensor<string, []>("op_35913_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2962869504)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2966146368)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_35913_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_out_0_bias_to_fp16, dilations = var_35911, groups = var_30385, pad = var_35913_pad_0, pad_type = var_35913_pad_type_0, strides = var_35909, weight = up_blocks_0_attentions_0_transformer_blocks_5_attn2_to_out_0_weight_to_fp16, x = input_483_cast)[name = tensor<string, []>("op_35913_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_239_cast = add(x = var_35913_cast, y = inputs_237_cast)[name = tensor<string, []>("inputs_239_cast")];
            tensor<int32, [1]> var_35917 = const()[name = tensor<string, []>("op_35917"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_239_cast = reduce_mean(axes = var_35917, keep_dims = var_30380, x = inputs_239_cast)[name = tensor<string, []>("channels_mean_239_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_239_cast = sub(x = inputs_239_cast, y = channels_mean_239_cast)[name = tensor<string, []>("zero_mean_239_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_239_cast = mul(x = zero_mean_239_cast, y = zero_mean_239_cast)[name = tensor<string, []>("zero_mean_sq_239_cast")];
            tensor<int32, [1]> var_35921 = const()[name = tensor<string, []>("op_35921"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_35922_cast = reduce_mean(axes = var_35921, keep_dims = var_30380, x = zero_mean_sq_239_cast)[name = tensor<string, []>("op_35922_cast")];
            tensor<fp16, []> var_35923_to_fp16 = const()[name = tensor<string, []>("op_35923_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_35924_cast = add(x = var_35922_cast, y = var_35923_to_fp16)[name = tensor<string, []>("op_35924_cast")];
            tensor<fp16, []> denom_239_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_239_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_239_cast = rsqrt(epsilon = denom_239_epsilon_0_to_fp16, x = var_35924_cast)[name = tensor<string, []>("denom_239_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_239_cast = mul(x = zero_mean_239_cast, y = denom_239_cast)[name = tensor<string, []>("out_239_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_35928_to_fp16 = const()[name = tensor<string, []>("op_35928_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2966148992)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_35929_cast = add(x = out_239_cast, y = var_35928_to_fp16)[name = tensor<string, []>("op_35929_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_35931_to_fp16 = const()[name = tensor<string, []>("op_35931_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2966151616)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_485_cast = mul(x = var_35929_cast, y = var_35931_to_fp16)[name = tensor<string, []>("input_485_cast")];
            tensor<int32, [2]> var_35939 = const()[name = tensor<string, []>("op_35939"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_35941 = const()[name = tensor<string, []>("op_35941"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_35943_pad_type_0 = const()[name = tensor<string, []>("op_35943_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_35943_pad_0 = const()[name = tensor<string, []>("op_35943_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2966154240)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2992368704)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_35943_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_5_ff_net_0_proj_bias_to_fp16, dilations = var_35941, groups = var_30385, pad = var_35943_pad_0, pad_type = var_35943_pad_type_0, strides = var_35939, weight = up_blocks_0_attentions_0_transformer_blocks_5_ff_net_0_proj_weight_to_fp16, x = input_485_cast)[name = tensor<string, []>("op_35943_cast")];
            tensor<int32, [2]> var_35944_split_sizes_0 = const()[name = tensor<string, []>("op_35944_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_35944_axis_0 = const()[name = tensor<string, []>("op_35944_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_35944_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_35944_cast_1 = split(axis = var_35944_axis_0, split_sizes = var_35944_split_sizes_0, x = var_35943_cast)[name = tensor<string, []>("op_35944_cast")];
            tensor<string, []> var_35946_mode_0 = const()[name = tensor<string, []>("op_35946_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_35946_cast = gelu(mode = var_35946_mode_0, x = var_35944_cast_1)[name = tensor<string, []>("op_35946_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_487_cast = mul(x = var_35944_cast_0, y = var_35946_cast)[name = tensor<string, []>("input_487_cast")];
            tensor<int32, [2]> var_35950 = const()[name = tensor<string, []>("op_35950"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_35952 = const()[name = tensor<string, []>("op_35952"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_35954_pad_type_0 = const()[name = tensor<string, []>("op_35954_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_35954_pad_0 = const()[name = tensor<string, []>("op_35954_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2992389248)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3005496512)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_35954_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_5_ff_net_2_bias_to_fp16, dilations = var_35952, groups = var_30385, pad = var_35954_pad_0, pad_type = var_35954_pad_type_0, strides = var_35950, weight = up_blocks_0_attentions_0_transformer_blocks_5_ff_net_2_weight_to_fp16, x = input_487_cast)[name = tensor<string, []>("op_35954_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_241_cast = add(x = var_35954_cast, y = inputs_239_cast)[name = tensor<string, []>("inputs_241_cast")];
            tensor<int32, [1]> var_35964 = const()[name = tensor<string, []>("op_35964"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_241_cast = reduce_mean(axes = var_35964, keep_dims = var_30380, x = inputs_241_cast)[name = tensor<string, []>("channels_mean_241_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_241_cast = sub(x = inputs_241_cast, y = channels_mean_241_cast)[name = tensor<string, []>("zero_mean_241_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_241_cast = mul(x = zero_mean_241_cast, y = zero_mean_241_cast)[name = tensor<string, []>("zero_mean_sq_241_cast")];
            tensor<int32, [1]> var_35968 = const()[name = tensor<string, []>("op_35968"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_35969_cast = reduce_mean(axes = var_35968, keep_dims = var_30380, x = zero_mean_sq_241_cast)[name = tensor<string, []>("op_35969_cast")];
            tensor<fp16, []> var_35970_to_fp16 = const()[name = tensor<string, []>("op_35970_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_35971_cast = add(x = var_35969_cast, y = var_35970_to_fp16)[name = tensor<string, []>("op_35971_cast")];
            tensor<fp16, []> denom_241_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_241_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_241_cast = rsqrt(epsilon = denom_241_epsilon_0_to_fp16, x = var_35971_cast)[name = tensor<string, []>("denom_241_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_241_cast = mul(x = zero_mean_241_cast, y = denom_241_cast)[name = tensor<string, []>("out_241_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_35975_to_fp16 = const()[name = tensor<string, []>("op_35975_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3005499136)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_35976_cast = add(x = out_241_cast, y = var_35975_to_fp16)[name = tensor<string, []>("op_35976_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_35978_to_fp16 = const()[name = tensor<string, []>("op_35978_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3005501760)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_323_cast = mul(x = var_35976_cast, y = var_35978_to_fp16)[name = tensor<string, []>("hidden_states_323_cast")];
            tensor<int32, [2]> var_35985 = const()[name = tensor<string, []>("op_35985"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_35987 = const()[name = tensor<string, []>("op_35987"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_161_pad_type_0 = const()[name = tensor<string, []>("q_161_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_161_pad_0 = const()[name = tensor<string, []>("q_161_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3005504384)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_161_cast = conv(dilations = var_35987, groups = var_30385, pad = q_161_pad_0, pad_type = q_161_pad_type_0, strides = var_35985, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_q_weight_to_fp16, x = hidden_states_323_cast)[name = tensor<string, []>("q_161_cast")];
            tensor<int32, [2]> var_35991 = const()[name = tensor<string, []>("op_35991"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_35993 = const()[name = tensor<string, []>("op_35993"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_321_pad_type_0 = const()[name = tensor<string, []>("k_321_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_321_pad_0 = const()[name = tensor<string, []>("k_321_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3008781248)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_321_cast = conv(dilations = var_35993, groups = var_30385, pad = k_321_pad_0, pad_type = k_321_pad_type_0, strides = var_35991, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_k_weight_to_fp16, x = hidden_states_323_cast)[name = tensor<string, []>("k_321_cast")];
            tensor<int32, [2]> var_35997 = const()[name = tensor<string, []>("op_35997"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_35999 = const()[name = tensor<string, []>("op_35999"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_161_pad_type_0 = const()[name = tensor<string, []>("v_161_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_161_pad_0 = const()[name = tensor<string, []>("v_161_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3012058112)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_161_cast = conv(dilations = var_35999, groups = var_30385, pad = v_161_pad_0, pad_type = v_161_pad_type_0, strides = var_35997, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_v_weight_to_fp16, x = hidden_states_323_cast)[name = tensor<string, []>("v_161_cast")];
            tensor<int32, [4]> var_36003_begin_0 = const()[name = tensor<string, []>("op_36003_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_36003_end_0 = const()[name = tensor<string, []>("op_36003_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_36003_end_mask_0 = const()[name = tensor<string, []>("op_36003_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36003_cast = slice_by_index(begin = var_36003_begin_0, end = var_36003_end_0, end_mask = var_36003_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36003_cast")];
            tensor<int32, [4]> var_36007_begin_0 = const()[name = tensor<string, []>("op_36007_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_36007_end_0 = const()[name = tensor<string, []>("op_36007_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_36007_end_mask_0 = const()[name = tensor<string, []>("op_36007_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36007_cast = slice_by_index(begin = var_36007_begin_0, end = var_36007_end_0, end_mask = var_36007_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36007_cast")];
            tensor<int32, [4]> var_36011_begin_0 = const()[name = tensor<string, []>("op_36011_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_36011_end_0 = const()[name = tensor<string, []>("op_36011_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_36011_end_mask_0 = const()[name = tensor<string, []>("op_36011_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36011_cast = slice_by_index(begin = var_36011_begin_0, end = var_36011_end_0, end_mask = var_36011_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36011_cast")];
            tensor<int32, [4]> var_36015_begin_0 = const()[name = tensor<string, []>("op_36015_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_36015_end_0 = const()[name = tensor<string, []>("op_36015_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_36015_end_mask_0 = const()[name = tensor<string, []>("op_36015_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36015_cast = slice_by_index(begin = var_36015_begin_0, end = var_36015_end_0, end_mask = var_36015_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36015_cast")];
            tensor<int32, [4]> var_36019_begin_0 = const()[name = tensor<string, []>("op_36019_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_36019_end_0 = const()[name = tensor<string, []>("op_36019_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_36019_end_mask_0 = const()[name = tensor<string, []>("op_36019_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36019_cast = slice_by_index(begin = var_36019_begin_0, end = var_36019_end_0, end_mask = var_36019_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36019_cast")];
            tensor<int32, [4]> var_36023_begin_0 = const()[name = tensor<string, []>("op_36023_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_36023_end_0 = const()[name = tensor<string, []>("op_36023_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_36023_end_mask_0 = const()[name = tensor<string, []>("op_36023_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36023_cast = slice_by_index(begin = var_36023_begin_0, end = var_36023_end_0, end_mask = var_36023_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36023_cast")];
            tensor<int32, [4]> var_36027_begin_0 = const()[name = tensor<string, []>("op_36027_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_36027_end_0 = const()[name = tensor<string, []>("op_36027_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_36027_end_mask_0 = const()[name = tensor<string, []>("op_36027_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36027_cast = slice_by_index(begin = var_36027_begin_0, end = var_36027_end_0, end_mask = var_36027_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36027_cast")];
            tensor<int32, [4]> var_36031_begin_0 = const()[name = tensor<string, []>("op_36031_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_36031_end_0 = const()[name = tensor<string, []>("op_36031_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_36031_end_mask_0 = const()[name = tensor<string, []>("op_36031_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36031_cast = slice_by_index(begin = var_36031_begin_0, end = var_36031_end_0, end_mask = var_36031_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36031_cast")];
            tensor<int32, [4]> var_36035_begin_0 = const()[name = tensor<string, []>("op_36035_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_36035_end_0 = const()[name = tensor<string, []>("op_36035_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_36035_end_mask_0 = const()[name = tensor<string, []>("op_36035_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36035_cast = slice_by_index(begin = var_36035_begin_0, end = var_36035_end_0, end_mask = var_36035_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36035_cast")];
            tensor<int32, [4]> var_36039_begin_0 = const()[name = tensor<string, []>("op_36039_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_36039_end_0 = const()[name = tensor<string, []>("op_36039_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_36039_end_mask_0 = const()[name = tensor<string, []>("op_36039_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36039_cast = slice_by_index(begin = var_36039_begin_0, end = var_36039_end_0, end_mask = var_36039_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36039_cast")];
            tensor<int32, [4]> var_36043_begin_0 = const()[name = tensor<string, []>("op_36043_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_36043_end_0 = const()[name = tensor<string, []>("op_36043_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_36043_end_mask_0 = const()[name = tensor<string, []>("op_36043_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36043_cast = slice_by_index(begin = var_36043_begin_0, end = var_36043_end_0, end_mask = var_36043_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36043_cast")];
            tensor<int32, [4]> var_36047_begin_0 = const()[name = tensor<string, []>("op_36047_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_36047_end_0 = const()[name = tensor<string, []>("op_36047_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_36047_end_mask_0 = const()[name = tensor<string, []>("op_36047_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36047_cast = slice_by_index(begin = var_36047_begin_0, end = var_36047_end_0, end_mask = var_36047_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36047_cast")];
            tensor<int32, [4]> var_36051_begin_0 = const()[name = tensor<string, []>("op_36051_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_36051_end_0 = const()[name = tensor<string, []>("op_36051_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_36051_end_mask_0 = const()[name = tensor<string, []>("op_36051_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36051_cast = slice_by_index(begin = var_36051_begin_0, end = var_36051_end_0, end_mask = var_36051_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36051_cast")];
            tensor<int32, [4]> var_36055_begin_0 = const()[name = tensor<string, []>("op_36055_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_36055_end_0 = const()[name = tensor<string, []>("op_36055_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_36055_end_mask_0 = const()[name = tensor<string, []>("op_36055_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36055_cast = slice_by_index(begin = var_36055_begin_0, end = var_36055_end_0, end_mask = var_36055_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36055_cast")];
            tensor<int32, [4]> var_36059_begin_0 = const()[name = tensor<string, []>("op_36059_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_36059_end_0 = const()[name = tensor<string, []>("op_36059_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_36059_end_mask_0 = const()[name = tensor<string, []>("op_36059_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36059_cast = slice_by_index(begin = var_36059_begin_0, end = var_36059_end_0, end_mask = var_36059_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36059_cast")];
            tensor<int32, [4]> var_36063_begin_0 = const()[name = tensor<string, []>("op_36063_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_36063_end_0 = const()[name = tensor<string, []>("op_36063_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_36063_end_mask_0 = const()[name = tensor<string, []>("op_36063_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36063_cast = slice_by_index(begin = var_36063_begin_0, end = var_36063_end_0, end_mask = var_36063_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36063_cast")];
            tensor<int32, [4]> var_36067_begin_0 = const()[name = tensor<string, []>("op_36067_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_36067_end_0 = const()[name = tensor<string, []>("op_36067_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_36067_end_mask_0 = const()[name = tensor<string, []>("op_36067_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36067_cast = slice_by_index(begin = var_36067_begin_0, end = var_36067_end_0, end_mask = var_36067_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36067_cast")];
            tensor<int32, [4]> var_36071_begin_0 = const()[name = tensor<string, []>("op_36071_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_36071_end_0 = const()[name = tensor<string, []>("op_36071_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_36071_end_mask_0 = const()[name = tensor<string, []>("op_36071_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36071_cast = slice_by_index(begin = var_36071_begin_0, end = var_36071_end_0, end_mask = var_36071_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36071_cast")];
            tensor<int32, [4]> var_36075_begin_0 = const()[name = tensor<string, []>("op_36075_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_36075_end_0 = const()[name = tensor<string, []>("op_36075_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_36075_end_mask_0 = const()[name = tensor<string, []>("op_36075_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36075_cast = slice_by_index(begin = var_36075_begin_0, end = var_36075_end_0, end_mask = var_36075_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36075_cast")];
            tensor<int32, [4]> var_36079_begin_0 = const()[name = tensor<string, []>("op_36079_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_36079_end_0 = const()[name = tensor<string, []>("op_36079_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_36079_end_mask_0 = const()[name = tensor<string, []>("op_36079_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36079_cast = slice_by_index(begin = var_36079_begin_0, end = var_36079_end_0, end_mask = var_36079_end_mask_0, x = q_161_cast)[name = tensor<string, []>("op_36079_cast")];
            tensor<int32, [4]> k_323_perm_0 = const()[name = tensor<string, []>("k_323_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_36086_begin_0 = const()[name = tensor<string, []>("op_36086_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_36086_end_0 = const()[name = tensor<string, []>("op_36086_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_36086_end_mask_0 = const()[name = tensor<string, []>("op_36086_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_59 = transpose(perm = k_323_perm_0, x = k_321_cast)[name = tensor<string, []>("transpose_59")];
            tensor<fp16, [2, 1024, 1, 64]> var_36086_cast = slice_by_index(begin = var_36086_begin_0, end = var_36086_end_0, end_mask = var_36086_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36086_cast")];
            tensor<int32, [4]> var_36090_begin_0 = const()[name = tensor<string, []>("op_36090_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_36090_end_0 = const()[name = tensor<string, []>("op_36090_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_36090_end_mask_0 = const()[name = tensor<string, []>("op_36090_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36090_cast = slice_by_index(begin = var_36090_begin_0, end = var_36090_end_0, end_mask = var_36090_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36090_cast")];
            tensor<int32, [4]> var_36094_begin_0 = const()[name = tensor<string, []>("op_36094_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_36094_end_0 = const()[name = tensor<string, []>("op_36094_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_36094_end_mask_0 = const()[name = tensor<string, []>("op_36094_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36094_cast = slice_by_index(begin = var_36094_begin_0, end = var_36094_end_0, end_mask = var_36094_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36094_cast")];
            tensor<int32, [4]> var_36098_begin_0 = const()[name = tensor<string, []>("op_36098_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_36098_end_0 = const()[name = tensor<string, []>("op_36098_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_36098_end_mask_0 = const()[name = tensor<string, []>("op_36098_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36098_cast = slice_by_index(begin = var_36098_begin_0, end = var_36098_end_0, end_mask = var_36098_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36098_cast")];
            tensor<int32, [4]> var_36102_begin_0 = const()[name = tensor<string, []>("op_36102_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_36102_end_0 = const()[name = tensor<string, []>("op_36102_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_36102_end_mask_0 = const()[name = tensor<string, []>("op_36102_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36102_cast = slice_by_index(begin = var_36102_begin_0, end = var_36102_end_0, end_mask = var_36102_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36102_cast")];
            tensor<int32, [4]> var_36106_begin_0 = const()[name = tensor<string, []>("op_36106_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_36106_end_0 = const()[name = tensor<string, []>("op_36106_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_36106_end_mask_0 = const()[name = tensor<string, []>("op_36106_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36106_cast = slice_by_index(begin = var_36106_begin_0, end = var_36106_end_0, end_mask = var_36106_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36106_cast")];
            tensor<int32, [4]> var_36110_begin_0 = const()[name = tensor<string, []>("op_36110_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_36110_end_0 = const()[name = tensor<string, []>("op_36110_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_36110_end_mask_0 = const()[name = tensor<string, []>("op_36110_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36110_cast = slice_by_index(begin = var_36110_begin_0, end = var_36110_end_0, end_mask = var_36110_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36110_cast")];
            tensor<int32, [4]> var_36114_begin_0 = const()[name = tensor<string, []>("op_36114_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_36114_end_0 = const()[name = tensor<string, []>("op_36114_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_36114_end_mask_0 = const()[name = tensor<string, []>("op_36114_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36114_cast = slice_by_index(begin = var_36114_begin_0, end = var_36114_end_0, end_mask = var_36114_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36114_cast")];
            tensor<int32, [4]> var_36118_begin_0 = const()[name = tensor<string, []>("op_36118_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_36118_end_0 = const()[name = tensor<string, []>("op_36118_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_36118_end_mask_0 = const()[name = tensor<string, []>("op_36118_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36118_cast = slice_by_index(begin = var_36118_begin_0, end = var_36118_end_0, end_mask = var_36118_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36118_cast")];
            tensor<int32, [4]> var_36122_begin_0 = const()[name = tensor<string, []>("op_36122_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_36122_end_0 = const()[name = tensor<string, []>("op_36122_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_36122_end_mask_0 = const()[name = tensor<string, []>("op_36122_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36122_cast = slice_by_index(begin = var_36122_begin_0, end = var_36122_end_0, end_mask = var_36122_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36122_cast")];
            tensor<int32, [4]> var_36126_begin_0 = const()[name = tensor<string, []>("op_36126_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_36126_end_0 = const()[name = tensor<string, []>("op_36126_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_36126_end_mask_0 = const()[name = tensor<string, []>("op_36126_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36126_cast = slice_by_index(begin = var_36126_begin_0, end = var_36126_end_0, end_mask = var_36126_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36126_cast")];
            tensor<int32, [4]> var_36130_begin_0 = const()[name = tensor<string, []>("op_36130_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_36130_end_0 = const()[name = tensor<string, []>("op_36130_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_36130_end_mask_0 = const()[name = tensor<string, []>("op_36130_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36130_cast = slice_by_index(begin = var_36130_begin_0, end = var_36130_end_0, end_mask = var_36130_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36130_cast")];
            tensor<int32, [4]> var_36134_begin_0 = const()[name = tensor<string, []>("op_36134_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_36134_end_0 = const()[name = tensor<string, []>("op_36134_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_36134_end_mask_0 = const()[name = tensor<string, []>("op_36134_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36134_cast = slice_by_index(begin = var_36134_begin_0, end = var_36134_end_0, end_mask = var_36134_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36134_cast")];
            tensor<int32, [4]> var_36138_begin_0 = const()[name = tensor<string, []>("op_36138_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_36138_end_0 = const()[name = tensor<string, []>("op_36138_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_36138_end_mask_0 = const()[name = tensor<string, []>("op_36138_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36138_cast = slice_by_index(begin = var_36138_begin_0, end = var_36138_end_0, end_mask = var_36138_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36138_cast")];
            tensor<int32, [4]> var_36142_begin_0 = const()[name = tensor<string, []>("op_36142_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_36142_end_0 = const()[name = tensor<string, []>("op_36142_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_36142_end_mask_0 = const()[name = tensor<string, []>("op_36142_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36142_cast = slice_by_index(begin = var_36142_begin_0, end = var_36142_end_0, end_mask = var_36142_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36142_cast")];
            tensor<int32, [4]> var_36146_begin_0 = const()[name = tensor<string, []>("op_36146_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_36146_end_0 = const()[name = tensor<string, []>("op_36146_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_36146_end_mask_0 = const()[name = tensor<string, []>("op_36146_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36146_cast = slice_by_index(begin = var_36146_begin_0, end = var_36146_end_0, end_mask = var_36146_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36146_cast")];
            tensor<int32, [4]> var_36150_begin_0 = const()[name = tensor<string, []>("op_36150_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_36150_end_0 = const()[name = tensor<string, []>("op_36150_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_36150_end_mask_0 = const()[name = tensor<string, []>("op_36150_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36150_cast = slice_by_index(begin = var_36150_begin_0, end = var_36150_end_0, end_mask = var_36150_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36150_cast")];
            tensor<int32, [4]> var_36154_begin_0 = const()[name = tensor<string, []>("op_36154_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_36154_end_0 = const()[name = tensor<string, []>("op_36154_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_36154_end_mask_0 = const()[name = tensor<string, []>("op_36154_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36154_cast = slice_by_index(begin = var_36154_begin_0, end = var_36154_end_0, end_mask = var_36154_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36154_cast")];
            tensor<int32, [4]> var_36158_begin_0 = const()[name = tensor<string, []>("op_36158_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_36158_end_0 = const()[name = tensor<string, []>("op_36158_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_36158_end_mask_0 = const()[name = tensor<string, []>("op_36158_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36158_cast = slice_by_index(begin = var_36158_begin_0, end = var_36158_end_0, end_mask = var_36158_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36158_cast")];
            tensor<int32, [4]> var_36162_begin_0 = const()[name = tensor<string, []>("op_36162_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_36162_end_0 = const()[name = tensor<string, []>("op_36162_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_36162_end_mask_0 = const()[name = tensor<string, []>("op_36162_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_36162_cast = slice_by_index(begin = var_36162_begin_0, end = var_36162_end_0, end_mask = var_36162_end_mask_0, x = transpose_59)[name = tensor<string, []>("op_36162_cast")];
            tensor<int32, [4]> var_36164_begin_0 = const()[name = tensor<string, []>("op_36164_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_36164_end_0 = const()[name = tensor<string, []>("op_36164_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_36164_end_mask_0 = const()[name = tensor<string, []>("op_36164_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36164_cast = slice_by_index(begin = var_36164_begin_0, end = var_36164_end_0, end_mask = var_36164_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36164_cast")];
            tensor<int32, [4]> var_36168_begin_0 = const()[name = tensor<string, []>("op_36168_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_36168_end_0 = const()[name = tensor<string, []>("op_36168_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_36168_end_mask_0 = const()[name = tensor<string, []>("op_36168_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36168_cast = slice_by_index(begin = var_36168_begin_0, end = var_36168_end_0, end_mask = var_36168_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36168_cast")];
            tensor<int32, [4]> var_36172_begin_0 = const()[name = tensor<string, []>("op_36172_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_36172_end_0 = const()[name = tensor<string, []>("op_36172_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_36172_end_mask_0 = const()[name = tensor<string, []>("op_36172_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36172_cast = slice_by_index(begin = var_36172_begin_0, end = var_36172_end_0, end_mask = var_36172_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36172_cast")];
            tensor<int32, [4]> var_36176_begin_0 = const()[name = tensor<string, []>("op_36176_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_36176_end_0 = const()[name = tensor<string, []>("op_36176_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_36176_end_mask_0 = const()[name = tensor<string, []>("op_36176_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36176_cast = slice_by_index(begin = var_36176_begin_0, end = var_36176_end_0, end_mask = var_36176_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36176_cast")];
            tensor<int32, [4]> var_36180_begin_0 = const()[name = tensor<string, []>("op_36180_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_36180_end_0 = const()[name = tensor<string, []>("op_36180_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_36180_end_mask_0 = const()[name = tensor<string, []>("op_36180_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36180_cast = slice_by_index(begin = var_36180_begin_0, end = var_36180_end_0, end_mask = var_36180_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36180_cast")];
            tensor<int32, [4]> var_36184_begin_0 = const()[name = tensor<string, []>("op_36184_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_36184_end_0 = const()[name = tensor<string, []>("op_36184_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_36184_end_mask_0 = const()[name = tensor<string, []>("op_36184_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36184_cast = slice_by_index(begin = var_36184_begin_0, end = var_36184_end_0, end_mask = var_36184_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36184_cast")];
            tensor<int32, [4]> var_36188_begin_0 = const()[name = tensor<string, []>("op_36188_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_36188_end_0 = const()[name = tensor<string, []>("op_36188_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_36188_end_mask_0 = const()[name = tensor<string, []>("op_36188_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36188_cast = slice_by_index(begin = var_36188_begin_0, end = var_36188_end_0, end_mask = var_36188_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36188_cast")];
            tensor<int32, [4]> var_36192_begin_0 = const()[name = tensor<string, []>("op_36192_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_36192_end_0 = const()[name = tensor<string, []>("op_36192_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_36192_end_mask_0 = const()[name = tensor<string, []>("op_36192_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36192_cast = slice_by_index(begin = var_36192_begin_0, end = var_36192_end_0, end_mask = var_36192_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36192_cast")];
            tensor<int32, [4]> var_36196_begin_0 = const()[name = tensor<string, []>("op_36196_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_36196_end_0 = const()[name = tensor<string, []>("op_36196_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_36196_end_mask_0 = const()[name = tensor<string, []>("op_36196_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36196_cast = slice_by_index(begin = var_36196_begin_0, end = var_36196_end_0, end_mask = var_36196_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36196_cast")];
            tensor<int32, [4]> var_36200_begin_0 = const()[name = tensor<string, []>("op_36200_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_36200_end_0 = const()[name = tensor<string, []>("op_36200_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_36200_end_mask_0 = const()[name = tensor<string, []>("op_36200_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36200_cast = slice_by_index(begin = var_36200_begin_0, end = var_36200_end_0, end_mask = var_36200_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36200_cast")];
            tensor<int32, [4]> var_36204_begin_0 = const()[name = tensor<string, []>("op_36204_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_36204_end_0 = const()[name = tensor<string, []>("op_36204_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_36204_end_mask_0 = const()[name = tensor<string, []>("op_36204_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36204_cast = slice_by_index(begin = var_36204_begin_0, end = var_36204_end_0, end_mask = var_36204_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36204_cast")];
            tensor<int32, [4]> var_36208_begin_0 = const()[name = tensor<string, []>("op_36208_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_36208_end_0 = const()[name = tensor<string, []>("op_36208_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_36208_end_mask_0 = const()[name = tensor<string, []>("op_36208_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36208_cast = slice_by_index(begin = var_36208_begin_0, end = var_36208_end_0, end_mask = var_36208_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36208_cast")];
            tensor<int32, [4]> var_36212_begin_0 = const()[name = tensor<string, []>("op_36212_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_36212_end_0 = const()[name = tensor<string, []>("op_36212_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_36212_end_mask_0 = const()[name = tensor<string, []>("op_36212_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36212_cast = slice_by_index(begin = var_36212_begin_0, end = var_36212_end_0, end_mask = var_36212_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36212_cast")];
            tensor<int32, [4]> var_36216_begin_0 = const()[name = tensor<string, []>("op_36216_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_36216_end_0 = const()[name = tensor<string, []>("op_36216_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_36216_end_mask_0 = const()[name = tensor<string, []>("op_36216_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36216_cast = slice_by_index(begin = var_36216_begin_0, end = var_36216_end_0, end_mask = var_36216_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36216_cast")];
            tensor<int32, [4]> var_36220_begin_0 = const()[name = tensor<string, []>("op_36220_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_36220_end_0 = const()[name = tensor<string, []>("op_36220_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_36220_end_mask_0 = const()[name = tensor<string, []>("op_36220_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36220_cast = slice_by_index(begin = var_36220_begin_0, end = var_36220_end_0, end_mask = var_36220_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36220_cast")];
            tensor<int32, [4]> var_36224_begin_0 = const()[name = tensor<string, []>("op_36224_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_36224_end_0 = const()[name = tensor<string, []>("op_36224_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_36224_end_mask_0 = const()[name = tensor<string, []>("op_36224_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36224_cast = slice_by_index(begin = var_36224_begin_0, end = var_36224_end_0, end_mask = var_36224_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36224_cast")];
            tensor<int32, [4]> var_36228_begin_0 = const()[name = tensor<string, []>("op_36228_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_36228_end_0 = const()[name = tensor<string, []>("op_36228_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_36228_end_mask_0 = const()[name = tensor<string, []>("op_36228_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36228_cast = slice_by_index(begin = var_36228_begin_0, end = var_36228_end_0, end_mask = var_36228_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36228_cast")];
            tensor<int32, [4]> var_36232_begin_0 = const()[name = tensor<string, []>("op_36232_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_36232_end_0 = const()[name = tensor<string, []>("op_36232_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_36232_end_mask_0 = const()[name = tensor<string, []>("op_36232_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36232_cast = slice_by_index(begin = var_36232_begin_0, end = var_36232_end_0, end_mask = var_36232_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36232_cast")];
            tensor<int32, [4]> var_36236_begin_0 = const()[name = tensor<string, []>("op_36236_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_36236_end_0 = const()[name = tensor<string, []>("op_36236_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_36236_end_mask_0 = const()[name = tensor<string, []>("op_36236_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36236_cast = slice_by_index(begin = var_36236_begin_0, end = var_36236_end_0, end_mask = var_36236_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36236_cast")];
            tensor<int32, [4]> var_36240_begin_0 = const()[name = tensor<string, []>("op_36240_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_36240_end_0 = const()[name = tensor<string, []>("op_36240_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_36240_end_mask_0 = const()[name = tensor<string, []>("op_36240_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36240_cast = slice_by_index(begin = var_36240_begin_0, end = var_36240_end_0, end_mask = var_36240_end_mask_0, x = v_161_cast)[name = tensor<string, []>("op_36240_cast")];
            tensor<string, []> var_36244_equation_0 = const()[name = tensor<string, []>("op_36244_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36244_cast = einsum(equation = var_36244_equation_0, values = (var_36086_cast, var_36003_cast))[name = tensor<string, []>("op_36244_cast")];
            tensor<fp16, []> var_36245_to_fp16 = const()[name = tensor<string, []>("op_36245_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3041_cast = mul(x = var_36244_cast, y = var_36245_to_fp16)[name = tensor<string, []>("aw_3041_cast")];
            tensor<string, []> var_36248_equation_0 = const()[name = tensor<string, []>("op_36248_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36248_cast = einsum(equation = var_36248_equation_0, values = (var_36090_cast, var_36007_cast))[name = tensor<string, []>("op_36248_cast")];
            tensor<fp16, []> var_36249_to_fp16 = const()[name = tensor<string, []>("op_36249_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3043_cast = mul(x = var_36248_cast, y = var_36249_to_fp16)[name = tensor<string, []>("aw_3043_cast")];
            tensor<string, []> var_36252_equation_0 = const()[name = tensor<string, []>("op_36252_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36252_cast = einsum(equation = var_36252_equation_0, values = (var_36094_cast, var_36011_cast))[name = tensor<string, []>("op_36252_cast")];
            tensor<fp16, []> var_36253_to_fp16 = const()[name = tensor<string, []>("op_36253_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3045_cast = mul(x = var_36252_cast, y = var_36253_to_fp16)[name = tensor<string, []>("aw_3045_cast")];
            tensor<string, []> var_36256_equation_0 = const()[name = tensor<string, []>("op_36256_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36256_cast = einsum(equation = var_36256_equation_0, values = (var_36098_cast, var_36015_cast))[name = tensor<string, []>("op_36256_cast")];
            tensor<fp16, []> var_36257_to_fp16 = const()[name = tensor<string, []>("op_36257_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3047_cast = mul(x = var_36256_cast, y = var_36257_to_fp16)[name = tensor<string, []>("aw_3047_cast")];
            tensor<string, []> var_36260_equation_0 = const()[name = tensor<string, []>("op_36260_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36260_cast = einsum(equation = var_36260_equation_0, values = (var_36102_cast, var_36019_cast))[name = tensor<string, []>("op_36260_cast")];
            tensor<fp16, []> var_36261_to_fp16 = const()[name = tensor<string, []>("op_36261_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3049_cast = mul(x = var_36260_cast, y = var_36261_to_fp16)[name = tensor<string, []>("aw_3049_cast")];
            tensor<string, []> var_36264_equation_0 = const()[name = tensor<string, []>("op_36264_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36264_cast = einsum(equation = var_36264_equation_0, values = (var_36106_cast, var_36023_cast))[name = tensor<string, []>("op_36264_cast")];
            tensor<fp16, []> var_36265_to_fp16 = const()[name = tensor<string, []>("op_36265_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3051_cast = mul(x = var_36264_cast, y = var_36265_to_fp16)[name = tensor<string, []>("aw_3051_cast")];
            tensor<string, []> var_36268_equation_0 = const()[name = tensor<string, []>("op_36268_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36268_cast = einsum(equation = var_36268_equation_0, values = (var_36110_cast, var_36027_cast))[name = tensor<string, []>("op_36268_cast")];
            tensor<fp16, []> var_36269_to_fp16 = const()[name = tensor<string, []>("op_36269_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3053_cast = mul(x = var_36268_cast, y = var_36269_to_fp16)[name = tensor<string, []>("aw_3053_cast")];
            tensor<string, []> var_36272_equation_0 = const()[name = tensor<string, []>("op_36272_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36272_cast = einsum(equation = var_36272_equation_0, values = (var_36114_cast, var_36031_cast))[name = tensor<string, []>("op_36272_cast")];
            tensor<fp16, []> var_36273_to_fp16 = const()[name = tensor<string, []>("op_36273_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3055_cast = mul(x = var_36272_cast, y = var_36273_to_fp16)[name = tensor<string, []>("aw_3055_cast")];
            tensor<string, []> var_36276_equation_0 = const()[name = tensor<string, []>("op_36276_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36276_cast = einsum(equation = var_36276_equation_0, values = (var_36118_cast, var_36035_cast))[name = tensor<string, []>("op_36276_cast")];
            tensor<fp16, []> var_36277_to_fp16 = const()[name = tensor<string, []>("op_36277_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3057_cast = mul(x = var_36276_cast, y = var_36277_to_fp16)[name = tensor<string, []>("aw_3057_cast")];
            tensor<string, []> var_36280_equation_0 = const()[name = tensor<string, []>("op_36280_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36280_cast = einsum(equation = var_36280_equation_0, values = (var_36122_cast, var_36039_cast))[name = tensor<string, []>("op_36280_cast")];
            tensor<fp16, []> var_36281_to_fp16 = const()[name = tensor<string, []>("op_36281_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3059_cast = mul(x = var_36280_cast, y = var_36281_to_fp16)[name = tensor<string, []>("aw_3059_cast")];
            tensor<string, []> var_36284_equation_0 = const()[name = tensor<string, []>("op_36284_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36284_cast = einsum(equation = var_36284_equation_0, values = (var_36126_cast, var_36043_cast))[name = tensor<string, []>("op_36284_cast")];
            tensor<fp16, []> var_36285_to_fp16 = const()[name = tensor<string, []>("op_36285_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3061_cast = mul(x = var_36284_cast, y = var_36285_to_fp16)[name = tensor<string, []>("aw_3061_cast")];
            tensor<string, []> var_36288_equation_0 = const()[name = tensor<string, []>("op_36288_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36288_cast = einsum(equation = var_36288_equation_0, values = (var_36130_cast, var_36047_cast))[name = tensor<string, []>("op_36288_cast")];
            tensor<fp16, []> var_36289_to_fp16 = const()[name = tensor<string, []>("op_36289_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3063_cast = mul(x = var_36288_cast, y = var_36289_to_fp16)[name = tensor<string, []>("aw_3063_cast")];
            tensor<string, []> var_36292_equation_0 = const()[name = tensor<string, []>("op_36292_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36292_cast = einsum(equation = var_36292_equation_0, values = (var_36134_cast, var_36051_cast))[name = tensor<string, []>("op_36292_cast")];
            tensor<fp16, []> var_36293_to_fp16 = const()[name = tensor<string, []>("op_36293_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3065_cast = mul(x = var_36292_cast, y = var_36293_to_fp16)[name = tensor<string, []>("aw_3065_cast")];
            tensor<string, []> var_36296_equation_0 = const()[name = tensor<string, []>("op_36296_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36296_cast = einsum(equation = var_36296_equation_0, values = (var_36138_cast, var_36055_cast))[name = tensor<string, []>("op_36296_cast")];
            tensor<fp16, []> var_36297_to_fp16 = const()[name = tensor<string, []>("op_36297_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3067_cast = mul(x = var_36296_cast, y = var_36297_to_fp16)[name = tensor<string, []>("aw_3067_cast")];
            tensor<string, []> var_36300_equation_0 = const()[name = tensor<string, []>("op_36300_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36300_cast = einsum(equation = var_36300_equation_0, values = (var_36142_cast, var_36059_cast))[name = tensor<string, []>("op_36300_cast")];
            tensor<fp16, []> var_36301_to_fp16 = const()[name = tensor<string, []>("op_36301_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3069_cast = mul(x = var_36300_cast, y = var_36301_to_fp16)[name = tensor<string, []>("aw_3069_cast")];
            tensor<string, []> var_36304_equation_0 = const()[name = tensor<string, []>("op_36304_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36304_cast = einsum(equation = var_36304_equation_0, values = (var_36146_cast, var_36063_cast))[name = tensor<string, []>("op_36304_cast")];
            tensor<fp16, []> var_36305_to_fp16 = const()[name = tensor<string, []>("op_36305_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3071_cast = mul(x = var_36304_cast, y = var_36305_to_fp16)[name = tensor<string, []>("aw_3071_cast")];
            tensor<string, []> var_36308_equation_0 = const()[name = tensor<string, []>("op_36308_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36308_cast = einsum(equation = var_36308_equation_0, values = (var_36150_cast, var_36067_cast))[name = tensor<string, []>("op_36308_cast")];
            tensor<fp16, []> var_36309_to_fp16 = const()[name = tensor<string, []>("op_36309_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3073_cast = mul(x = var_36308_cast, y = var_36309_to_fp16)[name = tensor<string, []>("aw_3073_cast")];
            tensor<string, []> var_36312_equation_0 = const()[name = tensor<string, []>("op_36312_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36312_cast = einsum(equation = var_36312_equation_0, values = (var_36154_cast, var_36071_cast))[name = tensor<string, []>("op_36312_cast")];
            tensor<fp16, []> var_36313_to_fp16 = const()[name = tensor<string, []>("op_36313_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3075_cast = mul(x = var_36312_cast, y = var_36313_to_fp16)[name = tensor<string, []>("aw_3075_cast")];
            tensor<string, []> var_36316_equation_0 = const()[name = tensor<string, []>("op_36316_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36316_cast = einsum(equation = var_36316_equation_0, values = (var_36158_cast, var_36075_cast))[name = tensor<string, []>("op_36316_cast")];
            tensor<fp16, []> var_36317_to_fp16 = const()[name = tensor<string, []>("op_36317_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3077_cast = mul(x = var_36316_cast, y = var_36317_to_fp16)[name = tensor<string, []>("aw_3077_cast")];
            tensor<string, []> var_36320_equation_0 = const()[name = tensor<string, []>("op_36320_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36320_cast = einsum(equation = var_36320_equation_0, values = (var_36162_cast, var_36079_cast))[name = tensor<string, []>("op_36320_cast")];
            tensor<fp16, []> var_36321_to_fp16 = const()[name = tensor<string, []>("op_36321_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3079_cast = mul(x = var_36320_cast, y = var_36321_to_fp16)[name = tensor<string, []>("aw_3079_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36323_cast = softmax(axis = var_30385, x = aw_3041_cast)[name = tensor<string, []>("op_36323_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36324_cast = softmax(axis = var_30385, x = aw_3043_cast)[name = tensor<string, []>("op_36324_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36325_cast = softmax(axis = var_30385, x = aw_3045_cast)[name = tensor<string, []>("op_36325_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36326_cast = softmax(axis = var_30385, x = aw_3047_cast)[name = tensor<string, []>("op_36326_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36327_cast = softmax(axis = var_30385, x = aw_3049_cast)[name = tensor<string, []>("op_36327_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36328_cast = softmax(axis = var_30385, x = aw_3051_cast)[name = tensor<string, []>("op_36328_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36329_cast = softmax(axis = var_30385, x = aw_3053_cast)[name = tensor<string, []>("op_36329_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36330_cast = softmax(axis = var_30385, x = aw_3055_cast)[name = tensor<string, []>("op_36330_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36331_cast = softmax(axis = var_30385, x = aw_3057_cast)[name = tensor<string, []>("op_36331_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36332_cast = softmax(axis = var_30385, x = aw_3059_cast)[name = tensor<string, []>("op_36332_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36333_cast = softmax(axis = var_30385, x = aw_3061_cast)[name = tensor<string, []>("op_36333_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36334_cast = softmax(axis = var_30385, x = aw_3063_cast)[name = tensor<string, []>("op_36334_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36335_cast = softmax(axis = var_30385, x = aw_3065_cast)[name = tensor<string, []>("op_36335_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36336_cast = softmax(axis = var_30385, x = aw_3067_cast)[name = tensor<string, []>("op_36336_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36337_cast = softmax(axis = var_30385, x = aw_3069_cast)[name = tensor<string, []>("op_36337_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36338_cast = softmax(axis = var_30385, x = aw_3071_cast)[name = tensor<string, []>("op_36338_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36339_cast = softmax(axis = var_30385, x = aw_3073_cast)[name = tensor<string, []>("op_36339_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36340_cast = softmax(axis = var_30385, x = aw_3075_cast)[name = tensor<string, []>("op_36340_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36341_cast = softmax(axis = var_30385, x = aw_3077_cast)[name = tensor<string, []>("op_36341_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_36342_cast = softmax(axis = var_30385, x = aw_3079_cast)[name = tensor<string, []>("op_36342_cast")];
            tensor<string, []> var_36344_equation_0 = const()[name = tensor<string, []>("op_36344_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36344_cast = einsum(equation = var_36344_equation_0, values = (var_36164_cast, var_36323_cast))[name = tensor<string, []>("op_36344_cast")];
            tensor<string, []> var_36346_equation_0 = const()[name = tensor<string, []>("op_36346_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36346_cast = einsum(equation = var_36346_equation_0, values = (var_36168_cast, var_36324_cast))[name = tensor<string, []>("op_36346_cast")];
            tensor<string, []> var_36348_equation_0 = const()[name = tensor<string, []>("op_36348_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36348_cast = einsum(equation = var_36348_equation_0, values = (var_36172_cast, var_36325_cast))[name = tensor<string, []>("op_36348_cast")];
            tensor<string, []> var_36350_equation_0 = const()[name = tensor<string, []>("op_36350_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36350_cast = einsum(equation = var_36350_equation_0, values = (var_36176_cast, var_36326_cast))[name = tensor<string, []>("op_36350_cast")];
            tensor<string, []> var_36352_equation_0 = const()[name = tensor<string, []>("op_36352_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36352_cast = einsum(equation = var_36352_equation_0, values = (var_36180_cast, var_36327_cast))[name = tensor<string, []>("op_36352_cast")];
            tensor<string, []> var_36354_equation_0 = const()[name = tensor<string, []>("op_36354_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36354_cast = einsum(equation = var_36354_equation_0, values = (var_36184_cast, var_36328_cast))[name = tensor<string, []>("op_36354_cast")];
            tensor<string, []> var_36356_equation_0 = const()[name = tensor<string, []>("op_36356_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36356_cast = einsum(equation = var_36356_equation_0, values = (var_36188_cast, var_36329_cast))[name = tensor<string, []>("op_36356_cast")];
            tensor<string, []> var_36358_equation_0 = const()[name = tensor<string, []>("op_36358_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36358_cast = einsum(equation = var_36358_equation_0, values = (var_36192_cast, var_36330_cast))[name = tensor<string, []>("op_36358_cast")];
            tensor<string, []> var_36360_equation_0 = const()[name = tensor<string, []>("op_36360_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36360_cast = einsum(equation = var_36360_equation_0, values = (var_36196_cast, var_36331_cast))[name = tensor<string, []>("op_36360_cast")];
            tensor<string, []> var_36362_equation_0 = const()[name = tensor<string, []>("op_36362_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36362_cast = einsum(equation = var_36362_equation_0, values = (var_36200_cast, var_36332_cast))[name = tensor<string, []>("op_36362_cast")];
            tensor<string, []> var_36364_equation_0 = const()[name = tensor<string, []>("op_36364_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36364_cast = einsum(equation = var_36364_equation_0, values = (var_36204_cast, var_36333_cast))[name = tensor<string, []>("op_36364_cast")];
            tensor<string, []> var_36366_equation_0 = const()[name = tensor<string, []>("op_36366_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36366_cast = einsum(equation = var_36366_equation_0, values = (var_36208_cast, var_36334_cast))[name = tensor<string, []>("op_36366_cast")];
            tensor<string, []> var_36368_equation_0 = const()[name = tensor<string, []>("op_36368_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36368_cast = einsum(equation = var_36368_equation_0, values = (var_36212_cast, var_36335_cast))[name = tensor<string, []>("op_36368_cast")];
            tensor<string, []> var_36370_equation_0 = const()[name = tensor<string, []>("op_36370_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36370_cast = einsum(equation = var_36370_equation_0, values = (var_36216_cast, var_36336_cast))[name = tensor<string, []>("op_36370_cast")];
            tensor<string, []> var_36372_equation_0 = const()[name = tensor<string, []>("op_36372_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36372_cast = einsum(equation = var_36372_equation_0, values = (var_36220_cast, var_36337_cast))[name = tensor<string, []>("op_36372_cast")];
            tensor<string, []> var_36374_equation_0 = const()[name = tensor<string, []>("op_36374_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36374_cast = einsum(equation = var_36374_equation_0, values = (var_36224_cast, var_36338_cast))[name = tensor<string, []>("op_36374_cast")];
            tensor<string, []> var_36376_equation_0 = const()[name = tensor<string, []>("op_36376_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36376_cast = einsum(equation = var_36376_equation_0, values = (var_36228_cast, var_36339_cast))[name = tensor<string, []>("op_36376_cast")];
            tensor<string, []> var_36378_equation_0 = const()[name = tensor<string, []>("op_36378_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36378_cast = einsum(equation = var_36378_equation_0, values = (var_36232_cast, var_36340_cast))[name = tensor<string, []>("op_36378_cast")];
            tensor<string, []> var_36380_equation_0 = const()[name = tensor<string, []>("op_36380_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36380_cast = einsum(equation = var_36380_equation_0, values = (var_36236_cast, var_36341_cast))[name = tensor<string, []>("op_36380_cast")];
            tensor<string, []> var_36382_equation_0 = const()[name = tensor<string, []>("op_36382_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36382_cast = einsum(equation = var_36382_equation_0, values = (var_36240_cast, var_36342_cast))[name = tensor<string, []>("op_36382_cast")];
            tensor<bool, []> input_489_interleave_0 = const()[name = tensor<string, []>("input_489_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_489_cast = concat(axis = var_30385, interleave = input_489_interleave_0, values = (var_36344_cast, var_36346_cast, var_36348_cast, var_36350_cast, var_36352_cast, var_36354_cast, var_36356_cast, var_36358_cast, var_36360_cast, var_36362_cast, var_36364_cast, var_36366_cast, var_36368_cast, var_36370_cast, var_36372_cast, var_36374_cast, var_36376_cast, var_36378_cast, var_36380_cast, var_36382_cast))[name = tensor<string, []>("input_489_cast")];
            tensor<int32, [2]> var_36388 = const()[name = tensor<string, []>("op_36388"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_36390 = const()[name = tensor<string, []>("op_36390"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_36392_pad_type_0 = const()[name = tensor<string, []>("op_36392_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_36392_pad_0 = const()[name = tensor<string, []>("op_36392_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3015334976)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3018611840)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_36392_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_out_0_bias_to_fp16, dilations = var_36390, groups = var_30385, pad = var_36392_pad_0, pad_type = var_36392_pad_type_0, strides = var_36388, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn1_to_out_0_weight_to_fp16, x = input_489_cast)[name = tensor<string, []>("op_36392_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_243_cast = add(x = var_36392_cast, y = inputs_241_cast)[name = tensor<string, []>("inputs_243_cast")];
            tensor<int32, [1]> var_36396 = const()[name = tensor<string, []>("op_36396"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_243_cast = reduce_mean(axes = var_36396, keep_dims = var_30380, x = inputs_243_cast)[name = tensor<string, []>("channels_mean_243_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_243_cast = sub(x = inputs_243_cast, y = channels_mean_243_cast)[name = tensor<string, []>("zero_mean_243_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_243_cast = mul(x = zero_mean_243_cast, y = zero_mean_243_cast)[name = tensor<string, []>("zero_mean_sq_243_cast")];
            tensor<int32, [1]> var_36400 = const()[name = tensor<string, []>("op_36400"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_36401_cast = reduce_mean(axes = var_36400, keep_dims = var_30380, x = zero_mean_sq_243_cast)[name = tensor<string, []>("op_36401_cast")];
            tensor<fp16, []> var_36402_to_fp16 = const()[name = tensor<string, []>("op_36402_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_36403_cast = add(x = var_36401_cast, y = var_36402_to_fp16)[name = tensor<string, []>("op_36403_cast")];
            tensor<fp16, []> denom_243_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_243_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_243_cast = rsqrt(epsilon = denom_243_epsilon_0_to_fp16, x = var_36403_cast)[name = tensor<string, []>("denom_243_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_243_cast = mul(x = zero_mean_243_cast, y = denom_243_cast)[name = tensor<string, []>("out_243_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_36407_to_fp16 = const()[name = tensor<string, []>("op_36407_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3018614464)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_36408_cast = add(x = out_243_cast, y = var_36407_to_fp16)[name = tensor<string, []>("op_36408_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_36410_to_fp16 = const()[name = tensor<string, []>("op_36410_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3018617088)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_325_cast = mul(x = var_36408_cast, y = var_36410_to_fp16)[name = tensor<string, []>("hidden_states_325_cast")];
            tensor<int32, [2]> var_36417 = const()[name = tensor<string, []>("op_36417"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_36419 = const()[name = tensor<string, []>("op_36419"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_163_pad_type_0 = const()[name = tensor<string, []>("q_163_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_163_pad_0 = const()[name = tensor<string, []>("q_163_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3018619712)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_163_cast = conv(dilations = var_36419, groups = var_30385, pad = q_163_pad_0, pad_type = q_163_pad_type_0, strides = var_36417, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_q_weight_to_fp16, x = hidden_states_325_cast)[name = tensor<string, []>("q_163_cast")];
            tensor<int32, [2]> var_36423 = const()[name = tensor<string, []>("op_36423"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_36425 = const()[name = tensor<string, []>("op_36425"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_325_pad_type_0 = const()[name = tensor<string, []>("k_325_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_325_pad_0 = const()[name = tensor<string, []>("k_325_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3021896576)))];
            tensor<fp16, [2, 1280, 1, 77]> k_325_cast = conv(dilations = var_36425, groups = var_30385, pad = k_325_pad_0, pad_type = k_325_pad_type_0, strides = var_36423, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_325_cast")];
            tensor<int32, [2]> var_36429 = const()[name = tensor<string, []>("op_36429"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_36431 = const()[name = tensor<string, []>("op_36431"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_163_pad_type_0 = const()[name = tensor<string, []>("v_163_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_163_pad_0 = const()[name = tensor<string, []>("v_163_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3027139520)))];
            tensor<fp16, [2, 1280, 1, 77]> v_163_cast = conv(dilations = var_36431, groups = var_30385, pad = v_163_pad_0, pad_type = v_163_pad_type_0, strides = var_36429, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_163_cast")];
            tensor<int32, [4]> var_36435_begin_0 = const()[name = tensor<string, []>("op_36435_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_36435_end_0 = const()[name = tensor<string, []>("op_36435_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_36435_end_mask_0 = const()[name = tensor<string, []>("op_36435_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36435_cast = slice_by_index(begin = var_36435_begin_0, end = var_36435_end_0, end_mask = var_36435_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36435_cast")];
            tensor<int32, [4]> var_36439_begin_0 = const()[name = tensor<string, []>("op_36439_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_36439_end_0 = const()[name = tensor<string, []>("op_36439_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_36439_end_mask_0 = const()[name = tensor<string, []>("op_36439_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36439_cast = slice_by_index(begin = var_36439_begin_0, end = var_36439_end_0, end_mask = var_36439_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36439_cast")];
            tensor<int32, [4]> var_36443_begin_0 = const()[name = tensor<string, []>("op_36443_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_36443_end_0 = const()[name = tensor<string, []>("op_36443_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_36443_end_mask_0 = const()[name = tensor<string, []>("op_36443_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36443_cast = slice_by_index(begin = var_36443_begin_0, end = var_36443_end_0, end_mask = var_36443_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36443_cast")];
            tensor<int32, [4]> var_36447_begin_0 = const()[name = tensor<string, []>("op_36447_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_36447_end_0 = const()[name = tensor<string, []>("op_36447_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_36447_end_mask_0 = const()[name = tensor<string, []>("op_36447_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36447_cast = slice_by_index(begin = var_36447_begin_0, end = var_36447_end_0, end_mask = var_36447_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36447_cast")];
            tensor<int32, [4]> var_36451_begin_0 = const()[name = tensor<string, []>("op_36451_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_36451_end_0 = const()[name = tensor<string, []>("op_36451_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_36451_end_mask_0 = const()[name = tensor<string, []>("op_36451_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36451_cast = slice_by_index(begin = var_36451_begin_0, end = var_36451_end_0, end_mask = var_36451_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36451_cast")];
            tensor<int32, [4]> var_36455_begin_0 = const()[name = tensor<string, []>("op_36455_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_36455_end_0 = const()[name = tensor<string, []>("op_36455_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_36455_end_mask_0 = const()[name = tensor<string, []>("op_36455_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36455_cast = slice_by_index(begin = var_36455_begin_0, end = var_36455_end_0, end_mask = var_36455_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36455_cast")];
            tensor<int32, [4]> var_36459_begin_0 = const()[name = tensor<string, []>("op_36459_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_36459_end_0 = const()[name = tensor<string, []>("op_36459_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_36459_end_mask_0 = const()[name = tensor<string, []>("op_36459_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36459_cast = slice_by_index(begin = var_36459_begin_0, end = var_36459_end_0, end_mask = var_36459_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36459_cast")];
            tensor<int32, [4]> var_36463_begin_0 = const()[name = tensor<string, []>("op_36463_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_36463_end_0 = const()[name = tensor<string, []>("op_36463_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_36463_end_mask_0 = const()[name = tensor<string, []>("op_36463_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36463_cast = slice_by_index(begin = var_36463_begin_0, end = var_36463_end_0, end_mask = var_36463_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36463_cast")];
            tensor<int32, [4]> var_36467_begin_0 = const()[name = tensor<string, []>("op_36467_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_36467_end_0 = const()[name = tensor<string, []>("op_36467_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_36467_end_mask_0 = const()[name = tensor<string, []>("op_36467_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36467_cast = slice_by_index(begin = var_36467_begin_0, end = var_36467_end_0, end_mask = var_36467_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36467_cast")];
            tensor<int32, [4]> var_36471_begin_0 = const()[name = tensor<string, []>("op_36471_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_36471_end_0 = const()[name = tensor<string, []>("op_36471_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_36471_end_mask_0 = const()[name = tensor<string, []>("op_36471_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36471_cast = slice_by_index(begin = var_36471_begin_0, end = var_36471_end_0, end_mask = var_36471_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36471_cast")];
            tensor<int32, [4]> var_36475_begin_0 = const()[name = tensor<string, []>("op_36475_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_36475_end_0 = const()[name = tensor<string, []>("op_36475_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_36475_end_mask_0 = const()[name = tensor<string, []>("op_36475_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36475_cast = slice_by_index(begin = var_36475_begin_0, end = var_36475_end_0, end_mask = var_36475_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36475_cast")];
            tensor<int32, [4]> var_36479_begin_0 = const()[name = tensor<string, []>("op_36479_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_36479_end_0 = const()[name = tensor<string, []>("op_36479_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_36479_end_mask_0 = const()[name = tensor<string, []>("op_36479_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36479_cast = slice_by_index(begin = var_36479_begin_0, end = var_36479_end_0, end_mask = var_36479_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36479_cast")];
            tensor<int32, [4]> var_36483_begin_0 = const()[name = tensor<string, []>("op_36483_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_36483_end_0 = const()[name = tensor<string, []>("op_36483_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_36483_end_mask_0 = const()[name = tensor<string, []>("op_36483_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36483_cast = slice_by_index(begin = var_36483_begin_0, end = var_36483_end_0, end_mask = var_36483_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36483_cast")];
            tensor<int32, [4]> var_36487_begin_0 = const()[name = tensor<string, []>("op_36487_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_36487_end_0 = const()[name = tensor<string, []>("op_36487_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_36487_end_mask_0 = const()[name = tensor<string, []>("op_36487_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36487_cast = slice_by_index(begin = var_36487_begin_0, end = var_36487_end_0, end_mask = var_36487_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36487_cast")];
            tensor<int32, [4]> var_36491_begin_0 = const()[name = tensor<string, []>("op_36491_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_36491_end_0 = const()[name = tensor<string, []>("op_36491_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_36491_end_mask_0 = const()[name = tensor<string, []>("op_36491_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36491_cast = slice_by_index(begin = var_36491_begin_0, end = var_36491_end_0, end_mask = var_36491_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36491_cast")];
            tensor<int32, [4]> var_36495_begin_0 = const()[name = tensor<string, []>("op_36495_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_36495_end_0 = const()[name = tensor<string, []>("op_36495_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_36495_end_mask_0 = const()[name = tensor<string, []>("op_36495_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36495_cast = slice_by_index(begin = var_36495_begin_0, end = var_36495_end_0, end_mask = var_36495_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36495_cast")];
            tensor<int32, [4]> var_36499_begin_0 = const()[name = tensor<string, []>("op_36499_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_36499_end_0 = const()[name = tensor<string, []>("op_36499_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_36499_end_mask_0 = const()[name = tensor<string, []>("op_36499_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36499_cast = slice_by_index(begin = var_36499_begin_0, end = var_36499_end_0, end_mask = var_36499_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36499_cast")];
            tensor<int32, [4]> var_36503_begin_0 = const()[name = tensor<string, []>("op_36503_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_36503_end_0 = const()[name = tensor<string, []>("op_36503_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_36503_end_mask_0 = const()[name = tensor<string, []>("op_36503_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36503_cast = slice_by_index(begin = var_36503_begin_0, end = var_36503_end_0, end_mask = var_36503_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36503_cast")];
            tensor<int32, [4]> var_36507_begin_0 = const()[name = tensor<string, []>("op_36507_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_36507_end_0 = const()[name = tensor<string, []>("op_36507_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_36507_end_mask_0 = const()[name = tensor<string, []>("op_36507_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36507_cast = slice_by_index(begin = var_36507_begin_0, end = var_36507_end_0, end_mask = var_36507_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36507_cast")];
            tensor<int32, [4]> var_36511_begin_0 = const()[name = tensor<string, []>("op_36511_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_36511_end_0 = const()[name = tensor<string, []>("op_36511_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_36511_end_mask_0 = const()[name = tensor<string, []>("op_36511_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36511_cast = slice_by_index(begin = var_36511_begin_0, end = var_36511_end_0, end_mask = var_36511_end_mask_0, x = q_163_cast)[name = tensor<string, []>("op_36511_cast")];
            tensor<int32, [4]> k_327_perm_0 = const()[name = tensor<string, []>("k_327_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_36518_begin_0 = const()[name = tensor<string, []>("op_36518_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_36518_end_0 = const()[name = tensor<string, []>("op_36518_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_36518_end_mask_0 = const()[name = tensor<string, []>("op_36518_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_58 = transpose(perm = k_327_perm_0, x = k_325_cast)[name = tensor<string, []>("transpose_58")];
            tensor<fp16, [2, 77, 1, 64]> var_36518_cast = slice_by_index(begin = var_36518_begin_0, end = var_36518_end_0, end_mask = var_36518_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36518_cast")];
            tensor<int32, [4]> var_36522_begin_0 = const()[name = tensor<string, []>("op_36522_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_36522_end_0 = const()[name = tensor<string, []>("op_36522_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_36522_end_mask_0 = const()[name = tensor<string, []>("op_36522_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36522_cast = slice_by_index(begin = var_36522_begin_0, end = var_36522_end_0, end_mask = var_36522_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36522_cast")];
            tensor<int32, [4]> var_36526_begin_0 = const()[name = tensor<string, []>("op_36526_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_36526_end_0 = const()[name = tensor<string, []>("op_36526_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_36526_end_mask_0 = const()[name = tensor<string, []>("op_36526_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36526_cast = slice_by_index(begin = var_36526_begin_0, end = var_36526_end_0, end_mask = var_36526_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36526_cast")];
            tensor<int32, [4]> var_36530_begin_0 = const()[name = tensor<string, []>("op_36530_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_36530_end_0 = const()[name = tensor<string, []>("op_36530_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_36530_end_mask_0 = const()[name = tensor<string, []>("op_36530_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36530_cast = slice_by_index(begin = var_36530_begin_0, end = var_36530_end_0, end_mask = var_36530_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36530_cast")];
            tensor<int32, [4]> var_36534_begin_0 = const()[name = tensor<string, []>("op_36534_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_36534_end_0 = const()[name = tensor<string, []>("op_36534_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_36534_end_mask_0 = const()[name = tensor<string, []>("op_36534_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36534_cast = slice_by_index(begin = var_36534_begin_0, end = var_36534_end_0, end_mask = var_36534_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36534_cast")];
            tensor<int32, [4]> var_36538_begin_0 = const()[name = tensor<string, []>("op_36538_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_36538_end_0 = const()[name = tensor<string, []>("op_36538_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_36538_end_mask_0 = const()[name = tensor<string, []>("op_36538_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36538_cast = slice_by_index(begin = var_36538_begin_0, end = var_36538_end_0, end_mask = var_36538_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36538_cast")];
            tensor<int32, [4]> var_36542_begin_0 = const()[name = tensor<string, []>("op_36542_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_36542_end_0 = const()[name = tensor<string, []>("op_36542_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_36542_end_mask_0 = const()[name = tensor<string, []>("op_36542_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36542_cast = slice_by_index(begin = var_36542_begin_0, end = var_36542_end_0, end_mask = var_36542_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36542_cast")];
            tensor<int32, [4]> var_36546_begin_0 = const()[name = tensor<string, []>("op_36546_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_36546_end_0 = const()[name = tensor<string, []>("op_36546_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_36546_end_mask_0 = const()[name = tensor<string, []>("op_36546_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36546_cast = slice_by_index(begin = var_36546_begin_0, end = var_36546_end_0, end_mask = var_36546_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36546_cast")];
            tensor<int32, [4]> var_36550_begin_0 = const()[name = tensor<string, []>("op_36550_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_36550_end_0 = const()[name = tensor<string, []>("op_36550_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_36550_end_mask_0 = const()[name = tensor<string, []>("op_36550_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36550_cast = slice_by_index(begin = var_36550_begin_0, end = var_36550_end_0, end_mask = var_36550_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36550_cast")];
            tensor<int32, [4]> var_36554_begin_0 = const()[name = tensor<string, []>("op_36554_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_36554_end_0 = const()[name = tensor<string, []>("op_36554_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_36554_end_mask_0 = const()[name = tensor<string, []>("op_36554_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36554_cast = slice_by_index(begin = var_36554_begin_0, end = var_36554_end_0, end_mask = var_36554_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36554_cast")];
            tensor<int32, [4]> var_36558_begin_0 = const()[name = tensor<string, []>("op_36558_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_36558_end_0 = const()[name = tensor<string, []>("op_36558_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_36558_end_mask_0 = const()[name = tensor<string, []>("op_36558_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36558_cast = slice_by_index(begin = var_36558_begin_0, end = var_36558_end_0, end_mask = var_36558_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36558_cast")];
            tensor<int32, [4]> var_36562_begin_0 = const()[name = tensor<string, []>("op_36562_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_36562_end_0 = const()[name = tensor<string, []>("op_36562_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_36562_end_mask_0 = const()[name = tensor<string, []>("op_36562_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36562_cast = slice_by_index(begin = var_36562_begin_0, end = var_36562_end_0, end_mask = var_36562_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36562_cast")];
            tensor<int32, [4]> var_36566_begin_0 = const()[name = tensor<string, []>("op_36566_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_36566_end_0 = const()[name = tensor<string, []>("op_36566_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_36566_end_mask_0 = const()[name = tensor<string, []>("op_36566_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36566_cast = slice_by_index(begin = var_36566_begin_0, end = var_36566_end_0, end_mask = var_36566_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36566_cast")];
            tensor<int32, [4]> var_36570_begin_0 = const()[name = tensor<string, []>("op_36570_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_36570_end_0 = const()[name = tensor<string, []>("op_36570_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_36570_end_mask_0 = const()[name = tensor<string, []>("op_36570_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36570_cast = slice_by_index(begin = var_36570_begin_0, end = var_36570_end_0, end_mask = var_36570_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36570_cast")];
            tensor<int32, [4]> var_36574_begin_0 = const()[name = tensor<string, []>("op_36574_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_36574_end_0 = const()[name = tensor<string, []>("op_36574_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_36574_end_mask_0 = const()[name = tensor<string, []>("op_36574_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36574_cast = slice_by_index(begin = var_36574_begin_0, end = var_36574_end_0, end_mask = var_36574_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36574_cast")];
            tensor<int32, [4]> var_36578_begin_0 = const()[name = tensor<string, []>("op_36578_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_36578_end_0 = const()[name = tensor<string, []>("op_36578_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_36578_end_mask_0 = const()[name = tensor<string, []>("op_36578_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36578_cast = slice_by_index(begin = var_36578_begin_0, end = var_36578_end_0, end_mask = var_36578_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36578_cast")];
            tensor<int32, [4]> var_36582_begin_0 = const()[name = tensor<string, []>("op_36582_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_36582_end_0 = const()[name = tensor<string, []>("op_36582_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_36582_end_mask_0 = const()[name = tensor<string, []>("op_36582_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36582_cast = slice_by_index(begin = var_36582_begin_0, end = var_36582_end_0, end_mask = var_36582_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36582_cast")];
            tensor<int32, [4]> var_36586_begin_0 = const()[name = tensor<string, []>("op_36586_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_36586_end_0 = const()[name = tensor<string, []>("op_36586_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_36586_end_mask_0 = const()[name = tensor<string, []>("op_36586_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36586_cast = slice_by_index(begin = var_36586_begin_0, end = var_36586_end_0, end_mask = var_36586_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36586_cast")];
            tensor<int32, [4]> var_36590_begin_0 = const()[name = tensor<string, []>("op_36590_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_36590_end_0 = const()[name = tensor<string, []>("op_36590_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_36590_end_mask_0 = const()[name = tensor<string, []>("op_36590_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36590_cast = slice_by_index(begin = var_36590_begin_0, end = var_36590_end_0, end_mask = var_36590_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36590_cast")];
            tensor<int32, [4]> var_36594_begin_0 = const()[name = tensor<string, []>("op_36594_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_36594_end_0 = const()[name = tensor<string, []>("op_36594_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_36594_end_mask_0 = const()[name = tensor<string, []>("op_36594_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_36594_cast = slice_by_index(begin = var_36594_begin_0, end = var_36594_end_0, end_mask = var_36594_end_mask_0, x = transpose_58)[name = tensor<string, []>("op_36594_cast")];
            tensor<int32, [4]> var_36596_begin_0 = const()[name = tensor<string, []>("op_36596_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_36596_end_0 = const()[name = tensor<string, []>("op_36596_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_36596_end_mask_0 = const()[name = tensor<string, []>("op_36596_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36596_cast = slice_by_index(begin = var_36596_begin_0, end = var_36596_end_0, end_mask = var_36596_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36596_cast")];
            tensor<int32, [4]> var_36600_begin_0 = const()[name = tensor<string, []>("op_36600_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_36600_end_0 = const()[name = tensor<string, []>("op_36600_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_36600_end_mask_0 = const()[name = tensor<string, []>("op_36600_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36600_cast = slice_by_index(begin = var_36600_begin_0, end = var_36600_end_0, end_mask = var_36600_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36600_cast")];
            tensor<int32, [4]> var_36604_begin_0 = const()[name = tensor<string, []>("op_36604_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_36604_end_0 = const()[name = tensor<string, []>("op_36604_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_36604_end_mask_0 = const()[name = tensor<string, []>("op_36604_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36604_cast = slice_by_index(begin = var_36604_begin_0, end = var_36604_end_0, end_mask = var_36604_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36604_cast")];
            tensor<int32, [4]> var_36608_begin_0 = const()[name = tensor<string, []>("op_36608_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_36608_end_0 = const()[name = tensor<string, []>("op_36608_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_36608_end_mask_0 = const()[name = tensor<string, []>("op_36608_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36608_cast = slice_by_index(begin = var_36608_begin_0, end = var_36608_end_0, end_mask = var_36608_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36608_cast")];
            tensor<int32, [4]> var_36612_begin_0 = const()[name = tensor<string, []>("op_36612_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_36612_end_0 = const()[name = tensor<string, []>("op_36612_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_36612_end_mask_0 = const()[name = tensor<string, []>("op_36612_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36612_cast = slice_by_index(begin = var_36612_begin_0, end = var_36612_end_0, end_mask = var_36612_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36612_cast")];
            tensor<int32, [4]> var_36616_begin_0 = const()[name = tensor<string, []>("op_36616_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_36616_end_0 = const()[name = tensor<string, []>("op_36616_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_36616_end_mask_0 = const()[name = tensor<string, []>("op_36616_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36616_cast = slice_by_index(begin = var_36616_begin_0, end = var_36616_end_0, end_mask = var_36616_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36616_cast")];
            tensor<int32, [4]> var_36620_begin_0 = const()[name = tensor<string, []>("op_36620_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_36620_end_0 = const()[name = tensor<string, []>("op_36620_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_36620_end_mask_0 = const()[name = tensor<string, []>("op_36620_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36620_cast = slice_by_index(begin = var_36620_begin_0, end = var_36620_end_0, end_mask = var_36620_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36620_cast")];
            tensor<int32, [4]> var_36624_begin_0 = const()[name = tensor<string, []>("op_36624_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_36624_end_0 = const()[name = tensor<string, []>("op_36624_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_36624_end_mask_0 = const()[name = tensor<string, []>("op_36624_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36624_cast = slice_by_index(begin = var_36624_begin_0, end = var_36624_end_0, end_mask = var_36624_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36624_cast")];
            tensor<int32, [4]> var_36628_begin_0 = const()[name = tensor<string, []>("op_36628_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_36628_end_0 = const()[name = tensor<string, []>("op_36628_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_36628_end_mask_0 = const()[name = tensor<string, []>("op_36628_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36628_cast = slice_by_index(begin = var_36628_begin_0, end = var_36628_end_0, end_mask = var_36628_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36628_cast")];
            tensor<int32, [4]> var_36632_begin_0 = const()[name = tensor<string, []>("op_36632_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_36632_end_0 = const()[name = tensor<string, []>("op_36632_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_36632_end_mask_0 = const()[name = tensor<string, []>("op_36632_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36632_cast = slice_by_index(begin = var_36632_begin_0, end = var_36632_end_0, end_mask = var_36632_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36632_cast")];
            tensor<int32, [4]> var_36636_begin_0 = const()[name = tensor<string, []>("op_36636_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_36636_end_0 = const()[name = tensor<string, []>("op_36636_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_36636_end_mask_0 = const()[name = tensor<string, []>("op_36636_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36636_cast = slice_by_index(begin = var_36636_begin_0, end = var_36636_end_0, end_mask = var_36636_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36636_cast")];
            tensor<int32, [4]> var_36640_begin_0 = const()[name = tensor<string, []>("op_36640_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_36640_end_0 = const()[name = tensor<string, []>("op_36640_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_36640_end_mask_0 = const()[name = tensor<string, []>("op_36640_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36640_cast = slice_by_index(begin = var_36640_begin_0, end = var_36640_end_0, end_mask = var_36640_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36640_cast")];
            tensor<int32, [4]> var_36644_begin_0 = const()[name = tensor<string, []>("op_36644_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_36644_end_0 = const()[name = tensor<string, []>("op_36644_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_36644_end_mask_0 = const()[name = tensor<string, []>("op_36644_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36644_cast = slice_by_index(begin = var_36644_begin_0, end = var_36644_end_0, end_mask = var_36644_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36644_cast")];
            tensor<int32, [4]> var_36648_begin_0 = const()[name = tensor<string, []>("op_36648_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_36648_end_0 = const()[name = tensor<string, []>("op_36648_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_36648_end_mask_0 = const()[name = tensor<string, []>("op_36648_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36648_cast = slice_by_index(begin = var_36648_begin_0, end = var_36648_end_0, end_mask = var_36648_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36648_cast")];
            tensor<int32, [4]> var_36652_begin_0 = const()[name = tensor<string, []>("op_36652_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_36652_end_0 = const()[name = tensor<string, []>("op_36652_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_36652_end_mask_0 = const()[name = tensor<string, []>("op_36652_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36652_cast = slice_by_index(begin = var_36652_begin_0, end = var_36652_end_0, end_mask = var_36652_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36652_cast")];
            tensor<int32, [4]> var_36656_begin_0 = const()[name = tensor<string, []>("op_36656_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_36656_end_0 = const()[name = tensor<string, []>("op_36656_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_36656_end_mask_0 = const()[name = tensor<string, []>("op_36656_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36656_cast = slice_by_index(begin = var_36656_begin_0, end = var_36656_end_0, end_mask = var_36656_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36656_cast")];
            tensor<int32, [4]> var_36660_begin_0 = const()[name = tensor<string, []>("op_36660_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_36660_end_0 = const()[name = tensor<string, []>("op_36660_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_36660_end_mask_0 = const()[name = tensor<string, []>("op_36660_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36660_cast = slice_by_index(begin = var_36660_begin_0, end = var_36660_end_0, end_mask = var_36660_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36660_cast")];
            tensor<int32, [4]> var_36664_begin_0 = const()[name = tensor<string, []>("op_36664_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_36664_end_0 = const()[name = tensor<string, []>("op_36664_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_36664_end_mask_0 = const()[name = tensor<string, []>("op_36664_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36664_cast = slice_by_index(begin = var_36664_begin_0, end = var_36664_end_0, end_mask = var_36664_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36664_cast")];
            tensor<int32, [4]> var_36668_begin_0 = const()[name = tensor<string, []>("op_36668_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_36668_end_0 = const()[name = tensor<string, []>("op_36668_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_36668_end_mask_0 = const()[name = tensor<string, []>("op_36668_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36668_cast = slice_by_index(begin = var_36668_begin_0, end = var_36668_end_0, end_mask = var_36668_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36668_cast")];
            tensor<int32, [4]> var_36672_begin_0 = const()[name = tensor<string, []>("op_36672_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_36672_end_0 = const()[name = tensor<string, []>("op_36672_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_36672_end_mask_0 = const()[name = tensor<string, []>("op_36672_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_36672_cast = slice_by_index(begin = var_36672_begin_0, end = var_36672_end_0, end_mask = var_36672_end_mask_0, x = v_163_cast)[name = tensor<string, []>("op_36672_cast")];
            tensor<string, []> var_36676_equation_0 = const()[name = tensor<string, []>("op_36676_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36676_cast = einsum(equation = var_36676_equation_0, values = (var_36518_cast, var_36435_cast))[name = tensor<string, []>("op_36676_cast")];
            tensor<fp16, []> var_36677_to_fp16 = const()[name = tensor<string, []>("op_36677_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3081_cast = mul(x = var_36676_cast, y = var_36677_to_fp16)[name = tensor<string, []>("aw_3081_cast")];
            tensor<string, []> var_36680_equation_0 = const()[name = tensor<string, []>("op_36680_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36680_cast = einsum(equation = var_36680_equation_0, values = (var_36522_cast, var_36439_cast))[name = tensor<string, []>("op_36680_cast")];
            tensor<fp16, []> var_36681_to_fp16 = const()[name = tensor<string, []>("op_36681_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3083_cast = mul(x = var_36680_cast, y = var_36681_to_fp16)[name = tensor<string, []>("aw_3083_cast")];
            tensor<string, []> var_36684_equation_0 = const()[name = tensor<string, []>("op_36684_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36684_cast = einsum(equation = var_36684_equation_0, values = (var_36526_cast, var_36443_cast))[name = tensor<string, []>("op_36684_cast")];
            tensor<fp16, []> var_36685_to_fp16 = const()[name = tensor<string, []>("op_36685_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3085_cast = mul(x = var_36684_cast, y = var_36685_to_fp16)[name = tensor<string, []>("aw_3085_cast")];
            tensor<string, []> var_36688_equation_0 = const()[name = tensor<string, []>("op_36688_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36688_cast = einsum(equation = var_36688_equation_0, values = (var_36530_cast, var_36447_cast))[name = tensor<string, []>("op_36688_cast")];
            tensor<fp16, []> var_36689_to_fp16 = const()[name = tensor<string, []>("op_36689_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3087_cast = mul(x = var_36688_cast, y = var_36689_to_fp16)[name = tensor<string, []>("aw_3087_cast")];
            tensor<string, []> var_36692_equation_0 = const()[name = tensor<string, []>("op_36692_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36692_cast = einsum(equation = var_36692_equation_0, values = (var_36534_cast, var_36451_cast))[name = tensor<string, []>("op_36692_cast")];
            tensor<fp16, []> var_36693_to_fp16 = const()[name = tensor<string, []>("op_36693_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3089_cast = mul(x = var_36692_cast, y = var_36693_to_fp16)[name = tensor<string, []>("aw_3089_cast")];
            tensor<string, []> var_36696_equation_0 = const()[name = tensor<string, []>("op_36696_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36696_cast = einsum(equation = var_36696_equation_0, values = (var_36538_cast, var_36455_cast))[name = tensor<string, []>("op_36696_cast")];
            tensor<fp16, []> var_36697_to_fp16 = const()[name = tensor<string, []>("op_36697_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3091_cast = mul(x = var_36696_cast, y = var_36697_to_fp16)[name = tensor<string, []>("aw_3091_cast")];
            tensor<string, []> var_36700_equation_0 = const()[name = tensor<string, []>("op_36700_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36700_cast = einsum(equation = var_36700_equation_0, values = (var_36542_cast, var_36459_cast))[name = tensor<string, []>("op_36700_cast")];
            tensor<fp16, []> var_36701_to_fp16 = const()[name = tensor<string, []>("op_36701_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3093_cast = mul(x = var_36700_cast, y = var_36701_to_fp16)[name = tensor<string, []>("aw_3093_cast")];
            tensor<string, []> var_36704_equation_0 = const()[name = tensor<string, []>("op_36704_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36704_cast = einsum(equation = var_36704_equation_0, values = (var_36546_cast, var_36463_cast))[name = tensor<string, []>("op_36704_cast")];
            tensor<fp16, []> var_36705_to_fp16 = const()[name = tensor<string, []>("op_36705_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3095_cast = mul(x = var_36704_cast, y = var_36705_to_fp16)[name = tensor<string, []>("aw_3095_cast")];
            tensor<string, []> var_36708_equation_0 = const()[name = tensor<string, []>("op_36708_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36708_cast = einsum(equation = var_36708_equation_0, values = (var_36550_cast, var_36467_cast))[name = tensor<string, []>("op_36708_cast")];
            tensor<fp16, []> var_36709_to_fp16 = const()[name = tensor<string, []>("op_36709_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3097_cast = mul(x = var_36708_cast, y = var_36709_to_fp16)[name = tensor<string, []>("aw_3097_cast")];
            tensor<string, []> var_36712_equation_0 = const()[name = tensor<string, []>("op_36712_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36712_cast = einsum(equation = var_36712_equation_0, values = (var_36554_cast, var_36471_cast))[name = tensor<string, []>("op_36712_cast")];
            tensor<fp16, []> var_36713_to_fp16 = const()[name = tensor<string, []>("op_36713_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3099_cast = mul(x = var_36712_cast, y = var_36713_to_fp16)[name = tensor<string, []>("aw_3099_cast")];
            tensor<string, []> var_36716_equation_0 = const()[name = tensor<string, []>("op_36716_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36716_cast = einsum(equation = var_36716_equation_0, values = (var_36558_cast, var_36475_cast))[name = tensor<string, []>("op_36716_cast")];
            tensor<fp16, []> var_36717_to_fp16 = const()[name = tensor<string, []>("op_36717_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3101_cast = mul(x = var_36716_cast, y = var_36717_to_fp16)[name = tensor<string, []>("aw_3101_cast")];
            tensor<string, []> var_36720_equation_0 = const()[name = tensor<string, []>("op_36720_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36720_cast = einsum(equation = var_36720_equation_0, values = (var_36562_cast, var_36479_cast))[name = tensor<string, []>("op_36720_cast")];
            tensor<fp16, []> var_36721_to_fp16 = const()[name = tensor<string, []>("op_36721_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3103_cast = mul(x = var_36720_cast, y = var_36721_to_fp16)[name = tensor<string, []>("aw_3103_cast")];
            tensor<string, []> var_36724_equation_0 = const()[name = tensor<string, []>("op_36724_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36724_cast = einsum(equation = var_36724_equation_0, values = (var_36566_cast, var_36483_cast))[name = tensor<string, []>("op_36724_cast")];
            tensor<fp16, []> var_36725_to_fp16 = const()[name = tensor<string, []>("op_36725_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3105_cast = mul(x = var_36724_cast, y = var_36725_to_fp16)[name = tensor<string, []>("aw_3105_cast")];
            tensor<string, []> var_36728_equation_0 = const()[name = tensor<string, []>("op_36728_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36728_cast = einsum(equation = var_36728_equation_0, values = (var_36570_cast, var_36487_cast))[name = tensor<string, []>("op_36728_cast")];
            tensor<fp16, []> var_36729_to_fp16 = const()[name = tensor<string, []>("op_36729_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3107_cast = mul(x = var_36728_cast, y = var_36729_to_fp16)[name = tensor<string, []>("aw_3107_cast")];
            tensor<string, []> var_36732_equation_0 = const()[name = tensor<string, []>("op_36732_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36732_cast = einsum(equation = var_36732_equation_0, values = (var_36574_cast, var_36491_cast))[name = tensor<string, []>("op_36732_cast")];
            tensor<fp16, []> var_36733_to_fp16 = const()[name = tensor<string, []>("op_36733_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3109_cast = mul(x = var_36732_cast, y = var_36733_to_fp16)[name = tensor<string, []>("aw_3109_cast")];
            tensor<string, []> var_36736_equation_0 = const()[name = tensor<string, []>("op_36736_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36736_cast = einsum(equation = var_36736_equation_0, values = (var_36578_cast, var_36495_cast))[name = tensor<string, []>("op_36736_cast")];
            tensor<fp16, []> var_36737_to_fp16 = const()[name = tensor<string, []>("op_36737_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3111_cast = mul(x = var_36736_cast, y = var_36737_to_fp16)[name = tensor<string, []>("aw_3111_cast")];
            tensor<string, []> var_36740_equation_0 = const()[name = tensor<string, []>("op_36740_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36740_cast = einsum(equation = var_36740_equation_0, values = (var_36582_cast, var_36499_cast))[name = tensor<string, []>("op_36740_cast")];
            tensor<fp16, []> var_36741_to_fp16 = const()[name = tensor<string, []>("op_36741_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3113_cast = mul(x = var_36740_cast, y = var_36741_to_fp16)[name = tensor<string, []>("aw_3113_cast")];
            tensor<string, []> var_36744_equation_0 = const()[name = tensor<string, []>("op_36744_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36744_cast = einsum(equation = var_36744_equation_0, values = (var_36586_cast, var_36503_cast))[name = tensor<string, []>("op_36744_cast")];
            tensor<fp16, []> var_36745_to_fp16 = const()[name = tensor<string, []>("op_36745_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3115_cast = mul(x = var_36744_cast, y = var_36745_to_fp16)[name = tensor<string, []>("aw_3115_cast")];
            tensor<string, []> var_36748_equation_0 = const()[name = tensor<string, []>("op_36748_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36748_cast = einsum(equation = var_36748_equation_0, values = (var_36590_cast, var_36507_cast))[name = tensor<string, []>("op_36748_cast")];
            tensor<fp16, []> var_36749_to_fp16 = const()[name = tensor<string, []>("op_36749_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3117_cast = mul(x = var_36748_cast, y = var_36749_to_fp16)[name = tensor<string, []>("aw_3117_cast")];
            tensor<string, []> var_36752_equation_0 = const()[name = tensor<string, []>("op_36752_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_36752_cast = einsum(equation = var_36752_equation_0, values = (var_36594_cast, var_36511_cast))[name = tensor<string, []>("op_36752_cast")];
            tensor<fp16, []> var_36753_to_fp16 = const()[name = tensor<string, []>("op_36753_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3119_cast = mul(x = var_36752_cast, y = var_36753_to_fp16)[name = tensor<string, []>("aw_3119_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36755_cast = softmax(axis = var_30385, x = aw_3081_cast)[name = tensor<string, []>("op_36755_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36756_cast = softmax(axis = var_30385, x = aw_3083_cast)[name = tensor<string, []>("op_36756_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36757_cast = softmax(axis = var_30385, x = aw_3085_cast)[name = tensor<string, []>("op_36757_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36758_cast = softmax(axis = var_30385, x = aw_3087_cast)[name = tensor<string, []>("op_36758_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36759_cast = softmax(axis = var_30385, x = aw_3089_cast)[name = tensor<string, []>("op_36759_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36760_cast = softmax(axis = var_30385, x = aw_3091_cast)[name = tensor<string, []>("op_36760_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36761_cast = softmax(axis = var_30385, x = aw_3093_cast)[name = tensor<string, []>("op_36761_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36762_cast = softmax(axis = var_30385, x = aw_3095_cast)[name = tensor<string, []>("op_36762_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36763_cast = softmax(axis = var_30385, x = aw_3097_cast)[name = tensor<string, []>("op_36763_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36764_cast = softmax(axis = var_30385, x = aw_3099_cast)[name = tensor<string, []>("op_36764_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36765_cast = softmax(axis = var_30385, x = aw_3101_cast)[name = tensor<string, []>("op_36765_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36766_cast = softmax(axis = var_30385, x = aw_3103_cast)[name = tensor<string, []>("op_36766_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36767_cast = softmax(axis = var_30385, x = aw_3105_cast)[name = tensor<string, []>("op_36767_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36768_cast = softmax(axis = var_30385, x = aw_3107_cast)[name = tensor<string, []>("op_36768_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36769_cast = softmax(axis = var_30385, x = aw_3109_cast)[name = tensor<string, []>("op_36769_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36770_cast = softmax(axis = var_30385, x = aw_3111_cast)[name = tensor<string, []>("op_36770_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36771_cast = softmax(axis = var_30385, x = aw_3113_cast)[name = tensor<string, []>("op_36771_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36772_cast = softmax(axis = var_30385, x = aw_3115_cast)[name = tensor<string, []>("op_36772_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36773_cast = softmax(axis = var_30385, x = aw_3117_cast)[name = tensor<string, []>("op_36773_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_36774_cast = softmax(axis = var_30385, x = aw_3119_cast)[name = tensor<string, []>("op_36774_cast")];
            tensor<string, []> var_36776_equation_0 = const()[name = tensor<string, []>("op_36776_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36776_cast = einsum(equation = var_36776_equation_0, values = (var_36596_cast, var_36755_cast))[name = tensor<string, []>("op_36776_cast")];
            tensor<string, []> var_36778_equation_0 = const()[name = tensor<string, []>("op_36778_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36778_cast = einsum(equation = var_36778_equation_0, values = (var_36600_cast, var_36756_cast))[name = tensor<string, []>("op_36778_cast")];
            tensor<string, []> var_36780_equation_0 = const()[name = tensor<string, []>("op_36780_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36780_cast = einsum(equation = var_36780_equation_0, values = (var_36604_cast, var_36757_cast))[name = tensor<string, []>("op_36780_cast")];
            tensor<string, []> var_36782_equation_0 = const()[name = tensor<string, []>("op_36782_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36782_cast = einsum(equation = var_36782_equation_0, values = (var_36608_cast, var_36758_cast))[name = tensor<string, []>("op_36782_cast")];
            tensor<string, []> var_36784_equation_0 = const()[name = tensor<string, []>("op_36784_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36784_cast = einsum(equation = var_36784_equation_0, values = (var_36612_cast, var_36759_cast))[name = tensor<string, []>("op_36784_cast")];
            tensor<string, []> var_36786_equation_0 = const()[name = tensor<string, []>("op_36786_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36786_cast = einsum(equation = var_36786_equation_0, values = (var_36616_cast, var_36760_cast))[name = tensor<string, []>("op_36786_cast")];
            tensor<string, []> var_36788_equation_0 = const()[name = tensor<string, []>("op_36788_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36788_cast = einsum(equation = var_36788_equation_0, values = (var_36620_cast, var_36761_cast))[name = tensor<string, []>("op_36788_cast")];
            tensor<string, []> var_36790_equation_0 = const()[name = tensor<string, []>("op_36790_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36790_cast = einsum(equation = var_36790_equation_0, values = (var_36624_cast, var_36762_cast))[name = tensor<string, []>("op_36790_cast")];
            tensor<string, []> var_36792_equation_0 = const()[name = tensor<string, []>("op_36792_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36792_cast = einsum(equation = var_36792_equation_0, values = (var_36628_cast, var_36763_cast))[name = tensor<string, []>("op_36792_cast")];
            tensor<string, []> var_36794_equation_0 = const()[name = tensor<string, []>("op_36794_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36794_cast = einsum(equation = var_36794_equation_0, values = (var_36632_cast, var_36764_cast))[name = tensor<string, []>("op_36794_cast")];
            tensor<string, []> var_36796_equation_0 = const()[name = tensor<string, []>("op_36796_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36796_cast = einsum(equation = var_36796_equation_0, values = (var_36636_cast, var_36765_cast))[name = tensor<string, []>("op_36796_cast")];
            tensor<string, []> var_36798_equation_0 = const()[name = tensor<string, []>("op_36798_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36798_cast = einsum(equation = var_36798_equation_0, values = (var_36640_cast, var_36766_cast))[name = tensor<string, []>("op_36798_cast")];
            tensor<string, []> var_36800_equation_0 = const()[name = tensor<string, []>("op_36800_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36800_cast = einsum(equation = var_36800_equation_0, values = (var_36644_cast, var_36767_cast))[name = tensor<string, []>("op_36800_cast")];
            tensor<string, []> var_36802_equation_0 = const()[name = tensor<string, []>("op_36802_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36802_cast = einsum(equation = var_36802_equation_0, values = (var_36648_cast, var_36768_cast))[name = tensor<string, []>("op_36802_cast")];
            tensor<string, []> var_36804_equation_0 = const()[name = tensor<string, []>("op_36804_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36804_cast = einsum(equation = var_36804_equation_0, values = (var_36652_cast, var_36769_cast))[name = tensor<string, []>("op_36804_cast")];
            tensor<string, []> var_36806_equation_0 = const()[name = tensor<string, []>("op_36806_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36806_cast = einsum(equation = var_36806_equation_0, values = (var_36656_cast, var_36770_cast))[name = tensor<string, []>("op_36806_cast")];
            tensor<string, []> var_36808_equation_0 = const()[name = tensor<string, []>("op_36808_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36808_cast = einsum(equation = var_36808_equation_0, values = (var_36660_cast, var_36771_cast))[name = tensor<string, []>("op_36808_cast")];
            tensor<string, []> var_36810_equation_0 = const()[name = tensor<string, []>("op_36810_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36810_cast = einsum(equation = var_36810_equation_0, values = (var_36664_cast, var_36772_cast))[name = tensor<string, []>("op_36810_cast")];
            tensor<string, []> var_36812_equation_0 = const()[name = tensor<string, []>("op_36812_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36812_cast = einsum(equation = var_36812_equation_0, values = (var_36668_cast, var_36773_cast))[name = tensor<string, []>("op_36812_cast")];
            tensor<string, []> var_36814_equation_0 = const()[name = tensor<string, []>("op_36814_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_36814_cast = einsum(equation = var_36814_equation_0, values = (var_36672_cast, var_36774_cast))[name = tensor<string, []>("op_36814_cast")];
            tensor<bool, []> input_491_interleave_0 = const()[name = tensor<string, []>("input_491_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_491_cast = concat(axis = var_30385, interleave = input_491_interleave_0, values = (var_36776_cast, var_36778_cast, var_36780_cast, var_36782_cast, var_36784_cast, var_36786_cast, var_36788_cast, var_36790_cast, var_36792_cast, var_36794_cast, var_36796_cast, var_36798_cast, var_36800_cast, var_36802_cast, var_36804_cast, var_36806_cast, var_36808_cast, var_36810_cast, var_36812_cast, var_36814_cast))[name = tensor<string, []>("input_491_cast")];
            tensor<int32, [2]> var_36820 = const()[name = tensor<string, []>("op_36820"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_36822 = const()[name = tensor<string, []>("op_36822"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_36824_pad_type_0 = const()[name = tensor<string, []>("op_36824_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_36824_pad_0 = const()[name = tensor<string, []>("op_36824_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3032382464)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3035659328)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_36824_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_out_0_bias_to_fp16, dilations = var_36822, groups = var_30385, pad = var_36824_pad_0, pad_type = var_36824_pad_type_0, strides = var_36820, weight = up_blocks_0_attentions_0_transformer_blocks_6_attn2_to_out_0_weight_to_fp16, x = input_491_cast)[name = tensor<string, []>("op_36824_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_245_cast = add(x = var_36824_cast, y = inputs_243_cast)[name = tensor<string, []>("inputs_245_cast")];
            tensor<int32, [1]> var_36828 = const()[name = tensor<string, []>("op_36828"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_245_cast = reduce_mean(axes = var_36828, keep_dims = var_30380, x = inputs_245_cast)[name = tensor<string, []>("channels_mean_245_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_245_cast = sub(x = inputs_245_cast, y = channels_mean_245_cast)[name = tensor<string, []>("zero_mean_245_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_245_cast = mul(x = zero_mean_245_cast, y = zero_mean_245_cast)[name = tensor<string, []>("zero_mean_sq_245_cast")];
            tensor<int32, [1]> var_36832 = const()[name = tensor<string, []>("op_36832"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_36833_cast = reduce_mean(axes = var_36832, keep_dims = var_30380, x = zero_mean_sq_245_cast)[name = tensor<string, []>("op_36833_cast")];
            tensor<fp16, []> var_36834_to_fp16 = const()[name = tensor<string, []>("op_36834_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_36835_cast = add(x = var_36833_cast, y = var_36834_to_fp16)[name = tensor<string, []>("op_36835_cast")];
            tensor<fp16, []> denom_245_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_245_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_245_cast = rsqrt(epsilon = denom_245_epsilon_0_to_fp16, x = var_36835_cast)[name = tensor<string, []>("denom_245_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_245_cast = mul(x = zero_mean_245_cast, y = denom_245_cast)[name = tensor<string, []>("out_245_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_36839_to_fp16 = const()[name = tensor<string, []>("op_36839_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3035661952)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_36840_cast = add(x = out_245_cast, y = var_36839_to_fp16)[name = tensor<string, []>("op_36840_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_36842_to_fp16 = const()[name = tensor<string, []>("op_36842_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3035664576)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_493_cast = mul(x = var_36840_cast, y = var_36842_to_fp16)[name = tensor<string, []>("input_493_cast")];
            tensor<int32, [2]> var_36850 = const()[name = tensor<string, []>("op_36850"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_36852 = const()[name = tensor<string, []>("op_36852"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_36854_pad_type_0 = const()[name = tensor<string, []>("op_36854_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_36854_pad_0 = const()[name = tensor<string, []>("op_36854_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3035667200)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3061881664)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_36854_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_6_ff_net_0_proj_bias_to_fp16, dilations = var_36852, groups = var_30385, pad = var_36854_pad_0, pad_type = var_36854_pad_type_0, strides = var_36850, weight = up_blocks_0_attentions_0_transformer_blocks_6_ff_net_0_proj_weight_to_fp16, x = input_493_cast)[name = tensor<string, []>("op_36854_cast")];
            tensor<int32, [2]> var_36855_split_sizes_0 = const()[name = tensor<string, []>("op_36855_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_36855_axis_0 = const()[name = tensor<string, []>("op_36855_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_36855_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_36855_cast_1 = split(axis = var_36855_axis_0, split_sizes = var_36855_split_sizes_0, x = var_36854_cast)[name = tensor<string, []>("op_36855_cast")];
            tensor<string, []> var_36857_mode_0 = const()[name = tensor<string, []>("op_36857_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_36857_cast = gelu(mode = var_36857_mode_0, x = var_36855_cast_1)[name = tensor<string, []>("op_36857_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_495_cast = mul(x = var_36855_cast_0, y = var_36857_cast)[name = tensor<string, []>("input_495_cast")];
            tensor<int32, [2]> var_36861 = const()[name = tensor<string, []>("op_36861"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_36863 = const()[name = tensor<string, []>("op_36863"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_36865_pad_type_0 = const()[name = tensor<string, []>("op_36865_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_36865_pad_0 = const()[name = tensor<string, []>("op_36865_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3061902208)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3075009472)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_36865_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_6_ff_net_2_bias_to_fp16, dilations = var_36863, groups = var_30385, pad = var_36865_pad_0, pad_type = var_36865_pad_type_0, strides = var_36861, weight = up_blocks_0_attentions_0_transformer_blocks_6_ff_net_2_weight_to_fp16, x = input_495_cast)[name = tensor<string, []>("op_36865_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_247_cast = add(x = var_36865_cast, y = inputs_245_cast)[name = tensor<string, []>("inputs_247_cast")];
            tensor<int32, [1]> var_36875 = const()[name = tensor<string, []>("op_36875"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_247_cast = reduce_mean(axes = var_36875, keep_dims = var_30380, x = inputs_247_cast)[name = tensor<string, []>("channels_mean_247_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_247_cast = sub(x = inputs_247_cast, y = channels_mean_247_cast)[name = tensor<string, []>("zero_mean_247_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_247_cast = mul(x = zero_mean_247_cast, y = zero_mean_247_cast)[name = tensor<string, []>("zero_mean_sq_247_cast")];
            tensor<int32, [1]> var_36879 = const()[name = tensor<string, []>("op_36879"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_36880_cast = reduce_mean(axes = var_36879, keep_dims = var_30380, x = zero_mean_sq_247_cast)[name = tensor<string, []>("op_36880_cast")];
            tensor<fp16, []> var_36881_to_fp16 = const()[name = tensor<string, []>("op_36881_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_36882_cast = add(x = var_36880_cast, y = var_36881_to_fp16)[name = tensor<string, []>("op_36882_cast")];
            tensor<fp16, []> denom_247_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_247_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_247_cast = rsqrt(epsilon = denom_247_epsilon_0_to_fp16, x = var_36882_cast)[name = tensor<string, []>("denom_247_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_247_cast = mul(x = zero_mean_247_cast, y = denom_247_cast)[name = tensor<string, []>("out_247_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_36886_to_fp16 = const()[name = tensor<string, []>("op_36886_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3075012096)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_36887_cast = add(x = out_247_cast, y = var_36886_to_fp16)[name = tensor<string, []>("op_36887_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_36889_to_fp16 = const()[name = tensor<string, []>("op_36889_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3075014720)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_329_cast = mul(x = var_36887_cast, y = var_36889_to_fp16)[name = tensor<string, []>("hidden_states_329_cast")];
            tensor<int32, [2]> var_36896 = const()[name = tensor<string, []>("op_36896"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_36898 = const()[name = tensor<string, []>("op_36898"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_165_pad_type_0 = const()[name = tensor<string, []>("q_165_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_165_pad_0 = const()[name = tensor<string, []>("q_165_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3075017344)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_165_cast = conv(dilations = var_36898, groups = var_30385, pad = q_165_pad_0, pad_type = q_165_pad_type_0, strides = var_36896, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_q_weight_to_fp16, x = hidden_states_329_cast)[name = tensor<string, []>("q_165_cast")];
            tensor<int32, [2]> var_36902 = const()[name = tensor<string, []>("op_36902"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_36904 = const()[name = tensor<string, []>("op_36904"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_329_pad_type_0 = const()[name = tensor<string, []>("k_329_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_329_pad_0 = const()[name = tensor<string, []>("k_329_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3078294208)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_329_cast = conv(dilations = var_36904, groups = var_30385, pad = k_329_pad_0, pad_type = k_329_pad_type_0, strides = var_36902, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_k_weight_to_fp16, x = hidden_states_329_cast)[name = tensor<string, []>("k_329_cast")];
            tensor<int32, [2]> var_36908 = const()[name = tensor<string, []>("op_36908"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_36910 = const()[name = tensor<string, []>("op_36910"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_165_pad_type_0 = const()[name = tensor<string, []>("v_165_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_165_pad_0 = const()[name = tensor<string, []>("v_165_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3081571072)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_165_cast = conv(dilations = var_36910, groups = var_30385, pad = v_165_pad_0, pad_type = v_165_pad_type_0, strides = var_36908, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_v_weight_to_fp16, x = hidden_states_329_cast)[name = tensor<string, []>("v_165_cast")];
            tensor<int32, [4]> var_36914_begin_0 = const()[name = tensor<string, []>("op_36914_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_36914_end_0 = const()[name = tensor<string, []>("op_36914_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_36914_end_mask_0 = const()[name = tensor<string, []>("op_36914_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36914_cast = slice_by_index(begin = var_36914_begin_0, end = var_36914_end_0, end_mask = var_36914_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36914_cast")];
            tensor<int32, [4]> var_36918_begin_0 = const()[name = tensor<string, []>("op_36918_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_36918_end_0 = const()[name = tensor<string, []>("op_36918_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_36918_end_mask_0 = const()[name = tensor<string, []>("op_36918_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36918_cast = slice_by_index(begin = var_36918_begin_0, end = var_36918_end_0, end_mask = var_36918_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36918_cast")];
            tensor<int32, [4]> var_36922_begin_0 = const()[name = tensor<string, []>("op_36922_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_36922_end_0 = const()[name = tensor<string, []>("op_36922_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_36922_end_mask_0 = const()[name = tensor<string, []>("op_36922_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36922_cast = slice_by_index(begin = var_36922_begin_0, end = var_36922_end_0, end_mask = var_36922_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36922_cast")];
            tensor<int32, [4]> var_36926_begin_0 = const()[name = tensor<string, []>("op_36926_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_36926_end_0 = const()[name = tensor<string, []>("op_36926_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_36926_end_mask_0 = const()[name = tensor<string, []>("op_36926_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36926_cast = slice_by_index(begin = var_36926_begin_0, end = var_36926_end_0, end_mask = var_36926_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36926_cast")];
            tensor<int32, [4]> var_36930_begin_0 = const()[name = tensor<string, []>("op_36930_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_36930_end_0 = const()[name = tensor<string, []>("op_36930_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_36930_end_mask_0 = const()[name = tensor<string, []>("op_36930_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36930_cast = slice_by_index(begin = var_36930_begin_0, end = var_36930_end_0, end_mask = var_36930_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36930_cast")];
            tensor<int32, [4]> var_36934_begin_0 = const()[name = tensor<string, []>("op_36934_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_36934_end_0 = const()[name = tensor<string, []>("op_36934_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_36934_end_mask_0 = const()[name = tensor<string, []>("op_36934_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36934_cast = slice_by_index(begin = var_36934_begin_0, end = var_36934_end_0, end_mask = var_36934_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36934_cast")];
            tensor<int32, [4]> var_36938_begin_0 = const()[name = tensor<string, []>("op_36938_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_36938_end_0 = const()[name = tensor<string, []>("op_36938_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_36938_end_mask_0 = const()[name = tensor<string, []>("op_36938_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36938_cast = slice_by_index(begin = var_36938_begin_0, end = var_36938_end_0, end_mask = var_36938_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36938_cast")];
            tensor<int32, [4]> var_36942_begin_0 = const()[name = tensor<string, []>("op_36942_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_36942_end_0 = const()[name = tensor<string, []>("op_36942_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_36942_end_mask_0 = const()[name = tensor<string, []>("op_36942_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36942_cast = slice_by_index(begin = var_36942_begin_0, end = var_36942_end_0, end_mask = var_36942_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36942_cast")];
            tensor<int32, [4]> var_36946_begin_0 = const()[name = tensor<string, []>("op_36946_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_36946_end_0 = const()[name = tensor<string, []>("op_36946_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_36946_end_mask_0 = const()[name = tensor<string, []>("op_36946_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36946_cast = slice_by_index(begin = var_36946_begin_0, end = var_36946_end_0, end_mask = var_36946_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36946_cast")];
            tensor<int32, [4]> var_36950_begin_0 = const()[name = tensor<string, []>("op_36950_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_36950_end_0 = const()[name = tensor<string, []>("op_36950_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_36950_end_mask_0 = const()[name = tensor<string, []>("op_36950_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36950_cast = slice_by_index(begin = var_36950_begin_0, end = var_36950_end_0, end_mask = var_36950_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36950_cast")];
            tensor<int32, [4]> var_36954_begin_0 = const()[name = tensor<string, []>("op_36954_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_36954_end_0 = const()[name = tensor<string, []>("op_36954_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_36954_end_mask_0 = const()[name = tensor<string, []>("op_36954_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36954_cast = slice_by_index(begin = var_36954_begin_0, end = var_36954_end_0, end_mask = var_36954_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36954_cast")];
            tensor<int32, [4]> var_36958_begin_0 = const()[name = tensor<string, []>("op_36958_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_36958_end_0 = const()[name = tensor<string, []>("op_36958_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_36958_end_mask_0 = const()[name = tensor<string, []>("op_36958_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36958_cast = slice_by_index(begin = var_36958_begin_0, end = var_36958_end_0, end_mask = var_36958_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36958_cast")];
            tensor<int32, [4]> var_36962_begin_0 = const()[name = tensor<string, []>("op_36962_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_36962_end_0 = const()[name = tensor<string, []>("op_36962_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_36962_end_mask_0 = const()[name = tensor<string, []>("op_36962_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36962_cast = slice_by_index(begin = var_36962_begin_0, end = var_36962_end_0, end_mask = var_36962_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36962_cast")];
            tensor<int32, [4]> var_36966_begin_0 = const()[name = tensor<string, []>("op_36966_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_36966_end_0 = const()[name = tensor<string, []>("op_36966_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_36966_end_mask_0 = const()[name = tensor<string, []>("op_36966_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36966_cast = slice_by_index(begin = var_36966_begin_0, end = var_36966_end_0, end_mask = var_36966_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36966_cast")];
            tensor<int32, [4]> var_36970_begin_0 = const()[name = tensor<string, []>("op_36970_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_36970_end_0 = const()[name = tensor<string, []>("op_36970_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_36970_end_mask_0 = const()[name = tensor<string, []>("op_36970_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36970_cast = slice_by_index(begin = var_36970_begin_0, end = var_36970_end_0, end_mask = var_36970_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36970_cast")];
            tensor<int32, [4]> var_36974_begin_0 = const()[name = tensor<string, []>("op_36974_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_36974_end_0 = const()[name = tensor<string, []>("op_36974_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_36974_end_mask_0 = const()[name = tensor<string, []>("op_36974_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36974_cast = slice_by_index(begin = var_36974_begin_0, end = var_36974_end_0, end_mask = var_36974_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36974_cast")];
            tensor<int32, [4]> var_36978_begin_0 = const()[name = tensor<string, []>("op_36978_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_36978_end_0 = const()[name = tensor<string, []>("op_36978_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_36978_end_mask_0 = const()[name = tensor<string, []>("op_36978_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36978_cast = slice_by_index(begin = var_36978_begin_0, end = var_36978_end_0, end_mask = var_36978_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36978_cast")];
            tensor<int32, [4]> var_36982_begin_0 = const()[name = tensor<string, []>("op_36982_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_36982_end_0 = const()[name = tensor<string, []>("op_36982_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_36982_end_mask_0 = const()[name = tensor<string, []>("op_36982_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36982_cast = slice_by_index(begin = var_36982_begin_0, end = var_36982_end_0, end_mask = var_36982_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36982_cast")];
            tensor<int32, [4]> var_36986_begin_0 = const()[name = tensor<string, []>("op_36986_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_36986_end_0 = const()[name = tensor<string, []>("op_36986_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_36986_end_mask_0 = const()[name = tensor<string, []>("op_36986_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36986_cast = slice_by_index(begin = var_36986_begin_0, end = var_36986_end_0, end_mask = var_36986_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36986_cast")];
            tensor<int32, [4]> var_36990_begin_0 = const()[name = tensor<string, []>("op_36990_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_36990_end_0 = const()[name = tensor<string, []>("op_36990_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_36990_end_mask_0 = const()[name = tensor<string, []>("op_36990_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_36990_cast = slice_by_index(begin = var_36990_begin_0, end = var_36990_end_0, end_mask = var_36990_end_mask_0, x = q_165_cast)[name = tensor<string, []>("op_36990_cast")];
            tensor<int32, [4]> k_331_perm_0 = const()[name = tensor<string, []>("k_331_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_36997_begin_0 = const()[name = tensor<string, []>("op_36997_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_36997_end_0 = const()[name = tensor<string, []>("op_36997_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_36997_end_mask_0 = const()[name = tensor<string, []>("op_36997_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_57 = transpose(perm = k_331_perm_0, x = k_329_cast)[name = tensor<string, []>("transpose_57")];
            tensor<fp16, [2, 1024, 1, 64]> var_36997_cast = slice_by_index(begin = var_36997_begin_0, end = var_36997_end_0, end_mask = var_36997_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_36997_cast")];
            tensor<int32, [4]> var_37001_begin_0 = const()[name = tensor<string, []>("op_37001_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_37001_end_0 = const()[name = tensor<string, []>("op_37001_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_37001_end_mask_0 = const()[name = tensor<string, []>("op_37001_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37001_cast = slice_by_index(begin = var_37001_begin_0, end = var_37001_end_0, end_mask = var_37001_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_37001_cast")];
            tensor<int32, [4]> var_37005_begin_0 = const()[name = tensor<string, []>("op_37005_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_37005_end_0 = const()[name = tensor<string, []>("op_37005_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_37005_end_mask_0 = const()[name = tensor<string, []>("op_37005_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37005_cast = slice_by_index(begin = var_37005_begin_0, end = var_37005_end_0, end_mask = var_37005_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_37005_cast")];
            tensor<int32, [4]> var_37009_begin_0 = const()[name = tensor<string, []>("op_37009_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_37009_end_0 = const()[name = tensor<string, []>("op_37009_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_37009_end_mask_0 = const()[name = tensor<string, []>("op_37009_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37009_cast = slice_by_index(begin = var_37009_begin_0, end = var_37009_end_0, end_mask = var_37009_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_37009_cast")];
            tensor<int32, [4]> var_37013_begin_0 = const()[name = tensor<string, []>("op_37013_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_37013_end_0 = const()[name = tensor<string, []>("op_37013_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_37013_end_mask_0 = const()[name = tensor<string, []>("op_37013_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37013_cast = slice_by_index(begin = var_37013_begin_0, end = var_37013_end_0, end_mask = var_37013_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_37013_cast")];
            tensor<int32, [4]> var_37017_begin_0 = const()[name = tensor<string, []>("op_37017_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_37017_end_0 = const()[name = tensor<string, []>("op_37017_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_37017_end_mask_0 = const()[name = tensor<string, []>("op_37017_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37017_cast = slice_by_index(begin = var_37017_begin_0, end = var_37017_end_0, end_mask = var_37017_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_37017_cast")];
            tensor<int32, [4]> var_37021_begin_0 = const()[name = tensor<string, []>("op_37021_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_37021_end_0 = const()[name = tensor<string, []>("op_37021_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_37021_end_mask_0 = const()[name = tensor<string, []>("op_37021_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37021_cast = slice_by_index(begin = var_37021_begin_0, end = var_37021_end_0, end_mask = var_37021_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_37021_cast")];
            tensor<int32, [4]> var_37025_begin_0 = const()[name = tensor<string, []>("op_37025_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_37025_end_0 = const()[name = tensor<string, []>("op_37025_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_37025_end_mask_0 = const()[name = tensor<string, []>("op_37025_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37025_cast = slice_by_index(begin = var_37025_begin_0, end = var_37025_end_0, end_mask = var_37025_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_37025_cast")];
            tensor<int32, [4]> var_37029_begin_0 = const()[name = tensor<string, []>("op_37029_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_37029_end_0 = const()[name = tensor<string, []>("op_37029_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_37029_end_mask_0 = const()[name = tensor<string, []>("op_37029_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37029_cast = slice_by_index(begin = var_37029_begin_0, end = var_37029_end_0, end_mask = var_37029_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_37029_cast")];
            tensor<int32, [4]> var_37033_begin_0 = const()[name = tensor<string, []>("op_37033_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_37033_end_0 = const()[name = tensor<string, []>("op_37033_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_37033_end_mask_0 = const()[name = tensor<string, []>("op_37033_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37033_cast = slice_by_index(begin = var_37033_begin_0, end = var_37033_end_0, end_mask = var_37033_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_37033_cast")];
            tensor<int32, [4]> var_37037_begin_0 = const()[name = tensor<string, []>("op_37037_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_37037_end_0 = const()[name = tensor<string, []>("op_37037_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_37037_end_mask_0 = const()[name = tensor<string, []>("op_37037_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37037_cast = slice_by_index(begin = var_37037_begin_0, end = var_37037_end_0, end_mask = var_37037_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_37037_cast")];
            tensor<int32, [4]> var_37041_begin_0 = const()[name = tensor<string, []>("op_37041_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_37041_end_0 = const()[name = tensor<string, []>("op_37041_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_37041_end_mask_0 = const()[name = tensor<string, []>("op_37041_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37041_cast = slice_by_index(begin = var_37041_begin_0, end = var_37041_end_0, end_mask = var_37041_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_37041_cast")];
            tensor<int32, [4]> var_37045_begin_0 = const()[name = tensor<string, []>("op_37045_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_37045_end_0 = const()[name = tensor<string, []>("op_37045_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_37045_end_mask_0 = const()[name = tensor<string, []>("op_37045_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37045_cast = slice_by_index(begin = var_37045_begin_0, end = var_37045_end_0, end_mask = var_37045_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_37045_cast")];
            tensor<int32, [4]> var_37049_begin_0 = const()[name = tensor<string, []>("op_37049_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_37049_end_0 = const()[name = tensor<string, []>("op_37049_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_37049_end_mask_0 = const()[name = tensor<string, []>("op_37049_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37049_cast = slice_by_index(begin = var_37049_begin_0, end = var_37049_end_0, end_mask = var_37049_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_37049_cast")];
            tensor<int32, [4]> var_37053_begin_0 = const()[name = tensor<string, []>("op_37053_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_37053_end_0 = const()[name = tensor<string, []>("op_37053_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_37053_end_mask_0 = const()[name = tensor<string, []>("op_37053_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37053_cast = slice_by_index(begin = var_37053_begin_0, end = var_37053_end_0, end_mask = var_37053_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_37053_cast")];
            tensor<int32, [4]> var_37057_begin_0 = const()[name = tensor<string, []>("op_37057_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_37057_end_0 = const()[name = tensor<string, []>("op_37057_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_37057_end_mask_0 = const()[name = tensor<string, []>("op_37057_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37057_cast = slice_by_index(begin = var_37057_begin_0, end = var_37057_end_0, end_mask = var_37057_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_37057_cast")];
            tensor<int32, [4]> var_37061_begin_0 = const()[name = tensor<string, []>("op_37061_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_37061_end_0 = const()[name = tensor<string, []>("op_37061_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_37061_end_mask_0 = const()[name = tensor<string, []>("op_37061_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37061_cast = slice_by_index(begin = var_37061_begin_0, end = var_37061_end_0, end_mask = var_37061_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_37061_cast")];
            tensor<int32, [4]> var_37065_begin_0 = const()[name = tensor<string, []>("op_37065_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_37065_end_0 = const()[name = tensor<string, []>("op_37065_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_37065_end_mask_0 = const()[name = tensor<string, []>("op_37065_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37065_cast = slice_by_index(begin = var_37065_begin_0, end = var_37065_end_0, end_mask = var_37065_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_37065_cast")];
            tensor<int32, [4]> var_37069_begin_0 = const()[name = tensor<string, []>("op_37069_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_37069_end_0 = const()[name = tensor<string, []>("op_37069_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_37069_end_mask_0 = const()[name = tensor<string, []>("op_37069_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37069_cast = slice_by_index(begin = var_37069_begin_0, end = var_37069_end_0, end_mask = var_37069_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_37069_cast")];
            tensor<int32, [4]> var_37073_begin_0 = const()[name = tensor<string, []>("op_37073_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_37073_end_0 = const()[name = tensor<string, []>("op_37073_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_37073_end_mask_0 = const()[name = tensor<string, []>("op_37073_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37073_cast = slice_by_index(begin = var_37073_begin_0, end = var_37073_end_0, end_mask = var_37073_end_mask_0, x = transpose_57)[name = tensor<string, []>("op_37073_cast")];
            tensor<int32, [4]> var_37075_begin_0 = const()[name = tensor<string, []>("op_37075_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_37075_end_0 = const()[name = tensor<string, []>("op_37075_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_37075_end_mask_0 = const()[name = tensor<string, []>("op_37075_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37075_cast = slice_by_index(begin = var_37075_begin_0, end = var_37075_end_0, end_mask = var_37075_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37075_cast")];
            tensor<int32, [4]> var_37079_begin_0 = const()[name = tensor<string, []>("op_37079_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_37079_end_0 = const()[name = tensor<string, []>("op_37079_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_37079_end_mask_0 = const()[name = tensor<string, []>("op_37079_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37079_cast = slice_by_index(begin = var_37079_begin_0, end = var_37079_end_0, end_mask = var_37079_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37079_cast")];
            tensor<int32, [4]> var_37083_begin_0 = const()[name = tensor<string, []>("op_37083_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_37083_end_0 = const()[name = tensor<string, []>("op_37083_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_37083_end_mask_0 = const()[name = tensor<string, []>("op_37083_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37083_cast = slice_by_index(begin = var_37083_begin_0, end = var_37083_end_0, end_mask = var_37083_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37083_cast")];
            tensor<int32, [4]> var_37087_begin_0 = const()[name = tensor<string, []>("op_37087_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_37087_end_0 = const()[name = tensor<string, []>("op_37087_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_37087_end_mask_0 = const()[name = tensor<string, []>("op_37087_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37087_cast = slice_by_index(begin = var_37087_begin_0, end = var_37087_end_0, end_mask = var_37087_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37087_cast")];
            tensor<int32, [4]> var_37091_begin_0 = const()[name = tensor<string, []>("op_37091_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_37091_end_0 = const()[name = tensor<string, []>("op_37091_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_37091_end_mask_0 = const()[name = tensor<string, []>("op_37091_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37091_cast = slice_by_index(begin = var_37091_begin_0, end = var_37091_end_0, end_mask = var_37091_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37091_cast")];
            tensor<int32, [4]> var_37095_begin_0 = const()[name = tensor<string, []>("op_37095_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_37095_end_0 = const()[name = tensor<string, []>("op_37095_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_37095_end_mask_0 = const()[name = tensor<string, []>("op_37095_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37095_cast = slice_by_index(begin = var_37095_begin_0, end = var_37095_end_0, end_mask = var_37095_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37095_cast")];
            tensor<int32, [4]> var_37099_begin_0 = const()[name = tensor<string, []>("op_37099_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_37099_end_0 = const()[name = tensor<string, []>("op_37099_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_37099_end_mask_0 = const()[name = tensor<string, []>("op_37099_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37099_cast = slice_by_index(begin = var_37099_begin_0, end = var_37099_end_0, end_mask = var_37099_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37099_cast")];
            tensor<int32, [4]> var_37103_begin_0 = const()[name = tensor<string, []>("op_37103_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_37103_end_0 = const()[name = tensor<string, []>("op_37103_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_37103_end_mask_0 = const()[name = tensor<string, []>("op_37103_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37103_cast = slice_by_index(begin = var_37103_begin_0, end = var_37103_end_0, end_mask = var_37103_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37103_cast")];
            tensor<int32, [4]> var_37107_begin_0 = const()[name = tensor<string, []>("op_37107_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_37107_end_0 = const()[name = tensor<string, []>("op_37107_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_37107_end_mask_0 = const()[name = tensor<string, []>("op_37107_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37107_cast = slice_by_index(begin = var_37107_begin_0, end = var_37107_end_0, end_mask = var_37107_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37107_cast")];
            tensor<int32, [4]> var_37111_begin_0 = const()[name = tensor<string, []>("op_37111_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_37111_end_0 = const()[name = tensor<string, []>("op_37111_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_37111_end_mask_0 = const()[name = tensor<string, []>("op_37111_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37111_cast = slice_by_index(begin = var_37111_begin_0, end = var_37111_end_0, end_mask = var_37111_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37111_cast")];
            tensor<int32, [4]> var_37115_begin_0 = const()[name = tensor<string, []>("op_37115_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_37115_end_0 = const()[name = tensor<string, []>("op_37115_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_37115_end_mask_0 = const()[name = tensor<string, []>("op_37115_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37115_cast = slice_by_index(begin = var_37115_begin_0, end = var_37115_end_0, end_mask = var_37115_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37115_cast")];
            tensor<int32, [4]> var_37119_begin_0 = const()[name = tensor<string, []>("op_37119_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_37119_end_0 = const()[name = tensor<string, []>("op_37119_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_37119_end_mask_0 = const()[name = tensor<string, []>("op_37119_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37119_cast = slice_by_index(begin = var_37119_begin_0, end = var_37119_end_0, end_mask = var_37119_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37119_cast")];
            tensor<int32, [4]> var_37123_begin_0 = const()[name = tensor<string, []>("op_37123_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_37123_end_0 = const()[name = tensor<string, []>("op_37123_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_37123_end_mask_0 = const()[name = tensor<string, []>("op_37123_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37123_cast = slice_by_index(begin = var_37123_begin_0, end = var_37123_end_0, end_mask = var_37123_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37123_cast")];
            tensor<int32, [4]> var_37127_begin_0 = const()[name = tensor<string, []>("op_37127_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_37127_end_0 = const()[name = tensor<string, []>("op_37127_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_37127_end_mask_0 = const()[name = tensor<string, []>("op_37127_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37127_cast = slice_by_index(begin = var_37127_begin_0, end = var_37127_end_0, end_mask = var_37127_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37127_cast")];
            tensor<int32, [4]> var_37131_begin_0 = const()[name = tensor<string, []>("op_37131_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_37131_end_0 = const()[name = tensor<string, []>("op_37131_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_37131_end_mask_0 = const()[name = tensor<string, []>("op_37131_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37131_cast = slice_by_index(begin = var_37131_begin_0, end = var_37131_end_0, end_mask = var_37131_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37131_cast")];
            tensor<int32, [4]> var_37135_begin_0 = const()[name = tensor<string, []>("op_37135_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_37135_end_0 = const()[name = tensor<string, []>("op_37135_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_37135_end_mask_0 = const()[name = tensor<string, []>("op_37135_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37135_cast = slice_by_index(begin = var_37135_begin_0, end = var_37135_end_0, end_mask = var_37135_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37135_cast")];
            tensor<int32, [4]> var_37139_begin_0 = const()[name = tensor<string, []>("op_37139_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_37139_end_0 = const()[name = tensor<string, []>("op_37139_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_37139_end_mask_0 = const()[name = tensor<string, []>("op_37139_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37139_cast = slice_by_index(begin = var_37139_begin_0, end = var_37139_end_0, end_mask = var_37139_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37139_cast")];
            tensor<int32, [4]> var_37143_begin_0 = const()[name = tensor<string, []>("op_37143_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_37143_end_0 = const()[name = tensor<string, []>("op_37143_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_37143_end_mask_0 = const()[name = tensor<string, []>("op_37143_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37143_cast = slice_by_index(begin = var_37143_begin_0, end = var_37143_end_0, end_mask = var_37143_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37143_cast")];
            tensor<int32, [4]> var_37147_begin_0 = const()[name = tensor<string, []>("op_37147_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_37147_end_0 = const()[name = tensor<string, []>("op_37147_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_37147_end_mask_0 = const()[name = tensor<string, []>("op_37147_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37147_cast = slice_by_index(begin = var_37147_begin_0, end = var_37147_end_0, end_mask = var_37147_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37147_cast")];
            tensor<int32, [4]> var_37151_begin_0 = const()[name = tensor<string, []>("op_37151_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_37151_end_0 = const()[name = tensor<string, []>("op_37151_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_37151_end_mask_0 = const()[name = tensor<string, []>("op_37151_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37151_cast = slice_by_index(begin = var_37151_begin_0, end = var_37151_end_0, end_mask = var_37151_end_mask_0, x = v_165_cast)[name = tensor<string, []>("op_37151_cast")];
            tensor<string, []> var_37155_equation_0 = const()[name = tensor<string, []>("op_37155_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37155_cast = einsum(equation = var_37155_equation_0, values = (var_36997_cast, var_36914_cast))[name = tensor<string, []>("op_37155_cast")];
            tensor<fp16, []> var_37156_to_fp16 = const()[name = tensor<string, []>("op_37156_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3121_cast = mul(x = var_37155_cast, y = var_37156_to_fp16)[name = tensor<string, []>("aw_3121_cast")];
            tensor<string, []> var_37159_equation_0 = const()[name = tensor<string, []>("op_37159_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37159_cast = einsum(equation = var_37159_equation_0, values = (var_37001_cast, var_36918_cast))[name = tensor<string, []>("op_37159_cast")];
            tensor<fp16, []> var_37160_to_fp16 = const()[name = tensor<string, []>("op_37160_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3123_cast = mul(x = var_37159_cast, y = var_37160_to_fp16)[name = tensor<string, []>("aw_3123_cast")];
            tensor<string, []> var_37163_equation_0 = const()[name = tensor<string, []>("op_37163_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37163_cast = einsum(equation = var_37163_equation_0, values = (var_37005_cast, var_36922_cast))[name = tensor<string, []>("op_37163_cast")];
            tensor<fp16, []> var_37164_to_fp16 = const()[name = tensor<string, []>("op_37164_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3125_cast = mul(x = var_37163_cast, y = var_37164_to_fp16)[name = tensor<string, []>("aw_3125_cast")];
            tensor<string, []> var_37167_equation_0 = const()[name = tensor<string, []>("op_37167_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37167_cast = einsum(equation = var_37167_equation_0, values = (var_37009_cast, var_36926_cast))[name = tensor<string, []>("op_37167_cast")];
            tensor<fp16, []> var_37168_to_fp16 = const()[name = tensor<string, []>("op_37168_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3127_cast = mul(x = var_37167_cast, y = var_37168_to_fp16)[name = tensor<string, []>("aw_3127_cast")];
            tensor<string, []> var_37171_equation_0 = const()[name = tensor<string, []>("op_37171_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37171_cast = einsum(equation = var_37171_equation_0, values = (var_37013_cast, var_36930_cast))[name = tensor<string, []>("op_37171_cast")];
            tensor<fp16, []> var_37172_to_fp16 = const()[name = tensor<string, []>("op_37172_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3129_cast = mul(x = var_37171_cast, y = var_37172_to_fp16)[name = tensor<string, []>("aw_3129_cast")];
            tensor<string, []> var_37175_equation_0 = const()[name = tensor<string, []>("op_37175_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37175_cast = einsum(equation = var_37175_equation_0, values = (var_37017_cast, var_36934_cast))[name = tensor<string, []>("op_37175_cast")];
            tensor<fp16, []> var_37176_to_fp16 = const()[name = tensor<string, []>("op_37176_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3131_cast = mul(x = var_37175_cast, y = var_37176_to_fp16)[name = tensor<string, []>("aw_3131_cast")];
            tensor<string, []> var_37179_equation_0 = const()[name = tensor<string, []>("op_37179_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37179_cast = einsum(equation = var_37179_equation_0, values = (var_37021_cast, var_36938_cast))[name = tensor<string, []>("op_37179_cast")];
            tensor<fp16, []> var_37180_to_fp16 = const()[name = tensor<string, []>("op_37180_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3133_cast = mul(x = var_37179_cast, y = var_37180_to_fp16)[name = tensor<string, []>("aw_3133_cast")];
            tensor<string, []> var_37183_equation_0 = const()[name = tensor<string, []>("op_37183_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37183_cast = einsum(equation = var_37183_equation_0, values = (var_37025_cast, var_36942_cast))[name = tensor<string, []>("op_37183_cast")];
            tensor<fp16, []> var_37184_to_fp16 = const()[name = tensor<string, []>("op_37184_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3135_cast = mul(x = var_37183_cast, y = var_37184_to_fp16)[name = tensor<string, []>("aw_3135_cast")];
            tensor<string, []> var_37187_equation_0 = const()[name = tensor<string, []>("op_37187_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37187_cast = einsum(equation = var_37187_equation_0, values = (var_37029_cast, var_36946_cast))[name = tensor<string, []>("op_37187_cast")];
            tensor<fp16, []> var_37188_to_fp16 = const()[name = tensor<string, []>("op_37188_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3137_cast = mul(x = var_37187_cast, y = var_37188_to_fp16)[name = tensor<string, []>("aw_3137_cast")];
            tensor<string, []> var_37191_equation_0 = const()[name = tensor<string, []>("op_37191_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37191_cast = einsum(equation = var_37191_equation_0, values = (var_37033_cast, var_36950_cast))[name = tensor<string, []>("op_37191_cast")];
            tensor<fp16, []> var_37192_to_fp16 = const()[name = tensor<string, []>("op_37192_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3139_cast = mul(x = var_37191_cast, y = var_37192_to_fp16)[name = tensor<string, []>("aw_3139_cast")];
            tensor<string, []> var_37195_equation_0 = const()[name = tensor<string, []>("op_37195_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37195_cast = einsum(equation = var_37195_equation_0, values = (var_37037_cast, var_36954_cast))[name = tensor<string, []>("op_37195_cast")];
            tensor<fp16, []> var_37196_to_fp16 = const()[name = tensor<string, []>("op_37196_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3141_cast = mul(x = var_37195_cast, y = var_37196_to_fp16)[name = tensor<string, []>("aw_3141_cast")];
            tensor<string, []> var_37199_equation_0 = const()[name = tensor<string, []>("op_37199_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37199_cast = einsum(equation = var_37199_equation_0, values = (var_37041_cast, var_36958_cast))[name = tensor<string, []>("op_37199_cast")];
            tensor<fp16, []> var_37200_to_fp16 = const()[name = tensor<string, []>("op_37200_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3143_cast = mul(x = var_37199_cast, y = var_37200_to_fp16)[name = tensor<string, []>("aw_3143_cast")];
            tensor<string, []> var_37203_equation_0 = const()[name = tensor<string, []>("op_37203_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37203_cast = einsum(equation = var_37203_equation_0, values = (var_37045_cast, var_36962_cast))[name = tensor<string, []>("op_37203_cast")];
            tensor<fp16, []> var_37204_to_fp16 = const()[name = tensor<string, []>("op_37204_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3145_cast = mul(x = var_37203_cast, y = var_37204_to_fp16)[name = tensor<string, []>("aw_3145_cast")];
            tensor<string, []> var_37207_equation_0 = const()[name = tensor<string, []>("op_37207_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37207_cast = einsum(equation = var_37207_equation_0, values = (var_37049_cast, var_36966_cast))[name = tensor<string, []>("op_37207_cast")];
            tensor<fp16, []> var_37208_to_fp16 = const()[name = tensor<string, []>("op_37208_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3147_cast = mul(x = var_37207_cast, y = var_37208_to_fp16)[name = tensor<string, []>("aw_3147_cast")];
            tensor<string, []> var_37211_equation_0 = const()[name = tensor<string, []>("op_37211_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37211_cast = einsum(equation = var_37211_equation_0, values = (var_37053_cast, var_36970_cast))[name = tensor<string, []>("op_37211_cast")];
            tensor<fp16, []> var_37212_to_fp16 = const()[name = tensor<string, []>("op_37212_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3149_cast = mul(x = var_37211_cast, y = var_37212_to_fp16)[name = tensor<string, []>("aw_3149_cast")];
            tensor<string, []> var_37215_equation_0 = const()[name = tensor<string, []>("op_37215_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37215_cast = einsum(equation = var_37215_equation_0, values = (var_37057_cast, var_36974_cast))[name = tensor<string, []>("op_37215_cast")];
            tensor<fp16, []> var_37216_to_fp16 = const()[name = tensor<string, []>("op_37216_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3151_cast = mul(x = var_37215_cast, y = var_37216_to_fp16)[name = tensor<string, []>("aw_3151_cast")];
            tensor<string, []> var_37219_equation_0 = const()[name = tensor<string, []>("op_37219_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37219_cast = einsum(equation = var_37219_equation_0, values = (var_37061_cast, var_36978_cast))[name = tensor<string, []>("op_37219_cast")];
            tensor<fp16, []> var_37220_to_fp16 = const()[name = tensor<string, []>("op_37220_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3153_cast = mul(x = var_37219_cast, y = var_37220_to_fp16)[name = tensor<string, []>("aw_3153_cast")];
            tensor<string, []> var_37223_equation_0 = const()[name = tensor<string, []>("op_37223_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37223_cast = einsum(equation = var_37223_equation_0, values = (var_37065_cast, var_36982_cast))[name = tensor<string, []>("op_37223_cast")];
            tensor<fp16, []> var_37224_to_fp16 = const()[name = tensor<string, []>("op_37224_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3155_cast = mul(x = var_37223_cast, y = var_37224_to_fp16)[name = tensor<string, []>("aw_3155_cast")];
            tensor<string, []> var_37227_equation_0 = const()[name = tensor<string, []>("op_37227_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37227_cast = einsum(equation = var_37227_equation_0, values = (var_37069_cast, var_36986_cast))[name = tensor<string, []>("op_37227_cast")];
            tensor<fp16, []> var_37228_to_fp16 = const()[name = tensor<string, []>("op_37228_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3157_cast = mul(x = var_37227_cast, y = var_37228_to_fp16)[name = tensor<string, []>("aw_3157_cast")];
            tensor<string, []> var_37231_equation_0 = const()[name = tensor<string, []>("op_37231_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37231_cast = einsum(equation = var_37231_equation_0, values = (var_37073_cast, var_36990_cast))[name = tensor<string, []>("op_37231_cast")];
            tensor<fp16, []> var_37232_to_fp16 = const()[name = tensor<string, []>("op_37232_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3159_cast = mul(x = var_37231_cast, y = var_37232_to_fp16)[name = tensor<string, []>("aw_3159_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37234_cast = softmax(axis = var_30385, x = aw_3121_cast)[name = tensor<string, []>("op_37234_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37235_cast = softmax(axis = var_30385, x = aw_3123_cast)[name = tensor<string, []>("op_37235_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37236_cast = softmax(axis = var_30385, x = aw_3125_cast)[name = tensor<string, []>("op_37236_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37237_cast = softmax(axis = var_30385, x = aw_3127_cast)[name = tensor<string, []>("op_37237_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37238_cast = softmax(axis = var_30385, x = aw_3129_cast)[name = tensor<string, []>("op_37238_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37239_cast = softmax(axis = var_30385, x = aw_3131_cast)[name = tensor<string, []>("op_37239_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37240_cast = softmax(axis = var_30385, x = aw_3133_cast)[name = tensor<string, []>("op_37240_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37241_cast = softmax(axis = var_30385, x = aw_3135_cast)[name = tensor<string, []>("op_37241_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37242_cast = softmax(axis = var_30385, x = aw_3137_cast)[name = tensor<string, []>("op_37242_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37243_cast = softmax(axis = var_30385, x = aw_3139_cast)[name = tensor<string, []>("op_37243_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37244_cast = softmax(axis = var_30385, x = aw_3141_cast)[name = tensor<string, []>("op_37244_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37245_cast = softmax(axis = var_30385, x = aw_3143_cast)[name = tensor<string, []>("op_37245_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37246_cast = softmax(axis = var_30385, x = aw_3145_cast)[name = tensor<string, []>("op_37246_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37247_cast = softmax(axis = var_30385, x = aw_3147_cast)[name = tensor<string, []>("op_37247_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37248_cast = softmax(axis = var_30385, x = aw_3149_cast)[name = tensor<string, []>("op_37248_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37249_cast = softmax(axis = var_30385, x = aw_3151_cast)[name = tensor<string, []>("op_37249_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37250_cast = softmax(axis = var_30385, x = aw_3153_cast)[name = tensor<string, []>("op_37250_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37251_cast = softmax(axis = var_30385, x = aw_3155_cast)[name = tensor<string, []>("op_37251_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37252_cast = softmax(axis = var_30385, x = aw_3157_cast)[name = tensor<string, []>("op_37252_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_37253_cast = softmax(axis = var_30385, x = aw_3159_cast)[name = tensor<string, []>("op_37253_cast")];
            tensor<string, []> var_37255_equation_0 = const()[name = tensor<string, []>("op_37255_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37255_cast = einsum(equation = var_37255_equation_0, values = (var_37075_cast, var_37234_cast))[name = tensor<string, []>("op_37255_cast")];
            tensor<string, []> var_37257_equation_0 = const()[name = tensor<string, []>("op_37257_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37257_cast = einsum(equation = var_37257_equation_0, values = (var_37079_cast, var_37235_cast))[name = tensor<string, []>("op_37257_cast")];
            tensor<string, []> var_37259_equation_0 = const()[name = tensor<string, []>("op_37259_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37259_cast = einsum(equation = var_37259_equation_0, values = (var_37083_cast, var_37236_cast))[name = tensor<string, []>("op_37259_cast")];
            tensor<string, []> var_37261_equation_0 = const()[name = tensor<string, []>("op_37261_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37261_cast = einsum(equation = var_37261_equation_0, values = (var_37087_cast, var_37237_cast))[name = tensor<string, []>("op_37261_cast")];
            tensor<string, []> var_37263_equation_0 = const()[name = tensor<string, []>("op_37263_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37263_cast = einsum(equation = var_37263_equation_0, values = (var_37091_cast, var_37238_cast))[name = tensor<string, []>("op_37263_cast")];
            tensor<string, []> var_37265_equation_0 = const()[name = tensor<string, []>("op_37265_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37265_cast = einsum(equation = var_37265_equation_0, values = (var_37095_cast, var_37239_cast))[name = tensor<string, []>("op_37265_cast")];
            tensor<string, []> var_37267_equation_0 = const()[name = tensor<string, []>("op_37267_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37267_cast = einsum(equation = var_37267_equation_0, values = (var_37099_cast, var_37240_cast))[name = tensor<string, []>("op_37267_cast")];
            tensor<string, []> var_37269_equation_0 = const()[name = tensor<string, []>("op_37269_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37269_cast = einsum(equation = var_37269_equation_0, values = (var_37103_cast, var_37241_cast))[name = tensor<string, []>("op_37269_cast")];
            tensor<string, []> var_37271_equation_0 = const()[name = tensor<string, []>("op_37271_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37271_cast = einsum(equation = var_37271_equation_0, values = (var_37107_cast, var_37242_cast))[name = tensor<string, []>("op_37271_cast")];
            tensor<string, []> var_37273_equation_0 = const()[name = tensor<string, []>("op_37273_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37273_cast = einsum(equation = var_37273_equation_0, values = (var_37111_cast, var_37243_cast))[name = tensor<string, []>("op_37273_cast")];
            tensor<string, []> var_37275_equation_0 = const()[name = tensor<string, []>("op_37275_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37275_cast = einsum(equation = var_37275_equation_0, values = (var_37115_cast, var_37244_cast))[name = tensor<string, []>("op_37275_cast")];
            tensor<string, []> var_37277_equation_0 = const()[name = tensor<string, []>("op_37277_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37277_cast = einsum(equation = var_37277_equation_0, values = (var_37119_cast, var_37245_cast))[name = tensor<string, []>("op_37277_cast")];
            tensor<string, []> var_37279_equation_0 = const()[name = tensor<string, []>("op_37279_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37279_cast = einsum(equation = var_37279_equation_0, values = (var_37123_cast, var_37246_cast))[name = tensor<string, []>("op_37279_cast")];
            tensor<string, []> var_37281_equation_0 = const()[name = tensor<string, []>("op_37281_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37281_cast = einsum(equation = var_37281_equation_0, values = (var_37127_cast, var_37247_cast))[name = tensor<string, []>("op_37281_cast")];
            tensor<string, []> var_37283_equation_0 = const()[name = tensor<string, []>("op_37283_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37283_cast = einsum(equation = var_37283_equation_0, values = (var_37131_cast, var_37248_cast))[name = tensor<string, []>("op_37283_cast")];
            tensor<string, []> var_37285_equation_0 = const()[name = tensor<string, []>("op_37285_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37285_cast = einsum(equation = var_37285_equation_0, values = (var_37135_cast, var_37249_cast))[name = tensor<string, []>("op_37285_cast")];
            tensor<string, []> var_37287_equation_0 = const()[name = tensor<string, []>("op_37287_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37287_cast = einsum(equation = var_37287_equation_0, values = (var_37139_cast, var_37250_cast))[name = tensor<string, []>("op_37287_cast")];
            tensor<string, []> var_37289_equation_0 = const()[name = tensor<string, []>("op_37289_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37289_cast = einsum(equation = var_37289_equation_0, values = (var_37143_cast, var_37251_cast))[name = tensor<string, []>("op_37289_cast")];
            tensor<string, []> var_37291_equation_0 = const()[name = tensor<string, []>("op_37291_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37291_cast = einsum(equation = var_37291_equation_0, values = (var_37147_cast, var_37252_cast))[name = tensor<string, []>("op_37291_cast")];
            tensor<string, []> var_37293_equation_0 = const()[name = tensor<string, []>("op_37293_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37293_cast = einsum(equation = var_37293_equation_0, values = (var_37151_cast, var_37253_cast))[name = tensor<string, []>("op_37293_cast")];
            tensor<bool, []> input_497_interleave_0 = const()[name = tensor<string, []>("input_497_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_497_cast = concat(axis = var_30385, interleave = input_497_interleave_0, values = (var_37255_cast, var_37257_cast, var_37259_cast, var_37261_cast, var_37263_cast, var_37265_cast, var_37267_cast, var_37269_cast, var_37271_cast, var_37273_cast, var_37275_cast, var_37277_cast, var_37279_cast, var_37281_cast, var_37283_cast, var_37285_cast, var_37287_cast, var_37289_cast, var_37291_cast, var_37293_cast))[name = tensor<string, []>("input_497_cast")];
            tensor<int32, [2]> var_37299 = const()[name = tensor<string, []>("op_37299"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_37301 = const()[name = tensor<string, []>("op_37301"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_37303_pad_type_0 = const()[name = tensor<string, []>("op_37303_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_37303_pad_0 = const()[name = tensor<string, []>("op_37303_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3084847936)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3088124800)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_37303_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_out_0_bias_to_fp16, dilations = var_37301, groups = var_30385, pad = var_37303_pad_0, pad_type = var_37303_pad_type_0, strides = var_37299, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn1_to_out_0_weight_to_fp16, x = input_497_cast)[name = tensor<string, []>("op_37303_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_249_cast = add(x = var_37303_cast, y = inputs_247_cast)[name = tensor<string, []>("inputs_249_cast")];
            tensor<int32, [1]> var_37307 = const()[name = tensor<string, []>("op_37307"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_249_cast = reduce_mean(axes = var_37307, keep_dims = var_30380, x = inputs_249_cast)[name = tensor<string, []>("channels_mean_249_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_249_cast = sub(x = inputs_249_cast, y = channels_mean_249_cast)[name = tensor<string, []>("zero_mean_249_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_249_cast = mul(x = zero_mean_249_cast, y = zero_mean_249_cast)[name = tensor<string, []>("zero_mean_sq_249_cast")];
            tensor<int32, [1]> var_37311 = const()[name = tensor<string, []>("op_37311"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_37312_cast = reduce_mean(axes = var_37311, keep_dims = var_30380, x = zero_mean_sq_249_cast)[name = tensor<string, []>("op_37312_cast")];
            tensor<fp16, []> var_37313_to_fp16 = const()[name = tensor<string, []>("op_37313_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_37314_cast = add(x = var_37312_cast, y = var_37313_to_fp16)[name = tensor<string, []>("op_37314_cast")];
            tensor<fp16, []> denom_249_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_249_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_249_cast = rsqrt(epsilon = denom_249_epsilon_0_to_fp16, x = var_37314_cast)[name = tensor<string, []>("denom_249_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_249_cast = mul(x = zero_mean_249_cast, y = denom_249_cast)[name = tensor<string, []>("out_249_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_37318_to_fp16 = const()[name = tensor<string, []>("op_37318_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3088127424)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_37319_cast = add(x = out_249_cast, y = var_37318_to_fp16)[name = tensor<string, []>("op_37319_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_37321_to_fp16 = const()[name = tensor<string, []>("op_37321_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3088130048)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_331_cast = mul(x = var_37319_cast, y = var_37321_to_fp16)[name = tensor<string, []>("hidden_states_331_cast")];
            tensor<int32, [2]> var_37328 = const()[name = tensor<string, []>("op_37328"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_37330 = const()[name = tensor<string, []>("op_37330"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_167_pad_type_0 = const()[name = tensor<string, []>("q_167_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_167_pad_0 = const()[name = tensor<string, []>("q_167_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3088132672)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_167_cast = conv(dilations = var_37330, groups = var_30385, pad = q_167_pad_0, pad_type = q_167_pad_type_0, strides = var_37328, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_q_weight_to_fp16, x = hidden_states_331_cast)[name = tensor<string, []>("q_167_cast")];
            tensor<int32, [2]> var_37334 = const()[name = tensor<string, []>("op_37334"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_37336 = const()[name = tensor<string, []>("op_37336"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_333_pad_type_0 = const()[name = tensor<string, []>("k_333_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_333_pad_0 = const()[name = tensor<string, []>("k_333_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3091409536)))];
            tensor<fp16, [2, 1280, 1, 77]> k_333_cast = conv(dilations = var_37336, groups = var_30385, pad = k_333_pad_0, pad_type = k_333_pad_type_0, strides = var_37334, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_333_cast")];
            tensor<int32, [2]> var_37340 = const()[name = tensor<string, []>("op_37340"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_37342 = const()[name = tensor<string, []>("op_37342"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_167_pad_type_0 = const()[name = tensor<string, []>("v_167_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_167_pad_0 = const()[name = tensor<string, []>("v_167_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3096652480)))];
            tensor<fp16, [2, 1280, 1, 77]> v_167_cast = conv(dilations = var_37342, groups = var_30385, pad = v_167_pad_0, pad_type = v_167_pad_type_0, strides = var_37340, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_167_cast")];
            tensor<int32, [4]> var_37346_begin_0 = const()[name = tensor<string, []>("op_37346_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_37346_end_0 = const()[name = tensor<string, []>("op_37346_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_37346_end_mask_0 = const()[name = tensor<string, []>("op_37346_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37346_cast = slice_by_index(begin = var_37346_begin_0, end = var_37346_end_0, end_mask = var_37346_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37346_cast")];
            tensor<int32, [4]> var_37350_begin_0 = const()[name = tensor<string, []>("op_37350_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_37350_end_0 = const()[name = tensor<string, []>("op_37350_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_37350_end_mask_0 = const()[name = tensor<string, []>("op_37350_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37350_cast = slice_by_index(begin = var_37350_begin_0, end = var_37350_end_0, end_mask = var_37350_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37350_cast")];
            tensor<int32, [4]> var_37354_begin_0 = const()[name = tensor<string, []>("op_37354_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_37354_end_0 = const()[name = tensor<string, []>("op_37354_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_37354_end_mask_0 = const()[name = tensor<string, []>("op_37354_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37354_cast = slice_by_index(begin = var_37354_begin_0, end = var_37354_end_0, end_mask = var_37354_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37354_cast")];
            tensor<int32, [4]> var_37358_begin_0 = const()[name = tensor<string, []>("op_37358_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_37358_end_0 = const()[name = tensor<string, []>("op_37358_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_37358_end_mask_0 = const()[name = tensor<string, []>("op_37358_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37358_cast = slice_by_index(begin = var_37358_begin_0, end = var_37358_end_0, end_mask = var_37358_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37358_cast")];
            tensor<int32, [4]> var_37362_begin_0 = const()[name = tensor<string, []>("op_37362_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_37362_end_0 = const()[name = tensor<string, []>("op_37362_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_37362_end_mask_0 = const()[name = tensor<string, []>("op_37362_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37362_cast = slice_by_index(begin = var_37362_begin_0, end = var_37362_end_0, end_mask = var_37362_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37362_cast")];
            tensor<int32, [4]> var_37366_begin_0 = const()[name = tensor<string, []>("op_37366_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_37366_end_0 = const()[name = tensor<string, []>("op_37366_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_37366_end_mask_0 = const()[name = tensor<string, []>("op_37366_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37366_cast = slice_by_index(begin = var_37366_begin_0, end = var_37366_end_0, end_mask = var_37366_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37366_cast")];
            tensor<int32, [4]> var_37370_begin_0 = const()[name = tensor<string, []>("op_37370_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_37370_end_0 = const()[name = tensor<string, []>("op_37370_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_37370_end_mask_0 = const()[name = tensor<string, []>("op_37370_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37370_cast = slice_by_index(begin = var_37370_begin_0, end = var_37370_end_0, end_mask = var_37370_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37370_cast")];
            tensor<int32, [4]> var_37374_begin_0 = const()[name = tensor<string, []>("op_37374_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_37374_end_0 = const()[name = tensor<string, []>("op_37374_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_37374_end_mask_0 = const()[name = tensor<string, []>("op_37374_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37374_cast = slice_by_index(begin = var_37374_begin_0, end = var_37374_end_0, end_mask = var_37374_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37374_cast")];
            tensor<int32, [4]> var_37378_begin_0 = const()[name = tensor<string, []>("op_37378_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_37378_end_0 = const()[name = tensor<string, []>("op_37378_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_37378_end_mask_0 = const()[name = tensor<string, []>("op_37378_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37378_cast = slice_by_index(begin = var_37378_begin_0, end = var_37378_end_0, end_mask = var_37378_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37378_cast")];
            tensor<int32, [4]> var_37382_begin_0 = const()[name = tensor<string, []>("op_37382_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_37382_end_0 = const()[name = tensor<string, []>("op_37382_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_37382_end_mask_0 = const()[name = tensor<string, []>("op_37382_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37382_cast = slice_by_index(begin = var_37382_begin_0, end = var_37382_end_0, end_mask = var_37382_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37382_cast")];
            tensor<int32, [4]> var_37386_begin_0 = const()[name = tensor<string, []>("op_37386_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_37386_end_0 = const()[name = tensor<string, []>("op_37386_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_37386_end_mask_0 = const()[name = tensor<string, []>("op_37386_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37386_cast = slice_by_index(begin = var_37386_begin_0, end = var_37386_end_0, end_mask = var_37386_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37386_cast")];
            tensor<int32, [4]> var_37390_begin_0 = const()[name = tensor<string, []>("op_37390_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_37390_end_0 = const()[name = tensor<string, []>("op_37390_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_37390_end_mask_0 = const()[name = tensor<string, []>("op_37390_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37390_cast = slice_by_index(begin = var_37390_begin_0, end = var_37390_end_0, end_mask = var_37390_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37390_cast")];
            tensor<int32, [4]> var_37394_begin_0 = const()[name = tensor<string, []>("op_37394_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_37394_end_0 = const()[name = tensor<string, []>("op_37394_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_37394_end_mask_0 = const()[name = tensor<string, []>("op_37394_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37394_cast = slice_by_index(begin = var_37394_begin_0, end = var_37394_end_0, end_mask = var_37394_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37394_cast")];
            tensor<int32, [4]> var_37398_begin_0 = const()[name = tensor<string, []>("op_37398_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_37398_end_0 = const()[name = tensor<string, []>("op_37398_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_37398_end_mask_0 = const()[name = tensor<string, []>("op_37398_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37398_cast = slice_by_index(begin = var_37398_begin_0, end = var_37398_end_0, end_mask = var_37398_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37398_cast")];
            tensor<int32, [4]> var_37402_begin_0 = const()[name = tensor<string, []>("op_37402_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_37402_end_0 = const()[name = tensor<string, []>("op_37402_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_37402_end_mask_0 = const()[name = tensor<string, []>("op_37402_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37402_cast = slice_by_index(begin = var_37402_begin_0, end = var_37402_end_0, end_mask = var_37402_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37402_cast")];
            tensor<int32, [4]> var_37406_begin_0 = const()[name = tensor<string, []>("op_37406_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_37406_end_0 = const()[name = tensor<string, []>("op_37406_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_37406_end_mask_0 = const()[name = tensor<string, []>("op_37406_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37406_cast = slice_by_index(begin = var_37406_begin_0, end = var_37406_end_0, end_mask = var_37406_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37406_cast")];
            tensor<int32, [4]> var_37410_begin_0 = const()[name = tensor<string, []>("op_37410_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_37410_end_0 = const()[name = tensor<string, []>("op_37410_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_37410_end_mask_0 = const()[name = tensor<string, []>("op_37410_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37410_cast = slice_by_index(begin = var_37410_begin_0, end = var_37410_end_0, end_mask = var_37410_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37410_cast")];
            tensor<int32, [4]> var_37414_begin_0 = const()[name = tensor<string, []>("op_37414_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_37414_end_0 = const()[name = tensor<string, []>("op_37414_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_37414_end_mask_0 = const()[name = tensor<string, []>("op_37414_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37414_cast = slice_by_index(begin = var_37414_begin_0, end = var_37414_end_0, end_mask = var_37414_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37414_cast")];
            tensor<int32, [4]> var_37418_begin_0 = const()[name = tensor<string, []>("op_37418_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_37418_end_0 = const()[name = tensor<string, []>("op_37418_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_37418_end_mask_0 = const()[name = tensor<string, []>("op_37418_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37418_cast = slice_by_index(begin = var_37418_begin_0, end = var_37418_end_0, end_mask = var_37418_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37418_cast")];
            tensor<int32, [4]> var_37422_begin_0 = const()[name = tensor<string, []>("op_37422_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_37422_end_0 = const()[name = tensor<string, []>("op_37422_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_37422_end_mask_0 = const()[name = tensor<string, []>("op_37422_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37422_cast = slice_by_index(begin = var_37422_begin_0, end = var_37422_end_0, end_mask = var_37422_end_mask_0, x = q_167_cast)[name = tensor<string, []>("op_37422_cast")];
            tensor<int32, [4]> k_335_perm_0 = const()[name = tensor<string, []>("k_335_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_37429_begin_0 = const()[name = tensor<string, []>("op_37429_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_37429_end_0 = const()[name = tensor<string, []>("op_37429_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_37429_end_mask_0 = const()[name = tensor<string, []>("op_37429_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_56 = transpose(perm = k_335_perm_0, x = k_333_cast)[name = tensor<string, []>("transpose_56")];
            tensor<fp16, [2, 77, 1, 64]> var_37429_cast = slice_by_index(begin = var_37429_begin_0, end = var_37429_end_0, end_mask = var_37429_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37429_cast")];
            tensor<int32, [4]> var_37433_begin_0 = const()[name = tensor<string, []>("op_37433_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_37433_end_0 = const()[name = tensor<string, []>("op_37433_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_37433_end_mask_0 = const()[name = tensor<string, []>("op_37433_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37433_cast = slice_by_index(begin = var_37433_begin_0, end = var_37433_end_0, end_mask = var_37433_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37433_cast")];
            tensor<int32, [4]> var_37437_begin_0 = const()[name = tensor<string, []>("op_37437_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_37437_end_0 = const()[name = tensor<string, []>("op_37437_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_37437_end_mask_0 = const()[name = tensor<string, []>("op_37437_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37437_cast = slice_by_index(begin = var_37437_begin_0, end = var_37437_end_0, end_mask = var_37437_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37437_cast")];
            tensor<int32, [4]> var_37441_begin_0 = const()[name = tensor<string, []>("op_37441_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_37441_end_0 = const()[name = tensor<string, []>("op_37441_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_37441_end_mask_0 = const()[name = tensor<string, []>("op_37441_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37441_cast = slice_by_index(begin = var_37441_begin_0, end = var_37441_end_0, end_mask = var_37441_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37441_cast")];
            tensor<int32, [4]> var_37445_begin_0 = const()[name = tensor<string, []>("op_37445_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_37445_end_0 = const()[name = tensor<string, []>("op_37445_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_37445_end_mask_0 = const()[name = tensor<string, []>("op_37445_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37445_cast = slice_by_index(begin = var_37445_begin_0, end = var_37445_end_0, end_mask = var_37445_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37445_cast")];
            tensor<int32, [4]> var_37449_begin_0 = const()[name = tensor<string, []>("op_37449_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_37449_end_0 = const()[name = tensor<string, []>("op_37449_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_37449_end_mask_0 = const()[name = tensor<string, []>("op_37449_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37449_cast = slice_by_index(begin = var_37449_begin_0, end = var_37449_end_0, end_mask = var_37449_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37449_cast")];
            tensor<int32, [4]> var_37453_begin_0 = const()[name = tensor<string, []>("op_37453_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_37453_end_0 = const()[name = tensor<string, []>("op_37453_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_37453_end_mask_0 = const()[name = tensor<string, []>("op_37453_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37453_cast = slice_by_index(begin = var_37453_begin_0, end = var_37453_end_0, end_mask = var_37453_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37453_cast")];
            tensor<int32, [4]> var_37457_begin_0 = const()[name = tensor<string, []>("op_37457_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_37457_end_0 = const()[name = tensor<string, []>("op_37457_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_37457_end_mask_0 = const()[name = tensor<string, []>("op_37457_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37457_cast = slice_by_index(begin = var_37457_begin_0, end = var_37457_end_0, end_mask = var_37457_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37457_cast")];
            tensor<int32, [4]> var_37461_begin_0 = const()[name = tensor<string, []>("op_37461_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_37461_end_0 = const()[name = tensor<string, []>("op_37461_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_37461_end_mask_0 = const()[name = tensor<string, []>("op_37461_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37461_cast = slice_by_index(begin = var_37461_begin_0, end = var_37461_end_0, end_mask = var_37461_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37461_cast")];
            tensor<int32, [4]> var_37465_begin_0 = const()[name = tensor<string, []>("op_37465_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_37465_end_0 = const()[name = tensor<string, []>("op_37465_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_37465_end_mask_0 = const()[name = tensor<string, []>("op_37465_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37465_cast = slice_by_index(begin = var_37465_begin_0, end = var_37465_end_0, end_mask = var_37465_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37465_cast")];
            tensor<int32, [4]> var_37469_begin_0 = const()[name = tensor<string, []>("op_37469_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_37469_end_0 = const()[name = tensor<string, []>("op_37469_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_37469_end_mask_0 = const()[name = tensor<string, []>("op_37469_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37469_cast = slice_by_index(begin = var_37469_begin_0, end = var_37469_end_0, end_mask = var_37469_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37469_cast")];
            tensor<int32, [4]> var_37473_begin_0 = const()[name = tensor<string, []>("op_37473_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_37473_end_0 = const()[name = tensor<string, []>("op_37473_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_37473_end_mask_0 = const()[name = tensor<string, []>("op_37473_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37473_cast = slice_by_index(begin = var_37473_begin_0, end = var_37473_end_0, end_mask = var_37473_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37473_cast")];
            tensor<int32, [4]> var_37477_begin_0 = const()[name = tensor<string, []>("op_37477_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_37477_end_0 = const()[name = tensor<string, []>("op_37477_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_37477_end_mask_0 = const()[name = tensor<string, []>("op_37477_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37477_cast = slice_by_index(begin = var_37477_begin_0, end = var_37477_end_0, end_mask = var_37477_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37477_cast")];
            tensor<int32, [4]> var_37481_begin_0 = const()[name = tensor<string, []>("op_37481_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_37481_end_0 = const()[name = tensor<string, []>("op_37481_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_37481_end_mask_0 = const()[name = tensor<string, []>("op_37481_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37481_cast = slice_by_index(begin = var_37481_begin_0, end = var_37481_end_0, end_mask = var_37481_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37481_cast")];
            tensor<int32, [4]> var_37485_begin_0 = const()[name = tensor<string, []>("op_37485_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_37485_end_0 = const()[name = tensor<string, []>("op_37485_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_37485_end_mask_0 = const()[name = tensor<string, []>("op_37485_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37485_cast = slice_by_index(begin = var_37485_begin_0, end = var_37485_end_0, end_mask = var_37485_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37485_cast")];
            tensor<int32, [4]> var_37489_begin_0 = const()[name = tensor<string, []>("op_37489_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_37489_end_0 = const()[name = tensor<string, []>("op_37489_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_37489_end_mask_0 = const()[name = tensor<string, []>("op_37489_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37489_cast = slice_by_index(begin = var_37489_begin_0, end = var_37489_end_0, end_mask = var_37489_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37489_cast")];
            tensor<int32, [4]> var_37493_begin_0 = const()[name = tensor<string, []>("op_37493_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_37493_end_0 = const()[name = tensor<string, []>("op_37493_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_37493_end_mask_0 = const()[name = tensor<string, []>("op_37493_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37493_cast = slice_by_index(begin = var_37493_begin_0, end = var_37493_end_0, end_mask = var_37493_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37493_cast")];
            tensor<int32, [4]> var_37497_begin_0 = const()[name = tensor<string, []>("op_37497_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_37497_end_0 = const()[name = tensor<string, []>("op_37497_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_37497_end_mask_0 = const()[name = tensor<string, []>("op_37497_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37497_cast = slice_by_index(begin = var_37497_begin_0, end = var_37497_end_0, end_mask = var_37497_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37497_cast")];
            tensor<int32, [4]> var_37501_begin_0 = const()[name = tensor<string, []>("op_37501_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_37501_end_0 = const()[name = tensor<string, []>("op_37501_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_37501_end_mask_0 = const()[name = tensor<string, []>("op_37501_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37501_cast = slice_by_index(begin = var_37501_begin_0, end = var_37501_end_0, end_mask = var_37501_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37501_cast")];
            tensor<int32, [4]> var_37505_begin_0 = const()[name = tensor<string, []>("op_37505_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_37505_end_0 = const()[name = tensor<string, []>("op_37505_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_37505_end_mask_0 = const()[name = tensor<string, []>("op_37505_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_37505_cast = slice_by_index(begin = var_37505_begin_0, end = var_37505_end_0, end_mask = var_37505_end_mask_0, x = transpose_56)[name = tensor<string, []>("op_37505_cast")];
            tensor<int32, [4]> var_37507_begin_0 = const()[name = tensor<string, []>("op_37507_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_37507_end_0 = const()[name = tensor<string, []>("op_37507_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_37507_end_mask_0 = const()[name = tensor<string, []>("op_37507_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37507_cast = slice_by_index(begin = var_37507_begin_0, end = var_37507_end_0, end_mask = var_37507_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37507_cast")];
            tensor<int32, [4]> var_37511_begin_0 = const()[name = tensor<string, []>("op_37511_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_37511_end_0 = const()[name = tensor<string, []>("op_37511_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_37511_end_mask_0 = const()[name = tensor<string, []>("op_37511_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37511_cast = slice_by_index(begin = var_37511_begin_0, end = var_37511_end_0, end_mask = var_37511_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37511_cast")];
            tensor<int32, [4]> var_37515_begin_0 = const()[name = tensor<string, []>("op_37515_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_37515_end_0 = const()[name = tensor<string, []>("op_37515_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_37515_end_mask_0 = const()[name = tensor<string, []>("op_37515_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37515_cast = slice_by_index(begin = var_37515_begin_0, end = var_37515_end_0, end_mask = var_37515_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37515_cast")];
            tensor<int32, [4]> var_37519_begin_0 = const()[name = tensor<string, []>("op_37519_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_37519_end_0 = const()[name = tensor<string, []>("op_37519_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_37519_end_mask_0 = const()[name = tensor<string, []>("op_37519_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37519_cast = slice_by_index(begin = var_37519_begin_0, end = var_37519_end_0, end_mask = var_37519_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37519_cast")];
            tensor<int32, [4]> var_37523_begin_0 = const()[name = tensor<string, []>("op_37523_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_37523_end_0 = const()[name = tensor<string, []>("op_37523_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_37523_end_mask_0 = const()[name = tensor<string, []>("op_37523_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37523_cast = slice_by_index(begin = var_37523_begin_0, end = var_37523_end_0, end_mask = var_37523_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37523_cast")];
            tensor<int32, [4]> var_37527_begin_0 = const()[name = tensor<string, []>("op_37527_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_37527_end_0 = const()[name = tensor<string, []>("op_37527_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_37527_end_mask_0 = const()[name = tensor<string, []>("op_37527_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37527_cast = slice_by_index(begin = var_37527_begin_0, end = var_37527_end_0, end_mask = var_37527_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37527_cast")];
            tensor<int32, [4]> var_37531_begin_0 = const()[name = tensor<string, []>("op_37531_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_37531_end_0 = const()[name = tensor<string, []>("op_37531_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_37531_end_mask_0 = const()[name = tensor<string, []>("op_37531_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37531_cast = slice_by_index(begin = var_37531_begin_0, end = var_37531_end_0, end_mask = var_37531_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37531_cast")];
            tensor<int32, [4]> var_37535_begin_0 = const()[name = tensor<string, []>("op_37535_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_37535_end_0 = const()[name = tensor<string, []>("op_37535_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_37535_end_mask_0 = const()[name = tensor<string, []>("op_37535_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37535_cast = slice_by_index(begin = var_37535_begin_0, end = var_37535_end_0, end_mask = var_37535_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37535_cast")];
            tensor<int32, [4]> var_37539_begin_0 = const()[name = tensor<string, []>("op_37539_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_37539_end_0 = const()[name = tensor<string, []>("op_37539_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_37539_end_mask_0 = const()[name = tensor<string, []>("op_37539_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37539_cast = slice_by_index(begin = var_37539_begin_0, end = var_37539_end_0, end_mask = var_37539_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37539_cast")];
            tensor<int32, [4]> var_37543_begin_0 = const()[name = tensor<string, []>("op_37543_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_37543_end_0 = const()[name = tensor<string, []>("op_37543_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_37543_end_mask_0 = const()[name = tensor<string, []>("op_37543_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37543_cast = slice_by_index(begin = var_37543_begin_0, end = var_37543_end_0, end_mask = var_37543_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37543_cast")];
            tensor<int32, [4]> var_37547_begin_0 = const()[name = tensor<string, []>("op_37547_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_37547_end_0 = const()[name = tensor<string, []>("op_37547_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_37547_end_mask_0 = const()[name = tensor<string, []>("op_37547_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37547_cast = slice_by_index(begin = var_37547_begin_0, end = var_37547_end_0, end_mask = var_37547_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37547_cast")];
            tensor<int32, [4]> var_37551_begin_0 = const()[name = tensor<string, []>("op_37551_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_37551_end_0 = const()[name = tensor<string, []>("op_37551_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_37551_end_mask_0 = const()[name = tensor<string, []>("op_37551_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37551_cast = slice_by_index(begin = var_37551_begin_0, end = var_37551_end_0, end_mask = var_37551_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37551_cast")];
            tensor<int32, [4]> var_37555_begin_0 = const()[name = tensor<string, []>("op_37555_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_37555_end_0 = const()[name = tensor<string, []>("op_37555_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_37555_end_mask_0 = const()[name = tensor<string, []>("op_37555_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37555_cast = slice_by_index(begin = var_37555_begin_0, end = var_37555_end_0, end_mask = var_37555_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37555_cast")];
            tensor<int32, [4]> var_37559_begin_0 = const()[name = tensor<string, []>("op_37559_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_37559_end_0 = const()[name = tensor<string, []>("op_37559_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_37559_end_mask_0 = const()[name = tensor<string, []>("op_37559_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37559_cast = slice_by_index(begin = var_37559_begin_0, end = var_37559_end_0, end_mask = var_37559_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37559_cast")];
            tensor<int32, [4]> var_37563_begin_0 = const()[name = tensor<string, []>("op_37563_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_37563_end_0 = const()[name = tensor<string, []>("op_37563_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_37563_end_mask_0 = const()[name = tensor<string, []>("op_37563_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37563_cast = slice_by_index(begin = var_37563_begin_0, end = var_37563_end_0, end_mask = var_37563_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37563_cast")];
            tensor<int32, [4]> var_37567_begin_0 = const()[name = tensor<string, []>("op_37567_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_37567_end_0 = const()[name = tensor<string, []>("op_37567_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_37567_end_mask_0 = const()[name = tensor<string, []>("op_37567_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37567_cast = slice_by_index(begin = var_37567_begin_0, end = var_37567_end_0, end_mask = var_37567_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37567_cast")];
            tensor<int32, [4]> var_37571_begin_0 = const()[name = tensor<string, []>("op_37571_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_37571_end_0 = const()[name = tensor<string, []>("op_37571_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_37571_end_mask_0 = const()[name = tensor<string, []>("op_37571_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37571_cast = slice_by_index(begin = var_37571_begin_0, end = var_37571_end_0, end_mask = var_37571_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37571_cast")];
            tensor<int32, [4]> var_37575_begin_0 = const()[name = tensor<string, []>("op_37575_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_37575_end_0 = const()[name = tensor<string, []>("op_37575_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_37575_end_mask_0 = const()[name = tensor<string, []>("op_37575_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37575_cast = slice_by_index(begin = var_37575_begin_0, end = var_37575_end_0, end_mask = var_37575_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37575_cast")];
            tensor<int32, [4]> var_37579_begin_0 = const()[name = tensor<string, []>("op_37579_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_37579_end_0 = const()[name = tensor<string, []>("op_37579_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_37579_end_mask_0 = const()[name = tensor<string, []>("op_37579_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37579_cast = slice_by_index(begin = var_37579_begin_0, end = var_37579_end_0, end_mask = var_37579_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37579_cast")];
            tensor<int32, [4]> var_37583_begin_0 = const()[name = tensor<string, []>("op_37583_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_37583_end_0 = const()[name = tensor<string, []>("op_37583_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_37583_end_mask_0 = const()[name = tensor<string, []>("op_37583_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_37583_cast = slice_by_index(begin = var_37583_begin_0, end = var_37583_end_0, end_mask = var_37583_end_mask_0, x = v_167_cast)[name = tensor<string, []>("op_37583_cast")];
            tensor<string, []> var_37587_equation_0 = const()[name = tensor<string, []>("op_37587_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37587_cast = einsum(equation = var_37587_equation_0, values = (var_37429_cast, var_37346_cast))[name = tensor<string, []>("op_37587_cast")];
            tensor<fp16, []> var_37588_to_fp16 = const()[name = tensor<string, []>("op_37588_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3161_cast = mul(x = var_37587_cast, y = var_37588_to_fp16)[name = tensor<string, []>("aw_3161_cast")];
            tensor<string, []> var_37591_equation_0 = const()[name = tensor<string, []>("op_37591_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37591_cast = einsum(equation = var_37591_equation_0, values = (var_37433_cast, var_37350_cast))[name = tensor<string, []>("op_37591_cast")];
            tensor<fp16, []> var_37592_to_fp16 = const()[name = tensor<string, []>("op_37592_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3163_cast = mul(x = var_37591_cast, y = var_37592_to_fp16)[name = tensor<string, []>("aw_3163_cast")];
            tensor<string, []> var_37595_equation_0 = const()[name = tensor<string, []>("op_37595_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37595_cast = einsum(equation = var_37595_equation_0, values = (var_37437_cast, var_37354_cast))[name = tensor<string, []>("op_37595_cast")];
            tensor<fp16, []> var_37596_to_fp16 = const()[name = tensor<string, []>("op_37596_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3165_cast = mul(x = var_37595_cast, y = var_37596_to_fp16)[name = tensor<string, []>("aw_3165_cast")];
            tensor<string, []> var_37599_equation_0 = const()[name = tensor<string, []>("op_37599_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37599_cast = einsum(equation = var_37599_equation_0, values = (var_37441_cast, var_37358_cast))[name = tensor<string, []>("op_37599_cast")];
            tensor<fp16, []> var_37600_to_fp16 = const()[name = tensor<string, []>("op_37600_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3167_cast = mul(x = var_37599_cast, y = var_37600_to_fp16)[name = tensor<string, []>("aw_3167_cast")];
            tensor<string, []> var_37603_equation_0 = const()[name = tensor<string, []>("op_37603_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37603_cast = einsum(equation = var_37603_equation_0, values = (var_37445_cast, var_37362_cast))[name = tensor<string, []>("op_37603_cast")];
            tensor<fp16, []> var_37604_to_fp16 = const()[name = tensor<string, []>("op_37604_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3169_cast = mul(x = var_37603_cast, y = var_37604_to_fp16)[name = tensor<string, []>("aw_3169_cast")];
            tensor<string, []> var_37607_equation_0 = const()[name = tensor<string, []>("op_37607_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37607_cast = einsum(equation = var_37607_equation_0, values = (var_37449_cast, var_37366_cast))[name = tensor<string, []>("op_37607_cast")];
            tensor<fp16, []> var_37608_to_fp16 = const()[name = tensor<string, []>("op_37608_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3171_cast = mul(x = var_37607_cast, y = var_37608_to_fp16)[name = tensor<string, []>("aw_3171_cast")];
            tensor<string, []> var_37611_equation_0 = const()[name = tensor<string, []>("op_37611_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37611_cast = einsum(equation = var_37611_equation_0, values = (var_37453_cast, var_37370_cast))[name = tensor<string, []>("op_37611_cast")];
            tensor<fp16, []> var_37612_to_fp16 = const()[name = tensor<string, []>("op_37612_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3173_cast = mul(x = var_37611_cast, y = var_37612_to_fp16)[name = tensor<string, []>("aw_3173_cast")];
            tensor<string, []> var_37615_equation_0 = const()[name = tensor<string, []>("op_37615_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37615_cast = einsum(equation = var_37615_equation_0, values = (var_37457_cast, var_37374_cast))[name = tensor<string, []>("op_37615_cast")];
            tensor<fp16, []> var_37616_to_fp16 = const()[name = tensor<string, []>("op_37616_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3175_cast = mul(x = var_37615_cast, y = var_37616_to_fp16)[name = tensor<string, []>("aw_3175_cast")];
            tensor<string, []> var_37619_equation_0 = const()[name = tensor<string, []>("op_37619_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37619_cast = einsum(equation = var_37619_equation_0, values = (var_37461_cast, var_37378_cast))[name = tensor<string, []>("op_37619_cast")];
            tensor<fp16, []> var_37620_to_fp16 = const()[name = tensor<string, []>("op_37620_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3177_cast = mul(x = var_37619_cast, y = var_37620_to_fp16)[name = tensor<string, []>("aw_3177_cast")];
            tensor<string, []> var_37623_equation_0 = const()[name = tensor<string, []>("op_37623_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37623_cast = einsum(equation = var_37623_equation_0, values = (var_37465_cast, var_37382_cast))[name = tensor<string, []>("op_37623_cast")];
            tensor<fp16, []> var_37624_to_fp16 = const()[name = tensor<string, []>("op_37624_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3179_cast = mul(x = var_37623_cast, y = var_37624_to_fp16)[name = tensor<string, []>("aw_3179_cast")];
            tensor<string, []> var_37627_equation_0 = const()[name = tensor<string, []>("op_37627_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37627_cast = einsum(equation = var_37627_equation_0, values = (var_37469_cast, var_37386_cast))[name = tensor<string, []>("op_37627_cast")];
            tensor<fp16, []> var_37628_to_fp16 = const()[name = tensor<string, []>("op_37628_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3181_cast = mul(x = var_37627_cast, y = var_37628_to_fp16)[name = tensor<string, []>("aw_3181_cast")];
            tensor<string, []> var_37631_equation_0 = const()[name = tensor<string, []>("op_37631_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37631_cast = einsum(equation = var_37631_equation_0, values = (var_37473_cast, var_37390_cast))[name = tensor<string, []>("op_37631_cast")];
            tensor<fp16, []> var_37632_to_fp16 = const()[name = tensor<string, []>("op_37632_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3183_cast = mul(x = var_37631_cast, y = var_37632_to_fp16)[name = tensor<string, []>("aw_3183_cast")];
            tensor<string, []> var_37635_equation_0 = const()[name = tensor<string, []>("op_37635_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37635_cast = einsum(equation = var_37635_equation_0, values = (var_37477_cast, var_37394_cast))[name = tensor<string, []>("op_37635_cast")];
            tensor<fp16, []> var_37636_to_fp16 = const()[name = tensor<string, []>("op_37636_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3185_cast = mul(x = var_37635_cast, y = var_37636_to_fp16)[name = tensor<string, []>("aw_3185_cast")];
            tensor<string, []> var_37639_equation_0 = const()[name = tensor<string, []>("op_37639_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37639_cast = einsum(equation = var_37639_equation_0, values = (var_37481_cast, var_37398_cast))[name = tensor<string, []>("op_37639_cast")];
            tensor<fp16, []> var_37640_to_fp16 = const()[name = tensor<string, []>("op_37640_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3187_cast = mul(x = var_37639_cast, y = var_37640_to_fp16)[name = tensor<string, []>("aw_3187_cast")];
            tensor<string, []> var_37643_equation_0 = const()[name = tensor<string, []>("op_37643_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37643_cast = einsum(equation = var_37643_equation_0, values = (var_37485_cast, var_37402_cast))[name = tensor<string, []>("op_37643_cast")];
            tensor<fp16, []> var_37644_to_fp16 = const()[name = tensor<string, []>("op_37644_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3189_cast = mul(x = var_37643_cast, y = var_37644_to_fp16)[name = tensor<string, []>("aw_3189_cast")];
            tensor<string, []> var_37647_equation_0 = const()[name = tensor<string, []>("op_37647_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37647_cast = einsum(equation = var_37647_equation_0, values = (var_37489_cast, var_37406_cast))[name = tensor<string, []>("op_37647_cast")];
            tensor<fp16, []> var_37648_to_fp16 = const()[name = tensor<string, []>("op_37648_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3191_cast = mul(x = var_37647_cast, y = var_37648_to_fp16)[name = tensor<string, []>("aw_3191_cast")];
            tensor<string, []> var_37651_equation_0 = const()[name = tensor<string, []>("op_37651_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37651_cast = einsum(equation = var_37651_equation_0, values = (var_37493_cast, var_37410_cast))[name = tensor<string, []>("op_37651_cast")];
            tensor<fp16, []> var_37652_to_fp16 = const()[name = tensor<string, []>("op_37652_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3193_cast = mul(x = var_37651_cast, y = var_37652_to_fp16)[name = tensor<string, []>("aw_3193_cast")];
            tensor<string, []> var_37655_equation_0 = const()[name = tensor<string, []>("op_37655_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37655_cast = einsum(equation = var_37655_equation_0, values = (var_37497_cast, var_37414_cast))[name = tensor<string, []>("op_37655_cast")];
            tensor<fp16, []> var_37656_to_fp16 = const()[name = tensor<string, []>("op_37656_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3195_cast = mul(x = var_37655_cast, y = var_37656_to_fp16)[name = tensor<string, []>("aw_3195_cast")];
            tensor<string, []> var_37659_equation_0 = const()[name = tensor<string, []>("op_37659_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37659_cast = einsum(equation = var_37659_equation_0, values = (var_37501_cast, var_37418_cast))[name = tensor<string, []>("op_37659_cast")];
            tensor<fp16, []> var_37660_to_fp16 = const()[name = tensor<string, []>("op_37660_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3197_cast = mul(x = var_37659_cast, y = var_37660_to_fp16)[name = tensor<string, []>("aw_3197_cast")];
            tensor<string, []> var_37663_equation_0 = const()[name = tensor<string, []>("op_37663_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_37663_cast = einsum(equation = var_37663_equation_0, values = (var_37505_cast, var_37422_cast))[name = tensor<string, []>("op_37663_cast")];
            tensor<fp16, []> var_37664_to_fp16 = const()[name = tensor<string, []>("op_37664_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3199_cast = mul(x = var_37663_cast, y = var_37664_to_fp16)[name = tensor<string, []>("aw_3199_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37666_cast = softmax(axis = var_30385, x = aw_3161_cast)[name = tensor<string, []>("op_37666_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37667_cast = softmax(axis = var_30385, x = aw_3163_cast)[name = tensor<string, []>("op_37667_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37668_cast = softmax(axis = var_30385, x = aw_3165_cast)[name = tensor<string, []>("op_37668_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37669_cast = softmax(axis = var_30385, x = aw_3167_cast)[name = tensor<string, []>("op_37669_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37670_cast = softmax(axis = var_30385, x = aw_3169_cast)[name = tensor<string, []>("op_37670_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37671_cast = softmax(axis = var_30385, x = aw_3171_cast)[name = tensor<string, []>("op_37671_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37672_cast = softmax(axis = var_30385, x = aw_3173_cast)[name = tensor<string, []>("op_37672_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37673_cast = softmax(axis = var_30385, x = aw_3175_cast)[name = tensor<string, []>("op_37673_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37674_cast = softmax(axis = var_30385, x = aw_3177_cast)[name = tensor<string, []>("op_37674_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37675_cast = softmax(axis = var_30385, x = aw_3179_cast)[name = tensor<string, []>("op_37675_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37676_cast = softmax(axis = var_30385, x = aw_3181_cast)[name = tensor<string, []>("op_37676_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37677_cast = softmax(axis = var_30385, x = aw_3183_cast)[name = tensor<string, []>("op_37677_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37678_cast = softmax(axis = var_30385, x = aw_3185_cast)[name = tensor<string, []>("op_37678_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37679_cast = softmax(axis = var_30385, x = aw_3187_cast)[name = tensor<string, []>("op_37679_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37680_cast = softmax(axis = var_30385, x = aw_3189_cast)[name = tensor<string, []>("op_37680_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37681_cast = softmax(axis = var_30385, x = aw_3191_cast)[name = tensor<string, []>("op_37681_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37682_cast = softmax(axis = var_30385, x = aw_3193_cast)[name = tensor<string, []>("op_37682_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37683_cast = softmax(axis = var_30385, x = aw_3195_cast)[name = tensor<string, []>("op_37683_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37684_cast = softmax(axis = var_30385, x = aw_3197_cast)[name = tensor<string, []>("op_37684_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_37685_cast = softmax(axis = var_30385, x = aw_3199_cast)[name = tensor<string, []>("op_37685_cast")];
            tensor<string, []> var_37687_equation_0 = const()[name = tensor<string, []>("op_37687_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37687_cast = einsum(equation = var_37687_equation_0, values = (var_37507_cast, var_37666_cast))[name = tensor<string, []>("op_37687_cast")];
            tensor<string, []> var_37689_equation_0 = const()[name = tensor<string, []>("op_37689_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37689_cast = einsum(equation = var_37689_equation_0, values = (var_37511_cast, var_37667_cast))[name = tensor<string, []>("op_37689_cast")];
            tensor<string, []> var_37691_equation_0 = const()[name = tensor<string, []>("op_37691_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37691_cast = einsum(equation = var_37691_equation_0, values = (var_37515_cast, var_37668_cast))[name = tensor<string, []>("op_37691_cast")];
            tensor<string, []> var_37693_equation_0 = const()[name = tensor<string, []>("op_37693_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37693_cast = einsum(equation = var_37693_equation_0, values = (var_37519_cast, var_37669_cast))[name = tensor<string, []>("op_37693_cast")];
            tensor<string, []> var_37695_equation_0 = const()[name = tensor<string, []>("op_37695_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37695_cast = einsum(equation = var_37695_equation_0, values = (var_37523_cast, var_37670_cast))[name = tensor<string, []>("op_37695_cast")];
            tensor<string, []> var_37697_equation_0 = const()[name = tensor<string, []>("op_37697_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37697_cast = einsum(equation = var_37697_equation_0, values = (var_37527_cast, var_37671_cast))[name = tensor<string, []>("op_37697_cast")];
            tensor<string, []> var_37699_equation_0 = const()[name = tensor<string, []>("op_37699_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37699_cast = einsum(equation = var_37699_equation_0, values = (var_37531_cast, var_37672_cast))[name = tensor<string, []>("op_37699_cast")];
            tensor<string, []> var_37701_equation_0 = const()[name = tensor<string, []>("op_37701_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37701_cast = einsum(equation = var_37701_equation_0, values = (var_37535_cast, var_37673_cast))[name = tensor<string, []>("op_37701_cast")];
            tensor<string, []> var_37703_equation_0 = const()[name = tensor<string, []>("op_37703_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37703_cast = einsum(equation = var_37703_equation_0, values = (var_37539_cast, var_37674_cast))[name = tensor<string, []>("op_37703_cast")];
            tensor<string, []> var_37705_equation_0 = const()[name = tensor<string, []>("op_37705_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37705_cast = einsum(equation = var_37705_equation_0, values = (var_37543_cast, var_37675_cast))[name = tensor<string, []>("op_37705_cast")];
            tensor<string, []> var_37707_equation_0 = const()[name = tensor<string, []>("op_37707_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37707_cast = einsum(equation = var_37707_equation_0, values = (var_37547_cast, var_37676_cast))[name = tensor<string, []>("op_37707_cast")];
            tensor<string, []> var_37709_equation_0 = const()[name = tensor<string, []>("op_37709_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37709_cast = einsum(equation = var_37709_equation_0, values = (var_37551_cast, var_37677_cast))[name = tensor<string, []>("op_37709_cast")];
            tensor<string, []> var_37711_equation_0 = const()[name = tensor<string, []>("op_37711_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37711_cast = einsum(equation = var_37711_equation_0, values = (var_37555_cast, var_37678_cast))[name = tensor<string, []>("op_37711_cast")];
            tensor<string, []> var_37713_equation_0 = const()[name = tensor<string, []>("op_37713_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37713_cast = einsum(equation = var_37713_equation_0, values = (var_37559_cast, var_37679_cast))[name = tensor<string, []>("op_37713_cast")];
            tensor<string, []> var_37715_equation_0 = const()[name = tensor<string, []>("op_37715_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37715_cast = einsum(equation = var_37715_equation_0, values = (var_37563_cast, var_37680_cast))[name = tensor<string, []>("op_37715_cast")];
            tensor<string, []> var_37717_equation_0 = const()[name = tensor<string, []>("op_37717_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37717_cast = einsum(equation = var_37717_equation_0, values = (var_37567_cast, var_37681_cast))[name = tensor<string, []>("op_37717_cast")];
            tensor<string, []> var_37719_equation_0 = const()[name = tensor<string, []>("op_37719_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37719_cast = einsum(equation = var_37719_equation_0, values = (var_37571_cast, var_37682_cast))[name = tensor<string, []>("op_37719_cast")];
            tensor<string, []> var_37721_equation_0 = const()[name = tensor<string, []>("op_37721_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37721_cast = einsum(equation = var_37721_equation_0, values = (var_37575_cast, var_37683_cast))[name = tensor<string, []>("op_37721_cast")];
            tensor<string, []> var_37723_equation_0 = const()[name = tensor<string, []>("op_37723_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37723_cast = einsum(equation = var_37723_equation_0, values = (var_37579_cast, var_37684_cast))[name = tensor<string, []>("op_37723_cast")];
            tensor<string, []> var_37725_equation_0 = const()[name = tensor<string, []>("op_37725_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_37725_cast = einsum(equation = var_37725_equation_0, values = (var_37583_cast, var_37685_cast))[name = tensor<string, []>("op_37725_cast")];
            tensor<bool, []> input_499_interleave_0 = const()[name = tensor<string, []>("input_499_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_499_cast = concat(axis = var_30385, interleave = input_499_interleave_0, values = (var_37687_cast, var_37689_cast, var_37691_cast, var_37693_cast, var_37695_cast, var_37697_cast, var_37699_cast, var_37701_cast, var_37703_cast, var_37705_cast, var_37707_cast, var_37709_cast, var_37711_cast, var_37713_cast, var_37715_cast, var_37717_cast, var_37719_cast, var_37721_cast, var_37723_cast, var_37725_cast))[name = tensor<string, []>("input_499_cast")];
            tensor<int32, [2]> var_37731 = const()[name = tensor<string, []>("op_37731"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_37733 = const()[name = tensor<string, []>("op_37733"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_37735_pad_type_0 = const()[name = tensor<string, []>("op_37735_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_37735_pad_0 = const()[name = tensor<string, []>("op_37735_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3101895424)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3105172288)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_37735_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_out_0_bias_to_fp16, dilations = var_37733, groups = var_30385, pad = var_37735_pad_0, pad_type = var_37735_pad_type_0, strides = var_37731, weight = up_blocks_0_attentions_0_transformer_blocks_7_attn2_to_out_0_weight_to_fp16, x = input_499_cast)[name = tensor<string, []>("op_37735_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_251_cast = add(x = var_37735_cast, y = inputs_249_cast)[name = tensor<string, []>("inputs_251_cast")];
            tensor<int32, [1]> var_37739 = const()[name = tensor<string, []>("op_37739"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_251_cast = reduce_mean(axes = var_37739, keep_dims = var_30380, x = inputs_251_cast)[name = tensor<string, []>("channels_mean_251_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_251_cast = sub(x = inputs_251_cast, y = channels_mean_251_cast)[name = tensor<string, []>("zero_mean_251_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_251_cast = mul(x = zero_mean_251_cast, y = zero_mean_251_cast)[name = tensor<string, []>("zero_mean_sq_251_cast")];
            tensor<int32, [1]> var_37743 = const()[name = tensor<string, []>("op_37743"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_37744_cast = reduce_mean(axes = var_37743, keep_dims = var_30380, x = zero_mean_sq_251_cast)[name = tensor<string, []>("op_37744_cast")];
            tensor<fp16, []> var_37745_to_fp16 = const()[name = tensor<string, []>("op_37745_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_37746_cast = add(x = var_37744_cast, y = var_37745_to_fp16)[name = tensor<string, []>("op_37746_cast")];
            tensor<fp16, []> denom_251_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_251_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_251_cast = rsqrt(epsilon = denom_251_epsilon_0_to_fp16, x = var_37746_cast)[name = tensor<string, []>("denom_251_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_251_cast = mul(x = zero_mean_251_cast, y = denom_251_cast)[name = tensor<string, []>("out_251_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_37750_to_fp16 = const()[name = tensor<string, []>("op_37750_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3105174912)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_37751_cast = add(x = out_251_cast, y = var_37750_to_fp16)[name = tensor<string, []>("op_37751_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_37753_to_fp16 = const()[name = tensor<string, []>("op_37753_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3105177536)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_501_cast = mul(x = var_37751_cast, y = var_37753_to_fp16)[name = tensor<string, []>("input_501_cast")];
            tensor<int32, [2]> var_37761 = const()[name = tensor<string, []>("op_37761"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_37763 = const()[name = tensor<string, []>("op_37763"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_37765_pad_type_0 = const()[name = tensor<string, []>("op_37765_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_37765_pad_0 = const()[name = tensor<string, []>("op_37765_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3105180160)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3131394624)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_37765_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_7_ff_net_0_proj_bias_to_fp16, dilations = var_37763, groups = var_30385, pad = var_37765_pad_0, pad_type = var_37765_pad_type_0, strides = var_37761, weight = up_blocks_0_attentions_0_transformer_blocks_7_ff_net_0_proj_weight_to_fp16, x = input_501_cast)[name = tensor<string, []>("op_37765_cast")];
            tensor<int32, [2]> var_37766_split_sizes_0 = const()[name = tensor<string, []>("op_37766_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_37766_axis_0 = const()[name = tensor<string, []>("op_37766_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_37766_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_37766_cast_1 = split(axis = var_37766_axis_0, split_sizes = var_37766_split_sizes_0, x = var_37765_cast)[name = tensor<string, []>("op_37766_cast")];
            tensor<string, []> var_37768_mode_0 = const()[name = tensor<string, []>("op_37768_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_37768_cast = gelu(mode = var_37768_mode_0, x = var_37766_cast_1)[name = tensor<string, []>("op_37768_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_503_cast = mul(x = var_37766_cast_0, y = var_37768_cast)[name = tensor<string, []>("input_503_cast")];
            tensor<int32, [2]> var_37772 = const()[name = tensor<string, []>("op_37772"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_37774 = const()[name = tensor<string, []>("op_37774"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_37776_pad_type_0 = const()[name = tensor<string, []>("op_37776_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_37776_pad_0 = const()[name = tensor<string, []>("op_37776_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3131415168)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3144522432)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_37776_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_7_ff_net_2_bias_to_fp16, dilations = var_37774, groups = var_30385, pad = var_37776_pad_0, pad_type = var_37776_pad_type_0, strides = var_37772, weight = up_blocks_0_attentions_0_transformer_blocks_7_ff_net_2_weight_to_fp16, x = input_503_cast)[name = tensor<string, []>("op_37776_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_253_cast = add(x = var_37776_cast, y = inputs_251_cast)[name = tensor<string, []>("inputs_253_cast")];
            tensor<int32, [1]> var_37786 = const()[name = tensor<string, []>("op_37786"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_253_cast = reduce_mean(axes = var_37786, keep_dims = var_30380, x = inputs_253_cast)[name = tensor<string, []>("channels_mean_253_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_253_cast = sub(x = inputs_253_cast, y = channels_mean_253_cast)[name = tensor<string, []>("zero_mean_253_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_253_cast = mul(x = zero_mean_253_cast, y = zero_mean_253_cast)[name = tensor<string, []>("zero_mean_sq_253_cast")];
            tensor<int32, [1]> var_37790 = const()[name = tensor<string, []>("op_37790"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_37791_cast = reduce_mean(axes = var_37790, keep_dims = var_30380, x = zero_mean_sq_253_cast)[name = tensor<string, []>("op_37791_cast")];
            tensor<fp16, []> var_37792_to_fp16 = const()[name = tensor<string, []>("op_37792_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_37793_cast = add(x = var_37791_cast, y = var_37792_to_fp16)[name = tensor<string, []>("op_37793_cast")];
            tensor<fp16, []> denom_253_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_253_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_253_cast = rsqrt(epsilon = denom_253_epsilon_0_to_fp16, x = var_37793_cast)[name = tensor<string, []>("denom_253_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_253_cast = mul(x = zero_mean_253_cast, y = denom_253_cast)[name = tensor<string, []>("out_253_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_37797_to_fp16 = const()[name = tensor<string, []>("op_37797_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3144525056)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_37798_cast = add(x = out_253_cast, y = var_37797_to_fp16)[name = tensor<string, []>("op_37798_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_37800_to_fp16 = const()[name = tensor<string, []>("op_37800_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3144527680)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_335_cast = mul(x = var_37798_cast, y = var_37800_to_fp16)[name = tensor<string, []>("hidden_states_335_cast")];
            tensor<int32, [2]> var_37807 = const()[name = tensor<string, []>("op_37807"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_37809 = const()[name = tensor<string, []>("op_37809"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_169_pad_type_0 = const()[name = tensor<string, []>("q_169_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_169_pad_0 = const()[name = tensor<string, []>("q_169_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3144530304)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_169_cast = conv(dilations = var_37809, groups = var_30385, pad = q_169_pad_0, pad_type = q_169_pad_type_0, strides = var_37807, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_q_weight_to_fp16, x = hidden_states_335_cast)[name = tensor<string, []>("q_169_cast")];
            tensor<int32, [2]> var_37813 = const()[name = tensor<string, []>("op_37813"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_37815 = const()[name = tensor<string, []>("op_37815"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_337_pad_type_0 = const()[name = tensor<string, []>("k_337_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_337_pad_0 = const()[name = tensor<string, []>("k_337_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3147807168)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_337_cast = conv(dilations = var_37815, groups = var_30385, pad = k_337_pad_0, pad_type = k_337_pad_type_0, strides = var_37813, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_k_weight_to_fp16, x = hidden_states_335_cast)[name = tensor<string, []>("k_337_cast")];
            tensor<int32, [2]> var_37819 = const()[name = tensor<string, []>("op_37819"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_37821 = const()[name = tensor<string, []>("op_37821"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_169_pad_type_0 = const()[name = tensor<string, []>("v_169_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_169_pad_0 = const()[name = tensor<string, []>("v_169_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3151084032)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_169_cast = conv(dilations = var_37821, groups = var_30385, pad = v_169_pad_0, pad_type = v_169_pad_type_0, strides = var_37819, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_v_weight_to_fp16, x = hidden_states_335_cast)[name = tensor<string, []>("v_169_cast")];
            tensor<int32, [4]> var_37825_begin_0 = const()[name = tensor<string, []>("op_37825_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_37825_end_0 = const()[name = tensor<string, []>("op_37825_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_37825_end_mask_0 = const()[name = tensor<string, []>("op_37825_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37825_cast = slice_by_index(begin = var_37825_begin_0, end = var_37825_end_0, end_mask = var_37825_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37825_cast")];
            tensor<int32, [4]> var_37829_begin_0 = const()[name = tensor<string, []>("op_37829_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_37829_end_0 = const()[name = tensor<string, []>("op_37829_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_37829_end_mask_0 = const()[name = tensor<string, []>("op_37829_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37829_cast = slice_by_index(begin = var_37829_begin_0, end = var_37829_end_0, end_mask = var_37829_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37829_cast")];
            tensor<int32, [4]> var_37833_begin_0 = const()[name = tensor<string, []>("op_37833_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_37833_end_0 = const()[name = tensor<string, []>("op_37833_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_37833_end_mask_0 = const()[name = tensor<string, []>("op_37833_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37833_cast = slice_by_index(begin = var_37833_begin_0, end = var_37833_end_0, end_mask = var_37833_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37833_cast")];
            tensor<int32, [4]> var_37837_begin_0 = const()[name = tensor<string, []>("op_37837_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_37837_end_0 = const()[name = tensor<string, []>("op_37837_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_37837_end_mask_0 = const()[name = tensor<string, []>("op_37837_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37837_cast = slice_by_index(begin = var_37837_begin_0, end = var_37837_end_0, end_mask = var_37837_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37837_cast")];
            tensor<int32, [4]> var_37841_begin_0 = const()[name = tensor<string, []>("op_37841_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_37841_end_0 = const()[name = tensor<string, []>("op_37841_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_37841_end_mask_0 = const()[name = tensor<string, []>("op_37841_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37841_cast = slice_by_index(begin = var_37841_begin_0, end = var_37841_end_0, end_mask = var_37841_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37841_cast")];
            tensor<int32, [4]> var_37845_begin_0 = const()[name = tensor<string, []>("op_37845_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_37845_end_0 = const()[name = tensor<string, []>("op_37845_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_37845_end_mask_0 = const()[name = tensor<string, []>("op_37845_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37845_cast = slice_by_index(begin = var_37845_begin_0, end = var_37845_end_0, end_mask = var_37845_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37845_cast")];
            tensor<int32, [4]> var_37849_begin_0 = const()[name = tensor<string, []>("op_37849_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_37849_end_0 = const()[name = tensor<string, []>("op_37849_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_37849_end_mask_0 = const()[name = tensor<string, []>("op_37849_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37849_cast = slice_by_index(begin = var_37849_begin_0, end = var_37849_end_0, end_mask = var_37849_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37849_cast")];
            tensor<int32, [4]> var_37853_begin_0 = const()[name = tensor<string, []>("op_37853_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_37853_end_0 = const()[name = tensor<string, []>("op_37853_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_37853_end_mask_0 = const()[name = tensor<string, []>("op_37853_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37853_cast = slice_by_index(begin = var_37853_begin_0, end = var_37853_end_0, end_mask = var_37853_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37853_cast")];
            tensor<int32, [4]> var_37857_begin_0 = const()[name = tensor<string, []>("op_37857_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_37857_end_0 = const()[name = tensor<string, []>("op_37857_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_37857_end_mask_0 = const()[name = tensor<string, []>("op_37857_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37857_cast = slice_by_index(begin = var_37857_begin_0, end = var_37857_end_0, end_mask = var_37857_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37857_cast")];
            tensor<int32, [4]> var_37861_begin_0 = const()[name = tensor<string, []>("op_37861_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_37861_end_0 = const()[name = tensor<string, []>("op_37861_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_37861_end_mask_0 = const()[name = tensor<string, []>("op_37861_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37861_cast = slice_by_index(begin = var_37861_begin_0, end = var_37861_end_0, end_mask = var_37861_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37861_cast")];
            tensor<int32, [4]> var_37865_begin_0 = const()[name = tensor<string, []>("op_37865_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_37865_end_0 = const()[name = tensor<string, []>("op_37865_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_37865_end_mask_0 = const()[name = tensor<string, []>("op_37865_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37865_cast = slice_by_index(begin = var_37865_begin_0, end = var_37865_end_0, end_mask = var_37865_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37865_cast")];
            tensor<int32, [4]> var_37869_begin_0 = const()[name = tensor<string, []>("op_37869_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_37869_end_0 = const()[name = tensor<string, []>("op_37869_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_37869_end_mask_0 = const()[name = tensor<string, []>("op_37869_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37869_cast = slice_by_index(begin = var_37869_begin_0, end = var_37869_end_0, end_mask = var_37869_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37869_cast")];
            tensor<int32, [4]> var_37873_begin_0 = const()[name = tensor<string, []>("op_37873_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_37873_end_0 = const()[name = tensor<string, []>("op_37873_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_37873_end_mask_0 = const()[name = tensor<string, []>("op_37873_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37873_cast = slice_by_index(begin = var_37873_begin_0, end = var_37873_end_0, end_mask = var_37873_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37873_cast")];
            tensor<int32, [4]> var_37877_begin_0 = const()[name = tensor<string, []>("op_37877_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_37877_end_0 = const()[name = tensor<string, []>("op_37877_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_37877_end_mask_0 = const()[name = tensor<string, []>("op_37877_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37877_cast = slice_by_index(begin = var_37877_begin_0, end = var_37877_end_0, end_mask = var_37877_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37877_cast")];
            tensor<int32, [4]> var_37881_begin_0 = const()[name = tensor<string, []>("op_37881_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_37881_end_0 = const()[name = tensor<string, []>("op_37881_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_37881_end_mask_0 = const()[name = tensor<string, []>("op_37881_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37881_cast = slice_by_index(begin = var_37881_begin_0, end = var_37881_end_0, end_mask = var_37881_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37881_cast")];
            tensor<int32, [4]> var_37885_begin_0 = const()[name = tensor<string, []>("op_37885_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_37885_end_0 = const()[name = tensor<string, []>("op_37885_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_37885_end_mask_0 = const()[name = tensor<string, []>("op_37885_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37885_cast = slice_by_index(begin = var_37885_begin_0, end = var_37885_end_0, end_mask = var_37885_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37885_cast")];
            tensor<int32, [4]> var_37889_begin_0 = const()[name = tensor<string, []>("op_37889_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_37889_end_0 = const()[name = tensor<string, []>("op_37889_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_37889_end_mask_0 = const()[name = tensor<string, []>("op_37889_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37889_cast = slice_by_index(begin = var_37889_begin_0, end = var_37889_end_0, end_mask = var_37889_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37889_cast")];
            tensor<int32, [4]> var_37893_begin_0 = const()[name = tensor<string, []>("op_37893_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_37893_end_0 = const()[name = tensor<string, []>("op_37893_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_37893_end_mask_0 = const()[name = tensor<string, []>("op_37893_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37893_cast = slice_by_index(begin = var_37893_begin_0, end = var_37893_end_0, end_mask = var_37893_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37893_cast")];
            tensor<int32, [4]> var_37897_begin_0 = const()[name = tensor<string, []>("op_37897_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_37897_end_0 = const()[name = tensor<string, []>("op_37897_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_37897_end_mask_0 = const()[name = tensor<string, []>("op_37897_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37897_cast = slice_by_index(begin = var_37897_begin_0, end = var_37897_end_0, end_mask = var_37897_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37897_cast")];
            tensor<int32, [4]> var_37901_begin_0 = const()[name = tensor<string, []>("op_37901_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_37901_end_0 = const()[name = tensor<string, []>("op_37901_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_37901_end_mask_0 = const()[name = tensor<string, []>("op_37901_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37901_cast = slice_by_index(begin = var_37901_begin_0, end = var_37901_end_0, end_mask = var_37901_end_mask_0, x = q_169_cast)[name = tensor<string, []>("op_37901_cast")];
            tensor<int32, [4]> k_339_perm_0 = const()[name = tensor<string, []>("k_339_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_37908_begin_0 = const()[name = tensor<string, []>("op_37908_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_37908_end_0 = const()[name = tensor<string, []>("op_37908_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_37908_end_mask_0 = const()[name = tensor<string, []>("op_37908_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_55 = transpose(perm = k_339_perm_0, x = k_337_cast)[name = tensor<string, []>("transpose_55")];
            tensor<fp16, [2, 1024, 1, 64]> var_37908_cast = slice_by_index(begin = var_37908_begin_0, end = var_37908_end_0, end_mask = var_37908_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37908_cast")];
            tensor<int32, [4]> var_37912_begin_0 = const()[name = tensor<string, []>("op_37912_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_37912_end_0 = const()[name = tensor<string, []>("op_37912_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_37912_end_mask_0 = const()[name = tensor<string, []>("op_37912_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37912_cast = slice_by_index(begin = var_37912_begin_0, end = var_37912_end_0, end_mask = var_37912_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37912_cast")];
            tensor<int32, [4]> var_37916_begin_0 = const()[name = tensor<string, []>("op_37916_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_37916_end_0 = const()[name = tensor<string, []>("op_37916_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_37916_end_mask_0 = const()[name = tensor<string, []>("op_37916_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37916_cast = slice_by_index(begin = var_37916_begin_0, end = var_37916_end_0, end_mask = var_37916_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37916_cast")];
            tensor<int32, [4]> var_37920_begin_0 = const()[name = tensor<string, []>("op_37920_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_37920_end_0 = const()[name = tensor<string, []>("op_37920_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_37920_end_mask_0 = const()[name = tensor<string, []>("op_37920_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37920_cast = slice_by_index(begin = var_37920_begin_0, end = var_37920_end_0, end_mask = var_37920_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37920_cast")];
            tensor<int32, [4]> var_37924_begin_0 = const()[name = tensor<string, []>("op_37924_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_37924_end_0 = const()[name = tensor<string, []>("op_37924_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_37924_end_mask_0 = const()[name = tensor<string, []>("op_37924_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37924_cast = slice_by_index(begin = var_37924_begin_0, end = var_37924_end_0, end_mask = var_37924_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37924_cast")];
            tensor<int32, [4]> var_37928_begin_0 = const()[name = tensor<string, []>("op_37928_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_37928_end_0 = const()[name = tensor<string, []>("op_37928_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_37928_end_mask_0 = const()[name = tensor<string, []>("op_37928_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37928_cast = slice_by_index(begin = var_37928_begin_0, end = var_37928_end_0, end_mask = var_37928_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37928_cast")];
            tensor<int32, [4]> var_37932_begin_0 = const()[name = tensor<string, []>("op_37932_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_37932_end_0 = const()[name = tensor<string, []>("op_37932_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_37932_end_mask_0 = const()[name = tensor<string, []>("op_37932_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37932_cast = slice_by_index(begin = var_37932_begin_0, end = var_37932_end_0, end_mask = var_37932_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37932_cast")];
            tensor<int32, [4]> var_37936_begin_0 = const()[name = tensor<string, []>("op_37936_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_37936_end_0 = const()[name = tensor<string, []>("op_37936_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_37936_end_mask_0 = const()[name = tensor<string, []>("op_37936_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37936_cast = slice_by_index(begin = var_37936_begin_0, end = var_37936_end_0, end_mask = var_37936_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37936_cast")];
            tensor<int32, [4]> var_37940_begin_0 = const()[name = tensor<string, []>("op_37940_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_37940_end_0 = const()[name = tensor<string, []>("op_37940_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_37940_end_mask_0 = const()[name = tensor<string, []>("op_37940_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37940_cast = slice_by_index(begin = var_37940_begin_0, end = var_37940_end_0, end_mask = var_37940_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37940_cast")];
            tensor<int32, [4]> var_37944_begin_0 = const()[name = tensor<string, []>("op_37944_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_37944_end_0 = const()[name = tensor<string, []>("op_37944_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_37944_end_mask_0 = const()[name = tensor<string, []>("op_37944_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37944_cast = slice_by_index(begin = var_37944_begin_0, end = var_37944_end_0, end_mask = var_37944_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37944_cast")];
            tensor<int32, [4]> var_37948_begin_0 = const()[name = tensor<string, []>("op_37948_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_37948_end_0 = const()[name = tensor<string, []>("op_37948_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_37948_end_mask_0 = const()[name = tensor<string, []>("op_37948_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37948_cast = slice_by_index(begin = var_37948_begin_0, end = var_37948_end_0, end_mask = var_37948_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37948_cast")];
            tensor<int32, [4]> var_37952_begin_0 = const()[name = tensor<string, []>("op_37952_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_37952_end_0 = const()[name = tensor<string, []>("op_37952_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_37952_end_mask_0 = const()[name = tensor<string, []>("op_37952_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37952_cast = slice_by_index(begin = var_37952_begin_0, end = var_37952_end_0, end_mask = var_37952_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37952_cast")];
            tensor<int32, [4]> var_37956_begin_0 = const()[name = tensor<string, []>("op_37956_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_37956_end_0 = const()[name = tensor<string, []>("op_37956_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_37956_end_mask_0 = const()[name = tensor<string, []>("op_37956_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37956_cast = slice_by_index(begin = var_37956_begin_0, end = var_37956_end_0, end_mask = var_37956_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37956_cast")];
            tensor<int32, [4]> var_37960_begin_0 = const()[name = tensor<string, []>("op_37960_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_37960_end_0 = const()[name = tensor<string, []>("op_37960_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_37960_end_mask_0 = const()[name = tensor<string, []>("op_37960_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37960_cast = slice_by_index(begin = var_37960_begin_0, end = var_37960_end_0, end_mask = var_37960_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37960_cast")];
            tensor<int32, [4]> var_37964_begin_0 = const()[name = tensor<string, []>("op_37964_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_37964_end_0 = const()[name = tensor<string, []>("op_37964_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_37964_end_mask_0 = const()[name = tensor<string, []>("op_37964_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37964_cast = slice_by_index(begin = var_37964_begin_0, end = var_37964_end_0, end_mask = var_37964_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37964_cast")];
            tensor<int32, [4]> var_37968_begin_0 = const()[name = tensor<string, []>("op_37968_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_37968_end_0 = const()[name = tensor<string, []>("op_37968_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_37968_end_mask_0 = const()[name = tensor<string, []>("op_37968_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37968_cast = slice_by_index(begin = var_37968_begin_0, end = var_37968_end_0, end_mask = var_37968_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37968_cast")];
            tensor<int32, [4]> var_37972_begin_0 = const()[name = tensor<string, []>("op_37972_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_37972_end_0 = const()[name = tensor<string, []>("op_37972_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_37972_end_mask_0 = const()[name = tensor<string, []>("op_37972_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37972_cast = slice_by_index(begin = var_37972_begin_0, end = var_37972_end_0, end_mask = var_37972_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37972_cast")];
            tensor<int32, [4]> var_37976_begin_0 = const()[name = tensor<string, []>("op_37976_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_37976_end_0 = const()[name = tensor<string, []>("op_37976_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_37976_end_mask_0 = const()[name = tensor<string, []>("op_37976_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37976_cast = slice_by_index(begin = var_37976_begin_0, end = var_37976_end_0, end_mask = var_37976_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37976_cast")];
            tensor<int32, [4]> var_37980_begin_0 = const()[name = tensor<string, []>("op_37980_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_37980_end_0 = const()[name = tensor<string, []>("op_37980_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_37980_end_mask_0 = const()[name = tensor<string, []>("op_37980_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37980_cast = slice_by_index(begin = var_37980_begin_0, end = var_37980_end_0, end_mask = var_37980_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37980_cast")];
            tensor<int32, [4]> var_37984_begin_0 = const()[name = tensor<string, []>("op_37984_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_37984_end_0 = const()[name = tensor<string, []>("op_37984_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_37984_end_mask_0 = const()[name = tensor<string, []>("op_37984_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_37984_cast = slice_by_index(begin = var_37984_begin_0, end = var_37984_end_0, end_mask = var_37984_end_mask_0, x = transpose_55)[name = tensor<string, []>("op_37984_cast")];
            tensor<int32, [4]> var_37986_begin_0 = const()[name = tensor<string, []>("op_37986_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_37986_end_0 = const()[name = tensor<string, []>("op_37986_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_37986_end_mask_0 = const()[name = tensor<string, []>("op_37986_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37986_cast = slice_by_index(begin = var_37986_begin_0, end = var_37986_end_0, end_mask = var_37986_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_37986_cast")];
            tensor<int32, [4]> var_37990_begin_0 = const()[name = tensor<string, []>("op_37990_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_37990_end_0 = const()[name = tensor<string, []>("op_37990_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_37990_end_mask_0 = const()[name = tensor<string, []>("op_37990_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37990_cast = slice_by_index(begin = var_37990_begin_0, end = var_37990_end_0, end_mask = var_37990_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_37990_cast")];
            tensor<int32, [4]> var_37994_begin_0 = const()[name = tensor<string, []>("op_37994_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_37994_end_0 = const()[name = tensor<string, []>("op_37994_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_37994_end_mask_0 = const()[name = tensor<string, []>("op_37994_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37994_cast = slice_by_index(begin = var_37994_begin_0, end = var_37994_end_0, end_mask = var_37994_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_37994_cast")];
            tensor<int32, [4]> var_37998_begin_0 = const()[name = tensor<string, []>("op_37998_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_37998_end_0 = const()[name = tensor<string, []>("op_37998_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_37998_end_mask_0 = const()[name = tensor<string, []>("op_37998_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_37998_cast = slice_by_index(begin = var_37998_begin_0, end = var_37998_end_0, end_mask = var_37998_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_37998_cast")];
            tensor<int32, [4]> var_38002_begin_0 = const()[name = tensor<string, []>("op_38002_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_38002_end_0 = const()[name = tensor<string, []>("op_38002_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_38002_end_mask_0 = const()[name = tensor<string, []>("op_38002_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38002_cast = slice_by_index(begin = var_38002_begin_0, end = var_38002_end_0, end_mask = var_38002_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_38002_cast")];
            tensor<int32, [4]> var_38006_begin_0 = const()[name = tensor<string, []>("op_38006_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_38006_end_0 = const()[name = tensor<string, []>("op_38006_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_38006_end_mask_0 = const()[name = tensor<string, []>("op_38006_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38006_cast = slice_by_index(begin = var_38006_begin_0, end = var_38006_end_0, end_mask = var_38006_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_38006_cast")];
            tensor<int32, [4]> var_38010_begin_0 = const()[name = tensor<string, []>("op_38010_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_38010_end_0 = const()[name = tensor<string, []>("op_38010_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_38010_end_mask_0 = const()[name = tensor<string, []>("op_38010_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38010_cast = slice_by_index(begin = var_38010_begin_0, end = var_38010_end_0, end_mask = var_38010_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_38010_cast")];
            tensor<int32, [4]> var_38014_begin_0 = const()[name = tensor<string, []>("op_38014_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_38014_end_0 = const()[name = tensor<string, []>("op_38014_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_38014_end_mask_0 = const()[name = tensor<string, []>("op_38014_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38014_cast = slice_by_index(begin = var_38014_begin_0, end = var_38014_end_0, end_mask = var_38014_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_38014_cast")];
            tensor<int32, [4]> var_38018_begin_0 = const()[name = tensor<string, []>("op_38018_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_38018_end_0 = const()[name = tensor<string, []>("op_38018_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_38018_end_mask_0 = const()[name = tensor<string, []>("op_38018_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38018_cast = slice_by_index(begin = var_38018_begin_0, end = var_38018_end_0, end_mask = var_38018_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_38018_cast")];
            tensor<int32, [4]> var_38022_begin_0 = const()[name = tensor<string, []>("op_38022_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_38022_end_0 = const()[name = tensor<string, []>("op_38022_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_38022_end_mask_0 = const()[name = tensor<string, []>("op_38022_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38022_cast = slice_by_index(begin = var_38022_begin_0, end = var_38022_end_0, end_mask = var_38022_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_38022_cast")];
            tensor<int32, [4]> var_38026_begin_0 = const()[name = tensor<string, []>("op_38026_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_38026_end_0 = const()[name = tensor<string, []>("op_38026_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_38026_end_mask_0 = const()[name = tensor<string, []>("op_38026_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38026_cast = slice_by_index(begin = var_38026_begin_0, end = var_38026_end_0, end_mask = var_38026_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_38026_cast")];
            tensor<int32, [4]> var_38030_begin_0 = const()[name = tensor<string, []>("op_38030_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_38030_end_0 = const()[name = tensor<string, []>("op_38030_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_38030_end_mask_0 = const()[name = tensor<string, []>("op_38030_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38030_cast = slice_by_index(begin = var_38030_begin_0, end = var_38030_end_0, end_mask = var_38030_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_38030_cast")];
            tensor<int32, [4]> var_38034_begin_0 = const()[name = tensor<string, []>("op_38034_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_38034_end_0 = const()[name = tensor<string, []>("op_38034_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_38034_end_mask_0 = const()[name = tensor<string, []>("op_38034_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38034_cast = slice_by_index(begin = var_38034_begin_0, end = var_38034_end_0, end_mask = var_38034_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_38034_cast")];
            tensor<int32, [4]> var_38038_begin_0 = const()[name = tensor<string, []>("op_38038_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_38038_end_0 = const()[name = tensor<string, []>("op_38038_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_38038_end_mask_0 = const()[name = tensor<string, []>("op_38038_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38038_cast = slice_by_index(begin = var_38038_begin_0, end = var_38038_end_0, end_mask = var_38038_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_38038_cast")];
            tensor<int32, [4]> var_38042_begin_0 = const()[name = tensor<string, []>("op_38042_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_38042_end_0 = const()[name = tensor<string, []>("op_38042_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_38042_end_mask_0 = const()[name = tensor<string, []>("op_38042_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38042_cast = slice_by_index(begin = var_38042_begin_0, end = var_38042_end_0, end_mask = var_38042_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_38042_cast")];
            tensor<int32, [4]> var_38046_begin_0 = const()[name = tensor<string, []>("op_38046_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_38046_end_0 = const()[name = tensor<string, []>("op_38046_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_38046_end_mask_0 = const()[name = tensor<string, []>("op_38046_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38046_cast = slice_by_index(begin = var_38046_begin_0, end = var_38046_end_0, end_mask = var_38046_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_38046_cast")];
            tensor<int32, [4]> var_38050_begin_0 = const()[name = tensor<string, []>("op_38050_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_38050_end_0 = const()[name = tensor<string, []>("op_38050_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_38050_end_mask_0 = const()[name = tensor<string, []>("op_38050_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38050_cast = slice_by_index(begin = var_38050_begin_0, end = var_38050_end_0, end_mask = var_38050_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_38050_cast")];
            tensor<int32, [4]> var_38054_begin_0 = const()[name = tensor<string, []>("op_38054_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_38054_end_0 = const()[name = tensor<string, []>("op_38054_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_38054_end_mask_0 = const()[name = tensor<string, []>("op_38054_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38054_cast = slice_by_index(begin = var_38054_begin_0, end = var_38054_end_0, end_mask = var_38054_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_38054_cast")];
            tensor<int32, [4]> var_38058_begin_0 = const()[name = tensor<string, []>("op_38058_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_38058_end_0 = const()[name = tensor<string, []>("op_38058_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_38058_end_mask_0 = const()[name = tensor<string, []>("op_38058_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38058_cast = slice_by_index(begin = var_38058_begin_0, end = var_38058_end_0, end_mask = var_38058_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_38058_cast")];
            tensor<int32, [4]> var_38062_begin_0 = const()[name = tensor<string, []>("op_38062_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_38062_end_0 = const()[name = tensor<string, []>("op_38062_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_38062_end_mask_0 = const()[name = tensor<string, []>("op_38062_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38062_cast = slice_by_index(begin = var_38062_begin_0, end = var_38062_end_0, end_mask = var_38062_end_mask_0, x = v_169_cast)[name = tensor<string, []>("op_38062_cast")];
            tensor<string, []> var_38066_equation_0 = const()[name = tensor<string, []>("op_38066_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38066_cast = einsum(equation = var_38066_equation_0, values = (var_37908_cast, var_37825_cast))[name = tensor<string, []>("op_38066_cast")];
            tensor<fp16, []> var_38067_to_fp16 = const()[name = tensor<string, []>("op_38067_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3201_cast = mul(x = var_38066_cast, y = var_38067_to_fp16)[name = tensor<string, []>("aw_3201_cast")];
            tensor<string, []> var_38070_equation_0 = const()[name = tensor<string, []>("op_38070_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38070_cast = einsum(equation = var_38070_equation_0, values = (var_37912_cast, var_37829_cast))[name = tensor<string, []>("op_38070_cast")];
            tensor<fp16, []> var_38071_to_fp16 = const()[name = tensor<string, []>("op_38071_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3203_cast = mul(x = var_38070_cast, y = var_38071_to_fp16)[name = tensor<string, []>("aw_3203_cast")];
            tensor<string, []> var_38074_equation_0 = const()[name = tensor<string, []>("op_38074_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38074_cast = einsum(equation = var_38074_equation_0, values = (var_37916_cast, var_37833_cast))[name = tensor<string, []>("op_38074_cast")];
            tensor<fp16, []> var_38075_to_fp16 = const()[name = tensor<string, []>("op_38075_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3205_cast = mul(x = var_38074_cast, y = var_38075_to_fp16)[name = tensor<string, []>("aw_3205_cast")];
            tensor<string, []> var_38078_equation_0 = const()[name = tensor<string, []>("op_38078_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38078_cast = einsum(equation = var_38078_equation_0, values = (var_37920_cast, var_37837_cast))[name = tensor<string, []>("op_38078_cast")];
            tensor<fp16, []> var_38079_to_fp16 = const()[name = tensor<string, []>("op_38079_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3207_cast = mul(x = var_38078_cast, y = var_38079_to_fp16)[name = tensor<string, []>("aw_3207_cast")];
            tensor<string, []> var_38082_equation_0 = const()[name = tensor<string, []>("op_38082_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38082_cast = einsum(equation = var_38082_equation_0, values = (var_37924_cast, var_37841_cast))[name = tensor<string, []>("op_38082_cast")];
            tensor<fp16, []> var_38083_to_fp16 = const()[name = tensor<string, []>("op_38083_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3209_cast = mul(x = var_38082_cast, y = var_38083_to_fp16)[name = tensor<string, []>("aw_3209_cast")];
            tensor<string, []> var_38086_equation_0 = const()[name = tensor<string, []>("op_38086_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38086_cast = einsum(equation = var_38086_equation_0, values = (var_37928_cast, var_37845_cast))[name = tensor<string, []>("op_38086_cast")];
            tensor<fp16, []> var_38087_to_fp16 = const()[name = tensor<string, []>("op_38087_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3211_cast = mul(x = var_38086_cast, y = var_38087_to_fp16)[name = tensor<string, []>("aw_3211_cast")];
            tensor<string, []> var_38090_equation_0 = const()[name = tensor<string, []>("op_38090_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38090_cast = einsum(equation = var_38090_equation_0, values = (var_37932_cast, var_37849_cast))[name = tensor<string, []>("op_38090_cast")];
            tensor<fp16, []> var_38091_to_fp16 = const()[name = tensor<string, []>("op_38091_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3213_cast = mul(x = var_38090_cast, y = var_38091_to_fp16)[name = tensor<string, []>("aw_3213_cast")];
            tensor<string, []> var_38094_equation_0 = const()[name = tensor<string, []>("op_38094_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38094_cast = einsum(equation = var_38094_equation_0, values = (var_37936_cast, var_37853_cast))[name = tensor<string, []>("op_38094_cast")];
            tensor<fp16, []> var_38095_to_fp16 = const()[name = tensor<string, []>("op_38095_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3215_cast = mul(x = var_38094_cast, y = var_38095_to_fp16)[name = tensor<string, []>("aw_3215_cast")];
            tensor<string, []> var_38098_equation_0 = const()[name = tensor<string, []>("op_38098_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38098_cast = einsum(equation = var_38098_equation_0, values = (var_37940_cast, var_37857_cast))[name = tensor<string, []>("op_38098_cast")];
            tensor<fp16, []> var_38099_to_fp16 = const()[name = tensor<string, []>("op_38099_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3217_cast = mul(x = var_38098_cast, y = var_38099_to_fp16)[name = tensor<string, []>("aw_3217_cast")];
            tensor<string, []> var_38102_equation_0 = const()[name = tensor<string, []>("op_38102_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38102_cast = einsum(equation = var_38102_equation_0, values = (var_37944_cast, var_37861_cast))[name = tensor<string, []>("op_38102_cast")];
            tensor<fp16, []> var_38103_to_fp16 = const()[name = tensor<string, []>("op_38103_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3219_cast = mul(x = var_38102_cast, y = var_38103_to_fp16)[name = tensor<string, []>("aw_3219_cast")];
            tensor<string, []> var_38106_equation_0 = const()[name = tensor<string, []>("op_38106_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38106_cast = einsum(equation = var_38106_equation_0, values = (var_37948_cast, var_37865_cast))[name = tensor<string, []>("op_38106_cast")];
            tensor<fp16, []> var_38107_to_fp16 = const()[name = tensor<string, []>("op_38107_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3221_cast = mul(x = var_38106_cast, y = var_38107_to_fp16)[name = tensor<string, []>("aw_3221_cast")];
            tensor<string, []> var_38110_equation_0 = const()[name = tensor<string, []>("op_38110_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38110_cast = einsum(equation = var_38110_equation_0, values = (var_37952_cast, var_37869_cast))[name = tensor<string, []>("op_38110_cast")];
            tensor<fp16, []> var_38111_to_fp16 = const()[name = tensor<string, []>("op_38111_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3223_cast = mul(x = var_38110_cast, y = var_38111_to_fp16)[name = tensor<string, []>("aw_3223_cast")];
            tensor<string, []> var_38114_equation_0 = const()[name = tensor<string, []>("op_38114_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38114_cast = einsum(equation = var_38114_equation_0, values = (var_37956_cast, var_37873_cast))[name = tensor<string, []>("op_38114_cast")];
            tensor<fp16, []> var_38115_to_fp16 = const()[name = tensor<string, []>("op_38115_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3225_cast = mul(x = var_38114_cast, y = var_38115_to_fp16)[name = tensor<string, []>("aw_3225_cast")];
            tensor<string, []> var_38118_equation_0 = const()[name = tensor<string, []>("op_38118_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38118_cast = einsum(equation = var_38118_equation_0, values = (var_37960_cast, var_37877_cast))[name = tensor<string, []>("op_38118_cast")];
            tensor<fp16, []> var_38119_to_fp16 = const()[name = tensor<string, []>("op_38119_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3227_cast = mul(x = var_38118_cast, y = var_38119_to_fp16)[name = tensor<string, []>("aw_3227_cast")];
            tensor<string, []> var_38122_equation_0 = const()[name = tensor<string, []>("op_38122_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38122_cast = einsum(equation = var_38122_equation_0, values = (var_37964_cast, var_37881_cast))[name = tensor<string, []>("op_38122_cast")];
            tensor<fp16, []> var_38123_to_fp16 = const()[name = tensor<string, []>("op_38123_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3229_cast = mul(x = var_38122_cast, y = var_38123_to_fp16)[name = tensor<string, []>("aw_3229_cast")];
            tensor<string, []> var_38126_equation_0 = const()[name = tensor<string, []>("op_38126_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38126_cast = einsum(equation = var_38126_equation_0, values = (var_37968_cast, var_37885_cast))[name = tensor<string, []>("op_38126_cast")];
            tensor<fp16, []> var_38127_to_fp16 = const()[name = tensor<string, []>("op_38127_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3231_cast = mul(x = var_38126_cast, y = var_38127_to_fp16)[name = tensor<string, []>("aw_3231_cast")];
            tensor<string, []> var_38130_equation_0 = const()[name = tensor<string, []>("op_38130_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38130_cast = einsum(equation = var_38130_equation_0, values = (var_37972_cast, var_37889_cast))[name = tensor<string, []>("op_38130_cast")];
            tensor<fp16, []> var_38131_to_fp16 = const()[name = tensor<string, []>("op_38131_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3233_cast = mul(x = var_38130_cast, y = var_38131_to_fp16)[name = tensor<string, []>("aw_3233_cast")];
            tensor<string, []> var_38134_equation_0 = const()[name = tensor<string, []>("op_38134_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38134_cast = einsum(equation = var_38134_equation_0, values = (var_37976_cast, var_37893_cast))[name = tensor<string, []>("op_38134_cast")];
            tensor<fp16, []> var_38135_to_fp16 = const()[name = tensor<string, []>("op_38135_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3235_cast = mul(x = var_38134_cast, y = var_38135_to_fp16)[name = tensor<string, []>("aw_3235_cast")];
            tensor<string, []> var_38138_equation_0 = const()[name = tensor<string, []>("op_38138_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38138_cast = einsum(equation = var_38138_equation_0, values = (var_37980_cast, var_37897_cast))[name = tensor<string, []>("op_38138_cast")];
            tensor<fp16, []> var_38139_to_fp16 = const()[name = tensor<string, []>("op_38139_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3237_cast = mul(x = var_38138_cast, y = var_38139_to_fp16)[name = tensor<string, []>("aw_3237_cast")];
            tensor<string, []> var_38142_equation_0 = const()[name = tensor<string, []>("op_38142_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38142_cast = einsum(equation = var_38142_equation_0, values = (var_37984_cast, var_37901_cast))[name = tensor<string, []>("op_38142_cast")];
            tensor<fp16, []> var_38143_to_fp16 = const()[name = tensor<string, []>("op_38143_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3239_cast = mul(x = var_38142_cast, y = var_38143_to_fp16)[name = tensor<string, []>("aw_3239_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38145_cast = softmax(axis = var_30385, x = aw_3201_cast)[name = tensor<string, []>("op_38145_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38146_cast = softmax(axis = var_30385, x = aw_3203_cast)[name = tensor<string, []>("op_38146_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38147_cast = softmax(axis = var_30385, x = aw_3205_cast)[name = tensor<string, []>("op_38147_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38148_cast = softmax(axis = var_30385, x = aw_3207_cast)[name = tensor<string, []>("op_38148_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38149_cast = softmax(axis = var_30385, x = aw_3209_cast)[name = tensor<string, []>("op_38149_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38150_cast = softmax(axis = var_30385, x = aw_3211_cast)[name = tensor<string, []>("op_38150_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38151_cast = softmax(axis = var_30385, x = aw_3213_cast)[name = tensor<string, []>("op_38151_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38152_cast = softmax(axis = var_30385, x = aw_3215_cast)[name = tensor<string, []>("op_38152_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38153_cast = softmax(axis = var_30385, x = aw_3217_cast)[name = tensor<string, []>("op_38153_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38154_cast = softmax(axis = var_30385, x = aw_3219_cast)[name = tensor<string, []>("op_38154_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38155_cast = softmax(axis = var_30385, x = aw_3221_cast)[name = tensor<string, []>("op_38155_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38156_cast = softmax(axis = var_30385, x = aw_3223_cast)[name = tensor<string, []>("op_38156_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38157_cast = softmax(axis = var_30385, x = aw_3225_cast)[name = tensor<string, []>("op_38157_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38158_cast = softmax(axis = var_30385, x = aw_3227_cast)[name = tensor<string, []>("op_38158_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38159_cast = softmax(axis = var_30385, x = aw_3229_cast)[name = tensor<string, []>("op_38159_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38160_cast = softmax(axis = var_30385, x = aw_3231_cast)[name = tensor<string, []>("op_38160_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38161_cast = softmax(axis = var_30385, x = aw_3233_cast)[name = tensor<string, []>("op_38161_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38162_cast = softmax(axis = var_30385, x = aw_3235_cast)[name = tensor<string, []>("op_38162_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38163_cast = softmax(axis = var_30385, x = aw_3237_cast)[name = tensor<string, []>("op_38163_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38164_cast = softmax(axis = var_30385, x = aw_3239_cast)[name = tensor<string, []>("op_38164_cast")];
            tensor<string, []> var_38166_equation_0 = const()[name = tensor<string, []>("op_38166_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38166_cast = einsum(equation = var_38166_equation_0, values = (var_37986_cast, var_38145_cast))[name = tensor<string, []>("op_38166_cast")];
            tensor<string, []> var_38168_equation_0 = const()[name = tensor<string, []>("op_38168_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38168_cast = einsum(equation = var_38168_equation_0, values = (var_37990_cast, var_38146_cast))[name = tensor<string, []>("op_38168_cast")];
            tensor<string, []> var_38170_equation_0 = const()[name = tensor<string, []>("op_38170_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38170_cast = einsum(equation = var_38170_equation_0, values = (var_37994_cast, var_38147_cast))[name = tensor<string, []>("op_38170_cast")];
            tensor<string, []> var_38172_equation_0 = const()[name = tensor<string, []>("op_38172_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38172_cast = einsum(equation = var_38172_equation_0, values = (var_37998_cast, var_38148_cast))[name = tensor<string, []>("op_38172_cast")];
            tensor<string, []> var_38174_equation_0 = const()[name = tensor<string, []>("op_38174_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38174_cast = einsum(equation = var_38174_equation_0, values = (var_38002_cast, var_38149_cast))[name = tensor<string, []>("op_38174_cast")];
            tensor<string, []> var_38176_equation_0 = const()[name = tensor<string, []>("op_38176_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38176_cast = einsum(equation = var_38176_equation_0, values = (var_38006_cast, var_38150_cast))[name = tensor<string, []>("op_38176_cast")];
            tensor<string, []> var_38178_equation_0 = const()[name = tensor<string, []>("op_38178_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38178_cast = einsum(equation = var_38178_equation_0, values = (var_38010_cast, var_38151_cast))[name = tensor<string, []>("op_38178_cast")];
            tensor<string, []> var_38180_equation_0 = const()[name = tensor<string, []>("op_38180_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38180_cast = einsum(equation = var_38180_equation_0, values = (var_38014_cast, var_38152_cast))[name = tensor<string, []>("op_38180_cast")];
            tensor<string, []> var_38182_equation_0 = const()[name = tensor<string, []>("op_38182_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38182_cast = einsum(equation = var_38182_equation_0, values = (var_38018_cast, var_38153_cast))[name = tensor<string, []>("op_38182_cast")];
            tensor<string, []> var_38184_equation_0 = const()[name = tensor<string, []>("op_38184_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38184_cast = einsum(equation = var_38184_equation_0, values = (var_38022_cast, var_38154_cast))[name = tensor<string, []>("op_38184_cast")];
            tensor<string, []> var_38186_equation_0 = const()[name = tensor<string, []>("op_38186_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38186_cast = einsum(equation = var_38186_equation_0, values = (var_38026_cast, var_38155_cast))[name = tensor<string, []>("op_38186_cast")];
            tensor<string, []> var_38188_equation_0 = const()[name = tensor<string, []>("op_38188_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38188_cast = einsum(equation = var_38188_equation_0, values = (var_38030_cast, var_38156_cast))[name = tensor<string, []>("op_38188_cast")];
            tensor<string, []> var_38190_equation_0 = const()[name = tensor<string, []>("op_38190_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38190_cast = einsum(equation = var_38190_equation_0, values = (var_38034_cast, var_38157_cast))[name = tensor<string, []>("op_38190_cast")];
            tensor<string, []> var_38192_equation_0 = const()[name = tensor<string, []>("op_38192_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38192_cast = einsum(equation = var_38192_equation_0, values = (var_38038_cast, var_38158_cast))[name = tensor<string, []>("op_38192_cast")];
            tensor<string, []> var_38194_equation_0 = const()[name = tensor<string, []>("op_38194_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38194_cast = einsum(equation = var_38194_equation_0, values = (var_38042_cast, var_38159_cast))[name = tensor<string, []>("op_38194_cast")];
            tensor<string, []> var_38196_equation_0 = const()[name = tensor<string, []>("op_38196_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38196_cast = einsum(equation = var_38196_equation_0, values = (var_38046_cast, var_38160_cast))[name = tensor<string, []>("op_38196_cast")];
            tensor<string, []> var_38198_equation_0 = const()[name = tensor<string, []>("op_38198_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38198_cast = einsum(equation = var_38198_equation_0, values = (var_38050_cast, var_38161_cast))[name = tensor<string, []>("op_38198_cast")];
            tensor<string, []> var_38200_equation_0 = const()[name = tensor<string, []>("op_38200_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38200_cast = einsum(equation = var_38200_equation_0, values = (var_38054_cast, var_38162_cast))[name = tensor<string, []>("op_38200_cast")];
            tensor<string, []> var_38202_equation_0 = const()[name = tensor<string, []>("op_38202_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38202_cast = einsum(equation = var_38202_equation_0, values = (var_38058_cast, var_38163_cast))[name = tensor<string, []>("op_38202_cast")];
            tensor<string, []> var_38204_equation_0 = const()[name = tensor<string, []>("op_38204_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38204_cast = einsum(equation = var_38204_equation_0, values = (var_38062_cast, var_38164_cast))[name = tensor<string, []>("op_38204_cast")];
            tensor<bool, []> input_505_interleave_0 = const()[name = tensor<string, []>("input_505_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_505_cast = concat(axis = var_30385, interleave = input_505_interleave_0, values = (var_38166_cast, var_38168_cast, var_38170_cast, var_38172_cast, var_38174_cast, var_38176_cast, var_38178_cast, var_38180_cast, var_38182_cast, var_38184_cast, var_38186_cast, var_38188_cast, var_38190_cast, var_38192_cast, var_38194_cast, var_38196_cast, var_38198_cast, var_38200_cast, var_38202_cast, var_38204_cast))[name = tensor<string, []>("input_505_cast")];
            tensor<int32, [2]> var_38210 = const()[name = tensor<string, []>("op_38210"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_38212 = const()[name = tensor<string, []>("op_38212"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_38214_pad_type_0 = const()[name = tensor<string, []>("op_38214_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_38214_pad_0 = const()[name = tensor<string, []>("op_38214_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3154360896)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3157637760)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_38214_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_out_0_bias_to_fp16, dilations = var_38212, groups = var_30385, pad = var_38214_pad_0, pad_type = var_38214_pad_type_0, strides = var_38210, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn1_to_out_0_weight_to_fp16, x = input_505_cast)[name = tensor<string, []>("op_38214_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_255_cast = add(x = var_38214_cast, y = inputs_253_cast)[name = tensor<string, []>("inputs_255_cast")];
            tensor<int32, [1]> var_38218 = const()[name = tensor<string, []>("op_38218"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_255_cast = reduce_mean(axes = var_38218, keep_dims = var_30380, x = inputs_255_cast)[name = tensor<string, []>("channels_mean_255_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_255_cast = sub(x = inputs_255_cast, y = channels_mean_255_cast)[name = tensor<string, []>("zero_mean_255_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_255_cast = mul(x = zero_mean_255_cast, y = zero_mean_255_cast)[name = tensor<string, []>("zero_mean_sq_255_cast")];
            tensor<int32, [1]> var_38222 = const()[name = tensor<string, []>("op_38222"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_38223_cast = reduce_mean(axes = var_38222, keep_dims = var_30380, x = zero_mean_sq_255_cast)[name = tensor<string, []>("op_38223_cast")];
            tensor<fp16, []> var_38224_to_fp16 = const()[name = tensor<string, []>("op_38224_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_38225_cast = add(x = var_38223_cast, y = var_38224_to_fp16)[name = tensor<string, []>("op_38225_cast")];
            tensor<fp16, []> denom_255_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_255_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_255_cast = rsqrt(epsilon = denom_255_epsilon_0_to_fp16, x = var_38225_cast)[name = tensor<string, []>("denom_255_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_255_cast = mul(x = zero_mean_255_cast, y = denom_255_cast)[name = tensor<string, []>("out_255_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_38229_to_fp16 = const()[name = tensor<string, []>("op_38229_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3157640384)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_38230_cast = add(x = out_255_cast, y = var_38229_to_fp16)[name = tensor<string, []>("op_38230_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_38232_to_fp16 = const()[name = tensor<string, []>("op_38232_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3157643008)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_337_cast = mul(x = var_38230_cast, y = var_38232_to_fp16)[name = tensor<string, []>("hidden_states_337_cast")];
            tensor<int32, [2]> var_38239 = const()[name = tensor<string, []>("op_38239"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_38241 = const()[name = tensor<string, []>("op_38241"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_171_pad_type_0 = const()[name = tensor<string, []>("q_171_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_171_pad_0 = const()[name = tensor<string, []>("q_171_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3157645632)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_171_cast = conv(dilations = var_38241, groups = var_30385, pad = q_171_pad_0, pad_type = q_171_pad_type_0, strides = var_38239, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_q_weight_to_fp16, x = hidden_states_337_cast)[name = tensor<string, []>("q_171_cast")];
            tensor<int32, [2]> var_38245 = const()[name = tensor<string, []>("op_38245"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_38247 = const()[name = tensor<string, []>("op_38247"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_341_pad_type_0 = const()[name = tensor<string, []>("k_341_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_341_pad_0 = const()[name = tensor<string, []>("k_341_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3160922496)))];
            tensor<fp16, [2, 1280, 1, 77]> k_341_cast = conv(dilations = var_38247, groups = var_30385, pad = k_341_pad_0, pad_type = k_341_pad_type_0, strides = var_38245, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_341_cast")];
            tensor<int32, [2]> var_38251 = const()[name = tensor<string, []>("op_38251"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_38253 = const()[name = tensor<string, []>("op_38253"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_171_pad_type_0 = const()[name = tensor<string, []>("v_171_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_171_pad_0 = const()[name = tensor<string, []>("v_171_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3166165440)))];
            tensor<fp16, [2, 1280, 1, 77]> v_171_cast = conv(dilations = var_38253, groups = var_30385, pad = v_171_pad_0, pad_type = v_171_pad_type_0, strides = var_38251, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_171_cast")];
            tensor<int32, [4]> var_38257_begin_0 = const()[name = tensor<string, []>("op_38257_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_38257_end_0 = const()[name = tensor<string, []>("op_38257_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_38257_end_mask_0 = const()[name = tensor<string, []>("op_38257_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38257_cast = slice_by_index(begin = var_38257_begin_0, end = var_38257_end_0, end_mask = var_38257_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38257_cast")];
            tensor<int32, [4]> var_38261_begin_0 = const()[name = tensor<string, []>("op_38261_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_38261_end_0 = const()[name = tensor<string, []>("op_38261_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_38261_end_mask_0 = const()[name = tensor<string, []>("op_38261_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38261_cast = slice_by_index(begin = var_38261_begin_0, end = var_38261_end_0, end_mask = var_38261_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38261_cast")];
            tensor<int32, [4]> var_38265_begin_0 = const()[name = tensor<string, []>("op_38265_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_38265_end_0 = const()[name = tensor<string, []>("op_38265_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_38265_end_mask_0 = const()[name = tensor<string, []>("op_38265_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38265_cast = slice_by_index(begin = var_38265_begin_0, end = var_38265_end_0, end_mask = var_38265_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38265_cast")];
            tensor<int32, [4]> var_38269_begin_0 = const()[name = tensor<string, []>("op_38269_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_38269_end_0 = const()[name = tensor<string, []>("op_38269_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_38269_end_mask_0 = const()[name = tensor<string, []>("op_38269_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38269_cast = slice_by_index(begin = var_38269_begin_0, end = var_38269_end_0, end_mask = var_38269_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38269_cast")];
            tensor<int32, [4]> var_38273_begin_0 = const()[name = tensor<string, []>("op_38273_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_38273_end_0 = const()[name = tensor<string, []>("op_38273_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_38273_end_mask_0 = const()[name = tensor<string, []>("op_38273_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38273_cast = slice_by_index(begin = var_38273_begin_0, end = var_38273_end_0, end_mask = var_38273_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38273_cast")];
            tensor<int32, [4]> var_38277_begin_0 = const()[name = tensor<string, []>("op_38277_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_38277_end_0 = const()[name = tensor<string, []>("op_38277_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_38277_end_mask_0 = const()[name = tensor<string, []>("op_38277_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38277_cast = slice_by_index(begin = var_38277_begin_0, end = var_38277_end_0, end_mask = var_38277_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38277_cast")];
            tensor<int32, [4]> var_38281_begin_0 = const()[name = tensor<string, []>("op_38281_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_38281_end_0 = const()[name = tensor<string, []>("op_38281_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_38281_end_mask_0 = const()[name = tensor<string, []>("op_38281_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38281_cast = slice_by_index(begin = var_38281_begin_0, end = var_38281_end_0, end_mask = var_38281_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38281_cast")];
            tensor<int32, [4]> var_38285_begin_0 = const()[name = tensor<string, []>("op_38285_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_38285_end_0 = const()[name = tensor<string, []>("op_38285_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_38285_end_mask_0 = const()[name = tensor<string, []>("op_38285_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38285_cast = slice_by_index(begin = var_38285_begin_0, end = var_38285_end_0, end_mask = var_38285_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38285_cast")];
            tensor<int32, [4]> var_38289_begin_0 = const()[name = tensor<string, []>("op_38289_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_38289_end_0 = const()[name = tensor<string, []>("op_38289_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_38289_end_mask_0 = const()[name = tensor<string, []>("op_38289_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38289_cast = slice_by_index(begin = var_38289_begin_0, end = var_38289_end_0, end_mask = var_38289_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38289_cast")];
            tensor<int32, [4]> var_38293_begin_0 = const()[name = tensor<string, []>("op_38293_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_38293_end_0 = const()[name = tensor<string, []>("op_38293_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_38293_end_mask_0 = const()[name = tensor<string, []>("op_38293_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38293_cast = slice_by_index(begin = var_38293_begin_0, end = var_38293_end_0, end_mask = var_38293_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38293_cast")];
            tensor<int32, [4]> var_38297_begin_0 = const()[name = tensor<string, []>("op_38297_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_38297_end_0 = const()[name = tensor<string, []>("op_38297_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_38297_end_mask_0 = const()[name = tensor<string, []>("op_38297_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38297_cast = slice_by_index(begin = var_38297_begin_0, end = var_38297_end_0, end_mask = var_38297_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38297_cast")];
            tensor<int32, [4]> var_38301_begin_0 = const()[name = tensor<string, []>("op_38301_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_38301_end_0 = const()[name = tensor<string, []>("op_38301_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_38301_end_mask_0 = const()[name = tensor<string, []>("op_38301_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38301_cast = slice_by_index(begin = var_38301_begin_0, end = var_38301_end_0, end_mask = var_38301_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38301_cast")];
            tensor<int32, [4]> var_38305_begin_0 = const()[name = tensor<string, []>("op_38305_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_38305_end_0 = const()[name = tensor<string, []>("op_38305_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_38305_end_mask_0 = const()[name = tensor<string, []>("op_38305_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38305_cast = slice_by_index(begin = var_38305_begin_0, end = var_38305_end_0, end_mask = var_38305_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38305_cast")];
            tensor<int32, [4]> var_38309_begin_0 = const()[name = tensor<string, []>("op_38309_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_38309_end_0 = const()[name = tensor<string, []>("op_38309_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_38309_end_mask_0 = const()[name = tensor<string, []>("op_38309_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38309_cast = slice_by_index(begin = var_38309_begin_0, end = var_38309_end_0, end_mask = var_38309_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38309_cast")];
            tensor<int32, [4]> var_38313_begin_0 = const()[name = tensor<string, []>("op_38313_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_38313_end_0 = const()[name = tensor<string, []>("op_38313_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_38313_end_mask_0 = const()[name = tensor<string, []>("op_38313_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38313_cast = slice_by_index(begin = var_38313_begin_0, end = var_38313_end_0, end_mask = var_38313_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38313_cast")];
            tensor<int32, [4]> var_38317_begin_0 = const()[name = tensor<string, []>("op_38317_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_38317_end_0 = const()[name = tensor<string, []>("op_38317_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_38317_end_mask_0 = const()[name = tensor<string, []>("op_38317_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38317_cast = slice_by_index(begin = var_38317_begin_0, end = var_38317_end_0, end_mask = var_38317_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38317_cast")];
            tensor<int32, [4]> var_38321_begin_0 = const()[name = tensor<string, []>("op_38321_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_38321_end_0 = const()[name = tensor<string, []>("op_38321_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_38321_end_mask_0 = const()[name = tensor<string, []>("op_38321_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38321_cast = slice_by_index(begin = var_38321_begin_0, end = var_38321_end_0, end_mask = var_38321_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38321_cast")];
            tensor<int32, [4]> var_38325_begin_0 = const()[name = tensor<string, []>("op_38325_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_38325_end_0 = const()[name = tensor<string, []>("op_38325_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_38325_end_mask_0 = const()[name = tensor<string, []>("op_38325_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38325_cast = slice_by_index(begin = var_38325_begin_0, end = var_38325_end_0, end_mask = var_38325_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38325_cast")];
            tensor<int32, [4]> var_38329_begin_0 = const()[name = tensor<string, []>("op_38329_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_38329_end_0 = const()[name = tensor<string, []>("op_38329_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_38329_end_mask_0 = const()[name = tensor<string, []>("op_38329_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38329_cast = slice_by_index(begin = var_38329_begin_0, end = var_38329_end_0, end_mask = var_38329_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38329_cast")];
            tensor<int32, [4]> var_38333_begin_0 = const()[name = tensor<string, []>("op_38333_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_38333_end_0 = const()[name = tensor<string, []>("op_38333_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_38333_end_mask_0 = const()[name = tensor<string, []>("op_38333_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38333_cast = slice_by_index(begin = var_38333_begin_0, end = var_38333_end_0, end_mask = var_38333_end_mask_0, x = q_171_cast)[name = tensor<string, []>("op_38333_cast")];
            tensor<int32, [4]> k_343_perm_0 = const()[name = tensor<string, []>("k_343_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_38340_begin_0 = const()[name = tensor<string, []>("op_38340_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_38340_end_0 = const()[name = tensor<string, []>("op_38340_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_38340_end_mask_0 = const()[name = tensor<string, []>("op_38340_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_54 = transpose(perm = k_343_perm_0, x = k_341_cast)[name = tensor<string, []>("transpose_54")];
            tensor<fp16, [2, 77, 1, 64]> var_38340_cast = slice_by_index(begin = var_38340_begin_0, end = var_38340_end_0, end_mask = var_38340_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38340_cast")];
            tensor<int32, [4]> var_38344_begin_0 = const()[name = tensor<string, []>("op_38344_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_38344_end_0 = const()[name = tensor<string, []>("op_38344_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_38344_end_mask_0 = const()[name = tensor<string, []>("op_38344_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38344_cast = slice_by_index(begin = var_38344_begin_0, end = var_38344_end_0, end_mask = var_38344_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38344_cast")];
            tensor<int32, [4]> var_38348_begin_0 = const()[name = tensor<string, []>("op_38348_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_38348_end_0 = const()[name = tensor<string, []>("op_38348_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_38348_end_mask_0 = const()[name = tensor<string, []>("op_38348_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38348_cast = slice_by_index(begin = var_38348_begin_0, end = var_38348_end_0, end_mask = var_38348_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38348_cast")];
            tensor<int32, [4]> var_38352_begin_0 = const()[name = tensor<string, []>("op_38352_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_38352_end_0 = const()[name = tensor<string, []>("op_38352_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_38352_end_mask_0 = const()[name = tensor<string, []>("op_38352_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38352_cast = slice_by_index(begin = var_38352_begin_0, end = var_38352_end_0, end_mask = var_38352_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38352_cast")];
            tensor<int32, [4]> var_38356_begin_0 = const()[name = tensor<string, []>("op_38356_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_38356_end_0 = const()[name = tensor<string, []>("op_38356_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_38356_end_mask_0 = const()[name = tensor<string, []>("op_38356_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38356_cast = slice_by_index(begin = var_38356_begin_0, end = var_38356_end_0, end_mask = var_38356_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38356_cast")];
            tensor<int32, [4]> var_38360_begin_0 = const()[name = tensor<string, []>("op_38360_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_38360_end_0 = const()[name = tensor<string, []>("op_38360_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_38360_end_mask_0 = const()[name = tensor<string, []>("op_38360_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38360_cast = slice_by_index(begin = var_38360_begin_0, end = var_38360_end_0, end_mask = var_38360_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38360_cast")];
            tensor<int32, [4]> var_38364_begin_0 = const()[name = tensor<string, []>("op_38364_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_38364_end_0 = const()[name = tensor<string, []>("op_38364_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_38364_end_mask_0 = const()[name = tensor<string, []>("op_38364_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38364_cast = slice_by_index(begin = var_38364_begin_0, end = var_38364_end_0, end_mask = var_38364_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38364_cast")];
            tensor<int32, [4]> var_38368_begin_0 = const()[name = tensor<string, []>("op_38368_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_38368_end_0 = const()[name = tensor<string, []>("op_38368_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_38368_end_mask_0 = const()[name = tensor<string, []>("op_38368_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38368_cast = slice_by_index(begin = var_38368_begin_0, end = var_38368_end_0, end_mask = var_38368_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38368_cast")];
            tensor<int32, [4]> var_38372_begin_0 = const()[name = tensor<string, []>("op_38372_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_38372_end_0 = const()[name = tensor<string, []>("op_38372_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_38372_end_mask_0 = const()[name = tensor<string, []>("op_38372_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38372_cast = slice_by_index(begin = var_38372_begin_0, end = var_38372_end_0, end_mask = var_38372_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38372_cast")];
            tensor<int32, [4]> var_38376_begin_0 = const()[name = tensor<string, []>("op_38376_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_38376_end_0 = const()[name = tensor<string, []>("op_38376_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_38376_end_mask_0 = const()[name = tensor<string, []>("op_38376_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38376_cast = slice_by_index(begin = var_38376_begin_0, end = var_38376_end_0, end_mask = var_38376_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38376_cast")];
            tensor<int32, [4]> var_38380_begin_0 = const()[name = tensor<string, []>("op_38380_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_38380_end_0 = const()[name = tensor<string, []>("op_38380_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_38380_end_mask_0 = const()[name = tensor<string, []>("op_38380_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38380_cast = slice_by_index(begin = var_38380_begin_0, end = var_38380_end_0, end_mask = var_38380_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38380_cast")];
            tensor<int32, [4]> var_38384_begin_0 = const()[name = tensor<string, []>("op_38384_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_38384_end_0 = const()[name = tensor<string, []>("op_38384_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_38384_end_mask_0 = const()[name = tensor<string, []>("op_38384_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38384_cast = slice_by_index(begin = var_38384_begin_0, end = var_38384_end_0, end_mask = var_38384_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38384_cast")];
            tensor<int32, [4]> var_38388_begin_0 = const()[name = tensor<string, []>("op_38388_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_38388_end_0 = const()[name = tensor<string, []>("op_38388_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_38388_end_mask_0 = const()[name = tensor<string, []>("op_38388_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38388_cast = slice_by_index(begin = var_38388_begin_0, end = var_38388_end_0, end_mask = var_38388_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38388_cast")];
            tensor<int32, [4]> var_38392_begin_0 = const()[name = tensor<string, []>("op_38392_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_38392_end_0 = const()[name = tensor<string, []>("op_38392_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_38392_end_mask_0 = const()[name = tensor<string, []>("op_38392_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38392_cast = slice_by_index(begin = var_38392_begin_0, end = var_38392_end_0, end_mask = var_38392_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38392_cast")];
            tensor<int32, [4]> var_38396_begin_0 = const()[name = tensor<string, []>("op_38396_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_38396_end_0 = const()[name = tensor<string, []>("op_38396_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_38396_end_mask_0 = const()[name = tensor<string, []>("op_38396_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38396_cast = slice_by_index(begin = var_38396_begin_0, end = var_38396_end_0, end_mask = var_38396_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38396_cast")];
            tensor<int32, [4]> var_38400_begin_0 = const()[name = tensor<string, []>("op_38400_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_38400_end_0 = const()[name = tensor<string, []>("op_38400_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_38400_end_mask_0 = const()[name = tensor<string, []>("op_38400_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38400_cast = slice_by_index(begin = var_38400_begin_0, end = var_38400_end_0, end_mask = var_38400_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38400_cast")];
            tensor<int32, [4]> var_38404_begin_0 = const()[name = tensor<string, []>("op_38404_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_38404_end_0 = const()[name = tensor<string, []>("op_38404_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_38404_end_mask_0 = const()[name = tensor<string, []>("op_38404_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38404_cast = slice_by_index(begin = var_38404_begin_0, end = var_38404_end_0, end_mask = var_38404_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38404_cast")];
            tensor<int32, [4]> var_38408_begin_0 = const()[name = tensor<string, []>("op_38408_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_38408_end_0 = const()[name = tensor<string, []>("op_38408_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_38408_end_mask_0 = const()[name = tensor<string, []>("op_38408_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38408_cast = slice_by_index(begin = var_38408_begin_0, end = var_38408_end_0, end_mask = var_38408_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38408_cast")];
            tensor<int32, [4]> var_38412_begin_0 = const()[name = tensor<string, []>("op_38412_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_38412_end_0 = const()[name = tensor<string, []>("op_38412_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_38412_end_mask_0 = const()[name = tensor<string, []>("op_38412_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38412_cast = slice_by_index(begin = var_38412_begin_0, end = var_38412_end_0, end_mask = var_38412_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38412_cast")];
            tensor<int32, [4]> var_38416_begin_0 = const()[name = tensor<string, []>("op_38416_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_38416_end_0 = const()[name = tensor<string, []>("op_38416_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_38416_end_mask_0 = const()[name = tensor<string, []>("op_38416_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_38416_cast = slice_by_index(begin = var_38416_begin_0, end = var_38416_end_0, end_mask = var_38416_end_mask_0, x = transpose_54)[name = tensor<string, []>("op_38416_cast")];
            tensor<int32, [4]> var_38418_begin_0 = const()[name = tensor<string, []>("op_38418_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_38418_end_0 = const()[name = tensor<string, []>("op_38418_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_38418_end_mask_0 = const()[name = tensor<string, []>("op_38418_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38418_cast = slice_by_index(begin = var_38418_begin_0, end = var_38418_end_0, end_mask = var_38418_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38418_cast")];
            tensor<int32, [4]> var_38422_begin_0 = const()[name = tensor<string, []>("op_38422_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_38422_end_0 = const()[name = tensor<string, []>("op_38422_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_38422_end_mask_0 = const()[name = tensor<string, []>("op_38422_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38422_cast = slice_by_index(begin = var_38422_begin_0, end = var_38422_end_0, end_mask = var_38422_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38422_cast")];
            tensor<int32, [4]> var_38426_begin_0 = const()[name = tensor<string, []>("op_38426_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_38426_end_0 = const()[name = tensor<string, []>("op_38426_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_38426_end_mask_0 = const()[name = tensor<string, []>("op_38426_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38426_cast = slice_by_index(begin = var_38426_begin_0, end = var_38426_end_0, end_mask = var_38426_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38426_cast")];
            tensor<int32, [4]> var_38430_begin_0 = const()[name = tensor<string, []>("op_38430_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_38430_end_0 = const()[name = tensor<string, []>("op_38430_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_38430_end_mask_0 = const()[name = tensor<string, []>("op_38430_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38430_cast = slice_by_index(begin = var_38430_begin_0, end = var_38430_end_0, end_mask = var_38430_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38430_cast")];
            tensor<int32, [4]> var_38434_begin_0 = const()[name = tensor<string, []>("op_38434_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_38434_end_0 = const()[name = tensor<string, []>("op_38434_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_38434_end_mask_0 = const()[name = tensor<string, []>("op_38434_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38434_cast = slice_by_index(begin = var_38434_begin_0, end = var_38434_end_0, end_mask = var_38434_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38434_cast")];
            tensor<int32, [4]> var_38438_begin_0 = const()[name = tensor<string, []>("op_38438_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_38438_end_0 = const()[name = tensor<string, []>("op_38438_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_38438_end_mask_0 = const()[name = tensor<string, []>("op_38438_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38438_cast = slice_by_index(begin = var_38438_begin_0, end = var_38438_end_0, end_mask = var_38438_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38438_cast")];
            tensor<int32, [4]> var_38442_begin_0 = const()[name = tensor<string, []>("op_38442_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_38442_end_0 = const()[name = tensor<string, []>("op_38442_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_38442_end_mask_0 = const()[name = tensor<string, []>("op_38442_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38442_cast = slice_by_index(begin = var_38442_begin_0, end = var_38442_end_0, end_mask = var_38442_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38442_cast")];
            tensor<int32, [4]> var_38446_begin_0 = const()[name = tensor<string, []>("op_38446_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_38446_end_0 = const()[name = tensor<string, []>("op_38446_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_38446_end_mask_0 = const()[name = tensor<string, []>("op_38446_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38446_cast = slice_by_index(begin = var_38446_begin_0, end = var_38446_end_0, end_mask = var_38446_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38446_cast")];
            tensor<int32, [4]> var_38450_begin_0 = const()[name = tensor<string, []>("op_38450_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_38450_end_0 = const()[name = tensor<string, []>("op_38450_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_38450_end_mask_0 = const()[name = tensor<string, []>("op_38450_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38450_cast = slice_by_index(begin = var_38450_begin_0, end = var_38450_end_0, end_mask = var_38450_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38450_cast")];
            tensor<int32, [4]> var_38454_begin_0 = const()[name = tensor<string, []>("op_38454_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_38454_end_0 = const()[name = tensor<string, []>("op_38454_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_38454_end_mask_0 = const()[name = tensor<string, []>("op_38454_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38454_cast = slice_by_index(begin = var_38454_begin_0, end = var_38454_end_0, end_mask = var_38454_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38454_cast")];
            tensor<int32, [4]> var_38458_begin_0 = const()[name = tensor<string, []>("op_38458_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_38458_end_0 = const()[name = tensor<string, []>("op_38458_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_38458_end_mask_0 = const()[name = tensor<string, []>("op_38458_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38458_cast = slice_by_index(begin = var_38458_begin_0, end = var_38458_end_0, end_mask = var_38458_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38458_cast")];
            tensor<int32, [4]> var_38462_begin_0 = const()[name = tensor<string, []>("op_38462_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_38462_end_0 = const()[name = tensor<string, []>("op_38462_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_38462_end_mask_0 = const()[name = tensor<string, []>("op_38462_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38462_cast = slice_by_index(begin = var_38462_begin_0, end = var_38462_end_0, end_mask = var_38462_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38462_cast")];
            tensor<int32, [4]> var_38466_begin_0 = const()[name = tensor<string, []>("op_38466_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_38466_end_0 = const()[name = tensor<string, []>("op_38466_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_38466_end_mask_0 = const()[name = tensor<string, []>("op_38466_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38466_cast = slice_by_index(begin = var_38466_begin_0, end = var_38466_end_0, end_mask = var_38466_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38466_cast")];
            tensor<int32, [4]> var_38470_begin_0 = const()[name = tensor<string, []>("op_38470_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_38470_end_0 = const()[name = tensor<string, []>("op_38470_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_38470_end_mask_0 = const()[name = tensor<string, []>("op_38470_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38470_cast = slice_by_index(begin = var_38470_begin_0, end = var_38470_end_0, end_mask = var_38470_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38470_cast")];
            tensor<int32, [4]> var_38474_begin_0 = const()[name = tensor<string, []>("op_38474_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_38474_end_0 = const()[name = tensor<string, []>("op_38474_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_38474_end_mask_0 = const()[name = tensor<string, []>("op_38474_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38474_cast = slice_by_index(begin = var_38474_begin_0, end = var_38474_end_0, end_mask = var_38474_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38474_cast")];
            tensor<int32, [4]> var_38478_begin_0 = const()[name = tensor<string, []>("op_38478_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_38478_end_0 = const()[name = tensor<string, []>("op_38478_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_38478_end_mask_0 = const()[name = tensor<string, []>("op_38478_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38478_cast = slice_by_index(begin = var_38478_begin_0, end = var_38478_end_0, end_mask = var_38478_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38478_cast")];
            tensor<int32, [4]> var_38482_begin_0 = const()[name = tensor<string, []>("op_38482_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_38482_end_0 = const()[name = tensor<string, []>("op_38482_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_38482_end_mask_0 = const()[name = tensor<string, []>("op_38482_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38482_cast = slice_by_index(begin = var_38482_begin_0, end = var_38482_end_0, end_mask = var_38482_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38482_cast")];
            tensor<int32, [4]> var_38486_begin_0 = const()[name = tensor<string, []>("op_38486_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_38486_end_0 = const()[name = tensor<string, []>("op_38486_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_38486_end_mask_0 = const()[name = tensor<string, []>("op_38486_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38486_cast = slice_by_index(begin = var_38486_begin_0, end = var_38486_end_0, end_mask = var_38486_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38486_cast")];
            tensor<int32, [4]> var_38490_begin_0 = const()[name = tensor<string, []>("op_38490_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_38490_end_0 = const()[name = tensor<string, []>("op_38490_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_38490_end_mask_0 = const()[name = tensor<string, []>("op_38490_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38490_cast = slice_by_index(begin = var_38490_begin_0, end = var_38490_end_0, end_mask = var_38490_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38490_cast")];
            tensor<int32, [4]> var_38494_begin_0 = const()[name = tensor<string, []>("op_38494_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_38494_end_0 = const()[name = tensor<string, []>("op_38494_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_38494_end_mask_0 = const()[name = tensor<string, []>("op_38494_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_38494_cast = slice_by_index(begin = var_38494_begin_0, end = var_38494_end_0, end_mask = var_38494_end_mask_0, x = v_171_cast)[name = tensor<string, []>("op_38494_cast")];
            tensor<string, []> var_38498_equation_0 = const()[name = tensor<string, []>("op_38498_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38498_cast = einsum(equation = var_38498_equation_0, values = (var_38340_cast, var_38257_cast))[name = tensor<string, []>("op_38498_cast")];
            tensor<fp16, []> var_38499_to_fp16 = const()[name = tensor<string, []>("op_38499_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3241_cast = mul(x = var_38498_cast, y = var_38499_to_fp16)[name = tensor<string, []>("aw_3241_cast")];
            tensor<string, []> var_38502_equation_0 = const()[name = tensor<string, []>("op_38502_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38502_cast = einsum(equation = var_38502_equation_0, values = (var_38344_cast, var_38261_cast))[name = tensor<string, []>("op_38502_cast")];
            tensor<fp16, []> var_38503_to_fp16 = const()[name = tensor<string, []>("op_38503_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3243_cast = mul(x = var_38502_cast, y = var_38503_to_fp16)[name = tensor<string, []>("aw_3243_cast")];
            tensor<string, []> var_38506_equation_0 = const()[name = tensor<string, []>("op_38506_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38506_cast = einsum(equation = var_38506_equation_0, values = (var_38348_cast, var_38265_cast))[name = tensor<string, []>("op_38506_cast")];
            tensor<fp16, []> var_38507_to_fp16 = const()[name = tensor<string, []>("op_38507_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3245_cast = mul(x = var_38506_cast, y = var_38507_to_fp16)[name = tensor<string, []>("aw_3245_cast")];
            tensor<string, []> var_38510_equation_0 = const()[name = tensor<string, []>("op_38510_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38510_cast = einsum(equation = var_38510_equation_0, values = (var_38352_cast, var_38269_cast))[name = tensor<string, []>("op_38510_cast")];
            tensor<fp16, []> var_38511_to_fp16 = const()[name = tensor<string, []>("op_38511_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3247_cast = mul(x = var_38510_cast, y = var_38511_to_fp16)[name = tensor<string, []>("aw_3247_cast")];
            tensor<string, []> var_38514_equation_0 = const()[name = tensor<string, []>("op_38514_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38514_cast = einsum(equation = var_38514_equation_0, values = (var_38356_cast, var_38273_cast))[name = tensor<string, []>("op_38514_cast")];
            tensor<fp16, []> var_38515_to_fp16 = const()[name = tensor<string, []>("op_38515_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3249_cast = mul(x = var_38514_cast, y = var_38515_to_fp16)[name = tensor<string, []>("aw_3249_cast")];
            tensor<string, []> var_38518_equation_0 = const()[name = tensor<string, []>("op_38518_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38518_cast = einsum(equation = var_38518_equation_0, values = (var_38360_cast, var_38277_cast))[name = tensor<string, []>("op_38518_cast")];
            tensor<fp16, []> var_38519_to_fp16 = const()[name = tensor<string, []>("op_38519_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3251_cast = mul(x = var_38518_cast, y = var_38519_to_fp16)[name = tensor<string, []>("aw_3251_cast")];
            tensor<string, []> var_38522_equation_0 = const()[name = tensor<string, []>("op_38522_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38522_cast = einsum(equation = var_38522_equation_0, values = (var_38364_cast, var_38281_cast))[name = tensor<string, []>("op_38522_cast")];
            tensor<fp16, []> var_38523_to_fp16 = const()[name = tensor<string, []>("op_38523_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3253_cast = mul(x = var_38522_cast, y = var_38523_to_fp16)[name = tensor<string, []>("aw_3253_cast")];
            tensor<string, []> var_38526_equation_0 = const()[name = tensor<string, []>("op_38526_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38526_cast = einsum(equation = var_38526_equation_0, values = (var_38368_cast, var_38285_cast))[name = tensor<string, []>("op_38526_cast")];
            tensor<fp16, []> var_38527_to_fp16 = const()[name = tensor<string, []>("op_38527_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3255_cast = mul(x = var_38526_cast, y = var_38527_to_fp16)[name = tensor<string, []>("aw_3255_cast")];
            tensor<string, []> var_38530_equation_0 = const()[name = tensor<string, []>("op_38530_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38530_cast = einsum(equation = var_38530_equation_0, values = (var_38372_cast, var_38289_cast))[name = tensor<string, []>("op_38530_cast")];
            tensor<fp16, []> var_38531_to_fp16 = const()[name = tensor<string, []>("op_38531_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3257_cast = mul(x = var_38530_cast, y = var_38531_to_fp16)[name = tensor<string, []>("aw_3257_cast")];
            tensor<string, []> var_38534_equation_0 = const()[name = tensor<string, []>("op_38534_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38534_cast = einsum(equation = var_38534_equation_0, values = (var_38376_cast, var_38293_cast))[name = tensor<string, []>("op_38534_cast")];
            tensor<fp16, []> var_38535_to_fp16 = const()[name = tensor<string, []>("op_38535_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3259_cast = mul(x = var_38534_cast, y = var_38535_to_fp16)[name = tensor<string, []>("aw_3259_cast")];
            tensor<string, []> var_38538_equation_0 = const()[name = tensor<string, []>("op_38538_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38538_cast = einsum(equation = var_38538_equation_0, values = (var_38380_cast, var_38297_cast))[name = tensor<string, []>("op_38538_cast")];
            tensor<fp16, []> var_38539_to_fp16 = const()[name = tensor<string, []>("op_38539_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3261_cast = mul(x = var_38538_cast, y = var_38539_to_fp16)[name = tensor<string, []>("aw_3261_cast")];
            tensor<string, []> var_38542_equation_0 = const()[name = tensor<string, []>("op_38542_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38542_cast = einsum(equation = var_38542_equation_0, values = (var_38384_cast, var_38301_cast))[name = tensor<string, []>("op_38542_cast")];
            tensor<fp16, []> var_38543_to_fp16 = const()[name = tensor<string, []>("op_38543_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3263_cast = mul(x = var_38542_cast, y = var_38543_to_fp16)[name = tensor<string, []>("aw_3263_cast")];
            tensor<string, []> var_38546_equation_0 = const()[name = tensor<string, []>("op_38546_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38546_cast = einsum(equation = var_38546_equation_0, values = (var_38388_cast, var_38305_cast))[name = tensor<string, []>("op_38546_cast")];
            tensor<fp16, []> var_38547_to_fp16 = const()[name = tensor<string, []>("op_38547_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3265_cast = mul(x = var_38546_cast, y = var_38547_to_fp16)[name = tensor<string, []>("aw_3265_cast")];
            tensor<string, []> var_38550_equation_0 = const()[name = tensor<string, []>("op_38550_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38550_cast = einsum(equation = var_38550_equation_0, values = (var_38392_cast, var_38309_cast))[name = tensor<string, []>("op_38550_cast")];
            tensor<fp16, []> var_38551_to_fp16 = const()[name = tensor<string, []>("op_38551_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3267_cast = mul(x = var_38550_cast, y = var_38551_to_fp16)[name = tensor<string, []>("aw_3267_cast")];
            tensor<string, []> var_38554_equation_0 = const()[name = tensor<string, []>("op_38554_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38554_cast = einsum(equation = var_38554_equation_0, values = (var_38396_cast, var_38313_cast))[name = tensor<string, []>("op_38554_cast")];
            tensor<fp16, []> var_38555_to_fp16 = const()[name = tensor<string, []>("op_38555_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3269_cast = mul(x = var_38554_cast, y = var_38555_to_fp16)[name = tensor<string, []>("aw_3269_cast")];
            tensor<string, []> var_38558_equation_0 = const()[name = tensor<string, []>("op_38558_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38558_cast = einsum(equation = var_38558_equation_0, values = (var_38400_cast, var_38317_cast))[name = tensor<string, []>("op_38558_cast")];
            tensor<fp16, []> var_38559_to_fp16 = const()[name = tensor<string, []>("op_38559_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3271_cast = mul(x = var_38558_cast, y = var_38559_to_fp16)[name = tensor<string, []>("aw_3271_cast")];
            tensor<string, []> var_38562_equation_0 = const()[name = tensor<string, []>("op_38562_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38562_cast = einsum(equation = var_38562_equation_0, values = (var_38404_cast, var_38321_cast))[name = tensor<string, []>("op_38562_cast")];
            tensor<fp16, []> var_38563_to_fp16 = const()[name = tensor<string, []>("op_38563_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3273_cast = mul(x = var_38562_cast, y = var_38563_to_fp16)[name = tensor<string, []>("aw_3273_cast")];
            tensor<string, []> var_38566_equation_0 = const()[name = tensor<string, []>("op_38566_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38566_cast = einsum(equation = var_38566_equation_0, values = (var_38408_cast, var_38325_cast))[name = tensor<string, []>("op_38566_cast")];
            tensor<fp16, []> var_38567_to_fp16 = const()[name = tensor<string, []>("op_38567_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3275_cast = mul(x = var_38566_cast, y = var_38567_to_fp16)[name = tensor<string, []>("aw_3275_cast")];
            tensor<string, []> var_38570_equation_0 = const()[name = tensor<string, []>("op_38570_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38570_cast = einsum(equation = var_38570_equation_0, values = (var_38412_cast, var_38329_cast))[name = tensor<string, []>("op_38570_cast")];
            tensor<fp16, []> var_38571_to_fp16 = const()[name = tensor<string, []>("op_38571_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3277_cast = mul(x = var_38570_cast, y = var_38571_to_fp16)[name = tensor<string, []>("aw_3277_cast")];
            tensor<string, []> var_38574_equation_0 = const()[name = tensor<string, []>("op_38574_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_38574_cast = einsum(equation = var_38574_equation_0, values = (var_38416_cast, var_38333_cast))[name = tensor<string, []>("op_38574_cast")];
            tensor<fp16, []> var_38575_to_fp16 = const()[name = tensor<string, []>("op_38575_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3279_cast = mul(x = var_38574_cast, y = var_38575_to_fp16)[name = tensor<string, []>("aw_3279_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38577_cast = softmax(axis = var_30385, x = aw_3241_cast)[name = tensor<string, []>("op_38577_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38578_cast = softmax(axis = var_30385, x = aw_3243_cast)[name = tensor<string, []>("op_38578_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38579_cast = softmax(axis = var_30385, x = aw_3245_cast)[name = tensor<string, []>("op_38579_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38580_cast = softmax(axis = var_30385, x = aw_3247_cast)[name = tensor<string, []>("op_38580_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38581_cast = softmax(axis = var_30385, x = aw_3249_cast)[name = tensor<string, []>("op_38581_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38582_cast = softmax(axis = var_30385, x = aw_3251_cast)[name = tensor<string, []>("op_38582_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38583_cast = softmax(axis = var_30385, x = aw_3253_cast)[name = tensor<string, []>("op_38583_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38584_cast = softmax(axis = var_30385, x = aw_3255_cast)[name = tensor<string, []>("op_38584_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38585_cast = softmax(axis = var_30385, x = aw_3257_cast)[name = tensor<string, []>("op_38585_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38586_cast = softmax(axis = var_30385, x = aw_3259_cast)[name = tensor<string, []>("op_38586_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38587_cast = softmax(axis = var_30385, x = aw_3261_cast)[name = tensor<string, []>("op_38587_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38588_cast = softmax(axis = var_30385, x = aw_3263_cast)[name = tensor<string, []>("op_38588_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38589_cast = softmax(axis = var_30385, x = aw_3265_cast)[name = tensor<string, []>("op_38589_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38590_cast = softmax(axis = var_30385, x = aw_3267_cast)[name = tensor<string, []>("op_38590_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38591_cast = softmax(axis = var_30385, x = aw_3269_cast)[name = tensor<string, []>("op_38591_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38592_cast = softmax(axis = var_30385, x = aw_3271_cast)[name = tensor<string, []>("op_38592_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38593_cast = softmax(axis = var_30385, x = aw_3273_cast)[name = tensor<string, []>("op_38593_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38594_cast = softmax(axis = var_30385, x = aw_3275_cast)[name = tensor<string, []>("op_38594_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38595_cast = softmax(axis = var_30385, x = aw_3277_cast)[name = tensor<string, []>("op_38595_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_38596_cast = softmax(axis = var_30385, x = aw_3279_cast)[name = tensor<string, []>("op_38596_cast")];
            tensor<string, []> var_38598_equation_0 = const()[name = tensor<string, []>("op_38598_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38598_cast = einsum(equation = var_38598_equation_0, values = (var_38418_cast, var_38577_cast))[name = tensor<string, []>("op_38598_cast")];
            tensor<string, []> var_38600_equation_0 = const()[name = tensor<string, []>("op_38600_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38600_cast = einsum(equation = var_38600_equation_0, values = (var_38422_cast, var_38578_cast))[name = tensor<string, []>("op_38600_cast")];
            tensor<string, []> var_38602_equation_0 = const()[name = tensor<string, []>("op_38602_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38602_cast = einsum(equation = var_38602_equation_0, values = (var_38426_cast, var_38579_cast))[name = tensor<string, []>("op_38602_cast")];
            tensor<string, []> var_38604_equation_0 = const()[name = tensor<string, []>("op_38604_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38604_cast = einsum(equation = var_38604_equation_0, values = (var_38430_cast, var_38580_cast))[name = tensor<string, []>("op_38604_cast")];
            tensor<string, []> var_38606_equation_0 = const()[name = tensor<string, []>("op_38606_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38606_cast = einsum(equation = var_38606_equation_0, values = (var_38434_cast, var_38581_cast))[name = tensor<string, []>("op_38606_cast")];
            tensor<string, []> var_38608_equation_0 = const()[name = tensor<string, []>("op_38608_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38608_cast = einsum(equation = var_38608_equation_0, values = (var_38438_cast, var_38582_cast))[name = tensor<string, []>("op_38608_cast")];
            tensor<string, []> var_38610_equation_0 = const()[name = tensor<string, []>("op_38610_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38610_cast = einsum(equation = var_38610_equation_0, values = (var_38442_cast, var_38583_cast))[name = tensor<string, []>("op_38610_cast")];
            tensor<string, []> var_38612_equation_0 = const()[name = tensor<string, []>("op_38612_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38612_cast = einsum(equation = var_38612_equation_0, values = (var_38446_cast, var_38584_cast))[name = tensor<string, []>("op_38612_cast")];
            tensor<string, []> var_38614_equation_0 = const()[name = tensor<string, []>("op_38614_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38614_cast = einsum(equation = var_38614_equation_0, values = (var_38450_cast, var_38585_cast))[name = tensor<string, []>("op_38614_cast")];
            tensor<string, []> var_38616_equation_0 = const()[name = tensor<string, []>("op_38616_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38616_cast = einsum(equation = var_38616_equation_0, values = (var_38454_cast, var_38586_cast))[name = tensor<string, []>("op_38616_cast")];
            tensor<string, []> var_38618_equation_0 = const()[name = tensor<string, []>("op_38618_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38618_cast = einsum(equation = var_38618_equation_0, values = (var_38458_cast, var_38587_cast))[name = tensor<string, []>("op_38618_cast")];
            tensor<string, []> var_38620_equation_0 = const()[name = tensor<string, []>("op_38620_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38620_cast = einsum(equation = var_38620_equation_0, values = (var_38462_cast, var_38588_cast))[name = tensor<string, []>("op_38620_cast")];
            tensor<string, []> var_38622_equation_0 = const()[name = tensor<string, []>("op_38622_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38622_cast = einsum(equation = var_38622_equation_0, values = (var_38466_cast, var_38589_cast))[name = tensor<string, []>("op_38622_cast")];
            tensor<string, []> var_38624_equation_0 = const()[name = tensor<string, []>("op_38624_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38624_cast = einsum(equation = var_38624_equation_0, values = (var_38470_cast, var_38590_cast))[name = tensor<string, []>("op_38624_cast")];
            tensor<string, []> var_38626_equation_0 = const()[name = tensor<string, []>("op_38626_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38626_cast = einsum(equation = var_38626_equation_0, values = (var_38474_cast, var_38591_cast))[name = tensor<string, []>("op_38626_cast")];
            tensor<string, []> var_38628_equation_0 = const()[name = tensor<string, []>("op_38628_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38628_cast = einsum(equation = var_38628_equation_0, values = (var_38478_cast, var_38592_cast))[name = tensor<string, []>("op_38628_cast")];
            tensor<string, []> var_38630_equation_0 = const()[name = tensor<string, []>("op_38630_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38630_cast = einsum(equation = var_38630_equation_0, values = (var_38482_cast, var_38593_cast))[name = tensor<string, []>("op_38630_cast")];
            tensor<string, []> var_38632_equation_0 = const()[name = tensor<string, []>("op_38632_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38632_cast = einsum(equation = var_38632_equation_0, values = (var_38486_cast, var_38594_cast))[name = tensor<string, []>("op_38632_cast")];
            tensor<string, []> var_38634_equation_0 = const()[name = tensor<string, []>("op_38634_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38634_cast = einsum(equation = var_38634_equation_0, values = (var_38490_cast, var_38595_cast))[name = tensor<string, []>("op_38634_cast")];
            tensor<string, []> var_38636_equation_0 = const()[name = tensor<string, []>("op_38636_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_38636_cast = einsum(equation = var_38636_equation_0, values = (var_38494_cast, var_38596_cast))[name = tensor<string, []>("op_38636_cast")];
            tensor<bool, []> input_507_interleave_0 = const()[name = tensor<string, []>("input_507_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_507_cast = concat(axis = var_30385, interleave = input_507_interleave_0, values = (var_38598_cast, var_38600_cast, var_38602_cast, var_38604_cast, var_38606_cast, var_38608_cast, var_38610_cast, var_38612_cast, var_38614_cast, var_38616_cast, var_38618_cast, var_38620_cast, var_38622_cast, var_38624_cast, var_38626_cast, var_38628_cast, var_38630_cast, var_38632_cast, var_38634_cast, var_38636_cast))[name = tensor<string, []>("input_507_cast")];
            tensor<int32, [2]> var_38642 = const()[name = tensor<string, []>("op_38642"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_38644 = const()[name = tensor<string, []>("op_38644"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_38646_pad_type_0 = const()[name = tensor<string, []>("op_38646_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_38646_pad_0 = const()[name = tensor<string, []>("op_38646_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3171408384)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3174685248)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_38646_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_out_0_bias_to_fp16, dilations = var_38644, groups = var_30385, pad = var_38646_pad_0, pad_type = var_38646_pad_type_0, strides = var_38642, weight = up_blocks_0_attentions_0_transformer_blocks_8_attn2_to_out_0_weight_to_fp16, x = input_507_cast)[name = tensor<string, []>("op_38646_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_257_cast = add(x = var_38646_cast, y = inputs_255_cast)[name = tensor<string, []>("inputs_257_cast")];
            tensor<int32, [1]> var_38650 = const()[name = tensor<string, []>("op_38650"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_257_cast = reduce_mean(axes = var_38650, keep_dims = var_30380, x = inputs_257_cast)[name = tensor<string, []>("channels_mean_257_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_257_cast = sub(x = inputs_257_cast, y = channels_mean_257_cast)[name = tensor<string, []>("zero_mean_257_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_257_cast = mul(x = zero_mean_257_cast, y = zero_mean_257_cast)[name = tensor<string, []>("zero_mean_sq_257_cast")];
            tensor<int32, [1]> var_38654 = const()[name = tensor<string, []>("op_38654"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_38655_cast = reduce_mean(axes = var_38654, keep_dims = var_30380, x = zero_mean_sq_257_cast)[name = tensor<string, []>("op_38655_cast")];
            tensor<fp16, []> var_38656_to_fp16 = const()[name = tensor<string, []>("op_38656_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_38657_cast = add(x = var_38655_cast, y = var_38656_to_fp16)[name = tensor<string, []>("op_38657_cast")];
            tensor<fp16, []> denom_257_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_257_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_257_cast = rsqrt(epsilon = denom_257_epsilon_0_to_fp16, x = var_38657_cast)[name = tensor<string, []>("denom_257_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_257_cast = mul(x = zero_mean_257_cast, y = denom_257_cast)[name = tensor<string, []>("out_257_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_38661_to_fp16 = const()[name = tensor<string, []>("op_38661_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3174687872)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_38662_cast = add(x = out_257_cast, y = var_38661_to_fp16)[name = tensor<string, []>("op_38662_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_38664_to_fp16 = const()[name = tensor<string, []>("op_38664_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3174690496)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_509_cast = mul(x = var_38662_cast, y = var_38664_to_fp16)[name = tensor<string, []>("input_509_cast")];
            tensor<int32, [2]> var_38672 = const()[name = tensor<string, []>("op_38672"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_38674 = const()[name = tensor<string, []>("op_38674"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_38676_pad_type_0 = const()[name = tensor<string, []>("op_38676_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_38676_pad_0 = const()[name = tensor<string, []>("op_38676_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3174693120)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3200907584)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_38676_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_8_ff_net_0_proj_bias_to_fp16, dilations = var_38674, groups = var_30385, pad = var_38676_pad_0, pad_type = var_38676_pad_type_0, strides = var_38672, weight = up_blocks_0_attentions_0_transformer_blocks_8_ff_net_0_proj_weight_to_fp16, x = input_509_cast)[name = tensor<string, []>("op_38676_cast")];
            tensor<int32, [2]> var_38677_split_sizes_0 = const()[name = tensor<string, []>("op_38677_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_38677_axis_0 = const()[name = tensor<string, []>("op_38677_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_38677_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_38677_cast_1 = split(axis = var_38677_axis_0, split_sizes = var_38677_split_sizes_0, x = var_38676_cast)[name = tensor<string, []>("op_38677_cast")];
            tensor<string, []> var_38679_mode_0 = const()[name = tensor<string, []>("op_38679_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_38679_cast = gelu(mode = var_38679_mode_0, x = var_38677_cast_1)[name = tensor<string, []>("op_38679_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_511_cast = mul(x = var_38677_cast_0, y = var_38679_cast)[name = tensor<string, []>("input_511_cast")];
            tensor<int32, [2]> var_38683 = const()[name = tensor<string, []>("op_38683"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_38685 = const()[name = tensor<string, []>("op_38685"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_38687_pad_type_0 = const()[name = tensor<string, []>("op_38687_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_38687_pad_0 = const()[name = tensor<string, []>("op_38687_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3200928128)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3214035392)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_38687_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_8_ff_net_2_bias_to_fp16, dilations = var_38685, groups = var_30385, pad = var_38687_pad_0, pad_type = var_38687_pad_type_0, strides = var_38683, weight = up_blocks_0_attentions_0_transformer_blocks_8_ff_net_2_weight_to_fp16, x = input_511_cast)[name = tensor<string, []>("op_38687_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_259_cast = add(x = var_38687_cast, y = inputs_257_cast)[name = tensor<string, []>("inputs_259_cast")];
            tensor<int32, [1]> var_38697 = const()[name = tensor<string, []>("op_38697"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_259_cast = reduce_mean(axes = var_38697, keep_dims = var_30380, x = inputs_259_cast)[name = tensor<string, []>("channels_mean_259_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_259_cast = sub(x = inputs_259_cast, y = channels_mean_259_cast)[name = tensor<string, []>("zero_mean_259_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_259_cast = mul(x = zero_mean_259_cast, y = zero_mean_259_cast)[name = tensor<string, []>("zero_mean_sq_259_cast")];
            tensor<int32, [1]> var_38701 = const()[name = tensor<string, []>("op_38701"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_38702_cast = reduce_mean(axes = var_38701, keep_dims = var_30380, x = zero_mean_sq_259_cast)[name = tensor<string, []>("op_38702_cast")];
            tensor<fp16, []> var_38703_to_fp16 = const()[name = tensor<string, []>("op_38703_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_38704_cast = add(x = var_38702_cast, y = var_38703_to_fp16)[name = tensor<string, []>("op_38704_cast")];
            tensor<fp16, []> denom_259_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_259_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_259_cast = rsqrt(epsilon = denom_259_epsilon_0_to_fp16, x = var_38704_cast)[name = tensor<string, []>("denom_259_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_259_cast = mul(x = zero_mean_259_cast, y = denom_259_cast)[name = tensor<string, []>("out_259_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_38708_to_fp16 = const()[name = tensor<string, []>("op_38708_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3214038016)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_38709_cast = add(x = out_259_cast, y = var_38708_to_fp16)[name = tensor<string, []>("op_38709_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_38711_to_fp16 = const()[name = tensor<string, []>("op_38711_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3214040640)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_341_cast = mul(x = var_38709_cast, y = var_38711_to_fp16)[name = tensor<string, []>("hidden_states_341_cast")];
            tensor<int32, [2]> var_38718 = const()[name = tensor<string, []>("op_38718"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_38720 = const()[name = tensor<string, []>("op_38720"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_173_pad_type_0 = const()[name = tensor<string, []>("q_173_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_173_pad_0 = const()[name = tensor<string, []>("q_173_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3214043264)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_173_cast = conv(dilations = var_38720, groups = var_30385, pad = q_173_pad_0, pad_type = q_173_pad_type_0, strides = var_38718, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_q_weight_to_fp16, x = hidden_states_341_cast)[name = tensor<string, []>("q_173_cast")];
            tensor<int32, [2]> var_38724 = const()[name = tensor<string, []>("op_38724"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_38726 = const()[name = tensor<string, []>("op_38726"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_345_pad_type_0 = const()[name = tensor<string, []>("k_345_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_345_pad_0 = const()[name = tensor<string, []>("k_345_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3217320128)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_345_cast = conv(dilations = var_38726, groups = var_30385, pad = k_345_pad_0, pad_type = k_345_pad_type_0, strides = var_38724, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_k_weight_to_fp16, x = hidden_states_341_cast)[name = tensor<string, []>("k_345_cast")];
            tensor<int32, [2]> var_38730 = const()[name = tensor<string, []>("op_38730"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_38732 = const()[name = tensor<string, []>("op_38732"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_173_pad_type_0 = const()[name = tensor<string, []>("v_173_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_173_pad_0 = const()[name = tensor<string, []>("v_173_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3220596992)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_173_cast = conv(dilations = var_38732, groups = var_30385, pad = v_173_pad_0, pad_type = v_173_pad_type_0, strides = var_38730, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_v_weight_to_fp16, x = hidden_states_341_cast)[name = tensor<string, []>("v_173_cast")];
            tensor<int32, [4]> var_38736_begin_0 = const()[name = tensor<string, []>("op_38736_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_38736_end_0 = const()[name = tensor<string, []>("op_38736_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_38736_end_mask_0 = const()[name = tensor<string, []>("op_38736_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38736_cast = slice_by_index(begin = var_38736_begin_0, end = var_38736_end_0, end_mask = var_38736_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38736_cast")];
            tensor<int32, [4]> var_38740_begin_0 = const()[name = tensor<string, []>("op_38740_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_38740_end_0 = const()[name = tensor<string, []>("op_38740_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_38740_end_mask_0 = const()[name = tensor<string, []>("op_38740_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38740_cast = slice_by_index(begin = var_38740_begin_0, end = var_38740_end_0, end_mask = var_38740_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38740_cast")];
            tensor<int32, [4]> var_38744_begin_0 = const()[name = tensor<string, []>("op_38744_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_38744_end_0 = const()[name = tensor<string, []>("op_38744_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_38744_end_mask_0 = const()[name = tensor<string, []>("op_38744_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38744_cast = slice_by_index(begin = var_38744_begin_0, end = var_38744_end_0, end_mask = var_38744_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38744_cast")];
            tensor<int32, [4]> var_38748_begin_0 = const()[name = tensor<string, []>("op_38748_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_38748_end_0 = const()[name = tensor<string, []>("op_38748_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_38748_end_mask_0 = const()[name = tensor<string, []>("op_38748_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38748_cast = slice_by_index(begin = var_38748_begin_0, end = var_38748_end_0, end_mask = var_38748_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38748_cast")];
            tensor<int32, [4]> var_38752_begin_0 = const()[name = tensor<string, []>("op_38752_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_38752_end_0 = const()[name = tensor<string, []>("op_38752_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_38752_end_mask_0 = const()[name = tensor<string, []>("op_38752_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38752_cast = slice_by_index(begin = var_38752_begin_0, end = var_38752_end_0, end_mask = var_38752_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38752_cast")];
            tensor<int32, [4]> var_38756_begin_0 = const()[name = tensor<string, []>("op_38756_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_38756_end_0 = const()[name = tensor<string, []>("op_38756_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_38756_end_mask_0 = const()[name = tensor<string, []>("op_38756_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38756_cast = slice_by_index(begin = var_38756_begin_0, end = var_38756_end_0, end_mask = var_38756_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38756_cast")];
            tensor<int32, [4]> var_38760_begin_0 = const()[name = tensor<string, []>("op_38760_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_38760_end_0 = const()[name = tensor<string, []>("op_38760_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_38760_end_mask_0 = const()[name = tensor<string, []>("op_38760_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38760_cast = slice_by_index(begin = var_38760_begin_0, end = var_38760_end_0, end_mask = var_38760_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38760_cast")];
            tensor<int32, [4]> var_38764_begin_0 = const()[name = tensor<string, []>("op_38764_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_38764_end_0 = const()[name = tensor<string, []>("op_38764_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_38764_end_mask_0 = const()[name = tensor<string, []>("op_38764_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38764_cast = slice_by_index(begin = var_38764_begin_0, end = var_38764_end_0, end_mask = var_38764_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38764_cast")];
            tensor<int32, [4]> var_38768_begin_0 = const()[name = tensor<string, []>("op_38768_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_38768_end_0 = const()[name = tensor<string, []>("op_38768_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_38768_end_mask_0 = const()[name = tensor<string, []>("op_38768_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38768_cast = slice_by_index(begin = var_38768_begin_0, end = var_38768_end_0, end_mask = var_38768_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38768_cast")];
            tensor<int32, [4]> var_38772_begin_0 = const()[name = tensor<string, []>("op_38772_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_38772_end_0 = const()[name = tensor<string, []>("op_38772_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_38772_end_mask_0 = const()[name = tensor<string, []>("op_38772_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38772_cast = slice_by_index(begin = var_38772_begin_0, end = var_38772_end_0, end_mask = var_38772_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38772_cast")];
            tensor<int32, [4]> var_38776_begin_0 = const()[name = tensor<string, []>("op_38776_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_38776_end_0 = const()[name = tensor<string, []>("op_38776_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_38776_end_mask_0 = const()[name = tensor<string, []>("op_38776_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38776_cast = slice_by_index(begin = var_38776_begin_0, end = var_38776_end_0, end_mask = var_38776_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38776_cast")];
            tensor<int32, [4]> var_38780_begin_0 = const()[name = tensor<string, []>("op_38780_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_38780_end_0 = const()[name = tensor<string, []>("op_38780_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_38780_end_mask_0 = const()[name = tensor<string, []>("op_38780_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38780_cast = slice_by_index(begin = var_38780_begin_0, end = var_38780_end_0, end_mask = var_38780_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38780_cast")];
            tensor<int32, [4]> var_38784_begin_0 = const()[name = tensor<string, []>("op_38784_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_38784_end_0 = const()[name = tensor<string, []>("op_38784_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_38784_end_mask_0 = const()[name = tensor<string, []>("op_38784_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38784_cast = slice_by_index(begin = var_38784_begin_0, end = var_38784_end_0, end_mask = var_38784_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38784_cast")];
            tensor<int32, [4]> var_38788_begin_0 = const()[name = tensor<string, []>("op_38788_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_38788_end_0 = const()[name = tensor<string, []>("op_38788_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_38788_end_mask_0 = const()[name = tensor<string, []>("op_38788_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38788_cast = slice_by_index(begin = var_38788_begin_0, end = var_38788_end_0, end_mask = var_38788_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38788_cast")];
            tensor<int32, [4]> var_38792_begin_0 = const()[name = tensor<string, []>("op_38792_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_38792_end_0 = const()[name = tensor<string, []>("op_38792_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_38792_end_mask_0 = const()[name = tensor<string, []>("op_38792_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38792_cast = slice_by_index(begin = var_38792_begin_0, end = var_38792_end_0, end_mask = var_38792_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38792_cast")];
            tensor<int32, [4]> var_38796_begin_0 = const()[name = tensor<string, []>("op_38796_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_38796_end_0 = const()[name = tensor<string, []>("op_38796_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_38796_end_mask_0 = const()[name = tensor<string, []>("op_38796_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38796_cast = slice_by_index(begin = var_38796_begin_0, end = var_38796_end_0, end_mask = var_38796_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38796_cast")];
            tensor<int32, [4]> var_38800_begin_0 = const()[name = tensor<string, []>("op_38800_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_38800_end_0 = const()[name = tensor<string, []>("op_38800_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_38800_end_mask_0 = const()[name = tensor<string, []>("op_38800_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38800_cast = slice_by_index(begin = var_38800_begin_0, end = var_38800_end_0, end_mask = var_38800_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38800_cast")];
            tensor<int32, [4]> var_38804_begin_0 = const()[name = tensor<string, []>("op_38804_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_38804_end_0 = const()[name = tensor<string, []>("op_38804_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_38804_end_mask_0 = const()[name = tensor<string, []>("op_38804_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38804_cast = slice_by_index(begin = var_38804_begin_0, end = var_38804_end_0, end_mask = var_38804_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38804_cast")];
            tensor<int32, [4]> var_38808_begin_0 = const()[name = tensor<string, []>("op_38808_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_38808_end_0 = const()[name = tensor<string, []>("op_38808_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_38808_end_mask_0 = const()[name = tensor<string, []>("op_38808_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38808_cast = slice_by_index(begin = var_38808_begin_0, end = var_38808_end_0, end_mask = var_38808_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38808_cast")];
            tensor<int32, [4]> var_38812_begin_0 = const()[name = tensor<string, []>("op_38812_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_38812_end_0 = const()[name = tensor<string, []>("op_38812_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_38812_end_mask_0 = const()[name = tensor<string, []>("op_38812_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38812_cast = slice_by_index(begin = var_38812_begin_0, end = var_38812_end_0, end_mask = var_38812_end_mask_0, x = q_173_cast)[name = tensor<string, []>("op_38812_cast")];
            tensor<int32, [4]> k_347_perm_0 = const()[name = tensor<string, []>("k_347_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_38819_begin_0 = const()[name = tensor<string, []>("op_38819_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_38819_end_0 = const()[name = tensor<string, []>("op_38819_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_38819_end_mask_0 = const()[name = tensor<string, []>("op_38819_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_53 = transpose(perm = k_347_perm_0, x = k_345_cast)[name = tensor<string, []>("transpose_53")];
            tensor<fp16, [2, 1024, 1, 64]> var_38819_cast = slice_by_index(begin = var_38819_begin_0, end = var_38819_end_0, end_mask = var_38819_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38819_cast")];
            tensor<int32, [4]> var_38823_begin_0 = const()[name = tensor<string, []>("op_38823_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_38823_end_0 = const()[name = tensor<string, []>("op_38823_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_38823_end_mask_0 = const()[name = tensor<string, []>("op_38823_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38823_cast = slice_by_index(begin = var_38823_begin_0, end = var_38823_end_0, end_mask = var_38823_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38823_cast")];
            tensor<int32, [4]> var_38827_begin_0 = const()[name = tensor<string, []>("op_38827_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_38827_end_0 = const()[name = tensor<string, []>("op_38827_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_38827_end_mask_0 = const()[name = tensor<string, []>("op_38827_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38827_cast = slice_by_index(begin = var_38827_begin_0, end = var_38827_end_0, end_mask = var_38827_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38827_cast")];
            tensor<int32, [4]> var_38831_begin_0 = const()[name = tensor<string, []>("op_38831_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_38831_end_0 = const()[name = tensor<string, []>("op_38831_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_38831_end_mask_0 = const()[name = tensor<string, []>("op_38831_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38831_cast = slice_by_index(begin = var_38831_begin_0, end = var_38831_end_0, end_mask = var_38831_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38831_cast")];
            tensor<int32, [4]> var_38835_begin_0 = const()[name = tensor<string, []>("op_38835_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_38835_end_0 = const()[name = tensor<string, []>("op_38835_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_38835_end_mask_0 = const()[name = tensor<string, []>("op_38835_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38835_cast = slice_by_index(begin = var_38835_begin_0, end = var_38835_end_0, end_mask = var_38835_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38835_cast")];
            tensor<int32, [4]> var_38839_begin_0 = const()[name = tensor<string, []>("op_38839_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_38839_end_0 = const()[name = tensor<string, []>("op_38839_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_38839_end_mask_0 = const()[name = tensor<string, []>("op_38839_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38839_cast = slice_by_index(begin = var_38839_begin_0, end = var_38839_end_0, end_mask = var_38839_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38839_cast")];
            tensor<int32, [4]> var_38843_begin_0 = const()[name = tensor<string, []>("op_38843_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_38843_end_0 = const()[name = tensor<string, []>("op_38843_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_38843_end_mask_0 = const()[name = tensor<string, []>("op_38843_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38843_cast = slice_by_index(begin = var_38843_begin_0, end = var_38843_end_0, end_mask = var_38843_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38843_cast")];
            tensor<int32, [4]> var_38847_begin_0 = const()[name = tensor<string, []>("op_38847_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_38847_end_0 = const()[name = tensor<string, []>("op_38847_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_38847_end_mask_0 = const()[name = tensor<string, []>("op_38847_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38847_cast = slice_by_index(begin = var_38847_begin_0, end = var_38847_end_0, end_mask = var_38847_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38847_cast")];
            tensor<int32, [4]> var_38851_begin_0 = const()[name = tensor<string, []>("op_38851_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_38851_end_0 = const()[name = tensor<string, []>("op_38851_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_38851_end_mask_0 = const()[name = tensor<string, []>("op_38851_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38851_cast = slice_by_index(begin = var_38851_begin_0, end = var_38851_end_0, end_mask = var_38851_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38851_cast")];
            tensor<int32, [4]> var_38855_begin_0 = const()[name = tensor<string, []>("op_38855_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_38855_end_0 = const()[name = tensor<string, []>("op_38855_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_38855_end_mask_0 = const()[name = tensor<string, []>("op_38855_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38855_cast = slice_by_index(begin = var_38855_begin_0, end = var_38855_end_0, end_mask = var_38855_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38855_cast")];
            tensor<int32, [4]> var_38859_begin_0 = const()[name = tensor<string, []>("op_38859_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_38859_end_0 = const()[name = tensor<string, []>("op_38859_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_38859_end_mask_0 = const()[name = tensor<string, []>("op_38859_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38859_cast = slice_by_index(begin = var_38859_begin_0, end = var_38859_end_0, end_mask = var_38859_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38859_cast")];
            tensor<int32, [4]> var_38863_begin_0 = const()[name = tensor<string, []>("op_38863_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_38863_end_0 = const()[name = tensor<string, []>("op_38863_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_38863_end_mask_0 = const()[name = tensor<string, []>("op_38863_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38863_cast = slice_by_index(begin = var_38863_begin_0, end = var_38863_end_0, end_mask = var_38863_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38863_cast")];
            tensor<int32, [4]> var_38867_begin_0 = const()[name = tensor<string, []>("op_38867_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_38867_end_0 = const()[name = tensor<string, []>("op_38867_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_38867_end_mask_0 = const()[name = tensor<string, []>("op_38867_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38867_cast = slice_by_index(begin = var_38867_begin_0, end = var_38867_end_0, end_mask = var_38867_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38867_cast")];
            tensor<int32, [4]> var_38871_begin_0 = const()[name = tensor<string, []>("op_38871_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_38871_end_0 = const()[name = tensor<string, []>("op_38871_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_38871_end_mask_0 = const()[name = tensor<string, []>("op_38871_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38871_cast = slice_by_index(begin = var_38871_begin_0, end = var_38871_end_0, end_mask = var_38871_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38871_cast")];
            tensor<int32, [4]> var_38875_begin_0 = const()[name = tensor<string, []>("op_38875_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_38875_end_0 = const()[name = tensor<string, []>("op_38875_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_38875_end_mask_0 = const()[name = tensor<string, []>("op_38875_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38875_cast = slice_by_index(begin = var_38875_begin_0, end = var_38875_end_0, end_mask = var_38875_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38875_cast")];
            tensor<int32, [4]> var_38879_begin_0 = const()[name = tensor<string, []>("op_38879_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_38879_end_0 = const()[name = tensor<string, []>("op_38879_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_38879_end_mask_0 = const()[name = tensor<string, []>("op_38879_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38879_cast = slice_by_index(begin = var_38879_begin_0, end = var_38879_end_0, end_mask = var_38879_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38879_cast")];
            tensor<int32, [4]> var_38883_begin_0 = const()[name = tensor<string, []>("op_38883_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_38883_end_0 = const()[name = tensor<string, []>("op_38883_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_38883_end_mask_0 = const()[name = tensor<string, []>("op_38883_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38883_cast = slice_by_index(begin = var_38883_begin_0, end = var_38883_end_0, end_mask = var_38883_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38883_cast")];
            tensor<int32, [4]> var_38887_begin_0 = const()[name = tensor<string, []>("op_38887_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_38887_end_0 = const()[name = tensor<string, []>("op_38887_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_38887_end_mask_0 = const()[name = tensor<string, []>("op_38887_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38887_cast = slice_by_index(begin = var_38887_begin_0, end = var_38887_end_0, end_mask = var_38887_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38887_cast")];
            tensor<int32, [4]> var_38891_begin_0 = const()[name = tensor<string, []>("op_38891_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_38891_end_0 = const()[name = tensor<string, []>("op_38891_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_38891_end_mask_0 = const()[name = tensor<string, []>("op_38891_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38891_cast = slice_by_index(begin = var_38891_begin_0, end = var_38891_end_0, end_mask = var_38891_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38891_cast")];
            tensor<int32, [4]> var_38895_begin_0 = const()[name = tensor<string, []>("op_38895_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_38895_end_0 = const()[name = tensor<string, []>("op_38895_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_38895_end_mask_0 = const()[name = tensor<string, []>("op_38895_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_38895_cast = slice_by_index(begin = var_38895_begin_0, end = var_38895_end_0, end_mask = var_38895_end_mask_0, x = transpose_53)[name = tensor<string, []>("op_38895_cast")];
            tensor<int32, [4]> var_38897_begin_0 = const()[name = tensor<string, []>("op_38897_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_38897_end_0 = const()[name = tensor<string, []>("op_38897_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_38897_end_mask_0 = const()[name = tensor<string, []>("op_38897_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38897_cast = slice_by_index(begin = var_38897_begin_0, end = var_38897_end_0, end_mask = var_38897_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38897_cast")];
            tensor<int32, [4]> var_38901_begin_0 = const()[name = tensor<string, []>("op_38901_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_38901_end_0 = const()[name = tensor<string, []>("op_38901_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_38901_end_mask_0 = const()[name = tensor<string, []>("op_38901_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38901_cast = slice_by_index(begin = var_38901_begin_0, end = var_38901_end_0, end_mask = var_38901_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38901_cast")];
            tensor<int32, [4]> var_38905_begin_0 = const()[name = tensor<string, []>("op_38905_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_38905_end_0 = const()[name = tensor<string, []>("op_38905_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_38905_end_mask_0 = const()[name = tensor<string, []>("op_38905_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38905_cast = slice_by_index(begin = var_38905_begin_0, end = var_38905_end_0, end_mask = var_38905_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38905_cast")];
            tensor<int32, [4]> var_38909_begin_0 = const()[name = tensor<string, []>("op_38909_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_38909_end_0 = const()[name = tensor<string, []>("op_38909_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_38909_end_mask_0 = const()[name = tensor<string, []>("op_38909_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38909_cast = slice_by_index(begin = var_38909_begin_0, end = var_38909_end_0, end_mask = var_38909_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38909_cast")];
            tensor<int32, [4]> var_38913_begin_0 = const()[name = tensor<string, []>("op_38913_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_38913_end_0 = const()[name = tensor<string, []>("op_38913_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_38913_end_mask_0 = const()[name = tensor<string, []>("op_38913_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38913_cast = slice_by_index(begin = var_38913_begin_0, end = var_38913_end_0, end_mask = var_38913_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38913_cast")];
            tensor<int32, [4]> var_38917_begin_0 = const()[name = tensor<string, []>("op_38917_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_38917_end_0 = const()[name = tensor<string, []>("op_38917_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_38917_end_mask_0 = const()[name = tensor<string, []>("op_38917_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38917_cast = slice_by_index(begin = var_38917_begin_0, end = var_38917_end_0, end_mask = var_38917_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38917_cast")];
            tensor<int32, [4]> var_38921_begin_0 = const()[name = tensor<string, []>("op_38921_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_38921_end_0 = const()[name = tensor<string, []>("op_38921_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_38921_end_mask_0 = const()[name = tensor<string, []>("op_38921_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38921_cast = slice_by_index(begin = var_38921_begin_0, end = var_38921_end_0, end_mask = var_38921_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38921_cast")];
            tensor<int32, [4]> var_38925_begin_0 = const()[name = tensor<string, []>("op_38925_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_38925_end_0 = const()[name = tensor<string, []>("op_38925_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_38925_end_mask_0 = const()[name = tensor<string, []>("op_38925_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38925_cast = slice_by_index(begin = var_38925_begin_0, end = var_38925_end_0, end_mask = var_38925_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38925_cast")];
            tensor<int32, [4]> var_38929_begin_0 = const()[name = tensor<string, []>("op_38929_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_38929_end_0 = const()[name = tensor<string, []>("op_38929_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_38929_end_mask_0 = const()[name = tensor<string, []>("op_38929_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38929_cast = slice_by_index(begin = var_38929_begin_0, end = var_38929_end_0, end_mask = var_38929_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38929_cast")];
            tensor<int32, [4]> var_38933_begin_0 = const()[name = tensor<string, []>("op_38933_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_38933_end_0 = const()[name = tensor<string, []>("op_38933_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_38933_end_mask_0 = const()[name = tensor<string, []>("op_38933_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38933_cast = slice_by_index(begin = var_38933_begin_0, end = var_38933_end_0, end_mask = var_38933_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38933_cast")];
            tensor<int32, [4]> var_38937_begin_0 = const()[name = tensor<string, []>("op_38937_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_38937_end_0 = const()[name = tensor<string, []>("op_38937_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_38937_end_mask_0 = const()[name = tensor<string, []>("op_38937_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38937_cast = slice_by_index(begin = var_38937_begin_0, end = var_38937_end_0, end_mask = var_38937_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38937_cast")];
            tensor<int32, [4]> var_38941_begin_0 = const()[name = tensor<string, []>("op_38941_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_38941_end_0 = const()[name = tensor<string, []>("op_38941_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_38941_end_mask_0 = const()[name = tensor<string, []>("op_38941_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38941_cast = slice_by_index(begin = var_38941_begin_0, end = var_38941_end_0, end_mask = var_38941_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38941_cast")];
            tensor<int32, [4]> var_38945_begin_0 = const()[name = tensor<string, []>("op_38945_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_38945_end_0 = const()[name = tensor<string, []>("op_38945_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_38945_end_mask_0 = const()[name = tensor<string, []>("op_38945_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38945_cast = slice_by_index(begin = var_38945_begin_0, end = var_38945_end_0, end_mask = var_38945_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38945_cast")];
            tensor<int32, [4]> var_38949_begin_0 = const()[name = tensor<string, []>("op_38949_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_38949_end_0 = const()[name = tensor<string, []>("op_38949_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_38949_end_mask_0 = const()[name = tensor<string, []>("op_38949_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38949_cast = slice_by_index(begin = var_38949_begin_0, end = var_38949_end_0, end_mask = var_38949_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38949_cast")];
            tensor<int32, [4]> var_38953_begin_0 = const()[name = tensor<string, []>("op_38953_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_38953_end_0 = const()[name = tensor<string, []>("op_38953_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_38953_end_mask_0 = const()[name = tensor<string, []>("op_38953_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38953_cast = slice_by_index(begin = var_38953_begin_0, end = var_38953_end_0, end_mask = var_38953_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38953_cast")];
            tensor<int32, [4]> var_38957_begin_0 = const()[name = tensor<string, []>("op_38957_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_38957_end_0 = const()[name = tensor<string, []>("op_38957_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_38957_end_mask_0 = const()[name = tensor<string, []>("op_38957_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38957_cast = slice_by_index(begin = var_38957_begin_0, end = var_38957_end_0, end_mask = var_38957_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38957_cast")];
            tensor<int32, [4]> var_38961_begin_0 = const()[name = tensor<string, []>("op_38961_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_38961_end_0 = const()[name = tensor<string, []>("op_38961_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_38961_end_mask_0 = const()[name = tensor<string, []>("op_38961_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38961_cast = slice_by_index(begin = var_38961_begin_0, end = var_38961_end_0, end_mask = var_38961_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38961_cast")];
            tensor<int32, [4]> var_38965_begin_0 = const()[name = tensor<string, []>("op_38965_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_38965_end_0 = const()[name = tensor<string, []>("op_38965_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_38965_end_mask_0 = const()[name = tensor<string, []>("op_38965_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38965_cast = slice_by_index(begin = var_38965_begin_0, end = var_38965_end_0, end_mask = var_38965_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38965_cast")];
            tensor<int32, [4]> var_38969_begin_0 = const()[name = tensor<string, []>("op_38969_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_38969_end_0 = const()[name = tensor<string, []>("op_38969_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_38969_end_mask_0 = const()[name = tensor<string, []>("op_38969_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38969_cast = slice_by_index(begin = var_38969_begin_0, end = var_38969_end_0, end_mask = var_38969_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38969_cast")];
            tensor<int32, [4]> var_38973_begin_0 = const()[name = tensor<string, []>("op_38973_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_38973_end_0 = const()[name = tensor<string, []>("op_38973_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_38973_end_mask_0 = const()[name = tensor<string, []>("op_38973_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_38973_cast = slice_by_index(begin = var_38973_begin_0, end = var_38973_end_0, end_mask = var_38973_end_mask_0, x = v_173_cast)[name = tensor<string, []>("op_38973_cast")];
            tensor<string, []> var_38977_equation_0 = const()[name = tensor<string, []>("op_38977_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38977_cast = einsum(equation = var_38977_equation_0, values = (var_38819_cast, var_38736_cast))[name = tensor<string, []>("op_38977_cast")];
            tensor<fp16, []> var_38978_to_fp16 = const()[name = tensor<string, []>("op_38978_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3281_cast = mul(x = var_38977_cast, y = var_38978_to_fp16)[name = tensor<string, []>("aw_3281_cast")];
            tensor<string, []> var_38981_equation_0 = const()[name = tensor<string, []>("op_38981_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38981_cast = einsum(equation = var_38981_equation_0, values = (var_38823_cast, var_38740_cast))[name = tensor<string, []>("op_38981_cast")];
            tensor<fp16, []> var_38982_to_fp16 = const()[name = tensor<string, []>("op_38982_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3283_cast = mul(x = var_38981_cast, y = var_38982_to_fp16)[name = tensor<string, []>("aw_3283_cast")];
            tensor<string, []> var_38985_equation_0 = const()[name = tensor<string, []>("op_38985_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38985_cast = einsum(equation = var_38985_equation_0, values = (var_38827_cast, var_38744_cast))[name = tensor<string, []>("op_38985_cast")];
            tensor<fp16, []> var_38986_to_fp16 = const()[name = tensor<string, []>("op_38986_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3285_cast = mul(x = var_38985_cast, y = var_38986_to_fp16)[name = tensor<string, []>("aw_3285_cast")];
            tensor<string, []> var_38989_equation_0 = const()[name = tensor<string, []>("op_38989_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38989_cast = einsum(equation = var_38989_equation_0, values = (var_38831_cast, var_38748_cast))[name = tensor<string, []>("op_38989_cast")];
            tensor<fp16, []> var_38990_to_fp16 = const()[name = tensor<string, []>("op_38990_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3287_cast = mul(x = var_38989_cast, y = var_38990_to_fp16)[name = tensor<string, []>("aw_3287_cast")];
            tensor<string, []> var_38993_equation_0 = const()[name = tensor<string, []>("op_38993_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38993_cast = einsum(equation = var_38993_equation_0, values = (var_38835_cast, var_38752_cast))[name = tensor<string, []>("op_38993_cast")];
            tensor<fp16, []> var_38994_to_fp16 = const()[name = tensor<string, []>("op_38994_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3289_cast = mul(x = var_38993_cast, y = var_38994_to_fp16)[name = tensor<string, []>("aw_3289_cast")];
            tensor<string, []> var_38997_equation_0 = const()[name = tensor<string, []>("op_38997_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_38997_cast = einsum(equation = var_38997_equation_0, values = (var_38839_cast, var_38756_cast))[name = tensor<string, []>("op_38997_cast")];
            tensor<fp16, []> var_38998_to_fp16 = const()[name = tensor<string, []>("op_38998_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3291_cast = mul(x = var_38997_cast, y = var_38998_to_fp16)[name = tensor<string, []>("aw_3291_cast")];
            tensor<string, []> var_39001_equation_0 = const()[name = tensor<string, []>("op_39001_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39001_cast = einsum(equation = var_39001_equation_0, values = (var_38843_cast, var_38760_cast))[name = tensor<string, []>("op_39001_cast")];
            tensor<fp16, []> var_39002_to_fp16 = const()[name = tensor<string, []>("op_39002_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3293_cast = mul(x = var_39001_cast, y = var_39002_to_fp16)[name = tensor<string, []>("aw_3293_cast")];
            tensor<string, []> var_39005_equation_0 = const()[name = tensor<string, []>("op_39005_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39005_cast = einsum(equation = var_39005_equation_0, values = (var_38847_cast, var_38764_cast))[name = tensor<string, []>("op_39005_cast")];
            tensor<fp16, []> var_39006_to_fp16 = const()[name = tensor<string, []>("op_39006_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3295_cast = mul(x = var_39005_cast, y = var_39006_to_fp16)[name = tensor<string, []>("aw_3295_cast")];
            tensor<string, []> var_39009_equation_0 = const()[name = tensor<string, []>("op_39009_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39009_cast = einsum(equation = var_39009_equation_0, values = (var_38851_cast, var_38768_cast))[name = tensor<string, []>("op_39009_cast")];
            tensor<fp16, []> var_39010_to_fp16 = const()[name = tensor<string, []>("op_39010_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3297_cast = mul(x = var_39009_cast, y = var_39010_to_fp16)[name = tensor<string, []>("aw_3297_cast")];
            tensor<string, []> var_39013_equation_0 = const()[name = tensor<string, []>("op_39013_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39013_cast = einsum(equation = var_39013_equation_0, values = (var_38855_cast, var_38772_cast))[name = tensor<string, []>("op_39013_cast")];
            tensor<fp16, []> var_39014_to_fp16 = const()[name = tensor<string, []>("op_39014_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3299_cast = mul(x = var_39013_cast, y = var_39014_to_fp16)[name = tensor<string, []>("aw_3299_cast")];
            tensor<string, []> var_39017_equation_0 = const()[name = tensor<string, []>("op_39017_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39017_cast = einsum(equation = var_39017_equation_0, values = (var_38859_cast, var_38776_cast))[name = tensor<string, []>("op_39017_cast")];
            tensor<fp16, []> var_39018_to_fp16 = const()[name = tensor<string, []>("op_39018_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3301_cast = mul(x = var_39017_cast, y = var_39018_to_fp16)[name = tensor<string, []>("aw_3301_cast")];
            tensor<string, []> var_39021_equation_0 = const()[name = tensor<string, []>("op_39021_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39021_cast = einsum(equation = var_39021_equation_0, values = (var_38863_cast, var_38780_cast))[name = tensor<string, []>("op_39021_cast")];
            tensor<fp16, []> var_39022_to_fp16 = const()[name = tensor<string, []>("op_39022_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3303_cast = mul(x = var_39021_cast, y = var_39022_to_fp16)[name = tensor<string, []>("aw_3303_cast")];
            tensor<string, []> var_39025_equation_0 = const()[name = tensor<string, []>("op_39025_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39025_cast = einsum(equation = var_39025_equation_0, values = (var_38867_cast, var_38784_cast))[name = tensor<string, []>("op_39025_cast")];
            tensor<fp16, []> var_39026_to_fp16 = const()[name = tensor<string, []>("op_39026_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3305_cast = mul(x = var_39025_cast, y = var_39026_to_fp16)[name = tensor<string, []>("aw_3305_cast")];
            tensor<string, []> var_39029_equation_0 = const()[name = tensor<string, []>("op_39029_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39029_cast = einsum(equation = var_39029_equation_0, values = (var_38871_cast, var_38788_cast))[name = tensor<string, []>("op_39029_cast")];
            tensor<fp16, []> var_39030_to_fp16 = const()[name = tensor<string, []>("op_39030_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3307_cast = mul(x = var_39029_cast, y = var_39030_to_fp16)[name = tensor<string, []>("aw_3307_cast")];
            tensor<string, []> var_39033_equation_0 = const()[name = tensor<string, []>("op_39033_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39033_cast = einsum(equation = var_39033_equation_0, values = (var_38875_cast, var_38792_cast))[name = tensor<string, []>("op_39033_cast")];
            tensor<fp16, []> var_39034_to_fp16 = const()[name = tensor<string, []>("op_39034_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3309_cast = mul(x = var_39033_cast, y = var_39034_to_fp16)[name = tensor<string, []>("aw_3309_cast")];
            tensor<string, []> var_39037_equation_0 = const()[name = tensor<string, []>("op_39037_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39037_cast = einsum(equation = var_39037_equation_0, values = (var_38879_cast, var_38796_cast))[name = tensor<string, []>("op_39037_cast")];
            tensor<fp16, []> var_39038_to_fp16 = const()[name = tensor<string, []>("op_39038_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3311_cast = mul(x = var_39037_cast, y = var_39038_to_fp16)[name = tensor<string, []>("aw_3311_cast")];
            tensor<string, []> var_39041_equation_0 = const()[name = tensor<string, []>("op_39041_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39041_cast = einsum(equation = var_39041_equation_0, values = (var_38883_cast, var_38800_cast))[name = tensor<string, []>("op_39041_cast")];
            tensor<fp16, []> var_39042_to_fp16 = const()[name = tensor<string, []>("op_39042_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3313_cast = mul(x = var_39041_cast, y = var_39042_to_fp16)[name = tensor<string, []>("aw_3313_cast")];
            tensor<string, []> var_39045_equation_0 = const()[name = tensor<string, []>("op_39045_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39045_cast = einsum(equation = var_39045_equation_0, values = (var_38887_cast, var_38804_cast))[name = tensor<string, []>("op_39045_cast")];
            tensor<fp16, []> var_39046_to_fp16 = const()[name = tensor<string, []>("op_39046_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3315_cast = mul(x = var_39045_cast, y = var_39046_to_fp16)[name = tensor<string, []>("aw_3315_cast")];
            tensor<string, []> var_39049_equation_0 = const()[name = tensor<string, []>("op_39049_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39049_cast = einsum(equation = var_39049_equation_0, values = (var_38891_cast, var_38808_cast))[name = tensor<string, []>("op_39049_cast")];
            tensor<fp16, []> var_39050_to_fp16 = const()[name = tensor<string, []>("op_39050_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3317_cast = mul(x = var_39049_cast, y = var_39050_to_fp16)[name = tensor<string, []>("aw_3317_cast")];
            tensor<string, []> var_39053_equation_0 = const()[name = tensor<string, []>("op_39053_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39053_cast = einsum(equation = var_39053_equation_0, values = (var_38895_cast, var_38812_cast))[name = tensor<string, []>("op_39053_cast")];
            tensor<fp16, []> var_39054_to_fp16 = const()[name = tensor<string, []>("op_39054_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3319_cast = mul(x = var_39053_cast, y = var_39054_to_fp16)[name = tensor<string, []>("aw_3319_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39056_cast = softmax(axis = var_30385, x = aw_3281_cast)[name = tensor<string, []>("op_39056_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39057_cast = softmax(axis = var_30385, x = aw_3283_cast)[name = tensor<string, []>("op_39057_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39058_cast = softmax(axis = var_30385, x = aw_3285_cast)[name = tensor<string, []>("op_39058_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39059_cast = softmax(axis = var_30385, x = aw_3287_cast)[name = tensor<string, []>("op_39059_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39060_cast = softmax(axis = var_30385, x = aw_3289_cast)[name = tensor<string, []>("op_39060_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39061_cast = softmax(axis = var_30385, x = aw_3291_cast)[name = tensor<string, []>("op_39061_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39062_cast = softmax(axis = var_30385, x = aw_3293_cast)[name = tensor<string, []>("op_39062_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39063_cast = softmax(axis = var_30385, x = aw_3295_cast)[name = tensor<string, []>("op_39063_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39064_cast = softmax(axis = var_30385, x = aw_3297_cast)[name = tensor<string, []>("op_39064_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39065_cast = softmax(axis = var_30385, x = aw_3299_cast)[name = tensor<string, []>("op_39065_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39066_cast = softmax(axis = var_30385, x = aw_3301_cast)[name = tensor<string, []>("op_39066_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39067_cast = softmax(axis = var_30385, x = aw_3303_cast)[name = tensor<string, []>("op_39067_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39068_cast = softmax(axis = var_30385, x = aw_3305_cast)[name = tensor<string, []>("op_39068_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39069_cast = softmax(axis = var_30385, x = aw_3307_cast)[name = tensor<string, []>("op_39069_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39070_cast = softmax(axis = var_30385, x = aw_3309_cast)[name = tensor<string, []>("op_39070_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39071_cast = softmax(axis = var_30385, x = aw_3311_cast)[name = tensor<string, []>("op_39071_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39072_cast = softmax(axis = var_30385, x = aw_3313_cast)[name = tensor<string, []>("op_39072_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39073_cast = softmax(axis = var_30385, x = aw_3315_cast)[name = tensor<string, []>("op_39073_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39074_cast = softmax(axis = var_30385, x = aw_3317_cast)[name = tensor<string, []>("op_39074_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39075_cast = softmax(axis = var_30385, x = aw_3319_cast)[name = tensor<string, []>("op_39075_cast")];
            tensor<string, []> var_39077_equation_0 = const()[name = tensor<string, []>("op_39077_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39077_cast = einsum(equation = var_39077_equation_0, values = (var_38897_cast, var_39056_cast))[name = tensor<string, []>("op_39077_cast")];
            tensor<string, []> var_39079_equation_0 = const()[name = tensor<string, []>("op_39079_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39079_cast = einsum(equation = var_39079_equation_0, values = (var_38901_cast, var_39057_cast))[name = tensor<string, []>("op_39079_cast")];
            tensor<string, []> var_39081_equation_0 = const()[name = tensor<string, []>("op_39081_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39081_cast = einsum(equation = var_39081_equation_0, values = (var_38905_cast, var_39058_cast))[name = tensor<string, []>("op_39081_cast")];
            tensor<string, []> var_39083_equation_0 = const()[name = tensor<string, []>("op_39083_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39083_cast = einsum(equation = var_39083_equation_0, values = (var_38909_cast, var_39059_cast))[name = tensor<string, []>("op_39083_cast")];
            tensor<string, []> var_39085_equation_0 = const()[name = tensor<string, []>("op_39085_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39085_cast = einsum(equation = var_39085_equation_0, values = (var_38913_cast, var_39060_cast))[name = tensor<string, []>("op_39085_cast")];
            tensor<string, []> var_39087_equation_0 = const()[name = tensor<string, []>("op_39087_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39087_cast = einsum(equation = var_39087_equation_0, values = (var_38917_cast, var_39061_cast))[name = tensor<string, []>("op_39087_cast")];
            tensor<string, []> var_39089_equation_0 = const()[name = tensor<string, []>("op_39089_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39089_cast = einsum(equation = var_39089_equation_0, values = (var_38921_cast, var_39062_cast))[name = tensor<string, []>("op_39089_cast")];
            tensor<string, []> var_39091_equation_0 = const()[name = tensor<string, []>("op_39091_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39091_cast = einsum(equation = var_39091_equation_0, values = (var_38925_cast, var_39063_cast))[name = tensor<string, []>("op_39091_cast")];
            tensor<string, []> var_39093_equation_0 = const()[name = tensor<string, []>("op_39093_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39093_cast = einsum(equation = var_39093_equation_0, values = (var_38929_cast, var_39064_cast))[name = tensor<string, []>("op_39093_cast")];
            tensor<string, []> var_39095_equation_0 = const()[name = tensor<string, []>("op_39095_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39095_cast = einsum(equation = var_39095_equation_0, values = (var_38933_cast, var_39065_cast))[name = tensor<string, []>("op_39095_cast")];
            tensor<string, []> var_39097_equation_0 = const()[name = tensor<string, []>("op_39097_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39097_cast = einsum(equation = var_39097_equation_0, values = (var_38937_cast, var_39066_cast))[name = tensor<string, []>("op_39097_cast")];
            tensor<string, []> var_39099_equation_0 = const()[name = tensor<string, []>("op_39099_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39099_cast = einsum(equation = var_39099_equation_0, values = (var_38941_cast, var_39067_cast))[name = tensor<string, []>("op_39099_cast")];
            tensor<string, []> var_39101_equation_0 = const()[name = tensor<string, []>("op_39101_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39101_cast = einsum(equation = var_39101_equation_0, values = (var_38945_cast, var_39068_cast))[name = tensor<string, []>("op_39101_cast")];
            tensor<string, []> var_39103_equation_0 = const()[name = tensor<string, []>("op_39103_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39103_cast = einsum(equation = var_39103_equation_0, values = (var_38949_cast, var_39069_cast))[name = tensor<string, []>("op_39103_cast")];
            tensor<string, []> var_39105_equation_0 = const()[name = tensor<string, []>("op_39105_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39105_cast = einsum(equation = var_39105_equation_0, values = (var_38953_cast, var_39070_cast))[name = tensor<string, []>("op_39105_cast")];
            tensor<string, []> var_39107_equation_0 = const()[name = tensor<string, []>("op_39107_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39107_cast = einsum(equation = var_39107_equation_0, values = (var_38957_cast, var_39071_cast))[name = tensor<string, []>("op_39107_cast")];
            tensor<string, []> var_39109_equation_0 = const()[name = tensor<string, []>("op_39109_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39109_cast = einsum(equation = var_39109_equation_0, values = (var_38961_cast, var_39072_cast))[name = tensor<string, []>("op_39109_cast")];
            tensor<string, []> var_39111_equation_0 = const()[name = tensor<string, []>("op_39111_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39111_cast = einsum(equation = var_39111_equation_0, values = (var_38965_cast, var_39073_cast))[name = tensor<string, []>("op_39111_cast")];
            tensor<string, []> var_39113_equation_0 = const()[name = tensor<string, []>("op_39113_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39113_cast = einsum(equation = var_39113_equation_0, values = (var_38969_cast, var_39074_cast))[name = tensor<string, []>("op_39113_cast")];
            tensor<string, []> var_39115_equation_0 = const()[name = tensor<string, []>("op_39115_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39115_cast = einsum(equation = var_39115_equation_0, values = (var_38973_cast, var_39075_cast))[name = tensor<string, []>("op_39115_cast")];
            tensor<bool, []> input_513_interleave_0 = const()[name = tensor<string, []>("input_513_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_513_cast = concat(axis = var_30385, interleave = input_513_interleave_0, values = (var_39077_cast, var_39079_cast, var_39081_cast, var_39083_cast, var_39085_cast, var_39087_cast, var_39089_cast, var_39091_cast, var_39093_cast, var_39095_cast, var_39097_cast, var_39099_cast, var_39101_cast, var_39103_cast, var_39105_cast, var_39107_cast, var_39109_cast, var_39111_cast, var_39113_cast, var_39115_cast))[name = tensor<string, []>("input_513_cast")];
            tensor<int32, [2]> var_39121 = const()[name = tensor<string, []>("op_39121"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_39123 = const()[name = tensor<string, []>("op_39123"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_39125_pad_type_0 = const()[name = tensor<string, []>("op_39125_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_39125_pad_0 = const()[name = tensor<string, []>("op_39125_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3223873856)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3227150720)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_39125_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_out_0_bias_to_fp16, dilations = var_39123, groups = var_30385, pad = var_39125_pad_0, pad_type = var_39125_pad_type_0, strides = var_39121, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn1_to_out_0_weight_to_fp16, x = input_513_cast)[name = tensor<string, []>("op_39125_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_261_cast = add(x = var_39125_cast, y = inputs_259_cast)[name = tensor<string, []>("inputs_261_cast")];
            tensor<int32, [1]> var_39129 = const()[name = tensor<string, []>("op_39129"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_261_cast = reduce_mean(axes = var_39129, keep_dims = var_30380, x = inputs_261_cast)[name = tensor<string, []>("channels_mean_261_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_261_cast = sub(x = inputs_261_cast, y = channels_mean_261_cast)[name = tensor<string, []>("zero_mean_261_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_261_cast = mul(x = zero_mean_261_cast, y = zero_mean_261_cast)[name = tensor<string, []>("zero_mean_sq_261_cast")];
            tensor<int32, [1]> var_39133 = const()[name = tensor<string, []>("op_39133"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_39134_cast = reduce_mean(axes = var_39133, keep_dims = var_30380, x = zero_mean_sq_261_cast)[name = tensor<string, []>("op_39134_cast")];
            tensor<fp16, []> var_39135_to_fp16 = const()[name = tensor<string, []>("op_39135_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_39136_cast = add(x = var_39134_cast, y = var_39135_to_fp16)[name = tensor<string, []>("op_39136_cast")];
            tensor<fp16, []> denom_261_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_261_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_261_cast = rsqrt(epsilon = denom_261_epsilon_0_to_fp16, x = var_39136_cast)[name = tensor<string, []>("denom_261_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_261_cast = mul(x = zero_mean_261_cast, y = denom_261_cast)[name = tensor<string, []>("out_261_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_39140_to_fp16 = const()[name = tensor<string, []>("op_39140_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3227153344)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_39141_cast = add(x = out_261_cast, y = var_39140_to_fp16)[name = tensor<string, []>("op_39141_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_39143_to_fp16 = const()[name = tensor<string, []>("op_39143_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3227155968)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_343_cast = mul(x = var_39141_cast, y = var_39143_to_fp16)[name = tensor<string, []>("hidden_states_343_cast")];
            tensor<int32, [2]> var_39150 = const()[name = tensor<string, []>("op_39150"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_39152 = const()[name = tensor<string, []>("op_39152"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_175_pad_type_0 = const()[name = tensor<string, []>("q_175_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_175_pad_0 = const()[name = tensor<string, []>("q_175_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3227158592)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_175_cast = conv(dilations = var_39152, groups = var_30385, pad = q_175_pad_0, pad_type = q_175_pad_type_0, strides = var_39150, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_q_weight_to_fp16, x = hidden_states_343_cast)[name = tensor<string, []>("q_175_cast")];
            tensor<int32, [2]> var_39156 = const()[name = tensor<string, []>("op_39156"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_39158 = const()[name = tensor<string, []>("op_39158"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_349_pad_type_0 = const()[name = tensor<string, []>("k_349_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_349_pad_0 = const()[name = tensor<string, []>("k_349_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3230435456)))];
            tensor<fp16, [2, 1280, 1, 77]> k_349_cast = conv(dilations = var_39158, groups = var_30385, pad = k_349_pad_0, pad_type = k_349_pad_type_0, strides = var_39156, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_349_cast")];
            tensor<int32, [2]> var_39162 = const()[name = tensor<string, []>("op_39162"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_39164 = const()[name = tensor<string, []>("op_39164"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_175_pad_type_0 = const()[name = tensor<string, []>("v_175_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_175_pad_0 = const()[name = tensor<string, []>("v_175_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3235678400)))];
            tensor<fp16, [2, 1280, 1, 77]> v_175_cast = conv(dilations = var_39164, groups = var_30385, pad = v_175_pad_0, pad_type = v_175_pad_type_0, strides = var_39162, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_175_cast")];
            tensor<int32, [4]> var_39168_begin_0 = const()[name = tensor<string, []>("op_39168_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_39168_end_0 = const()[name = tensor<string, []>("op_39168_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_39168_end_mask_0 = const()[name = tensor<string, []>("op_39168_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39168_cast = slice_by_index(begin = var_39168_begin_0, end = var_39168_end_0, end_mask = var_39168_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39168_cast")];
            tensor<int32, [4]> var_39172_begin_0 = const()[name = tensor<string, []>("op_39172_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_39172_end_0 = const()[name = tensor<string, []>("op_39172_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_39172_end_mask_0 = const()[name = tensor<string, []>("op_39172_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39172_cast = slice_by_index(begin = var_39172_begin_0, end = var_39172_end_0, end_mask = var_39172_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39172_cast")];
            tensor<int32, [4]> var_39176_begin_0 = const()[name = tensor<string, []>("op_39176_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_39176_end_0 = const()[name = tensor<string, []>("op_39176_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_39176_end_mask_0 = const()[name = tensor<string, []>("op_39176_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39176_cast = slice_by_index(begin = var_39176_begin_0, end = var_39176_end_0, end_mask = var_39176_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39176_cast")];
            tensor<int32, [4]> var_39180_begin_0 = const()[name = tensor<string, []>("op_39180_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_39180_end_0 = const()[name = tensor<string, []>("op_39180_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_39180_end_mask_0 = const()[name = tensor<string, []>("op_39180_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39180_cast = slice_by_index(begin = var_39180_begin_0, end = var_39180_end_0, end_mask = var_39180_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39180_cast")];
            tensor<int32, [4]> var_39184_begin_0 = const()[name = tensor<string, []>("op_39184_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_39184_end_0 = const()[name = tensor<string, []>("op_39184_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_39184_end_mask_0 = const()[name = tensor<string, []>("op_39184_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39184_cast = slice_by_index(begin = var_39184_begin_0, end = var_39184_end_0, end_mask = var_39184_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39184_cast")];
            tensor<int32, [4]> var_39188_begin_0 = const()[name = tensor<string, []>("op_39188_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_39188_end_0 = const()[name = tensor<string, []>("op_39188_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_39188_end_mask_0 = const()[name = tensor<string, []>("op_39188_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39188_cast = slice_by_index(begin = var_39188_begin_0, end = var_39188_end_0, end_mask = var_39188_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39188_cast")];
            tensor<int32, [4]> var_39192_begin_0 = const()[name = tensor<string, []>("op_39192_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_39192_end_0 = const()[name = tensor<string, []>("op_39192_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_39192_end_mask_0 = const()[name = tensor<string, []>("op_39192_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39192_cast = slice_by_index(begin = var_39192_begin_0, end = var_39192_end_0, end_mask = var_39192_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39192_cast")];
            tensor<int32, [4]> var_39196_begin_0 = const()[name = tensor<string, []>("op_39196_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_39196_end_0 = const()[name = tensor<string, []>("op_39196_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_39196_end_mask_0 = const()[name = tensor<string, []>("op_39196_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39196_cast = slice_by_index(begin = var_39196_begin_0, end = var_39196_end_0, end_mask = var_39196_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39196_cast")];
            tensor<int32, [4]> var_39200_begin_0 = const()[name = tensor<string, []>("op_39200_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_39200_end_0 = const()[name = tensor<string, []>("op_39200_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_39200_end_mask_0 = const()[name = tensor<string, []>("op_39200_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39200_cast = slice_by_index(begin = var_39200_begin_0, end = var_39200_end_0, end_mask = var_39200_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39200_cast")];
            tensor<int32, [4]> var_39204_begin_0 = const()[name = tensor<string, []>("op_39204_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_39204_end_0 = const()[name = tensor<string, []>("op_39204_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_39204_end_mask_0 = const()[name = tensor<string, []>("op_39204_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39204_cast = slice_by_index(begin = var_39204_begin_0, end = var_39204_end_0, end_mask = var_39204_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39204_cast")];
            tensor<int32, [4]> var_39208_begin_0 = const()[name = tensor<string, []>("op_39208_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_39208_end_0 = const()[name = tensor<string, []>("op_39208_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_39208_end_mask_0 = const()[name = tensor<string, []>("op_39208_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39208_cast = slice_by_index(begin = var_39208_begin_0, end = var_39208_end_0, end_mask = var_39208_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39208_cast")];
            tensor<int32, [4]> var_39212_begin_0 = const()[name = tensor<string, []>("op_39212_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_39212_end_0 = const()[name = tensor<string, []>("op_39212_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_39212_end_mask_0 = const()[name = tensor<string, []>("op_39212_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39212_cast = slice_by_index(begin = var_39212_begin_0, end = var_39212_end_0, end_mask = var_39212_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39212_cast")];
            tensor<int32, [4]> var_39216_begin_0 = const()[name = tensor<string, []>("op_39216_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_39216_end_0 = const()[name = tensor<string, []>("op_39216_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_39216_end_mask_0 = const()[name = tensor<string, []>("op_39216_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39216_cast = slice_by_index(begin = var_39216_begin_0, end = var_39216_end_0, end_mask = var_39216_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39216_cast")];
            tensor<int32, [4]> var_39220_begin_0 = const()[name = tensor<string, []>("op_39220_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_39220_end_0 = const()[name = tensor<string, []>("op_39220_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_39220_end_mask_0 = const()[name = tensor<string, []>("op_39220_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39220_cast = slice_by_index(begin = var_39220_begin_0, end = var_39220_end_0, end_mask = var_39220_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39220_cast")];
            tensor<int32, [4]> var_39224_begin_0 = const()[name = tensor<string, []>("op_39224_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_39224_end_0 = const()[name = tensor<string, []>("op_39224_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_39224_end_mask_0 = const()[name = tensor<string, []>("op_39224_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39224_cast = slice_by_index(begin = var_39224_begin_0, end = var_39224_end_0, end_mask = var_39224_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39224_cast")];
            tensor<int32, [4]> var_39228_begin_0 = const()[name = tensor<string, []>("op_39228_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_39228_end_0 = const()[name = tensor<string, []>("op_39228_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_39228_end_mask_0 = const()[name = tensor<string, []>("op_39228_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39228_cast = slice_by_index(begin = var_39228_begin_0, end = var_39228_end_0, end_mask = var_39228_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39228_cast")];
            tensor<int32, [4]> var_39232_begin_0 = const()[name = tensor<string, []>("op_39232_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_39232_end_0 = const()[name = tensor<string, []>("op_39232_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_39232_end_mask_0 = const()[name = tensor<string, []>("op_39232_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39232_cast = slice_by_index(begin = var_39232_begin_0, end = var_39232_end_0, end_mask = var_39232_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39232_cast")];
            tensor<int32, [4]> var_39236_begin_0 = const()[name = tensor<string, []>("op_39236_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_39236_end_0 = const()[name = tensor<string, []>("op_39236_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_39236_end_mask_0 = const()[name = tensor<string, []>("op_39236_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39236_cast = slice_by_index(begin = var_39236_begin_0, end = var_39236_end_0, end_mask = var_39236_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39236_cast")];
            tensor<int32, [4]> var_39240_begin_0 = const()[name = tensor<string, []>("op_39240_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_39240_end_0 = const()[name = tensor<string, []>("op_39240_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_39240_end_mask_0 = const()[name = tensor<string, []>("op_39240_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39240_cast = slice_by_index(begin = var_39240_begin_0, end = var_39240_end_0, end_mask = var_39240_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39240_cast")];
            tensor<int32, [4]> var_39244_begin_0 = const()[name = tensor<string, []>("op_39244_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_39244_end_0 = const()[name = tensor<string, []>("op_39244_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_39244_end_mask_0 = const()[name = tensor<string, []>("op_39244_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39244_cast = slice_by_index(begin = var_39244_begin_0, end = var_39244_end_0, end_mask = var_39244_end_mask_0, x = q_175_cast)[name = tensor<string, []>("op_39244_cast")];
            tensor<int32, [4]> k_351_perm_0 = const()[name = tensor<string, []>("k_351_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_39251_begin_0 = const()[name = tensor<string, []>("op_39251_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_39251_end_0 = const()[name = tensor<string, []>("op_39251_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_39251_end_mask_0 = const()[name = tensor<string, []>("op_39251_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_52 = transpose(perm = k_351_perm_0, x = k_349_cast)[name = tensor<string, []>("transpose_52")];
            tensor<fp16, [2, 77, 1, 64]> var_39251_cast = slice_by_index(begin = var_39251_begin_0, end = var_39251_end_0, end_mask = var_39251_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39251_cast")];
            tensor<int32, [4]> var_39255_begin_0 = const()[name = tensor<string, []>("op_39255_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_39255_end_0 = const()[name = tensor<string, []>("op_39255_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_39255_end_mask_0 = const()[name = tensor<string, []>("op_39255_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39255_cast = slice_by_index(begin = var_39255_begin_0, end = var_39255_end_0, end_mask = var_39255_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39255_cast")];
            tensor<int32, [4]> var_39259_begin_0 = const()[name = tensor<string, []>("op_39259_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_39259_end_0 = const()[name = tensor<string, []>("op_39259_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_39259_end_mask_0 = const()[name = tensor<string, []>("op_39259_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39259_cast = slice_by_index(begin = var_39259_begin_0, end = var_39259_end_0, end_mask = var_39259_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39259_cast")];
            tensor<int32, [4]> var_39263_begin_0 = const()[name = tensor<string, []>("op_39263_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_39263_end_0 = const()[name = tensor<string, []>("op_39263_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_39263_end_mask_0 = const()[name = tensor<string, []>("op_39263_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39263_cast = slice_by_index(begin = var_39263_begin_0, end = var_39263_end_0, end_mask = var_39263_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39263_cast")];
            tensor<int32, [4]> var_39267_begin_0 = const()[name = tensor<string, []>("op_39267_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_39267_end_0 = const()[name = tensor<string, []>("op_39267_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_39267_end_mask_0 = const()[name = tensor<string, []>("op_39267_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39267_cast = slice_by_index(begin = var_39267_begin_0, end = var_39267_end_0, end_mask = var_39267_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39267_cast")];
            tensor<int32, [4]> var_39271_begin_0 = const()[name = tensor<string, []>("op_39271_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_39271_end_0 = const()[name = tensor<string, []>("op_39271_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_39271_end_mask_0 = const()[name = tensor<string, []>("op_39271_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39271_cast = slice_by_index(begin = var_39271_begin_0, end = var_39271_end_0, end_mask = var_39271_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39271_cast")];
            tensor<int32, [4]> var_39275_begin_0 = const()[name = tensor<string, []>("op_39275_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_39275_end_0 = const()[name = tensor<string, []>("op_39275_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_39275_end_mask_0 = const()[name = tensor<string, []>("op_39275_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39275_cast = slice_by_index(begin = var_39275_begin_0, end = var_39275_end_0, end_mask = var_39275_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39275_cast")];
            tensor<int32, [4]> var_39279_begin_0 = const()[name = tensor<string, []>("op_39279_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_39279_end_0 = const()[name = tensor<string, []>("op_39279_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_39279_end_mask_0 = const()[name = tensor<string, []>("op_39279_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39279_cast = slice_by_index(begin = var_39279_begin_0, end = var_39279_end_0, end_mask = var_39279_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39279_cast")];
            tensor<int32, [4]> var_39283_begin_0 = const()[name = tensor<string, []>("op_39283_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_39283_end_0 = const()[name = tensor<string, []>("op_39283_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_39283_end_mask_0 = const()[name = tensor<string, []>("op_39283_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39283_cast = slice_by_index(begin = var_39283_begin_0, end = var_39283_end_0, end_mask = var_39283_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39283_cast")];
            tensor<int32, [4]> var_39287_begin_0 = const()[name = tensor<string, []>("op_39287_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_39287_end_0 = const()[name = tensor<string, []>("op_39287_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_39287_end_mask_0 = const()[name = tensor<string, []>("op_39287_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39287_cast = slice_by_index(begin = var_39287_begin_0, end = var_39287_end_0, end_mask = var_39287_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39287_cast")];
            tensor<int32, [4]> var_39291_begin_0 = const()[name = tensor<string, []>("op_39291_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_39291_end_0 = const()[name = tensor<string, []>("op_39291_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_39291_end_mask_0 = const()[name = tensor<string, []>("op_39291_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39291_cast = slice_by_index(begin = var_39291_begin_0, end = var_39291_end_0, end_mask = var_39291_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39291_cast")];
            tensor<int32, [4]> var_39295_begin_0 = const()[name = tensor<string, []>("op_39295_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_39295_end_0 = const()[name = tensor<string, []>("op_39295_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_39295_end_mask_0 = const()[name = tensor<string, []>("op_39295_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39295_cast = slice_by_index(begin = var_39295_begin_0, end = var_39295_end_0, end_mask = var_39295_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39295_cast")];
            tensor<int32, [4]> var_39299_begin_0 = const()[name = tensor<string, []>("op_39299_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_39299_end_0 = const()[name = tensor<string, []>("op_39299_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_39299_end_mask_0 = const()[name = tensor<string, []>("op_39299_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39299_cast = slice_by_index(begin = var_39299_begin_0, end = var_39299_end_0, end_mask = var_39299_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39299_cast")];
            tensor<int32, [4]> var_39303_begin_0 = const()[name = tensor<string, []>("op_39303_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_39303_end_0 = const()[name = tensor<string, []>("op_39303_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_39303_end_mask_0 = const()[name = tensor<string, []>("op_39303_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39303_cast = slice_by_index(begin = var_39303_begin_0, end = var_39303_end_0, end_mask = var_39303_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39303_cast")];
            tensor<int32, [4]> var_39307_begin_0 = const()[name = tensor<string, []>("op_39307_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_39307_end_0 = const()[name = tensor<string, []>("op_39307_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_39307_end_mask_0 = const()[name = tensor<string, []>("op_39307_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39307_cast = slice_by_index(begin = var_39307_begin_0, end = var_39307_end_0, end_mask = var_39307_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39307_cast")];
            tensor<int32, [4]> var_39311_begin_0 = const()[name = tensor<string, []>("op_39311_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_39311_end_0 = const()[name = tensor<string, []>("op_39311_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_39311_end_mask_0 = const()[name = tensor<string, []>("op_39311_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39311_cast = slice_by_index(begin = var_39311_begin_0, end = var_39311_end_0, end_mask = var_39311_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39311_cast")];
            tensor<int32, [4]> var_39315_begin_0 = const()[name = tensor<string, []>("op_39315_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_39315_end_0 = const()[name = tensor<string, []>("op_39315_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_39315_end_mask_0 = const()[name = tensor<string, []>("op_39315_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39315_cast = slice_by_index(begin = var_39315_begin_0, end = var_39315_end_0, end_mask = var_39315_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39315_cast")];
            tensor<int32, [4]> var_39319_begin_0 = const()[name = tensor<string, []>("op_39319_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_39319_end_0 = const()[name = tensor<string, []>("op_39319_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_39319_end_mask_0 = const()[name = tensor<string, []>("op_39319_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39319_cast = slice_by_index(begin = var_39319_begin_0, end = var_39319_end_0, end_mask = var_39319_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39319_cast")];
            tensor<int32, [4]> var_39323_begin_0 = const()[name = tensor<string, []>("op_39323_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_39323_end_0 = const()[name = tensor<string, []>("op_39323_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_39323_end_mask_0 = const()[name = tensor<string, []>("op_39323_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39323_cast = slice_by_index(begin = var_39323_begin_0, end = var_39323_end_0, end_mask = var_39323_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39323_cast")];
            tensor<int32, [4]> var_39327_begin_0 = const()[name = tensor<string, []>("op_39327_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_39327_end_0 = const()[name = tensor<string, []>("op_39327_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_39327_end_mask_0 = const()[name = tensor<string, []>("op_39327_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_39327_cast = slice_by_index(begin = var_39327_begin_0, end = var_39327_end_0, end_mask = var_39327_end_mask_0, x = transpose_52)[name = tensor<string, []>("op_39327_cast")];
            tensor<int32, [4]> var_39329_begin_0 = const()[name = tensor<string, []>("op_39329_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_39329_end_0 = const()[name = tensor<string, []>("op_39329_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_39329_end_mask_0 = const()[name = tensor<string, []>("op_39329_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39329_cast = slice_by_index(begin = var_39329_begin_0, end = var_39329_end_0, end_mask = var_39329_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39329_cast")];
            tensor<int32, [4]> var_39333_begin_0 = const()[name = tensor<string, []>("op_39333_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_39333_end_0 = const()[name = tensor<string, []>("op_39333_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_39333_end_mask_0 = const()[name = tensor<string, []>("op_39333_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39333_cast = slice_by_index(begin = var_39333_begin_0, end = var_39333_end_0, end_mask = var_39333_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39333_cast")];
            tensor<int32, [4]> var_39337_begin_0 = const()[name = tensor<string, []>("op_39337_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_39337_end_0 = const()[name = tensor<string, []>("op_39337_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_39337_end_mask_0 = const()[name = tensor<string, []>("op_39337_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39337_cast = slice_by_index(begin = var_39337_begin_0, end = var_39337_end_0, end_mask = var_39337_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39337_cast")];
            tensor<int32, [4]> var_39341_begin_0 = const()[name = tensor<string, []>("op_39341_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_39341_end_0 = const()[name = tensor<string, []>("op_39341_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_39341_end_mask_0 = const()[name = tensor<string, []>("op_39341_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39341_cast = slice_by_index(begin = var_39341_begin_0, end = var_39341_end_0, end_mask = var_39341_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39341_cast")];
            tensor<int32, [4]> var_39345_begin_0 = const()[name = tensor<string, []>("op_39345_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_39345_end_0 = const()[name = tensor<string, []>("op_39345_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_39345_end_mask_0 = const()[name = tensor<string, []>("op_39345_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39345_cast = slice_by_index(begin = var_39345_begin_0, end = var_39345_end_0, end_mask = var_39345_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39345_cast")];
            tensor<int32, [4]> var_39349_begin_0 = const()[name = tensor<string, []>("op_39349_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_39349_end_0 = const()[name = tensor<string, []>("op_39349_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_39349_end_mask_0 = const()[name = tensor<string, []>("op_39349_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39349_cast = slice_by_index(begin = var_39349_begin_0, end = var_39349_end_0, end_mask = var_39349_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39349_cast")];
            tensor<int32, [4]> var_39353_begin_0 = const()[name = tensor<string, []>("op_39353_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_39353_end_0 = const()[name = tensor<string, []>("op_39353_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_39353_end_mask_0 = const()[name = tensor<string, []>("op_39353_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39353_cast = slice_by_index(begin = var_39353_begin_0, end = var_39353_end_0, end_mask = var_39353_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39353_cast")];
            tensor<int32, [4]> var_39357_begin_0 = const()[name = tensor<string, []>("op_39357_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_39357_end_0 = const()[name = tensor<string, []>("op_39357_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_39357_end_mask_0 = const()[name = tensor<string, []>("op_39357_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39357_cast = slice_by_index(begin = var_39357_begin_0, end = var_39357_end_0, end_mask = var_39357_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39357_cast")];
            tensor<int32, [4]> var_39361_begin_0 = const()[name = tensor<string, []>("op_39361_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_39361_end_0 = const()[name = tensor<string, []>("op_39361_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_39361_end_mask_0 = const()[name = tensor<string, []>("op_39361_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39361_cast = slice_by_index(begin = var_39361_begin_0, end = var_39361_end_0, end_mask = var_39361_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39361_cast")];
            tensor<int32, [4]> var_39365_begin_0 = const()[name = tensor<string, []>("op_39365_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_39365_end_0 = const()[name = tensor<string, []>("op_39365_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_39365_end_mask_0 = const()[name = tensor<string, []>("op_39365_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39365_cast = slice_by_index(begin = var_39365_begin_0, end = var_39365_end_0, end_mask = var_39365_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39365_cast")];
            tensor<int32, [4]> var_39369_begin_0 = const()[name = tensor<string, []>("op_39369_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_39369_end_0 = const()[name = tensor<string, []>("op_39369_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_39369_end_mask_0 = const()[name = tensor<string, []>("op_39369_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39369_cast = slice_by_index(begin = var_39369_begin_0, end = var_39369_end_0, end_mask = var_39369_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39369_cast")];
            tensor<int32, [4]> var_39373_begin_0 = const()[name = tensor<string, []>("op_39373_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_39373_end_0 = const()[name = tensor<string, []>("op_39373_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_39373_end_mask_0 = const()[name = tensor<string, []>("op_39373_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39373_cast = slice_by_index(begin = var_39373_begin_0, end = var_39373_end_0, end_mask = var_39373_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39373_cast")];
            tensor<int32, [4]> var_39377_begin_0 = const()[name = tensor<string, []>("op_39377_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_39377_end_0 = const()[name = tensor<string, []>("op_39377_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_39377_end_mask_0 = const()[name = tensor<string, []>("op_39377_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39377_cast = slice_by_index(begin = var_39377_begin_0, end = var_39377_end_0, end_mask = var_39377_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39377_cast")];
            tensor<int32, [4]> var_39381_begin_0 = const()[name = tensor<string, []>("op_39381_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_39381_end_0 = const()[name = tensor<string, []>("op_39381_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_39381_end_mask_0 = const()[name = tensor<string, []>("op_39381_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39381_cast = slice_by_index(begin = var_39381_begin_0, end = var_39381_end_0, end_mask = var_39381_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39381_cast")];
            tensor<int32, [4]> var_39385_begin_0 = const()[name = tensor<string, []>("op_39385_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_39385_end_0 = const()[name = tensor<string, []>("op_39385_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_39385_end_mask_0 = const()[name = tensor<string, []>("op_39385_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39385_cast = slice_by_index(begin = var_39385_begin_0, end = var_39385_end_0, end_mask = var_39385_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39385_cast")];
            tensor<int32, [4]> var_39389_begin_0 = const()[name = tensor<string, []>("op_39389_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_39389_end_0 = const()[name = tensor<string, []>("op_39389_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_39389_end_mask_0 = const()[name = tensor<string, []>("op_39389_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39389_cast = slice_by_index(begin = var_39389_begin_0, end = var_39389_end_0, end_mask = var_39389_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39389_cast")];
            tensor<int32, [4]> var_39393_begin_0 = const()[name = tensor<string, []>("op_39393_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_39393_end_0 = const()[name = tensor<string, []>("op_39393_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_39393_end_mask_0 = const()[name = tensor<string, []>("op_39393_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39393_cast = slice_by_index(begin = var_39393_begin_0, end = var_39393_end_0, end_mask = var_39393_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39393_cast")];
            tensor<int32, [4]> var_39397_begin_0 = const()[name = tensor<string, []>("op_39397_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_39397_end_0 = const()[name = tensor<string, []>("op_39397_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_39397_end_mask_0 = const()[name = tensor<string, []>("op_39397_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39397_cast = slice_by_index(begin = var_39397_begin_0, end = var_39397_end_0, end_mask = var_39397_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39397_cast")];
            tensor<int32, [4]> var_39401_begin_0 = const()[name = tensor<string, []>("op_39401_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_39401_end_0 = const()[name = tensor<string, []>("op_39401_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_39401_end_mask_0 = const()[name = tensor<string, []>("op_39401_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39401_cast = slice_by_index(begin = var_39401_begin_0, end = var_39401_end_0, end_mask = var_39401_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39401_cast")];
            tensor<int32, [4]> var_39405_begin_0 = const()[name = tensor<string, []>("op_39405_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_39405_end_0 = const()[name = tensor<string, []>("op_39405_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_39405_end_mask_0 = const()[name = tensor<string, []>("op_39405_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_39405_cast = slice_by_index(begin = var_39405_begin_0, end = var_39405_end_0, end_mask = var_39405_end_mask_0, x = v_175_cast)[name = tensor<string, []>("op_39405_cast")];
            tensor<string, []> var_39409_equation_0 = const()[name = tensor<string, []>("op_39409_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39409_cast = einsum(equation = var_39409_equation_0, values = (var_39251_cast, var_39168_cast))[name = tensor<string, []>("op_39409_cast")];
            tensor<fp16, []> var_39410_to_fp16 = const()[name = tensor<string, []>("op_39410_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3321_cast = mul(x = var_39409_cast, y = var_39410_to_fp16)[name = tensor<string, []>("aw_3321_cast")];
            tensor<string, []> var_39413_equation_0 = const()[name = tensor<string, []>("op_39413_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39413_cast = einsum(equation = var_39413_equation_0, values = (var_39255_cast, var_39172_cast))[name = tensor<string, []>("op_39413_cast")];
            tensor<fp16, []> var_39414_to_fp16 = const()[name = tensor<string, []>("op_39414_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3323_cast = mul(x = var_39413_cast, y = var_39414_to_fp16)[name = tensor<string, []>("aw_3323_cast")];
            tensor<string, []> var_39417_equation_0 = const()[name = tensor<string, []>("op_39417_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39417_cast = einsum(equation = var_39417_equation_0, values = (var_39259_cast, var_39176_cast))[name = tensor<string, []>("op_39417_cast")];
            tensor<fp16, []> var_39418_to_fp16 = const()[name = tensor<string, []>("op_39418_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3325_cast = mul(x = var_39417_cast, y = var_39418_to_fp16)[name = tensor<string, []>("aw_3325_cast")];
            tensor<string, []> var_39421_equation_0 = const()[name = tensor<string, []>("op_39421_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39421_cast = einsum(equation = var_39421_equation_0, values = (var_39263_cast, var_39180_cast))[name = tensor<string, []>("op_39421_cast")];
            tensor<fp16, []> var_39422_to_fp16 = const()[name = tensor<string, []>("op_39422_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3327_cast = mul(x = var_39421_cast, y = var_39422_to_fp16)[name = tensor<string, []>("aw_3327_cast")];
            tensor<string, []> var_39425_equation_0 = const()[name = tensor<string, []>("op_39425_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39425_cast = einsum(equation = var_39425_equation_0, values = (var_39267_cast, var_39184_cast))[name = tensor<string, []>("op_39425_cast")];
            tensor<fp16, []> var_39426_to_fp16 = const()[name = tensor<string, []>("op_39426_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3329_cast = mul(x = var_39425_cast, y = var_39426_to_fp16)[name = tensor<string, []>("aw_3329_cast")];
            tensor<string, []> var_39429_equation_0 = const()[name = tensor<string, []>("op_39429_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39429_cast = einsum(equation = var_39429_equation_0, values = (var_39271_cast, var_39188_cast))[name = tensor<string, []>("op_39429_cast")];
            tensor<fp16, []> var_39430_to_fp16 = const()[name = tensor<string, []>("op_39430_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3331_cast = mul(x = var_39429_cast, y = var_39430_to_fp16)[name = tensor<string, []>("aw_3331_cast")];
            tensor<string, []> var_39433_equation_0 = const()[name = tensor<string, []>("op_39433_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39433_cast = einsum(equation = var_39433_equation_0, values = (var_39275_cast, var_39192_cast))[name = tensor<string, []>("op_39433_cast")];
            tensor<fp16, []> var_39434_to_fp16 = const()[name = tensor<string, []>("op_39434_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3333_cast = mul(x = var_39433_cast, y = var_39434_to_fp16)[name = tensor<string, []>("aw_3333_cast")];
            tensor<string, []> var_39437_equation_0 = const()[name = tensor<string, []>("op_39437_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39437_cast = einsum(equation = var_39437_equation_0, values = (var_39279_cast, var_39196_cast))[name = tensor<string, []>("op_39437_cast")];
            tensor<fp16, []> var_39438_to_fp16 = const()[name = tensor<string, []>("op_39438_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3335_cast = mul(x = var_39437_cast, y = var_39438_to_fp16)[name = tensor<string, []>("aw_3335_cast")];
            tensor<string, []> var_39441_equation_0 = const()[name = tensor<string, []>("op_39441_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39441_cast = einsum(equation = var_39441_equation_0, values = (var_39283_cast, var_39200_cast))[name = tensor<string, []>("op_39441_cast")];
            tensor<fp16, []> var_39442_to_fp16 = const()[name = tensor<string, []>("op_39442_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3337_cast = mul(x = var_39441_cast, y = var_39442_to_fp16)[name = tensor<string, []>("aw_3337_cast")];
            tensor<string, []> var_39445_equation_0 = const()[name = tensor<string, []>("op_39445_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39445_cast = einsum(equation = var_39445_equation_0, values = (var_39287_cast, var_39204_cast))[name = tensor<string, []>("op_39445_cast")];
            tensor<fp16, []> var_39446_to_fp16 = const()[name = tensor<string, []>("op_39446_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3339_cast = mul(x = var_39445_cast, y = var_39446_to_fp16)[name = tensor<string, []>("aw_3339_cast")];
            tensor<string, []> var_39449_equation_0 = const()[name = tensor<string, []>("op_39449_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39449_cast = einsum(equation = var_39449_equation_0, values = (var_39291_cast, var_39208_cast))[name = tensor<string, []>("op_39449_cast")];
            tensor<fp16, []> var_39450_to_fp16 = const()[name = tensor<string, []>("op_39450_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3341_cast = mul(x = var_39449_cast, y = var_39450_to_fp16)[name = tensor<string, []>("aw_3341_cast")];
            tensor<string, []> var_39453_equation_0 = const()[name = tensor<string, []>("op_39453_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39453_cast = einsum(equation = var_39453_equation_0, values = (var_39295_cast, var_39212_cast))[name = tensor<string, []>("op_39453_cast")];
            tensor<fp16, []> var_39454_to_fp16 = const()[name = tensor<string, []>("op_39454_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3343_cast = mul(x = var_39453_cast, y = var_39454_to_fp16)[name = tensor<string, []>("aw_3343_cast")];
            tensor<string, []> var_39457_equation_0 = const()[name = tensor<string, []>("op_39457_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39457_cast = einsum(equation = var_39457_equation_0, values = (var_39299_cast, var_39216_cast))[name = tensor<string, []>("op_39457_cast")];
            tensor<fp16, []> var_39458_to_fp16 = const()[name = tensor<string, []>("op_39458_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3345_cast = mul(x = var_39457_cast, y = var_39458_to_fp16)[name = tensor<string, []>("aw_3345_cast")];
            tensor<string, []> var_39461_equation_0 = const()[name = tensor<string, []>("op_39461_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39461_cast = einsum(equation = var_39461_equation_0, values = (var_39303_cast, var_39220_cast))[name = tensor<string, []>("op_39461_cast")];
            tensor<fp16, []> var_39462_to_fp16 = const()[name = tensor<string, []>("op_39462_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3347_cast = mul(x = var_39461_cast, y = var_39462_to_fp16)[name = tensor<string, []>("aw_3347_cast")];
            tensor<string, []> var_39465_equation_0 = const()[name = tensor<string, []>("op_39465_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39465_cast = einsum(equation = var_39465_equation_0, values = (var_39307_cast, var_39224_cast))[name = tensor<string, []>("op_39465_cast")];
            tensor<fp16, []> var_39466_to_fp16 = const()[name = tensor<string, []>("op_39466_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3349_cast = mul(x = var_39465_cast, y = var_39466_to_fp16)[name = tensor<string, []>("aw_3349_cast")];
            tensor<string, []> var_39469_equation_0 = const()[name = tensor<string, []>("op_39469_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39469_cast = einsum(equation = var_39469_equation_0, values = (var_39311_cast, var_39228_cast))[name = tensor<string, []>("op_39469_cast")];
            tensor<fp16, []> var_39470_to_fp16 = const()[name = tensor<string, []>("op_39470_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3351_cast = mul(x = var_39469_cast, y = var_39470_to_fp16)[name = tensor<string, []>("aw_3351_cast")];
            tensor<string, []> var_39473_equation_0 = const()[name = tensor<string, []>("op_39473_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39473_cast = einsum(equation = var_39473_equation_0, values = (var_39315_cast, var_39232_cast))[name = tensor<string, []>("op_39473_cast")];
            tensor<fp16, []> var_39474_to_fp16 = const()[name = tensor<string, []>("op_39474_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3353_cast = mul(x = var_39473_cast, y = var_39474_to_fp16)[name = tensor<string, []>("aw_3353_cast")];
            tensor<string, []> var_39477_equation_0 = const()[name = tensor<string, []>("op_39477_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39477_cast = einsum(equation = var_39477_equation_0, values = (var_39319_cast, var_39236_cast))[name = tensor<string, []>("op_39477_cast")];
            tensor<fp16, []> var_39478_to_fp16 = const()[name = tensor<string, []>("op_39478_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3355_cast = mul(x = var_39477_cast, y = var_39478_to_fp16)[name = tensor<string, []>("aw_3355_cast")];
            tensor<string, []> var_39481_equation_0 = const()[name = tensor<string, []>("op_39481_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39481_cast = einsum(equation = var_39481_equation_0, values = (var_39323_cast, var_39240_cast))[name = tensor<string, []>("op_39481_cast")];
            tensor<fp16, []> var_39482_to_fp16 = const()[name = tensor<string, []>("op_39482_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3357_cast = mul(x = var_39481_cast, y = var_39482_to_fp16)[name = tensor<string, []>("aw_3357_cast")];
            tensor<string, []> var_39485_equation_0 = const()[name = tensor<string, []>("op_39485_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_39485_cast = einsum(equation = var_39485_equation_0, values = (var_39327_cast, var_39244_cast))[name = tensor<string, []>("op_39485_cast")];
            tensor<fp16, []> var_39486_to_fp16 = const()[name = tensor<string, []>("op_39486_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3359_cast = mul(x = var_39485_cast, y = var_39486_to_fp16)[name = tensor<string, []>("aw_3359_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39488_cast = softmax(axis = var_30385, x = aw_3321_cast)[name = tensor<string, []>("op_39488_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39489_cast = softmax(axis = var_30385, x = aw_3323_cast)[name = tensor<string, []>("op_39489_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39490_cast = softmax(axis = var_30385, x = aw_3325_cast)[name = tensor<string, []>("op_39490_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39491_cast = softmax(axis = var_30385, x = aw_3327_cast)[name = tensor<string, []>("op_39491_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39492_cast = softmax(axis = var_30385, x = aw_3329_cast)[name = tensor<string, []>("op_39492_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39493_cast = softmax(axis = var_30385, x = aw_3331_cast)[name = tensor<string, []>("op_39493_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39494_cast = softmax(axis = var_30385, x = aw_3333_cast)[name = tensor<string, []>("op_39494_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39495_cast = softmax(axis = var_30385, x = aw_3335_cast)[name = tensor<string, []>("op_39495_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39496_cast = softmax(axis = var_30385, x = aw_3337_cast)[name = tensor<string, []>("op_39496_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39497_cast = softmax(axis = var_30385, x = aw_3339_cast)[name = tensor<string, []>("op_39497_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39498_cast = softmax(axis = var_30385, x = aw_3341_cast)[name = tensor<string, []>("op_39498_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39499_cast = softmax(axis = var_30385, x = aw_3343_cast)[name = tensor<string, []>("op_39499_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39500_cast = softmax(axis = var_30385, x = aw_3345_cast)[name = tensor<string, []>("op_39500_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39501_cast = softmax(axis = var_30385, x = aw_3347_cast)[name = tensor<string, []>("op_39501_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39502_cast = softmax(axis = var_30385, x = aw_3349_cast)[name = tensor<string, []>("op_39502_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39503_cast = softmax(axis = var_30385, x = aw_3351_cast)[name = tensor<string, []>("op_39503_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39504_cast = softmax(axis = var_30385, x = aw_3353_cast)[name = tensor<string, []>("op_39504_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39505_cast = softmax(axis = var_30385, x = aw_3355_cast)[name = tensor<string, []>("op_39505_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39506_cast = softmax(axis = var_30385, x = aw_3357_cast)[name = tensor<string, []>("op_39506_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_39507_cast = softmax(axis = var_30385, x = aw_3359_cast)[name = tensor<string, []>("op_39507_cast")];
            tensor<string, []> var_39509_equation_0 = const()[name = tensor<string, []>("op_39509_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39509_cast = einsum(equation = var_39509_equation_0, values = (var_39329_cast, var_39488_cast))[name = tensor<string, []>("op_39509_cast")];
            tensor<string, []> var_39511_equation_0 = const()[name = tensor<string, []>("op_39511_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39511_cast = einsum(equation = var_39511_equation_0, values = (var_39333_cast, var_39489_cast))[name = tensor<string, []>("op_39511_cast")];
            tensor<string, []> var_39513_equation_0 = const()[name = tensor<string, []>("op_39513_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39513_cast = einsum(equation = var_39513_equation_0, values = (var_39337_cast, var_39490_cast))[name = tensor<string, []>("op_39513_cast")];
            tensor<string, []> var_39515_equation_0 = const()[name = tensor<string, []>("op_39515_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39515_cast = einsum(equation = var_39515_equation_0, values = (var_39341_cast, var_39491_cast))[name = tensor<string, []>("op_39515_cast")];
            tensor<string, []> var_39517_equation_0 = const()[name = tensor<string, []>("op_39517_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39517_cast = einsum(equation = var_39517_equation_0, values = (var_39345_cast, var_39492_cast))[name = tensor<string, []>("op_39517_cast")];
            tensor<string, []> var_39519_equation_0 = const()[name = tensor<string, []>("op_39519_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39519_cast = einsum(equation = var_39519_equation_0, values = (var_39349_cast, var_39493_cast))[name = tensor<string, []>("op_39519_cast")];
            tensor<string, []> var_39521_equation_0 = const()[name = tensor<string, []>("op_39521_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39521_cast = einsum(equation = var_39521_equation_0, values = (var_39353_cast, var_39494_cast))[name = tensor<string, []>("op_39521_cast")];
            tensor<string, []> var_39523_equation_0 = const()[name = tensor<string, []>("op_39523_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39523_cast = einsum(equation = var_39523_equation_0, values = (var_39357_cast, var_39495_cast))[name = tensor<string, []>("op_39523_cast")];
            tensor<string, []> var_39525_equation_0 = const()[name = tensor<string, []>("op_39525_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39525_cast = einsum(equation = var_39525_equation_0, values = (var_39361_cast, var_39496_cast))[name = tensor<string, []>("op_39525_cast")];
            tensor<string, []> var_39527_equation_0 = const()[name = tensor<string, []>("op_39527_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39527_cast = einsum(equation = var_39527_equation_0, values = (var_39365_cast, var_39497_cast))[name = tensor<string, []>("op_39527_cast")];
            tensor<string, []> var_39529_equation_0 = const()[name = tensor<string, []>("op_39529_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39529_cast = einsum(equation = var_39529_equation_0, values = (var_39369_cast, var_39498_cast))[name = tensor<string, []>("op_39529_cast")];
            tensor<string, []> var_39531_equation_0 = const()[name = tensor<string, []>("op_39531_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39531_cast = einsum(equation = var_39531_equation_0, values = (var_39373_cast, var_39499_cast))[name = tensor<string, []>("op_39531_cast")];
            tensor<string, []> var_39533_equation_0 = const()[name = tensor<string, []>("op_39533_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39533_cast = einsum(equation = var_39533_equation_0, values = (var_39377_cast, var_39500_cast))[name = tensor<string, []>("op_39533_cast")];
            tensor<string, []> var_39535_equation_0 = const()[name = tensor<string, []>("op_39535_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39535_cast = einsum(equation = var_39535_equation_0, values = (var_39381_cast, var_39501_cast))[name = tensor<string, []>("op_39535_cast")];
            tensor<string, []> var_39537_equation_0 = const()[name = tensor<string, []>("op_39537_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39537_cast = einsum(equation = var_39537_equation_0, values = (var_39385_cast, var_39502_cast))[name = tensor<string, []>("op_39537_cast")];
            tensor<string, []> var_39539_equation_0 = const()[name = tensor<string, []>("op_39539_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39539_cast = einsum(equation = var_39539_equation_0, values = (var_39389_cast, var_39503_cast))[name = tensor<string, []>("op_39539_cast")];
            tensor<string, []> var_39541_equation_0 = const()[name = tensor<string, []>("op_39541_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39541_cast = einsum(equation = var_39541_equation_0, values = (var_39393_cast, var_39504_cast))[name = tensor<string, []>("op_39541_cast")];
            tensor<string, []> var_39543_equation_0 = const()[name = tensor<string, []>("op_39543_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39543_cast = einsum(equation = var_39543_equation_0, values = (var_39397_cast, var_39505_cast))[name = tensor<string, []>("op_39543_cast")];
            tensor<string, []> var_39545_equation_0 = const()[name = tensor<string, []>("op_39545_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39545_cast = einsum(equation = var_39545_equation_0, values = (var_39401_cast, var_39506_cast))[name = tensor<string, []>("op_39545_cast")];
            tensor<string, []> var_39547_equation_0 = const()[name = tensor<string, []>("op_39547_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_39547_cast = einsum(equation = var_39547_equation_0, values = (var_39405_cast, var_39507_cast))[name = tensor<string, []>("op_39547_cast")];
            tensor<bool, []> input_515_interleave_0 = const()[name = tensor<string, []>("input_515_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_515_cast = concat(axis = var_30385, interleave = input_515_interleave_0, values = (var_39509_cast, var_39511_cast, var_39513_cast, var_39515_cast, var_39517_cast, var_39519_cast, var_39521_cast, var_39523_cast, var_39525_cast, var_39527_cast, var_39529_cast, var_39531_cast, var_39533_cast, var_39535_cast, var_39537_cast, var_39539_cast, var_39541_cast, var_39543_cast, var_39545_cast, var_39547_cast))[name = tensor<string, []>("input_515_cast")];
            tensor<int32, [2]> var_39553 = const()[name = tensor<string, []>("op_39553"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_39555 = const()[name = tensor<string, []>("op_39555"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_39557_pad_type_0 = const()[name = tensor<string, []>("op_39557_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_39557_pad_0 = const()[name = tensor<string, []>("op_39557_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3240921344)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3244198208)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_39557_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_out_0_bias_to_fp16, dilations = var_39555, groups = var_30385, pad = var_39557_pad_0, pad_type = var_39557_pad_type_0, strides = var_39553, weight = up_blocks_0_attentions_0_transformer_blocks_9_attn2_to_out_0_weight_to_fp16, x = input_515_cast)[name = tensor<string, []>("op_39557_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_263_cast = add(x = var_39557_cast, y = inputs_261_cast)[name = tensor<string, []>("inputs_263_cast")];
            tensor<int32, [1]> var_39561 = const()[name = tensor<string, []>("op_39561"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_263_cast = reduce_mean(axes = var_39561, keep_dims = var_30380, x = inputs_263_cast)[name = tensor<string, []>("channels_mean_263_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_263_cast = sub(x = inputs_263_cast, y = channels_mean_263_cast)[name = tensor<string, []>("zero_mean_263_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_263_cast = mul(x = zero_mean_263_cast, y = zero_mean_263_cast)[name = tensor<string, []>("zero_mean_sq_263_cast")];
            tensor<int32, [1]> var_39565 = const()[name = tensor<string, []>("op_39565"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_39566_cast = reduce_mean(axes = var_39565, keep_dims = var_30380, x = zero_mean_sq_263_cast)[name = tensor<string, []>("op_39566_cast")];
            tensor<fp16, []> var_39567_to_fp16 = const()[name = tensor<string, []>("op_39567_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_39568_cast = add(x = var_39566_cast, y = var_39567_to_fp16)[name = tensor<string, []>("op_39568_cast")];
            tensor<fp16, []> denom_263_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_263_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_263_cast = rsqrt(epsilon = denom_263_epsilon_0_to_fp16, x = var_39568_cast)[name = tensor<string, []>("denom_263_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_263_cast = mul(x = zero_mean_263_cast, y = denom_263_cast)[name = tensor<string, []>("out_263_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_39572_to_fp16 = const()[name = tensor<string, []>("op_39572_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3244200832)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_39573_cast = add(x = out_263_cast, y = var_39572_to_fp16)[name = tensor<string, []>("op_39573_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_39575_to_fp16 = const()[name = tensor<string, []>("op_39575_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3244203456)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_517_cast = mul(x = var_39573_cast, y = var_39575_to_fp16)[name = tensor<string, []>("input_517_cast")];
            tensor<int32, [2]> var_39583 = const()[name = tensor<string, []>("op_39583"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_39585 = const()[name = tensor<string, []>("op_39585"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_39587_pad_type_0 = const()[name = tensor<string, []>("op_39587_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_39587_pad_0 = const()[name = tensor<string, []>("op_39587_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3244206080)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3270420544)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_39587_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_9_ff_net_0_proj_bias_to_fp16, dilations = var_39585, groups = var_30385, pad = var_39587_pad_0, pad_type = var_39587_pad_type_0, strides = var_39583, weight = up_blocks_0_attentions_0_transformer_blocks_9_ff_net_0_proj_weight_to_fp16, x = input_517_cast)[name = tensor<string, []>("op_39587_cast")];
            tensor<int32, [2]> var_39588_split_sizes_0 = const()[name = tensor<string, []>("op_39588_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_39588_axis_0 = const()[name = tensor<string, []>("op_39588_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_39588_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_39588_cast_1 = split(axis = var_39588_axis_0, split_sizes = var_39588_split_sizes_0, x = var_39587_cast)[name = tensor<string, []>("op_39588_cast")];
            tensor<string, []> var_39590_mode_0 = const()[name = tensor<string, []>("op_39590_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_39590_cast = gelu(mode = var_39590_mode_0, x = var_39588_cast_1)[name = tensor<string, []>("op_39590_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_519_cast = mul(x = var_39588_cast_0, y = var_39590_cast)[name = tensor<string, []>("input_519_cast")];
            tensor<int32, [2]> var_39594 = const()[name = tensor<string, []>("op_39594"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_39596 = const()[name = tensor<string, []>("op_39596"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_39598_pad_type_0 = const()[name = tensor<string, []>("op_39598_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_39598_pad_0 = const()[name = tensor<string, []>("op_39598_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3270441088)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3283548352)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_39598_cast = conv(bias = up_blocks_0_attentions_0_transformer_blocks_9_ff_net_2_bias_to_fp16, dilations = var_39596, groups = var_30385, pad = var_39598_pad_0, pad_type = var_39598_pad_type_0, strides = var_39594, weight = up_blocks_0_attentions_0_transformer_blocks_9_ff_net_2_weight_to_fp16, x = input_519_cast)[name = tensor<string, []>("op_39598_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_347_cast = add(x = var_39598_cast, y = inputs_263_cast)[name = tensor<string, []>("hidden_states_347_cast")];
            tensor<int32, [4]> var_39600 = const()[name = tensor<string, []>("op_39600"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> input_521_cast = reshape(shape = var_39600, x = hidden_states_347_cast)[name = tensor<string, []>("input_521_cast")];
            tensor<int32, [2]> var_39604 = const()[name = tensor<string, []>("op_39604"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_39606 = const()[name = tensor<string, []>("op_39606"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_349_pad_type_0 = const()[name = tensor<string, []>("hidden_states_349_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_349_pad_0 = const()[name = tensor<string, []>("hidden_states_349_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_proj_out_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3283550976)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_0_proj_out_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3286827840)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_349_cast = conv(bias = up_blocks_0_attentions_0_proj_out_bias_to_fp16, dilations = var_39606, groups = var_30385, pad = hidden_states_349_pad_0, pad_type = hidden_states_349_pad_type_0, strides = var_39604, weight = up_blocks_0_attentions_0_proj_out_weight_to_fp16, x = input_521_cast)[name = tensor<string, []>("hidden_states_349_cast")];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_351_cast = add(x = hidden_states_349_cast, y = hidden_states_283_cast)[name = tensor<string, []>("hidden_states_351_cast")];
            tensor<bool, []> input_523_interleave_0 = const()[name = tensor<string, []>("input_523_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 2560, 32, 32]> input_523_cast = concat(axis = var_30385, interleave = input_523_interleave_0, values = (hidden_states_351_cast, res_hidden_states_3_cast))[name = tensor<string, []>("input_523_cast")];
            tensor<int32, [5]> reshape_96_shape_0 = const()[name = tensor<string, []>("reshape_96_shape_0"), val = tensor<int32, [5]>([2, 32, 80, 32, 32])];
            tensor<fp16, [2, 32, 80, 32, 32]> reshape_96_cast = reshape(shape = reshape_96_shape_0, x = input_523_cast)[name = tensor<string, []>("reshape_96_cast")];
            tensor<int32, [3]> reduce_mean_72_axes_0 = const()[name = tensor<string, []>("reduce_mean_72_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_72_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_72_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_72_cast = reduce_mean(axes = reduce_mean_72_axes_0, keep_dims = reduce_mean_72_keep_dims_0, x = reshape_96_cast)[name = tensor<string, []>("reduce_mean_72_cast")];
            tensor<fp16, [2, 32, 80, 32, 32]> sub_48_cast = sub(x = reshape_96_cast, y = reduce_mean_72_cast)[name = tensor<string, []>("sub_48_cast")];
            tensor<fp16, [2, 32, 80, 32, 32]> square_24_cast = square(x = sub_48_cast)[name = tensor<string, []>("square_24_cast")];
            tensor<int32, [3]> reduce_mean_74_axes_0 = const()[name = tensor<string, []>("reduce_mean_74_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_74_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_74_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_74_cast = reduce_mean(axes = reduce_mean_74_axes_0, keep_dims = reduce_mean_74_keep_dims_0, x = square_24_cast)[name = tensor<string, []>("reduce_mean_74_cast")];
            tensor<fp16, []> add_48_y_0_to_fp16 = const()[name = tensor<string, []>("add_48_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_48_cast = add(x = reduce_mean_74_cast, y = add_48_y_0_to_fp16)[name = tensor<string, []>("add_48_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_24_cast = sqrt(x = add_48_cast)[name = tensor<string, []>("sqrt_24_cast")];
            tensor<fp16, [2, 32, 80, 32, 32]> real_div_24_cast = real_div(x = sub_48_cast, y = sqrt_24_cast)[name = tensor<string, []>("real_div_24_cast")];
            tensor<int32, [4]> reshape_97_shape_0 = const()[name = tensor<string, []>("reshape_97_shape_0"), val = tensor<int32, [4]>([2, 2560, 32, 32])];
            tensor<fp16, [2, 2560, 32, 32]> reshape_97_cast = reshape(shape = reshape_97_shape_0, x = real_div_24_cast)[name = tensor<string, []>("reshape_97_cast")];
            tensor<fp16, [2560]> add_49_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_49_gamma_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3286830464)))];
            tensor<fp16, [2560]> add_49_beta_0_to_fp16 = const()[name = tensor<string, []>("add_49_beta_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3286835648)))];
            tensor<fp16, []> add_49_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_49_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 2560, 32, 32]> add_49_cast = batch_norm(beta = add_49_beta_0_to_fp16, epsilon = add_49_epsilon_0_to_fp16, gamma = add_49_gamma_0_to_fp16, mean = add_43_mean_0_to_fp16, variance = add_43_variance_0_to_fp16, x = reshape_97_cast)[name = tensor<string, []>("add_49_cast")];
            tensor<fp16, [2, 2560, 32, 32]> input_527_cast = silu(x = add_49_cast)[name = tensor<string, []>("input_527_cast")];
            tensor<int32, [2]> var_39624 = const()[name = tensor<string, []>("op_39624"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_39626 = const()[name = tensor<string, []>("op_39626"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_353_pad_type_0 = const()[name = tensor<string, []>("hidden_states_353_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_353_pad_0 = const()[name = tensor<string, []>("hidden_states_353_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 2560, 3, 3]> up_blocks_0_resnets_1_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_1_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 2560, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3286840832)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_1_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_1_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3345823296)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_353_cast = conv(bias = up_blocks_0_resnets_1_conv1_bias_to_fp16, dilations = var_39626, groups = var_30385, pad = hidden_states_353_pad_0, pad_type = hidden_states_353_pad_type_0, strides = var_39624, weight = up_blocks_0_resnets_1_conv1_weight_to_fp16, x = input_527_cast)[name = tensor<string, []>("hidden_states_353_cast")];
            tensor<int32, [2]> var_39632 = const()[name = tensor<string, []>("op_39632"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_39634 = const()[name = tensor<string, []>("op_39634"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_19_pad_type_0 = const()[name = tensor<string, []>("temb_19_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_19_pad_0 = const()[name = tensor<string, []>("temb_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3345825920)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3349102784)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_19_cast = conv(bias = up_blocks_0_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_39634, groups = var_30385, pad = temb_19_pad_0, pad_type = temb_19_pad_type_0, strides = var_39632, weight = up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16, x = input_23_cast)[name = tensor<string, []>("temb_19_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_531_cast = add(x = hidden_states_353_cast, y = temb_19_cast)[name = tensor<string, []>("input_531_cast")];
            tensor<int32, [5]> reshape_100_shape_0 = const()[name = tensor<string, []>("reshape_100_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_100_cast = reshape(shape = reshape_100_shape_0, x = input_531_cast)[name = tensor<string, []>("reshape_100_cast")];
            tensor<int32, [3]> reduce_mean_75_axes_0 = const()[name = tensor<string, []>("reduce_mean_75_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_75_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_75_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_75_cast = reduce_mean(axes = reduce_mean_75_axes_0, keep_dims = reduce_mean_75_keep_dims_0, x = reshape_100_cast)[name = tensor<string, []>("reduce_mean_75_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_50_cast = sub(x = reshape_100_cast, y = reduce_mean_75_cast)[name = tensor<string, []>("sub_50_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_25_cast = square(x = sub_50_cast)[name = tensor<string, []>("square_25_cast")];
            tensor<int32, [3]> reduce_mean_77_axes_0 = const()[name = tensor<string, []>("reduce_mean_77_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_77_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_77_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_77_cast = reduce_mean(axes = reduce_mean_77_axes_0, keep_dims = reduce_mean_77_keep_dims_0, x = square_25_cast)[name = tensor<string, []>("reduce_mean_77_cast")];
            tensor<fp16, []> add_50_y_0_to_fp16 = const()[name = tensor<string, []>("add_50_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_50_cast = add(x = reduce_mean_77_cast, y = add_50_y_0_to_fp16)[name = tensor<string, []>("add_50_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_25_cast = sqrt(x = add_50_cast)[name = tensor<string, []>("sqrt_25_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_25_cast = real_div(x = sub_50_cast, y = sqrt_25_cast)[name = tensor<string, []>("real_div_25_cast")];
            tensor<int32, [4]> reshape_101_shape_0 = const()[name = tensor<string, []>("reshape_101_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_101_cast = reshape(shape = reshape_101_shape_0, x = real_div_25_cast)[name = tensor<string, []>("reshape_101_cast")];
            tensor<fp16, [1280]> add_51_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_51_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3349105408)))];
            tensor<fp16, [1280]> add_51_beta_0_to_fp16 = const()[name = tensor<string, []>("add_51_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3349108032)))];
            tensor<fp16, []> add_51_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_51_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 32, 32]> add_51_cast = batch_norm(beta = add_51_beta_0_to_fp16, epsilon = add_51_epsilon_0_to_fp16, gamma = add_51_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_101_cast)[name = tensor<string, []>("add_51_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_535_cast = silu(x = add_51_cast)[name = tensor<string, []>("input_535_cast")];
            tensor<int32, [2]> var_39644 = const()[name = tensor<string, []>("op_39644"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_39646 = const()[name = tensor<string, []>("op_39646"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_355_pad_type_0 = const()[name = tensor<string, []>("hidden_states_355_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_355_pad_0 = const()[name = tensor<string, []>("hidden_states_355_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> up_blocks_0_resnets_1_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_1_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3349110656)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_1_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_1_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3378601920)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_355_cast = conv(bias = up_blocks_0_resnets_1_conv2_bias_to_fp16, dilations = var_39646, groups = var_30385, pad = hidden_states_355_pad_0, pad_type = hidden_states_355_pad_type_0, strides = var_39644, weight = up_blocks_0_resnets_1_conv2_weight_to_fp16, x = input_535_cast)[name = tensor<string, []>("hidden_states_355_cast")];
            tensor<int32, [2]> var_39651 = const()[name = tensor<string, []>("op_39651"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_39653 = const()[name = tensor<string, []>("op_39653"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_7_pad_type_0 = const()[name = tensor<string, []>("x_7_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_7_pad_0 = const()[name = tensor<string, []>("x_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2560, 1, 1]> up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [1280, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3378604544)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3385158208)))];
            tensor<fp16, [2, 1280, 32, 32]> x_7_cast = conv(bias = up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_39653, groups = var_30385, pad = x_7_pad_0, pad_type = x_7_pad_type_0, strides = var_39651, weight = up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16, x = input_523_cast)[name = tensor<string, []>("x_7_cast")];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_357_cast = add(x = x_7_cast, y = hidden_states_355_cast)[name = tensor<string, []>("hidden_states_357_cast")];
            tensor<int32, [5]> reshape_104_shape_0 = const()[name = tensor<string, []>("reshape_104_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_104_cast = reshape(shape = reshape_104_shape_0, x = hidden_states_357_cast)[name = tensor<string, []>("reshape_104_cast")];
            tensor<int32, [3]> reduce_mean_78_axes_0 = const()[name = tensor<string, []>("reduce_mean_78_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_78_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_78_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_78_cast = reduce_mean(axes = reduce_mean_78_axes_0, keep_dims = reduce_mean_78_keep_dims_0, x = reshape_104_cast)[name = tensor<string, []>("reduce_mean_78_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_52_cast = sub(x = reshape_104_cast, y = reduce_mean_78_cast)[name = tensor<string, []>("sub_52_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_26_cast = square(x = sub_52_cast)[name = tensor<string, []>("square_26_cast")];
            tensor<int32, [3]> reduce_mean_80_axes_0 = const()[name = tensor<string, []>("reduce_mean_80_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_80_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_80_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_80_cast = reduce_mean(axes = reduce_mean_80_axes_0, keep_dims = reduce_mean_80_keep_dims_0, x = square_26_cast)[name = tensor<string, []>("reduce_mean_80_cast")];
            tensor<fp16, []> add_52_y_0_to_fp16 = const()[name = tensor<string, []>("add_52_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_52_cast = add(x = reduce_mean_80_cast, y = add_52_y_0_to_fp16)[name = tensor<string, []>("add_52_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_26_cast = sqrt(x = add_52_cast)[name = tensor<string, []>("sqrt_26_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_26_cast = real_div(x = sub_52_cast, y = sqrt_26_cast)[name = tensor<string, []>("real_div_26_cast")];
            tensor<int32, [4]> reshape_105_shape_0 = const()[name = tensor<string, []>("reshape_105_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_105_cast = reshape(shape = reshape_105_shape_0, x = real_div_26_cast)[name = tensor<string, []>("reshape_105_cast")];
            tensor<fp16, [1280]> add_53_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_53_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3385160832)))];
            tensor<fp16, [1280]> add_53_beta_0_to_fp16 = const()[name = tensor<string, []>("add_53_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3385163456)))];
            tensor<fp16, []> add_53_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_53_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 32, 32]> add_53_cast = batch_norm(beta = add_53_beta_0_to_fp16, epsilon = add_53_epsilon_0_to_fp16, gamma = add_53_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_105_cast)[name = tensor<string, []>("add_53_cast")];
            tensor<int32, [2]> var_39691 = const()[name = tensor<string, []>("op_39691"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_39693 = const()[name = tensor<string, []>("op_39693"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_359_pad_type_0 = const()[name = tensor<string, []>("hidden_states_359_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_359_pad_0 = const()[name = tensor<string, []>("hidden_states_359_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_proj_in_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3385166080)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_proj_in_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3388442944)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_359_cast = conv(bias = up_blocks_0_attentions_1_proj_in_bias_to_fp16, dilations = var_39693, groups = var_30385, pad = hidden_states_359_pad_0, pad_type = hidden_states_359_pad_type_0, strides = var_39691, weight = up_blocks_0_attentions_1_proj_in_weight_to_fp16, x = add_53_cast)[name = tensor<string, []>("hidden_states_359_cast")];
            tensor<int32, [4]> var_39698 = const()[name = tensor<string, []>("op_39698"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_265_cast = reshape(shape = var_39698, x = hidden_states_359_cast)[name = tensor<string, []>("inputs_265_cast")];
            tensor<int32, [1]> var_39708 = const()[name = tensor<string, []>("op_39708"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_265_cast = reduce_mean(axes = var_39708, keep_dims = var_30380, x = inputs_265_cast)[name = tensor<string, []>("channels_mean_265_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_265_cast = sub(x = inputs_265_cast, y = channels_mean_265_cast)[name = tensor<string, []>("zero_mean_265_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_265_cast = mul(x = zero_mean_265_cast, y = zero_mean_265_cast)[name = tensor<string, []>("zero_mean_sq_265_cast")];
            tensor<int32, [1]> var_39712 = const()[name = tensor<string, []>("op_39712"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_39713_cast = reduce_mean(axes = var_39712, keep_dims = var_30380, x = zero_mean_sq_265_cast)[name = tensor<string, []>("op_39713_cast")];
            tensor<fp16, []> var_39714_to_fp16 = const()[name = tensor<string, []>("op_39714_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_39715_cast = add(x = var_39713_cast, y = var_39714_to_fp16)[name = tensor<string, []>("op_39715_cast")];
            tensor<fp16, []> denom_265_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_265_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_265_cast = rsqrt(epsilon = denom_265_epsilon_0_to_fp16, x = var_39715_cast)[name = tensor<string, []>("denom_265_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_265_cast = mul(x = zero_mean_265_cast, y = denom_265_cast)[name = tensor<string, []>("out_265_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_39719_to_fp16 = const()[name = tensor<string, []>("op_39719_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3388445568)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_39720_cast = add(x = out_265_cast, y = var_39719_to_fp16)[name = tensor<string, []>("op_39720_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_39722_to_fp16 = const()[name = tensor<string, []>("op_39722_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3388448192)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_361_cast = mul(x = var_39720_cast, y = var_39722_to_fp16)[name = tensor<string, []>("hidden_states_361_cast")];
            tensor<int32, [2]> var_39729 = const()[name = tensor<string, []>("op_39729"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_39731 = const()[name = tensor<string, []>("op_39731"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_177_pad_type_0 = const()[name = tensor<string, []>("q_177_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_177_pad_0 = const()[name = tensor<string, []>("q_177_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3388450816)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_177_cast = conv(dilations = var_39731, groups = var_30385, pad = q_177_pad_0, pad_type = q_177_pad_type_0, strides = var_39729, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_361_cast)[name = tensor<string, []>("q_177_cast")];
            tensor<int32, [2]> var_39735 = const()[name = tensor<string, []>("op_39735"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_39737 = const()[name = tensor<string, []>("op_39737"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_353_pad_type_0 = const()[name = tensor<string, []>("k_353_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_353_pad_0 = const()[name = tensor<string, []>("k_353_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3391727680)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_353_cast = conv(dilations = var_39737, groups = var_30385, pad = k_353_pad_0, pad_type = k_353_pad_type_0, strides = var_39735, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_361_cast)[name = tensor<string, []>("k_353_cast")];
            tensor<int32, [2]> var_39741 = const()[name = tensor<string, []>("op_39741"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_39743 = const()[name = tensor<string, []>("op_39743"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_177_pad_type_0 = const()[name = tensor<string, []>("v_177_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_177_pad_0 = const()[name = tensor<string, []>("v_177_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3395004544)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_177_cast = conv(dilations = var_39743, groups = var_30385, pad = v_177_pad_0, pad_type = v_177_pad_type_0, strides = var_39741, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_361_cast)[name = tensor<string, []>("v_177_cast")];
            tensor<int32, [4]> var_39747_begin_0 = const()[name = tensor<string, []>("op_39747_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_39747_end_0 = const()[name = tensor<string, []>("op_39747_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_39747_end_mask_0 = const()[name = tensor<string, []>("op_39747_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39747_cast = slice_by_index(begin = var_39747_begin_0, end = var_39747_end_0, end_mask = var_39747_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39747_cast")];
            tensor<int32, [4]> var_39751_begin_0 = const()[name = tensor<string, []>("op_39751_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_39751_end_0 = const()[name = tensor<string, []>("op_39751_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_39751_end_mask_0 = const()[name = tensor<string, []>("op_39751_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39751_cast = slice_by_index(begin = var_39751_begin_0, end = var_39751_end_0, end_mask = var_39751_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39751_cast")];
            tensor<int32, [4]> var_39755_begin_0 = const()[name = tensor<string, []>("op_39755_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_39755_end_0 = const()[name = tensor<string, []>("op_39755_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_39755_end_mask_0 = const()[name = tensor<string, []>("op_39755_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39755_cast = slice_by_index(begin = var_39755_begin_0, end = var_39755_end_0, end_mask = var_39755_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39755_cast")];
            tensor<int32, [4]> var_39759_begin_0 = const()[name = tensor<string, []>("op_39759_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_39759_end_0 = const()[name = tensor<string, []>("op_39759_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_39759_end_mask_0 = const()[name = tensor<string, []>("op_39759_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39759_cast = slice_by_index(begin = var_39759_begin_0, end = var_39759_end_0, end_mask = var_39759_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39759_cast")];
            tensor<int32, [4]> var_39763_begin_0 = const()[name = tensor<string, []>("op_39763_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_39763_end_0 = const()[name = tensor<string, []>("op_39763_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_39763_end_mask_0 = const()[name = tensor<string, []>("op_39763_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39763_cast = slice_by_index(begin = var_39763_begin_0, end = var_39763_end_0, end_mask = var_39763_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39763_cast")];
            tensor<int32, [4]> var_39767_begin_0 = const()[name = tensor<string, []>("op_39767_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_39767_end_0 = const()[name = tensor<string, []>("op_39767_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_39767_end_mask_0 = const()[name = tensor<string, []>("op_39767_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39767_cast = slice_by_index(begin = var_39767_begin_0, end = var_39767_end_0, end_mask = var_39767_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39767_cast")];
            tensor<int32, [4]> var_39771_begin_0 = const()[name = tensor<string, []>("op_39771_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_39771_end_0 = const()[name = tensor<string, []>("op_39771_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_39771_end_mask_0 = const()[name = tensor<string, []>("op_39771_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39771_cast = slice_by_index(begin = var_39771_begin_0, end = var_39771_end_0, end_mask = var_39771_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39771_cast")];
            tensor<int32, [4]> var_39775_begin_0 = const()[name = tensor<string, []>("op_39775_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_39775_end_0 = const()[name = tensor<string, []>("op_39775_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_39775_end_mask_0 = const()[name = tensor<string, []>("op_39775_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39775_cast = slice_by_index(begin = var_39775_begin_0, end = var_39775_end_0, end_mask = var_39775_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39775_cast")];
            tensor<int32, [4]> var_39779_begin_0 = const()[name = tensor<string, []>("op_39779_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_39779_end_0 = const()[name = tensor<string, []>("op_39779_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_39779_end_mask_0 = const()[name = tensor<string, []>("op_39779_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39779_cast = slice_by_index(begin = var_39779_begin_0, end = var_39779_end_0, end_mask = var_39779_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39779_cast")];
            tensor<int32, [4]> var_39783_begin_0 = const()[name = tensor<string, []>("op_39783_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_39783_end_0 = const()[name = tensor<string, []>("op_39783_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_39783_end_mask_0 = const()[name = tensor<string, []>("op_39783_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39783_cast = slice_by_index(begin = var_39783_begin_0, end = var_39783_end_0, end_mask = var_39783_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39783_cast")];
            tensor<int32, [4]> var_39787_begin_0 = const()[name = tensor<string, []>("op_39787_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_39787_end_0 = const()[name = tensor<string, []>("op_39787_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_39787_end_mask_0 = const()[name = tensor<string, []>("op_39787_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39787_cast = slice_by_index(begin = var_39787_begin_0, end = var_39787_end_0, end_mask = var_39787_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39787_cast")];
            tensor<int32, [4]> var_39791_begin_0 = const()[name = tensor<string, []>("op_39791_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_39791_end_0 = const()[name = tensor<string, []>("op_39791_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_39791_end_mask_0 = const()[name = tensor<string, []>("op_39791_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39791_cast = slice_by_index(begin = var_39791_begin_0, end = var_39791_end_0, end_mask = var_39791_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39791_cast")];
            tensor<int32, [4]> var_39795_begin_0 = const()[name = tensor<string, []>("op_39795_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_39795_end_0 = const()[name = tensor<string, []>("op_39795_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_39795_end_mask_0 = const()[name = tensor<string, []>("op_39795_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39795_cast = slice_by_index(begin = var_39795_begin_0, end = var_39795_end_0, end_mask = var_39795_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39795_cast")];
            tensor<int32, [4]> var_39799_begin_0 = const()[name = tensor<string, []>("op_39799_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_39799_end_0 = const()[name = tensor<string, []>("op_39799_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_39799_end_mask_0 = const()[name = tensor<string, []>("op_39799_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39799_cast = slice_by_index(begin = var_39799_begin_0, end = var_39799_end_0, end_mask = var_39799_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39799_cast")];
            tensor<int32, [4]> var_39803_begin_0 = const()[name = tensor<string, []>("op_39803_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_39803_end_0 = const()[name = tensor<string, []>("op_39803_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_39803_end_mask_0 = const()[name = tensor<string, []>("op_39803_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39803_cast = slice_by_index(begin = var_39803_begin_0, end = var_39803_end_0, end_mask = var_39803_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39803_cast")];
            tensor<int32, [4]> var_39807_begin_0 = const()[name = tensor<string, []>("op_39807_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_39807_end_0 = const()[name = tensor<string, []>("op_39807_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_39807_end_mask_0 = const()[name = tensor<string, []>("op_39807_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39807_cast = slice_by_index(begin = var_39807_begin_0, end = var_39807_end_0, end_mask = var_39807_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39807_cast")];
            tensor<int32, [4]> var_39811_begin_0 = const()[name = tensor<string, []>("op_39811_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_39811_end_0 = const()[name = tensor<string, []>("op_39811_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_39811_end_mask_0 = const()[name = tensor<string, []>("op_39811_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39811_cast = slice_by_index(begin = var_39811_begin_0, end = var_39811_end_0, end_mask = var_39811_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39811_cast")];
            tensor<int32, [4]> var_39815_begin_0 = const()[name = tensor<string, []>("op_39815_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_39815_end_0 = const()[name = tensor<string, []>("op_39815_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_39815_end_mask_0 = const()[name = tensor<string, []>("op_39815_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39815_cast = slice_by_index(begin = var_39815_begin_0, end = var_39815_end_0, end_mask = var_39815_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39815_cast")];
            tensor<int32, [4]> var_39819_begin_0 = const()[name = tensor<string, []>("op_39819_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_39819_end_0 = const()[name = tensor<string, []>("op_39819_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_39819_end_mask_0 = const()[name = tensor<string, []>("op_39819_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39819_cast = slice_by_index(begin = var_39819_begin_0, end = var_39819_end_0, end_mask = var_39819_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39819_cast")];
            tensor<int32, [4]> var_39823_begin_0 = const()[name = tensor<string, []>("op_39823_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_39823_end_0 = const()[name = tensor<string, []>("op_39823_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_39823_end_mask_0 = const()[name = tensor<string, []>("op_39823_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39823_cast = slice_by_index(begin = var_39823_begin_0, end = var_39823_end_0, end_mask = var_39823_end_mask_0, x = q_177_cast)[name = tensor<string, []>("op_39823_cast")];
            tensor<int32, [4]> k_355_perm_0 = const()[name = tensor<string, []>("k_355_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_39830_begin_0 = const()[name = tensor<string, []>("op_39830_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_39830_end_0 = const()[name = tensor<string, []>("op_39830_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_39830_end_mask_0 = const()[name = tensor<string, []>("op_39830_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_51 = transpose(perm = k_355_perm_0, x = k_353_cast)[name = tensor<string, []>("transpose_51")];
            tensor<fp16, [2, 1024, 1, 64]> var_39830_cast = slice_by_index(begin = var_39830_begin_0, end = var_39830_end_0, end_mask = var_39830_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39830_cast")];
            tensor<int32, [4]> var_39834_begin_0 = const()[name = tensor<string, []>("op_39834_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_39834_end_0 = const()[name = tensor<string, []>("op_39834_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_39834_end_mask_0 = const()[name = tensor<string, []>("op_39834_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39834_cast = slice_by_index(begin = var_39834_begin_0, end = var_39834_end_0, end_mask = var_39834_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39834_cast")];
            tensor<int32, [4]> var_39838_begin_0 = const()[name = tensor<string, []>("op_39838_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_39838_end_0 = const()[name = tensor<string, []>("op_39838_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_39838_end_mask_0 = const()[name = tensor<string, []>("op_39838_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39838_cast = slice_by_index(begin = var_39838_begin_0, end = var_39838_end_0, end_mask = var_39838_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39838_cast")];
            tensor<int32, [4]> var_39842_begin_0 = const()[name = tensor<string, []>("op_39842_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_39842_end_0 = const()[name = tensor<string, []>("op_39842_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_39842_end_mask_0 = const()[name = tensor<string, []>("op_39842_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39842_cast = slice_by_index(begin = var_39842_begin_0, end = var_39842_end_0, end_mask = var_39842_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39842_cast")];
            tensor<int32, [4]> var_39846_begin_0 = const()[name = tensor<string, []>("op_39846_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_39846_end_0 = const()[name = tensor<string, []>("op_39846_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_39846_end_mask_0 = const()[name = tensor<string, []>("op_39846_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39846_cast = slice_by_index(begin = var_39846_begin_0, end = var_39846_end_0, end_mask = var_39846_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39846_cast")];
            tensor<int32, [4]> var_39850_begin_0 = const()[name = tensor<string, []>("op_39850_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_39850_end_0 = const()[name = tensor<string, []>("op_39850_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_39850_end_mask_0 = const()[name = tensor<string, []>("op_39850_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39850_cast = slice_by_index(begin = var_39850_begin_0, end = var_39850_end_0, end_mask = var_39850_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39850_cast")];
            tensor<int32, [4]> var_39854_begin_0 = const()[name = tensor<string, []>("op_39854_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_39854_end_0 = const()[name = tensor<string, []>("op_39854_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_39854_end_mask_0 = const()[name = tensor<string, []>("op_39854_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39854_cast = slice_by_index(begin = var_39854_begin_0, end = var_39854_end_0, end_mask = var_39854_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39854_cast")];
            tensor<int32, [4]> var_39858_begin_0 = const()[name = tensor<string, []>("op_39858_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_39858_end_0 = const()[name = tensor<string, []>("op_39858_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_39858_end_mask_0 = const()[name = tensor<string, []>("op_39858_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39858_cast = slice_by_index(begin = var_39858_begin_0, end = var_39858_end_0, end_mask = var_39858_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39858_cast")];
            tensor<int32, [4]> var_39862_begin_0 = const()[name = tensor<string, []>("op_39862_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_39862_end_0 = const()[name = tensor<string, []>("op_39862_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_39862_end_mask_0 = const()[name = tensor<string, []>("op_39862_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39862_cast = slice_by_index(begin = var_39862_begin_0, end = var_39862_end_0, end_mask = var_39862_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39862_cast")];
            tensor<int32, [4]> var_39866_begin_0 = const()[name = tensor<string, []>("op_39866_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_39866_end_0 = const()[name = tensor<string, []>("op_39866_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_39866_end_mask_0 = const()[name = tensor<string, []>("op_39866_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39866_cast = slice_by_index(begin = var_39866_begin_0, end = var_39866_end_0, end_mask = var_39866_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39866_cast")];
            tensor<int32, [4]> var_39870_begin_0 = const()[name = tensor<string, []>("op_39870_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_39870_end_0 = const()[name = tensor<string, []>("op_39870_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_39870_end_mask_0 = const()[name = tensor<string, []>("op_39870_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39870_cast = slice_by_index(begin = var_39870_begin_0, end = var_39870_end_0, end_mask = var_39870_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39870_cast")];
            tensor<int32, [4]> var_39874_begin_0 = const()[name = tensor<string, []>("op_39874_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_39874_end_0 = const()[name = tensor<string, []>("op_39874_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_39874_end_mask_0 = const()[name = tensor<string, []>("op_39874_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39874_cast = slice_by_index(begin = var_39874_begin_0, end = var_39874_end_0, end_mask = var_39874_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39874_cast")];
            tensor<int32, [4]> var_39878_begin_0 = const()[name = tensor<string, []>("op_39878_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_39878_end_0 = const()[name = tensor<string, []>("op_39878_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_39878_end_mask_0 = const()[name = tensor<string, []>("op_39878_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39878_cast = slice_by_index(begin = var_39878_begin_0, end = var_39878_end_0, end_mask = var_39878_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39878_cast")];
            tensor<int32, [4]> var_39882_begin_0 = const()[name = tensor<string, []>("op_39882_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_39882_end_0 = const()[name = tensor<string, []>("op_39882_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_39882_end_mask_0 = const()[name = tensor<string, []>("op_39882_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39882_cast = slice_by_index(begin = var_39882_begin_0, end = var_39882_end_0, end_mask = var_39882_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39882_cast")];
            tensor<int32, [4]> var_39886_begin_0 = const()[name = tensor<string, []>("op_39886_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_39886_end_0 = const()[name = tensor<string, []>("op_39886_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_39886_end_mask_0 = const()[name = tensor<string, []>("op_39886_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39886_cast = slice_by_index(begin = var_39886_begin_0, end = var_39886_end_0, end_mask = var_39886_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39886_cast")];
            tensor<int32, [4]> var_39890_begin_0 = const()[name = tensor<string, []>("op_39890_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_39890_end_0 = const()[name = tensor<string, []>("op_39890_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_39890_end_mask_0 = const()[name = tensor<string, []>("op_39890_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39890_cast = slice_by_index(begin = var_39890_begin_0, end = var_39890_end_0, end_mask = var_39890_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39890_cast")];
            tensor<int32, [4]> var_39894_begin_0 = const()[name = tensor<string, []>("op_39894_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_39894_end_0 = const()[name = tensor<string, []>("op_39894_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_39894_end_mask_0 = const()[name = tensor<string, []>("op_39894_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39894_cast = slice_by_index(begin = var_39894_begin_0, end = var_39894_end_0, end_mask = var_39894_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39894_cast")];
            tensor<int32, [4]> var_39898_begin_0 = const()[name = tensor<string, []>("op_39898_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_39898_end_0 = const()[name = tensor<string, []>("op_39898_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_39898_end_mask_0 = const()[name = tensor<string, []>("op_39898_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39898_cast = slice_by_index(begin = var_39898_begin_0, end = var_39898_end_0, end_mask = var_39898_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39898_cast")];
            tensor<int32, [4]> var_39902_begin_0 = const()[name = tensor<string, []>("op_39902_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_39902_end_0 = const()[name = tensor<string, []>("op_39902_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_39902_end_mask_0 = const()[name = tensor<string, []>("op_39902_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39902_cast = slice_by_index(begin = var_39902_begin_0, end = var_39902_end_0, end_mask = var_39902_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39902_cast")];
            tensor<int32, [4]> var_39906_begin_0 = const()[name = tensor<string, []>("op_39906_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_39906_end_0 = const()[name = tensor<string, []>("op_39906_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_39906_end_mask_0 = const()[name = tensor<string, []>("op_39906_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_39906_cast = slice_by_index(begin = var_39906_begin_0, end = var_39906_end_0, end_mask = var_39906_end_mask_0, x = transpose_51)[name = tensor<string, []>("op_39906_cast")];
            tensor<int32, [4]> var_39908_begin_0 = const()[name = tensor<string, []>("op_39908_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_39908_end_0 = const()[name = tensor<string, []>("op_39908_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_39908_end_mask_0 = const()[name = tensor<string, []>("op_39908_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39908_cast = slice_by_index(begin = var_39908_begin_0, end = var_39908_end_0, end_mask = var_39908_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39908_cast")];
            tensor<int32, [4]> var_39912_begin_0 = const()[name = tensor<string, []>("op_39912_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_39912_end_0 = const()[name = tensor<string, []>("op_39912_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_39912_end_mask_0 = const()[name = tensor<string, []>("op_39912_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39912_cast = slice_by_index(begin = var_39912_begin_0, end = var_39912_end_0, end_mask = var_39912_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39912_cast")];
            tensor<int32, [4]> var_39916_begin_0 = const()[name = tensor<string, []>("op_39916_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_39916_end_0 = const()[name = tensor<string, []>("op_39916_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_39916_end_mask_0 = const()[name = tensor<string, []>("op_39916_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39916_cast = slice_by_index(begin = var_39916_begin_0, end = var_39916_end_0, end_mask = var_39916_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39916_cast")];
            tensor<int32, [4]> var_39920_begin_0 = const()[name = tensor<string, []>("op_39920_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_39920_end_0 = const()[name = tensor<string, []>("op_39920_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_39920_end_mask_0 = const()[name = tensor<string, []>("op_39920_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39920_cast = slice_by_index(begin = var_39920_begin_0, end = var_39920_end_0, end_mask = var_39920_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39920_cast")];
            tensor<int32, [4]> var_39924_begin_0 = const()[name = tensor<string, []>("op_39924_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_39924_end_0 = const()[name = tensor<string, []>("op_39924_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_39924_end_mask_0 = const()[name = tensor<string, []>("op_39924_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39924_cast = slice_by_index(begin = var_39924_begin_0, end = var_39924_end_0, end_mask = var_39924_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39924_cast")];
            tensor<int32, [4]> var_39928_begin_0 = const()[name = tensor<string, []>("op_39928_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_39928_end_0 = const()[name = tensor<string, []>("op_39928_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_39928_end_mask_0 = const()[name = tensor<string, []>("op_39928_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39928_cast = slice_by_index(begin = var_39928_begin_0, end = var_39928_end_0, end_mask = var_39928_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39928_cast")];
            tensor<int32, [4]> var_39932_begin_0 = const()[name = tensor<string, []>("op_39932_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_39932_end_0 = const()[name = tensor<string, []>("op_39932_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_39932_end_mask_0 = const()[name = tensor<string, []>("op_39932_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39932_cast = slice_by_index(begin = var_39932_begin_0, end = var_39932_end_0, end_mask = var_39932_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39932_cast")];
            tensor<int32, [4]> var_39936_begin_0 = const()[name = tensor<string, []>("op_39936_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_39936_end_0 = const()[name = tensor<string, []>("op_39936_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_39936_end_mask_0 = const()[name = tensor<string, []>("op_39936_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39936_cast = slice_by_index(begin = var_39936_begin_0, end = var_39936_end_0, end_mask = var_39936_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39936_cast")];
            tensor<int32, [4]> var_39940_begin_0 = const()[name = tensor<string, []>("op_39940_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_39940_end_0 = const()[name = tensor<string, []>("op_39940_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_39940_end_mask_0 = const()[name = tensor<string, []>("op_39940_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39940_cast = slice_by_index(begin = var_39940_begin_0, end = var_39940_end_0, end_mask = var_39940_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39940_cast")];
            tensor<int32, [4]> var_39944_begin_0 = const()[name = tensor<string, []>("op_39944_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_39944_end_0 = const()[name = tensor<string, []>("op_39944_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_39944_end_mask_0 = const()[name = tensor<string, []>("op_39944_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39944_cast = slice_by_index(begin = var_39944_begin_0, end = var_39944_end_0, end_mask = var_39944_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39944_cast")];
            tensor<int32, [4]> var_39948_begin_0 = const()[name = tensor<string, []>("op_39948_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_39948_end_0 = const()[name = tensor<string, []>("op_39948_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_39948_end_mask_0 = const()[name = tensor<string, []>("op_39948_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39948_cast = slice_by_index(begin = var_39948_begin_0, end = var_39948_end_0, end_mask = var_39948_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39948_cast")];
            tensor<int32, [4]> var_39952_begin_0 = const()[name = tensor<string, []>("op_39952_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_39952_end_0 = const()[name = tensor<string, []>("op_39952_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_39952_end_mask_0 = const()[name = tensor<string, []>("op_39952_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39952_cast = slice_by_index(begin = var_39952_begin_0, end = var_39952_end_0, end_mask = var_39952_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39952_cast")];
            tensor<int32, [4]> var_39956_begin_0 = const()[name = tensor<string, []>("op_39956_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_39956_end_0 = const()[name = tensor<string, []>("op_39956_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_39956_end_mask_0 = const()[name = tensor<string, []>("op_39956_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39956_cast = slice_by_index(begin = var_39956_begin_0, end = var_39956_end_0, end_mask = var_39956_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39956_cast")];
            tensor<int32, [4]> var_39960_begin_0 = const()[name = tensor<string, []>("op_39960_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_39960_end_0 = const()[name = tensor<string, []>("op_39960_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_39960_end_mask_0 = const()[name = tensor<string, []>("op_39960_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39960_cast = slice_by_index(begin = var_39960_begin_0, end = var_39960_end_0, end_mask = var_39960_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39960_cast")];
            tensor<int32, [4]> var_39964_begin_0 = const()[name = tensor<string, []>("op_39964_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_39964_end_0 = const()[name = tensor<string, []>("op_39964_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_39964_end_mask_0 = const()[name = tensor<string, []>("op_39964_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39964_cast = slice_by_index(begin = var_39964_begin_0, end = var_39964_end_0, end_mask = var_39964_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39964_cast")];
            tensor<int32, [4]> var_39968_begin_0 = const()[name = tensor<string, []>("op_39968_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_39968_end_0 = const()[name = tensor<string, []>("op_39968_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_39968_end_mask_0 = const()[name = tensor<string, []>("op_39968_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39968_cast = slice_by_index(begin = var_39968_begin_0, end = var_39968_end_0, end_mask = var_39968_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39968_cast")];
            tensor<int32, [4]> var_39972_begin_0 = const()[name = tensor<string, []>("op_39972_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_39972_end_0 = const()[name = tensor<string, []>("op_39972_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_39972_end_mask_0 = const()[name = tensor<string, []>("op_39972_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39972_cast = slice_by_index(begin = var_39972_begin_0, end = var_39972_end_0, end_mask = var_39972_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39972_cast")];
            tensor<int32, [4]> var_39976_begin_0 = const()[name = tensor<string, []>("op_39976_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_39976_end_0 = const()[name = tensor<string, []>("op_39976_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_39976_end_mask_0 = const()[name = tensor<string, []>("op_39976_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39976_cast = slice_by_index(begin = var_39976_begin_0, end = var_39976_end_0, end_mask = var_39976_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39976_cast")];
            tensor<int32, [4]> var_39980_begin_0 = const()[name = tensor<string, []>("op_39980_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_39980_end_0 = const()[name = tensor<string, []>("op_39980_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_39980_end_mask_0 = const()[name = tensor<string, []>("op_39980_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39980_cast = slice_by_index(begin = var_39980_begin_0, end = var_39980_end_0, end_mask = var_39980_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39980_cast")];
            tensor<int32, [4]> var_39984_begin_0 = const()[name = tensor<string, []>("op_39984_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_39984_end_0 = const()[name = tensor<string, []>("op_39984_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_39984_end_mask_0 = const()[name = tensor<string, []>("op_39984_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_39984_cast = slice_by_index(begin = var_39984_begin_0, end = var_39984_end_0, end_mask = var_39984_end_mask_0, x = v_177_cast)[name = tensor<string, []>("op_39984_cast")];
            tensor<string, []> var_39988_equation_0 = const()[name = tensor<string, []>("op_39988_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39988_cast = einsum(equation = var_39988_equation_0, values = (var_39830_cast, var_39747_cast))[name = tensor<string, []>("op_39988_cast")];
            tensor<fp16, []> var_39989_to_fp16 = const()[name = tensor<string, []>("op_39989_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3361_cast = mul(x = var_39988_cast, y = var_39989_to_fp16)[name = tensor<string, []>("aw_3361_cast")];
            tensor<string, []> var_39992_equation_0 = const()[name = tensor<string, []>("op_39992_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39992_cast = einsum(equation = var_39992_equation_0, values = (var_39834_cast, var_39751_cast))[name = tensor<string, []>("op_39992_cast")];
            tensor<fp16, []> var_39993_to_fp16 = const()[name = tensor<string, []>("op_39993_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3363_cast = mul(x = var_39992_cast, y = var_39993_to_fp16)[name = tensor<string, []>("aw_3363_cast")];
            tensor<string, []> var_39996_equation_0 = const()[name = tensor<string, []>("op_39996_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_39996_cast = einsum(equation = var_39996_equation_0, values = (var_39838_cast, var_39755_cast))[name = tensor<string, []>("op_39996_cast")];
            tensor<fp16, []> var_39997_to_fp16 = const()[name = tensor<string, []>("op_39997_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3365_cast = mul(x = var_39996_cast, y = var_39997_to_fp16)[name = tensor<string, []>("aw_3365_cast")];
            tensor<string, []> var_40000_equation_0 = const()[name = tensor<string, []>("op_40000_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40000_cast = einsum(equation = var_40000_equation_0, values = (var_39842_cast, var_39759_cast))[name = tensor<string, []>("op_40000_cast")];
            tensor<fp16, []> var_40001_to_fp16 = const()[name = tensor<string, []>("op_40001_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3367_cast = mul(x = var_40000_cast, y = var_40001_to_fp16)[name = tensor<string, []>("aw_3367_cast")];
            tensor<string, []> var_40004_equation_0 = const()[name = tensor<string, []>("op_40004_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40004_cast = einsum(equation = var_40004_equation_0, values = (var_39846_cast, var_39763_cast))[name = tensor<string, []>("op_40004_cast")];
            tensor<fp16, []> var_40005_to_fp16 = const()[name = tensor<string, []>("op_40005_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3369_cast = mul(x = var_40004_cast, y = var_40005_to_fp16)[name = tensor<string, []>("aw_3369_cast")];
            tensor<string, []> var_40008_equation_0 = const()[name = tensor<string, []>("op_40008_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40008_cast = einsum(equation = var_40008_equation_0, values = (var_39850_cast, var_39767_cast))[name = tensor<string, []>("op_40008_cast")];
            tensor<fp16, []> var_40009_to_fp16 = const()[name = tensor<string, []>("op_40009_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3371_cast = mul(x = var_40008_cast, y = var_40009_to_fp16)[name = tensor<string, []>("aw_3371_cast")];
            tensor<string, []> var_40012_equation_0 = const()[name = tensor<string, []>("op_40012_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40012_cast = einsum(equation = var_40012_equation_0, values = (var_39854_cast, var_39771_cast))[name = tensor<string, []>("op_40012_cast")];
            tensor<fp16, []> var_40013_to_fp16 = const()[name = tensor<string, []>("op_40013_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3373_cast = mul(x = var_40012_cast, y = var_40013_to_fp16)[name = tensor<string, []>("aw_3373_cast")];
            tensor<string, []> var_40016_equation_0 = const()[name = tensor<string, []>("op_40016_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40016_cast = einsum(equation = var_40016_equation_0, values = (var_39858_cast, var_39775_cast))[name = tensor<string, []>("op_40016_cast")];
            tensor<fp16, []> var_40017_to_fp16 = const()[name = tensor<string, []>("op_40017_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3375_cast = mul(x = var_40016_cast, y = var_40017_to_fp16)[name = tensor<string, []>("aw_3375_cast")];
            tensor<string, []> var_40020_equation_0 = const()[name = tensor<string, []>("op_40020_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40020_cast = einsum(equation = var_40020_equation_0, values = (var_39862_cast, var_39779_cast))[name = tensor<string, []>("op_40020_cast")];
            tensor<fp16, []> var_40021_to_fp16 = const()[name = tensor<string, []>("op_40021_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3377_cast = mul(x = var_40020_cast, y = var_40021_to_fp16)[name = tensor<string, []>("aw_3377_cast")];
            tensor<string, []> var_40024_equation_0 = const()[name = tensor<string, []>("op_40024_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40024_cast = einsum(equation = var_40024_equation_0, values = (var_39866_cast, var_39783_cast))[name = tensor<string, []>("op_40024_cast")];
            tensor<fp16, []> var_40025_to_fp16 = const()[name = tensor<string, []>("op_40025_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3379_cast = mul(x = var_40024_cast, y = var_40025_to_fp16)[name = tensor<string, []>("aw_3379_cast")];
            tensor<string, []> var_40028_equation_0 = const()[name = tensor<string, []>("op_40028_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40028_cast = einsum(equation = var_40028_equation_0, values = (var_39870_cast, var_39787_cast))[name = tensor<string, []>("op_40028_cast")];
            tensor<fp16, []> var_40029_to_fp16 = const()[name = tensor<string, []>("op_40029_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3381_cast = mul(x = var_40028_cast, y = var_40029_to_fp16)[name = tensor<string, []>("aw_3381_cast")];
            tensor<string, []> var_40032_equation_0 = const()[name = tensor<string, []>("op_40032_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40032_cast = einsum(equation = var_40032_equation_0, values = (var_39874_cast, var_39791_cast))[name = tensor<string, []>("op_40032_cast")];
            tensor<fp16, []> var_40033_to_fp16 = const()[name = tensor<string, []>("op_40033_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3383_cast = mul(x = var_40032_cast, y = var_40033_to_fp16)[name = tensor<string, []>("aw_3383_cast")];
            tensor<string, []> var_40036_equation_0 = const()[name = tensor<string, []>("op_40036_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40036_cast = einsum(equation = var_40036_equation_0, values = (var_39878_cast, var_39795_cast))[name = tensor<string, []>("op_40036_cast")];
            tensor<fp16, []> var_40037_to_fp16 = const()[name = tensor<string, []>("op_40037_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3385_cast = mul(x = var_40036_cast, y = var_40037_to_fp16)[name = tensor<string, []>("aw_3385_cast")];
            tensor<string, []> var_40040_equation_0 = const()[name = tensor<string, []>("op_40040_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40040_cast = einsum(equation = var_40040_equation_0, values = (var_39882_cast, var_39799_cast))[name = tensor<string, []>("op_40040_cast")];
            tensor<fp16, []> var_40041_to_fp16 = const()[name = tensor<string, []>("op_40041_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3387_cast = mul(x = var_40040_cast, y = var_40041_to_fp16)[name = tensor<string, []>("aw_3387_cast")];
            tensor<string, []> var_40044_equation_0 = const()[name = tensor<string, []>("op_40044_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40044_cast = einsum(equation = var_40044_equation_0, values = (var_39886_cast, var_39803_cast))[name = tensor<string, []>("op_40044_cast")];
            tensor<fp16, []> var_40045_to_fp16 = const()[name = tensor<string, []>("op_40045_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3389_cast = mul(x = var_40044_cast, y = var_40045_to_fp16)[name = tensor<string, []>("aw_3389_cast")];
            tensor<string, []> var_40048_equation_0 = const()[name = tensor<string, []>("op_40048_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40048_cast = einsum(equation = var_40048_equation_0, values = (var_39890_cast, var_39807_cast))[name = tensor<string, []>("op_40048_cast")];
            tensor<fp16, []> var_40049_to_fp16 = const()[name = tensor<string, []>("op_40049_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3391_cast = mul(x = var_40048_cast, y = var_40049_to_fp16)[name = tensor<string, []>("aw_3391_cast")];
            tensor<string, []> var_40052_equation_0 = const()[name = tensor<string, []>("op_40052_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40052_cast = einsum(equation = var_40052_equation_0, values = (var_39894_cast, var_39811_cast))[name = tensor<string, []>("op_40052_cast")];
            tensor<fp16, []> var_40053_to_fp16 = const()[name = tensor<string, []>("op_40053_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3393_cast = mul(x = var_40052_cast, y = var_40053_to_fp16)[name = tensor<string, []>("aw_3393_cast")];
            tensor<string, []> var_40056_equation_0 = const()[name = tensor<string, []>("op_40056_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40056_cast = einsum(equation = var_40056_equation_0, values = (var_39898_cast, var_39815_cast))[name = tensor<string, []>("op_40056_cast")];
            tensor<fp16, []> var_40057_to_fp16 = const()[name = tensor<string, []>("op_40057_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3395_cast = mul(x = var_40056_cast, y = var_40057_to_fp16)[name = tensor<string, []>("aw_3395_cast")];
            tensor<string, []> var_40060_equation_0 = const()[name = tensor<string, []>("op_40060_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40060_cast = einsum(equation = var_40060_equation_0, values = (var_39902_cast, var_39819_cast))[name = tensor<string, []>("op_40060_cast")];
            tensor<fp16, []> var_40061_to_fp16 = const()[name = tensor<string, []>("op_40061_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3397_cast = mul(x = var_40060_cast, y = var_40061_to_fp16)[name = tensor<string, []>("aw_3397_cast")];
            tensor<string, []> var_40064_equation_0 = const()[name = tensor<string, []>("op_40064_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40064_cast = einsum(equation = var_40064_equation_0, values = (var_39906_cast, var_39823_cast))[name = tensor<string, []>("op_40064_cast")];
            tensor<fp16, []> var_40065_to_fp16 = const()[name = tensor<string, []>("op_40065_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3399_cast = mul(x = var_40064_cast, y = var_40065_to_fp16)[name = tensor<string, []>("aw_3399_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40067_cast = softmax(axis = var_30385, x = aw_3361_cast)[name = tensor<string, []>("op_40067_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40068_cast = softmax(axis = var_30385, x = aw_3363_cast)[name = tensor<string, []>("op_40068_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40069_cast = softmax(axis = var_30385, x = aw_3365_cast)[name = tensor<string, []>("op_40069_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40070_cast = softmax(axis = var_30385, x = aw_3367_cast)[name = tensor<string, []>("op_40070_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40071_cast = softmax(axis = var_30385, x = aw_3369_cast)[name = tensor<string, []>("op_40071_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40072_cast = softmax(axis = var_30385, x = aw_3371_cast)[name = tensor<string, []>("op_40072_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40073_cast = softmax(axis = var_30385, x = aw_3373_cast)[name = tensor<string, []>("op_40073_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40074_cast = softmax(axis = var_30385, x = aw_3375_cast)[name = tensor<string, []>("op_40074_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40075_cast = softmax(axis = var_30385, x = aw_3377_cast)[name = tensor<string, []>("op_40075_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40076_cast = softmax(axis = var_30385, x = aw_3379_cast)[name = tensor<string, []>("op_40076_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40077_cast = softmax(axis = var_30385, x = aw_3381_cast)[name = tensor<string, []>("op_40077_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40078_cast = softmax(axis = var_30385, x = aw_3383_cast)[name = tensor<string, []>("op_40078_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40079_cast = softmax(axis = var_30385, x = aw_3385_cast)[name = tensor<string, []>("op_40079_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40080_cast = softmax(axis = var_30385, x = aw_3387_cast)[name = tensor<string, []>("op_40080_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40081_cast = softmax(axis = var_30385, x = aw_3389_cast)[name = tensor<string, []>("op_40081_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40082_cast = softmax(axis = var_30385, x = aw_3391_cast)[name = tensor<string, []>("op_40082_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40083_cast = softmax(axis = var_30385, x = aw_3393_cast)[name = tensor<string, []>("op_40083_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40084_cast = softmax(axis = var_30385, x = aw_3395_cast)[name = tensor<string, []>("op_40084_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40085_cast = softmax(axis = var_30385, x = aw_3397_cast)[name = tensor<string, []>("op_40085_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40086_cast = softmax(axis = var_30385, x = aw_3399_cast)[name = tensor<string, []>("op_40086_cast")];
            tensor<string, []> var_40088_equation_0 = const()[name = tensor<string, []>("op_40088_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40088_cast = einsum(equation = var_40088_equation_0, values = (var_39908_cast, var_40067_cast))[name = tensor<string, []>("op_40088_cast")];
            tensor<string, []> var_40090_equation_0 = const()[name = tensor<string, []>("op_40090_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40090_cast = einsum(equation = var_40090_equation_0, values = (var_39912_cast, var_40068_cast))[name = tensor<string, []>("op_40090_cast")];
            tensor<string, []> var_40092_equation_0 = const()[name = tensor<string, []>("op_40092_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40092_cast = einsum(equation = var_40092_equation_0, values = (var_39916_cast, var_40069_cast))[name = tensor<string, []>("op_40092_cast")];
            tensor<string, []> var_40094_equation_0 = const()[name = tensor<string, []>("op_40094_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40094_cast = einsum(equation = var_40094_equation_0, values = (var_39920_cast, var_40070_cast))[name = tensor<string, []>("op_40094_cast")];
            tensor<string, []> var_40096_equation_0 = const()[name = tensor<string, []>("op_40096_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40096_cast = einsum(equation = var_40096_equation_0, values = (var_39924_cast, var_40071_cast))[name = tensor<string, []>("op_40096_cast")];
            tensor<string, []> var_40098_equation_0 = const()[name = tensor<string, []>("op_40098_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40098_cast = einsum(equation = var_40098_equation_0, values = (var_39928_cast, var_40072_cast))[name = tensor<string, []>("op_40098_cast")];
            tensor<string, []> var_40100_equation_0 = const()[name = tensor<string, []>("op_40100_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40100_cast = einsum(equation = var_40100_equation_0, values = (var_39932_cast, var_40073_cast))[name = tensor<string, []>("op_40100_cast")];
            tensor<string, []> var_40102_equation_0 = const()[name = tensor<string, []>("op_40102_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40102_cast = einsum(equation = var_40102_equation_0, values = (var_39936_cast, var_40074_cast))[name = tensor<string, []>("op_40102_cast")];
            tensor<string, []> var_40104_equation_0 = const()[name = tensor<string, []>("op_40104_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40104_cast = einsum(equation = var_40104_equation_0, values = (var_39940_cast, var_40075_cast))[name = tensor<string, []>("op_40104_cast")];
            tensor<string, []> var_40106_equation_0 = const()[name = tensor<string, []>("op_40106_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40106_cast = einsum(equation = var_40106_equation_0, values = (var_39944_cast, var_40076_cast))[name = tensor<string, []>("op_40106_cast")];
            tensor<string, []> var_40108_equation_0 = const()[name = tensor<string, []>("op_40108_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40108_cast = einsum(equation = var_40108_equation_0, values = (var_39948_cast, var_40077_cast))[name = tensor<string, []>("op_40108_cast")];
            tensor<string, []> var_40110_equation_0 = const()[name = tensor<string, []>("op_40110_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40110_cast = einsum(equation = var_40110_equation_0, values = (var_39952_cast, var_40078_cast))[name = tensor<string, []>("op_40110_cast")];
            tensor<string, []> var_40112_equation_0 = const()[name = tensor<string, []>("op_40112_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40112_cast = einsum(equation = var_40112_equation_0, values = (var_39956_cast, var_40079_cast))[name = tensor<string, []>("op_40112_cast")];
            tensor<string, []> var_40114_equation_0 = const()[name = tensor<string, []>("op_40114_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40114_cast = einsum(equation = var_40114_equation_0, values = (var_39960_cast, var_40080_cast))[name = tensor<string, []>("op_40114_cast")];
            tensor<string, []> var_40116_equation_0 = const()[name = tensor<string, []>("op_40116_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40116_cast = einsum(equation = var_40116_equation_0, values = (var_39964_cast, var_40081_cast))[name = tensor<string, []>("op_40116_cast")];
            tensor<string, []> var_40118_equation_0 = const()[name = tensor<string, []>("op_40118_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40118_cast = einsum(equation = var_40118_equation_0, values = (var_39968_cast, var_40082_cast))[name = tensor<string, []>("op_40118_cast")];
            tensor<string, []> var_40120_equation_0 = const()[name = tensor<string, []>("op_40120_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40120_cast = einsum(equation = var_40120_equation_0, values = (var_39972_cast, var_40083_cast))[name = tensor<string, []>("op_40120_cast")];
            tensor<string, []> var_40122_equation_0 = const()[name = tensor<string, []>("op_40122_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40122_cast = einsum(equation = var_40122_equation_0, values = (var_39976_cast, var_40084_cast))[name = tensor<string, []>("op_40122_cast")];
            tensor<string, []> var_40124_equation_0 = const()[name = tensor<string, []>("op_40124_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40124_cast = einsum(equation = var_40124_equation_0, values = (var_39980_cast, var_40085_cast))[name = tensor<string, []>("op_40124_cast")];
            tensor<string, []> var_40126_equation_0 = const()[name = tensor<string, []>("op_40126_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40126_cast = einsum(equation = var_40126_equation_0, values = (var_39984_cast, var_40086_cast))[name = tensor<string, []>("op_40126_cast")];
            tensor<bool, []> input_539_interleave_0 = const()[name = tensor<string, []>("input_539_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_539_cast = concat(axis = var_30385, interleave = input_539_interleave_0, values = (var_40088_cast, var_40090_cast, var_40092_cast, var_40094_cast, var_40096_cast, var_40098_cast, var_40100_cast, var_40102_cast, var_40104_cast, var_40106_cast, var_40108_cast, var_40110_cast, var_40112_cast, var_40114_cast, var_40116_cast, var_40118_cast, var_40120_cast, var_40122_cast, var_40124_cast, var_40126_cast))[name = tensor<string, []>("input_539_cast")];
            tensor<int32, [2]> var_40132 = const()[name = tensor<string, []>("op_40132"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_40134 = const()[name = tensor<string, []>("op_40134"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_40136_pad_type_0 = const()[name = tensor<string, []>("op_40136_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_40136_pad_0 = const()[name = tensor<string, []>("op_40136_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3398281408)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3401558272)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_40136_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_40134, groups = var_30385, pad = var_40136_pad_0, pad_type = var_40136_pad_type_0, strides = var_40132, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_539_cast)[name = tensor<string, []>("op_40136_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_267_cast = add(x = var_40136_cast, y = inputs_265_cast)[name = tensor<string, []>("inputs_267_cast")];
            tensor<int32, [1]> var_40140 = const()[name = tensor<string, []>("op_40140"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_267_cast = reduce_mean(axes = var_40140, keep_dims = var_30380, x = inputs_267_cast)[name = tensor<string, []>("channels_mean_267_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_267_cast = sub(x = inputs_267_cast, y = channels_mean_267_cast)[name = tensor<string, []>("zero_mean_267_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_267_cast = mul(x = zero_mean_267_cast, y = zero_mean_267_cast)[name = tensor<string, []>("zero_mean_sq_267_cast")];
            tensor<int32, [1]> var_40144 = const()[name = tensor<string, []>("op_40144"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_40145_cast = reduce_mean(axes = var_40144, keep_dims = var_30380, x = zero_mean_sq_267_cast)[name = tensor<string, []>("op_40145_cast")];
            tensor<fp16, []> var_40146_to_fp16 = const()[name = tensor<string, []>("op_40146_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_40147_cast = add(x = var_40145_cast, y = var_40146_to_fp16)[name = tensor<string, []>("op_40147_cast")];
            tensor<fp16, []> denom_267_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_267_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_267_cast = rsqrt(epsilon = denom_267_epsilon_0_to_fp16, x = var_40147_cast)[name = tensor<string, []>("denom_267_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_267_cast = mul(x = zero_mean_267_cast, y = denom_267_cast)[name = tensor<string, []>("out_267_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_40151_to_fp16 = const()[name = tensor<string, []>("op_40151_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3401560896)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_40152_cast = add(x = out_267_cast, y = var_40151_to_fp16)[name = tensor<string, []>("op_40152_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_40154_to_fp16 = const()[name = tensor<string, []>("op_40154_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3401563520)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_363_cast = mul(x = var_40152_cast, y = var_40154_to_fp16)[name = tensor<string, []>("hidden_states_363_cast")];
            tensor<int32, [2]> var_40161 = const()[name = tensor<string, []>("op_40161"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_40163 = const()[name = tensor<string, []>("op_40163"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_179_pad_type_0 = const()[name = tensor<string, []>("q_179_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_179_pad_0 = const()[name = tensor<string, []>("q_179_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3401566144)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_179_cast = conv(dilations = var_40163, groups = var_30385, pad = q_179_pad_0, pad_type = q_179_pad_type_0, strides = var_40161, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_363_cast)[name = tensor<string, []>("q_179_cast")];
            tensor<int32, [2]> var_40167 = const()[name = tensor<string, []>("op_40167"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_40169 = const()[name = tensor<string, []>("op_40169"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_357_pad_type_0 = const()[name = tensor<string, []>("k_357_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_357_pad_0 = const()[name = tensor<string, []>("k_357_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3404843008)))];
            tensor<fp16, [2, 1280, 1, 77]> k_357_cast = conv(dilations = var_40169, groups = var_30385, pad = k_357_pad_0, pad_type = k_357_pad_type_0, strides = var_40167, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_357_cast")];
            tensor<int32, [2]> var_40173 = const()[name = tensor<string, []>("op_40173"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_40175 = const()[name = tensor<string, []>("op_40175"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_179_pad_type_0 = const()[name = tensor<string, []>("v_179_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_179_pad_0 = const()[name = tensor<string, []>("v_179_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3410085952)))];
            tensor<fp16, [2, 1280, 1, 77]> v_179_cast = conv(dilations = var_40175, groups = var_30385, pad = v_179_pad_0, pad_type = v_179_pad_type_0, strides = var_40173, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_179_cast")];
            tensor<int32, [4]> var_40179_begin_0 = const()[name = tensor<string, []>("op_40179_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_40179_end_0 = const()[name = tensor<string, []>("op_40179_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_40179_end_mask_0 = const()[name = tensor<string, []>("op_40179_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40179_cast = slice_by_index(begin = var_40179_begin_0, end = var_40179_end_0, end_mask = var_40179_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40179_cast")];
            tensor<int32, [4]> var_40183_begin_0 = const()[name = tensor<string, []>("op_40183_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_40183_end_0 = const()[name = tensor<string, []>("op_40183_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_40183_end_mask_0 = const()[name = tensor<string, []>("op_40183_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40183_cast = slice_by_index(begin = var_40183_begin_0, end = var_40183_end_0, end_mask = var_40183_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40183_cast")];
            tensor<int32, [4]> var_40187_begin_0 = const()[name = tensor<string, []>("op_40187_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_40187_end_0 = const()[name = tensor<string, []>("op_40187_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_40187_end_mask_0 = const()[name = tensor<string, []>("op_40187_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40187_cast = slice_by_index(begin = var_40187_begin_0, end = var_40187_end_0, end_mask = var_40187_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40187_cast")];
            tensor<int32, [4]> var_40191_begin_0 = const()[name = tensor<string, []>("op_40191_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_40191_end_0 = const()[name = tensor<string, []>("op_40191_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_40191_end_mask_0 = const()[name = tensor<string, []>("op_40191_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40191_cast = slice_by_index(begin = var_40191_begin_0, end = var_40191_end_0, end_mask = var_40191_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40191_cast")];
            tensor<int32, [4]> var_40195_begin_0 = const()[name = tensor<string, []>("op_40195_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_40195_end_0 = const()[name = tensor<string, []>("op_40195_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_40195_end_mask_0 = const()[name = tensor<string, []>("op_40195_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40195_cast = slice_by_index(begin = var_40195_begin_0, end = var_40195_end_0, end_mask = var_40195_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40195_cast")];
            tensor<int32, [4]> var_40199_begin_0 = const()[name = tensor<string, []>("op_40199_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_40199_end_0 = const()[name = tensor<string, []>("op_40199_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_40199_end_mask_0 = const()[name = tensor<string, []>("op_40199_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40199_cast = slice_by_index(begin = var_40199_begin_0, end = var_40199_end_0, end_mask = var_40199_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40199_cast")];
            tensor<int32, [4]> var_40203_begin_0 = const()[name = tensor<string, []>("op_40203_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_40203_end_0 = const()[name = tensor<string, []>("op_40203_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_40203_end_mask_0 = const()[name = tensor<string, []>("op_40203_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40203_cast = slice_by_index(begin = var_40203_begin_0, end = var_40203_end_0, end_mask = var_40203_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40203_cast")];
            tensor<int32, [4]> var_40207_begin_0 = const()[name = tensor<string, []>("op_40207_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_40207_end_0 = const()[name = tensor<string, []>("op_40207_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_40207_end_mask_0 = const()[name = tensor<string, []>("op_40207_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40207_cast = slice_by_index(begin = var_40207_begin_0, end = var_40207_end_0, end_mask = var_40207_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40207_cast")];
            tensor<int32, [4]> var_40211_begin_0 = const()[name = tensor<string, []>("op_40211_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_40211_end_0 = const()[name = tensor<string, []>("op_40211_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_40211_end_mask_0 = const()[name = tensor<string, []>("op_40211_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40211_cast = slice_by_index(begin = var_40211_begin_0, end = var_40211_end_0, end_mask = var_40211_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40211_cast")];
            tensor<int32, [4]> var_40215_begin_0 = const()[name = tensor<string, []>("op_40215_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_40215_end_0 = const()[name = tensor<string, []>("op_40215_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_40215_end_mask_0 = const()[name = tensor<string, []>("op_40215_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40215_cast = slice_by_index(begin = var_40215_begin_0, end = var_40215_end_0, end_mask = var_40215_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40215_cast")];
            tensor<int32, [4]> var_40219_begin_0 = const()[name = tensor<string, []>("op_40219_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_40219_end_0 = const()[name = tensor<string, []>("op_40219_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_40219_end_mask_0 = const()[name = tensor<string, []>("op_40219_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40219_cast = slice_by_index(begin = var_40219_begin_0, end = var_40219_end_0, end_mask = var_40219_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40219_cast")];
            tensor<int32, [4]> var_40223_begin_0 = const()[name = tensor<string, []>("op_40223_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_40223_end_0 = const()[name = tensor<string, []>("op_40223_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_40223_end_mask_0 = const()[name = tensor<string, []>("op_40223_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40223_cast = slice_by_index(begin = var_40223_begin_0, end = var_40223_end_0, end_mask = var_40223_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40223_cast")];
            tensor<int32, [4]> var_40227_begin_0 = const()[name = tensor<string, []>("op_40227_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_40227_end_0 = const()[name = tensor<string, []>("op_40227_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_40227_end_mask_0 = const()[name = tensor<string, []>("op_40227_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40227_cast = slice_by_index(begin = var_40227_begin_0, end = var_40227_end_0, end_mask = var_40227_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40227_cast")];
            tensor<int32, [4]> var_40231_begin_0 = const()[name = tensor<string, []>("op_40231_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_40231_end_0 = const()[name = tensor<string, []>("op_40231_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_40231_end_mask_0 = const()[name = tensor<string, []>("op_40231_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40231_cast = slice_by_index(begin = var_40231_begin_0, end = var_40231_end_0, end_mask = var_40231_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40231_cast")];
            tensor<int32, [4]> var_40235_begin_0 = const()[name = tensor<string, []>("op_40235_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_40235_end_0 = const()[name = tensor<string, []>("op_40235_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_40235_end_mask_0 = const()[name = tensor<string, []>("op_40235_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40235_cast = slice_by_index(begin = var_40235_begin_0, end = var_40235_end_0, end_mask = var_40235_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40235_cast")];
            tensor<int32, [4]> var_40239_begin_0 = const()[name = tensor<string, []>("op_40239_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_40239_end_0 = const()[name = tensor<string, []>("op_40239_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_40239_end_mask_0 = const()[name = tensor<string, []>("op_40239_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40239_cast = slice_by_index(begin = var_40239_begin_0, end = var_40239_end_0, end_mask = var_40239_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40239_cast")];
            tensor<int32, [4]> var_40243_begin_0 = const()[name = tensor<string, []>("op_40243_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_40243_end_0 = const()[name = tensor<string, []>("op_40243_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_40243_end_mask_0 = const()[name = tensor<string, []>("op_40243_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40243_cast = slice_by_index(begin = var_40243_begin_0, end = var_40243_end_0, end_mask = var_40243_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40243_cast")];
            tensor<int32, [4]> var_40247_begin_0 = const()[name = tensor<string, []>("op_40247_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_40247_end_0 = const()[name = tensor<string, []>("op_40247_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_40247_end_mask_0 = const()[name = tensor<string, []>("op_40247_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40247_cast = slice_by_index(begin = var_40247_begin_0, end = var_40247_end_0, end_mask = var_40247_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40247_cast")];
            tensor<int32, [4]> var_40251_begin_0 = const()[name = tensor<string, []>("op_40251_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_40251_end_0 = const()[name = tensor<string, []>("op_40251_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_40251_end_mask_0 = const()[name = tensor<string, []>("op_40251_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40251_cast = slice_by_index(begin = var_40251_begin_0, end = var_40251_end_0, end_mask = var_40251_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40251_cast")];
            tensor<int32, [4]> var_40255_begin_0 = const()[name = tensor<string, []>("op_40255_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_40255_end_0 = const()[name = tensor<string, []>("op_40255_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_40255_end_mask_0 = const()[name = tensor<string, []>("op_40255_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40255_cast = slice_by_index(begin = var_40255_begin_0, end = var_40255_end_0, end_mask = var_40255_end_mask_0, x = q_179_cast)[name = tensor<string, []>("op_40255_cast")];
            tensor<int32, [4]> k_359_perm_0 = const()[name = tensor<string, []>("k_359_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_40262_begin_0 = const()[name = tensor<string, []>("op_40262_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_40262_end_0 = const()[name = tensor<string, []>("op_40262_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_40262_end_mask_0 = const()[name = tensor<string, []>("op_40262_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_50 = transpose(perm = k_359_perm_0, x = k_357_cast)[name = tensor<string, []>("transpose_50")];
            tensor<fp16, [2, 77, 1, 64]> var_40262_cast = slice_by_index(begin = var_40262_begin_0, end = var_40262_end_0, end_mask = var_40262_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40262_cast")];
            tensor<int32, [4]> var_40266_begin_0 = const()[name = tensor<string, []>("op_40266_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_40266_end_0 = const()[name = tensor<string, []>("op_40266_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_40266_end_mask_0 = const()[name = tensor<string, []>("op_40266_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40266_cast = slice_by_index(begin = var_40266_begin_0, end = var_40266_end_0, end_mask = var_40266_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40266_cast")];
            tensor<int32, [4]> var_40270_begin_0 = const()[name = tensor<string, []>("op_40270_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_40270_end_0 = const()[name = tensor<string, []>("op_40270_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_40270_end_mask_0 = const()[name = tensor<string, []>("op_40270_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40270_cast = slice_by_index(begin = var_40270_begin_0, end = var_40270_end_0, end_mask = var_40270_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40270_cast")];
            tensor<int32, [4]> var_40274_begin_0 = const()[name = tensor<string, []>("op_40274_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_40274_end_0 = const()[name = tensor<string, []>("op_40274_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_40274_end_mask_0 = const()[name = tensor<string, []>("op_40274_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40274_cast = slice_by_index(begin = var_40274_begin_0, end = var_40274_end_0, end_mask = var_40274_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40274_cast")];
            tensor<int32, [4]> var_40278_begin_0 = const()[name = tensor<string, []>("op_40278_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_40278_end_0 = const()[name = tensor<string, []>("op_40278_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_40278_end_mask_0 = const()[name = tensor<string, []>("op_40278_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40278_cast = slice_by_index(begin = var_40278_begin_0, end = var_40278_end_0, end_mask = var_40278_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40278_cast")];
            tensor<int32, [4]> var_40282_begin_0 = const()[name = tensor<string, []>("op_40282_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_40282_end_0 = const()[name = tensor<string, []>("op_40282_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_40282_end_mask_0 = const()[name = tensor<string, []>("op_40282_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40282_cast = slice_by_index(begin = var_40282_begin_0, end = var_40282_end_0, end_mask = var_40282_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40282_cast")];
            tensor<int32, [4]> var_40286_begin_0 = const()[name = tensor<string, []>("op_40286_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_40286_end_0 = const()[name = tensor<string, []>("op_40286_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_40286_end_mask_0 = const()[name = tensor<string, []>("op_40286_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40286_cast = slice_by_index(begin = var_40286_begin_0, end = var_40286_end_0, end_mask = var_40286_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40286_cast")];
            tensor<int32, [4]> var_40290_begin_0 = const()[name = tensor<string, []>("op_40290_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_40290_end_0 = const()[name = tensor<string, []>("op_40290_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_40290_end_mask_0 = const()[name = tensor<string, []>("op_40290_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40290_cast = slice_by_index(begin = var_40290_begin_0, end = var_40290_end_0, end_mask = var_40290_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40290_cast")];
            tensor<int32, [4]> var_40294_begin_0 = const()[name = tensor<string, []>("op_40294_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_40294_end_0 = const()[name = tensor<string, []>("op_40294_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_40294_end_mask_0 = const()[name = tensor<string, []>("op_40294_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40294_cast = slice_by_index(begin = var_40294_begin_0, end = var_40294_end_0, end_mask = var_40294_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40294_cast")];
            tensor<int32, [4]> var_40298_begin_0 = const()[name = tensor<string, []>("op_40298_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_40298_end_0 = const()[name = tensor<string, []>("op_40298_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_40298_end_mask_0 = const()[name = tensor<string, []>("op_40298_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40298_cast = slice_by_index(begin = var_40298_begin_0, end = var_40298_end_0, end_mask = var_40298_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40298_cast")];
            tensor<int32, [4]> var_40302_begin_0 = const()[name = tensor<string, []>("op_40302_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_40302_end_0 = const()[name = tensor<string, []>("op_40302_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_40302_end_mask_0 = const()[name = tensor<string, []>("op_40302_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40302_cast = slice_by_index(begin = var_40302_begin_0, end = var_40302_end_0, end_mask = var_40302_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40302_cast")];
            tensor<int32, [4]> var_40306_begin_0 = const()[name = tensor<string, []>("op_40306_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_40306_end_0 = const()[name = tensor<string, []>("op_40306_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_40306_end_mask_0 = const()[name = tensor<string, []>("op_40306_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40306_cast = slice_by_index(begin = var_40306_begin_0, end = var_40306_end_0, end_mask = var_40306_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40306_cast")];
            tensor<int32, [4]> var_40310_begin_0 = const()[name = tensor<string, []>("op_40310_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_40310_end_0 = const()[name = tensor<string, []>("op_40310_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_40310_end_mask_0 = const()[name = tensor<string, []>("op_40310_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40310_cast = slice_by_index(begin = var_40310_begin_0, end = var_40310_end_0, end_mask = var_40310_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40310_cast")];
            tensor<int32, [4]> var_40314_begin_0 = const()[name = tensor<string, []>("op_40314_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_40314_end_0 = const()[name = tensor<string, []>("op_40314_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_40314_end_mask_0 = const()[name = tensor<string, []>("op_40314_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40314_cast = slice_by_index(begin = var_40314_begin_0, end = var_40314_end_0, end_mask = var_40314_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40314_cast")];
            tensor<int32, [4]> var_40318_begin_0 = const()[name = tensor<string, []>("op_40318_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_40318_end_0 = const()[name = tensor<string, []>("op_40318_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_40318_end_mask_0 = const()[name = tensor<string, []>("op_40318_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40318_cast = slice_by_index(begin = var_40318_begin_0, end = var_40318_end_0, end_mask = var_40318_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40318_cast")];
            tensor<int32, [4]> var_40322_begin_0 = const()[name = tensor<string, []>("op_40322_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_40322_end_0 = const()[name = tensor<string, []>("op_40322_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_40322_end_mask_0 = const()[name = tensor<string, []>("op_40322_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40322_cast = slice_by_index(begin = var_40322_begin_0, end = var_40322_end_0, end_mask = var_40322_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40322_cast")];
            tensor<int32, [4]> var_40326_begin_0 = const()[name = tensor<string, []>("op_40326_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_40326_end_0 = const()[name = tensor<string, []>("op_40326_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_40326_end_mask_0 = const()[name = tensor<string, []>("op_40326_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40326_cast = slice_by_index(begin = var_40326_begin_0, end = var_40326_end_0, end_mask = var_40326_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40326_cast")];
            tensor<int32, [4]> var_40330_begin_0 = const()[name = tensor<string, []>("op_40330_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_40330_end_0 = const()[name = tensor<string, []>("op_40330_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_40330_end_mask_0 = const()[name = tensor<string, []>("op_40330_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40330_cast = slice_by_index(begin = var_40330_begin_0, end = var_40330_end_0, end_mask = var_40330_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40330_cast")];
            tensor<int32, [4]> var_40334_begin_0 = const()[name = tensor<string, []>("op_40334_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_40334_end_0 = const()[name = tensor<string, []>("op_40334_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_40334_end_mask_0 = const()[name = tensor<string, []>("op_40334_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40334_cast = slice_by_index(begin = var_40334_begin_0, end = var_40334_end_0, end_mask = var_40334_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40334_cast")];
            tensor<int32, [4]> var_40338_begin_0 = const()[name = tensor<string, []>("op_40338_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_40338_end_0 = const()[name = tensor<string, []>("op_40338_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_40338_end_mask_0 = const()[name = tensor<string, []>("op_40338_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_40338_cast = slice_by_index(begin = var_40338_begin_0, end = var_40338_end_0, end_mask = var_40338_end_mask_0, x = transpose_50)[name = tensor<string, []>("op_40338_cast")];
            tensor<int32, [4]> var_40340_begin_0 = const()[name = tensor<string, []>("op_40340_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_40340_end_0 = const()[name = tensor<string, []>("op_40340_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_40340_end_mask_0 = const()[name = tensor<string, []>("op_40340_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40340_cast = slice_by_index(begin = var_40340_begin_0, end = var_40340_end_0, end_mask = var_40340_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40340_cast")];
            tensor<int32, [4]> var_40344_begin_0 = const()[name = tensor<string, []>("op_40344_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_40344_end_0 = const()[name = tensor<string, []>("op_40344_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_40344_end_mask_0 = const()[name = tensor<string, []>("op_40344_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40344_cast = slice_by_index(begin = var_40344_begin_0, end = var_40344_end_0, end_mask = var_40344_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40344_cast")];
            tensor<int32, [4]> var_40348_begin_0 = const()[name = tensor<string, []>("op_40348_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_40348_end_0 = const()[name = tensor<string, []>("op_40348_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_40348_end_mask_0 = const()[name = tensor<string, []>("op_40348_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40348_cast = slice_by_index(begin = var_40348_begin_0, end = var_40348_end_0, end_mask = var_40348_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40348_cast")];
            tensor<int32, [4]> var_40352_begin_0 = const()[name = tensor<string, []>("op_40352_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_40352_end_0 = const()[name = tensor<string, []>("op_40352_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_40352_end_mask_0 = const()[name = tensor<string, []>("op_40352_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40352_cast = slice_by_index(begin = var_40352_begin_0, end = var_40352_end_0, end_mask = var_40352_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40352_cast")];
            tensor<int32, [4]> var_40356_begin_0 = const()[name = tensor<string, []>("op_40356_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_40356_end_0 = const()[name = tensor<string, []>("op_40356_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_40356_end_mask_0 = const()[name = tensor<string, []>("op_40356_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40356_cast = slice_by_index(begin = var_40356_begin_0, end = var_40356_end_0, end_mask = var_40356_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40356_cast")];
            tensor<int32, [4]> var_40360_begin_0 = const()[name = tensor<string, []>("op_40360_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_40360_end_0 = const()[name = tensor<string, []>("op_40360_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_40360_end_mask_0 = const()[name = tensor<string, []>("op_40360_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40360_cast = slice_by_index(begin = var_40360_begin_0, end = var_40360_end_0, end_mask = var_40360_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40360_cast")];
            tensor<int32, [4]> var_40364_begin_0 = const()[name = tensor<string, []>("op_40364_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_40364_end_0 = const()[name = tensor<string, []>("op_40364_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_40364_end_mask_0 = const()[name = tensor<string, []>("op_40364_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40364_cast = slice_by_index(begin = var_40364_begin_0, end = var_40364_end_0, end_mask = var_40364_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40364_cast")];
            tensor<int32, [4]> var_40368_begin_0 = const()[name = tensor<string, []>("op_40368_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_40368_end_0 = const()[name = tensor<string, []>("op_40368_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_40368_end_mask_0 = const()[name = tensor<string, []>("op_40368_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40368_cast = slice_by_index(begin = var_40368_begin_0, end = var_40368_end_0, end_mask = var_40368_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40368_cast")];
            tensor<int32, [4]> var_40372_begin_0 = const()[name = tensor<string, []>("op_40372_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_40372_end_0 = const()[name = tensor<string, []>("op_40372_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_40372_end_mask_0 = const()[name = tensor<string, []>("op_40372_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40372_cast = slice_by_index(begin = var_40372_begin_0, end = var_40372_end_0, end_mask = var_40372_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40372_cast")];
            tensor<int32, [4]> var_40376_begin_0 = const()[name = tensor<string, []>("op_40376_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_40376_end_0 = const()[name = tensor<string, []>("op_40376_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_40376_end_mask_0 = const()[name = tensor<string, []>("op_40376_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40376_cast = slice_by_index(begin = var_40376_begin_0, end = var_40376_end_0, end_mask = var_40376_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40376_cast")];
            tensor<int32, [4]> var_40380_begin_0 = const()[name = tensor<string, []>("op_40380_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_40380_end_0 = const()[name = tensor<string, []>("op_40380_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_40380_end_mask_0 = const()[name = tensor<string, []>("op_40380_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40380_cast = slice_by_index(begin = var_40380_begin_0, end = var_40380_end_0, end_mask = var_40380_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40380_cast")];
            tensor<int32, [4]> var_40384_begin_0 = const()[name = tensor<string, []>("op_40384_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_40384_end_0 = const()[name = tensor<string, []>("op_40384_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_40384_end_mask_0 = const()[name = tensor<string, []>("op_40384_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40384_cast = slice_by_index(begin = var_40384_begin_0, end = var_40384_end_0, end_mask = var_40384_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40384_cast")];
            tensor<int32, [4]> var_40388_begin_0 = const()[name = tensor<string, []>("op_40388_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_40388_end_0 = const()[name = tensor<string, []>("op_40388_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_40388_end_mask_0 = const()[name = tensor<string, []>("op_40388_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40388_cast = slice_by_index(begin = var_40388_begin_0, end = var_40388_end_0, end_mask = var_40388_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40388_cast")];
            tensor<int32, [4]> var_40392_begin_0 = const()[name = tensor<string, []>("op_40392_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_40392_end_0 = const()[name = tensor<string, []>("op_40392_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_40392_end_mask_0 = const()[name = tensor<string, []>("op_40392_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40392_cast = slice_by_index(begin = var_40392_begin_0, end = var_40392_end_0, end_mask = var_40392_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40392_cast")];
            tensor<int32, [4]> var_40396_begin_0 = const()[name = tensor<string, []>("op_40396_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_40396_end_0 = const()[name = tensor<string, []>("op_40396_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_40396_end_mask_0 = const()[name = tensor<string, []>("op_40396_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40396_cast = slice_by_index(begin = var_40396_begin_0, end = var_40396_end_0, end_mask = var_40396_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40396_cast")];
            tensor<int32, [4]> var_40400_begin_0 = const()[name = tensor<string, []>("op_40400_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_40400_end_0 = const()[name = tensor<string, []>("op_40400_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_40400_end_mask_0 = const()[name = tensor<string, []>("op_40400_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40400_cast = slice_by_index(begin = var_40400_begin_0, end = var_40400_end_0, end_mask = var_40400_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40400_cast")];
            tensor<int32, [4]> var_40404_begin_0 = const()[name = tensor<string, []>("op_40404_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_40404_end_0 = const()[name = tensor<string, []>("op_40404_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_40404_end_mask_0 = const()[name = tensor<string, []>("op_40404_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40404_cast = slice_by_index(begin = var_40404_begin_0, end = var_40404_end_0, end_mask = var_40404_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40404_cast")];
            tensor<int32, [4]> var_40408_begin_0 = const()[name = tensor<string, []>("op_40408_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_40408_end_0 = const()[name = tensor<string, []>("op_40408_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_40408_end_mask_0 = const()[name = tensor<string, []>("op_40408_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40408_cast = slice_by_index(begin = var_40408_begin_0, end = var_40408_end_0, end_mask = var_40408_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40408_cast")];
            tensor<int32, [4]> var_40412_begin_0 = const()[name = tensor<string, []>("op_40412_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_40412_end_0 = const()[name = tensor<string, []>("op_40412_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_40412_end_mask_0 = const()[name = tensor<string, []>("op_40412_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40412_cast = slice_by_index(begin = var_40412_begin_0, end = var_40412_end_0, end_mask = var_40412_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40412_cast")];
            tensor<int32, [4]> var_40416_begin_0 = const()[name = tensor<string, []>("op_40416_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_40416_end_0 = const()[name = tensor<string, []>("op_40416_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_40416_end_mask_0 = const()[name = tensor<string, []>("op_40416_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_40416_cast = slice_by_index(begin = var_40416_begin_0, end = var_40416_end_0, end_mask = var_40416_end_mask_0, x = v_179_cast)[name = tensor<string, []>("op_40416_cast")];
            tensor<string, []> var_40420_equation_0 = const()[name = tensor<string, []>("op_40420_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40420_cast = einsum(equation = var_40420_equation_0, values = (var_40262_cast, var_40179_cast))[name = tensor<string, []>("op_40420_cast")];
            tensor<fp16, []> var_40421_to_fp16 = const()[name = tensor<string, []>("op_40421_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3401_cast = mul(x = var_40420_cast, y = var_40421_to_fp16)[name = tensor<string, []>("aw_3401_cast")];
            tensor<string, []> var_40424_equation_0 = const()[name = tensor<string, []>("op_40424_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40424_cast = einsum(equation = var_40424_equation_0, values = (var_40266_cast, var_40183_cast))[name = tensor<string, []>("op_40424_cast")];
            tensor<fp16, []> var_40425_to_fp16 = const()[name = tensor<string, []>("op_40425_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3403_cast = mul(x = var_40424_cast, y = var_40425_to_fp16)[name = tensor<string, []>("aw_3403_cast")];
            tensor<string, []> var_40428_equation_0 = const()[name = tensor<string, []>("op_40428_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40428_cast = einsum(equation = var_40428_equation_0, values = (var_40270_cast, var_40187_cast))[name = tensor<string, []>("op_40428_cast")];
            tensor<fp16, []> var_40429_to_fp16 = const()[name = tensor<string, []>("op_40429_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3405_cast = mul(x = var_40428_cast, y = var_40429_to_fp16)[name = tensor<string, []>("aw_3405_cast")];
            tensor<string, []> var_40432_equation_0 = const()[name = tensor<string, []>("op_40432_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40432_cast = einsum(equation = var_40432_equation_0, values = (var_40274_cast, var_40191_cast))[name = tensor<string, []>("op_40432_cast")];
            tensor<fp16, []> var_40433_to_fp16 = const()[name = tensor<string, []>("op_40433_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3407_cast = mul(x = var_40432_cast, y = var_40433_to_fp16)[name = tensor<string, []>("aw_3407_cast")];
            tensor<string, []> var_40436_equation_0 = const()[name = tensor<string, []>("op_40436_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40436_cast = einsum(equation = var_40436_equation_0, values = (var_40278_cast, var_40195_cast))[name = tensor<string, []>("op_40436_cast")];
            tensor<fp16, []> var_40437_to_fp16 = const()[name = tensor<string, []>("op_40437_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3409_cast = mul(x = var_40436_cast, y = var_40437_to_fp16)[name = tensor<string, []>("aw_3409_cast")];
            tensor<string, []> var_40440_equation_0 = const()[name = tensor<string, []>("op_40440_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40440_cast = einsum(equation = var_40440_equation_0, values = (var_40282_cast, var_40199_cast))[name = tensor<string, []>("op_40440_cast")];
            tensor<fp16, []> var_40441_to_fp16 = const()[name = tensor<string, []>("op_40441_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3411_cast = mul(x = var_40440_cast, y = var_40441_to_fp16)[name = tensor<string, []>("aw_3411_cast")];
            tensor<string, []> var_40444_equation_0 = const()[name = tensor<string, []>("op_40444_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40444_cast = einsum(equation = var_40444_equation_0, values = (var_40286_cast, var_40203_cast))[name = tensor<string, []>("op_40444_cast")];
            tensor<fp16, []> var_40445_to_fp16 = const()[name = tensor<string, []>("op_40445_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3413_cast = mul(x = var_40444_cast, y = var_40445_to_fp16)[name = tensor<string, []>("aw_3413_cast")];
            tensor<string, []> var_40448_equation_0 = const()[name = tensor<string, []>("op_40448_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40448_cast = einsum(equation = var_40448_equation_0, values = (var_40290_cast, var_40207_cast))[name = tensor<string, []>("op_40448_cast")];
            tensor<fp16, []> var_40449_to_fp16 = const()[name = tensor<string, []>("op_40449_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3415_cast = mul(x = var_40448_cast, y = var_40449_to_fp16)[name = tensor<string, []>("aw_3415_cast")];
            tensor<string, []> var_40452_equation_0 = const()[name = tensor<string, []>("op_40452_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40452_cast = einsum(equation = var_40452_equation_0, values = (var_40294_cast, var_40211_cast))[name = tensor<string, []>("op_40452_cast")];
            tensor<fp16, []> var_40453_to_fp16 = const()[name = tensor<string, []>("op_40453_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3417_cast = mul(x = var_40452_cast, y = var_40453_to_fp16)[name = tensor<string, []>("aw_3417_cast")];
            tensor<string, []> var_40456_equation_0 = const()[name = tensor<string, []>("op_40456_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40456_cast = einsum(equation = var_40456_equation_0, values = (var_40298_cast, var_40215_cast))[name = tensor<string, []>("op_40456_cast")];
            tensor<fp16, []> var_40457_to_fp16 = const()[name = tensor<string, []>("op_40457_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3419_cast = mul(x = var_40456_cast, y = var_40457_to_fp16)[name = tensor<string, []>("aw_3419_cast")];
            tensor<string, []> var_40460_equation_0 = const()[name = tensor<string, []>("op_40460_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40460_cast = einsum(equation = var_40460_equation_0, values = (var_40302_cast, var_40219_cast))[name = tensor<string, []>("op_40460_cast")];
            tensor<fp16, []> var_40461_to_fp16 = const()[name = tensor<string, []>("op_40461_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3421_cast = mul(x = var_40460_cast, y = var_40461_to_fp16)[name = tensor<string, []>("aw_3421_cast")];
            tensor<string, []> var_40464_equation_0 = const()[name = tensor<string, []>("op_40464_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40464_cast = einsum(equation = var_40464_equation_0, values = (var_40306_cast, var_40223_cast))[name = tensor<string, []>("op_40464_cast")];
            tensor<fp16, []> var_40465_to_fp16 = const()[name = tensor<string, []>("op_40465_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3423_cast = mul(x = var_40464_cast, y = var_40465_to_fp16)[name = tensor<string, []>("aw_3423_cast")];
            tensor<string, []> var_40468_equation_0 = const()[name = tensor<string, []>("op_40468_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40468_cast = einsum(equation = var_40468_equation_0, values = (var_40310_cast, var_40227_cast))[name = tensor<string, []>("op_40468_cast")];
            tensor<fp16, []> var_40469_to_fp16 = const()[name = tensor<string, []>("op_40469_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3425_cast = mul(x = var_40468_cast, y = var_40469_to_fp16)[name = tensor<string, []>("aw_3425_cast")];
            tensor<string, []> var_40472_equation_0 = const()[name = tensor<string, []>("op_40472_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40472_cast = einsum(equation = var_40472_equation_0, values = (var_40314_cast, var_40231_cast))[name = tensor<string, []>("op_40472_cast")];
            tensor<fp16, []> var_40473_to_fp16 = const()[name = tensor<string, []>("op_40473_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3427_cast = mul(x = var_40472_cast, y = var_40473_to_fp16)[name = tensor<string, []>("aw_3427_cast")];
            tensor<string, []> var_40476_equation_0 = const()[name = tensor<string, []>("op_40476_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40476_cast = einsum(equation = var_40476_equation_0, values = (var_40318_cast, var_40235_cast))[name = tensor<string, []>("op_40476_cast")];
            tensor<fp16, []> var_40477_to_fp16 = const()[name = tensor<string, []>("op_40477_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3429_cast = mul(x = var_40476_cast, y = var_40477_to_fp16)[name = tensor<string, []>("aw_3429_cast")];
            tensor<string, []> var_40480_equation_0 = const()[name = tensor<string, []>("op_40480_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40480_cast = einsum(equation = var_40480_equation_0, values = (var_40322_cast, var_40239_cast))[name = tensor<string, []>("op_40480_cast")];
            tensor<fp16, []> var_40481_to_fp16 = const()[name = tensor<string, []>("op_40481_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3431_cast = mul(x = var_40480_cast, y = var_40481_to_fp16)[name = tensor<string, []>("aw_3431_cast")];
            tensor<string, []> var_40484_equation_0 = const()[name = tensor<string, []>("op_40484_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40484_cast = einsum(equation = var_40484_equation_0, values = (var_40326_cast, var_40243_cast))[name = tensor<string, []>("op_40484_cast")];
            tensor<fp16, []> var_40485_to_fp16 = const()[name = tensor<string, []>("op_40485_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3433_cast = mul(x = var_40484_cast, y = var_40485_to_fp16)[name = tensor<string, []>("aw_3433_cast")];
            tensor<string, []> var_40488_equation_0 = const()[name = tensor<string, []>("op_40488_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40488_cast = einsum(equation = var_40488_equation_0, values = (var_40330_cast, var_40247_cast))[name = tensor<string, []>("op_40488_cast")];
            tensor<fp16, []> var_40489_to_fp16 = const()[name = tensor<string, []>("op_40489_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3435_cast = mul(x = var_40488_cast, y = var_40489_to_fp16)[name = tensor<string, []>("aw_3435_cast")];
            tensor<string, []> var_40492_equation_0 = const()[name = tensor<string, []>("op_40492_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40492_cast = einsum(equation = var_40492_equation_0, values = (var_40334_cast, var_40251_cast))[name = tensor<string, []>("op_40492_cast")];
            tensor<fp16, []> var_40493_to_fp16 = const()[name = tensor<string, []>("op_40493_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3437_cast = mul(x = var_40492_cast, y = var_40493_to_fp16)[name = tensor<string, []>("aw_3437_cast")];
            tensor<string, []> var_40496_equation_0 = const()[name = tensor<string, []>("op_40496_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_40496_cast = einsum(equation = var_40496_equation_0, values = (var_40338_cast, var_40255_cast))[name = tensor<string, []>("op_40496_cast")];
            tensor<fp16, []> var_40497_to_fp16 = const()[name = tensor<string, []>("op_40497_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3439_cast = mul(x = var_40496_cast, y = var_40497_to_fp16)[name = tensor<string, []>("aw_3439_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40499_cast = softmax(axis = var_30385, x = aw_3401_cast)[name = tensor<string, []>("op_40499_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40500_cast = softmax(axis = var_30385, x = aw_3403_cast)[name = tensor<string, []>("op_40500_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40501_cast = softmax(axis = var_30385, x = aw_3405_cast)[name = tensor<string, []>("op_40501_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40502_cast = softmax(axis = var_30385, x = aw_3407_cast)[name = tensor<string, []>("op_40502_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40503_cast = softmax(axis = var_30385, x = aw_3409_cast)[name = tensor<string, []>("op_40503_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40504_cast = softmax(axis = var_30385, x = aw_3411_cast)[name = tensor<string, []>("op_40504_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40505_cast = softmax(axis = var_30385, x = aw_3413_cast)[name = tensor<string, []>("op_40505_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40506_cast = softmax(axis = var_30385, x = aw_3415_cast)[name = tensor<string, []>("op_40506_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40507_cast = softmax(axis = var_30385, x = aw_3417_cast)[name = tensor<string, []>("op_40507_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40508_cast = softmax(axis = var_30385, x = aw_3419_cast)[name = tensor<string, []>("op_40508_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40509_cast = softmax(axis = var_30385, x = aw_3421_cast)[name = tensor<string, []>("op_40509_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40510_cast = softmax(axis = var_30385, x = aw_3423_cast)[name = tensor<string, []>("op_40510_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40511_cast = softmax(axis = var_30385, x = aw_3425_cast)[name = tensor<string, []>("op_40511_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40512_cast = softmax(axis = var_30385, x = aw_3427_cast)[name = tensor<string, []>("op_40512_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40513_cast = softmax(axis = var_30385, x = aw_3429_cast)[name = tensor<string, []>("op_40513_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40514_cast = softmax(axis = var_30385, x = aw_3431_cast)[name = tensor<string, []>("op_40514_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40515_cast = softmax(axis = var_30385, x = aw_3433_cast)[name = tensor<string, []>("op_40515_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40516_cast = softmax(axis = var_30385, x = aw_3435_cast)[name = tensor<string, []>("op_40516_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40517_cast = softmax(axis = var_30385, x = aw_3437_cast)[name = tensor<string, []>("op_40517_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_40518_cast = softmax(axis = var_30385, x = aw_3439_cast)[name = tensor<string, []>("op_40518_cast")];
            tensor<string, []> var_40520_equation_0 = const()[name = tensor<string, []>("op_40520_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40520_cast = einsum(equation = var_40520_equation_0, values = (var_40340_cast, var_40499_cast))[name = tensor<string, []>("op_40520_cast")];
            tensor<string, []> var_40522_equation_0 = const()[name = tensor<string, []>("op_40522_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40522_cast = einsum(equation = var_40522_equation_0, values = (var_40344_cast, var_40500_cast))[name = tensor<string, []>("op_40522_cast")];
            tensor<string, []> var_40524_equation_0 = const()[name = tensor<string, []>("op_40524_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40524_cast = einsum(equation = var_40524_equation_0, values = (var_40348_cast, var_40501_cast))[name = tensor<string, []>("op_40524_cast")];
            tensor<string, []> var_40526_equation_0 = const()[name = tensor<string, []>("op_40526_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40526_cast = einsum(equation = var_40526_equation_0, values = (var_40352_cast, var_40502_cast))[name = tensor<string, []>("op_40526_cast")];
            tensor<string, []> var_40528_equation_0 = const()[name = tensor<string, []>("op_40528_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40528_cast = einsum(equation = var_40528_equation_0, values = (var_40356_cast, var_40503_cast))[name = tensor<string, []>("op_40528_cast")];
            tensor<string, []> var_40530_equation_0 = const()[name = tensor<string, []>("op_40530_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40530_cast = einsum(equation = var_40530_equation_0, values = (var_40360_cast, var_40504_cast))[name = tensor<string, []>("op_40530_cast")];
            tensor<string, []> var_40532_equation_0 = const()[name = tensor<string, []>("op_40532_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40532_cast = einsum(equation = var_40532_equation_0, values = (var_40364_cast, var_40505_cast))[name = tensor<string, []>("op_40532_cast")];
            tensor<string, []> var_40534_equation_0 = const()[name = tensor<string, []>("op_40534_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40534_cast = einsum(equation = var_40534_equation_0, values = (var_40368_cast, var_40506_cast))[name = tensor<string, []>("op_40534_cast")];
            tensor<string, []> var_40536_equation_0 = const()[name = tensor<string, []>("op_40536_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40536_cast = einsum(equation = var_40536_equation_0, values = (var_40372_cast, var_40507_cast))[name = tensor<string, []>("op_40536_cast")];
            tensor<string, []> var_40538_equation_0 = const()[name = tensor<string, []>("op_40538_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40538_cast = einsum(equation = var_40538_equation_0, values = (var_40376_cast, var_40508_cast))[name = tensor<string, []>("op_40538_cast")];
            tensor<string, []> var_40540_equation_0 = const()[name = tensor<string, []>("op_40540_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40540_cast = einsum(equation = var_40540_equation_0, values = (var_40380_cast, var_40509_cast))[name = tensor<string, []>("op_40540_cast")];
            tensor<string, []> var_40542_equation_0 = const()[name = tensor<string, []>("op_40542_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40542_cast = einsum(equation = var_40542_equation_0, values = (var_40384_cast, var_40510_cast))[name = tensor<string, []>("op_40542_cast")];
            tensor<string, []> var_40544_equation_0 = const()[name = tensor<string, []>("op_40544_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40544_cast = einsum(equation = var_40544_equation_0, values = (var_40388_cast, var_40511_cast))[name = tensor<string, []>("op_40544_cast")];
            tensor<string, []> var_40546_equation_0 = const()[name = tensor<string, []>("op_40546_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40546_cast = einsum(equation = var_40546_equation_0, values = (var_40392_cast, var_40512_cast))[name = tensor<string, []>("op_40546_cast")];
            tensor<string, []> var_40548_equation_0 = const()[name = tensor<string, []>("op_40548_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40548_cast = einsum(equation = var_40548_equation_0, values = (var_40396_cast, var_40513_cast))[name = tensor<string, []>("op_40548_cast")];
            tensor<string, []> var_40550_equation_0 = const()[name = tensor<string, []>("op_40550_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40550_cast = einsum(equation = var_40550_equation_0, values = (var_40400_cast, var_40514_cast))[name = tensor<string, []>("op_40550_cast")];
            tensor<string, []> var_40552_equation_0 = const()[name = tensor<string, []>("op_40552_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40552_cast = einsum(equation = var_40552_equation_0, values = (var_40404_cast, var_40515_cast))[name = tensor<string, []>("op_40552_cast")];
            tensor<string, []> var_40554_equation_0 = const()[name = tensor<string, []>("op_40554_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40554_cast = einsum(equation = var_40554_equation_0, values = (var_40408_cast, var_40516_cast))[name = tensor<string, []>("op_40554_cast")];
            tensor<string, []> var_40556_equation_0 = const()[name = tensor<string, []>("op_40556_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40556_cast = einsum(equation = var_40556_equation_0, values = (var_40412_cast, var_40517_cast))[name = tensor<string, []>("op_40556_cast")];
            tensor<string, []> var_40558_equation_0 = const()[name = tensor<string, []>("op_40558_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40558_cast = einsum(equation = var_40558_equation_0, values = (var_40416_cast, var_40518_cast))[name = tensor<string, []>("op_40558_cast")];
            tensor<bool, []> input_541_interleave_0 = const()[name = tensor<string, []>("input_541_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_541_cast = concat(axis = var_30385, interleave = input_541_interleave_0, values = (var_40520_cast, var_40522_cast, var_40524_cast, var_40526_cast, var_40528_cast, var_40530_cast, var_40532_cast, var_40534_cast, var_40536_cast, var_40538_cast, var_40540_cast, var_40542_cast, var_40544_cast, var_40546_cast, var_40548_cast, var_40550_cast, var_40552_cast, var_40554_cast, var_40556_cast, var_40558_cast))[name = tensor<string, []>("input_541_cast")];
            tensor<int32, [2]> var_40564 = const()[name = tensor<string, []>("op_40564"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_40566 = const()[name = tensor<string, []>("op_40566"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_40568_pad_type_0 = const()[name = tensor<string, []>("op_40568_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_40568_pad_0 = const()[name = tensor<string, []>("op_40568_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3415328896)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3418605760)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_40568_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_40566, groups = var_30385, pad = var_40568_pad_0, pad_type = var_40568_pad_type_0, strides = var_40564, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_541_cast)[name = tensor<string, []>("op_40568_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_269_cast = add(x = var_40568_cast, y = inputs_267_cast)[name = tensor<string, []>("inputs_269_cast")];
            tensor<int32, [1]> var_40572 = const()[name = tensor<string, []>("op_40572"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_269_cast = reduce_mean(axes = var_40572, keep_dims = var_30380, x = inputs_269_cast)[name = tensor<string, []>("channels_mean_269_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_269_cast = sub(x = inputs_269_cast, y = channels_mean_269_cast)[name = tensor<string, []>("zero_mean_269_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_269_cast = mul(x = zero_mean_269_cast, y = zero_mean_269_cast)[name = tensor<string, []>("zero_mean_sq_269_cast")];
            tensor<int32, [1]> var_40576 = const()[name = tensor<string, []>("op_40576"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_40577_cast = reduce_mean(axes = var_40576, keep_dims = var_30380, x = zero_mean_sq_269_cast)[name = tensor<string, []>("op_40577_cast")];
            tensor<fp16, []> var_40578_to_fp16 = const()[name = tensor<string, []>("op_40578_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_40579_cast = add(x = var_40577_cast, y = var_40578_to_fp16)[name = tensor<string, []>("op_40579_cast")];
            tensor<fp16, []> denom_269_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_269_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_269_cast = rsqrt(epsilon = denom_269_epsilon_0_to_fp16, x = var_40579_cast)[name = tensor<string, []>("denom_269_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_269_cast = mul(x = zero_mean_269_cast, y = denom_269_cast)[name = tensor<string, []>("out_269_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_40583_to_fp16 = const()[name = tensor<string, []>("op_40583_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3418608384)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_40584_cast = add(x = out_269_cast, y = var_40583_to_fp16)[name = tensor<string, []>("op_40584_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_40586_to_fp16 = const()[name = tensor<string, []>("op_40586_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3418611008)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_543_cast = mul(x = var_40584_cast, y = var_40586_to_fp16)[name = tensor<string, []>("input_543_cast")];
            tensor<int32, [2]> var_40594 = const()[name = tensor<string, []>("op_40594"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_40596 = const()[name = tensor<string, []>("op_40596"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_40598_pad_type_0 = const()[name = tensor<string, []>("op_40598_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_40598_pad_0 = const()[name = tensor<string, []>("op_40598_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3418613632)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3444828096)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_40598_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_40596, groups = var_30385, pad = var_40598_pad_0, pad_type = var_40598_pad_type_0, strides = var_40594, weight = up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_543_cast)[name = tensor<string, []>("op_40598_cast")];
            tensor<int32, [2]> var_40599_split_sizes_0 = const()[name = tensor<string, []>("op_40599_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_40599_axis_0 = const()[name = tensor<string, []>("op_40599_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_40599_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_40599_cast_1 = split(axis = var_40599_axis_0, split_sizes = var_40599_split_sizes_0, x = var_40598_cast)[name = tensor<string, []>("op_40599_cast")];
            tensor<string, []> var_40601_mode_0 = const()[name = tensor<string, []>("op_40601_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_40601_cast = gelu(mode = var_40601_mode_0, x = var_40599_cast_1)[name = tensor<string, []>("op_40601_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_545_cast = mul(x = var_40599_cast_0, y = var_40601_cast)[name = tensor<string, []>("input_545_cast")];
            tensor<int32, [2]> var_40605 = const()[name = tensor<string, []>("op_40605"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_40607 = const()[name = tensor<string, []>("op_40607"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_40609_pad_type_0 = const()[name = tensor<string, []>("op_40609_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_40609_pad_0 = const()[name = tensor<string, []>("op_40609_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3444848640)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3457955904)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_40609_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_40607, groups = var_30385, pad = var_40609_pad_0, pad_type = var_40609_pad_type_0, strides = var_40605, weight = up_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_545_cast)[name = tensor<string, []>("op_40609_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_271_cast = add(x = var_40609_cast, y = inputs_269_cast)[name = tensor<string, []>("inputs_271_cast")];
            tensor<int32, [1]> var_40619 = const()[name = tensor<string, []>("op_40619"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_271_cast = reduce_mean(axes = var_40619, keep_dims = var_30380, x = inputs_271_cast)[name = tensor<string, []>("channels_mean_271_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_271_cast = sub(x = inputs_271_cast, y = channels_mean_271_cast)[name = tensor<string, []>("zero_mean_271_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_271_cast = mul(x = zero_mean_271_cast, y = zero_mean_271_cast)[name = tensor<string, []>("zero_mean_sq_271_cast")];
            tensor<int32, [1]> var_40623 = const()[name = tensor<string, []>("op_40623"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_40624_cast = reduce_mean(axes = var_40623, keep_dims = var_30380, x = zero_mean_sq_271_cast)[name = tensor<string, []>("op_40624_cast")];
            tensor<fp16, []> var_40625_to_fp16 = const()[name = tensor<string, []>("op_40625_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_40626_cast = add(x = var_40624_cast, y = var_40625_to_fp16)[name = tensor<string, []>("op_40626_cast")];
            tensor<fp16, []> denom_271_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_271_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_271_cast = rsqrt(epsilon = denom_271_epsilon_0_to_fp16, x = var_40626_cast)[name = tensor<string, []>("denom_271_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_271_cast = mul(x = zero_mean_271_cast, y = denom_271_cast)[name = tensor<string, []>("out_271_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_40630_to_fp16 = const()[name = tensor<string, []>("op_40630_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3457958528)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_40631_cast = add(x = out_271_cast, y = var_40630_to_fp16)[name = tensor<string, []>("op_40631_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_40633_to_fp16 = const()[name = tensor<string, []>("op_40633_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3457961152)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_367_cast = mul(x = var_40631_cast, y = var_40633_to_fp16)[name = tensor<string, []>("hidden_states_367_cast")];
            tensor<int32, [2]> var_40640 = const()[name = tensor<string, []>("op_40640"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_40642 = const()[name = tensor<string, []>("op_40642"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_181_pad_type_0 = const()[name = tensor<string, []>("q_181_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_181_pad_0 = const()[name = tensor<string, []>("q_181_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3457963776)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_181_cast = conv(dilations = var_40642, groups = var_30385, pad = q_181_pad_0, pad_type = q_181_pad_type_0, strides = var_40640, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_367_cast)[name = tensor<string, []>("q_181_cast")];
            tensor<int32, [2]> var_40646 = const()[name = tensor<string, []>("op_40646"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_40648 = const()[name = tensor<string, []>("op_40648"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_361_pad_type_0 = const()[name = tensor<string, []>("k_361_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_361_pad_0 = const()[name = tensor<string, []>("k_361_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3461240640)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_361_cast = conv(dilations = var_40648, groups = var_30385, pad = k_361_pad_0, pad_type = k_361_pad_type_0, strides = var_40646, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_367_cast)[name = tensor<string, []>("k_361_cast")];
            tensor<int32, [2]> var_40652 = const()[name = tensor<string, []>("op_40652"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_40654 = const()[name = tensor<string, []>("op_40654"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_181_pad_type_0 = const()[name = tensor<string, []>("v_181_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_181_pad_0 = const()[name = tensor<string, []>("v_181_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3464517504)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_181_cast = conv(dilations = var_40654, groups = var_30385, pad = v_181_pad_0, pad_type = v_181_pad_type_0, strides = var_40652, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_367_cast)[name = tensor<string, []>("v_181_cast")];
            tensor<int32, [4]> var_40658_begin_0 = const()[name = tensor<string, []>("op_40658_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_40658_end_0 = const()[name = tensor<string, []>("op_40658_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_40658_end_mask_0 = const()[name = tensor<string, []>("op_40658_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40658_cast = slice_by_index(begin = var_40658_begin_0, end = var_40658_end_0, end_mask = var_40658_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40658_cast")];
            tensor<int32, [4]> var_40662_begin_0 = const()[name = tensor<string, []>("op_40662_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_40662_end_0 = const()[name = tensor<string, []>("op_40662_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_40662_end_mask_0 = const()[name = tensor<string, []>("op_40662_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40662_cast = slice_by_index(begin = var_40662_begin_0, end = var_40662_end_0, end_mask = var_40662_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40662_cast")];
            tensor<int32, [4]> var_40666_begin_0 = const()[name = tensor<string, []>("op_40666_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_40666_end_0 = const()[name = tensor<string, []>("op_40666_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_40666_end_mask_0 = const()[name = tensor<string, []>("op_40666_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40666_cast = slice_by_index(begin = var_40666_begin_0, end = var_40666_end_0, end_mask = var_40666_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40666_cast")];
            tensor<int32, [4]> var_40670_begin_0 = const()[name = tensor<string, []>("op_40670_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_40670_end_0 = const()[name = tensor<string, []>("op_40670_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_40670_end_mask_0 = const()[name = tensor<string, []>("op_40670_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40670_cast = slice_by_index(begin = var_40670_begin_0, end = var_40670_end_0, end_mask = var_40670_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40670_cast")];
            tensor<int32, [4]> var_40674_begin_0 = const()[name = tensor<string, []>("op_40674_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_40674_end_0 = const()[name = tensor<string, []>("op_40674_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_40674_end_mask_0 = const()[name = tensor<string, []>("op_40674_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40674_cast = slice_by_index(begin = var_40674_begin_0, end = var_40674_end_0, end_mask = var_40674_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40674_cast")];
            tensor<int32, [4]> var_40678_begin_0 = const()[name = tensor<string, []>("op_40678_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_40678_end_0 = const()[name = tensor<string, []>("op_40678_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_40678_end_mask_0 = const()[name = tensor<string, []>("op_40678_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40678_cast = slice_by_index(begin = var_40678_begin_0, end = var_40678_end_0, end_mask = var_40678_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40678_cast")];
            tensor<int32, [4]> var_40682_begin_0 = const()[name = tensor<string, []>("op_40682_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_40682_end_0 = const()[name = tensor<string, []>("op_40682_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_40682_end_mask_0 = const()[name = tensor<string, []>("op_40682_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40682_cast = slice_by_index(begin = var_40682_begin_0, end = var_40682_end_0, end_mask = var_40682_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40682_cast")];
            tensor<int32, [4]> var_40686_begin_0 = const()[name = tensor<string, []>("op_40686_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_40686_end_0 = const()[name = tensor<string, []>("op_40686_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_40686_end_mask_0 = const()[name = tensor<string, []>("op_40686_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40686_cast = slice_by_index(begin = var_40686_begin_0, end = var_40686_end_0, end_mask = var_40686_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40686_cast")];
            tensor<int32, [4]> var_40690_begin_0 = const()[name = tensor<string, []>("op_40690_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_40690_end_0 = const()[name = tensor<string, []>("op_40690_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_40690_end_mask_0 = const()[name = tensor<string, []>("op_40690_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40690_cast = slice_by_index(begin = var_40690_begin_0, end = var_40690_end_0, end_mask = var_40690_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40690_cast")];
            tensor<int32, [4]> var_40694_begin_0 = const()[name = tensor<string, []>("op_40694_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_40694_end_0 = const()[name = tensor<string, []>("op_40694_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_40694_end_mask_0 = const()[name = tensor<string, []>("op_40694_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40694_cast = slice_by_index(begin = var_40694_begin_0, end = var_40694_end_0, end_mask = var_40694_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40694_cast")];
            tensor<int32, [4]> var_40698_begin_0 = const()[name = tensor<string, []>("op_40698_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_40698_end_0 = const()[name = tensor<string, []>("op_40698_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_40698_end_mask_0 = const()[name = tensor<string, []>("op_40698_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40698_cast = slice_by_index(begin = var_40698_begin_0, end = var_40698_end_0, end_mask = var_40698_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40698_cast")];
            tensor<int32, [4]> var_40702_begin_0 = const()[name = tensor<string, []>("op_40702_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_40702_end_0 = const()[name = tensor<string, []>("op_40702_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_40702_end_mask_0 = const()[name = tensor<string, []>("op_40702_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40702_cast = slice_by_index(begin = var_40702_begin_0, end = var_40702_end_0, end_mask = var_40702_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40702_cast")];
            tensor<int32, [4]> var_40706_begin_0 = const()[name = tensor<string, []>("op_40706_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_40706_end_0 = const()[name = tensor<string, []>("op_40706_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_40706_end_mask_0 = const()[name = tensor<string, []>("op_40706_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40706_cast = slice_by_index(begin = var_40706_begin_0, end = var_40706_end_0, end_mask = var_40706_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40706_cast")];
            tensor<int32, [4]> var_40710_begin_0 = const()[name = tensor<string, []>("op_40710_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_40710_end_0 = const()[name = tensor<string, []>("op_40710_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_40710_end_mask_0 = const()[name = tensor<string, []>("op_40710_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40710_cast = slice_by_index(begin = var_40710_begin_0, end = var_40710_end_0, end_mask = var_40710_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40710_cast")];
            tensor<int32, [4]> var_40714_begin_0 = const()[name = tensor<string, []>("op_40714_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_40714_end_0 = const()[name = tensor<string, []>("op_40714_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_40714_end_mask_0 = const()[name = tensor<string, []>("op_40714_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40714_cast = slice_by_index(begin = var_40714_begin_0, end = var_40714_end_0, end_mask = var_40714_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40714_cast")];
            tensor<int32, [4]> var_40718_begin_0 = const()[name = tensor<string, []>("op_40718_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_40718_end_0 = const()[name = tensor<string, []>("op_40718_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_40718_end_mask_0 = const()[name = tensor<string, []>("op_40718_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40718_cast = slice_by_index(begin = var_40718_begin_0, end = var_40718_end_0, end_mask = var_40718_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40718_cast")];
            tensor<int32, [4]> var_40722_begin_0 = const()[name = tensor<string, []>("op_40722_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_40722_end_0 = const()[name = tensor<string, []>("op_40722_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_40722_end_mask_0 = const()[name = tensor<string, []>("op_40722_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40722_cast = slice_by_index(begin = var_40722_begin_0, end = var_40722_end_0, end_mask = var_40722_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40722_cast")];
            tensor<int32, [4]> var_40726_begin_0 = const()[name = tensor<string, []>("op_40726_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_40726_end_0 = const()[name = tensor<string, []>("op_40726_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_40726_end_mask_0 = const()[name = tensor<string, []>("op_40726_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40726_cast = slice_by_index(begin = var_40726_begin_0, end = var_40726_end_0, end_mask = var_40726_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40726_cast")];
            tensor<int32, [4]> var_40730_begin_0 = const()[name = tensor<string, []>("op_40730_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_40730_end_0 = const()[name = tensor<string, []>("op_40730_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_40730_end_mask_0 = const()[name = tensor<string, []>("op_40730_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40730_cast = slice_by_index(begin = var_40730_begin_0, end = var_40730_end_0, end_mask = var_40730_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40730_cast")];
            tensor<int32, [4]> var_40734_begin_0 = const()[name = tensor<string, []>("op_40734_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_40734_end_0 = const()[name = tensor<string, []>("op_40734_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_40734_end_mask_0 = const()[name = tensor<string, []>("op_40734_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40734_cast = slice_by_index(begin = var_40734_begin_0, end = var_40734_end_0, end_mask = var_40734_end_mask_0, x = q_181_cast)[name = tensor<string, []>("op_40734_cast")];
            tensor<int32, [4]> k_363_perm_0 = const()[name = tensor<string, []>("k_363_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_40741_begin_0 = const()[name = tensor<string, []>("op_40741_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_40741_end_0 = const()[name = tensor<string, []>("op_40741_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_40741_end_mask_0 = const()[name = tensor<string, []>("op_40741_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_49 = transpose(perm = k_363_perm_0, x = k_361_cast)[name = tensor<string, []>("transpose_49")];
            tensor<fp16, [2, 1024, 1, 64]> var_40741_cast = slice_by_index(begin = var_40741_begin_0, end = var_40741_end_0, end_mask = var_40741_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40741_cast")];
            tensor<int32, [4]> var_40745_begin_0 = const()[name = tensor<string, []>("op_40745_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_40745_end_0 = const()[name = tensor<string, []>("op_40745_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_40745_end_mask_0 = const()[name = tensor<string, []>("op_40745_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40745_cast = slice_by_index(begin = var_40745_begin_0, end = var_40745_end_0, end_mask = var_40745_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40745_cast")];
            tensor<int32, [4]> var_40749_begin_0 = const()[name = tensor<string, []>("op_40749_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_40749_end_0 = const()[name = tensor<string, []>("op_40749_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_40749_end_mask_0 = const()[name = tensor<string, []>("op_40749_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40749_cast = slice_by_index(begin = var_40749_begin_0, end = var_40749_end_0, end_mask = var_40749_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40749_cast")];
            tensor<int32, [4]> var_40753_begin_0 = const()[name = tensor<string, []>("op_40753_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_40753_end_0 = const()[name = tensor<string, []>("op_40753_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_40753_end_mask_0 = const()[name = tensor<string, []>("op_40753_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40753_cast = slice_by_index(begin = var_40753_begin_0, end = var_40753_end_0, end_mask = var_40753_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40753_cast")];
            tensor<int32, [4]> var_40757_begin_0 = const()[name = tensor<string, []>("op_40757_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_40757_end_0 = const()[name = tensor<string, []>("op_40757_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_40757_end_mask_0 = const()[name = tensor<string, []>("op_40757_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40757_cast = slice_by_index(begin = var_40757_begin_0, end = var_40757_end_0, end_mask = var_40757_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40757_cast")];
            tensor<int32, [4]> var_40761_begin_0 = const()[name = tensor<string, []>("op_40761_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_40761_end_0 = const()[name = tensor<string, []>("op_40761_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_40761_end_mask_0 = const()[name = tensor<string, []>("op_40761_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40761_cast = slice_by_index(begin = var_40761_begin_0, end = var_40761_end_0, end_mask = var_40761_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40761_cast")];
            tensor<int32, [4]> var_40765_begin_0 = const()[name = tensor<string, []>("op_40765_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_40765_end_0 = const()[name = tensor<string, []>("op_40765_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_40765_end_mask_0 = const()[name = tensor<string, []>("op_40765_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40765_cast = slice_by_index(begin = var_40765_begin_0, end = var_40765_end_0, end_mask = var_40765_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40765_cast")];
            tensor<int32, [4]> var_40769_begin_0 = const()[name = tensor<string, []>("op_40769_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_40769_end_0 = const()[name = tensor<string, []>("op_40769_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_40769_end_mask_0 = const()[name = tensor<string, []>("op_40769_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40769_cast = slice_by_index(begin = var_40769_begin_0, end = var_40769_end_0, end_mask = var_40769_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40769_cast")];
            tensor<int32, [4]> var_40773_begin_0 = const()[name = tensor<string, []>("op_40773_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_40773_end_0 = const()[name = tensor<string, []>("op_40773_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_40773_end_mask_0 = const()[name = tensor<string, []>("op_40773_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40773_cast = slice_by_index(begin = var_40773_begin_0, end = var_40773_end_0, end_mask = var_40773_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40773_cast")];
            tensor<int32, [4]> var_40777_begin_0 = const()[name = tensor<string, []>("op_40777_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_40777_end_0 = const()[name = tensor<string, []>("op_40777_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_40777_end_mask_0 = const()[name = tensor<string, []>("op_40777_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40777_cast = slice_by_index(begin = var_40777_begin_0, end = var_40777_end_0, end_mask = var_40777_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40777_cast")];
            tensor<int32, [4]> var_40781_begin_0 = const()[name = tensor<string, []>("op_40781_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_40781_end_0 = const()[name = tensor<string, []>("op_40781_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_40781_end_mask_0 = const()[name = tensor<string, []>("op_40781_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40781_cast = slice_by_index(begin = var_40781_begin_0, end = var_40781_end_0, end_mask = var_40781_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40781_cast")];
            tensor<int32, [4]> var_40785_begin_0 = const()[name = tensor<string, []>("op_40785_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_40785_end_0 = const()[name = tensor<string, []>("op_40785_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_40785_end_mask_0 = const()[name = tensor<string, []>("op_40785_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40785_cast = slice_by_index(begin = var_40785_begin_0, end = var_40785_end_0, end_mask = var_40785_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40785_cast")];
            tensor<int32, [4]> var_40789_begin_0 = const()[name = tensor<string, []>("op_40789_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_40789_end_0 = const()[name = tensor<string, []>("op_40789_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_40789_end_mask_0 = const()[name = tensor<string, []>("op_40789_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40789_cast = slice_by_index(begin = var_40789_begin_0, end = var_40789_end_0, end_mask = var_40789_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40789_cast")];
            tensor<int32, [4]> var_40793_begin_0 = const()[name = tensor<string, []>("op_40793_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_40793_end_0 = const()[name = tensor<string, []>("op_40793_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_40793_end_mask_0 = const()[name = tensor<string, []>("op_40793_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40793_cast = slice_by_index(begin = var_40793_begin_0, end = var_40793_end_0, end_mask = var_40793_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40793_cast")];
            tensor<int32, [4]> var_40797_begin_0 = const()[name = tensor<string, []>("op_40797_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_40797_end_0 = const()[name = tensor<string, []>("op_40797_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_40797_end_mask_0 = const()[name = tensor<string, []>("op_40797_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40797_cast = slice_by_index(begin = var_40797_begin_0, end = var_40797_end_0, end_mask = var_40797_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40797_cast")];
            tensor<int32, [4]> var_40801_begin_0 = const()[name = tensor<string, []>("op_40801_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_40801_end_0 = const()[name = tensor<string, []>("op_40801_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_40801_end_mask_0 = const()[name = tensor<string, []>("op_40801_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40801_cast = slice_by_index(begin = var_40801_begin_0, end = var_40801_end_0, end_mask = var_40801_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40801_cast")];
            tensor<int32, [4]> var_40805_begin_0 = const()[name = tensor<string, []>("op_40805_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_40805_end_0 = const()[name = tensor<string, []>("op_40805_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_40805_end_mask_0 = const()[name = tensor<string, []>("op_40805_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40805_cast = slice_by_index(begin = var_40805_begin_0, end = var_40805_end_0, end_mask = var_40805_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40805_cast")];
            tensor<int32, [4]> var_40809_begin_0 = const()[name = tensor<string, []>("op_40809_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_40809_end_0 = const()[name = tensor<string, []>("op_40809_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_40809_end_mask_0 = const()[name = tensor<string, []>("op_40809_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40809_cast = slice_by_index(begin = var_40809_begin_0, end = var_40809_end_0, end_mask = var_40809_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40809_cast")];
            tensor<int32, [4]> var_40813_begin_0 = const()[name = tensor<string, []>("op_40813_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_40813_end_0 = const()[name = tensor<string, []>("op_40813_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_40813_end_mask_0 = const()[name = tensor<string, []>("op_40813_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40813_cast = slice_by_index(begin = var_40813_begin_0, end = var_40813_end_0, end_mask = var_40813_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40813_cast")];
            tensor<int32, [4]> var_40817_begin_0 = const()[name = tensor<string, []>("op_40817_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_40817_end_0 = const()[name = tensor<string, []>("op_40817_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_40817_end_mask_0 = const()[name = tensor<string, []>("op_40817_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_40817_cast = slice_by_index(begin = var_40817_begin_0, end = var_40817_end_0, end_mask = var_40817_end_mask_0, x = transpose_49)[name = tensor<string, []>("op_40817_cast")];
            tensor<int32, [4]> var_40819_begin_0 = const()[name = tensor<string, []>("op_40819_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_40819_end_0 = const()[name = tensor<string, []>("op_40819_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_40819_end_mask_0 = const()[name = tensor<string, []>("op_40819_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40819_cast = slice_by_index(begin = var_40819_begin_0, end = var_40819_end_0, end_mask = var_40819_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40819_cast")];
            tensor<int32, [4]> var_40823_begin_0 = const()[name = tensor<string, []>("op_40823_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_40823_end_0 = const()[name = tensor<string, []>("op_40823_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_40823_end_mask_0 = const()[name = tensor<string, []>("op_40823_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40823_cast = slice_by_index(begin = var_40823_begin_0, end = var_40823_end_0, end_mask = var_40823_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40823_cast")];
            tensor<int32, [4]> var_40827_begin_0 = const()[name = tensor<string, []>("op_40827_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_40827_end_0 = const()[name = tensor<string, []>("op_40827_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_40827_end_mask_0 = const()[name = tensor<string, []>("op_40827_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40827_cast = slice_by_index(begin = var_40827_begin_0, end = var_40827_end_0, end_mask = var_40827_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40827_cast")];
            tensor<int32, [4]> var_40831_begin_0 = const()[name = tensor<string, []>("op_40831_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_40831_end_0 = const()[name = tensor<string, []>("op_40831_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_40831_end_mask_0 = const()[name = tensor<string, []>("op_40831_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40831_cast = slice_by_index(begin = var_40831_begin_0, end = var_40831_end_0, end_mask = var_40831_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40831_cast")];
            tensor<int32, [4]> var_40835_begin_0 = const()[name = tensor<string, []>("op_40835_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_40835_end_0 = const()[name = tensor<string, []>("op_40835_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_40835_end_mask_0 = const()[name = tensor<string, []>("op_40835_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40835_cast = slice_by_index(begin = var_40835_begin_0, end = var_40835_end_0, end_mask = var_40835_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40835_cast")];
            tensor<int32, [4]> var_40839_begin_0 = const()[name = tensor<string, []>("op_40839_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_40839_end_0 = const()[name = tensor<string, []>("op_40839_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_40839_end_mask_0 = const()[name = tensor<string, []>("op_40839_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40839_cast = slice_by_index(begin = var_40839_begin_0, end = var_40839_end_0, end_mask = var_40839_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40839_cast")];
            tensor<int32, [4]> var_40843_begin_0 = const()[name = tensor<string, []>("op_40843_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_40843_end_0 = const()[name = tensor<string, []>("op_40843_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_40843_end_mask_0 = const()[name = tensor<string, []>("op_40843_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40843_cast = slice_by_index(begin = var_40843_begin_0, end = var_40843_end_0, end_mask = var_40843_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40843_cast")];
            tensor<int32, [4]> var_40847_begin_0 = const()[name = tensor<string, []>("op_40847_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_40847_end_0 = const()[name = tensor<string, []>("op_40847_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_40847_end_mask_0 = const()[name = tensor<string, []>("op_40847_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40847_cast = slice_by_index(begin = var_40847_begin_0, end = var_40847_end_0, end_mask = var_40847_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40847_cast")];
            tensor<int32, [4]> var_40851_begin_0 = const()[name = tensor<string, []>("op_40851_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_40851_end_0 = const()[name = tensor<string, []>("op_40851_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_40851_end_mask_0 = const()[name = tensor<string, []>("op_40851_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40851_cast = slice_by_index(begin = var_40851_begin_0, end = var_40851_end_0, end_mask = var_40851_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40851_cast")];
            tensor<int32, [4]> var_40855_begin_0 = const()[name = tensor<string, []>("op_40855_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_40855_end_0 = const()[name = tensor<string, []>("op_40855_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_40855_end_mask_0 = const()[name = tensor<string, []>("op_40855_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40855_cast = slice_by_index(begin = var_40855_begin_0, end = var_40855_end_0, end_mask = var_40855_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40855_cast")];
            tensor<int32, [4]> var_40859_begin_0 = const()[name = tensor<string, []>("op_40859_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_40859_end_0 = const()[name = tensor<string, []>("op_40859_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_40859_end_mask_0 = const()[name = tensor<string, []>("op_40859_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40859_cast = slice_by_index(begin = var_40859_begin_0, end = var_40859_end_0, end_mask = var_40859_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40859_cast")];
            tensor<int32, [4]> var_40863_begin_0 = const()[name = tensor<string, []>("op_40863_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_40863_end_0 = const()[name = tensor<string, []>("op_40863_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_40863_end_mask_0 = const()[name = tensor<string, []>("op_40863_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40863_cast = slice_by_index(begin = var_40863_begin_0, end = var_40863_end_0, end_mask = var_40863_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40863_cast")];
            tensor<int32, [4]> var_40867_begin_0 = const()[name = tensor<string, []>("op_40867_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_40867_end_0 = const()[name = tensor<string, []>("op_40867_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_40867_end_mask_0 = const()[name = tensor<string, []>("op_40867_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40867_cast = slice_by_index(begin = var_40867_begin_0, end = var_40867_end_0, end_mask = var_40867_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40867_cast")];
            tensor<int32, [4]> var_40871_begin_0 = const()[name = tensor<string, []>("op_40871_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_40871_end_0 = const()[name = tensor<string, []>("op_40871_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_40871_end_mask_0 = const()[name = tensor<string, []>("op_40871_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40871_cast = slice_by_index(begin = var_40871_begin_0, end = var_40871_end_0, end_mask = var_40871_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40871_cast")];
            tensor<int32, [4]> var_40875_begin_0 = const()[name = tensor<string, []>("op_40875_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_40875_end_0 = const()[name = tensor<string, []>("op_40875_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_40875_end_mask_0 = const()[name = tensor<string, []>("op_40875_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40875_cast = slice_by_index(begin = var_40875_begin_0, end = var_40875_end_0, end_mask = var_40875_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40875_cast")];
            tensor<int32, [4]> var_40879_begin_0 = const()[name = tensor<string, []>("op_40879_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_40879_end_0 = const()[name = tensor<string, []>("op_40879_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_40879_end_mask_0 = const()[name = tensor<string, []>("op_40879_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40879_cast = slice_by_index(begin = var_40879_begin_0, end = var_40879_end_0, end_mask = var_40879_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40879_cast")];
            tensor<int32, [4]> var_40883_begin_0 = const()[name = tensor<string, []>("op_40883_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_40883_end_0 = const()[name = tensor<string, []>("op_40883_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_40883_end_mask_0 = const()[name = tensor<string, []>("op_40883_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40883_cast = slice_by_index(begin = var_40883_begin_0, end = var_40883_end_0, end_mask = var_40883_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40883_cast")];
            tensor<int32, [4]> var_40887_begin_0 = const()[name = tensor<string, []>("op_40887_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_40887_end_0 = const()[name = tensor<string, []>("op_40887_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_40887_end_mask_0 = const()[name = tensor<string, []>("op_40887_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40887_cast = slice_by_index(begin = var_40887_begin_0, end = var_40887_end_0, end_mask = var_40887_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40887_cast")];
            tensor<int32, [4]> var_40891_begin_0 = const()[name = tensor<string, []>("op_40891_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_40891_end_0 = const()[name = tensor<string, []>("op_40891_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_40891_end_mask_0 = const()[name = tensor<string, []>("op_40891_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40891_cast = slice_by_index(begin = var_40891_begin_0, end = var_40891_end_0, end_mask = var_40891_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40891_cast")];
            tensor<int32, [4]> var_40895_begin_0 = const()[name = tensor<string, []>("op_40895_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_40895_end_0 = const()[name = tensor<string, []>("op_40895_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_40895_end_mask_0 = const()[name = tensor<string, []>("op_40895_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_40895_cast = slice_by_index(begin = var_40895_begin_0, end = var_40895_end_0, end_mask = var_40895_end_mask_0, x = v_181_cast)[name = tensor<string, []>("op_40895_cast")];
            tensor<string, []> var_40899_equation_0 = const()[name = tensor<string, []>("op_40899_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40899_cast = einsum(equation = var_40899_equation_0, values = (var_40741_cast, var_40658_cast))[name = tensor<string, []>("op_40899_cast")];
            tensor<fp16, []> var_40900_to_fp16 = const()[name = tensor<string, []>("op_40900_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3441_cast = mul(x = var_40899_cast, y = var_40900_to_fp16)[name = tensor<string, []>("aw_3441_cast")];
            tensor<string, []> var_40903_equation_0 = const()[name = tensor<string, []>("op_40903_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40903_cast = einsum(equation = var_40903_equation_0, values = (var_40745_cast, var_40662_cast))[name = tensor<string, []>("op_40903_cast")];
            tensor<fp16, []> var_40904_to_fp16 = const()[name = tensor<string, []>("op_40904_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3443_cast = mul(x = var_40903_cast, y = var_40904_to_fp16)[name = tensor<string, []>("aw_3443_cast")];
            tensor<string, []> var_40907_equation_0 = const()[name = tensor<string, []>("op_40907_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40907_cast = einsum(equation = var_40907_equation_0, values = (var_40749_cast, var_40666_cast))[name = tensor<string, []>("op_40907_cast")];
            tensor<fp16, []> var_40908_to_fp16 = const()[name = tensor<string, []>("op_40908_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3445_cast = mul(x = var_40907_cast, y = var_40908_to_fp16)[name = tensor<string, []>("aw_3445_cast")];
            tensor<string, []> var_40911_equation_0 = const()[name = tensor<string, []>("op_40911_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40911_cast = einsum(equation = var_40911_equation_0, values = (var_40753_cast, var_40670_cast))[name = tensor<string, []>("op_40911_cast")];
            tensor<fp16, []> var_40912_to_fp16 = const()[name = tensor<string, []>("op_40912_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3447_cast = mul(x = var_40911_cast, y = var_40912_to_fp16)[name = tensor<string, []>("aw_3447_cast")];
            tensor<string, []> var_40915_equation_0 = const()[name = tensor<string, []>("op_40915_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40915_cast = einsum(equation = var_40915_equation_0, values = (var_40757_cast, var_40674_cast))[name = tensor<string, []>("op_40915_cast")];
            tensor<fp16, []> var_40916_to_fp16 = const()[name = tensor<string, []>("op_40916_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3449_cast = mul(x = var_40915_cast, y = var_40916_to_fp16)[name = tensor<string, []>("aw_3449_cast")];
            tensor<string, []> var_40919_equation_0 = const()[name = tensor<string, []>("op_40919_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40919_cast = einsum(equation = var_40919_equation_0, values = (var_40761_cast, var_40678_cast))[name = tensor<string, []>("op_40919_cast")];
            tensor<fp16, []> var_40920_to_fp16 = const()[name = tensor<string, []>("op_40920_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3451_cast = mul(x = var_40919_cast, y = var_40920_to_fp16)[name = tensor<string, []>("aw_3451_cast")];
            tensor<string, []> var_40923_equation_0 = const()[name = tensor<string, []>("op_40923_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40923_cast = einsum(equation = var_40923_equation_0, values = (var_40765_cast, var_40682_cast))[name = tensor<string, []>("op_40923_cast")];
            tensor<fp16, []> var_40924_to_fp16 = const()[name = tensor<string, []>("op_40924_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3453_cast = mul(x = var_40923_cast, y = var_40924_to_fp16)[name = tensor<string, []>("aw_3453_cast")];
            tensor<string, []> var_40927_equation_0 = const()[name = tensor<string, []>("op_40927_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40927_cast = einsum(equation = var_40927_equation_0, values = (var_40769_cast, var_40686_cast))[name = tensor<string, []>("op_40927_cast")];
            tensor<fp16, []> var_40928_to_fp16 = const()[name = tensor<string, []>("op_40928_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3455_cast = mul(x = var_40927_cast, y = var_40928_to_fp16)[name = tensor<string, []>("aw_3455_cast")];
            tensor<string, []> var_40931_equation_0 = const()[name = tensor<string, []>("op_40931_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40931_cast = einsum(equation = var_40931_equation_0, values = (var_40773_cast, var_40690_cast))[name = tensor<string, []>("op_40931_cast")];
            tensor<fp16, []> var_40932_to_fp16 = const()[name = tensor<string, []>("op_40932_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3457_cast = mul(x = var_40931_cast, y = var_40932_to_fp16)[name = tensor<string, []>("aw_3457_cast")];
            tensor<string, []> var_40935_equation_0 = const()[name = tensor<string, []>("op_40935_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40935_cast = einsum(equation = var_40935_equation_0, values = (var_40777_cast, var_40694_cast))[name = tensor<string, []>("op_40935_cast")];
            tensor<fp16, []> var_40936_to_fp16 = const()[name = tensor<string, []>("op_40936_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3459_cast = mul(x = var_40935_cast, y = var_40936_to_fp16)[name = tensor<string, []>("aw_3459_cast")];
            tensor<string, []> var_40939_equation_0 = const()[name = tensor<string, []>("op_40939_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40939_cast = einsum(equation = var_40939_equation_0, values = (var_40781_cast, var_40698_cast))[name = tensor<string, []>("op_40939_cast")];
            tensor<fp16, []> var_40940_to_fp16 = const()[name = tensor<string, []>("op_40940_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3461_cast = mul(x = var_40939_cast, y = var_40940_to_fp16)[name = tensor<string, []>("aw_3461_cast")];
            tensor<string, []> var_40943_equation_0 = const()[name = tensor<string, []>("op_40943_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40943_cast = einsum(equation = var_40943_equation_0, values = (var_40785_cast, var_40702_cast))[name = tensor<string, []>("op_40943_cast")];
            tensor<fp16, []> var_40944_to_fp16 = const()[name = tensor<string, []>("op_40944_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3463_cast = mul(x = var_40943_cast, y = var_40944_to_fp16)[name = tensor<string, []>("aw_3463_cast")];
            tensor<string, []> var_40947_equation_0 = const()[name = tensor<string, []>("op_40947_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40947_cast = einsum(equation = var_40947_equation_0, values = (var_40789_cast, var_40706_cast))[name = tensor<string, []>("op_40947_cast")];
            tensor<fp16, []> var_40948_to_fp16 = const()[name = tensor<string, []>("op_40948_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3465_cast = mul(x = var_40947_cast, y = var_40948_to_fp16)[name = tensor<string, []>("aw_3465_cast")];
            tensor<string, []> var_40951_equation_0 = const()[name = tensor<string, []>("op_40951_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40951_cast = einsum(equation = var_40951_equation_0, values = (var_40793_cast, var_40710_cast))[name = tensor<string, []>("op_40951_cast")];
            tensor<fp16, []> var_40952_to_fp16 = const()[name = tensor<string, []>("op_40952_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3467_cast = mul(x = var_40951_cast, y = var_40952_to_fp16)[name = tensor<string, []>("aw_3467_cast")];
            tensor<string, []> var_40955_equation_0 = const()[name = tensor<string, []>("op_40955_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40955_cast = einsum(equation = var_40955_equation_0, values = (var_40797_cast, var_40714_cast))[name = tensor<string, []>("op_40955_cast")];
            tensor<fp16, []> var_40956_to_fp16 = const()[name = tensor<string, []>("op_40956_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3469_cast = mul(x = var_40955_cast, y = var_40956_to_fp16)[name = tensor<string, []>("aw_3469_cast")];
            tensor<string, []> var_40959_equation_0 = const()[name = tensor<string, []>("op_40959_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40959_cast = einsum(equation = var_40959_equation_0, values = (var_40801_cast, var_40718_cast))[name = tensor<string, []>("op_40959_cast")];
            tensor<fp16, []> var_40960_to_fp16 = const()[name = tensor<string, []>("op_40960_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3471_cast = mul(x = var_40959_cast, y = var_40960_to_fp16)[name = tensor<string, []>("aw_3471_cast")];
            tensor<string, []> var_40963_equation_0 = const()[name = tensor<string, []>("op_40963_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40963_cast = einsum(equation = var_40963_equation_0, values = (var_40805_cast, var_40722_cast))[name = tensor<string, []>("op_40963_cast")];
            tensor<fp16, []> var_40964_to_fp16 = const()[name = tensor<string, []>("op_40964_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3473_cast = mul(x = var_40963_cast, y = var_40964_to_fp16)[name = tensor<string, []>("aw_3473_cast")];
            tensor<string, []> var_40967_equation_0 = const()[name = tensor<string, []>("op_40967_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40967_cast = einsum(equation = var_40967_equation_0, values = (var_40809_cast, var_40726_cast))[name = tensor<string, []>("op_40967_cast")];
            tensor<fp16, []> var_40968_to_fp16 = const()[name = tensor<string, []>("op_40968_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3475_cast = mul(x = var_40967_cast, y = var_40968_to_fp16)[name = tensor<string, []>("aw_3475_cast")];
            tensor<string, []> var_40971_equation_0 = const()[name = tensor<string, []>("op_40971_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40971_cast = einsum(equation = var_40971_equation_0, values = (var_40813_cast, var_40730_cast))[name = tensor<string, []>("op_40971_cast")];
            tensor<fp16, []> var_40972_to_fp16 = const()[name = tensor<string, []>("op_40972_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3477_cast = mul(x = var_40971_cast, y = var_40972_to_fp16)[name = tensor<string, []>("aw_3477_cast")];
            tensor<string, []> var_40975_equation_0 = const()[name = tensor<string, []>("op_40975_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40975_cast = einsum(equation = var_40975_equation_0, values = (var_40817_cast, var_40734_cast))[name = tensor<string, []>("op_40975_cast")];
            tensor<fp16, []> var_40976_to_fp16 = const()[name = tensor<string, []>("op_40976_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3479_cast = mul(x = var_40975_cast, y = var_40976_to_fp16)[name = tensor<string, []>("aw_3479_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40978_cast = softmax(axis = var_30385, x = aw_3441_cast)[name = tensor<string, []>("op_40978_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40979_cast = softmax(axis = var_30385, x = aw_3443_cast)[name = tensor<string, []>("op_40979_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40980_cast = softmax(axis = var_30385, x = aw_3445_cast)[name = tensor<string, []>("op_40980_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40981_cast = softmax(axis = var_30385, x = aw_3447_cast)[name = tensor<string, []>("op_40981_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40982_cast = softmax(axis = var_30385, x = aw_3449_cast)[name = tensor<string, []>("op_40982_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40983_cast = softmax(axis = var_30385, x = aw_3451_cast)[name = tensor<string, []>("op_40983_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40984_cast = softmax(axis = var_30385, x = aw_3453_cast)[name = tensor<string, []>("op_40984_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40985_cast = softmax(axis = var_30385, x = aw_3455_cast)[name = tensor<string, []>("op_40985_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40986_cast = softmax(axis = var_30385, x = aw_3457_cast)[name = tensor<string, []>("op_40986_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40987_cast = softmax(axis = var_30385, x = aw_3459_cast)[name = tensor<string, []>("op_40987_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40988_cast = softmax(axis = var_30385, x = aw_3461_cast)[name = tensor<string, []>("op_40988_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40989_cast = softmax(axis = var_30385, x = aw_3463_cast)[name = tensor<string, []>("op_40989_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40990_cast = softmax(axis = var_30385, x = aw_3465_cast)[name = tensor<string, []>("op_40990_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40991_cast = softmax(axis = var_30385, x = aw_3467_cast)[name = tensor<string, []>("op_40991_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40992_cast = softmax(axis = var_30385, x = aw_3469_cast)[name = tensor<string, []>("op_40992_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40993_cast = softmax(axis = var_30385, x = aw_3471_cast)[name = tensor<string, []>("op_40993_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40994_cast = softmax(axis = var_30385, x = aw_3473_cast)[name = tensor<string, []>("op_40994_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40995_cast = softmax(axis = var_30385, x = aw_3475_cast)[name = tensor<string, []>("op_40995_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40996_cast = softmax(axis = var_30385, x = aw_3477_cast)[name = tensor<string, []>("op_40996_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_40997_cast = softmax(axis = var_30385, x = aw_3479_cast)[name = tensor<string, []>("op_40997_cast")];
            tensor<string, []> var_40999_equation_0 = const()[name = tensor<string, []>("op_40999_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_40999_cast = einsum(equation = var_40999_equation_0, values = (var_40819_cast, var_40978_cast))[name = tensor<string, []>("op_40999_cast")];
            tensor<string, []> var_41001_equation_0 = const()[name = tensor<string, []>("op_41001_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41001_cast = einsum(equation = var_41001_equation_0, values = (var_40823_cast, var_40979_cast))[name = tensor<string, []>("op_41001_cast")];
            tensor<string, []> var_41003_equation_0 = const()[name = tensor<string, []>("op_41003_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41003_cast = einsum(equation = var_41003_equation_0, values = (var_40827_cast, var_40980_cast))[name = tensor<string, []>("op_41003_cast")];
            tensor<string, []> var_41005_equation_0 = const()[name = tensor<string, []>("op_41005_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41005_cast = einsum(equation = var_41005_equation_0, values = (var_40831_cast, var_40981_cast))[name = tensor<string, []>("op_41005_cast")];
            tensor<string, []> var_41007_equation_0 = const()[name = tensor<string, []>("op_41007_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41007_cast = einsum(equation = var_41007_equation_0, values = (var_40835_cast, var_40982_cast))[name = tensor<string, []>("op_41007_cast")];
            tensor<string, []> var_41009_equation_0 = const()[name = tensor<string, []>("op_41009_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41009_cast = einsum(equation = var_41009_equation_0, values = (var_40839_cast, var_40983_cast))[name = tensor<string, []>("op_41009_cast")];
            tensor<string, []> var_41011_equation_0 = const()[name = tensor<string, []>("op_41011_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41011_cast = einsum(equation = var_41011_equation_0, values = (var_40843_cast, var_40984_cast))[name = tensor<string, []>("op_41011_cast")];
            tensor<string, []> var_41013_equation_0 = const()[name = tensor<string, []>("op_41013_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41013_cast = einsum(equation = var_41013_equation_0, values = (var_40847_cast, var_40985_cast))[name = tensor<string, []>("op_41013_cast")];
            tensor<string, []> var_41015_equation_0 = const()[name = tensor<string, []>("op_41015_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41015_cast = einsum(equation = var_41015_equation_0, values = (var_40851_cast, var_40986_cast))[name = tensor<string, []>("op_41015_cast")];
            tensor<string, []> var_41017_equation_0 = const()[name = tensor<string, []>("op_41017_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41017_cast = einsum(equation = var_41017_equation_0, values = (var_40855_cast, var_40987_cast))[name = tensor<string, []>("op_41017_cast")];
            tensor<string, []> var_41019_equation_0 = const()[name = tensor<string, []>("op_41019_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41019_cast = einsum(equation = var_41019_equation_0, values = (var_40859_cast, var_40988_cast))[name = tensor<string, []>("op_41019_cast")];
            tensor<string, []> var_41021_equation_0 = const()[name = tensor<string, []>("op_41021_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41021_cast = einsum(equation = var_41021_equation_0, values = (var_40863_cast, var_40989_cast))[name = tensor<string, []>("op_41021_cast")];
            tensor<string, []> var_41023_equation_0 = const()[name = tensor<string, []>("op_41023_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41023_cast = einsum(equation = var_41023_equation_0, values = (var_40867_cast, var_40990_cast))[name = tensor<string, []>("op_41023_cast")];
            tensor<string, []> var_41025_equation_0 = const()[name = tensor<string, []>("op_41025_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41025_cast = einsum(equation = var_41025_equation_0, values = (var_40871_cast, var_40991_cast))[name = tensor<string, []>("op_41025_cast")];
            tensor<string, []> var_41027_equation_0 = const()[name = tensor<string, []>("op_41027_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41027_cast = einsum(equation = var_41027_equation_0, values = (var_40875_cast, var_40992_cast))[name = tensor<string, []>("op_41027_cast")];
            tensor<string, []> var_41029_equation_0 = const()[name = tensor<string, []>("op_41029_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41029_cast = einsum(equation = var_41029_equation_0, values = (var_40879_cast, var_40993_cast))[name = tensor<string, []>("op_41029_cast")];
            tensor<string, []> var_41031_equation_0 = const()[name = tensor<string, []>("op_41031_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41031_cast = einsum(equation = var_41031_equation_0, values = (var_40883_cast, var_40994_cast))[name = tensor<string, []>("op_41031_cast")];
            tensor<string, []> var_41033_equation_0 = const()[name = tensor<string, []>("op_41033_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41033_cast = einsum(equation = var_41033_equation_0, values = (var_40887_cast, var_40995_cast))[name = tensor<string, []>("op_41033_cast")];
            tensor<string, []> var_41035_equation_0 = const()[name = tensor<string, []>("op_41035_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41035_cast = einsum(equation = var_41035_equation_0, values = (var_40891_cast, var_40996_cast))[name = tensor<string, []>("op_41035_cast")];
            tensor<string, []> var_41037_equation_0 = const()[name = tensor<string, []>("op_41037_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41037_cast = einsum(equation = var_41037_equation_0, values = (var_40895_cast, var_40997_cast))[name = tensor<string, []>("op_41037_cast")];
            tensor<bool, []> input_547_interleave_0 = const()[name = tensor<string, []>("input_547_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_547_cast = concat(axis = var_30385, interleave = input_547_interleave_0, values = (var_40999_cast, var_41001_cast, var_41003_cast, var_41005_cast, var_41007_cast, var_41009_cast, var_41011_cast, var_41013_cast, var_41015_cast, var_41017_cast, var_41019_cast, var_41021_cast, var_41023_cast, var_41025_cast, var_41027_cast, var_41029_cast, var_41031_cast, var_41033_cast, var_41035_cast, var_41037_cast))[name = tensor<string, []>("input_547_cast")];
            tensor<int32, [2]> var_41043 = const()[name = tensor<string, []>("op_41043"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_41045 = const()[name = tensor<string, []>("op_41045"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_41047_pad_type_0 = const()[name = tensor<string, []>("op_41047_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_41047_pad_0 = const()[name = tensor<string, []>("op_41047_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3467794368)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3471071232)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_41047_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_41045, groups = var_30385, pad = var_41047_pad_0, pad_type = var_41047_pad_type_0, strides = var_41043, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_547_cast)[name = tensor<string, []>("op_41047_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_273_cast = add(x = var_41047_cast, y = inputs_271_cast)[name = tensor<string, []>("inputs_273_cast")];
            tensor<int32, [1]> var_41051 = const()[name = tensor<string, []>("op_41051"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_273_cast = reduce_mean(axes = var_41051, keep_dims = var_30380, x = inputs_273_cast)[name = tensor<string, []>("channels_mean_273_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_273_cast = sub(x = inputs_273_cast, y = channels_mean_273_cast)[name = tensor<string, []>("zero_mean_273_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_273_cast = mul(x = zero_mean_273_cast, y = zero_mean_273_cast)[name = tensor<string, []>("zero_mean_sq_273_cast")];
            tensor<int32, [1]> var_41055 = const()[name = tensor<string, []>("op_41055"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_41056_cast = reduce_mean(axes = var_41055, keep_dims = var_30380, x = zero_mean_sq_273_cast)[name = tensor<string, []>("op_41056_cast")];
            tensor<fp16, []> var_41057_to_fp16 = const()[name = tensor<string, []>("op_41057_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_41058_cast = add(x = var_41056_cast, y = var_41057_to_fp16)[name = tensor<string, []>("op_41058_cast")];
            tensor<fp16, []> denom_273_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_273_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_273_cast = rsqrt(epsilon = denom_273_epsilon_0_to_fp16, x = var_41058_cast)[name = tensor<string, []>("denom_273_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_273_cast = mul(x = zero_mean_273_cast, y = denom_273_cast)[name = tensor<string, []>("out_273_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_41062_to_fp16 = const()[name = tensor<string, []>("op_41062_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3471073856)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_41063_cast = add(x = out_273_cast, y = var_41062_to_fp16)[name = tensor<string, []>("op_41063_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_41065_to_fp16 = const()[name = tensor<string, []>("op_41065_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3471076480)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_369_cast = mul(x = var_41063_cast, y = var_41065_to_fp16)[name = tensor<string, []>("hidden_states_369_cast")];
            tensor<int32, [2]> var_41072 = const()[name = tensor<string, []>("op_41072"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_41074 = const()[name = tensor<string, []>("op_41074"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_183_pad_type_0 = const()[name = tensor<string, []>("q_183_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_183_pad_0 = const()[name = tensor<string, []>("q_183_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3471079104)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_183_cast = conv(dilations = var_41074, groups = var_30385, pad = q_183_pad_0, pad_type = q_183_pad_type_0, strides = var_41072, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_369_cast)[name = tensor<string, []>("q_183_cast")];
            tensor<int32, [2]> var_41078 = const()[name = tensor<string, []>("op_41078"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_41080 = const()[name = tensor<string, []>("op_41080"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_365_pad_type_0 = const()[name = tensor<string, []>("k_365_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_365_pad_0 = const()[name = tensor<string, []>("k_365_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3474355968)))];
            tensor<fp16, [2, 1280, 1, 77]> k_365_cast = conv(dilations = var_41080, groups = var_30385, pad = k_365_pad_0, pad_type = k_365_pad_type_0, strides = var_41078, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_365_cast")];
            tensor<int32, [2]> var_41084 = const()[name = tensor<string, []>("op_41084"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_41086 = const()[name = tensor<string, []>("op_41086"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_183_pad_type_0 = const()[name = tensor<string, []>("v_183_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_183_pad_0 = const()[name = tensor<string, []>("v_183_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3479598912)))];
            tensor<fp16, [2, 1280, 1, 77]> v_183_cast = conv(dilations = var_41086, groups = var_30385, pad = v_183_pad_0, pad_type = v_183_pad_type_0, strides = var_41084, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_183_cast")];
            tensor<int32, [4]> var_41090_begin_0 = const()[name = tensor<string, []>("op_41090_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_41090_end_0 = const()[name = tensor<string, []>("op_41090_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_41090_end_mask_0 = const()[name = tensor<string, []>("op_41090_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41090_cast = slice_by_index(begin = var_41090_begin_0, end = var_41090_end_0, end_mask = var_41090_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41090_cast")];
            tensor<int32, [4]> var_41094_begin_0 = const()[name = tensor<string, []>("op_41094_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_41094_end_0 = const()[name = tensor<string, []>("op_41094_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_41094_end_mask_0 = const()[name = tensor<string, []>("op_41094_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41094_cast = slice_by_index(begin = var_41094_begin_0, end = var_41094_end_0, end_mask = var_41094_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41094_cast")];
            tensor<int32, [4]> var_41098_begin_0 = const()[name = tensor<string, []>("op_41098_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_41098_end_0 = const()[name = tensor<string, []>("op_41098_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_41098_end_mask_0 = const()[name = tensor<string, []>("op_41098_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41098_cast = slice_by_index(begin = var_41098_begin_0, end = var_41098_end_0, end_mask = var_41098_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41098_cast")];
            tensor<int32, [4]> var_41102_begin_0 = const()[name = tensor<string, []>("op_41102_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_41102_end_0 = const()[name = tensor<string, []>("op_41102_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_41102_end_mask_0 = const()[name = tensor<string, []>("op_41102_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41102_cast = slice_by_index(begin = var_41102_begin_0, end = var_41102_end_0, end_mask = var_41102_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41102_cast")];
            tensor<int32, [4]> var_41106_begin_0 = const()[name = tensor<string, []>("op_41106_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_41106_end_0 = const()[name = tensor<string, []>("op_41106_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_41106_end_mask_0 = const()[name = tensor<string, []>("op_41106_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41106_cast = slice_by_index(begin = var_41106_begin_0, end = var_41106_end_0, end_mask = var_41106_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41106_cast")];
            tensor<int32, [4]> var_41110_begin_0 = const()[name = tensor<string, []>("op_41110_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_41110_end_0 = const()[name = tensor<string, []>("op_41110_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_41110_end_mask_0 = const()[name = tensor<string, []>("op_41110_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41110_cast = slice_by_index(begin = var_41110_begin_0, end = var_41110_end_0, end_mask = var_41110_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41110_cast")];
            tensor<int32, [4]> var_41114_begin_0 = const()[name = tensor<string, []>("op_41114_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_41114_end_0 = const()[name = tensor<string, []>("op_41114_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_41114_end_mask_0 = const()[name = tensor<string, []>("op_41114_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41114_cast = slice_by_index(begin = var_41114_begin_0, end = var_41114_end_0, end_mask = var_41114_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41114_cast")];
            tensor<int32, [4]> var_41118_begin_0 = const()[name = tensor<string, []>("op_41118_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_41118_end_0 = const()[name = tensor<string, []>("op_41118_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_41118_end_mask_0 = const()[name = tensor<string, []>("op_41118_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41118_cast = slice_by_index(begin = var_41118_begin_0, end = var_41118_end_0, end_mask = var_41118_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41118_cast")];
            tensor<int32, [4]> var_41122_begin_0 = const()[name = tensor<string, []>("op_41122_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_41122_end_0 = const()[name = tensor<string, []>("op_41122_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_41122_end_mask_0 = const()[name = tensor<string, []>("op_41122_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41122_cast = slice_by_index(begin = var_41122_begin_0, end = var_41122_end_0, end_mask = var_41122_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41122_cast")];
            tensor<int32, [4]> var_41126_begin_0 = const()[name = tensor<string, []>("op_41126_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_41126_end_0 = const()[name = tensor<string, []>("op_41126_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_41126_end_mask_0 = const()[name = tensor<string, []>("op_41126_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41126_cast = slice_by_index(begin = var_41126_begin_0, end = var_41126_end_0, end_mask = var_41126_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41126_cast")];
            tensor<int32, [4]> var_41130_begin_0 = const()[name = tensor<string, []>("op_41130_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_41130_end_0 = const()[name = tensor<string, []>("op_41130_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_41130_end_mask_0 = const()[name = tensor<string, []>("op_41130_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41130_cast = slice_by_index(begin = var_41130_begin_0, end = var_41130_end_0, end_mask = var_41130_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41130_cast")];
            tensor<int32, [4]> var_41134_begin_0 = const()[name = tensor<string, []>("op_41134_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_41134_end_0 = const()[name = tensor<string, []>("op_41134_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_41134_end_mask_0 = const()[name = tensor<string, []>("op_41134_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41134_cast = slice_by_index(begin = var_41134_begin_0, end = var_41134_end_0, end_mask = var_41134_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41134_cast")];
            tensor<int32, [4]> var_41138_begin_0 = const()[name = tensor<string, []>("op_41138_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_41138_end_0 = const()[name = tensor<string, []>("op_41138_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_41138_end_mask_0 = const()[name = tensor<string, []>("op_41138_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41138_cast = slice_by_index(begin = var_41138_begin_0, end = var_41138_end_0, end_mask = var_41138_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41138_cast")];
            tensor<int32, [4]> var_41142_begin_0 = const()[name = tensor<string, []>("op_41142_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_41142_end_0 = const()[name = tensor<string, []>("op_41142_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_41142_end_mask_0 = const()[name = tensor<string, []>("op_41142_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41142_cast = slice_by_index(begin = var_41142_begin_0, end = var_41142_end_0, end_mask = var_41142_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41142_cast")];
            tensor<int32, [4]> var_41146_begin_0 = const()[name = tensor<string, []>("op_41146_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_41146_end_0 = const()[name = tensor<string, []>("op_41146_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_41146_end_mask_0 = const()[name = tensor<string, []>("op_41146_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41146_cast = slice_by_index(begin = var_41146_begin_0, end = var_41146_end_0, end_mask = var_41146_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41146_cast")];
            tensor<int32, [4]> var_41150_begin_0 = const()[name = tensor<string, []>("op_41150_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_41150_end_0 = const()[name = tensor<string, []>("op_41150_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_41150_end_mask_0 = const()[name = tensor<string, []>("op_41150_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41150_cast = slice_by_index(begin = var_41150_begin_0, end = var_41150_end_0, end_mask = var_41150_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41150_cast")];
            tensor<int32, [4]> var_41154_begin_0 = const()[name = tensor<string, []>("op_41154_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_41154_end_0 = const()[name = tensor<string, []>("op_41154_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_41154_end_mask_0 = const()[name = tensor<string, []>("op_41154_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41154_cast = slice_by_index(begin = var_41154_begin_0, end = var_41154_end_0, end_mask = var_41154_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41154_cast")];
            tensor<int32, [4]> var_41158_begin_0 = const()[name = tensor<string, []>("op_41158_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_41158_end_0 = const()[name = tensor<string, []>("op_41158_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_41158_end_mask_0 = const()[name = tensor<string, []>("op_41158_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41158_cast = slice_by_index(begin = var_41158_begin_0, end = var_41158_end_0, end_mask = var_41158_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41158_cast")];
            tensor<int32, [4]> var_41162_begin_0 = const()[name = tensor<string, []>("op_41162_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_41162_end_0 = const()[name = tensor<string, []>("op_41162_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_41162_end_mask_0 = const()[name = tensor<string, []>("op_41162_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41162_cast = slice_by_index(begin = var_41162_begin_0, end = var_41162_end_0, end_mask = var_41162_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41162_cast")];
            tensor<int32, [4]> var_41166_begin_0 = const()[name = tensor<string, []>("op_41166_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_41166_end_0 = const()[name = tensor<string, []>("op_41166_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_41166_end_mask_0 = const()[name = tensor<string, []>("op_41166_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41166_cast = slice_by_index(begin = var_41166_begin_0, end = var_41166_end_0, end_mask = var_41166_end_mask_0, x = q_183_cast)[name = tensor<string, []>("op_41166_cast")];
            tensor<int32, [4]> k_367_perm_0 = const()[name = tensor<string, []>("k_367_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_41173_begin_0 = const()[name = tensor<string, []>("op_41173_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_41173_end_0 = const()[name = tensor<string, []>("op_41173_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_41173_end_mask_0 = const()[name = tensor<string, []>("op_41173_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_48 = transpose(perm = k_367_perm_0, x = k_365_cast)[name = tensor<string, []>("transpose_48")];
            tensor<fp16, [2, 77, 1, 64]> var_41173_cast = slice_by_index(begin = var_41173_begin_0, end = var_41173_end_0, end_mask = var_41173_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41173_cast")];
            tensor<int32, [4]> var_41177_begin_0 = const()[name = tensor<string, []>("op_41177_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_41177_end_0 = const()[name = tensor<string, []>("op_41177_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_41177_end_mask_0 = const()[name = tensor<string, []>("op_41177_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41177_cast = slice_by_index(begin = var_41177_begin_0, end = var_41177_end_0, end_mask = var_41177_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41177_cast")];
            tensor<int32, [4]> var_41181_begin_0 = const()[name = tensor<string, []>("op_41181_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_41181_end_0 = const()[name = tensor<string, []>("op_41181_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_41181_end_mask_0 = const()[name = tensor<string, []>("op_41181_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41181_cast = slice_by_index(begin = var_41181_begin_0, end = var_41181_end_0, end_mask = var_41181_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41181_cast")];
            tensor<int32, [4]> var_41185_begin_0 = const()[name = tensor<string, []>("op_41185_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_41185_end_0 = const()[name = tensor<string, []>("op_41185_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_41185_end_mask_0 = const()[name = tensor<string, []>("op_41185_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41185_cast = slice_by_index(begin = var_41185_begin_0, end = var_41185_end_0, end_mask = var_41185_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41185_cast")];
            tensor<int32, [4]> var_41189_begin_0 = const()[name = tensor<string, []>("op_41189_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_41189_end_0 = const()[name = tensor<string, []>("op_41189_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_41189_end_mask_0 = const()[name = tensor<string, []>("op_41189_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41189_cast = slice_by_index(begin = var_41189_begin_0, end = var_41189_end_0, end_mask = var_41189_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41189_cast")];
            tensor<int32, [4]> var_41193_begin_0 = const()[name = tensor<string, []>("op_41193_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_41193_end_0 = const()[name = tensor<string, []>("op_41193_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_41193_end_mask_0 = const()[name = tensor<string, []>("op_41193_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41193_cast = slice_by_index(begin = var_41193_begin_0, end = var_41193_end_0, end_mask = var_41193_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41193_cast")];
            tensor<int32, [4]> var_41197_begin_0 = const()[name = tensor<string, []>("op_41197_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_41197_end_0 = const()[name = tensor<string, []>("op_41197_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_41197_end_mask_0 = const()[name = tensor<string, []>("op_41197_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41197_cast = slice_by_index(begin = var_41197_begin_0, end = var_41197_end_0, end_mask = var_41197_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41197_cast")];
            tensor<int32, [4]> var_41201_begin_0 = const()[name = tensor<string, []>("op_41201_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_41201_end_0 = const()[name = tensor<string, []>("op_41201_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_41201_end_mask_0 = const()[name = tensor<string, []>("op_41201_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41201_cast = slice_by_index(begin = var_41201_begin_0, end = var_41201_end_0, end_mask = var_41201_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41201_cast")];
            tensor<int32, [4]> var_41205_begin_0 = const()[name = tensor<string, []>("op_41205_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_41205_end_0 = const()[name = tensor<string, []>("op_41205_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_41205_end_mask_0 = const()[name = tensor<string, []>("op_41205_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41205_cast = slice_by_index(begin = var_41205_begin_0, end = var_41205_end_0, end_mask = var_41205_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41205_cast")];
            tensor<int32, [4]> var_41209_begin_0 = const()[name = tensor<string, []>("op_41209_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_41209_end_0 = const()[name = tensor<string, []>("op_41209_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_41209_end_mask_0 = const()[name = tensor<string, []>("op_41209_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41209_cast = slice_by_index(begin = var_41209_begin_0, end = var_41209_end_0, end_mask = var_41209_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41209_cast")];
            tensor<int32, [4]> var_41213_begin_0 = const()[name = tensor<string, []>("op_41213_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_41213_end_0 = const()[name = tensor<string, []>("op_41213_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_41213_end_mask_0 = const()[name = tensor<string, []>("op_41213_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41213_cast = slice_by_index(begin = var_41213_begin_0, end = var_41213_end_0, end_mask = var_41213_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41213_cast")];
            tensor<int32, [4]> var_41217_begin_0 = const()[name = tensor<string, []>("op_41217_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_41217_end_0 = const()[name = tensor<string, []>("op_41217_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_41217_end_mask_0 = const()[name = tensor<string, []>("op_41217_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41217_cast = slice_by_index(begin = var_41217_begin_0, end = var_41217_end_0, end_mask = var_41217_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41217_cast")];
            tensor<int32, [4]> var_41221_begin_0 = const()[name = tensor<string, []>("op_41221_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_41221_end_0 = const()[name = tensor<string, []>("op_41221_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_41221_end_mask_0 = const()[name = tensor<string, []>("op_41221_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41221_cast = slice_by_index(begin = var_41221_begin_0, end = var_41221_end_0, end_mask = var_41221_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41221_cast")];
            tensor<int32, [4]> var_41225_begin_0 = const()[name = tensor<string, []>("op_41225_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_41225_end_0 = const()[name = tensor<string, []>("op_41225_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_41225_end_mask_0 = const()[name = tensor<string, []>("op_41225_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41225_cast = slice_by_index(begin = var_41225_begin_0, end = var_41225_end_0, end_mask = var_41225_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41225_cast")];
            tensor<int32, [4]> var_41229_begin_0 = const()[name = tensor<string, []>("op_41229_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_41229_end_0 = const()[name = tensor<string, []>("op_41229_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_41229_end_mask_0 = const()[name = tensor<string, []>("op_41229_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41229_cast = slice_by_index(begin = var_41229_begin_0, end = var_41229_end_0, end_mask = var_41229_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41229_cast")];
            tensor<int32, [4]> var_41233_begin_0 = const()[name = tensor<string, []>("op_41233_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_41233_end_0 = const()[name = tensor<string, []>("op_41233_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_41233_end_mask_0 = const()[name = tensor<string, []>("op_41233_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41233_cast = slice_by_index(begin = var_41233_begin_0, end = var_41233_end_0, end_mask = var_41233_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41233_cast")];
            tensor<int32, [4]> var_41237_begin_0 = const()[name = tensor<string, []>("op_41237_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_41237_end_0 = const()[name = tensor<string, []>("op_41237_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_41237_end_mask_0 = const()[name = tensor<string, []>("op_41237_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41237_cast = slice_by_index(begin = var_41237_begin_0, end = var_41237_end_0, end_mask = var_41237_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41237_cast")];
            tensor<int32, [4]> var_41241_begin_0 = const()[name = tensor<string, []>("op_41241_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_41241_end_0 = const()[name = tensor<string, []>("op_41241_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_41241_end_mask_0 = const()[name = tensor<string, []>("op_41241_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41241_cast = slice_by_index(begin = var_41241_begin_0, end = var_41241_end_0, end_mask = var_41241_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41241_cast")];
            tensor<int32, [4]> var_41245_begin_0 = const()[name = tensor<string, []>("op_41245_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_41245_end_0 = const()[name = tensor<string, []>("op_41245_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_41245_end_mask_0 = const()[name = tensor<string, []>("op_41245_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41245_cast = slice_by_index(begin = var_41245_begin_0, end = var_41245_end_0, end_mask = var_41245_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41245_cast")];
            tensor<int32, [4]> var_41249_begin_0 = const()[name = tensor<string, []>("op_41249_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_41249_end_0 = const()[name = tensor<string, []>("op_41249_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_41249_end_mask_0 = const()[name = tensor<string, []>("op_41249_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_41249_cast = slice_by_index(begin = var_41249_begin_0, end = var_41249_end_0, end_mask = var_41249_end_mask_0, x = transpose_48)[name = tensor<string, []>("op_41249_cast")];
            tensor<int32, [4]> var_41251_begin_0 = const()[name = tensor<string, []>("op_41251_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_41251_end_0 = const()[name = tensor<string, []>("op_41251_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_41251_end_mask_0 = const()[name = tensor<string, []>("op_41251_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41251_cast = slice_by_index(begin = var_41251_begin_0, end = var_41251_end_0, end_mask = var_41251_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41251_cast")];
            tensor<int32, [4]> var_41255_begin_0 = const()[name = tensor<string, []>("op_41255_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_41255_end_0 = const()[name = tensor<string, []>("op_41255_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_41255_end_mask_0 = const()[name = tensor<string, []>("op_41255_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41255_cast = slice_by_index(begin = var_41255_begin_0, end = var_41255_end_0, end_mask = var_41255_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41255_cast")];
            tensor<int32, [4]> var_41259_begin_0 = const()[name = tensor<string, []>("op_41259_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_41259_end_0 = const()[name = tensor<string, []>("op_41259_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_41259_end_mask_0 = const()[name = tensor<string, []>("op_41259_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41259_cast = slice_by_index(begin = var_41259_begin_0, end = var_41259_end_0, end_mask = var_41259_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41259_cast")];
            tensor<int32, [4]> var_41263_begin_0 = const()[name = tensor<string, []>("op_41263_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_41263_end_0 = const()[name = tensor<string, []>("op_41263_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_41263_end_mask_0 = const()[name = tensor<string, []>("op_41263_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41263_cast = slice_by_index(begin = var_41263_begin_0, end = var_41263_end_0, end_mask = var_41263_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41263_cast")];
            tensor<int32, [4]> var_41267_begin_0 = const()[name = tensor<string, []>("op_41267_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_41267_end_0 = const()[name = tensor<string, []>("op_41267_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_41267_end_mask_0 = const()[name = tensor<string, []>("op_41267_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41267_cast = slice_by_index(begin = var_41267_begin_0, end = var_41267_end_0, end_mask = var_41267_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41267_cast")];
            tensor<int32, [4]> var_41271_begin_0 = const()[name = tensor<string, []>("op_41271_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_41271_end_0 = const()[name = tensor<string, []>("op_41271_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_41271_end_mask_0 = const()[name = tensor<string, []>("op_41271_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41271_cast = slice_by_index(begin = var_41271_begin_0, end = var_41271_end_0, end_mask = var_41271_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41271_cast")];
            tensor<int32, [4]> var_41275_begin_0 = const()[name = tensor<string, []>("op_41275_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_41275_end_0 = const()[name = tensor<string, []>("op_41275_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_41275_end_mask_0 = const()[name = tensor<string, []>("op_41275_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41275_cast = slice_by_index(begin = var_41275_begin_0, end = var_41275_end_0, end_mask = var_41275_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41275_cast")];
            tensor<int32, [4]> var_41279_begin_0 = const()[name = tensor<string, []>("op_41279_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_41279_end_0 = const()[name = tensor<string, []>("op_41279_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_41279_end_mask_0 = const()[name = tensor<string, []>("op_41279_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41279_cast = slice_by_index(begin = var_41279_begin_0, end = var_41279_end_0, end_mask = var_41279_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41279_cast")];
            tensor<int32, [4]> var_41283_begin_0 = const()[name = tensor<string, []>("op_41283_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_41283_end_0 = const()[name = tensor<string, []>("op_41283_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_41283_end_mask_0 = const()[name = tensor<string, []>("op_41283_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41283_cast = slice_by_index(begin = var_41283_begin_0, end = var_41283_end_0, end_mask = var_41283_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41283_cast")];
            tensor<int32, [4]> var_41287_begin_0 = const()[name = tensor<string, []>("op_41287_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_41287_end_0 = const()[name = tensor<string, []>("op_41287_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_41287_end_mask_0 = const()[name = tensor<string, []>("op_41287_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41287_cast = slice_by_index(begin = var_41287_begin_0, end = var_41287_end_0, end_mask = var_41287_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41287_cast")];
            tensor<int32, [4]> var_41291_begin_0 = const()[name = tensor<string, []>("op_41291_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_41291_end_0 = const()[name = tensor<string, []>("op_41291_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_41291_end_mask_0 = const()[name = tensor<string, []>("op_41291_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41291_cast = slice_by_index(begin = var_41291_begin_0, end = var_41291_end_0, end_mask = var_41291_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41291_cast")];
            tensor<int32, [4]> var_41295_begin_0 = const()[name = tensor<string, []>("op_41295_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_41295_end_0 = const()[name = tensor<string, []>("op_41295_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_41295_end_mask_0 = const()[name = tensor<string, []>("op_41295_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41295_cast = slice_by_index(begin = var_41295_begin_0, end = var_41295_end_0, end_mask = var_41295_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41295_cast")];
            tensor<int32, [4]> var_41299_begin_0 = const()[name = tensor<string, []>("op_41299_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_41299_end_0 = const()[name = tensor<string, []>("op_41299_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_41299_end_mask_0 = const()[name = tensor<string, []>("op_41299_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41299_cast = slice_by_index(begin = var_41299_begin_0, end = var_41299_end_0, end_mask = var_41299_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41299_cast")];
            tensor<int32, [4]> var_41303_begin_0 = const()[name = tensor<string, []>("op_41303_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_41303_end_0 = const()[name = tensor<string, []>("op_41303_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_41303_end_mask_0 = const()[name = tensor<string, []>("op_41303_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41303_cast = slice_by_index(begin = var_41303_begin_0, end = var_41303_end_0, end_mask = var_41303_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41303_cast")];
            tensor<int32, [4]> var_41307_begin_0 = const()[name = tensor<string, []>("op_41307_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_41307_end_0 = const()[name = tensor<string, []>("op_41307_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_41307_end_mask_0 = const()[name = tensor<string, []>("op_41307_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41307_cast = slice_by_index(begin = var_41307_begin_0, end = var_41307_end_0, end_mask = var_41307_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41307_cast")];
            tensor<int32, [4]> var_41311_begin_0 = const()[name = tensor<string, []>("op_41311_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_41311_end_0 = const()[name = tensor<string, []>("op_41311_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_41311_end_mask_0 = const()[name = tensor<string, []>("op_41311_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41311_cast = slice_by_index(begin = var_41311_begin_0, end = var_41311_end_0, end_mask = var_41311_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41311_cast")];
            tensor<int32, [4]> var_41315_begin_0 = const()[name = tensor<string, []>("op_41315_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_41315_end_0 = const()[name = tensor<string, []>("op_41315_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_41315_end_mask_0 = const()[name = tensor<string, []>("op_41315_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41315_cast = slice_by_index(begin = var_41315_begin_0, end = var_41315_end_0, end_mask = var_41315_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41315_cast")];
            tensor<int32, [4]> var_41319_begin_0 = const()[name = tensor<string, []>("op_41319_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_41319_end_0 = const()[name = tensor<string, []>("op_41319_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_41319_end_mask_0 = const()[name = tensor<string, []>("op_41319_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41319_cast = slice_by_index(begin = var_41319_begin_0, end = var_41319_end_0, end_mask = var_41319_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41319_cast")];
            tensor<int32, [4]> var_41323_begin_0 = const()[name = tensor<string, []>("op_41323_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_41323_end_0 = const()[name = tensor<string, []>("op_41323_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_41323_end_mask_0 = const()[name = tensor<string, []>("op_41323_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41323_cast = slice_by_index(begin = var_41323_begin_0, end = var_41323_end_0, end_mask = var_41323_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41323_cast")];
            tensor<int32, [4]> var_41327_begin_0 = const()[name = tensor<string, []>("op_41327_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_41327_end_0 = const()[name = tensor<string, []>("op_41327_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_41327_end_mask_0 = const()[name = tensor<string, []>("op_41327_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_41327_cast = slice_by_index(begin = var_41327_begin_0, end = var_41327_end_0, end_mask = var_41327_end_mask_0, x = v_183_cast)[name = tensor<string, []>("op_41327_cast")];
            tensor<string, []> var_41331_equation_0 = const()[name = tensor<string, []>("op_41331_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41331_cast = einsum(equation = var_41331_equation_0, values = (var_41173_cast, var_41090_cast))[name = tensor<string, []>("op_41331_cast")];
            tensor<fp16, []> var_41332_to_fp16 = const()[name = tensor<string, []>("op_41332_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3481_cast = mul(x = var_41331_cast, y = var_41332_to_fp16)[name = tensor<string, []>("aw_3481_cast")];
            tensor<string, []> var_41335_equation_0 = const()[name = tensor<string, []>("op_41335_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41335_cast = einsum(equation = var_41335_equation_0, values = (var_41177_cast, var_41094_cast))[name = tensor<string, []>("op_41335_cast")];
            tensor<fp16, []> var_41336_to_fp16 = const()[name = tensor<string, []>("op_41336_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3483_cast = mul(x = var_41335_cast, y = var_41336_to_fp16)[name = tensor<string, []>("aw_3483_cast")];
            tensor<string, []> var_41339_equation_0 = const()[name = tensor<string, []>("op_41339_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41339_cast = einsum(equation = var_41339_equation_0, values = (var_41181_cast, var_41098_cast))[name = tensor<string, []>("op_41339_cast")];
            tensor<fp16, []> var_41340_to_fp16 = const()[name = tensor<string, []>("op_41340_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3485_cast = mul(x = var_41339_cast, y = var_41340_to_fp16)[name = tensor<string, []>("aw_3485_cast")];
            tensor<string, []> var_41343_equation_0 = const()[name = tensor<string, []>("op_41343_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41343_cast = einsum(equation = var_41343_equation_0, values = (var_41185_cast, var_41102_cast))[name = tensor<string, []>("op_41343_cast")];
            tensor<fp16, []> var_41344_to_fp16 = const()[name = tensor<string, []>("op_41344_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3487_cast = mul(x = var_41343_cast, y = var_41344_to_fp16)[name = tensor<string, []>("aw_3487_cast")];
            tensor<string, []> var_41347_equation_0 = const()[name = tensor<string, []>("op_41347_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41347_cast = einsum(equation = var_41347_equation_0, values = (var_41189_cast, var_41106_cast))[name = tensor<string, []>("op_41347_cast")];
            tensor<fp16, []> var_41348_to_fp16 = const()[name = tensor<string, []>("op_41348_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3489_cast = mul(x = var_41347_cast, y = var_41348_to_fp16)[name = tensor<string, []>("aw_3489_cast")];
            tensor<string, []> var_41351_equation_0 = const()[name = tensor<string, []>("op_41351_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41351_cast = einsum(equation = var_41351_equation_0, values = (var_41193_cast, var_41110_cast))[name = tensor<string, []>("op_41351_cast")];
            tensor<fp16, []> var_41352_to_fp16 = const()[name = tensor<string, []>("op_41352_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3491_cast = mul(x = var_41351_cast, y = var_41352_to_fp16)[name = tensor<string, []>("aw_3491_cast")];
            tensor<string, []> var_41355_equation_0 = const()[name = tensor<string, []>("op_41355_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41355_cast = einsum(equation = var_41355_equation_0, values = (var_41197_cast, var_41114_cast))[name = tensor<string, []>("op_41355_cast")];
            tensor<fp16, []> var_41356_to_fp16 = const()[name = tensor<string, []>("op_41356_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3493_cast = mul(x = var_41355_cast, y = var_41356_to_fp16)[name = tensor<string, []>("aw_3493_cast")];
            tensor<string, []> var_41359_equation_0 = const()[name = tensor<string, []>("op_41359_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41359_cast = einsum(equation = var_41359_equation_0, values = (var_41201_cast, var_41118_cast))[name = tensor<string, []>("op_41359_cast")];
            tensor<fp16, []> var_41360_to_fp16 = const()[name = tensor<string, []>("op_41360_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3495_cast = mul(x = var_41359_cast, y = var_41360_to_fp16)[name = tensor<string, []>("aw_3495_cast")];
            tensor<string, []> var_41363_equation_0 = const()[name = tensor<string, []>("op_41363_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41363_cast = einsum(equation = var_41363_equation_0, values = (var_41205_cast, var_41122_cast))[name = tensor<string, []>("op_41363_cast")];
            tensor<fp16, []> var_41364_to_fp16 = const()[name = tensor<string, []>("op_41364_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3497_cast = mul(x = var_41363_cast, y = var_41364_to_fp16)[name = tensor<string, []>("aw_3497_cast")];
            tensor<string, []> var_41367_equation_0 = const()[name = tensor<string, []>("op_41367_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41367_cast = einsum(equation = var_41367_equation_0, values = (var_41209_cast, var_41126_cast))[name = tensor<string, []>("op_41367_cast")];
            tensor<fp16, []> var_41368_to_fp16 = const()[name = tensor<string, []>("op_41368_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3499_cast = mul(x = var_41367_cast, y = var_41368_to_fp16)[name = tensor<string, []>("aw_3499_cast")];
            tensor<string, []> var_41371_equation_0 = const()[name = tensor<string, []>("op_41371_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41371_cast = einsum(equation = var_41371_equation_0, values = (var_41213_cast, var_41130_cast))[name = tensor<string, []>("op_41371_cast")];
            tensor<fp16, []> var_41372_to_fp16 = const()[name = tensor<string, []>("op_41372_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3501_cast = mul(x = var_41371_cast, y = var_41372_to_fp16)[name = tensor<string, []>("aw_3501_cast")];
            tensor<string, []> var_41375_equation_0 = const()[name = tensor<string, []>("op_41375_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41375_cast = einsum(equation = var_41375_equation_0, values = (var_41217_cast, var_41134_cast))[name = tensor<string, []>("op_41375_cast")];
            tensor<fp16, []> var_41376_to_fp16 = const()[name = tensor<string, []>("op_41376_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3503_cast = mul(x = var_41375_cast, y = var_41376_to_fp16)[name = tensor<string, []>("aw_3503_cast")];
            tensor<string, []> var_41379_equation_0 = const()[name = tensor<string, []>("op_41379_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41379_cast = einsum(equation = var_41379_equation_0, values = (var_41221_cast, var_41138_cast))[name = tensor<string, []>("op_41379_cast")];
            tensor<fp16, []> var_41380_to_fp16 = const()[name = tensor<string, []>("op_41380_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3505_cast = mul(x = var_41379_cast, y = var_41380_to_fp16)[name = tensor<string, []>("aw_3505_cast")];
            tensor<string, []> var_41383_equation_0 = const()[name = tensor<string, []>("op_41383_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41383_cast = einsum(equation = var_41383_equation_0, values = (var_41225_cast, var_41142_cast))[name = tensor<string, []>("op_41383_cast")];
            tensor<fp16, []> var_41384_to_fp16 = const()[name = tensor<string, []>("op_41384_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3507_cast = mul(x = var_41383_cast, y = var_41384_to_fp16)[name = tensor<string, []>("aw_3507_cast")];
            tensor<string, []> var_41387_equation_0 = const()[name = tensor<string, []>("op_41387_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41387_cast = einsum(equation = var_41387_equation_0, values = (var_41229_cast, var_41146_cast))[name = tensor<string, []>("op_41387_cast")];
            tensor<fp16, []> var_41388_to_fp16 = const()[name = tensor<string, []>("op_41388_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3509_cast = mul(x = var_41387_cast, y = var_41388_to_fp16)[name = tensor<string, []>("aw_3509_cast")];
            tensor<string, []> var_41391_equation_0 = const()[name = tensor<string, []>("op_41391_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41391_cast = einsum(equation = var_41391_equation_0, values = (var_41233_cast, var_41150_cast))[name = tensor<string, []>("op_41391_cast")];
            tensor<fp16, []> var_41392_to_fp16 = const()[name = tensor<string, []>("op_41392_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3511_cast = mul(x = var_41391_cast, y = var_41392_to_fp16)[name = tensor<string, []>("aw_3511_cast")];
            tensor<string, []> var_41395_equation_0 = const()[name = tensor<string, []>("op_41395_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41395_cast = einsum(equation = var_41395_equation_0, values = (var_41237_cast, var_41154_cast))[name = tensor<string, []>("op_41395_cast")];
            tensor<fp16, []> var_41396_to_fp16 = const()[name = tensor<string, []>("op_41396_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3513_cast = mul(x = var_41395_cast, y = var_41396_to_fp16)[name = tensor<string, []>("aw_3513_cast")];
            tensor<string, []> var_41399_equation_0 = const()[name = tensor<string, []>("op_41399_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41399_cast = einsum(equation = var_41399_equation_0, values = (var_41241_cast, var_41158_cast))[name = tensor<string, []>("op_41399_cast")];
            tensor<fp16, []> var_41400_to_fp16 = const()[name = tensor<string, []>("op_41400_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3515_cast = mul(x = var_41399_cast, y = var_41400_to_fp16)[name = tensor<string, []>("aw_3515_cast")];
            tensor<string, []> var_41403_equation_0 = const()[name = tensor<string, []>("op_41403_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41403_cast = einsum(equation = var_41403_equation_0, values = (var_41245_cast, var_41162_cast))[name = tensor<string, []>("op_41403_cast")];
            tensor<fp16, []> var_41404_to_fp16 = const()[name = tensor<string, []>("op_41404_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3517_cast = mul(x = var_41403_cast, y = var_41404_to_fp16)[name = tensor<string, []>("aw_3517_cast")];
            tensor<string, []> var_41407_equation_0 = const()[name = tensor<string, []>("op_41407_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_41407_cast = einsum(equation = var_41407_equation_0, values = (var_41249_cast, var_41166_cast))[name = tensor<string, []>("op_41407_cast")];
            tensor<fp16, []> var_41408_to_fp16 = const()[name = tensor<string, []>("op_41408_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3519_cast = mul(x = var_41407_cast, y = var_41408_to_fp16)[name = tensor<string, []>("aw_3519_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41410_cast = softmax(axis = var_30385, x = aw_3481_cast)[name = tensor<string, []>("op_41410_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41411_cast = softmax(axis = var_30385, x = aw_3483_cast)[name = tensor<string, []>("op_41411_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41412_cast = softmax(axis = var_30385, x = aw_3485_cast)[name = tensor<string, []>("op_41412_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41413_cast = softmax(axis = var_30385, x = aw_3487_cast)[name = tensor<string, []>("op_41413_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41414_cast = softmax(axis = var_30385, x = aw_3489_cast)[name = tensor<string, []>("op_41414_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41415_cast = softmax(axis = var_30385, x = aw_3491_cast)[name = tensor<string, []>("op_41415_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41416_cast = softmax(axis = var_30385, x = aw_3493_cast)[name = tensor<string, []>("op_41416_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41417_cast = softmax(axis = var_30385, x = aw_3495_cast)[name = tensor<string, []>("op_41417_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41418_cast = softmax(axis = var_30385, x = aw_3497_cast)[name = tensor<string, []>("op_41418_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41419_cast = softmax(axis = var_30385, x = aw_3499_cast)[name = tensor<string, []>("op_41419_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41420_cast = softmax(axis = var_30385, x = aw_3501_cast)[name = tensor<string, []>("op_41420_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41421_cast = softmax(axis = var_30385, x = aw_3503_cast)[name = tensor<string, []>("op_41421_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41422_cast = softmax(axis = var_30385, x = aw_3505_cast)[name = tensor<string, []>("op_41422_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41423_cast = softmax(axis = var_30385, x = aw_3507_cast)[name = tensor<string, []>("op_41423_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41424_cast = softmax(axis = var_30385, x = aw_3509_cast)[name = tensor<string, []>("op_41424_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41425_cast = softmax(axis = var_30385, x = aw_3511_cast)[name = tensor<string, []>("op_41425_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41426_cast = softmax(axis = var_30385, x = aw_3513_cast)[name = tensor<string, []>("op_41426_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41427_cast = softmax(axis = var_30385, x = aw_3515_cast)[name = tensor<string, []>("op_41427_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41428_cast = softmax(axis = var_30385, x = aw_3517_cast)[name = tensor<string, []>("op_41428_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_41429_cast = softmax(axis = var_30385, x = aw_3519_cast)[name = tensor<string, []>("op_41429_cast")];
            tensor<string, []> var_41431_equation_0 = const()[name = tensor<string, []>("op_41431_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41431_cast = einsum(equation = var_41431_equation_0, values = (var_41251_cast, var_41410_cast))[name = tensor<string, []>("op_41431_cast")];
            tensor<string, []> var_41433_equation_0 = const()[name = tensor<string, []>("op_41433_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41433_cast = einsum(equation = var_41433_equation_0, values = (var_41255_cast, var_41411_cast))[name = tensor<string, []>("op_41433_cast")];
            tensor<string, []> var_41435_equation_0 = const()[name = tensor<string, []>("op_41435_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41435_cast = einsum(equation = var_41435_equation_0, values = (var_41259_cast, var_41412_cast))[name = tensor<string, []>("op_41435_cast")];
            tensor<string, []> var_41437_equation_0 = const()[name = tensor<string, []>("op_41437_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41437_cast = einsum(equation = var_41437_equation_0, values = (var_41263_cast, var_41413_cast))[name = tensor<string, []>("op_41437_cast")];
            tensor<string, []> var_41439_equation_0 = const()[name = tensor<string, []>("op_41439_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41439_cast = einsum(equation = var_41439_equation_0, values = (var_41267_cast, var_41414_cast))[name = tensor<string, []>("op_41439_cast")];
            tensor<string, []> var_41441_equation_0 = const()[name = tensor<string, []>("op_41441_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41441_cast = einsum(equation = var_41441_equation_0, values = (var_41271_cast, var_41415_cast))[name = tensor<string, []>("op_41441_cast")];
            tensor<string, []> var_41443_equation_0 = const()[name = tensor<string, []>("op_41443_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41443_cast = einsum(equation = var_41443_equation_0, values = (var_41275_cast, var_41416_cast))[name = tensor<string, []>("op_41443_cast")];
            tensor<string, []> var_41445_equation_0 = const()[name = tensor<string, []>("op_41445_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41445_cast = einsum(equation = var_41445_equation_0, values = (var_41279_cast, var_41417_cast))[name = tensor<string, []>("op_41445_cast")];
            tensor<string, []> var_41447_equation_0 = const()[name = tensor<string, []>("op_41447_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41447_cast = einsum(equation = var_41447_equation_0, values = (var_41283_cast, var_41418_cast))[name = tensor<string, []>("op_41447_cast")];
            tensor<string, []> var_41449_equation_0 = const()[name = tensor<string, []>("op_41449_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41449_cast = einsum(equation = var_41449_equation_0, values = (var_41287_cast, var_41419_cast))[name = tensor<string, []>("op_41449_cast")];
            tensor<string, []> var_41451_equation_0 = const()[name = tensor<string, []>("op_41451_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41451_cast = einsum(equation = var_41451_equation_0, values = (var_41291_cast, var_41420_cast))[name = tensor<string, []>("op_41451_cast")];
            tensor<string, []> var_41453_equation_0 = const()[name = tensor<string, []>("op_41453_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41453_cast = einsum(equation = var_41453_equation_0, values = (var_41295_cast, var_41421_cast))[name = tensor<string, []>("op_41453_cast")];
            tensor<string, []> var_41455_equation_0 = const()[name = tensor<string, []>("op_41455_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41455_cast = einsum(equation = var_41455_equation_0, values = (var_41299_cast, var_41422_cast))[name = tensor<string, []>("op_41455_cast")];
            tensor<string, []> var_41457_equation_0 = const()[name = tensor<string, []>("op_41457_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41457_cast = einsum(equation = var_41457_equation_0, values = (var_41303_cast, var_41423_cast))[name = tensor<string, []>("op_41457_cast")];
            tensor<string, []> var_41459_equation_0 = const()[name = tensor<string, []>("op_41459_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41459_cast = einsum(equation = var_41459_equation_0, values = (var_41307_cast, var_41424_cast))[name = tensor<string, []>("op_41459_cast")];
            tensor<string, []> var_41461_equation_0 = const()[name = tensor<string, []>("op_41461_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41461_cast = einsum(equation = var_41461_equation_0, values = (var_41311_cast, var_41425_cast))[name = tensor<string, []>("op_41461_cast")];
            tensor<string, []> var_41463_equation_0 = const()[name = tensor<string, []>("op_41463_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41463_cast = einsum(equation = var_41463_equation_0, values = (var_41315_cast, var_41426_cast))[name = tensor<string, []>("op_41463_cast")];
            tensor<string, []> var_41465_equation_0 = const()[name = tensor<string, []>("op_41465_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41465_cast = einsum(equation = var_41465_equation_0, values = (var_41319_cast, var_41427_cast))[name = tensor<string, []>("op_41465_cast")];
            tensor<string, []> var_41467_equation_0 = const()[name = tensor<string, []>("op_41467_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41467_cast = einsum(equation = var_41467_equation_0, values = (var_41323_cast, var_41428_cast))[name = tensor<string, []>("op_41467_cast")];
            tensor<string, []> var_41469_equation_0 = const()[name = tensor<string, []>("op_41469_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41469_cast = einsum(equation = var_41469_equation_0, values = (var_41327_cast, var_41429_cast))[name = tensor<string, []>("op_41469_cast")];
            tensor<bool, []> input_549_interleave_0 = const()[name = tensor<string, []>("input_549_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_549_cast = concat(axis = var_30385, interleave = input_549_interleave_0, values = (var_41431_cast, var_41433_cast, var_41435_cast, var_41437_cast, var_41439_cast, var_41441_cast, var_41443_cast, var_41445_cast, var_41447_cast, var_41449_cast, var_41451_cast, var_41453_cast, var_41455_cast, var_41457_cast, var_41459_cast, var_41461_cast, var_41463_cast, var_41465_cast, var_41467_cast, var_41469_cast))[name = tensor<string, []>("input_549_cast")];
            tensor<int32, [2]> var_41475 = const()[name = tensor<string, []>("op_41475"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_41477 = const()[name = tensor<string, []>("op_41477"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_41479_pad_type_0 = const()[name = tensor<string, []>("op_41479_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_41479_pad_0 = const()[name = tensor<string, []>("op_41479_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3484841856)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3488118720)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_41479_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_41477, groups = var_30385, pad = var_41479_pad_0, pad_type = var_41479_pad_type_0, strides = var_41475, weight = up_blocks_0_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_549_cast)[name = tensor<string, []>("op_41479_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_275_cast = add(x = var_41479_cast, y = inputs_273_cast)[name = tensor<string, []>("inputs_275_cast")];
            tensor<int32, [1]> var_41483 = const()[name = tensor<string, []>("op_41483"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_275_cast = reduce_mean(axes = var_41483, keep_dims = var_30380, x = inputs_275_cast)[name = tensor<string, []>("channels_mean_275_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_275_cast = sub(x = inputs_275_cast, y = channels_mean_275_cast)[name = tensor<string, []>("zero_mean_275_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_275_cast = mul(x = zero_mean_275_cast, y = zero_mean_275_cast)[name = tensor<string, []>("zero_mean_sq_275_cast")];
            tensor<int32, [1]> var_41487 = const()[name = tensor<string, []>("op_41487"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_41488_cast = reduce_mean(axes = var_41487, keep_dims = var_30380, x = zero_mean_sq_275_cast)[name = tensor<string, []>("op_41488_cast")];
            tensor<fp16, []> var_41489_to_fp16 = const()[name = tensor<string, []>("op_41489_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_41490_cast = add(x = var_41488_cast, y = var_41489_to_fp16)[name = tensor<string, []>("op_41490_cast")];
            tensor<fp16, []> denom_275_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_275_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_275_cast = rsqrt(epsilon = denom_275_epsilon_0_to_fp16, x = var_41490_cast)[name = tensor<string, []>("denom_275_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_275_cast = mul(x = zero_mean_275_cast, y = denom_275_cast)[name = tensor<string, []>("out_275_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_41494_to_fp16 = const()[name = tensor<string, []>("op_41494_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3488121344)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_41495_cast = add(x = out_275_cast, y = var_41494_to_fp16)[name = tensor<string, []>("op_41495_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_41497_to_fp16 = const()[name = tensor<string, []>("op_41497_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3488123968)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_551_cast = mul(x = var_41495_cast, y = var_41497_to_fp16)[name = tensor<string, []>("input_551_cast")];
            tensor<int32, [2]> var_41505 = const()[name = tensor<string, []>("op_41505"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_41507 = const()[name = tensor<string, []>("op_41507"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_41509_pad_type_0 = const()[name = tensor<string, []>("op_41509_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_41509_pad_0 = const()[name = tensor<string, []>("op_41509_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3488126592)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3514341056)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_41509_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_41507, groups = var_30385, pad = var_41509_pad_0, pad_type = var_41509_pad_type_0, strides = var_41505, weight = up_blocks_0_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_551_cast)[name = tensor<string, []>("op_41509_cast")];
            tensor<int32, [2]> var_41510_split_sizes_0 = const()[name = tensor<string, []>("op_41510_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_41510_axis_0 = const()[name = tensor<string, []>("op_41510_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_41510_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_41510_cast_1 = split(axis = var_41510_axis_0, split_sizes = var_41510_split_sizes_0, x = var_41509_cast)[name = tensor<string, []>("op_41510_cast")];
            tensor<string, []> var_41512_mode_0 = const()[name = tensor<string, []>("op_41512_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_41512_cast = gelu(mode = var_41512_mode_0, x = var_41510_cast_1)[name = tensor<string, []>("op_41512_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_553_cast = mul(x = var_41510_cast_0, y = var_41512_cast)[name = tensor<string, []>("input_553_cast")];
            tensor<int32, [2]> var_41516 = const()[name = tensor<string, []>("op_41516"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_41518 = const()[name = tensor<string, []>("op_41518"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_41520_pad_type_0 = const()[name = tensor<string, []>("op_41520_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_41520_pad_0 = const()[name = tensor<string, []>("op_41520_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3514361600)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3527468864)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_41520_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_41518, groups = var_30385, pad = var_41520_pad_0, pad_type = var_41520_pad_type_0, strides = var_41516, weight = up_blocks_0_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_553_cast)[name = tensor<string, []>("op_41520_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_277_cast = add(x = var_41520_cast, y = inputs_275_cast)[name = tensor<string, []>("inputs_277_cast")];
            tensor<int32, [1]> var_41530 = const()[name = tensor<string, []>("op_41530"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_277_cast = reduce_mean(axes = var_41530, keep_dims = var_30380, x = inputs_277_cast)[name = tensor<string, []>("channels_mean_277_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_277_cast = sub(x = inputs_277_cast, y = channels_mean_277_cast)[name = tensor<string, []>("zero_mean_277_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_277_cast = mul(x = zero_mean_277_cast, y = zero_mean_277_cast)[name = tensor<string, []>("zero_mean_sq_277_cast")];
            tensor<int32, [1]> var_41534 = const()[name = tensor<string, []>("op_41534"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_41535_cast = reduce_mean(axes = var_41534, keep_dims = var_30380, x = zero_mean_sq_277_cast)[name = tensor<string, []>("op_41535_cast")];
            tensor<fp16, []> var_41536_to_fp16 = const()[name = tensor<string, []>("op_41536_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_41537_cast = add(x = var_41535_cast, y = var_41536_to_fp16)[name = tensor<string, []>("op_41537_cast")];
            tensor<fp16, []> denom_277_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_277_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_277_cast = rsqrt(epsilon = denom_277_epsilon_0_to_fp16, x = var_41537_cast)[name = tensor<string, []>("denom_277_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_277_cast = mul(x = zero_mean_277_cast, y = denom_277_cast)[name = tensor<string, []>("out_277_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_41541_to_fp16 = const()[name = tensor<string, []>("op_41541_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3527471488)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_41542_cast = add(x = out_277_cast, y = var_41541_to_fp16)[name = tensor<string, []>("op_41542_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_41544_to_fp16 = const()[name = tensor<string, []>("op_41544_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3527474112)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_373_cast = mul(x = var_41542_cast, y = var_41544_to_fp16)[name = tensor<string, []>("hidden_states_373_cast")];
            tensor<int32, [2]> var_41551 = const()[name = tensor<string, []>("op_41551"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_41553 = const()[name = tensor<string, []>("op_41553"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_185_pad_type_0 = const()[name = tensor<string, []>("q_185_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_185_pad_0 = const()[name = tensor<string, []>("q_185_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3527476736)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_185_cast = conv(dilations = var_41553, groups = var_30385, pad = q_185_pad_0, pad_type = q_185_pad_type_0, strides = var_41551, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_q_weight_to_fp16, x = hidden_states_373_cast)[name = tensor<string, []>("q_185_cast")];
            tensor<int32, [2]> var_41557 = const()[name = tensor<string, []>("op_41557"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_41559 = const()[name = tensor<string, []>("op_41559"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_369_pad_type_0 = const()[name = tensor<string, []>("k_369_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_369_pad_0 = const()[name = tensor<string, []>("k_369_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3530753600)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_369_cast = conv(dilations = var_41559, groups = var_30385, pad = k_369_pad_0, pad_type = k_369_pad_type_0, strides = var_41557, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_k_weight_to_fp16, x = hidden_states_373_cast)[name = tensor<string, []>("k_369_cast")];
            tensor<int32, [2]> var_41563 = const()[name = tensor<string, []>("op_41563"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_41565 = const()[name = tensor<string, []>("op_41565"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_185_pad_type_0 = const()[name = tensor<string, []>("v_185_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_185_pad_0 = const()[name = tensor<string, []>("v_185_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3534030464)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_185_cast = conv(dilations = var_41565, groups = var_30385, pad = v_185_pad_0, pad_type = v_185_pad_type_0, strides = var_41563, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_v_weight_to_fp16, x = hidden_states_373_cast)[name = tensor<string, []>("v_185_cast")];
            tensor<int32, [4]> var_41569_begin_0 = const()[name = tensor<string, []>("op_41569_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_41569_end_0 = const()[name = tensor<string, []>("op_41569_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_41569_end_mask_0 = const()[name = tensor<string, []>("op_41569_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41569_cast = slice_by_index(begin = var_41569_begin_0, end = var_41569_end_0, end_mask = var_41569_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41569_cast")];
            tensor<int32, [4]> var_41573_begin_0 = const()[name = tensor<string, []>("op_41573_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_41573_end_0 = const()[name = tensor<string, []>("op_41573_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_41573_end_mask_0 = const()[name = tensor<string, []>("op_41573_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41573_cast = slice_by_index(begin = var_41573_begin_0, end = var_41573_end_0, end_mask = var_41573_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41573_cast")];
            tensor<int32, [4]> var_41577_begin_0 = const()[name = tensor<string, []>("op_41577_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_41577_end_0 = const()[name = tensor<string, []>("op_41577_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_41577_end_mask_0 = const()[name = tensor<string, []>("op_41577_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41577_cast = slice_by_index(begin = var_41577_begin_0, end = var_41577_end_0, end_mask = var_41577_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41577_cast")];
            tensor<int32, [4]> var_41581_begin_0 = const()[name = tensor<string, []>("op_41581_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_41581_end_0 = const()[name = tensor<string, []>("op_41581_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_41581_end_mask_0 = const()[name = tensor<string, []>("op_41581_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41581_cast = slice_by_index(begin = var_41581_begin_0, end = var_41581_end_0, end_mask = var_41581_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41581_cast")];
            tensor<int32, [4]> var_41585_begin_0 = const()[name = tensor<string, []>("op_41585_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_41585_end_0 = const()[name = tensor<string, []>("op_41585_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_41585_end_mask_0 = const()[name = tensor<string, []>("op_41585_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41585_cast = slice_by_index(begin = var_41585_begin_0, end = var_41585_end_0, end_mask = var_41585_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41585_cast")];
            tensor<int32, [4]> var_41589_begin_0 = const()[name = tensor<string, []>("op_41589_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_41589_end_0 = const()[name = tensor<string, []>("op_41589_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_41589_end_mask_0 = const()[name = tensor<string, []>("op_41589_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41589_cast = slice_by_index(begin = var_41589_begin_0, end = var_41589_end_0, end_mask = var_41589_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41589_cast")];
            tensor<int32, [4]> var_41593_begin_0 = const()[name = tensor<string, []>("op_41593_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_41593_end_0 = const()[name = tensor<string, []>("op_41593_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_41593_end_mask_0 = const()[name = tensor<string, []>("op_41593_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41593_cast = slice_by_index(begin = var_41593_begin_0, end = var_41593_end_0, end_mask = var_41593_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41593_cast")];
            tensor<int32, [4]> var_41597_begin_0 = const()[name = tensor<string, []>("op_41597_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_41597_end_0 = const()[name = tensor<string, []>("op_41597_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_41597_end_mask_0 = const()[name = tensor<string, []>("op_41597_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41597_cast = slice_by_index(begin = var_41597_begin_0, end = var_41597_end_0, end_mask = var_41597_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41597_cast")];
            tensor<int32, [4]> var_41601_begin_0 = const()[name = tensor<string, []>("op_41601_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_41601_end_0 = const()[name = tensor<string, []>("op_41601_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_41601_end_mask_0 = const()[name = tensor<string, []>("op_41601_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41601_cast = slice_by_index(begin = var_41601_begin_0, end = var_41601_end_0, end_mask = var_41601_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41601_cast")];
            tensor<int32, [4]> var_41605_begin_0 = const()[name = tensor<string, []>("op_41605_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_41605_end_0 = const()[name = tensor<string, []>("op_41605_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_41605_end_mask_0 = const()[name = tensor<string, []>("op_41605_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41605_cast = slice_by_index(begin = var_41605_begin_0, end = var_41605_end_0, end_mask = var_41605_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41605_cast")];
            tensor<int32, [4]> var_41609_begin_0 = const()[name = tensor<string, []>("op_41609_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_41609_end_0 = const()[name = tensor<string, []>("op_41609_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_41609_end_mask_0 = const()[name = tensor<string, []>("op_41609_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41609_cast = slice_by_index(begin = var_41609_begin_0, end = var_41609_end_0, end_mask = var_41609_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41609_cast")];
            tensor<int32, [4]> var_41613_begin_0 = const()[name = tensor<string, []>("op_41613_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_41613_end_0 = const()[name = tensor<string, []>("op_41613_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_41613_end_mask_0 = const()[name = tensor<string, []>("op_41613_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41613_cast = slice_by_index(begin = var_41613_begin_0, end = var_41613_end_0, end_mask = var_41613_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41613_cast")];
            tensor<int32, [4]> var_41617_begin_0 = const()[name = tensor<string, []>("op_41617_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_41617_end_0 = const()[name = tensor<string, []>("op_41617_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_41617_end_mask_0 = const()[name = tensor<string, []>("op_41617_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41617_cast = slice_by_index(begin = var_41617_begin_0, end = var_41617_end_0, end_mask = var_41617_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41617_cast")];
            tensor<int32, [4]> var_41621_begin_0 = const()[name = tensor<string, []>("op_41621_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_41621_end_0 = const()[name = tensor<string, []>("op_41621_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_41621_end_mask_0 = const()[name = tensor<string, []>("op_41621_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41621_cast = slice_by_index(begin = var_41621_begin_0, end = var_41621_end_0, end_mask = var_41621_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41621_cast")];
            tensor<int32, [4]> var_41625_begin_0 = const()[name = tensor<string, []>("op_41625_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_41625_end_0 = const()[name = tensor<string, []>("op_41625_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_41625_end_mask_0 = const()[name = tensor<string, []>("op_41625_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41625_cast = slice_by_index(begin = var_41625_begin_0, end = var_41625_end_0, end_mask = var_41625_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41625_cast")];
            tensor<int32, [4]> var_41629_begin_0 = const()[name = tensor<string, []>("op_41629_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_41629_end_0 = const()[name = tensor<string, []>("op_41629_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_41629_end_mask_0 = const()[name = tensor<string, []>("op_41629_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41629_cast = slice_by_index(begin = var_41629_begin_0, end = var_41629_end_0, end_mask = var_41629_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41629_cast")];
            tensor<int32, [4]> var_41633_begin_0 = const()[name = tensor<string, []>("op_41633_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_41633_end_0 = const()[name = tensor<string, []>("op_41633_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_41633_end_mask_0 = const()[name = tensor<string, []>("op_41633_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41633_cast = slice_by_index(begin = var_41633_begin_0, end = var_41633_end_0, end_mask = var_41633_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41633_cast")];
            tensor<int32, [4]> var_41637_begin_0 = const()[name = tensor<string, []>("op_41637_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_41637_end_0 = const()[name = tensor<string, []>("op_41637_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_41637_end_mask_0 = const()[name = tensor<string, []>("op_41637_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41637_cast = slice_by_index(begin = var_41637_begin_0, end = var_41637_end_0, end_mask = var_41637_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41637_cast")];
            tensor<int32, [4]> var_41641_begin_0 = const()[name = tensor<string, []>("op_41641_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_41641_end_0 = const()[name = tensor<string, []>("op_41641_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_41641_end_mask_0 = const()[name = tensor<string, []>("op_41641_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41641_cast = slice_by_index(begin = var_41641_begin_0, end = var_41641_end_0, end_mask = var_41641_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41641_cast")];
            tensor<int32, [4]> var_41645_begin_0 = const()[name = tensor<string, []>("op_41645_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_41645_end_0 = const()[name = tensor<string, []>("op_41645_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_41645_end_mask_0 = const()[name = tensor<string, []>("op_41645_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41645_cast = slice_by_index(begin = var_41645_begin_0, end = var_41645_end_0, end_mask = var_41645_end_mask_0, x = q_185_cast)[name = tensor<string, []>("op_41645_cast")];
            tensor<int32, [4]> k_371_perm_0 = const()[name = tensor<string, []>("k_371_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_41652_begin_0 = const()[name = tensor<string, []>("op_41652_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_41652_end_0 = const()[name = tensor<string, []>("op_41652_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_41652_end_mask_0 = const()[name = tensor<string, []>("op_41652_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_47 = transpose(perm = k_371_perm_0, x = k_369_cast)[name = tensor<string, []>("transpose_47")];
            tensor<fp16, [2, 1024, 1, 64]> var_41652_cast = slice_by_index(begin = var_41652_begin_0, end = var_41652_end_0, end_mask = var_41652_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41652_cast")];
            tensor<int32, [4]> var_41656_begin_0 = const()[name = tensor<string, []>("op_41656_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_41656_end_0 = const()[name = tensor<string, []>("op_41656_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_41656_end_mask_0 = const()[name = tensor<string, []>("op_41656_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41656_cast = slice_by_index(begin = var_41656_begin_0, end = var_41656_end_0, end_mask = var_41656_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41656_cast")];
            tensor<int32, [4]> var_41660_begin_0 = const()[name = tensor<string, []>("op_41660_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_41660_end_0 = const()[name = tensor<string, []>("op_41660_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_41660_end_mask_0 = const()[name = tensor<string, []>("op_41660_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41660_cast = slice_by_index(begin = var_41660_begin_0, end = var_41660_end_0, end_mask = var_41660_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41660_cast")];
            tensor<int32, [4]> var_41664_begin_0 = const()[name = tensor<string, []>("op_41664_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_41664_end_0 = const()[name = tensor<string, []>("op_41664_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_41664_end_mask_0 = const()[name = tensor<string, []>("op_41664_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41664_cast = slice_by_index(begin = var_41664_begin_0, end = var_41664_end_0, end_mask = var_41664_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41664_cast")];
            tensor<int32, [4]> var_41668_begin_0 = const()[name = tensor<string, []>("op_41668_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_41668_end_0 = const()[name = tensor<string, []>("op_41668_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_41668_end_mask_0 = const()[name = tensor<string, []>("op_41668_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41668_cast = slice_by_index(begin = var_41668_begin_0, end = var_41668_end_0, end_mask = var_41668_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41668_cast")];
            tensor<int32, [4]> var_41672_begin_0 = const()[name = tensor<string, []>("op_41672_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_41672_end_0 = const()[name = tensor<string, []>("op_41672_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_41672_end_mask_0 = const()[name = tensor<string, []>("op_41672_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41672_cast = slice_by_index(begin = var_41672_begin_0, end = var_41672_end_0, end_mask = var_41672_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41672_cast")];
            tensor<int32, [4]> var_41676_begin_0 = const()[name = tensor<string, []>("op_41676_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_41676_end_0 = const()[name = tensor<string, []>("op_41676_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_41676_end_mask_0 = const()[name = tensor<string, []>("op_41676_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41676_cast = slice_by_index(begin = var_41676_begin_0, end = var_41676_end_0, end_mask = var_41676_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41676_cast")];
            tensor<int32, [4]> var_41680_begin_0 = const()[name = tensor<string, []>("op_41680_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_41680_end_0 = const()[name = tensor<string, []>("op_41680_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_41680_end_mask_0 = const()[name = tensor<string, []>("op_41680_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41680_cast = slice_by_index(begin = var_41680_begin_0, end = var_41680_end_0, end_mask = var_41680_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41680_cast")];
            tensor<int32, [4]> var_41684_begin_0 = const()[name = tensor<string, []>("op_41684_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_41684_end_0 = const()[name = tensor<string, []>("op_41684_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_41684_end_mask_0 = const()[name = tensor<string, []>("op_41684_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41684_cast = slice_by_index(begin = var_41684_begin_0, end = var_41684_end_0, end_mask = var_41684_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41684_cast")];
            tensor<int32, [4]> var_41688_begin_0 = const()[name = tensor<string, []>("op_41688_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_41688_end_0 = const()[name = tensor<string, []>("op_41688_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_41688_end_mask_0 = const()[name = tensor<string, []>("op_41688_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41688_cast = slice_by_index(begin = var_41688_begin_0, end = var_41688_end_0, end_mask = var_41688_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41688_cast")];
            tensor<int32, [4]> var_41692_begin_0 = const()[name = tensor<string, []>("op_41692_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_41692_end_0 = const()[name = tensor<string, []>("op_41692_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_41692_end_mask_0 = const()[name = tensor<string, []>("op_41692_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41692_cast = slice_by_index(begin = var_41692_begin_0, end = var_41692_end_0, end_mask = var_41692_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41692_cast")];
            tensor<int32, [4]> var_41696_begin_0 = const()[name = tensor<string, []>("op_41696_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_41696_end_0 = const()[name = tensor<string, []>("op_41696_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_41696_end_mask_0 = const()[name = tensor<string, []>("op_41696_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41696_cast = slice_by_index(begin = var_41696_begin_0, end = var_41696_end_0, end_mask = var_41696_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41696_cast")];
            tensor<int32, [4]> var_41700_begin_0 = const()[name = tensor<string, []>("op_41700_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_41700_end_0 = const()[name = tensor<string, []>("op_41700_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_41700_end_mask_0 = const()[name = tensor<string, []>("op_41700_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41700_cast = slice_by_index(begin = var_41700_begin_0, end = var_41700_end_0, end_mask = var_41700_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41700_cast")];
            tensor<int32, [4]> var_41704_begin_0 = const()[name = tensor<string, []>("op_41704_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_41704_end_0 = const()[name = tensor<string, []>("op_41704_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_41704_end_mask_0 = const()[name = tensor<string, []>("op_41704_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41704_cast = slice_by_index(begin = var_41704_begin_0, end = var_41704_end_0, end_mask = var_41704_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41704_cast")];
            tensor<int32, [4]> var_41708_begin_0 = const()[name = tensor<string, []>("op_41708_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_41708_end_0 = const()[name = tensor<string, []>("op_41708_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_41708_end_mask_0 = const()[name = tensor<string, []>("op_41708_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41708_cast = slice_by_index(begin = var_41708_begin_0, end = var_41708_end_0, end_mask = var_41708_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41708_cast")];
            tensor<int32, [4]> var_41712_begin_0 = const()[name = tensor<string, []>("op_41712_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_41712_end_0 = const()[name = tensor<string, []>("op_41712_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_41712_end_mask_0 = const()[name = tensor<string, []>("op_41712_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41712_cast = slice_by_index(begin = var_41712_begin_0, end = var_41712_end_0, end_mask = var_41712_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41712_cast")];
            tensor<int32, [4]> var_41716_begin_0 = const()[name = tensor<string, []>("op_41716_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_41716_end_0 = const()[name = tensor<string, []>("op_41716_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_41716_end_mask_0 = const()[name = tensor<string, []>("op_41716_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41716_cast = slice_by_index(begin = var_41716_begin_0, end = var_41716_end_0, end_mask = var_41716_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41716_cast")];
            tensor<int32, [4]> var_41720_begin_0 = const()[name = tensor<string, []>("op_41720_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_41720_end_0 = const()[name = tensor<string, []>("op_41720_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_41720_end_mask_0 = const()[name = tensor<string, []>("op_41720_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41720_cast = slice_by_index(begin = var_41720_begin_0, end = var_41720_end_0, end_mask = var_41720_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41720_cast")];
            tensor<int32, [4]> var_41724_begin_0 = const()[name = tensor<string, []>("op_41724_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_41724_end_0 = const()[name = tensor<string, []>("op_41724_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_41724_end_mask_0 = const()[name = tensor<string, []>("op_41724_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41724_cast = slice_by_index(begin = var_41724_begin_0, end = var_41724_end_0, end_mask = var_41724_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41724_cast")];
            tensor<int32, [4]> var_41728_begin_0 = const()[name = tensor<string, []>("op_41728_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_41728_end_0 = const()[name = tensor<string, []>("op_41728_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_41728_end_mask_0 = const()[name = tensor<string, []>("op_41728_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_41728_cast = slice_by_index(begin = var_41728_begin_0, end = var_41728_end_0, end_mask = var_41728_end_mask_0, x = transpose_47)[name = tensor<string, []>("op_41728_cast")];
            tensor<int32, [4]> var_41730_begin_0 = const()[name = tensor<string, []>("op_41730_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_41730_end_0 = const()[name = tensor<string, []>("op_41730_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_41730_end_mask_0 = const()[name = tensor<string, []>("op_41730_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41730_cast = slice_by_index(begin = var_41730_begin_0, end = var_41730_end_0, end_mask = var_41730_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41730_cast")];
            tensor<int32, [4]> var_41734_begin_0 = const()[name = tensor<string, []>("op_41734_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_41734_end_0 = const()[name = tensor<string, []>("op_41734_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_41734_end_mask_0 = const()[name = tensor<string, []>("op_41734_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41734_cast = slice_by_index(begin = var_41734_begin_0, end = var_41734_end_0, end_mask = var_41734_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41734_cast")];
            tensor<int32, [4]> var_41738_begin_0 = const()[name = tensor<string, []>("op_41738_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_41738_end_0 = const()[name = tensor<string, []>("op_41738_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_41738_end_mask_0 = const()[name = tensor<string, []>("op_41738_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41738_cast = slice_by_index(begin = var_41738_begin_0, end = var_41738_end_0, end_mask = var_41738_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41738_cast")];
            tensor<int32, [4]> var_41742_begin_0 = const()[name = tensor<string, []>("op_41742_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_41742_end_0 = const()[name = tensor<string, []>("op_41742_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_41742_end_mask_0 = const()[name = tensor<string, []>("op_41742_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41742_cast = slice_by_index(begin = var_41742_begin_0, end = var_41742_end_0, end_mask = var_41742_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41742_cast")];
            tensor<int32, [4]> var_41746_begin_0 = const()[name = tensor<string, []>("op_41746_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_41746_end_0 = const()[name = tensor<string, []>("op_41746_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_41746_end_mask_0 = const()[name = tensor<string, []>("op_41746_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41746_cast = slice_by_index(begin = var_41746_begin_0, end = var_41746_end_0, end_mask = var_41746_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41746_cast")];
            tensor<int32, [4]> var_41750_begin_0 = const()[name = tensor<string, []>("op_41750_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_41750_end_0 = const()[name = tensor<string, []>("op_41750_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_41750_end_mask_0 = const()[name = tensor<string, []>("op_41750_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41750_cast = slice_by_index(begin = var_41750_begin_0, end = var_41750_end_0, end_mask = var_41750_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41750_cast")];
            tensor<int32, [4]> var_41754_begin_0 = const()[name = tensor<string, []>("op_41754_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_41754_end_0 = const()[name = tensor<string, []>("op_41754_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_41754_end_mask_0 = const()[name = tensor<string, []>("op_41754_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41754_cast = slice_by_index(begin = var_41754_begin_0, end = var_41754_end_0, end_mask = var_41754_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41754_cast")];
            tensor<int32, [4]> var_41758_begin_0 = const()[name = tensor<string, []>("op_41758_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_41758_end_0 = const()[name = tensor<string, []>("op_41758_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_41758_end_mask_0 = const()[name = tensor<string, []>("op_41758_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41758_cast = slice_by_index(begin = var_41758_begin_0, end = var_41758_end_0, end_mask = var_41758_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41758_cast")];
            tensor<int32, [4]> var_41762_begin_0 = const()[name = tensor<string, []>("op_41762_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_41762_end_0 = const()[name = tensor<string, []>("op_41762_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_41762_end_mask_0 = const()[name = tensor<string, []>("op_41762_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41762_cast = slice_by_index(begin = var_41762_begin_0, end = var_41762_end_0, end_mask = var_41762_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41762_cast")];
            tensor<int32, [4]> var_41766_begin_0 = const()[name = tensor<string, []>("op_41766_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_41766_end_0 = const()[name = tensor<string, []>("op_41766_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_41766_end_mask_0 = const()[name = tensor<string, []>("op_41766_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41766_cast = slice_by_index(begin = var_41766_begin_0, end = var_41766_end_0, end_mask = var_41766_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41766_cast")];
            tensor<int32, [4]> var_41770_begin_0 = const()[name = tensor<string, []>("op_41770_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_41770_end_0 = const()[name = tensor<string, []>("op_41770_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_41770_end_mask_0 = const()[name = tensor<string, []>("op_41770_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41770_cast = slice_by_index(begin = var_41770_begin_0, end = var_41770_end_0, end_mask = var_41770_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41770_cast")];
            tensor<int32, [4]> var_41774_begin_0 = const()[name = tensor<string, []>("op_41774_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_41774_end_0 = const()[name = tensor<string, []>("op_41774_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_41774_end_mask_0 = const()[name = tensor<string, []>("op_41774_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41774_cast = slice_by_index(begin = var_41774_begin_0, end = var_41774_end_0, end_mask = var_41774_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41774_cast")];
            tensor<int32, [4]> var_41778_begin_0 = const()[name = tensor<string, []>("op_41778_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_41778_end_0 = const()[name = tensor<string, []>("op_41778_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_41778_end_mask_0 = const()[name = tensor<string, []>("op_41778_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41778_cast = slice_by_index(begin = var_41778_begin_0, end = var_41778_end_0, end_mask = var_41778_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41778_cast")];
            tensor<int32, [4]> var_41782_begin_0 = const()[name = tensor<string, []>("op_41782_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_41782_end_0 = const()[name = tensor<string, []>("op_41782_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_41782_end_mask_0 = const()[name = tensor<string, []>("op_41782_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41782_cast = slice_by_index(begin = var_41782_begin_0, end = var_41782_end_0, end_mask = var_41782_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41782_cast")];
            tensor<int32, [4]> var_41786_begin_0 = const()[name = tensor<string, []>("op_41786_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_41786_end_0 = const()[name = tensor<string, []>("op_41786_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_41786_end_mask_0 = const()[name = tensor<string, []>("op_41786_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41786_cast = slice_by_index(begin = var_41786_begin_0, end = var_41786_end_0, end_mask = var_41786_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41786_cast")];
            tensor<int32, [4]> var_41790_begin_0 = const()[name = tensor<string, []>("op_41790_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_41790_end_0 = const()[name = tensor<string, []>("op_41790_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_41790_end_mask_0 = const()[name = tensor<string, []>("op_41790_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41790_cast = slice_by_index(begin = var_41790_begin_0, end = var_41790_end_0, end_mask = var_41790_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41790_cast")];
            tensor<int32, [4]> var_41794_begin_0 = const()[name = tensor<string, []>("op_41794_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_41794_end_0 = const()[name = tensor<string, []>("op_41794_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_41794_end_mask_0 = const()[name = tensor<string, []>("op_41794_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41794_cast = slice_by_index(begin = var_41794_begin_0, end = var_41794_end_0, end_mask = var_41794_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41794_cast")];
            tensor<int32, [4]> var_41798_begin_0 = const()[name = tensor<string, []>("op_41798_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_41798_end_0 = const()[name = tensor<string, []>("op_41798_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_41798_end_mask_0 = const()[name = tensor<string, []>("op_41798_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41798_cast = slice_by_index(begin = var_41798_begin_0, end = var_41798_end_0, end_mask = var_41798_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41798_cast")];
            tensor<int32, [4]> var_41802_begin_0 = const()[name = tensor<string, []>("op_41802_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_41802_end_0 = const()[name = tensor<string, []>("op_41802_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_41802_end_mask_0 = const()[name = tensor<string, []>("op_41802_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41802_cast = slice_by_index(begin = var_41802_begin_0, end = var_41802_end_0, end_mask = var_41802_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41802_cast")];
            tensor<int32, [4]> var_41806_begin_0 = const()[name = tensor<string, []>("op_41806_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_41806_end_0 = const()[name = tensor<string, []>("op_41806_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_41806_end_mask_0 = const()[name = tensor<string, []>("op_41806_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_41806_cast = slice_by_index(begin = var_41806_begin_0, end = var_41806_end_0, end_mask = var_41806_end_mask_0, x = v_185_cast)[name = tensor<string, []>("op_41806_cast")];
            tensor<string, []> var_41810_equation_0 = const()[name = tensor<string, []>("op_41810_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41810_cast = einsum(equation = var_41810_equation_0, values = (var_41652_cast, var_41569_cast))[name = tensor<string, []>("op_41810_cast")];
            tensor<fp16, []> var_41811_to_fp16 = const()[name = tensor<string, []>("op_41811_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3521_cast = mul(x = var_41810_cast, y = var_41811_to_fp16)[name = tensor<string, []>("aw_3521_cast")];
            tensor<string, []> var_41814_equation_0 = const()[name = tensor<string, []>("op_41814_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41814_cast = einsum(equation = var_41814_equation_0, values = (var_41656_cast, var_41573_cast))[name = tensor<string, []>("op_41814_cast")];
            tensor<fp16, []> var_41815_to_fp16 = const()[name = tensor<string, []>("op_41815_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3523_cast = mul(x = var_41814_cast, y = var_41815_to_fp16)[name = tensor<string, []>("aw_3523_cast")];
            tensor<string, []> var_41818_equation_0 = const()[name = tensor<string, []>("op_41818_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41818_cast = einsum(equation = var_41818_equation_0, values = (var_41660_cast, var_41577_cast))[name = tensor<string, []>("op_41818_cast")];
            tensor<fp16, []> var_41819_to_fp16 = const()[name = tensor<string, []>("op_41819_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3525_cast = mul(x = var_41818_cast, y = var_41819_to_fp16)[name = tensor<string, []>("aw_3525_cast")];
            tensor<string, []> var_41822_equation_0 = const()[name = tensor<string, []>("op_41822_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41822_cast = einsum(equation = var_41822_equation_0, values = (var_41664_cast, var_41581_cast))[name = tensor<string, []>("op_41822_cast")];
            tensor<fp16, []> var_41823_to_fp16 = const()[name = tensor<string, []>("op_41823_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3527_cast = mul(x = var_41822_cast, y = var_41823_to_fp16)[name = tensor<string, []>("aw_3527_cast")];
            tensor<string, []> var_41826_equation_0 = const()[name = tensor<string, []>("op_41826_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41826_cast = einsum(equation = var_41826_equation_0, values = (var_41668_cast, var_41585_cast))[name = tensor<string, []>("op_41826_cast")];
            tensor<fp16, []> var_41827_to_fp16 = const()[name = tensor<string, []>("op_41827_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3529_cast = mul(x = var_41826_cast, y = var_41827_to_fp16)[name = tensor<string, []>("aw_3529_cast")];
            tensor<string, []> var_41830_equation_0 = const()[name = tensor<string, []>("op_41830_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41830_cast = einsum(equation = var_41830_equation_0, values = (var_41672_cast, var_41589_cast))[name = tensor<string, []>("op_41830_cast")];
            tensor<fp16, []> var_41831_to_fp16 = const()[name = tensor<string, []>("op_41831_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3531_cast = mul(x = var_41830_cast, y = var_41831_to_fp16)[name = tensor<string, []>("aw_3531_cast")];
            tensor<string, []> var_41834_equation_0 = const()[name = tensor<string, []>("op_41834_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41834_cast = einsum(equation = var_41834_equation_0, values = (var_41676_cast, var_41593_cast))[name = tensor<string, []>("op_41834_cast")];
            tensor<fp16, []> var_41835_to_fp16 = const()[name = tensor<string, []>("op_41835_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3533_cast = mul(x = var_41834_cast, y = var_41835_to_fp16)[name = tensor<string, []>("aw_3533_cast")];
            tensor<string, []> var_41838_equation_0 = const()[name = tensor<string, []>("op_41838_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41838_cast = einsum(equation = var_41838_equation_0, values = (var_41680_cast, var_41597_cast))[name = tensor<string, []>("op_41838_cast")];
            tensor<fp16, []> var_41839_to_fp16 = const()[name = tensor<string, []>("op_41839_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3535_cast = mul(x = var_41838_cast, y = var_41839_to_fp16)[name = tensor<string, []>("aw_3535_cast")];
            tensor<string, []> var_41842_equation_0 = const()[name = tensor<string, []>("op_41842_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41842_cast = einsum(equation = var_41842_equation_0, values = (var_41684_cast, var_41601_cast))[name = tensor<string, []>("op_41842_cast")];
            tensor<fp16, []> var_41843_to_fp16 = const()[name = tensor<string, []>("op_41843_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3537_cast = mul(x = var_41842_cast, y = var_41843_to_fp16)[name = tensor<string, []>("aw_3537_cast")];
            tensor<string, []> var_41846_equation_0 = const()[name = tensor<string, []>("op_41846_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41846_cast = einsum(equation = var_41846_equation_0, values = (var_41688_cast, var_41605_cast))[name = tensor<string, []>("op_41846_cast")];
            tensor<fp16, []> var_41847_to_fp16 = const()[name = tensor<string, []>("op_41847_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3539_cast = mul(x = var_41846_cast, y = var_41847_to_fp16)[name = tensor<string, []>("aw_3539_cast")];
            tensor<string, []> var_41850_equation_0 = const()[name = tensor<string, []>("op_41850_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41850_cast = einsum(equation = var_41850_equation_0, values = (var_41692_cast, var_41609_cast))[name = tensor<string, []>("op_41850_cast")];
            tensor<fp16, []> var_41851_to_fp16 = const()[name = tensor<string, []>("op_41851_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3541_cast = mul(x = var_41850_cast, y = var_41851_to_fp16)[name = tensor<string, []>("aw_3541_cast")];
            tensor<string, []> var_41854_equation_0 = const()[name = tensor<string, []>("op_41854_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41854_cast = einsum(equation = var_41854_equation_0, values = (var_41696_cast, var_41613_cast))[name = tensor<string, []>("op_41854_cast")];
            tensor<fp16, []> var_41855_to_fp16 = const()[name = tensor<string, []>("op_41855_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3543_cast = mul(x = var_41854_cast, y = var_41855_to_fp16)[name = tensor<string, []>("aw_3543_cast")];
            tensor<string, []> var_41858_equation_0 = const()[name = tensor<string, []>("op_41858_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41858_cast = einsum(equation = var_41858_equation_0, values = (var_41700_cast, var_41617_cast))[name = tensor<string, []>("op_41858_cast")];
            tensor<fp16, []> var_41859_to_fp16 = const()[name = tensor<string, []>("op_41859_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3545_cast = mul(x = var_41858_cast, y = var_41859_to_fp16)[name = tensor<string, []>("aw_3545_cast")];
            tensor<string, []> var_41862_equation_0 = const()[name = tensor<string, []>("op_41862_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41862_cast = einsum(equation = var_41862_equation_0, values = (var_41704_cast, var_41621_cast))[name = tensor<string, []>("op_41862_cast")];
            tensor<fp16, []> var_41863_to_fp16 = const()[name = tensor<string, []>("op_41863_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3547_cast = mul(x = var_41862_cast, y = var_41863_to_fp16)[name = tensor<string, []>("aw_3547_cast")];
            tensor<string, []> var_41866_equation_0 = const()[name = tensor<string, []>("op_41866_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41866_cast = einsum(equation = var_41866_equation_0, values = (var_41708_cast, var_41625_cast))[name = tensor<string, []>("op_41866_cast")];
            tensor<fp16, []> var_41867_to_fp16 = const()[name = tensor<string, []>("op_41867_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3549_cast = mul(x = var_41866_cast, y = var_41867_to_fp16)[name = tensor<string, []>("aw_3549_cast")];
            tensor<string, []> var_41870_equation_0 = const()[name = tensor<string, []>("op_41870_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41870_cast = einsum(equation = var_41870_equation_0, values = (var_41712_cast, var_41629_cast))[name = tensor<string, []>("op_41870_cast")];
            tensor<fp16, []> var_41871_to_fp16 = const()[name = tensor<string, []>("op_41871_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3551_cast = mul(x = var_41870_cast, y = var_41871_to_fp16)[name = tensor<string, []>("aw_3551_cast")];
            tensor<string, []> var_41874_equation_0 = const()[name = tensor<string, []>("op_41874_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41874_cast = einsum(equation = var_41874_equation_0, values = (var_41716_cast, var_41633_cast))[name = tensor<string, []>("op_41874_cast")];
            tensor<fp16, []> var_41875_to_fp16 = const()[name = tensor<string, []>("op_41875_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3553_cast = mul(x = var_41874_cast, y = var_41875_to_fp16)[name = tensor<string, []>("aw_3553_cast")];
            tensor<string, []> var_41878_equation_0 = const()[name = tensor<string, []>("op_41878_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41878_cast = einsum(equation = var_41878_equation_0, values = (var_41720_cast, var_41637_cast))[name = tensor<string, []>("op_41878_cast")];
            tensor<fp16, []> var_41879_to_fp16 = const()[name = tensor<string, []>("op_41879_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3555_cast = mul(x = var_41878_cast, y = var_41879_to_fp16)[name = tensor<string, []>("aw_3555_cast")];
            tensor<string, []> var_41882_equation_0 = const()[name = tensor<string, []>("op_41882_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41882_cast = einsum(equation = var_41882_equation_0, values = (var_41724_cast, var_41641_cast))[name = tensor<string, []>("op_41882_cast")];
            tensor<fp16, []> var_41883_to_fp16 = const()[name = tensor<string, []>("op_41883_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3557_cast = mul(x = var_41882_cast, y = var_41883_to_fp16)[name = tensor<string, []>("aw_3557_cast")];
            tensor<string, []> var_41886_equation_0 = const()[name = tensor<string, []>("op_41886_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41886_cast = einsum(equation = var_41886_equation_0, values = (var_41728_cast, var_41645_cast))[name = tensor<string, []>("op_41886_cast")];
            tensor<fp16, []> var_41887_to_fp16 = const()[name = tensor<string, []>("op_41887_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3559_cast = mul(x = var_41886_cast, y = var_41887_to_fp16)[name = tensor<string, []>("aw_3559_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41889_cast = softmax(axis = var_30385, x = aw_3521_cast)[name = tensor<string, []>("op_41889_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41890_cast = softmax(axis = var_30385, x = aw_3523_cast)[name = tensor<string, []>("op_41890_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41891_cast = softmax(axis = var_30385, x = aw_3525_cast)[name = tensor<string, []>("op_41891_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41892_cast = softmax(axis = var_30385, x = aw_3527_cast)[name = tensor<string, []>("op_41892_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41893_cast = softmax(axis = var_30385, x = aw_3529_cast)[name = tensor<string, []>("op_41893_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41894_cast = softmax(axis = var_30385, x = aw_3531_cast)[name = tensor<string, []>("op_41894_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41895_cast = softmax(axis = var_30385, x = aw_3533_cast)[name = tensor<string, []>("op_41895_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41896_cast = softmax(axis = var_30385, x = aw_3535_cast)[name = tensor<string, []>("op_41896_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41897_cast = softmax(axis = var_30385, x = aw_3537_cast)[name = tensor<string, []>("op_41897_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41898_cast = softmax(axis = var_30385, x = aw_3539_cast)[name = tensor<string, []>("op_41898_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41899_cast = softmax(axis = var_30385, x = aw_3541_cast)[name = tensor<string, []>("op_41899_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41900_cast = softmax(axis = var_30385, x = aw_3543_cast)[name = tensor<string, []>("op_41900_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41901_cast = softmax(axis = var_30385, x = aw_3545_cast)[name = tensor<string, []>("op_41901_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41902_cast = softmax(axis = var_30385, x = aw_3547_cast)[name = tensor<string, []>("op_41902_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41903_cast = softmax(axis = var_30385, x = aw_3549_cast)[name = tensor<string, []>("op_41903_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41904_cast = softmax(axis = var_30385, x = aw_3551_cast)[name = tensor<string, []>("op_41904_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41905_cast = softmax(axis = var_30385, x = aw_3553_cast)[name = tensor<string, []>("op_41905_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41906_cast = softmax(axis = var_30385, x = aw_3555_cast)[name = tensor<string, []>("op_41906_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41907_cast = softmax(axis = var_30385, x = aw_3557_cast)[name = tensor<string, []>("op_41907_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_41908_cast = softmax(axis = var_30385, x = aw_3559_cast)[name = tensor<string, []>("op_41908_cast")];
            tensor<string, []> var_41910_equation_0 = const()[name = tensor<string, []>("op_41910_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41910_cast = einsum(equation = var_41910_equation_0, values = (var_41730_cast, var_41889_cast))[name = tensor<string, []>("op_41910_cast")];
            tensor<string, []> var_41912_equation_0 = const()[name = tensor<string, []>("op_41912_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41912_cast = einsum(equation = var_41912_equation_0, values = (var_41734_cast, var_41890_cast))[name = tensor<string, []>("op_41912_cast")];
            tensor<string, []> var_41914_equation_0 = const()[name = tensor<string, []>("op_41914_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41914_cast = einsum(equation = var_41914_equation_0, values = (var_41738_cast, var_41891_cast))[name = tensor<string, []>("op_41914_cast")];
            tensor<string, []> var_41916_equation_0 = const()[name = tensor<string, []>("op_41916_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41916_cast = einsum(equation = var_41916_equation_0, values = (var_41742_cast, var_41892_cast))[name = tensor<string, []>("op_41916_cast")];
            tensor<string, []> var_41918_equation_0 = const()[name = tensor<string, []>("op_41918_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41918_cast = einsum(equation = var_41918_equation_0, values = (var_41746_cast, var_41893_cast))[name = tensor<string, []>("op_41918_cast")];
            tensor<string, []> var_41920_equation_0 = const()[name = tensor<string, []>("op_41920_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41920_cast = einsum(equation = var_41920_equation_0, values = (var_41750_cast, var_41894_cast))[name = tensor<string, []>("op_41920_cast")];
            tensor<string, []> var_41922_equation_0 = const()[name = tensor<string, []>("op_41922_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41922_cast = einsum(equation = var_41922_equation_0, values = (var_41754_cast, var_41895_cast))[name = tensor<string, []>("op_41922_cast")];
            tensor<string, []> var_41924_equation_0 = const()[name = tensor<string, []>("op_41924_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41924_cast = einsum(equation = var_41924_equation_0, values = (var_41758_cast, var_41896_cast))[name = tensor<string, []>("op_41924_cast")];
            tensor<string, []> var_41926_equation_0 = const()[name = tensor<string, []>("op_41926_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41926_cast = einsum(equation = var_41926_equation_0, values = (var_41762_cast, var_41897_cast))[name = tensor<string, []>("op_41926_cast")];
            tensor<string, []> var_41928_equation_0 = const()[name = tensor<string, []>("op_41928_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41928_cast = einsum(equation = var_41928_equation_0, values = (var_41766_cast, var_41898_cast))[name = tensor<string, []>("op_41928_cast")];
            tensor<string, []> var_41930_equation_0 = const()[name = tensor<string, []>("op_41930_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41930_cast = einsum(equation = var_41930_equation_0, values = (var_41770_cast, var_41899_cast))[name = tensor<string, []>("op_41930_cast")];
            tensor<string, []> var_41932_equation_0 = const()[name = tensor<string, []>("op_41932_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41932_cast = einsum(equation = var_41932_equation_0, values = (var_41774_cast, var_41900_cast))[name = tensor<string, []>("op_41932_cast")];
            tensor<string, []> var_41934_equation_0 = const()[name = tensor<string, []>("op_41934_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41934_cast = einsum(equation = var_41934_equation_0, values = (var_41778_cast, var_41901_cast))[name = tensor<string, []>("op_41934_cast")];
            tensor<string, []> var_41936_equation_0 = const()[name = tensor<string, []>("op_41936_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41936_cast = einsum(equation = var_41936_equation_0, values = (var_41782_cast, var_41902_cast))[name = tensor<string, []>("op_41936_cast")];
            tensor<string, []> var_41938_equation_0 = const()[name = tensor<string, []>("op_41938_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41938_cast = einsum(equation = var_41938_equation_0, values = (var_41786_cast, var_41903_cast))[name = tensor<string, []>("op_41938_cast")];
            tensor<string, []> var_41940_equation_0 = const()[name = tensor<string, []>("op_41940_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41940_cast = einsum(equation = var_41940_equation_0, values = (var_41790_cast, var_41904_cast))[name = tensor<string, []>("op_41940_cast")];
            tensor<string, []> var_41942_equation_0 = const()[name = tensor<string, []>("op_41942_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41942_cast = einsum(equation = var_41942_equation_0, values = (var_41794_cast, var_41905_cast))[name = tensor<string, []>("op_41942_cast")];
            tensor<string, []> var_41944_equation_0 = const()[name = tensor<string, []>("op_41944_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41944_cast = einsum(equation = var_41944_equation_0, values = (var_41798_cast, var_41906_cast))[name = tensor<string, []>("op_41944_cast")];
            tensor<string, []> var_41946_equation_0 = const()[name = tensor<string, []>("op_41946_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41946_cast = einsum(equation = var_41946_equation_0, values = (var_41802_cast, var_41907_cast))[name = tensor<string, []>("op_41946_cast")];
            tensor<string, []> var_41948_equation_0 = const()[name = tensor<string, []>("op_41948_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_41948_cast = einsum(equation = var_41948_equation_0, values = (var_41806_cast, var_41908_cast))[name = tensor<string, []>("op_41948_cast")];
            tensor<bool, []> input_555_interleave_0 = const()[name = tensor<string, []>("input_555_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_555_cast = concat(axis = var_30385, interleave = input_555_interleave_0, values = (var_41910_cast, var_41912_cast, var_41914_cast, var_41916_cast, var_41918_cast, var_41920_cast, var_41922_cast, var_41924_cast, var_41926_cast, var_41928_cast, var_41930_cast, var_41932_cast, var_41934_cast, var_41936_cast, var_41938_cast, var_41940_cast, var_41942_cast, var_41944_cast, var_41946_cast, var_41948_cast))[name = tensor<string, []>("input_555_cast")];
            tensor<int32, [2]> var_41954 = const()[name = tensor<string, []>("op_41954"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_41956 = const()[name = tensor<string, []>("op_41956"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_41958_pad_type_0 = const()[name = tensor<string, []>("op_41958_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_41958_pad_0 = const()[name = tensor<string, []>("op_41958_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3537307328)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3540584192)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_41958_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_out_0_bias_to_fp16, dilations = var_41956, groups = var_30385, pad = var_41958_pad_0, pad_type = var_41958_pad_type_0, strides = var_41954, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn1_to_out_0_weight_to_fp16, x = input_555_cast)[name = tensor<string, []>("op_41958_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_279_cast = add(x = var_41958_cast, y = inputs_277_cast)[name = tensor<string, []>("inputs_279_cast")];
            tensor<int32, [1]> var_41962 = const()[name = tensor<string, []>("op_41962"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_279_cast = reduce_mean(axes = var_41962, keep_dims = var_30380, x = inputs_279_cast)[name = tensor<string, []>("channels_mean_279_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_279_cast = sub(x = inputs_279_cast, y = channels_mean_279_cast)[name = tensor<string, []>("zero_mean_279_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_279_cast = mul(x = zero_mean_279_cast, y = zero_mean_279_cast)[name = tensor<string, []>("zero_mean_sq_279_cast")];
            tensor<int32, [1]> var_41966 = const()[name = tensor<string, []>("op_41966"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_41967_cast = reduce_mean(axes = var_41966, keep_dims = var_30380, x = zero_mean_sq_279_cast)[name = tensor<string, []>("op_41967_cast")];
            tensor<fp16, []> var_41968_to_fp16 = const()[name = tensor<string, []>("op_41968_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_41969_cast = add(x = var_41967_cast, y = var_41968_to_fp16)[name = tensor<string, []>("op_41969_cast")];
            tensor<fp16, []> denom_279_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_279_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_279_cast = rsqrt(epsilon = denom_279_epsilon_0_to_fp16, x = var_41969_cast)[name = tensor<string, []>("denom_279_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_279_cast = mul(x = zero_mean_279_cast, y = denom_279_cast)[name = tensor<string, []>("out_279_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_41973_to_fp16 = const()[name = tensor<string, []>("op_41973_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3540586816)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_41974_cast = add(x = out_279_cast, y = var_41973_to_fp16)[name = tensor<string, []>("op_41974_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_41976_to_fp16 = const()[name = tensor<string, []>("op_41976_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3540589440)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_375_cast = mul(x = var_41974_cast, y = var_41976_to_fp16)[name = tensor<string, []>("hidden_states_375_cast")];
            tensor<int32, [2]> var_41983 = const()[name = tensor<string, []>("op_41983"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_41985 = const()[name = tensor<string, []>("op_41985"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_187_pad_type_0 = const()[name = tensor<string, []>("q_187_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_187_pad_0 = const()[name = tensor<string, []>("q_187_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3540592064)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_187_cast = conv(dilations = var_41985, groups = var_30385, pad = q_187_pad_0, pad_type = q_187_pad_type_0, strides = var_41983, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_q_weight_to_fp16, x = hidden_states_375_cast)[name = tensor<string, []>("q_187_cast")];
            tensor<int32, [2]> var_41989 = const()[name = tensor<string, []>("op_41989"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_41991 = const()[name = tensor<string, []>("op_41991"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_373_pad_type_0 = const()[name = tensor<string, []>("k_373_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_373_pad_0 = const()[name = tensor<string, []>("k_373_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3543868928)))];
            tensor<fp16, [2, 1280, 1, 77]> k_373_cast = conv(dilations = var_41991, groups = var_30385, pad = k_373_pad_0, pad_type = k_373_pad_type_0, strides = var_41989, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_373_cast")];
            tensor<int32, [2]> var_41995 = const()[name = tensor<string, []>("op_41995"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_41997 = const()[name = tensor<string, []>("op_41997"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_187_pad_type_0 = const()[name = tensor<string, []>("v_187_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_187_pad_0 = const()[name = tensor<string, []>("v_187_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3549111872)))];
            tensor<fp16, [2, 1280, 1, 77]> v_187_cast = conv(dilations = var_41997, groups = var_30385, pad = v_187_pad_0, pad_type = v_187_pad_type_0, strides = var_41995, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_187_cast")];
            tensor<int32, [4]> var_42001_begin_0 = const()[name = tensor<string, []>("op_42001_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_42001_end_0 = const()[name = tensor<string, []>("op_42001_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_42001_end_mask_0 = const()[name = tensor<string, []>("op_42001_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42001_cast = slice_by_index(begin = var_42001_begin_0, end = var_42001_end_0, end_mask = var_42001_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42001_cast")];
            tensor<int32, [4]> var_42005_begin_0 = const()[name = tensor<string, []>("op_42005_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_42005_end_0 = const()[name = tensor<string, []>("op_42005_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_42005_end_mask_0 = const()[name = tensor<string, []>("op_42005_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42005_cast = slice_by_index(begin = var_42005_begin_0, end = var_42005_end_0, end_mask = var_42005_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42005_cast")];
            tensor<int32, [4]> var_42009_begin_0 = const()[name = tensor<string, []>("op_42009_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_42009_end_0 = const()[name = tensor<string, []>("op_42009_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_42009_end_mask_0 = const()[name = tensor<string, []>("op_42009_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42009_cast = slice_by_index(begin = var_42009_begin_0, end = var_42009_end_0, end_mask = var_42009_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42009_cast")];
            tensor<int32, [4]> var_42013_begin_0 = const()[name = tensor<string, []>("op_42013_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_42013_end_0 = const()[name = tensor<string, []>("op_42013_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_42013_end_mask_0 = const()[name = tensor<string, []>("op_42013_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42013_cast = slice_by_index(begin = var_42013_begin_0, end = var_42013_end_0, end_mask = var_42013_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42013_cast")];
            tensor<int32, [4]> var_42017_begin_0 = const()[name = tensor<string, []>("op_42017_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_42017_end_0 = const()[name = tensor<string, []>("op_42017_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_42017_end_mask_0 = const()[name = tensor<string, []>("op_42017_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42017_cast = slice_by_index(begin = var_42017_begin_0, end = var_42017_end_0, end_mask = var_42017_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42017_cast")];
            tensor<int32, [4]> var_42021_begin_0 = const()[name = tensor<string, []>("op_42021_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_42021_end_0 = const()[name = tensor<string, []>("op_42021_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_42021_end_mask_0 = const()[name = tensor<string, []>("op_42021_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42021_cast = slice_by_index(begin = var_42021_begin_0, end = var_42021_end_0, end_mask = var_42021_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42021_cast")];
            tensor<int32, [4]> var_42025_begin_0 = const()[name = tensor<string, []>("op_42025_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_42025_end_0 = const()[name = tensor<string, []>("op_42025_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_42025_end_mask_0 = const()[name = tensor<string, []>("op_42025_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42025_cast = slice_by_index(begin = var_42025_begin_0, end = var_42025_end_0, end_mask = var_42025_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42025_cast")];
            tensor<int32, [4]> var_42029_begin_0 = const()[name = tensor<string, []>("op_42029_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_42029_end_0 = const()[name = tensor<string, []>("op_42029_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_42029_end_mask_0 = const()[name = tensor<string, []>("op_42029_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42029_cast = slice_by_index(begin = var_42029_begin_0, end = var_42029_end_0, end_mask = var_42029_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42029_cast")];
            tensor<int32, [4]> var_42033_begin_0 = const()[name = tensor<string, []>("op_42033_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_42033_end_0 = const()[name = tensor<string, []>("op_42033_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_42033_end_mask_0 = const()[name = tensor<string, []>("op_42033_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42033_cast = slice_by_index(begin = var_42033_begin_0, end = var_42033_end_0, end_mask = var_42033_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42033_cast")];
            tensor<int32, [4]> var_42037_begin_0 = const()[name = tensor<string, []>("op_42037_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_42037_end_0 = const()[name = tensor<string, []>("op_42037_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_42037_end_mask_0 = const()[name = tensor<string, []>("op_42037_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42037_cast = slice_by_index(begin = var_42037_begin_0, end = var_42037_end_0, end_mask = var_42037_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42037_cast")];
            tensor<int32, [4]> var_42041_begin_0 = const()[name = tensor<string, []>("op_42041_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_42041_end_0 = const()[name = tensor<string, []>("op_42041_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_42041_end_mask_0 = const()[name = tensor<string, []>("op_42041_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42041_cast = slice_by_index(begin = var_42041_begin_0, end = var_42041_end_0, end_mask = var_42041_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42041_cast")];
            tensor<int32, [4]> var_42045_begin_0 = const()[name = tensor<string, []>("op_42045_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_42045_end_0 = const()[name = tensor<string, []>("op_42045_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_42045_end_mask_0 = const()[name = tensor<string, []>("op_42045_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42045_cast = slice_by_index(begin = var_42045_begin_0, end = var_42045_end_0, end_mask = var_42045_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42045_cast")];
            tensor<int32, [4]> var_42049_begin_0 = const()[name = tensor<string, []>("op_42049_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_42049_end_0 = const()[name = tensor<string, []>("op_42049_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_42049_end_mask_0 = const()[name = tensor<string, []>("op_42049_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42049_cast = slice_by_index(begin = var_42049_begin_0, end = var_42049_end_0, end_mask = var_42049_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42049_cast")];
            tensor<int32, [4]> var_42053_begin_0 = const()[name = tensor<string, []>("op_42053_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_42053_end_0 = const()[name = tensor<string, []>("op_42053_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_42053_end_mask_0 = const()[name = tensor<string, []>("op_42053_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42053_cast = slice_by_index(begin = var_42053_begin_0, end = var_42053_end_0, end_mask = var_42053_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42053_cast")];
            tensor<int32, [4]> var_42057_begin_0 = const()[name = tensor<string, []>("op_42057_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_42057_end_0 = const()[name = tensor<string, []>("op_42057_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_42057_end_mask_0 = const()[name = tensor<string, []>("op_42057_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42057_cast = slice_by_index(begin = var_42057_begin_0, end = var_42057_end_0, end_mask = var_42057_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42057_cast")];
            tensor<int32, [4]> var_42061_begin_0 = const()[name = tensor<string, []>("op_42061_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_42061_end_0 = const()[name = tensor<string, []>("op_42061_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_42061_end_mask_0 = const()[name = tensor<string, []>("op_42061_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42061_cast = slice_by_index(begin = var_42061_begin_0, end = var_42061_end_0, end_mask = var_42061_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42061_cast")];
            tensor<int32, [4]> var_42065_begin_0 = const()[name = tensor<string, []>("op_42065_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_42065_end_0 = const()[name = tensor<string, []>("op_42065_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_42065_end_mask_0 = const()[name = tensor<string, []>("op_42065_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42065_cast = slice_by_index(begin = var_42065_begin_0, end = var_42065_end_0, end_mask = var_42065_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42065_cast")];
            tensor<int32, [4]> var_42069_begin_0 = const()[name = tensor<string, []>("op_42069_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_42069_end_0 = const()[name = tensor<string, []>("op_42069_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_42069_end_mask_0 = const()[name = tensor<string, []>("op_42069_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42069_cast = slice_by_index(begin = var_42069_begin_0, end = var_42069_end_0, end_mask = var_42069_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42069_cast")];
            tensor<int32, [4]> var_42073_begin_0 = const()[name = tensor<string, []>("op_42073_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_42073_end_0 = const()[name = tensor<string, []>("op_42073_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_42073_end_mask_0 = const()[name = tensor<string, []>("op_42073_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42073_cast = slice_by_index(begin = var_42073_begin_0, end = var_42073_end_0, end_mask = var_42073_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42073_cast")];
            tensor<int32, [4]> var_42077_begin_0 = const()[name = tensor<string, []>("op_42077_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_42077_end_0 = const()[name = tensor<string, []>("op_42077_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_42077_end_mask_0 = const()[name = tensor<string, []>("op_42077_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42077_cast = slice_by_index(begin = var_42077_begin_0, end = var_42077_end_0, end_mask = var_42077_end_mask_0, x = q_187_cast)[name = tensor<string, []>("op_42077_cast")];
            tensor<int32, [4]> k_375_perm_0 = const()[name = tensor<string, []>("k_375_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_42084_begin_0 = const()[name = tensor<string, []>("op_42084_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_42084_end_0 = const()[name = tensor<string, []>("op_42084_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_42084_end_mask_0 = const()[name = tensor<string, []>("op_42084_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_46 = transpose(perm = k_375_perm_0, x = k_373_cast)[name = tensor<string, []>("transpose_46")];
            tensor<fp16, [2, 77, 1, 64]> var_42084_cast = slice_by_index(begin = var_42084_begin_0, end = var_42084_end_0, end_mask = var_42084_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42084_cast")];
            tensor<int32, [4]> var_42088_begin_0 = const()[name = tensor<string, []>("op_42088_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_42088_end_0 = const()[name = tensor<string, []>("op_42088_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_42088_end_mask_0 = const()[name = tensor<string, []>("op_42088_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42088_cast = slice_by_index(begin = var_42088_begin_0, end = var_42088_end_0, end_mask = var_42088_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42088_cast")];
            tensor<int32, [4]> var_42092_begin_0 = const()[name = tensor<string, []>("op_42092_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_42092_end_0 = const()[name = tensor<string, []>("op_42092_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_42092_end_mask_0 = const()[name = tensor<string, []>("op_42092_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42092_cast = slice_by_index(begin = var_42092_begin_0, end = var_42092_end_0, end_mask = var_42092_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42092_cast")];
            tensor<int32, [4]> var_42096_begin_0 = const()[name = tensor<string, []>("op_42096_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_42096_end_0 = const()[name = tensor<string, []>("op_42096_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_42096_end_mask_0 = const()[name = tensor<string, []>("op_42096_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42096_cast = slice_by_index(begin = var_42096_begin_0, end = var_42096_end_0, end_mask = var_42096_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42096_cast")];
            tensor<int32, [4]> var_42100_begin_0 = const()[name = tensor<string, []>("op_42100_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_42100_end_0 = const()[name = tensor<string, []>("op_42100_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_42100_end_mask_0 = const()[name = tensor<string, []>("op_42100_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42100_cast = slice_by_index(begin = var_42100_begin_0, end = var_42100_end_0, end_mask = var_42100_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42100_cast")];
            tensor<int32, [4]> var_42104_begin_0 = const()[name = tensor<string, []>("op_42104_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_42104_end_0 = const()[name = tensor<string, []>("op_42104_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_42104_end_mask_0 = const()[name = tensor<string, []>("op_42104_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42104_cast = slice_by_index(begin = var_42104_begin_0, end = var_42104_end_0, end_mask = var_42104_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42104_cast")];
            tensor<int32, [4]> var_42108_begin_0 = const()[name = tensor<string, []>("op_42108_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_42108_end_0 = const()[name = tensor<string, []>("op_42108_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_42108_end_mask_0 = const()[name = tensor<string, []>("op_42108_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42108_cast = slice_by_index(begin = var_42108_begin_0, end = var_42108_end_0, end_mask = var_42108_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42108_cast")];
            tensor<int32, [4]> var_42112_begin_0 = const()[name = tensor<string, []>("op_42112_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_42112_end_0 = const()[name = tensor<string, []>("op_42112_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_42112_end_mask_0 = const()[name = tensor<string, []>("op_42112_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42112_cast = slice_by_index(begin = var_42112_begin_0, end = var_42112_end_0, end_mask = var_42112_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42112_cast")];
            tensor<int32, [4]> var_42116_begin_0 = const()[name = tensor<string, []>("op_42116_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_42116_end_0 = const()[name = tensor<string, []>("op_42116_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_42116_end_mask_0 = const()[name = tensor<string, []>("op_42116_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42116_cast = slice_by_index(begin = var_42116_begin_0, end = var_42116_end_0, end_mask = var_42116_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42116_cast")];
            tensor<int32, [4]> var_42120_begin_0 = const()[name = tensor<string, []>("op_42120_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_42120_end_0 = const()[name = tensor<string, []>("op_42120_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_42120_end_mask_0 = const()[name = tensor<string, []>("op_42120_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42120_cast = slice_by_index(begin = var_42120_begin_0, end = var_42120_end_0, end_mask = var_42120_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42120_cast")];
            tensor<int32, [4]> var_42124_begin_0 = const()[name = tensor<string, []>("op_42124_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_42124_end_0 = const()[name = tensor<string, []>("op_42124_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_42124_end_mask_0 = const()[name = tensor<string, []>("op_42124_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42124_cast = slice_by_index(begin = var_42124_begin_0, end = var_42124_end_0, end_mask = var_42124_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42124_cast")];
            tensor<int32, [4]> var_42128_begin_0 = const()[name = tensor<string, []>("op_42128_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_42128_end_0 = const()[name = tensor<string, []>("op_42128_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_42128_end_mask_0 = const()[name = tensor<string, []>("op_42128_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42128_cast = slice_by_index(begin = var_42128_begin_0, end = var_42128_end_0, end_mask = var_42128_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42128_cast")];
            tensor<int32, [4]> var_42132_begin_0 = const()[name = tensor<string, []>("op_42132_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_42132_end_0 = const()[name = tensor<string, []>("op_42132_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_42132_end_mask_0 = const()[name = tensor<string, []>("op_42132_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42132_cast = slice_by_index(begin = var_42132_begin_0, end = var_42132_end_0, end_mask = var_42132_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42132_cast")];
            tensor<int32, [4]> var_42136_begin_0 = const()[name = tensor<string, []>("op_42136_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_42136_end_0 = const()[name = tensor<string, []>("op_42136_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_42136_end_mask_0 = const()[name = tensor<string, []>("op_42136_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42136_cast = slice_by_index(begin = var_42136_begin_0, end = var_42136_end_0, end_mask = var_42136_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42136_cast")];
            tensor<int32, [4]> var_42140_begin_0 = const()[name = tensor<string, []>("op_42140_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_42140_end_0 = const()[name = tensor<string, []>("op_42140_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_42140_end_mask_0 = const()[name = tensor<string, []>("op_42140_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42140_cast = slice_by_index(begin = var_42140_begin_0, end = var_42140_end_0, end_mask = var_42140_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42140_cast")];
            tensor<int32, [4]> var_42144_begin_0 = const()[name = tensor<string, []>("op_42144_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_42144_end_0 = const()[name = tensor<string, []>("op_42144_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_42144_end_mask_0 = const()[name = tensor<string, []>("op_42144_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42144_cast = slice_by_index(begin = var_42144_begin_0, end = var_42144_end_0, end_mask = var_42144_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42144_cast")];
            tensor<int32, [4]> var_42148_begin_0 = const()[name = tensor<string, []>("op_42148_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_42148_end_0 = const()[name = tensor<string, []>("op_42148_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_42148_end_mask_0 = const()[name = tensor<string, []>("op_42148_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42148_cast = slice_by_index(begin = var_42148_begin_0, end = var_42148_end_0, end_mask = var_42148_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42148_cast")];
            tensor<int32, [4]> var_42152_begin_0 = const()[name = tensor<string, []>("op_42152_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_42152_end_0 = const()[name = tensor<string, []>("op_42152_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_42152_end_mask_0 = const()[name = tensor<string, []>("op_42152_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42152_cast = slice_by_index(begin = var_42152_begin_0, end = var_42152_end_0, end_mask = var_42152_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42152_cast")];
            tensor<int32, [4]> var_42156_begin_0 = const()[name = tensor<string, []>("op_42156_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_42156_end_0 = const()[name = tensor<string, []>("op_42156_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_42156_end_mask_0 = const()[name = tensor<string, []>("op_42156_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42156_cast = slice_by_index(begin = var_42156_begin_0, end = var_42156_end_0, end_mask = var_42156_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42156_cast")];
            tensor<int32, [4]> var_42160_begin_0 = const()[name = tensor<string, []>("op_42160_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_42160_end_0 = const()[name = tensor<string, []>("op_42160_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_42160_end_mask_0 = const()[name = tensor<string, []>("op_42160_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42160_cast = slice_by_index(begin = var_42160_begin_0, end = var_42160_end_0, end_mask = var_42160_end_mask_0, x = transpose_46)[name = tensor<string, []>("op_42160_cast")];
            tensor<int32, [4]> var_42162_begin_0 = const()[name = tensor<string, []>("op_42162_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_42162_end_0 = const()[name = tensor<string, []>("op_42162_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_42162_end_mask_0 = const()[name = tensor<string, []>("op_42162_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42162_cast = slice_by_index(begin = var_42162_begin_0, end = var_42162_end_0, end_mask = var_42162_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42162_cast")];
            tensor<int32, [4]> var_42166_begin_0 = const()[name = tensor<string, []>("op_42166_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_42166_end_0 = const()[name = tensor<string, []>("op_42166_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_42166_end_mask_0 = const()[name = tensor<string, []>("op_42166_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42166_cast = slice_by_index(begin = var_42166_begin_0, end = var_42166_end_0, end_mask = var_42166_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42166_cast")];
            tensor<int32, [4]> var_42170_begin_0 = const()[name = tensor<string, []>("op_42170_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_42170_end_0 = const()[name = tensor<string, []>("op_42170_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_42170_end_mask_0 = const()[name = tensor<string, []>("op_42170_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42170_cast = slice_by_index(begin = var_42170_begin_0, end = var_42170_end_0, end_mask = var_42170_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42170_cast")];
            tensor<int32, [4]> var_42174_begin_0 = const()[name = tensor<string, []>("op_42174_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_42174_end_0 = const()[name = tensor<string, []>("op_42174_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_42174_end_mask_0 = const()[name = tensor<string, []>("op_42174_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42174_cast = slice_by_index(begin = var_42174_begin_0, end = var_42174_end_0, end_mask = var_42174_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42174_cast")];
            tensor<int32, [4]> var_42178_begin_0 = const()[name = tensor<string, []>("op_42178_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_42178_end_0 = const()[name = tensor<string, []>("op_42178_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_42178_end_mask_0 = const()[name = tensor<string, []>("op_42178_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42178_cast = slice_by_index(begin = var_42178_begin_0, end = var_42178_end_0, end_mask = var_42178_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42178_cast")];
            tensor<int32, [4]> var_42182_begin_0 = const()[name = tensor<string, []>("op_42182_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_42182_end_0 = const()[name = tensor<string, []>("op_42182_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_42182_end_mask_0 = const()[name = tensor<string, []>("op_42182_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42182_cast = slice_by_index(begin = var_42182_begin_0, end = var_42182_end_0, end_mask = var_42182_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42182_cast")];
            tensor<int32, [4]> var_42186_begin_0 = const()[name = tensor<string, []>("op_42186_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_42186_end_0 = const()[name = tensor<string, []>("op_42186_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_42186_end_mask_0 = const()[name = tensor<string, []>("op_42186_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42186_cast = slice_by_index(begin = var_42186_begin_0, end = var_42186_end_0, end_mask = var_42186_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42186_cast")];
            tensor<int32, [4]> var_42190_begin_0 = const()[name = tensor<string, []>("op_42190_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_42190_end_0 = const()[name = tensor<string, []>("op_42190_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_42190_end_mask_0 = const()[name = tensor<string, []>("op_42190_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42190_cast = slice_by_index(begin = var_42190_begin_0, end = var_42190_end_0, end_mask = var_42190_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42190_cast")];
            tensor<int32, [4]> var_42194_begin_0 = const()[name = tensor<string, []>("op_42194_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_42194_end_0 = const()[name = tensor<string, []>("op_42194_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_42194_end_mask_0 = const()[name = tensor<string, []>("op_42194_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42194_cast = slice_by_index(begin = var_42194_begin_0, end = var_42194_end_0, end_mask = var_42194_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42194_cast")];
            tensor<int32, [4]> var_42198_begin_0 = const()[name = tensor<string, []>("op_42198_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_42198_end_0 = const()[name = tensor<string, []>("op_42198_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_42198_end_mask_0 = const()[name = tensor<string, []>("op_42198_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42198_cast = slice_by_index(begin = var_42198_begin_0, end = var_42198_end_0, end_mask = var_42198_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42198_cast")];
            tensor<int32, [4]> var_42202_begin_0 = const()[name = tensor<string, []>("op_42202_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_42202_end_0 = const()[name = tensor<string, []>("op_42202_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_42202_end_mask_0 = const()[name = tensor<string, []>("op_42202_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42202_cast = slice_by_index(begin = var_42202_begin_0, end = var_42202_end_0, end_mask = var_42202_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42202_cast")];
            tensor<int32, [4]> var_42206_begin_0 = const()[name = tensor<string, []>("op_42206_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_42206_end_0 = const()[name = tensor<string, []>("op_42206_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_42206_end_mask_0 = const()[name = tensor<string, []>("op_42206_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42206_cast = slice_by_index(begin = var_42206_begin_0, end = var_42206_end_0, end_mask = var_42206_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42206_cast")];
            tensor<int32, [4]> var_42210_begin_0 = const()[name = tensor<string, []>("op_42210_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_42210_end_0 = const()[name = tensor<string, []>("op_42210_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_42210_end_mask_0 = const()[name = tensor<string, []>("op_42210_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42210_cast = slice_by_index(begin = var_42210_begin_0, end = var_42210_end_0, end_mask = var_42210_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42210_cast")];
            tensor<int32, [4]> var_42214_begin_0 = const()[name = tensor<string, []>("op_42214_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_42214_end_0 = const()[name = tensor<string, []>("op_42214_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_42214_end_mask_0 = const()[name = tensor<string, []>("op_42214_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42214_cast = slice_by_index(begin = var_42214_begin_0, end = var_42214_end_0, end_mask = var_42214_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42214_cast")];
            tensor<int32, [4]> var_42218_begin_0 = const()[name = tensor<string, []>("op_42218_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_42218_end_0 = const()[name = tensor<string, []>("op_42218_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_42218_end_mask_0 = const()[name = tensor<string, []>("op_42218_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42218_cast = slice_by_index(begin = var_42218_begin_0, end = var_42218_end_0, end_mask = var_42218_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42218_cast")];
            tensor<int32, [4]> var_42222_begin_0 = const()[name = tensor<string, []>("op_42222_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_42222_end_0 = const()[name = tensor<string, []>("op_42222_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_42222_end_mask_0 = const()[name = tensor<string, []>("op_42222_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42222_cast = slice_by_index(begin = var_42222_begin_0, end = var_42222_end_0, end_mask = var_42222_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42222_cast")];
            tensor<int32, [4]> var_42226_begin_0 = const()[name = tensor<string, []>("op_42226_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_42226_end_0 = const()[name = tensor<string, []>("op_42226_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_42226_end_mask_0 = const()[name = tensor<string, []>("op_42226_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42226_cast = slice_by_index(begin = var_42226_begin_0, end = var_42226_end_0, end_mask = var_42226_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42226_cast")];
            tensor<int32, [4]> var_42230_begin_0 = const()[name = tensor<string, []>("op_42230_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_42230_end_0 = const()[name = tensor<string, []>("op_42230_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_42230_end_mask_0 = const()[name = tensor<string, []>("op_42230_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42230_cast = slice_by_index(begin = var_42230_begin_0, end = var_42230_end_0, end_mask = var_42230_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42230_cast")];
            tensor<int32, [4]> var_42234_begin_0 = const()[name = tensor<string, []>("op_42234_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_42234_end_0 = const()[name = tensor<string, []>("op_42234_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_42234_end_mask_0 = const()[name = tensor<string, []>("op_42234_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42234_cast = slice_by_index(begin = var_42234_begin_0, end = var_42234_end_0, end_mask = var_42234_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42234_cast")];
            tensor<int32, [4]> var_42238_begin_0 = const()[name = tensor<string, []>("op_42238_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_42238_end_0 = const()[name = tensor<string, []>("op_42238_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_42238_end_mask_0 = const()[name = tensor<string, []>("op_42238_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_42238_cast = slice_by_index(begin = var_42238_begin_0, end = var_42238_end_0, end_mask = var_42238_end_mask_0, x = v_187_cast)[name = tensor<string, []>("op_42238_cast")];
            tensor<string, []> var_42242_equation_0 = const()[name = tensor<string, []>("op_42242_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42242_cast = einsum(equation = var_42242_equation_0, values = (var_42084_cast, var_42001_cast))[name = tensor<string, []>("op_42242_cast")];
            tensor<fp16, []> var_42243_to_fp16 = const()[name = tensor<string, []>("op_42243_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3561_cast = mul(x = var_42242_cast, y = var_42243_to_fp16)[name = tensor<string, []>("aw_3561_cast")];
            tensor<string, []> var_42246_equation_0 = const()[name = tensor<string, []>("op_42246_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42246_cast = einsum(equation = var_42246_equation_0, values = (var_42088_cast, var_42005_cast))[name = tensor<string, []>("op_42246_cast")];
            tensor<fp16, []> var_42247_to_fp16 = const()[name = tensor<string, []>("op_42247_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3563_cast = mul(x = var_42246_cast, y = var_42247_to_fp16)[name = tensor<string, []>("aw_3563_cast")];
            tensor<string, []> var_42250_equation_0 = const()[name = tensor<string, []>("op_42250_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42250_cast = einsum(equation = var_42250_equation_0, values = (var_42092_cast, var_42009_cast))[name = tensor<string, []>("op_42250_cast")];
            tensor<fp16, []> var_42251_to_fp16 = const()[name = tensor<string, []>("op_42251_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3565_cast = mul(x = var_42250_cast, y = var_42251_to_fp16)[name = tensor<string, []>("aw_3565_cast")];
            tensor<string, []> var_42254_equation_0 = const()[name = tensor<string, []>("op_42254_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42254_cast = einsum(equation = var_42254_equation_0, values = (var_42096_cast, var_42013_cast))[name = tensor<string, []>("op_42254_cast")];
            tensor<fp16, []> var_42255_to_fp16 = const()[name = tensor<string, []>("op_42255_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3567_cast = mul(x = var_42254_cast, y = var_42255_to_fp16)[name = tensor<string, []>("aw_3567_cast")];
            tensor<string, []> var_42258_equation_0 = const()[name = tensor<string, []>("op_42258_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42258_cast = einsum(equation = var_42258_equation_0, values = (var_42100_cast, var_42017_cast))[name = tensor<string, []>("op_42258_cast")];
            tensor<fp16, []> var_42259_to_fp16 = const()[name = tensor<string, []>("op_42259_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3569_cast = mul(x = var_42258_cast, y = var_42259_to_fp16)[name = tensor<string, []>("aw_3569_cast")];
            tensor<string, []> var_42262_equation_0 = const()[name = tensor<string, []>("op_42262_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42262_cast = einsum(equation = var_42262_equation_0, values = (var_42104_cast, var_42021_cast))[name = tensor<string, []>("op_42262_cast")];
            tensor<fp16, []> var_42263_to_fp16 = const()[name = tensor<string, []>("op_42263_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3571_cast = mul(x = var_42262_cast, y = var_42263_to_fp16)[name = tensor<string, []>("aw_3571_cast")];
            tensor<string, []> var_42266_equation_0 = const()[name = tensor<string, []>("op_42266_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42266_cast = einsum(equation = var_42266_equation_0, values = (var_42108_cast, var_42025_cast))[name = tensor<string, []>("op_42266_cast")];
            tensor<fp16, []> var_42267_to_fp16 = const()[name = tensor<string, []>("op_42267_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3573_cast = mul(x = var_42266_cast, y = var_42267_to_fp16)[name = tensor<string, []>("aw_3573_cast")];
            tensor<string, []> var_42270_equation_0 = const()[name = tensor<string, []>("op_42270_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42270_cast = einsum(equation = var_42270_equation_0, values = (var_42112_cast, var_42029_cast))[name = tensor<string, []>("op_42270_cast")];
            tensor<fp16, []> var_42271_to_fp16 = const()[name = tensor<string, []>("op_42271_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3575_cast = mul(x = var_42270_cast, y = var_42271_to_fp16)[name = tensor<string, []>("aw_3575_cast")];
            tensor<string, []> var_42274_equation_0 = const()[name = tensor<string, []>("op_42274_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42274_cast = einsum(equation = var_42274_equation_0, values = (var_42116_cast, var_42033_cast))[name = tensor<string, []>("op_42274_cast")];
            tensor<fp16, []> var_42275_to_fp16 = const()[name = tensor<string, []>("op_42275_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3577_cast = mul(x = var_42274_cast, y = var_42275_to_fp16)[name = tensor<string, []>("aw_3577_cast")];
            tensor<string, []> var_42278_equation_0 = const()[name = tensor<string, []>("op_42278_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42278_cast = einsum(equation = var_42278_equation_0, values = (var_42120_cast, var_42037_cast))[name = tensor<string, []>("op_42278_cast")];
            tensor<fp16, []> var_42279_to_fp16 = const()[name = tensor<string, []>("op_42279_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3579_cast = mul(x = var_42278_cast, y = var_42279_to_fp16)[name = tensor<string, []>("aw_3579_cast")];
            tensor<string, []> var_42282_equation_0 = const()[name = tensor<string, []>("op_42282_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42282_cast = einsum(equation = var_42282_equation_0, values = (var_42124_cast, var_42041_cast))[name = tensor<string, []>("op_42282_cast")];
            tensor<fp16, []> var_42283_to_fp16 = const()[name = tensor<string, []>("op_42283_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3581_cast = mul(x = var_42282_cast, y = var_42283_to_fp16)[name = tensor<string, []>("aw_3581_cast")];
            tensor<string, []> var_42286_equation_0 = const()[name = tensor<string, []>("op_42286_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42286_cast = einsum(equation = var_42286_equation_0, values = (var_42128_cast, var_42045_cast))[name = tensor<string, []>("op_42286_cast")];
            tensor<fp16, []> var_42287_to_fp16 = const()[name = tensor<string, []>("op_42287_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3583_cast = mul(x = var_42286_cast, y = var_42287_to_fp16)[name = tensor<string, []>("aw_3583_cast")];
            tensor<string, []> var_42290_equation_0 = const()[name = tensor<string, []>("op_42290_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42290_cast = einsum(equation = var_42290_equation_0, values = (var_42132_cast, var_42049_cast))[name = tensor<string, []>("op_42290_cast")];
            tensor<fp16, []> var_42291_to_fp16 = const()[name = tensor<string, []>("op_42291_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3585_cast = mul(x = var_42290_cast, y = var_42291_to_fp16)[name = tensor<string, []>("aw_3585_cast")];
            tensor<string, []> var_42294_equation_0 = const()[name = tensor<string, []>("op_42294_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42294_cast = einsum(equation = var_42294_equation_0, values = (var_42136_cast, var_42053_cast))[name = tensor<string, []>("op_42294_cast")];
            tensor<fp16, []> var_42295_to_fp16 = const()[name = tensor<string, []>("op_42295_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3587_cast = mul(x = var_42294_cast, y = var_42295_to_fp16)[name = tensor<string, []>("aw_3587_cast")];
            tensor<string, []> var_42298_equation_0 = const()[name = tensor<string, []>("op_42298_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42298_cast = einsum(equation = var_42298_equation_0, values = (var_42140_cast, var_42057_cast))[name = tensor<string, []>("op_42298_cast")];
            tensor<fp16, []> var_42299_to_fp16 = const()[name = tensor<string, []>("op_42299_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3589_cast = mul(x = var_42298_cast, y = var_42299_to_fp16)[name = tensor<string, []>("aw_3589_cast")];
            tensor<string, []> var_42302_equation_0 = const()[name = tensor<string, []>("op_42302_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42302_cast = einsum(equation = var_42302_equation_0, values = (var_42144_cast, var_42061_cast))[name = tensor<string, []>("op_42302_cast")];
            tensor<fp16, []> var_42303_to_fp16 = const()[name = tensor<string, []>("op_42303_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3591_cast = mul(x = var_42302_cast, y = var_42303_to_fp16)[name = tensor<string, []>("aw_3591_cast")];
            tensor<string, []> var_42306_equation_0 = const()[name = tensor<string, []>("op_42306_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42306_cast = einsum(equation = var_42306_equation_0, values = (var_42148_cast, var_42065_cast))[name = tensor<string, []>("op_42306_cast")];
            tensor<fp16, []> var_42307_to_fp16 = const()[name = tensor<string, []>("op_42307_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3593_cast = mul(x = var_42306_cast, y = var_42307_to_fp16)[name = tensor<string, []>("aw_3593_cast")];
            tensor<string, []> var_42310_equation_0 = const()[name = tensor<string, []>("op_42310_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42310_cast = einsum(equation = var_42310_equation_0, values = (var_42152_cast, var_42069_cast))[name = tensor<string, []>("op_42310_cast")];
            tensor<fp16, []> var_42311_to_fp16 = const()[name = tensor<string, []>("op_42311_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3595_cast = mul(x = var_42310_cast, y = var_42311_to_fp16)[name = tensor<string, []>("aw_3595_cast")];
            tensor<string, []> var_42314_equation_0 = const()[name = tensor<string, []>("op_42314_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42314_cast = einsum(equation = var_42314_equation_0, values = (var_42156_cast, var_42073_cast))[name = tensor<string, []>("op_42314_cast")];
            tensor<fp16, []> var_42315_to_fp16 = const()[name = tensor<string, []>("op_42315_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3597_cast = mul(x = var_42314_cast, y = var_42315_to_fp16)[name = tensor<string, []>("aw_3597_cast")];
            tensor<string, []> var_42318_equation_0 = const()[name = tensor<string, []>("op_42318_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_42318_cast = einsum(equation = var_42318_equation_0, values = (var_42160_cast, var_42077_cast))[name = tensor<string, []>("op_42318_cast")];
            tensor<fp16, []> var_42319_to_fp16 = const()[name = tensor<string, []>("op_42319_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3599_cast = mul(x = var_42318_cast, y = var_42319_to_fp16)[name = tensor<string, []>("aw_3599_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42321_cast = softmax(axis = var_30385, x = aw_3561_cast)[name = tensor<string, []>("op_42321_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42322_cast = softmax(axis = var_30385, x = aw_3563_cast)[name = tensor<string, []>("op_42322_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42323_cast = softmax(axis = var_30385, x = aw_3565_cast)[name = tensor<string, []>("op_42323_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42324_cast = softmax(axis = var_30385, x = aw_3567_cast)[name = tensor<string, []>("op_42324_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42325_cast = softmax(axis = var_30385, x = aw_3569_cast)[name = tensor<string, []>("op_42325_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42326_cast = softmax(axis = var_30385, x = aw_3571_cast)[name = tensor<string, []>("op_42326_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42327_cast = softmax(axis = var_30385, x = aw_3573_cast)[name = tensor<string, []>("op_42327_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42328_cast = softmax(axis = var_30385, x = aw_3575_cast)[name = tensor<string, []>("op_42328_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42329_cast = softmax(axis = var_30385, x = aw_3577_cast)[name = tensor<string, []>("op_42329_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42330_cast = softmax(axis = var_30385, x = aw_3579_cast)[name = tensor<string, []>("op_42330_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42331_cast = softmax(axis = var_30385, x = aw_3581_cast)[name = tensor<string, []>("op_42331_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42332_cast = softmax(axis = var_30385, x = aw_3583_cast)[name = tensor<string, []>("op_42332_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42333_cast = softmax(axis = var_30385, x = aw_3585_cast)[name = tensor<string, []>("op_42333_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42334_cast = softmax(axis = var_30385, x = aw_3587_cast)[name = tensor<string, []>("op_42334_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42335_cast = softmax(axis = var_30385, x = aw_3589_cast)[name = tensor<string, []>("op_42335_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42336_cast = softmax(axis = var_30385, x = aw_3591_cast)[name = tensor<string, []>("op_42336_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42337_cast = softmax(axis = var_30385, x = aw_3593_cast)[name = tensor<string, []>("op_42337_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42338_cast = softmax(axis = var_30385, x = aw_3595_cast)[name = tensor<string, []>("op_42338_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42339_cast = softmax(axis = var_30385, x = aw_3597_cast)[name = tensor<string, []>("op_42339_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_42340_cast = softmax(axis = var_30385, x = aw_3599_cast)[name = tensor<string, []>("op_42340_cast")];
            tensor<string, []> var_42342_equation_0 = const()[name = tensor<string, []>("op_42342_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42342_cast = einsum(equation = var_42342_equation_0, values = (var_42162_cast, var_42321_cast))[name = tensor<string, []>("op_42342_cast")];
            tensor<string, []> var_42344_equation_0 = const()[name = tensor<string, []>("op_42344_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42344_cast = einsum(equation = var_42344_equation_0, values = (var_42166_cast, var_42322_cast))[name = tensor<string, []>("op_42344_cast")];
            tensor<string, []> var_42346_equation_0 = const()[name = tensor<string, []>("op_42346_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42346_cast = einsum(equation = var_42346_equation_0, values = (var_42170_cast, var_42323_cast))[name = tensor<string, []>("op_42346_cast")];
            tensor<string, []> var_42348_equation_0 = const()[name = tensor<string, []>("op_42348_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42348_cast = einsum(equation = var_42348_equation_0, values = (var_42174_cast, var_42324_cast))[name = tensor<string, []>("op_42348_cast")];
            tensor<string, []> var_42350_equation_0 = const()[name = tensor<string, []>("op_42350_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42350_cast = einsum(equation = var_42350_equation_0, values = (var_42178_cast, var_42325_cast))[name = tensor<string, []>("op_42350_cast")];
            tensor<string, []> var_42352_equation_0 = const()[name = tensor<string, []>("op_42352_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42352_cast = einsum(equation = var_42352_equation_0, values = (var_42182_cast, var_42326_cast))[name = tensor<string, []>("op_42352_cast")];
            tensor<string, []> var_42354_equation_0 = const()[name = tensor<string, []>("op_42354_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42354_cast = einsum(equation = var_42354_equation_0, values = (var_42186_cast, var_42327_cast))[name = tensor<string, []>("op_42354_cast")];
            tensor<string, []> var_42356_equation_0 = const()[name = tensor<string, []>("op_42356_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42356_cast = einsum(equation = var_42356_equation_0, values = (var_42190_cast, var_42328_cast))[name = tensor<string, []>("op_42356_cast")];
            tensor<string, []> var_42358_equation_0 = const()[name = tensor<string, []>("op_42358_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42358_cast = einsum(equation = var_42358_equation_0, values = (var_42194_cast, var_42329_cast))[name = tensor<string, []>("op_42358_cast")];
            tensor<string, []> var_42360_equation_0 = const()[name = tensor<string, []>("op_42360_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42360_cast = einsum(equation = var_42360_equation_0, values = (var_42198_cast, var_42330_cast))[name = tensor<string, []>("op_42360_cast")];
            tensor<string, []> var_42362_equation_0 = const()[name = tensor<string, []>("op_42362_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42362_cast = einsum(equation = var_42362_equation_0, values = (var_42202_cast, var_42331_cast))[name = tensor<string, []>("op_42362_cast")];
            tensor<string, []> var_42364_equation_0 = const()[name = tensor<string, []>("op_42364_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42364_cast = einsum(equation = var_42364_equation_0, values = (var_42206_cast, var_42332_cast))[name = tensor<string, []>("op_42364_cast")];
            tensor<string, []> var_42366_equation_0 = const()[name = tensor<string, []>("op_42366_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42366_cast = einsum(equation = var_42366_equation_0, values = (var_42210_cast, var_42333_cast))[name = tensor<string, []>("op_42366_cast")];
            tensor<string, []> var_42368_equation_0 = const()[name = tensor<string, []>("op_42368_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42368_cast = einsum(equation = var_42368_equation_0, values = (var_42214_cast, var_42334_cast))[name = tensor<string, []>("op_42368_cast")];
            tensor<string, []> var_42370_equation_0 = const()[name = tensor<string, []>("op_42370_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42370_cast = einsum(equation = var_42370_equation_0, values = (var_42218_cast, var_42335_cast))[name = tensor<string, []>("op_42370_cast")];
            tensor<string, []> var_42372_equation_0 = const()[name = tensor<string, []>("op_42372_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42372_cast = einsum(equation = var_42372_equation_0, values = (var_42222_cast, var_42336_cast))[name = tensor<string, []>("op_42372_cast")];
            tensor<string, []> var_42374_equation_0 = const()[name = tensor<string, []>("op_42374_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42374_cast = einsum(equation = var_42374_equation_0, values = (var_42226_cast, var_42337_cast))[name = tensor<string, []>("op_42374_cast")];
            tensor<string, []> var_42376_equation_0 = const()[name = tensor<string, []>("op_42376_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42376_cast = einsum(equation = var_42376_equation_0, values = (var_42230_cast, var_42338_cast))[name = tensor<string, []>("op_42376_cast")];
            tensor<string, []> var_42378_equation_0 = const()[name = tensor<string, []>("op_42378_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42378_cast = einsum(equation = var_42378_equation_0, values = (var_42234_cast, var_42339_cast))[name = tensor<string, []>("op_42378_cast")];
            tensor<string, []> var_42380_equation_0 = const()[name = tensor<string, []>("op_42380_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42380_cast = einsum(equation = var_42380_equation_0, values = (var_42238_cast, var_42340_cast))[name = tensor<string, []>("op_42380_cast")];
            tensor<bool, []> input_557_interleave_0 = const()[name = tensor<string, []>("input_557_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_557_cast = concat(axis = var_30385, interleave = input_557_interleave_0, values = (var_42342_cast, var_42344_cast, var_42346_cast, var_42348_cast, var_42350_cast, var_42352_cast, var_42354_cast, var_42356_cast, var_42358_cast, var_42360_cast, var_42362_cast, var_42364_cast, var_42366_cast, var_42368_cast, var_42370_cast, var_42372_cast, var_42374_cast, var_42376_cast, var_42378_cast, var_42380_cast))[name = tensor<string, []>("input_557_cast")];
            tensor<int32, [2]> var_42386 = const()[name = tensor<string, []>("op_42386"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_42388 = const()[name = tensor<string, []>("op_42388"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_42390_pad_type_0 = const()[name = tensor<string, []>("op_42390_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_42390_pad_0 = const()[name = tensor<string, []>("op_42390_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3554354816)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3557631680)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_42390_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_out_0_bias_to_fp16, dilations = var_42388, groups = var_30385, pad = var_42390_pad_0, pad_type = var_42390_pad_type_0, strides = var_42386, weight = up_blocks_0_attentions_1_transformer_blocks_2_attn2_to_out_0_weight_to_fp16, x = input_557_cast)[name = tensor<string, []>("op_42390_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_281_cast = add(x = var_42390_cast, y = inputs_279_cast)[name = tensor<string, []>("inputs_281_cast")];
            tensor<int32, [1]> var_42394 = const()[name = tensor<string, []>("op_42394"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_281_cast = reduce_mean(axes = var_42394, keep_dims = var_30380, x = inputs_281_cast)[name = tensor<string, []>("channels_mean_281_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_281_cast = sub(x = inputs_281_cast, y = channels_mean_281_cast)[name = tensor<string, []>("zero_mean_281_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_281_cast = mul(x = zero_mean_281_cast, y = zero_mean_281_cast)[name = tensor<string, []>("zero_mean_sq_281_cast")];
            tensor<int32, [1]> var_42398 = const()[name = tensor<string, []>("op_42398"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_42399_cast = reduce_mean(axes = var_42398, keep_dims = var_30380, x = zero_mean_sq_281_cast)[name = tensor<string, []>("op_42399_cast")];
            tensor<fp16, []> var_42400_to_fp16 = const()[name = tensor<string, []>("op_42400_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_42401_cast = add(x = var_42399_cast, y = var_42400_to_fp16)[name = tensor<string, []>("op_42401_cast")];
            tensor<fp16, []> denom_281_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_281_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_281_cast = rsqrt(epsilon = denom_281_epsilon_0_to_fp16, x = var_42401_cast)[name = tensor<string, []>("denom_281_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_281_cast = mul(x = zero_mean_281_cast, y = denom_281_cast)[name = tensor<string, []>("out_281_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_42405_to_fp16 = const()[name = tensor<string, []>("op_42405_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3557634304)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_42406_cast = add(x = out_281_cast, y = var_42405_to_fp16)[name = tensor<string, []>("op_42406_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_42408_to_fp16 = const()[name = tensor<string, []>("op_42408_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3557636928)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_559_cast = mul(x = var_42406_cast, y = var_42408_to_fp16)[name = tensor<string, []>("input_559_cast")];
            tensor<int32, [2]> var_42416 = const()[name = tensor<string, []>("op_42416"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_42418 = const()[name = tensor<string, []>("op_42418"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_42420_pad_type_0 = const()[name = tensor<string, []>("op_42420_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_42420_pad_0 = const()[name = tensor<string, []>("op_42420_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3557639552)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_1_transformer_blocks_2_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3583854016)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_42420_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_2_ff_net_0_proj_bias_to_fp16, dilations = var_42418, groups = var_30385, pad = var_42420_pad_0, pad_type = var_42420_pad_type_0, strides = var_42416, weight = up_blocks_0_attentions_1_transformer_blocks_2_ff_net_0_proj_weight_to_fp16, x = input_559_cast)[name = tensor<string, []>("op_42420_cast")];
            tensor<int32, [2]> var_42421_split_sizes_0 = const()[name = tensor<string, []>("op_42421_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_42421_axis_0 = const()[name = tensor<string, []>("op_42421_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_42421_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_42421_cast_1 = split(axis = var_42421_axis_0, split_sizes = var_42421_split_sizes_0, x = var_42420_cast)[name = tensor<string, []>("op_42421_cast")];
            tensor<string, []> var_42423_mode_0 = const()[name = tensor<string, []>("op_42423_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_42423_cast = gelu(mode = var_42423_mode_0, x = var_42421_cast_1)[name = tensor<string, []>("op_42423_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_561_cast = mul(x = var_42421_cast_0, y = var_42423_cast)[name = tensor<string, []>("input_561_cast")];
            tensor<int32, [2]> var_42427 = const()[name = tensor<string, []>("op_42427"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_42429 = const()[name = tensor<string, []>("op_42429"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_42431_pad_type_0 = const()[name = tensor<string, []>("op_42431_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_42431_pad_0 = const()[name = tensor<string, []>("op_42431_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_2_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3583874560)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_2_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_2_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3596981824)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_42431_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_2_ff_net_2_bias_to_fp16, dilations = var_42429, groups = var_30385, pad = var_42431_pad_0, pad_type = var_42431_pad_type_0, strides = var_42427, weight = up_blocks_0_attentions_1_transformer_blocks_2_ff_net_2_weight_to_fp16, x = input_561_cast)[name = tensor<string, []>("op_42431_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_283_cast = add(x = var_42431_cast, y = inputs_281_cast)[name = tensor<string, []>("inputs_283_cast")];
            tensor<int32, [1]> var_42441 = const()[name = tensor<string, []>("op_42441"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_283_cast = reduce_mean(axes = var_42441, keep_dims = var_30380, x = inputs_283_cast)[name = tensor<string, []>("channels_mean_283_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_283_cast = sub(x = inputs_283_cast, y = channels_mean_283_cast)[name = tensor<string, []>("zero_mean_283_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_283_cast = mul(x = zero_mean_283_cast, y = zero_mean_283_cast)[name = tensor<string, []>("zero_mean_sq_283_cast")];
            tensor<int32, [1]> var_42445 = const()[name = tensor<string, []>("op_42445"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_42446_cast = reduce_mean(axes = var_42445, keep_dims = var_30380, x = zero_mean_sq_283_cast)[name = tensor<string, []>("op_42446_cast")];
            tensor<fp16, []> var_42447_to_fp16 = const()[name = tensor<string, []>("op_42447_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_42448_cast = add(x = var_42446_cast, y = var_42447_to_fp16)[name = tensor<string, []>("op_42448_cast")];
            tensor<fp16, []> denom_283_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_283_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_283_cast = rsqrt(epsilon = denom_283_epsilon_0_to_fp16, x = var_42448_cast)[name = tensor<string, []>("denom_283_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_283_cast = mul(x = zero_mean_283_cast, y = denom_283_cast)[name = tensor<string, []>("out_283_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_42452_to_fp16 = const()[name = tensor<string, []>("op_42452_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3596984448)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_42453_cast = add(x = out_283_cast, y = var_42452_to_fp16)[name = tensor<string, []>("op_42453_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_42455_to_fp16 = const()[name = tensor<string, []>("op_42455_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3596987072)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_379_cast = mul(x = var_42453_cast, y = var_42455_to_fp16)[name = tensor<string, []>("hidden_states_379_cast")];
            tensor<int32, [2]> var_42462 = const()[name = tensor<string, []>("op_42462"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_42464 = const()[name = tensor<string, []>("op_42464"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_189_pad_type_0 = const()[name = tensor<string, []>("q_189_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_189_pad_0 = const()[name = tensor<string, []>("q_189_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3596989696)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_189_cast = conv(dilations = var_42464, groups = var_30385, pad = q_189_pad_0, pad_type = q_189_pad_type_0, strides = var_42462, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_q_weight_to_fp16, x = hidden_states_379_cast)[name = tensor<string, []>("q_189_cast")];
            tensor<int32, [2]> var_42468 = const()[name = tensor<string, []>("op_42468"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_42470 = const()[name = tensor<string, []>("op_42470"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_377_pad_type_0 = const()[name = tensor<string, []>("k_377_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_377_pad_0 = const()[name = tensor<string, []>("k_377_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3600266560)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_377_cast = conv(dilations = var_42470, groups = var_30385, pad = k_377_pad_0, pad_type = k_377_pad_type_0, strides = var_42468, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_k_weight_to_fp16, x = hidden_states_379_cast)[name = tensor<string, []>("k_377_cast")];
            tensor<int32, [2]> var_42474 = const()[name = tensor<string, []>("op_42474"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_42476 = const()[name = tensor<string, []>("op_42476"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_189_pad_type_0 = const()[name = tensor<string, []>("v_189_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_189_pad_0 = const()[name = tensor<string, []>("v_189_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3603543424)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_189_cast = conv(dilations = var_42476, groups = var_30385, pad = v_189_pad_0, pad_type = v_189_pad_type_0, strides = var_42474, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_v_weight_to_fp16, x = hidden_states_379_cast)[name = tensor<string, []>("v_189_cast")];
            tensor<int32, [4]> var_42480_begin_0 = const()[name = tensor<string, []>("op_42480_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_42480_end_0 = const()[name = tensor<string, []>("op_42480_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_42480_end_mask_0 = const()[name = tensor<string, []>("op_42480_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42480_cast = slice_by_index(begin = var_42480_begin_0, end = var_42480_end_0, end_mask = var_42480_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42480_cast")];
            tensor<int32, [4]> var_42484_begin_0 = const()[name = tensor<string, []>("op_42484_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_42484_end_0 = const()[name = tensor<string, []>("op_42484_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_42484_end_mask_0 = const()[name = tensor<string, []>("op_42484_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42484_cast = slice_by_index(begin = var_42484_begin_0, end = var_42484_end_0, end_mask = var_42484_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42484_cast")];
            tensor<int32, [4]> var_42488_begin_0 = const()[name = tensor<string, []>("op_42488_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_42488_end_0 = const()[name = tensor<string, []>("op_42488_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_42488_end_mask_0 = const()[name = tensor<string, []>("op_42488_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42488_cast = slice_by_index(begin = var_42488_begin_0, end = var_42488_end_0, end_mask = var_42488_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42488_cast")];
            tensor<int32, [4]> var_42492_begin_0 = const()[name = tensor<string, []>("op_42492_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_42492_end_0 = const()[name = tensor<string, []>("op_42492_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_42492_end_mask_0 = const()[name = tensor<string, []>("op_42492_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42492_cast = slice_by_index(begin = var_42492_begin_0, end = var_42492_end_0, end_mask = var_42492_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42492_cast")];
            tensor<int32, [4]> var_42496_begin_0 = const()[name = tensor<string, []>("op_42496_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_42496_end_0 = const()[name = tensor<string, []>("op_42496_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_42496_end_mask_0 = const()[name = tensor<string, []>("op_42496_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42496_cast = slice_by_index(begin = var_42496_begin_0, end = var_42496_end_0, end_mask = var_42496_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42496_cast")];
            tensor<int32, [4]> var_42500_begin_0 = const()[name = tensor<string, []>("op_42500_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_42500_end_0 = const()[name = tensor<string, []>("op_42500_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_42500_end_mask_0 = const()[name = tensor<string, []>("op_42500_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42500_cast = slice_by_index(begin = var_42500_begin_0, end = var_42500_end_0, end_mask = var_42500_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42500_cast")];
            tensor<int32, [4]> var_42504_begin_0 = const()[name = tensor<string, []>("op_42504_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_42504_end_0 = const()[name = tensor<string, []>("op_42504_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_42504_end_mask_0 = const()[name = tensor<string, []>("op_42504_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42504_cast = slice_by_index(begin = var_42504_begin_0, end = var_42504_end_0, end_mask = var_42504_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42504_cast")];
            tensor<int32, [4]> var_42508_begin_0 = const()[name = tensor<string, []>("op_42508_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_42508_end_0 = const()[name = tensor<string, []>("op_42508_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_42508_end_mask_0 = const()[name = tensor<string, []>("op_42508_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42508_cast = slice_by_index(begin = var_42508_begin_0, end = var_42508_end_0, end_mask = var_42508_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42508_cast")];
            tensor<int32, [4]> var_42512_begin_0 = const()[name = tensor<string, []>("op_42512_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_42512_end_0 = const()[name = tensor<string, []>("op_42512_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_42512_end_mask_0 = const()[name = tensor<string, []>("op_42512_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42512_cast = slice_by_index(begin = var_42512_begin_0, end = var_42512_end_0, end_mask = var_42512_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42512_cast")];
            tensor<int32, [4]> var_42516_begin_0 = const()[name = tensor<string, []>("op_42516_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_42516_end_0 = const()[name = tensor<string, []>("op_42516_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_42516_end_mask_0 = const()[name = tensor<string, []>("op_42516_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42516_cast = slice_by_index(begin = var_42516_begin_0, end = var_42516_end_0, end_mask = var_42516_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42516_cast")];
            tensor<int32, [4]> var_42520_begin_0 = const()[name = tensor<string, []>("op_42520_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_42520_end_0 = const()[name = tensor<string, []>("op_42520_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_42520_end_mask_0 = const()[name = tensor<string, []>("op_42520_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42520_cast = slice_by_index(begin = var_42520_begin_0, end = var_42520_end_0, end_mask = var_42520_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42520_cast")];
            tensor<int32, [4]> var_42524_begin_0 = const()[name = tensor<string, []>("op_42524_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_42524_end_0 = const()[name = tensor<string, []>("op_42524_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_42524_end_mask_0 = const()[name = tensor<string, []>("op_42524_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42524_cast = slice_by_index(begin = var_42524_begin_0, end = var_42524_end_0, end_mask = var_42524_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42524_cast")];
            tensor<int32, [4]> var_42528_begin_0 = const()[name = tensor<string, []>("op_42528_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_42528_end_0 = const()[name = tensor<string, []>("op_42528_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_42528_end_mask_0 = const()[name = tensor<string, []>("op_42528_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42528_cast = slice_by_index(begin = var_42528_begin_0, end = var_42528_end_0, end_mask = var_42528_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42528_cast")];
            tensor<int32, [4]> var_42532_begin_0 = const()[name = tensor<string, []>("op_42532_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_42532_end_0 = const()[name = tensor<string, []>("op_42532_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_42532_end_mask_0 = const()[name = tensor<string, []>("op_42532_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42532_cast = slice_by_index(begin = var_42532_begin_0, end = var_42532_end_0, end_mask = var_42532_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42532_cast")];
            tensor<int32, [4]> var_42536_begin_0 = const()[name = tensor<string, []>("op_42536_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_42536_end_0 = const()[name = tensor<string, []>("op_42536_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_42536_end_mask_0 = const()[name = tensor<string, []>("op_42536_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42536_cast = slice_by_index(begin = var_42536_begin_0, end = var_42536_end_0, end_mask = var_42536_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42536_cast")];
            tensor<int32, [4]> var_42540_begin_0 = const()[name = tensor<string, []>("op_42540_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_42540_end_0 = const()[name = tensor<string, []>("op_42540_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_42540_end_mask_0 = const()[name = tensor<string, []>("op_42540_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42540_cast = slice_by_index(begin = var_42540_begin_0, end = var_42540_end_0, end_mask = var_42540_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42540_cast")];
            tensor<int32, [4]> var_42544_begin_0 = const()[name = tensor<string, []>("op_42544_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_42544_end_0 = const()[name = tensor<string, []>("op_42544_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_42544_end_mask_0 = const()[name = tensor<string, []>("op_42544_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42544_cast = slice_by_index(begin = var_42544_begin_0, end = var_42544_end_0, end_mask = var_42544_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42544_cast")];
            tensor<int32, [4]> var_42548_begin_0 = const()[name = tensor<string, []>("op_42548_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_42548_end_0 = const()[name = tensor<string, []>("op_42548_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_42548_end_mask_0 = const()[name = tensor<string, []>("op_42548_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42548_cast = slice_by_index(begin = var_42548_begin_0, end = var_42548_end_0, end_mask = var_42548_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42548_cast")];
            tensor<int32, [4]> var_42552_begin_0 = const()[name = tensor<string, []>("op_42552_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_42552_end_0 = const()[name = tensor<string, []>("op_42552_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_42552_end_mask_0 = const()[name = tensor<string, []>("op_42552_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42552_cast = slice_by_index(begin = var_42552_begin_0, end = var_42552_end_0, end_mask = var_42552_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42552_cast")];
            tensor<int32, [4]> var_42556_begin_0 = const()[name = tensor<string, []>("op_42556_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_42556_end_0 = const()[name = tensor<string, []>("op_42556_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_42556_end_mask_0 = const()[name = tensor<string, []>("op_42556_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42556_cast = slice_by_index(begin = var_42556_begin_0, end = var_42556_end_0, end_mask = var_42556_end_mask_0, x = q_189_cast)[name = tensor<string, []>("op_42556_cast")];
            tensor<int32, [4]> k_379_perm_0 = const()[name = tensor<string, []>("k_379_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_42563_begin_0 = const()[name = tensor<string, []>("op_42563_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_42563_end_0 = const()[name = tensor<string, []>("op_42563_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_42563_end_mask_0 = const()[name = tensor<string, []>("op_42563_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_45 = transpose(perm = k_379_perm_0, x = k_377_cast)[name = tensor<string, []>("transpose_45")];
            tensor<fp16, [2, 1024, 1, 64]> var_42563_cast = slice_by_index(begin = var_42563_begin_0, end = var_42563_end_0, end_mask = var_42563_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42563_cast")];
            tensor<int32, [4]> var_42567_begin_0 = const()[name = tensor<string, []>("op_42567_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_42567_end_0 = const()[name = tensor<string, []>("op_42567_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_42567_end_mask_0 = const()[name = tensor<string, []>("op_42567_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42567_cast = slice_by_index(begin = var_42567_begin_0, end = var_42567_end_0, end_mask = var_42567_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42567_cast")];
            tensor<int32, [4]> var_42571_begin_0 = const()[name = tensor<string, []>("op_42571_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_42571_end_0 = const()[name = tensor<string, []>("op_42571_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_42571_end_mask_0 = const()[name = tensor<string, []>("op_42571_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42571_cast = slice_by_index(begin = var_42571_begin_0, end = var_42571_end_0, end_mask = var_42571_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42571_cast")];
            tensor<int32, [4]> var_42575_begin_0 = const()[name = tensor<string, []>("op_42575_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_42575_end_0 = const()[name = tensor<string, []>("op_42575_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_42575_end_mask_0 = const()[name = tensor<string, []>("op_42575_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42575_cast = slice_by_index(begin = var_42575_begin_0, end = var_42575_end_0, end_mask = var_42575_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42575_cast")];
            tensor<int32, [4]> var_42579_begin_0 = const()[name = tensor<string, []>("op_42579_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_42579_end_0 = const()[name = tensor<string, []>("op_42579_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_42579_end_mask_0 = const()[name = tensor<string, []>("op_42579_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42579_cast = slice_by_index(begin = var_42579_begin_0, end = var_42579_end_0, end_mask = var_42579_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42579_cast")];
            tensor<int32, [4]> var_42583_begin_0 = const()[name = tensor<string, []>("op_42583_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_42583_end_0 = const()[name = tensor<string, []>("op_42583_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_42583_end_mask_0 = const()[name = tensor<string, []>("op_42583_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42583_cast = slice_by_index(begin = var_42583_begin_0, end = var_42583_end_0, end_mask = var_42583_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42583_cast")];
            tensor<int32, [4]> var_42587_begin_0 = const()[name = tensor<string, []>("op_42587_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_42587_end_0 = const()[name = tensor<string, []>("op_42587_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_42587_end_mask_0 = const()[name = tensor<string, []>("op_42587_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42587_cast = slice_by_index(begin = var_42587_begin_0, end = var_42587_end_0, end_mask = var_42587_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42587_cast")];
            tensor<int32, [4]> var_42591_begin_0 = const()[name = tensor<string, []>("op_42591_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_42591_end_0 = const()[name = tensor<string, []>("op_42591_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_42591_end_mask_0 = const()[name = tensor<string, []>("op_42591_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42591_cast = slice_by_index(begin = var_42591_begin_0, end = var_42591_end_0, end_mask = var_42591_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42591_cast")];
            tensor<int32, [4]> var_42595_begin_0 = const()[name = tensor<string, []>("op_42595_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_42595_end_0 = const()[name = tensor<string, []>("op_42595_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_42595_end_mask_0 = const()[name = tensor<string, []>("op_42595_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42595_cast = slice_by_index(begin = var_42595_begin_0, end = var_42595_end_0, end_mask = var_42595_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42595_cast")];
            tensor<int32, [4]> var_42599_begin_0 = const()[name = tensor<string, []>("op_42599_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_42599_end_0 = const()[name = tensor<string, []>("op_42599_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_42599_end_mask_0 = const()[name = tensor<string, []>("op_42599_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42599_cast = slice_by_index(begin = var_42599_begin_0, end = var_42599_end_0, end_mask = var_42599_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42599_cast")];
            tensor<int32, [4]> var_42603_begin_0 = const()[name = tensor<string, []>("op_42603_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_42603_end_0 = const()[name = tensor<string, []>("op_42603_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_42603_end_mask_0 = const()[name = tensor<string, []>("op_42603_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42603_cast = slice_by_index(begin = var_42603_begin_0, end = var_42603_end_0, end_mask = var_42603_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42603_cast")];
            tensor<int32, [4]> var_42607_begin_0 = const()[name = tensor<string, []>("op_42607_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_42607_end_0 = const()[name = tensor<string, []>("op_42607_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_42607_end_mask_0 = const()[name = tensor<string, []>("op_42607_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42607_cast = slice_by_index(begin = var_42607_begin_0, end = var_42607_end_0, end_mask = var_42607_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42607_cast")];
            tensor<int32, [4]> var_42611_begin_0 = const()[name = tensor<string, []>("op_42611_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_42611_end_0 = const()[name = tensor<string, []>("op_42611_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_42611_end_mask_0 = const()[name = tensor<string, []>("op_42611_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42611_cast = slice_by_index(begin = var_42611_begin_0, end = var_42611_end_0, end_mask = var_42611_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42611_cast")];
            tensor<int32, [4]> var_42615_begin_0 = const()[name = tensor<string, []>("op_42615_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_42615_end_0 = const()[name = tensor<string, []>("op_42615_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_42615_end_mask_0 = const()[name = tensor<string, []>("op_42615_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42615_cast = slice_by_index(begin = var_42615_begin_0, end = var_42615_end_0, end_mask = var_42615_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42615_cast")];
            tensor<int32, [4]> var_42619_begin_0 = const()[name = tensor<string, []>("op_42619_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_42619_end_0 = const()[name = tensor<string, []>("op_42619_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_42619_end_mask_0 = const()[name = tensor<string, []>("op_42619_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42619_cast = slice_by_index(begin = var_42619_begin_0, end = var_42619_end_0, end_mask = var_42619_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42619_cast")];
            tensor<int32, [4]> var_42623_begin_0 = const()[name = tensor<string, []>("op_42623_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_42623_end_0 = const()[name = tensor<string, []>("op_42623_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_42623_end_mask_0 = const()[name = tensor<string, []>("op_42623_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42623_cast = slice_by_index(begin = var_42623_begin_0, end = var_42623_end_0, end_mask = var_42623_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42623_cast")];
            tensor<int32, [4]> var_42627_begin_0 = const()[name = tensor<string, []>("op_42627_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_42627_end_0 = const()[name = tensor<string, []>("op_42627_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_42627_end_mask_0 = const()[name = tensor<string, []>("op_42627_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42627_cast = slice_by_index(begin = var_42627_begin_0, end = var_42627_end_0, end_mask = var_42627_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42627_cast")];
            tensor<int32, [4]> var_42631_begin_0 = const()[name = tensor<string, []>("op_42631_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_42631_end_0 = const()[name = tensor<string, []>("op_42631_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_42631_end_mask_0 = const()[name = tensor<string, []>("op_42631_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42631_cast = slice_by_index(begin = var_42631_begin_0, end = var_42631_end_0, end_mask = var_42631_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42631_cast")];
            tensor<int32, [4]> var_42635_begin_0 = const()[name = tensor<string, []>("op_42635_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_42635_end_0 = const()[name = tensor<string, []>("op_42635_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_42635_end_mask_0 = const()[name = tensor<string, []>("op_42635_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42635_cast = slice_by_index(begin = var_42635_begin_0, end = var_42635_end_0, end_mask = var_42635_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42635_cast")];
            tensor<int32, [4]> var_42639_begin_0 = const()[name = tensor<string, []>("op_42639_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_42639_end_0 = const()[name = tensor<string, []>("op_42639_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_42639_end_mask_0 = const()[name = tensor<string, []>("op_42639_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_42639_cast = slice_by_index(begin = var_42639_begin_0, end = var_42639_end_0, end_mask = var_42639_end_mask_0, x = transpose_45)[name = tensor<string, []>("op_42639_cast")];
            tensor<int32, [4]> var_42641_begin_0 = const()[name = tensor<string, []>("op_42641_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_42641_end_0 = const()[name = tensor<string, []>("op_42641_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_42641_end_mask_0 = const()[name = tensor<string, []>("op_42641_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42641_cast = slice_by_index(begin = var_42641_begin_0, end = var_42641_end_0, end_mask = var_42641_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42641_cast")];
            tensor<int32, [4]> var_42645_begin_0 = const()[name = tensor<string, []>("op_42645_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_42645_end_0 = const()[name = tensor<string, []>("op_42645_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_42645_end_mask_0 = const()[name = tensor<string, []>("op_42645_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42645_cast = slice_by_index(begin = var_42645_begin_0, end = var_42645_end_0, end_mask = var_42645_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42645_cast")];
            tensor<int32, [4]> var_42649_begin_0 = const()[name = tensor<string, []>("op_42649_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_42649_end_0 = const()[name = tensor<string, []>("op_42649_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_42649_end_mask_0 = const()[name = tensor<string, []>("op_42649_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42649_cast = slice_by_index(begin = var_42649_begin_0, end = var_42649_end_0, end_mask = var_42649_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42649_cast")];
            tensor<int32, [4]> var_42653_begin_0 = const()[name = tensor<string, []>("op_42653_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_42653_end_0 = const()[name = tensor<string, []>("op_42653_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_42653_end_mask_0 = const()[name = tensor<string, []>("op_42653_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42653_cast = slice_by_index(begin = var_42653_begin_0, end = var_42653_end_0, end_mask = var_42653_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42653_cast")];
            tensor<int32, [4]> var_42657_begin_0 = const()[name = tensor<string, []>("op_42657_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_42657_end_0 = const()[name = tensor<string, []>("op_42657_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_42657_end_mask_0 = const()[name = tensor<string, []>("op_42657_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42657_cast = slice_by_index(begin = var_42657_begin_0, end = var_42657_end_0, end_mask = var_42657_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42657_cast")];
            tensor<int32, [4]> var_42661_begin_0 = const()[name = tensor<string, []>("op_42661_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_42661_end_0 = const()[name = tensor<string, []>("op_42661_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_42661_end_mask_0 = const()[name = tensor<string, []>("op_42661_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42661_cast = slice_by_index(begin = var_42661_begin_0, end = var_42661_end_0, end_mask = var_42661_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42661_cast")];
            tensor<int32, [4]> var_42665_begin_0 = const()[name = tensor<string, []>("op_42665_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_42665_end_0 = const()[name = tensor<string, []>("op_42665_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_42665_end_mask_0 = const()[name = tensor<string, []>("op_42665_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42665_cast = slice_by_index(begin = var_42665_begin_0, end = var_42665_end_0, end_mask = var_42665_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42665_cast")];
            tensor<int32, [4]> var_42669_begin_0 = const()[name = tensor<string, []>("op_42669_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_42669_end_0 = const()[name = tensor<string, []>("op_42669_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_42669_end_mask_0 = const()[name = tensor<string, []>("op_42669_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42669_cast = slice_by_index(begin = var_42669_begin_0, end = var_42669_end_0, end_mask = var_42669_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42669_cast")];
            tensor<int32, [4]> var_42673_begin_0 = const()[name = tensor<string, []>("op_42673_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_42673_end_0 = const()[name = tensor<string, []>("op_42673_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_42673_end_mask_0 = const()[name = tensor<string, []>("op_42673_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42673_cast = slice_by_index(begin = var_42673_begin_0, end = var_42673_end_0, end_mask = var_42673_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42673_cast")];
            tensor<int32, [4]> var_42677_begin_0 = const()[name = tensor<string, []>("op_42677_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_42677_end_0 = const()[name = tensor<string, []>("op_42677_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_42677_end_mask_0 = const()[name = tensor<string, []>("op_42677_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42677_cast = slice_by_index(begin = var_42677_begin_0, end = var_42677_end_0, end_mask = var_42677_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42677_cast")];
            tensor<int32, [4]> var_42681_begin_0 = const()[name = tensor<string, []>("op_42681_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_42681_end_0 = const()[name = tensor<string, []>("op_42681_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_42681_end_mask_0 = const()[name = tensor<string, []>("op_42681_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42681_cast = slice_by_index(begin = var_42681_begin_0, end = var_42681_end_0, end_mask = var_42681_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42681_cast")];
            tensor<int32, [4]> var_42685_begin_0 = const()[name = tensor<string, []>("op_42685_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_42685_end_0 = const()[name = tensor<string, []>("op_42685_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_42685_end_mask_0 = const()[name = tensor<string, []>("op_42685_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42685_cast = slice_by_index(begin = var_42685_begin_0, end = var_42685_end_0, end_mask = var_42685_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42685_cast")];
            tensor<int32, [4]> var_42689_begin_0 = const()[name = tensor<string, []>("op_42689_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_42689_end_0 = const()[name = tensor<string, []>("op_42689_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_42689_end_mask_0 = const()[name = tensor<string, []>("op_42689_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42689_cast = slice_by_index(begin = var_42689_begin_0, end = var_42689_end_0, end_mask = var_42689_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42689_cast")];
            tensor<int32, [4]> var_42693_begin_0 = const()[name = tensor<string, []>("op_42693_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_42693_end_0 = const()[name = tensor<string, []>("op_42693_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_42693_end_mask_0 = const()[name = tensor<string, []>("op_42693_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42693_cast = slice_by_index(begin = var_42693_begin_0, end = var_42693_end_0, end_mask = var_42693_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42693_cast")];
            tensor<int32, [4]> var_42697_begin_0 = const()[name = tensor<string, []>("op_42697_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_42697_end_0 = const()[name = tensor<string, []>("op_42697_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_42697_end_mask_0 = const()[name = tensor<string, []>("op_42697_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42697_cast = slice_by_index(begin = var_42697_begin_0, end = var_42697_end_0, end_mask = var_42697_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42697_cast")];
            tensor<int32, [4]> var_42701_begin_0 = const()[name = tensor<string, []>("op_42701_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_42701_end_0 = const()[name = tensor<string, []>("op_42701_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_42701_end_mask_0 = const()[name = tensor<string, []>("op_42701_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42701_cast = slice_by_index(begin = var_42701_begin_0, end = var_42701_end_0, end_mask = var_42701_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42701_cast")];
            tensor<int32, [4]> var_42705_begin_0 = const()[name = tensor<string, []>("op_42705_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_42705_end_0 = const()[name = tensor<string, []>("op_42705_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_42705_end_mask_0 = const()[name = tensor<string, []>("op_42705_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42705_cast = slice_by_index(begin = var_42705_begin_0, end = var_42705_end_0, end_mask = var_42705_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42705_cast")];
            tensor<int32, [4]> var_42709_begin_0 = const()[name = tensor<string, []>("op_42709_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_42709_end_0 = const()[name = tensor<string, []>("op_42709_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_42709_end_mask_0 = const()[name = tensor<string, []>("op_42709_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42709_cast = slice_by_index(begin = var_42709_begin_0, end = var_42709_end_0, end_mask = var_42709_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42709_cast")];
            tensor<int32, [4]> var_42713_begin_0 = const()[name = tensor<string, []>("op_42713_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_42713_end_0 = const()[name = tensor<string, []>("op_42713_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_42713_end_mask_0 = const()[name = tensor<string, []>("op_42713_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42713_cast = slice_by_index(begin = var_42713_begin_0, end = var_42713_end_0, end_mask = var_42713_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42713_cast")];
            tensor<int32, [4]> var_42717_begin_0 = const()[name = tensor<string, []>("op_42717_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_42717_end_0 = const()[name = tensor<string, []>("op_42717_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_42717_end_mask_0 = const()[name = tensor<string, []>("op_42717_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42717_cast = slice_by_index(begin = var_42717_begin_0, end = var_42717_end_0, end_mask = var_42717_end_mask_0, x = v_189_cast)[name = tensor<string, []>("op_42717_cast")];
            tensor<string, []> var_42721_equation_0 = const()[name = tensor<string, []>("op_42721_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42721_cast = einsum(equation = var_42721_equation_0, values = (var_42563_cast, var_42480_cast))[name = tensor<string, []>("op_42721_cast")];
            tensor<fp16, []> var_42722_to_fp16 = const()[name = tensor<string, []>("op_42722_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3601_cast = mul(x = var_42721_cast, y = var_42722_to_fp16)[name = tensor<string, []>("aw_3601_cast")];
            tensor<string, []> var_42725_equation_0 = const()[name = tensor<string, []>("op_42725_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42725_cast = einsum(equation = var_42725_equation_0, values = (var_42567_cast, var_42484_cast))[name = tensor<string, []>("op_42725_cast")];
            tensor<fp16, []> var_42726_to_fp16 = const()[name = tensor<string, []>("op_42726_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3603_cast = mul(x = var_42725_cast, y = var_42726_to_fp16)[name = tensor<string, []>("aw_3603_cast")];
            tensor<string, []> var_42729_equation_0 = const()[name = tensor<string, []>("op_42729_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42729_cast = einsum(equation = var_42729_equation_0, values = (var_42571_cast, var_42488_cast))[name = tensor<string, []>("op_42729_cast")];
            tensor<fp16, []> var_42730_to_fp16 = const()[name = tensor<string, []>("op_42730_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3605_cast = mul(x = var_42729_cast, y = var_42730_to_fp16)[name = tensor<string, []>("aw_3605_cast")];
            tensor<string, []> var_42733_equation_0 = const()[name = tensor<string, []>("op_42733_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42733_cast = einsum(equation = var_42733_equation_0, values = (var_42575_cast, var_42492_cast))[name = tensor<string, []>("op_42733_cast")];
            tensor<fp16, []> var_42734_to_fp16 = const()[name = tensor<string, []>("op_42734_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3607_cast = mul(x = var_42733_cast, y = var_42734_to_fp16)[name = tensor<string, []>("aw_3607_cast")];
            tensor<string, []> var_42737_equation_0 = const()[name = tensor<string, []>("op_42737_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42737_cast = einsum(equation = var_42737_equation_0, values = (var_42579_cast, var_42496_cast))[name = tensor<string, []>("op_42737_cast")];
            tensor<fp16, []> var_42738_to_fp16 = const()[name = tensor<string, []>("op_42738_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3609_cast = mul(x = var_42737_cast, y = var_42738_to_fp16)[name = tensor<string, []>("aw_3609_cast")];
            tensor<string, []> var_42741_equation_0 = const()[name = tensor<string, []>("op_42741_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42741_cast = einsum(equation = var_42741_equation_0, values = (var_42583_cast, var_42500_cast))[name = tensor<string, []>("op_42741_cast")];
            tensor<fp16, []> var_42742_to_fp16 = const()[name = tensor<string, []>("op_42742_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3611_cast = mul(x = var_42741_cast, y = var_42742_to_fp16)[name = tensor<string, []>("aw_3611_cast")];
            tensor<string, []> var_42745_equation_0 = const()[name = tensor<string, []>("op_42745_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42745_cast = einsum(equation = var_42745_equation_0, values = (var_42587_cast, var_42504_cast))[name = tensor<string, []>("op_42745_cast")];
            tensor<fp16, []> var_42746_to_fp16 = const()[name = tensor<string, []>("op_42746_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3613_cast = mul(x = var_42745_cast, y = var_42746_to_fp16)[name = tensor<string, []>("aw_3613_cast")];
            tensor<string, []> var_42749_equation_0 = const()[name = tensor<string, []>("op_42749_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42749_cast = einsum(equation = var_42749_equation_0, values = (var_42591_cast, var_42508_cast))[name = tensor<string, []>("op_42749_cast")];
            tensor<fp16, []> var_42750_to_fp16 = const()[name = tensor<string, []>("op_42750_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3615_cast = mul(x = var_42749_cast, y = var_42750_to_fp16)[name = tensor<string, []>("aw_3615_cast")];
            tensor<string, []> var_42753_equation_0 = const()[name = tensor<string, []>("op_42753_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42753_cast = einsum(equation = var_42753_equation_0, values = (var_42595_cast, var_42512_cast))[name = tensor<string, []>("op_42753_cast")];
            tensor<fp16, []> var_42754_to_fp16 = const()[name = tensor<string, []>("op_42754_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3617_cast = mul(x = var_42753_cast, y = var_42754_to_fp16)[name = tensor<string, []>("aw_3617_cast")];
            tensor<string, []> var_42757_equation_0 = const()[name = tensor<string, []>("op_42757_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42757_cast = einsum(equation = var_42757_equation_0, values = (var_42599_cast, var_42516_cast))[name = tensor<string, []>("op_42757_cast")];
            tensor<fp16, []> var_42758_to_fp16 = const()[name = tensor<string, []>("op_42758_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3619_cast = mul(x = var_42757_cast, y = var_42758_to_fp16)[name = tensor<string, []>("aw_3619_cast")];
            tensor<string, []> var_42761_equation_0 = const()[name = tensor<string, []>("op_42761_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42761_cast = einsum(equation = var_42761_equation_0, values = (var_42603_cast, var_42520_cast))[name = tensor<string, []>("op_42761_cast")];
            tensor<fp16, []> var_42762_to_fp16 = const()[name = tensor<string, []>("op_42762_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3621_cast = mul(x = var_42761_cast, y = var_42762_to_fp16)[name = tensor<string, []>("aw_3621_cast")];
            tensor<string, []> var_42765_equation_0 = const()[name = tensor<string, []>("op_42765_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42765_cast = einsum(equation = var_42765_equation_0, values = (var_42607_cast, var_42524_cast))[name = tensor<string, []>("op_42765_cast")];
            tensor<fp16, []> var_42766_to_fp16 = const()[name = tensor<string, []>("op_42766_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3623_cast = mul(x = var_42765_cast, y = var_42766_to_fp16)[name = tensor<string, []>("aw_3623_cast")];
            tensor<string, []> var_42769_equation_0 = const()[name = tensor<string, []>("op_42769_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42769_cast = einsum(equation = var_42769_equation_0, values = (var_42611_cast, var_42528_cast))[name = tensor<string, []>("op_42769_cast")];
            tensor<fp16, []> var_42770_to_fp16 = const()[name = tensor<string, []>("op_42770_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3625_cast = mul(x = var_42769_cast, y = var_42770_to_fp16)[name = tensor<string, []>("aw_3625_cast")];
            tensor<string, []> var_42773_equation_0 = const()[name = tensor<string, []>("op_42773_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42773_cast = einsum(equation = var_42773_equation_0, values = (var_42615_cast, var_42532_cast))[name = tensor<string, []>("op_42773_cast")];
            tensor<fp16, []> var_42774_to_fp16 = const()[name = tensor<string, []>("op_42774_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3627_cast = mul(x = var_42773_cast, y = var_42774_to_fp16)[name = tensor<string, []>("aw_3627_cast")];
            tensor<string, []> var_42777_equation_0 = const()[name = tensor<string, []>("op_42777_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42777_cast = einsum(equation = var_42777_equation_0, values = (var_42619_cast, var_42536_cast))[name = tensor<string, []>("op_42777_cast")];
            tensor<fp16, []> var_42778_to_fp16 = const()[name = tensor<string, []>("op_42778_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3629_cast = mul(x = var_42777_cast, y = var_42778_to_fp16)[name = tensor<string, []>("aw_3629_cast")];
            tensor<string, []> var_42781_equation_0 = const()[name = tensor<string, []>("op_42781_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42781_cast = einsum(equation = var_42781_equation_0, values = (var_42623_cast, var_42540_cast))[name = tensor<string, []>("op_42781_cast")];
            tensor<fp16, []> var_42782_to_fp16 = const()[name = tensor<string, []>("op_42782_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3631_cast = mul(x = var_42781_cast, y = var_42782_to_fp16)[name = tensor<string, []>("aw_3631_cast")];
            tensor<string, []> var_42785_equation_0 = const()[name = tensor<string, []>("op_42785_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42785_cast = einsum(equation = var_42785_equation_0, values = (var_42627_cast, var_42544_cast))[name = tensor<string, []>("op_42785_cast")];
            tensor<fp16, []> var_42786_to_fp16 = const()[name = tensor<string, []>("op_42786_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3633_cast = mul(x = var_42785_cast, y = var_42786_to_fp16)[name = tensor<string, []>("aw_3633_cast")];
            tensor<string, []> var_42789_equation_0 = const()[name = tensor<string, []>("op_42789_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42789_cast = einsum(equation = var_42789_equation_0, values = (var_42631_cast, var_42548_cast))[name = tensor<string, []>("op_42789_cast")];
            tensor<fp16, []> var_42790_to_fp16 = const()[name = tensor<string, []>("op_42790_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3635_cast = mul(x = var_42789_cast, y = var_42790_to_fp16)[name = tensor<string, []>("aw_3635_cast")];
            tensor<string, []> var_42793_equation_0 = const()[name = tensor<string, []>("op_42793_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42793_cast = einsum(equation = var_42793_equation_0, values = (var_42635_cast, var_42552_cast))[name = tensor<string, []>("op_42793_cast")];
            tensor<fp16, []> var_42794_to_fp16 = const()[name = tensor<string, []>("op_42794_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3637_cast = mul(x = var_42793_cast, y = var_42794_to_fp16)[name = tensor<string, []>("aw_3637_cast")];
            tensor<string, []> var_42797_equation_0 = const()[name = tensor<string, []>("op_42797_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42797_cast = einsum(equation = var_42797_equation_0, values = (var_42639_cast, var_42556_cast))[name = tensor<string, []>("op_42797_cast")];
            tensor<fp16, []> var_42798_to_fp16 = const()[name = tensor<string, []>("op_42798_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3639_cast = mul(x = var_42797_cast, y = var_42798_to_fp16)[name = tensor<string, []>("aw_3639_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42800_cast = softmax(axis = var_30385, x = aw_3601_cast)[name = tensor<string, []>("op_42800_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42801_cast = softmax(axis = var_30385, x = aw_3603_cast)[name = tensor<string, []>("op_42801_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42802_cast = softmax(axis = var_30385, x = aw_3605_cast)[name = tensor<string, []>("op_42802_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42803_cast = softmax(axis = var_30385, x = aw_3607_cast)[name = tensor<string, []>("op_42803_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42804_cast = softmax(axis = var_30385, x = aw_3609_cast)[name = tensor<string, []>("op_42804_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42805_cast = softmax(axis = var_30385, x = aw_3611_cast)[name = tensor<string, []>("op_42805_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42806_cast = softmax(axis = var_30385, x = aw_3613_cast)[name = tensor<string, []>("op_42806_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42807_cast = softmax(axis = var_30385, x = aw_3615_cast)[name = tensor<string, []>("op_42807_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42808_cast = softmax(axis = var_30385, x = aw_3617_cast)[name = tensor<string, []>("op_42808_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42809_cast = softmax(axis = var_30385, x = aw_3619_cast)[name = tensor<string, []>("op_42809_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42810_cast = softmax(axis = var_30385, x = aw_3621_cast)[name = tensor<string, []>("op_42810_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42811_cast = softmax(axis = var_30385, x = aw_3623_cast)[name = tensor<string, []>("op_42811_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42812_cast = softmax(axis = var_30385, x = aw_3625_cast)[name = tensor<string, []>("op_42812_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42813_cast = softmax(axis = var_30385, x = aw_3627_cast)[name = tensor<string, []>("op_42813_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42814_cast = softmax(axis = var_30385, x = aw_3629_cast)[name = tensor<string, []>("op_42814_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42815_cast = softmax(axis = var_30385, x = aw_3631_cast)[name = tensor<string, []>("op_42815_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42816_cast = softmax(axis = var_30385, x = aw_3633_cast)[name = tensor<string, []>("op_42816_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42817_cast = softmax(axis = var_30385, x = aw_3635_cast)[name = tensor<string, []>("op_42817_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42818_cast = softmax(axis = var_30385, x = aw_3637_cast)[name = tensor<string, []>("op_42818_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_42819_cast = softmax(axis = var_30385, x = aw_3639_cast)[name = tensor<string, []>("op_42819_cast")];
            tensor<string, []> var_42821_equation_0 = const()[name = tensor<string, []>("op_42821_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42821_cast = einsum(equation = var_42821_equation_0, values = (var_42641_cast, var_42800_cast))[name = tensor<string, []>("op_42821_cast")];
            tensor<string, []> var_42823_equation_0 = const()[name = tensor<string, []>("op_42823_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42823_cast = einsum(equation = var_42823_equation_0, values = (var_42645_cast, var_42801_cast))[name = tensor<string, []>("op_42823_cast")];
            tensor<string, []> var_42825_equation_0 = const()[name = tensor<string, []>("op_42825_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42825_cast = einsum(equation = var_42825_equation_0, values = (var_42649_cast, var_42802_cast))[name = tensor<string, []>("op_42825_cast")];
            tensor<string, []> var_42827_equation_0 = const()[name = tensor<string, []>("op_42827_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42827_cast = einsum(equation = var_42827_equation_0, values = (var_42653_cast, var_42803_cast))[name = tensor<string, []>("op_42827_cast")];
            tensor<string, []> var_42829_equation_0 = const()[name = tensor<string, []>("op_42829_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42829_cast = einsum(equation = var_42829_equation_0, values = (var_42657_cast, var_42804_cast))[name = tensor<string, []>("op_42829_cast")];
            tensor<string, []> var_42831_equation_0 = const()[name = tensor<string, []>("op_42831_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42831_cast = einsum(equation = var_42831_equation_0, values = (var_42661_cast, var_42805_cast))[name = tensor<string, []>("op_42831_cast")];
            tensor<string, []> var_42833_equation_0 = const()[name = tensor<string, []>("op_42833_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42833_cast = einsum(equation = var_42833_equation_0, values = (var_42665_cast, var_42806_cast))[name = tensor<string, []>("op_42833_cast")];
            tensor<string, []> var_42835_equation_0 = const()[name = tensor<string, []>("op_42835_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42835_cast = einsum(equation = var_42835_equation_0, values = (var_42669_cast, var_42807_cast))[name = tensor<string, []>("op_42835_cast")];
            tensor<string, []> var_42837_equation_0 = const()[name = tensor<string, []>("op_42837_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42837_cast = einsum(equation = var_42837_equation_0, values = (var_42673_cast, var_42808_cast))[name = tensor<string, []>("op_42837_cast")];
            tensor<string, []> var_42839_equation_0 = const()[name = tensor<string, []>("op_42839_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42839_cast = einsum(equation = var_42839_equation_0, values = (var_42677_cast, var_42809_cast))[name = tensor<string, []>("op_42839_cast")];
            tensor<string, []> var_42841_equation_0 = const()[name = tensor<string, []>("op_42841_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42841_cast = einsum(equation = var_42841_equation_0, values = (var_42681_cast, var_42810_cast))[name = tensor<string, []>("op_42841_cast")];
            tensor<string, []> var_42843_equation_0 = const()[name = tensor<string, []>("op_42843_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42843_cast = einsum(equation = var_42843_equation_0, values = (var_42685_cast, var_42811_cast))[name = tensor<string, []>("op_42843_cast")];
            tensor<string, []> var_42845_equation_0 = const()[name = tensor<string, []>("op_42845_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42845_cast = einsum(equation = var_42845_equation_0, values = (var_42689_cast, var_42812_cast))[name = tensor<string, []>("op_42845_cast")];
            tensor<string, []> var_42847_equation_0 = const()[name = tensor<string, []>("op_42847_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42847_cast = einsum(equation = var_42847_equation_0, values = (var_42693_cast, var_42813_cast))[name = tensor<string, []>("op_42847_cast")];
            tensor<string, []> var_42849_equation_0 = const()[name = tensor<string, []>("op_42849_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42849_cast = einsum(equation = var_42849_equation_0, values = (var_42697_cast, var_42814_cast))[name = tensor<string, []>("op_42849_cast")];
            tensor<string, []> var_42851_equation_0 = const()[name = tensor<string, []>("op_42851_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42851_cast = einsum(equation = var_42851_equation_0, values = (var_42701_cast, var_42815_cast))[name = tensor<string, []>("op_42851_cast")];
            tensor<string, []> var_42853_equation_0 = const()[name = tensor<string, []>("op_42853_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42853_cast = einsum(equation = var_42853_equation_0, values = (var_42705_cast, var_42816_cast))[name = tensor<string, []>("op_42853_cast")];
            tensor<string, []> var_42855_equation_0 = const()[name = tensor<string, []>("op_42855_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42855_cast = einsum(equation = var_42855_equation_0, values = (var_42709_cast, var_42817_cast))[name = tensor<string, []>("op_42855_cast")];
            tensor<string, []> var_42857_equation_0 = const()[name = tensor<string, []>("op_42857_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42857_cast = einsum(equation = var_42857_equation_0, values = (var_42713_cast, var_42818_cast))[name = tensor<string, []>("op_42857_cast")];
            tensor<string, []> var_42859_equation_0 = const()[name = tensor<string, []>("op_42859_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_42859_cast = einsum(equation = var_42859_equation_0, values = (var_42717_cast, var_42819_cast))[name = tensor<string, []>("op_42859_cast")];
            tensor<bool, []> input_563_interleave_0 = const()[name = tensor<string, []>("input_563_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_563_cast = concat(axis = var_30385, interleave = input_563_interleave_0, values = (var_42821_cast, var_42823_cast, var_42825_cast, var_42827_cast, var_42829_cast, var_42831_cast, var_42833_cast, var_42835_cast, var_42837_cast, var_42839_cast, var_42841_cast, var_42843_cast, var_42845_cast, var_42847_cast, var_42849_cast, var_42851_cast, var_42853_cast, var_42855_cast, var_42857_cast, var_42859_cast))[name = tensor<string, []>("input_563_cast")];
            tensor<int32, [2]> var_42865 = const()[name = tensor<string, []>("op_42865"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_42867 = const()[name = tensor<string, []>("op_42867"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_42869_pad_type_0 = const()[name = tensor<string, []>("op_42869_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_42869_pad_0 = const()[name = tensor<string, []>("op_42869_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3606820288)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3610097152)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_42869_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_out_0_bias_to_fp16, dilations = var_42867, groups = var_30385, pad = var_42869_pad_0, pad_type = var_42869_pad_type_0, strides = var_42865, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn1_to_out_0_weight_to_fp16, x = input_563_cast)[name = tensor<string, []>("op_42869_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_285_cast = add(x = var_42869_cast, y = inputs_283_cast)[name = tensor<string, []>("inputs_285_cast")];
            tensor<int32, [1]> var_42873 = const()[name = tensor<string, []>("op_42873"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_285_cast = reduce_mean(axes = var_42873, keep_dims = var_30380, x = inputs_285_cast)[name = tensor<string, []>("channels_mean_285_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_285_cast = sub(x = inputs_285_cast, y = channels_mean_285_cast)[name = tensor<string, []>("zero_mean_285_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_285_cast = mul(x = zero_mean_285_cast, y = zero_mean_285_cast)[name = tensor<string, []>("zero_mean_sq_285_cast")];
            tensor<int32, [1]> var_42877 = const()[name = tensor<string, []>("op_42877"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_42878_cast = reduce_mean(axes = var_42877, keep_dims = var_30380, x = zero_mean_sq_285_cast)[name = tensor<string, []>("op_42878_cast")];
            tensor<fp16, []> var_42879_to_fp16 = const()[name = tensor<string, []>("op_42879_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_42880_cast = add(x = var_42878_cast, y = var_42879_to_fp16)[name = tensor<string, []>("op_42880_cast")];
            tensor<fp16, []> denom_285_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_285_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_285_cast = rsqrt(epsilon = denom_285_epsilon_0_to_fp16, x = var_42880_cast)[name = tensor<string, []>("denom_285_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_285_cast = mul(x = zero_mean_285_cast, y = denom_285_cast)[name = tensor<string, []>("out_285_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_42884_to_fp16 = const()[name = tensor<string, []>("op_42884_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3610099776)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_42885_cast = add(x = out_285_cast, y = var_42884_to_fp16)[name = tensor<string, []>("op_42885_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_42887_to_fp16 = const()[name = tensor<string, []>("op_42887_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3610102400)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_381_cast = mul(x = var_42885_cast, y = var_42887_to_fp16)[name = tensor<string, []>("hidden_states_381_cast")];
            tensor<int32, [2]> var_42894 = const()[name = tensor<string, []>("op_42894"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_42896 = const()[name = tensor<string, []>("op_42896"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_191_pad_type_0 = const()[name = tensor<string, []>("q_191_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_191_pad_0 = const()[name = tensor<string, []>("q_191_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3610105024)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_191_cast = conv(dilations = var_42896, groups = var_30385, pad = q_191_pad_0, pad_type = q_191_pad_type_0, strides = var_42894, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_q_weight_to_fp16, x = hidden_states_381_cast)[name = tensor<string, []>("q_191_cast")];
            tensor<int32, [2]> var_42900 = const()[name = tensor<string, []>("op_42900"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_42902 = const()[name = tensor<string, []>("op_42902"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_381_pad_type_0 = const()[name = tensor<string, []>("k_381_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_381_pad_0 = const()[name = tensor<string, []>("k_381_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3613381888)))];
            tensor<fp16, [2, 1280, 1, 77]> k_381_cast = conv(dilations = var_42902, groups = var_30385, pad = k_381_pad_0, pad_type = k_381_pad_type_0, strides = var_42900, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_381_cast")];
            tensor<int32, [2]> var_42906 = const()[name = tensor<string, []>("op_42906"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_42908 = const()[name = tensor<string, []>("op_42908"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_191_pad_type_0 = const()[name = tensor<string, []>("v_191_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_191_pad_0 = const()[name = tensor<string, []>("v_191_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3618624832)))];
            tensor<fp16, [2, 1280, 1, 77]> v_191_cast = conv(dilations = var_42908, groups = var_30385, pad = v_191_pad_0, pad_type = v_191_pad_type_0, strides = var_42906, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_191_cast")];
            tensor<int32, [4]> var_42912_begin_0 = const()[name = tensor<string, []>("op_42912_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_42912_end_0 = const()[name = tensor<string, []>("op_42912_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_42912_end_mask_0 = const()[name = tensor<string, []>("op_42912_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42912_cast = slice_by_index(begin = var_42912_begin_0, end = var_42912_end_0, end_mask = var_42912_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42912_cast")];
            tensor<int32, [4]> var_42916_begin_0 = const()[name = tensor<string, []>("op_42916_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_42916_end_0 = const()[name = tensor<string, []>("op_42916_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_42916_end_mask_0 = const()[name = tensor<string, []>("op_42916_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42916_cast = slice_by_index(begin = var_42916_begin_0, end = var_42916_end_0, end_mask = var_42916_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42916_cast")];
            tensor<int32, [4]> var_42920_begin_0 = const()[name = tensor<string, []>("op_42920_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_42920_end_0 = const()[name = tensor<string, []>("op_42920_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_42920_end_mask_0 = const()[name = tensor<string, []>("op_42920_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42920_cast = slice_by_index(begin = var_42920_begin_0, end = var_42920_end_0, end_mask = var_42920_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42920_cast")];
            tensor<int32, [4]> var_42924_begin_0 = const()[name = tensor<string, []>("op_42924_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_42924_end_0 = const()[name = tensor<string, []>("op_42924_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_42924_end_mask_0 = const()[name = tensor<string, []>("op_42924_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42924_cast = slice_by_index(begin = var_42924_begin_0, end = var_42924_end_0, end_mask = var_42924_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42924_cast")];
            tensor<int32, [4]> var_42928_begin_0 = const()[name = tensor<string, []>("op_42928_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_42928_end_0 = const()[name = tensor<string, []>("op_42928_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_42928_end_mask_0 = const()[name = tensor<string, []>("op_42928_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42928_cast = slice_by_index(begin = var_42928_begin_0, end = var_42928_end_0, end_mask = var_42928_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42928_cast")];
            tensor<int32, [4]> var_42932_begin_0 = const()[name = tensor<string, []>("op_42932_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_42932_end_0 = const()[name = tensor<string, []>("op_42932_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_42932_end_mask_0 = const()[name = tensor<string, []>("op_42932_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42932_cast = slice_by_index(begin = var_42932_begin_0, end = var_42932_end_0, end_mask = var_42932_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42932_cast")];
            tensor<int32, [4]> var_42936_begin_0 = const()[name = tensor<string, []>("op_42936_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_42936_end_0 = const()[name = tensor<string, []>("op_42936_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_42936_end_mask_0 = const()[name = tensor<string, []>("op_42936_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42936_cast = slice_by_index(begin = var_42936_begin_0, end = var_42936_end_0, end_mask = var_42936_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42936_cast")];
            tensor<int32, [4]> var_42940_begin_0 = const()[name = tensor<string, []>("op_42940_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_42940_end_0 = const()[name = tensor<string, []>("op_42940_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_42940_end_mask_0 = const()[name = tensor<string, []>("op_42940_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42940_cast = slice_by_index(begin = var_42940_begin_0, end = var_42940_end_0, end_mask = var_42940_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42940_cast")];
            tensor<int32, [4]> var_42944_begin_0 = const()[name = tensor<string, []>("op_42944_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_42944_end_0 = const()[name = tensor<string, []>("op_42944_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_42944_end_mask_0 = const()[name = tensor<string, []>("op_42944_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42944_cast = slice_by_index(begin = var_42944_begin_0, end = var_42944_end_0, end_mask = var_42944_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42944_cast")];
            tensor<int32, [4]> var_42948_begin_0 = const()[name = tensor<string, []>("op_42948_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_42948_end_0 = const()[name = tensor<string, []>("op_42948_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_42948_end_mask_0 = const()[name = tensor<string, []>("op_42948_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42948_cast = slice_by_index(begin = var_42948_begin_0, end = var_42948_end_0, end_mask = var_42948_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42948_cast")];
            tensor<int32, [4]> var_42952_begin_0 = const()[name = tensor<string, []>("op_42952_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_42952_end_0 = const()[name = tensor<string, []>("op_42952_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_42952_end_mask_0 = const()[name = tensor<string, []>("op_42952_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42952_cast = slice_by_index(begin = var_42952_begin_0, end = var_42952_end_0, end_mask = var_42952_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42952_cast")];
            tensor<int32, [4]> var_42956_begin_0 = const()[name = tensor<string, []>("op_42956_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_42956_end_0 = const()[name = tensor<string, []>("op_42956_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_42956_end_mask_0 = const()[name = tensor<string, []>("op_42956_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42956_cast = slice_by_index(begin = var_42956_begin_0, end = var_42956_end_0, end_mask = var_42956_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42956_cast")];
            tensor<int32, [4]> var_42960_begin_0 = const()[name = tensor<string, []>("op_42960_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_42960_end_0 = const()[name = tensor<string, []>("op_42960_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_42960_end_mask_0 = const()[name = tensor<string, []>("op_42960_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42960_cast = slice_by_index(begin = var_42960_begin_0, end = var_42960_end_0, end_mask = var_42960_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42960_cast")];
            tensor<int32, [4]> var_42964_begin_0 = const()[name = tensor<string, []>("op_42964_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_42964_end_0 = const()[name = tensor<string, []>("op_42964_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_42964_end_mask_0 = const()[name = tensor<string, []>("op_42964_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42964_cast = slice_by_index(begin = var_42964_begin_0, end = var_42964_end_0, end_mask = var_42964_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42964_cast")];
            tensor<int32, [4]> var_42968_begin_0 = const()[name = tensor<string, []>("op_42968_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_42968_end_0 = const()[name = tensor<string, []>("op_42968_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_42968_end_mask_0 = const()[name = tensor<string, []>("op_42968_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42968_cast = slice_by_index(begin = var_42968_begin_0, end = var_42968_end_0, end_mask = var_42968_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42968_cast")];
            tensor<int32, [4]> var_42972_begin_0 = const()[name = tensor<string, []>("op_42972_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_42972_end_0 = const()[name = tensor<string, []>("op_42972_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_42972_end_mask_0 = const()[name = tensor<string, []>("op_42972_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42972_cast = slice_by_index(begin = var_42972_begin_0, end = var_42972_end_0, end_mask = var_42972_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42972_cast")];
            tensor<int32, [4]> var_42976_begin_0 = const()[name = tensor<string, []>("op_42976_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_42976_end_0 = const()[name = tensor<string, []>("op_42976_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_42976_end_mask_0 = const()[name = tensor<string, []>("op_42976_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42976_cast = slice_by_index(begin = var_42976_begin_0, end = var_42976_end_0, end_mask = var_42976_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42976_cast")];
            tensor<int32, [4]> var_42980_begin_0 = const()[name = tensor<string, []>("op_42980_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_42980_end_0 = const()[name = tensor<string, []>("op_42980_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_42980_end_mask_0 = const()[name = tensor<string, []>("op_42980_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42980_cast = slice_by_index(begin = var_42980_begin_0, end = var_42980_end_0, end_mask = var_42980_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42980_cast")];
            tensor<int32, [4]> var_42984_begin_0 = const()[name = tensor<string, []>("op_42984_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_42984_end_0 = const()[name = tensor<string, []>("op_42984_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_42984_end_mask_0 = const()[name = tensor<string, []>("op_42984_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42984_cast = slice_by_index(begin = var_42984_begin_0, end = var_42984_end_0, end_mask = var_42984_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42984_cast")];
            tensor<int32, [4]> var_42988_begin_0 = const()[name = tensor<string, []>("op_42988_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_42988_end_0 = const()[name = tensor<string, []>("op_42988_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_42988_end_mask_0 = const()[name = tensor<string, []>("op_42988_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_42988_cast = slice_by_index(begin = var_42988_begin_0, end = var_42988_end_0, end_mask = var_42988_end_mask_0, x = q_191_cast)[name = tensor<string, []>("op_42988_cast")];
            tensor<int32, [4]> k_383_perm_0 = const()[name = tensor<string, []>("k_383_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_42995_begin_0 = const()[name = tensor<string, []>("op_42995_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_42995_end_0 = const()[name = tensor<string, []>("op_42995_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_42995_end_mask_0 = const()[name = tensor<string, []>("op_42995_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_44 = transpose(perm = k_383_perm_0, x = k_381_cast)[name = tensor<string, []>("transpose_44")];
            tensor<fp16, [2, 77, 1, 64]> var_42995_cast = slice_by_index(begin = var_42995_begin_0, end = var_42995_end_0, end_mask = var_42995_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_42995_cast")];
            tensor<int32, [4]> var_42999_begin_0 = const()[name = tensor<string, []>("op_42999_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_42999_end_0 = const()[name = tensor<string, []>("op_42999_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_42999_end_mask_0 = const()[name = tensor<string, []>("op_42999_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_42999_cast = slice_by_index(begin = var_42999_begin_0, end = var_42999_end_0, end_mask = var_42999_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_42999_cast")];
            tensor<int32, [4]> var_43003_begin_0 = const()[name = tensor<string, []>("op_43003_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_43003_end_0 = const()[name = tensor<string, []>("op_43003_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_43003_end_mask_0 = const()[name = tensor<string, []>("op_43003_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43003_cast = slice_by_index(begin = var_43003_begin_0, end = var_43003_end_0, end_mask = var_43003_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_43003_cast")];
            tensor<int32, [4]> var_43007_begin_0 = const()[name = tensor<string, []>("op_43007_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_43007_end_0 = const()[name = tensor<string, []>("op_43007_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_43007_end_mask_0 = const()[name = tensor<string, []>("op_43007_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43007_cast = slice_by_index(begin = var_43007_begin_0, end = var_43007_end_0, end_mask = var_43007_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_43007_cast")];
            tensor<int32, [4]> var_43011_begin_0 = const()[name = tensor<string, []>("op_43011_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_43011_end_0 = const()[name = tensor<string, []>("op_43011_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_43011_end_mask_0 = const()[name = tensor<string, []>("op_43011_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43011_cast = slice_by_index(begin = var_43011_begin_0, end = var_43011_end_0, end_mask = var_43011_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_43011_cast")];
            tensor<int32, [4]> var_43015_begin_0 = const()[name = tensor<string, []>("op_43015_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_43015_end_0 = const()[name = tensor<string, []>("op_43015_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_43015_end_mask_0 = const()[name = tensor<string, []>("op_43015_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43015_cast = slice_by_index(begin = var_43015_begin_0, end = var_43015_end_0, end_mask = var_43015_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_43015_cast")];
            tensor<int32, [4]> var_43019_begin_0 = const()[name = tensor<string, []>("op_43019_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_43019_end_0 = const()[name = tensor<string, []>("op_43019_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_43019_end_mask_0 = const()[name = tensor<string, []>("op_43019_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43019_cast = slice_by_index(begin = var_43019_begin_0, end = var_43019_end_0, end_mask = var_43019_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_43019_cast")];
            tensor<int32, [4]> var_43023_begin_0 = const()[name = tensor<string, []>("op_43023_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_43023_end_0 = const()[name = tensor<string, []>("op_43023_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_43023_end_mask_0 = const()[name = tensor<string, []>("op_43023_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43023_cast = slice_by_index(begin = var_43023_begin_0, end = var_43023_end_0, end_mask = var_43023_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_43023_cast")];
            tensor<int32, [4]> var_43027_begin_0 = const()[name = tensor<string, []>("op_43027_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_43027_end_0 = const()[name = tensor<string, []>("op_43027_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_43027_end_mask_0 = const()[name = tensor<string, []>("op_43027_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43027_cast = slice_by_index(begin = var_43027_begin_0, end = var_43027_end_0, end_mask = var_43027_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_43027_cast")];
            tensor<int32, [4]> var_43031_begin_0 = const()[name = tensor<string, []>("op_43031_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_43031_end_0 = const()[name = tensor<string, []>("op_43031_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_43031_end_mask_0 = const()[name = tensor<string, []>("op_43031_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43031_cast = slice_by_index(begin = var_43031_begin_0, end = var_43031_end_0, end_mask = var_43031_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_43031_cast")];
            tensor<int32, [4]> var_43035_begin_0 = const()[name = tensor<string, []>("op_43035_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_43035_end_0 = const()[name = tensor<string, []>("op_43035_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_43035_end_mask_0 = const()[name = tensor<string, []>("op_43035_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43035_cast = slice_by_index(begin = var_43035_begin_0, end = var_43035_end_0, end_mask = var_43035_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_43035_cast")];
            tensor<int32, [4]> var_43039_begin_0 = const()[name = tensor<string, []>("op_43039_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_43039_end_0 = const()[name = tensor<string, []>("op_43039_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_43039_end_mask_0 = const()[name = tensor<string, []>("op_43039_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43039_cast = slice_by_index(begin = var_43039_begin_0, end = var_43039_end_0, end_mask = var_43039_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_43039_cast")];
            tensor<int32, [4]> var_43043_begin_0 = const()[name = tensor<string, []>("op_43043_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_43043_end_0 = const()[name = tensor<string, []>("op_43043_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_43043_end_mask_0 = const()[name = tensor<string, []>("op_43043_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43043_cast = slice_by_index(begin = var_43043_begin_0, end = var_43043_end_0, end_mask = var_43043_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_43043_cast")];
            tensor<int32, [4]> var_43047_begin_0 = const()[name = tensor<string, []>("op_43047_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_43047_end_0 = const()[name = tensor<string, []>("op_43047_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_43047_end_mask_0 = const()[name = tensor<string, []>("op_43047_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43047_cast = slice_by_index(begin = var_43047_begin_0, end = var_43047_end_0, end_mask = var_43047_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_43047_cast")];
            tensor<int32, [4]> var_43051_begin_0 = const()[name = tensor<string, []>("op_43051_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_43051_end_0 = const()[name = tensor<string, []>("op_43051_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_43051_end_mask_0 = const()[name = tensor<string, []>("op_43051_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43051_cast = slice_by_index(begin = var_43051_begin_0, end = var_43051_end_0, end_mask = var_43051_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_43051_cast")];
            tensor<int32, [4]> var_43055_begin_0 = const()[name = tensor<string, []>("op_43055_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_43055_end_0 = const()[name = tensor<string, []>("op_43055_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_43055_end_mask_0 = const()[name = tensor<string, []>("op_43055_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43055_cast = slice_by_index(begin = var_43055_begin_0, end = var_43055_end_0, end_mask = var_43055_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_43055_cast")];
            tensor<int32, [4]> var_43059_begin_0 = const()[name = tensor<string, []>("op_43059_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_43059_end_0 = const()[name = tensor<string, []>("op_43059_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_43059_end_mask_0 = const()[name = tensor<string, []>("op_43059_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43059_cast = slice_by_index(begin = var_43059_begin_0, end = var_43059_end_0, end_mask = var_43059_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_43059_cast")];
            tensor<int32, [4]> var_43063_begin_0 = const()[name = tensor<string, []>("op_43063_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_43063_end_0 = const()[name = tensor<string, []>("op_43063_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_43063_end_mask_0 = const()[name = tensor<string, []>("op_43063_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43063_cast = slice_by_index(begin = var_43063_begin_0, end = var_43063_end_0, end_mask = var_43063_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_43063_cast")];
            tensor<int32, [4]> var_43067_begin_0 = const()[name = tensor<string, []>("op_43067_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_43067_end_0 = const()[name = tensor<string, []>("op_43067_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_43067_end_mask_0 = const()[name = tensor<string, []>("op_43067_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43067_cast = slice_by_index(begin = var_43067_begin_0, end = var_43067_end_0, end_mask = var_43067_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_43067_cast")];
            tensor<int32, [4]> var_43071_begin_0 = const()[name = tensor<string, []>("op_43071_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_43071_end_0 = const()[name = tensor<string, []>("op_43071_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_43071_end_mask_0 = const()[name = tensor<string, []>("op_43071_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43071_cast = slice_by_index(begin = var_43071_begin_0, end = var_43071_end_0, end_mask = var_43071_end_mask_0, x = transpose_44)[name = tensor<string, []>("op_43071_cast")];
            tensor<int32, [4]> var_43073_begin_0 = const()[name = tensor<string, []>("op_43073_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_43073_end_0 = const()[name = tensor<string, []>("op_43073_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_43073_end_mask_0 = const()[name = tensor<string, []>("op_43073_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43073_cast = slice_by_index(begin = var_43073_begin_0, end = var_43073_end_0, end_mask = var_43073_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43073_cast")];
            tensor<int32, [4]> var_43077_begin_0 = const()[name = tensor<string, []>("op_43077_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_43077_end_0 = const()[name = tensor<string, []>("op_43077_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_43077_end_mask_0 = const()[name = tensor<string, []>("op_43077_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43077_cast = slice_by_index(begin = var_43077_begin_0, end = var_43077_end_0, end_mask = var_43077_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43077_cast")];
            tensor<int32, [4]> var_43081_begin_0 = const()[name = tensor<string, []>("op_43081_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_43081_end_0 = const()[name = tensor<string, []>("op_43081_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_43081_end_mask_0 = const()[name = tensor<string, []>("op_43081_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43081_cast = slice_by_index(begin = var_43081_begin_0, end = var_43081_end_0, end_mask = var_43081_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43081_cast")];
            tensor<int32, [4]> var_43085_begin_0 = const()[name = tensor<string, []>("op_43085_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_43085_end_0 = const()[name = tensor<string, []>("op_43085_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_43085_end_mask_0 = const()[name = tensor<string, []>("op_43085_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43085_cast = slice_by_index(begin = var_43085_begin_0, end = var_43085_end_0, end_mask = var_43085_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43085_cast")];
            tensor<int32, [4]> var_43089_begin_0 = const()[name = tensor<string, []>("op_43089_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_43089_end_0 = const()[name = tensor<string, []>("op_43089_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_43089_end_mask_0 = const()[name = tensor<string, []>("op_43089_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43089_cast = slice_by_index(begin = var_43089_begin_0, end = var_43089_end_0, end_mask = var_43089_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43089_cast")];
            tensor<int32, [4]> var_43093_begin_0 = const()[name = tensor<string, []>("op_43093_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_43093_end_0 = const()[name = tensor<string, []>("op_43093_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_43093_end_mask_0 = const()[name = tensor<string, []>("op_43093_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43093_cast = slice_by_index(begin = var_43093_begin_0, end = var_43093_end_0, end_mask = var_43093_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43093_cast")];
            tensor<int32, [4]> var_43097_begin_0 = const()[name = tensor<string, []>("op_43097_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_43097_end_0 = const()[name = tensor<string, []>("op_43097_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_43097_end_mask_0 = const()[name = tensor<string, []>("op_43097_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43097_cast = slice_by_index(begin = var_43097_begin_0, end = var_43097_end_0, end_mask = var_43097_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43097_cast")];
            tensor<int32, [4]> var_43101_begin_0 = const()[name = tensor<string, []>("op_43101_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_43101_end_0 = const()[name = tensor<string, []>("op_43101_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_43101_end_mask_0 = const()[name = tensor<string, []>("op_43101_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43101_cast = slice_by_index(begin = var_43101_begin_0, end = var_43101_end_0, end_mask = var_43101_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43101_cast")];
            tensor<int32, [4]> var_43105_begin_0 = const()[name = tensor<string, []>("op_43105_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_43105_end_0 = const()[name = tensor<string, []>("op_43105_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_43105_end_mask_0 = const()[name = tensor<string, []>("op_43105_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43105_cast = slice_by_index(begin = var_43105_begin_0, end = var_43105_end_0, end_mask = var_43105_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43105_cast")];
            tensor<int32, [4]> var_43109_begin_0 = const()[name = tensor<string, []>("op_43109_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_43109_end_0 = const()[name = tensor<string, []>("op_43109_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_43109_end_mask_0 = const()[name = tensor<string, []>("op_43109_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43109_cast = slice_by_index(begin = var_43109_begin_0, end = var_43109_end_0, end_mask = var_43109_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43109_cast")];
            tensor<int32, [4]> var_43113_begin_0 = const()[name = tensor<string, []>("op_43113_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_43113_end_0 = const()[name = tensor<string, []>("op_43113_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_43113_end_mask_0 = const()[name = tensor<string, []>("op_43113_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43113_cast = slice_by_index(begin = var_43113_begin_0, end = var_43113_end_0, end_mask = var_43113_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43113_cast")];
            tensor<int32, [4]> var_43117_begin_0 = const()[name = tensor<string, []>("op_43117_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_43117_end_0 = const()[name = tensor<string, []>("op_43117_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_43117_end_mask_0 = const()[name = tensor<string, []>("op_43117_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43117_cast = slice_by_index(begin = var_43117_begin_0, end = var_43117_end_0, end_mask = var_43117_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43117_cast")];
            tensor<int32, [4]> var_43121_begin_0 = const()[name = tensor<string, []>("op_43121_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_43121_end_0 = const()[name = tensor<string, []>("op_43121_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_43121_end_mask_0 = const()[name = tensor<string, []>("op_43121_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43121_cast = slice_by_index(begin = var_43121_begin_0, end = var_43121_end_0, end_mask = var_43121_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43121_cast")];
            tensor<int32, [4]> var_43125_begin_0 = const()[name = tensor<string, []>("op_43125_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_43125_end_0 = const()[name = tensor<string, []>("op_43125_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_43125_end_mask_0 = const()[name = tensor<string, []>("op_43125_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43125_cast = slice_by_index(begin = var_43125_begin_0, end = var_43125_end_0, end_mask = var_43125_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43125_cast")];
            tensor<int32, [4]> var_43129_begin_0 = const()[name = tensor<string, []>("op_43129_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_43129_end_0 = const()[name = tensor<string, []>("op_43129_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_43129_end_mask_0 = const()[name = tensor<string, []>("op_43129_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43129_cast = slice_by_index(begin = var_43129_begin_0, end = var_43129_end_0, end_mask = var_43129_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43129_cast")];
            tensor<int32, [4]> var_43133_begin_0 = const()[name = tensor<string, []>("op_43133_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_43133_end_0 = const()[name = tensor<string, []>("op_43133_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_43133_end_mask_0 = const()[name = tensor<string, []>("op_43133_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43133_cast = slice_by_index(begin = var_43133_begin_0, end = var_43133_end_0, end_mask = var_43133_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43133_cast")];
            tensor<int32, [4]> var_43137_begin_0 = const()[name = tensor<string, []>("op_43137_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_43137_end_0 = const()[name = tensor<string, []>("op_43137_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_43137_end_mask_0 = const()[name = tensor<string, []>("op_43137_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43137_cast = slice_by_index(begin = var_43137_begin_0, end = var_43137_end_0, end_mask = var_43137_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43137_cast")];
            tensor<int32, [4]> var_43141_begin_0 = const()[name = tensor<string, []>("op_43141_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_43141_end_0 = const()[name = tensor<string, []>("op_43141_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_43141_end_mask_0 = const()[name = tensor<string, []>("op_43141_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43141_cast = slice_by_index(begin = var_43141_begin_0, end = var_43141_end_0, end_mask = var_43141_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43141_cast")];
            tensor<int32, [4]> var_43145_begin_0 = const()[name = tensor<string, []>("op_43145_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_43145_end_0 = const()[name = tensor<string, []>("op_43145_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_43145_end_mask_0 = const()[name = tensor<string, []>("op_43145_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43145_cast = slice_by_index(begin = var_43145_begin_0, end = var_43145_end_0, end_mask = var_43145_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43145_cast")];
            tensor<int32, [4]> var_43149_begin_0 = const()[name = tensor<string, []>("op_43149_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_43149_end_0 = const()[name = tensor<string, []>("op_43149_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_43149_end_mask_0 = const()[name = tensor<string, []>("op_43149_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43149_cast = slice_by_index(begin = var_43149_begin_0, end = var_43149_end_0, end_mask = var_43149_end_mask_0, x = v_191_cast)[name = tensor<string, []>("op_43149_cast")];
            tensor<string, []> var_43153_equation_0 = const()[name = tensor<string, []>("op_43153_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43153_cast = einsum(equation = var_43153_equation_0, values = (var_42995_cast, var_42912_cast))[name = tensor<string, []>("op_43153_cast")];
            tensor<fp16, []> var_43154_to_fp16 = const()[name = tensor<string, []>("op_43154_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3641_cast = mul(x = var_43153_cast, y = var_43154_to_fp16)[name = tensor<string, []>("aw_3641_cast")];
            tensor<string, []> var_43157_equation_0 = const()[name = tensor<string, []>("op_43157_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43157_cast = einsum(equation = var_43157_equation_0, values = (var_42999_cast, var_42916_cast))[name = tensor<string, []>("op_43157_cast")];
            tensor<fp16, []> var_43158_to_fp16 = const()[name = tensor<string, []>("op_43158_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3643_cast = mul(x = var_43157_cast, y = var_43158_to_fp16)[name = tensor<string, []>("aw_3643_cast")];
            tensor<string, []> var_43161_equation_0 = const()[name = tensor<string, []>("op_43161_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43161_cast = einsum(equation = var_43161_equation_0, values = (var_43003_cast, var_42920_cast))[name = tensor<string, []>("op_43161_cast")];
            tensor<fp16, []> var_43162_to_fp16 = const()[name = tensor<string, []>("op_43162_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3645_cast = mul(x = var_43161_cast, y = var_43162_to_fp16)[name = tensor<string, []>("aw_3645_cast")];
            tensor<string, []> var_43165_equation_0 = const()[name = tensor<string, []>("op_43165_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43165_cast = einsum(equation = var_43165_equation_0, values = (var_43007_cast, var_42924_cast))[name = tensor<string, []>("op_43165_cast")];
            tensor<fp16, []> var_43166_to_fp16 = const()[name = tensor<string, []>("op_43166_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3647_cast = mul(x = var_43165_cast, y = var_43166_to_fp16)[name = tensor<string, []>("aw_3647_cast")];
            tensor<string, []> var_43169_equation_0 = const()[name = tensor<string, []>("op_43169_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43169_cast = einsum(equation = var_43169_equation_0, values = (var_43011_cast, var_42928_cast))[name = tensor<string, []>("op_43169_cast")];
            tensor<fp16, []> var_43170_to_fp16 = const()[name = tensor<string, []>("op_43170_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3649_cast = mul(x = var_43169_cast, y = var_43170_to_fp16)[name = tensor<string, []>("aw_3649_cast")];
            tensor<string, []> var_43173_equation_0 = const()[name = tensor<string, []>("op_43173_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43173_cast = einsum(equation = var_43173_equation_0, values = (var_43015_cast, var_42932_cast))[name = tensor<string, []>("op_43173_cast")];
            tensor<fp16, []> var_43174_to_fp16 = const()[name = tensor<string, []>("op_43174_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3651_cast = mul(x = var_43173_cast, y = var_43174_to_fp16)[name = tensor<string, []>("aw_3651_cast")];
            tensor<string, []> var_43177_equation_0 = const()[name = tensor<string, []>("op_43177_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43177_cast = einsum(equation = var_43177_equation_0, values = (var_43019_cast, var_42936_cast))[name = tensor<string, []>("op_43177_cast")];
            tensor<fp16, []> var_43178_to_fp16 = const()[name = tensor<string, []>("op_43178_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3653_cast = mul(x = var_43177_cast, y = var_43178_to_fp16)[name = tensor<string, []>("aw_3653_cast")];
            tensor<string, []> var_43181_equation_0 = const()[name = tensor<string, []>("op_43181_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43181_cast = einsum(equation = var_43181_equation_0, values = (var_43023_cast, var_42940_cast))[name = tensor<string, []>("op_43181_cast")];
            tensor<fp16, []> var_43182_to_fp16 = const()[name = tensor<string, []>("op_43182_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3655_cast = mul(x = var_43181_cast, y = var_43182_to_fp16)[name = tensor<string, []>("aw_3655_cast")];
            tensor<string, []> var_43185_equation_0 = const()[name = tensor<string, []>("op_43185_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43185_cast = einsum(equation = var_43185_equation_0, values = (var_43027_cast, var_42944_cast))[name = tensor<string, []>("op_43185_cast")];
            tensor<fp16, []> var_43186_to_fp16 = const()[name = tensor<string, []>("op_43186_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3657_cast = mul(x = var_43185_cast, y = var_43186_to_fp16)[name = tensor<string, []>("aw_3657_cast")];
            tensor<string, []> var_43189_equation_0 = const()[name = tensor<string, []>("op_43189_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43189_cast = einsum(equation = var_43189_equation_0, values = (var_43031_cast, var_42948_cast))[name = tensor<string, []>("op_43189_cast")];
            tensor<fp16, []> var_43190_to_fp16 = const()[name = tensor<string, []>("op_43190_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3659_cast = mul(x = var_43189_cast, y = var_43190_to_fp16)[name = tensor<string, []>("aw_3659_cast")];
            tensor<string, []> var_43193_equation_0 = const()[name = tensor<string, []>("op_43193_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43193_cast = einsum(equation = var_43193_equation_0, values = (var_43035_cast, var_42952_cast))[name = tensor<string, []>("op_43193_cast")];
            tensor<fp16, []> var_43194_to_fp16 = const()[name = tensor<string, []>("op_43194_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3661_cast = mul(x = var_43193_cast, y = var_43194_to_fp16)[name = tensor<string, []>("aw_3661_cast")];
            tensor<string, []> var_43197_equation_0 = const()[name = tensor<string, []>("op_43197_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43197_cast = einsum(equation = var_43197_equation_0, values = (var_43039_cast, var_42956_cast))[name = tensor<string, []>("op_43197_cast")];
            tensor<fp16, []> var_43198_to_fp16 = const()[name = tensor<string, []>("op_43198_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3663_cast = mul(x = var_43197_cast, y = var_43198_to_fp16)[name = tensor<string, []>("aw_3663_cast")];
            tensor<string, []> var_43201_equation_0 = const()[name = tensor<string, []>("op_43201_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43201_cast = einsum(equation = var_43201_equation_0, values = (var_43043_cast, var_42960_cast))[name = tensor<string, []>("op_43201_cast")];
            tensor<fp16, []> var_43202_to_fp16 = const()[name = tensor<string, []>("op_43202_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3665_cast = mul(x = var_43201_cast, y = var_43202_to_fp16)[name = tensor<string, []>("aw_3665_cast")];
            tensor<string, []> var_43205_equation_0 = const()[name = tensor<string, []>("op_43205_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43205_cast = einsum(equation = var_43205_equation_0, values = (var_43047_cast, var_42964_cast))[name = tensor<string, []>("op_43205_cast")];
            tensor<fp16, []> var_43206_to_fp16 = const()[name = tensor<string, []>("op_43206_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3667_cast = mul(x = var_43205_cast, y = var_43206_to_fp16)[name = tensor<string, []>("aw_3667_cast")];
            tensor<string, []> var_43209_equation_0 = const()[name = tensor<string, []>("op_43209_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43209_cast = einsum(equation = var_43209_equation_0, values = (var_43051_cast, var_42968_cast))[name = tensor<string, []>("op_43209_cast")];
            tensor<fp16, []> var_43210_to_fp16 = const()[name = tensor<string, []>("op_43210_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3669_cast = mul(x = var_43209_cast, y = var_43210_to_fp16)[name = tensor<string, []>("aw_3669_cast")];
            tensor<string, []> var_43213_equation_0 = const()[name = tensor<string, []>("op_43213_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43213_cast = einsum(equation = var_43213_equation_0, values = (var_43055_cast, var_42972_cast))[name = tensor<string, []>("op_43213_cast")];
            tensor<fp16, []> var_43214_to_fp16 = const()[name = tensor<string, []>("op_43214_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3671_cast = mul(x = var_43213_cast, y = var_43214_to_fp16)[name = tensor<string, []>("aw_3671_cast")];
            tensor<string, []> var_43217_equation_0 = const()[name = tensor<string, []>("op_43217_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43217_cast = einsum(equation = var_43217_equation_0, values = (var_43059_cast, var_42976_cast))[name = tensor<string, []>("op_43217_cast")];
            tensor<fp16, []> var_43218_to_fp16 = const()[name = tensor<string, []>("op_43218_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3673_cast = mul(x = var_43217_cast, y = var_43218_to_fp16)[name = tensor<string, []>("aw_3673_cast")];
            tensor<string, []> var_43221_equation_0 = const()[name = tensor<string, []>("op_43221_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43221_cast = einsum(equation = var_43221_equation_0, values = (var_43063_cast, var_42980_cast))[name = tensor<string, []>("op_43221_cast")];
            tensor<fp16, []> var_43222_to_fp16 = const()[name = tensor<string, []>("op_43222_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3675_cast = mul(x = var_43221_cast, y = var_43222_to_fp16)[name = tensor<string, []>("aw_3675_cast")];
            tensor<string, []> var_43225_equation_0 = const()[name = tensor<string, []>("op_43225_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43225_cast = einsum(equation = var_43225_equation_0, values = (var_43067_cast, var_42984_cast))[name = tensor<string, []>("op_43225_cast")];
            tensor<fp16, []> var_43226_to_fp16 = const()[name = tensor<string, []>("op_43226_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3677_cast = mul(x = var_43225_cast, y = var_43226_to_fp16)[name = tensor<string, []>("aw_3677_cast")];
            tensor<string, []> var_43229_equation_0 = const()[name = tensor<string, []>("op_43229_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_43229_cast = einsum(equation = var_43229_equation_0, values = (var_43071_cast, var_42988_cast))[name = tensor<string, []>("op_43229_cast")];
            tensor<fp16, []> var_43230_to_fp16 = const()[name = tensor<string, []>("op_43230_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3679_cast = mul(x = var_43229_cast, y = var_43230_to_fp16)[name = tensor<string, []>("aw_3679_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43232_cast = softmax(axis = var_30385, x = aw_3641_cast)[name = tensor<string, []>("op_43232_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43233_cast = softmax(axis = var_30385, x = aw_3643_cast)[name = tensor<string, []>("op_43233_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43234_cast = softmax(axis = var_30385, x = aw_3645_cast)[name = tensor<string, []>("op_43234_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43235_cast = softmax(axis = var_30385, x = aw_3647_cast)[name = tensor<string, []>("op_43235_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43236_cast = softmax(axis = var_30385, x = aw_3649_cast)[name = tensor<string, []>("op_43236_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43237_cast = softmax(axis = var_30385, x = aw_3651_cast)[name = tensor<string, []>("op_43237_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43238_cast = softmax(axis = var_30385, x = aw_3653_cast)[name = tensor<string, []>("op_43238_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43239_cast = softmax(axis = var_30385, x = aw_3655_cast)[name = tensor<string, []>("op_43239_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43240_cast = softmax(axis = var_30385, x = aw_3657_cast)[name = tensor<string, []>("op_43240_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43241_cast = softmax(axis = var_30385, x = aw_3659_cast)[name = tensor<string, []>("op_43241_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43242_cast = softmax(axis = var_30385, x = aw_3661_cast)[name = tensor<string, []>("op_43242_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43243_cast = softmax(axis = var_30385, x = aw_3663_cast)[name = tensor<string, []>("op_43243_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43244_cast = softmax(axis = var_30385, x = aw_3665_cast)[name = tensor<string, []>("op_43244_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43245_cast = softmax(axis = var_30385, x = aw_3667_cast)[name = tensor<string, []>("op_43245_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43246_cast = softmax(axis = var_30385, x = aw_3669_cast)[name = tensor<string, []>("op_43246_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43247_cast = softmax(axis = var_30385, x = aw_3671_cast)[name = tensor<string, []>("op_43247_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43248_cast = softmax(axis = var_30385, x = aw_3673_cast)[name = tensor<string, []>("op_43248_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43249_cast = softmax(axis = var_30385, x = aw_3675_cast)[name = tensor<string, []>("op_43249_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43250_cast = softmax(axis = var_30385, x = aw_3677_cast)[name = tensor<string, []>("op_43250_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_43251_cast = softmax(axis = var_30385, x = aw_3679_cast)[name = tensor<string, []>("op_43251_cast")];
            tensor<string, []> var_43253_equation_0 = const()[name = tensor<string, []>("op_43253_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43253_cast = einsum(equation = var_43253_equation_0, values = (var_43073_cast, var_43232_cast))[name = tensor<string, []>("op_43253_cast")];
            tensor<string, []> var_43255_equation_0 = const()[name = tensor<string, []>("op_43255_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43255_cast = einsum(equation = var_43255_equation_0, values = (var_43077_cast, var_43233_cast))[name = tensor<string, []>("op_43255_cast")];
            tensor<string, []> var_43257_equation_0 = const()[name = tensor<string, []>("op_43257_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43257_cast = einsum(equation = var_43257_equation_0, values = (var_43081_cast, var_43234_cast))[name = tensor<string, []>("op_43257_cast")];
            tensor<string, []> var_43259_equation_0 = const()[name = tensor<string, []>("op_43259_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43259_cast = einsum(equation = var_43259_equation_0, values = (var_43085_cast, var_43235_cast))[name = tensor<string, []>("op_43259_cast")];
            tensor<string, []> var_43261_equation_0 = const()[name = tensor<string, []>("op_43261_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43261_cast = einsum(equation = var_43261_equation_0, values = (var_43089_cast, var_43236_cast))[name = tensor<string, []>("op_43261_cast")];
            tensor<string, []> var_43263_equation_0 = const()[name = tensor<string, []>("op_43263_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43263_cast = einsum(equation = var_43263_equation_0, values = (var_43093_cast, var_43237_cast))[name = tensor<string, []>("op_43263_cast")];
            tensor<string, []> var_43265_equation_0 = const()[name = tensor<string, []>("op_43265_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43265_cast = einsum(equation = var_43265_equation_0, values = (var_43097_cast, var_43238_cast))[name = tensor<string, []>("op_43265_cast")];
            tensor<string, []> var_43267_equation_0 = const()[name = tensor<string, []>("op_43267_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43267_cast = einsum(equation = var_43267_equation_0, values = (var_43101_cast, var_43239_cast))[name = tensor<string, []>("op_43267_cast")];
            tensor<string, []> var_43269_equation_0 = const()[name = tensor<string, []>("op_43269_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43269_cast = einsum(equation = var_43269_equation_0, values = (var_43105_cast, var_43240_cast))[name = tensor<string, []>("op_43269_cast")];
            tensor<string, []> var_43271_equation_0 = const()[name = tensor<string, []>("op_43271_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43271_cast = einsum(equation = var_43271_equation_0, values = (var_43109_cast, var_43241_cast))[name = tensor<string, []>("op_43271_cast")];
            tensor<string, []> var_43273_equation_0 = const()[name = tensor<string, []>("op_43273_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43273_cast = einsum(equation = var_43273_equation_0, values = (var_43113_cast, var_43242_cast))[name = tensor<string, []>("op_43273_cast")];
            tensor<string, []> var_43275_equation_0 = const()[name = tensor<string, []>("op_43275_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43275_cast = einsum(equation = var_43275_equation_0, values = (var_43117_cast, var_43243_cast))[name = tensor<string, []>("op_43275_cast")];
            tensor<string, []> var_43277_equation_0 = const()[name = tensor<string, []>("op_43277_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43277_cast = einsum(equation = var_43277_equation_0, values = (var_43121_cast, var_43244_cast))[name = tensor<string, []>("op_43277_cast")];
            tensor<string, []> var_43279_equation_0 = const()[name = tensor<string, []>("op_43279_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43279_cast = einsum(equation = var_43279_equation_0, values = (var_43125_cast, var_43245_cast))[name = tensor<string, []>("op_43279_cast")];
            tensor<string, []> var_43281_equation_0 = const()[name = tensor<string, []>("op_43281_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43281_cast = einsum(equation = var_43281_equation_0, values = (var_43129_cast, var_43246_cast))[name = tensor<string, []>("op_43281_cast")];
            tensor<string, []> var_43283_equation_0 = const()[name = tensor<string, []>("op_43283_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43283_cast = einsum(equation = var_43283_equation_0, values = (var_43133_cast, var_43247_cast))[name = tensor<string, []>("op_43283_cast")];
            tensor<string, []> var_43285_equation_0 = const()[name = tensor<string, []>("op_43285_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43285_cast = einsum(equation = var_43285_equation_0, values = (var_43137_cast, var_43248_cast))[name = tensor<string, []>("op_43285_cast")];
            tensor<string, []> var_43287_equation_0 = const()[name = tensor<string, []>("op_43287_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43287_cast = einsum(equation = var_43287_equation_0, values = (var_43141_cast, var_43249_cast))[name = tensor<string, []>("op_43287_cast")];
            tensor<string, []> var_43289_equation_0 = const()[name = tensor<string, []>("op_43289_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43289_cast = einsum(equation = var_43289_equation_0, values = (var_43145_cast, var_43250_cast))[name = tensor<string, []>("op_43289_cast")];
            tensor<string, []> var_43291_equation_0 = const()[name = tensor<string, []>("op_43291_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43291_cast = einsum(equation = var_43291_equation_0, values = (var_43149_cast, var_43251_cast))[name = tensor<string, []>("op_43291_cast")];
            tensor<bool, []> input_565_interleave_0 = const()[name = tensor<string, []>("input_565_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_565_cast = concat(axis = var_30385, interleave = input_565_interleave_0, values = (var_43253_cast, var_43255_cast, var_43257_cast, var_43259_cast, var_43261_cast, var_43263_cast, var_43265_cast, var_43267_cast, var_43269_cast, var_43271_cast, var_43273_cast, var_43275_cast, var_43277_cast, var_43279_cast, var_43281_cast, var_43283_cast, var_43285_cast, var_43287_cast, var_43289_cast, var_43291_cast))[name = tensor<string, []>("input_565_cast")];
            tensor<int32, [2]> var_43297 = const()[name = tensor<string, []>("op_43297"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_43299 = const()[name = tensor<string, []>("op_43299"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_43301_pad_type_0 = const()[name = tensor<string, []>("op_43301_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_43301_pad_0 = const()[name = tensor<string, []>("op_43301_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3623867776)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3627144640)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_43301_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_out_0_bias_to_fp16, dilations = var_43299, groups = var_30385, pad = var_43301_pad_0, pad_type = var_43301_pad_type_0, strides = var_43297, weight = up_blocks_0_attentions_1_transformer_blocks_3_attn2_to_out_0_weight_to_fp16, x = input_565_cast)[name = tensor<string, []>("op_43301_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_287_cast = add(x = var_43301_cast, y = inputs_285_cast)[name = tensor<string, []>("inputs_287_cast")];
            tensor<int32, [1]> var_43305 = const()[name = tensor<string, []>("op_43305"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_287_cast = reduce_mean(axes = var_43305, keep_dims = var_30380, x = inputs_287_cast)[name = tensor<string, []>("channels_mean_287_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_287_cast = sub(x = inputs_287_cast, y = channels_mean_287_cast)[name = tensor<string, []>("zero_mean_287_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_287_cast = mul(x = zero_mean_287_cast, y = zero_mean_287_cast)[name = tensor<string, []>("zero_mean_sq_287_cast")];
            tensor<int32, [1]> var_43309 = const()[name = tensor<string, []>("op_43309"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_43310_cast = reduce_mean(axes = var_43309, keep_dims = var_30380, x = zero_mean_sq_287_cast)[name = tensor<string, []>("op_43310_cast")];
            tensor<fp16, []> var_43311_to_fp16 = const()[name = tensor<string, []>("op_43311_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_43312_cast = add(x = var_43310_cast, y = var_43311_to_fp16)[name = tensor<string, []>("op_43312_cast")];
            tensor<fp16, []> denom_287_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_287_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_287_cast = rsqrt(epsilon = denom_287_epsilon_0_to_fp16, x = var_43312_cast)[name = tensor<string, []>("denom_287_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_287_cast = mul(x = zero_mean_287_cast, y = denom_287_cast)[name = tensor<string, []>("out_287_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_43316_to_fp16 = const()[name = tensor<string, []>("op_43316_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3627147264)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_43317_cast = add(x = out_287_cast, y = var_43316_to_fp16)[name = tensor<string, []>("op_43317_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_43319_to_fp16 = const()[name = tensor<string, []>("op_43319_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3627149888)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_567_cast = mul(x = var_43317_cast, y = var_43319_to_fp16)[name = tensor<string, []>("input_567_cast")];
            tensor<int32, [2]> var_43327 = const()[name = tensor<string, []>("op_43327"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_43329 = const()[name = tensor<string, []>("op_43329"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_43331_pad_type_0 = const()[name = tensor<string, []>("op_43331_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_43331_pad_0 = const()[name = tensor<string, []>("op_43331_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3627152512)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_1_transformer_blocks_3_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3653366976)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_43331_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_3_ff_net_0_proj_bias_to_fp16, dilations = var_43329, groups = var_30385, pad = var_43331_pad_0, pad_type = var_43331_pad_type_0, strides = var_43327, weight = up_blocks_0_attentions_1_transformer_blocks_3_ff_net_0_proj_weight_to_fp16, x = input_567_cast)[name = tensor<string, []>("op_43331_cast")];
            tensor<int32, [2]> var_43332_split_sizes_0 = const()[name = tensor<string, []>("op_43332_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_43332_axis_0 = const()[name = tensor<string, []>("op_43332_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_43332_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_43332_cast_1 = split(axis = var_43332_axis_0, split_sizes = var_43332_split_sizes_0, x = var_43331_cast)[name = tensor<string, []>("op_43332_cast")];
            tensor<string, []> var_43334_mode_0 = const()[name = tensor<string, []>("op_43334_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_43334_cast = gelu(mode = var_43334_mode_0, x = var_43332_cast_1)[name = tensor<string, []>("op_43334_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_569_cast = mul(x = var_43332_cast_0, y = var_43334_cast)[name = tensor<string, []>("input_569_cast")];
            tensor<int32, [2]> var_43338 = const()[name = tensor<string, []>("op_43338"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_43340 = const()[name = tensor<string, []>("op_43340"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_43342_pad_type_0 = const()[name = tensor<string, []>("op_43342_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_43342_pad_0 = const()[name = tensor<string, []>("op_43342_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_3_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3653387520)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_3_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_3_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3666494784)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_43342_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_3_ff_net_2_bias_to_fp16, dilations = var_43340, groups = var_30385, pad = var_43342_pad_0, pad_type = var_43342_pad_type_0, strides = var_43338, weight = up_blocks_0_attentions_1_transformer_blocks_3_ff_net_2_weight_to_fp16, x = input_569_cast)[name = tensor<string, []>("op_43342_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_289_cast = add(x = var_43342_cast, y = inputs_287_cast)[name = tensor<string, []>("inputs_289_cast")];
            tensor<int32, [1]> var_43352 = const()[name = tensor<string, []>("op_43352"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_289_cast = reduce_mean(axes = var_43352, keep_dims = var_30380, x = inputs_289_cast)[name = tensor<string, []>("channels_mean_289_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_289_cast = sub(x = inputs_289_cast, y = channels_mean_289_cast)[name = tensor<string, []>("zero_mean_289_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_289_cast = mul(x = zero_mean_289_cast, y = zero_mean_289_cast)[name = tensor<string, []>("zero_mean_sq_289_cast")];
            tensor<int32, [1]> var_43356 = const()[name = tensor<string, []>("op_43356"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_43357_cast = reduce_mean(axes = var_43356, keep_dims = var_30380, x = zero_mean_sq_289_cast)[name = tensor<string, []>("op_43357_cast")];
            tensor<fp16, []> var_43358_to_fp16 = const()[name = tensor<string, []>("op_43358_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_43359_cast = add(x = var_43357_cast, y = var_43358_to_fp16)[name = tensor<string, []>("op_43359_cast")];
            tensor<fp16, []> denom_289_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_289_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_289_cast = rsqrt(epsilon = denom_289_epsilon_0_to_fp16, x = var_43359_cast)[name = tensor<string, []>("denom_289_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_289_cast = mul(x = zero_mean_289_cast, y = denom_289_cast)[name = tensor<string, []>("out_289_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_43363_to_fp16 = const()[name = tensor<string, []>("op_43363_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3666497408)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_43364_cast = add(x = out_289_cast, y = var_43363_to_fp16)[name = tensor<string, []>("op_43364_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_43366_to_fp16 = const()[name = tensor<string, []>("op_43366_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3666500032)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_385_cast = mul(x = var_43364_cast, y = var_43366_to_fp16)[name = tensor<string, []>("hidden_states_385_cast")];
            tensor<int32, [2]> var_43373 = const()[name = tensor<string, []>("op_43373"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_43375 = const()[name = tensor<string, []>("op_43375"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_193_pad_type_0 = const()[name = tensor<string, []>("q_193_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_193_pad_0 = const()[name = tensor<string, []>("q_193_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3666502656)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_193_cast = conv(dilations = var_43375, groups = var_30385, pad = q_193_pad_0, pad_type = q_193_pad_type_0, strides = var_43373, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_q_weight_to_fp16, x = hidden_states_385_cast)[name = tensor<string, []>("q_193_cast")];
            tensor<int32, [2]> var_43379 = const()[name = tensor<string, []>("op_43379"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_43381 = const()[name = tensor<string, []>("op_43381"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_385_pad_type_0 = const()[name = tensor<string, []>("k_385_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_385_pad_0 = const()[name = tensor<string, []>("k_385_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3669779520)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_385_cast = conv(dilations = var_43381, groups = var_30385, pad = k_385_pad_0, pad_type = k_385_pad_type_0, strides = var_43379, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_k_weight_to_fp16, x = hidden_states_385_cast)[name = tensor<string, []>("k_385_cast")];
            tensor<int32, [2]> var_43385 = const()[name = tensor<string, []>("op_43385"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_43387 = const()[name = tensor<string, []>("op_43387"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_193_pad_type_0 = const()[name = tensor<string, []>("v_193_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_193_pad_0 = const()[name = tensor<string, []>("v_193_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3673056384)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_193_cast = conv(dilations = var_43387, groups = var_30385, pad = v_193_pad_0, pad_type = v_193_pad_type_0, strides = var_43385, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_v_weight_to_fp16, x = hidden_states_385_cast)[name = tensor<string, []>("v_193_cast")];
            tensor<int32, [4]> var_43391_begin_0 = const()[name = tensor<string, []>("op_43391_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_43391_end_0 = const()[name = tensor<string, []>("op_43391_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_43391_end_mask_0 = const()[name = tensor<string, []>("op_43391_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43391_cast = slice_by_index(begin = var_43391_begin_0, end = var_43391_end_0, end_mask = var_43391_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43391_cast")];
            tensor<int32, [4]> var_43395_begin_0 = const()[name = tensor<string, []>("op_43395_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_43395_end_0 = const()[name = tensor<string, []>("op_43395_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_43395_end_mask_0 = const()[name = tensor<string, []>("op_43395_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43395_cast = slice_by_index(begin = var_43395_begin_0, end = var_43395_end_0, end_mask = var_43395_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43395_cast")];
            tensor<int32, [4]> var_43399_begin_0 = const()[name = tensor<string, []>("op_43399_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_43399_end_0 = const()[name = tensor<string, []>("op_43399_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_43399_end_mask_0 = const()[name = tensor<string, []>("op_43399_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43399_cast = slice_by_index(begin = var_43399_begin_0, end = var_43399_end_0, end_mask = var_43399_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43399_cast")];
            tensor<int32, [4]> var_43403_begin_0 = const()[name = tensor<string, []>("op_43403_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_43403_end_0 = const()[name = tensor<string, []>("op_43403_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_43403_end_mask_0 = const()[name = tensor<string, []>("op_43403_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43403_cast = slice_by_index(begin = var_43403_begin_0, end = var_43403_end_0, end_mask = var_43403_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43403_cast")];
            tensor<int32, [4]> var_43407_begin_0 = const()[name = tensor<string, []>("op_43407_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_43407_end_0 = const()[name = tensor<string, []>("op_43407_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_43407_end_mask_0 = const()[name = tensor<string, []>("op_43407_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43407_cast = slice_by_index(begin = var_43407_begin_0, end = var_43407_end_0, end_mask = var_43407_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43407_cast")];
            tensor<int32, [4]> var_43411_begin_0 = const()[name = tensor<string, []>("op_43411_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_43411_end_0 = const()[name = tensor<string, []>("op_43411_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_43411_end_mask_0 = const()[name = tensor<string, []>("op_43411_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43411_cast = slice_by_index(begin = var_43411_begin_0, end = var_43411_end_0, end_mask = var_43411_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43411_cast")];
            tensor<int32, [4]> var_43415_begin_0 = const()[name = tensor<string, []>("op_43415_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_43415_end_0 = const()[name = tensor<string, []>("op_43415_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_43415_end_mask_0 = const()[name = tensor<string, []>("op_43415_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43415_cast = slice_by_index(begin = var_43415_begin_0, end = var_43415_end_0, end_mask = var_43415_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43415_cast")];
            tensor<int32, [4]> var_43419_begin_0 = const()[name = tensor<string, []>("op_43419_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_43419_end_0 = const()[name = tensor<string, []>("op_43419_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_43419_end_mask_0 = const()[name = tensor<string, []>("op_43419_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43419_cast = slice_by_index(begin = var_43419_begin_0, end = var_43419_end_0, end_mask = var_43419_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43419_cast")];
            tensor<int32, [4]> var_43423_begin_0 = const()[name = tensor<string, []>("op_43423_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_43423_end_0 = const()[name = tensor<string, []>("op_43423_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_43423_end_mask_0 = const()[name = tensor<string, []>("op_43423_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43423_cast = slice_by_index(begin = var_43423_begin_0, end = var_43423_end_0, end_mask = var_43423_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43423_cast")];
            tensor<int32, [4]> var_43427_begin_0 = const()[name = tensor<string, []>("op_43427_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_43427_end_0 = const()[name = tensor<string, []>("op_43427_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_43427_end_mask_0 = const()[name = tensor<string, []>("op_43427_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43427_cast = slice_by_index(begin = var_43427_begin_0, end = var_43427_end_0, end_mask = var_43427_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43427_cast")];
            tensor<int32, [4]> var_43431_begin_0 = const()[name = tensor<string, []>("op_43431_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_43431_end_0 = const()[name = tensor<string, []>("op_43431_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_43431_end_mask_0 = const()[name = tensor<string, []>("op_43431_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43431_cast = slice_by_index(begin = var_43431_begin_0, end = var_43431_end_0, end_mask = var_43431_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43431_cast")];
            tensor<int32, [4]> var_43435_begin_0 = const()[name = tensor<string, []>("op_43435_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_43435_end_0 = const()[name = tensor<string, []>("op_43435_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_43435_end_mask_0 = const()[name = tensor<string, []>("op_43435_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43435_cast = slice_by_index(begin = var_43435_begin_0, end = var_43435_end_0, end_mask = var_43435_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43435_cast")];
            tensor<int32, [4]> var_43439_begin_0 = const()[name = tensor<string, []>("op_43439_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_43439_end_0 = const()[name = tensor<string, []>("op_43439_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_43439_end_mask_0 = const()[name = tensor<string, []>("op_43439_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43439_cast = slice_by_index(begin = var_43439_begin_0, end = var_43439_end_0, end_mask = var_43439_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43439_cast")];
            tensor<int32, [4]> var_43443_begin_0 = const()[name = tensor<string, []>("op_43443_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_43443_end_0 = const()[name = tensor<string, []>("op_43443_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_43443_end_mask_0 = const()[name = tensor<string, []>("op_43443_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43443_cast = slice_by_index(begin = var_43443_begin_0, end = var_43443_end_0, end_mask = var_43443_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43443_cast")];
            tensor<int32, [4]> var_43447_begin_0 = const()[name = tensor<string, []>("op_43447_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_43447_end_0 = const()[name = tensor<string, []>("op_43447_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_43447_end_mask_0 = const()[name = tensor<string, []>("op_43447_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43447_cast = slice_by_index(begin = var_43447_begin_0, end = var_43447_end_0, end_mask = var_43447_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43447_cast")];
            tensor<int32, [4]> var_43451_begin_0 = const()[name = tensor<string, []>("op_43451_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_43451_end_0 = const()[name = tensor<string, []>("op_43451_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_43451_end_mask_0 = const()[name = tensor<string, []>("op_43451_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43451_cast = slice_by_index(begin = var_43451_begin_0, end = var_43451_end_0, end_mask = var_43451_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43451_cast")];
            tensor<int32, [4]> var_43455_begin_0 = const()[name = tensor<string, []>("op_43455_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_43455_end_0 = const()[name = tensor<string, []>("op_43455_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_43455_end_mask_0 = const()[name = tensor<string, []>("op_43455_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43455_cast = slice_by_index(begin = var_43455_begin_0, end = var_43455_end_0, end_mask = var_43455_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43455_cast")];
            tensor<int32, [4]> var_43459_begin_0 = const()[name = tensor<string, []>("op_43459_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_43459_end_0 = const()[name = tensor<string, []>("op_43459_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_43459_end_mask_0 = const()[name = tensor<string, []>("op_43459_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43459_cast = slice_by_index(begin = var_43459_begin_0, end = var_43459_end_0, end_mask = var_43459_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43459_cast")];
            tensor<int32, [4]> var_43463_begin_0 = const()[name = tensor<string, []>("op_43463_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_43463_end_0 = const()[name = tensor<string, []>("op_43463_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_43463_end_mask_0 = const()[name = tensor<string, []>("op_43463_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43463_cast = slice_by_index(begin = var_43463_begin_0, end = var_43463_end_0, end_mask = var_43463_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43463_cast")];
            tensor<int32, [4]> var_43467_begin_0 = const()[name = tensor<string, []>("op_43467_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_43467_end_0 = const()[name = tensor<string, []>("op_43467_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_43467_end_mask_0 = const()[name = tensor<string, []>("op_43467_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43467_cast = slice_by_index(begin = var_43467_begin_0, end = var_43467_end_0, end_mask = var_43467_end_mask_0, x = q_193_cast)[name = tensor<string, []>("op_43467_cast")];
            tensor<int32, [4]> k_387_perm_0 = const()[name = tensor<string, []>("k_387_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_43474_begin_0 = const()[name = tensor<string, []>("op_43474_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_43474_end_0 = const()[name = tensor<string, []>("op_43474_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_43474_end_mask_0 = const()[name = tensor<string, []>("op_43474_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_43 = transpose(perm = k_387_perm_0, x = k_385_cast)[name = tensor<string, []>("transpose_43")];
            tensor<fp16, [2, 1024, 1, 64]> var_43474_cast = slice_by_index(begin = var_43474_begin_0, end = var_43474_end_0, end_mask = var_43474_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43474_cast")];
            tensor<int32, [4]> var_43478_begin_0 = const()[name = tensor<string, []>("op_43478_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_43478_end_0 = const()[name = tensor<string, []>("op_43478_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_43478_end_mask_0 = const()[name = tensor<string, []>("op_43478_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43478_cast = slice_by_index(begin = var_43478_begin_0, end = var_43478_end_0, end_mask = var_43478_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43478_cast")];
            tensor<int32, [4]> var_43482_begin_0 = const()[name = tensor<string, []>("op_43482_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_43482_end_0 = const()[name = tensor<string, []>("op_43482_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_43482_end_mask_0 = const()[name = tensor<string, []>("op_43482_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43482_cast = slice_by_index(begin = var_43482_begin_0, end = var_43482_end_0, end_mask = var_43482_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43482_cast")];
            tensor<int32, [4]> var_43486_begin_0 = const()[name = tensor<string, []>("op_43486_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_43486_end_0 = const()[name = tensor<string, []>("op_43486_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_43486_end_mask_0 = const()[name = tensor<string, []>("op_43486_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43486_cast = slice_by_index(begin = var_43486_begin_0, end = var_43486_end_0, end_mask = var_43486_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43486_cast")];
            tensor<int32, [4]> var_43490_begin_0 = const()[name = tensor<string, []>("op_43490_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_43490_end_0 = const()[name = tensor<string, []>("op_43490_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_43490_end_mask_0 = const()[name = tensor<string, []>("op_43490_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43490_cast = slice_by_index(begin = var_43490_begin_0, end = var_43490_end_0, end_mask = var_43490_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43490_cast")];
            tensor<int32, [4]> var_43494_begin_0 = const()[name = tensor<string, []>("op_43494_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_43494_end_0 = const()[name = tensor<string, []>("op_43494_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_43494_end_mask_0 = const()[name = tensor<string, []>("op_43494_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43494_cast = slice_by_index(begin = var_43494_begin_0, end = var_43494_end_0, end_mask = var_43494_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43494_cast")];
            tensor<int32, [4]> var_43498_begin_0 = const()[name = tensor<string, []>("op_43498_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_43498_end_0 = const()[name = tensor<string, []>("op_43498_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_43498_end_mask_0 = const()[name = tensor<string, []>("op_43498_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43498_cast = slice_by_index(begin = var_43498_begin_0, end = var_43498_end_0, end_mask = var_43498_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43498_cast")];
            tensor<int32, [4]> var_43502_begin_0 = const()[name = tensor<string, []>("op_43502_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_43502_end_0 = const()[name = tensor<string, []>("op_43502_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_43502_end_mask_0 = const()[name = tensor<string, []>("op_43502_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43502_cast = slice_by_index(begin = var_43502_begin_0, end = var_43502_end_0, end_mask = var_43502_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43502_cast")];
            tensor<int32, [4]> var_43506_begin_0 = const()[name = tensor<string, []>("op_43506_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_43506_end_0 = const()[name = tensor<string, []>("op_43506_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_43506_end_mask_0 = const()[name = tensor<string, []>("op_43506_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43506_cast = slice_by_index(begin = var_43506_begin_0, end = var_43506_end_0, end_mask = var_43506_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43506_cast")];
            tensor<int32, [4]> var_43510_begin_0 = const()[name = tensor<string, []>("op_43510_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_43510_end_0 = const()[name = tensor<string, []>("op_43510_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_43510_end_mask_0 = const()[name = tensor<string, []>("op_43510_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43510_cast = slice_by_index(begin = var_43510_begin_0, end = var_43510_end_0, end_mask = var_43510_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43510_cast")];
            tensor<int32, [4]> var_43514_begin_0 = const()[name = tensor<string, []>("op_43514_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_43514_end_0 = const()[name = tensor<string, []>("op_43514_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_43514_end_mask_0 = const()[name = tensor<string, []>("op_43514_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43514_cast = slice_by_index(begin = var_43514_begin_0, end = var_43514_end_0, end_mask = var_43514_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43514_cast")];
            tensor<int32, [4]> var_43518_begin_0 = const()[name = tensor<string, []>("op_43518_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_43518_end_0 = const()[name = tensor<string, []>("op_43518_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_43518_end_mask_0 = const()[name = tensor<string, []>("op_43518_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43518_cast = slice_by_index(begin = var_43518_begin_0, end = var_43518_end_0, end_mask = var_43518_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43518_cast")];
            tensor<int32, [4]> var_43522_begin_0 = const()[name = tensor<string, []>("op_43522_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_43522_end_0 = const()[name = tensor<string, []>("op_43522_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_43522_end_mask_0 = const()[name = tensor<string, []>("op_43522_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43522_cast = slice_by_index(begin = var_43522_begin_0, end = var_43522_end_0, end_mask = var_43522_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43522_cast")];
            tensor<int32, [4]> var_43526_begin_0 = const()[name = tensor<string, []>("op_43526_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_43526_end_0 = const()[name = tensor<string, []>("op_43526_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_43526_end_mask_0 = const()[name = tensor<string, []>("op_43526_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43526_cast = slice_by_index(begin = var_43526_begin_0, end = var_43526_end_0, end_mask = var_43526_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43526_cast")];
            tensor<int32, [4]> var_43530_begin_0 = const()[name = tensor<string, []>("op_43530_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_43530_end_0 = const()[name = tensor<string, []>("op_43530_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_43530_end_mask_0 = const()[name = tensor<string, []>("op_43530_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43530_cast = slice_by_index(begin = var_43530_begin_0, end = var_43530_end_0, end_mask = var_43530_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43530_cast")];
            tensor<int32, [4]> var_43534_begin_0 = const()[name = tensor<string, []>("op_43534_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_43534_end_0 = const()[name = tensor<string, []>("op_43534_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_43534_end_mask_0 = const()[name = tensor<string, []>("op_43534_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43534_cast = slice_by_index(begin = var_43534_begin_0, end = var_43534_end_0, end_mask = var_43534_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43534_cast")];
            tensor<int32, [4]> var_43538_begin_0 = const()[name = tensor<string, []>("op_43538_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_43538_end_0 = const()[name = tensor<string, []>("op_43538_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_43538_end_mask_0 = const()[name = tensor<string, []>("op_43538_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43538_cast = slice_by_index(begin = var_43538_begin_0, end = var_43538_end_0, end_mask = var_43538_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43538_cast")];
            tensor<int32, [4]> var_43542_begin_0 = const()[name = tensor<string, []>("op_43542_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_43542_end_0 = const()[name = tensor<string, []>("op_43542_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_43542_end_mask_0 = const()[name = tensor<string, []>("op_43542_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43542_cast = slice_by_index(begin = var_43542_begin_0, end = var_43542_end_0, end_mask = var_43542_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43542_cast")];
            tensor<int32, [4]> var_43546_begin_0 = const()[name = tensor<string, []>("op_43546_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_43546_end_0 = const()[name = tensor<string, []>("op_43546_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_43546_end_mask_0 = const()[name = tensor<string, []>("op_43546_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43546_cast = slice_by_index(begin = var_43546_begin_0, end = var_43546_end_0, end_mask = var_43546_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43546_cast")];
            tensor<int32, [4]> var_43550_begin_0 = const()[name = tensor<string, []>("op_43550_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_43550_end_0 = const()[name = tensor<string, []>("op_43550_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_43550_end_mask_0 = const()[name = tensor<string, []>("op_43550_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_43550_cast = slice_by_index(begin = var_43550_begin_0, end = var_43550_end_0, end_mask = var_43550_end_mask_0, x = transpose_43)[name = tensor<string, []>("op_43550_cast")];
            tensor<int32, [4]> var_43552_begin_0 = const()[name = tensor<string, []>("op_43552_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_43552_end_0 = const()[name = tensor<string, []>("op_43552_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_43552_end_mask_0 = const()[name = tensor<string, []>("op_43552_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43552_cast = slice_by_index(begin = var_43552_begin_0, end = var_43552_end_0, end_mask = var_43552_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43552_cast")];
            tensor<int32, [4]> var_43556_begin_0 = const()[name = tensor<string, []>("op_43556_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_43556_end_0 = const()[name = tensor<string, []>("op_43556_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_43556_end_mask_0 = const()[name = tensor<string, []>("op_43556_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43556_cast = slice_by_index(begin = var_43556_begin_0, end = var_43556_end_0, end_mask = var_43556_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43556_cast")];
            tensor<int32, [4]> var_43560_begin_0 = const()[name = tensor<string, []>("op_43560_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_43560_end_0 = const()[name = tensor<string, []>("op_43560_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_43560_end_mask_0 = const()[name = tensor<string, []>("op_43560_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43560_cast = slice_by_index(begin = var_43560_begin_0, end = var_43560_end_0, end_mask = var_43560_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43560_cast")];
            tensor<int32, [4]> var_43564_begin_0 = const()[name = tensor<string, []>("op_43564_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_43564_end_0 = const()[name = tensor<string, []>("op_43564_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_43564_end_mask_0 = const()[name = tensor<string, []>("op_43564_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43564_cast = slice_by_index(begin = var_43564_begin_0, end = var_43564_end_0, end_mask = var_43564_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43564_cast")];
            tensor<int32, [4]> var_43568_begin_0 = const()[name = tensor<string, []>("op_43568_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_43568_end_0 = const()[name = tensor<string, []>("op_43568_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_43568_end_mask_0 = const()[name = tensor<string, []>("op_43568_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43568_cast = slice_by_index(begin = var_43568_begin_0, end = var_43568_end_0, end_mask = var_43568_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43568_cast")];
            tensor<int32, [4]> var_43572_begin_0 = const()[name = tensor<string, []>("op_43572_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_43572_end_0 = const()[name = tensor<string, []>("op_43572_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_43572_end_mask_0 = const()[name = tensor<string, []>("op_43572_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43572_cast = slice_by_index(begin = var_43572_begin_0, end = var_43572_end_0, end_mask = var_43572_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43572_cast")];
            tensor<int32, [4]> var_43576_begin_0 = const()[name = tensor<string, []>("op_43576_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_43576_end_0 = const()[name = tensor<string, []>("op_43576_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_43576_end_mask_0 = const()[name = tensor<string, []>("op_43576_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43576_cast = slice_by_index(begin = var_43576_begin_0, end = var_43576_end_0, end_mask = var_43576_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43576_cast")];
            tensor<int32, [4]> var_43580_begin_0 = const()[name = tensor<string, []>("op_43580_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_43580_end_0 = const()[name = tensor<string, []>("op_43580_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_43580_end_mask_0 = const()[name = tensor<string, []>("op_43580_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43580_cast = slice_by_index(begin = var_43580_begin_0, end = var_43580_end_0, end_mask = var_43580_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43580_cast")];
            tensor<int32, [4]> var_43584_begin_0 = const()[name = tensor<string, []>("op_43584_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_43584_end_0 = const()[name = tensor<string, []>("op_43584_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_43584_end_mask_0 = const()[name = tensor<string, []>("op_43584_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43584_cast = slice_by_index(begin = var_43584_begin_0, end = var_43584_end_0, end_mask = var_43584_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43584_cast")];
            tensor<int32, [4]> var_43588_begin_0 = const()[name = tensor<string, []>("op_43588_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_43588_end_0 = const()[name = tensor<string, []>("op_43588_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_43588_end_mask_0 = const()[name = tensor<string, []>("op_43588_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43588_cast = slice_by_index(begin = var_43588_begin_0, end = var_43588_end_0, end_mask = var_43588_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43588_cast")];
            tensor<int32, [4]> var_43592_begin_0 = const()[name = tensor<string, []>("op_43592_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_43592_end_0 = const()[name = tensor<string, []>("op_43592_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_43592_end_mask_0 = const()[name = tensor<string, []>("op_43592_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43592_cast = slice_by_index(begin = var_43592_begin_0, end = var_43592_end_0, end_mask = var_43592_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43592_cast")];
            tensor<int32, [4]> var_43596_begin_0 = const()[name = tensor<string, []>("op_43596_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_43596_end_0 = const()[name = tensor<string, []>("op_43596_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_43596_end_mask_0 = const()[name = tensor<string, []>("op_43596_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43596_cast = slice_by_index(begin = var_43596_begin_0, end = var_43596_end_0, end_mask = var_43596_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43596_cast")];
            tensor<int32, [4]> var_43600_begin_0 = const()[name = tensor<string, []>("op_43600_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_43600_end_0 = const()[name = tensor<string, []>("op_43600_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_43600_end_mask_0 = const()[name = tensor<string, []>("op_43600_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43600_cast = slice_by_index(begin = var_43600_begin_0, end = var_43600_end_0, end_mask = var_43600_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43600_cast")];
            tensor<int32, [4]> var_43604_begin_0 = const()[name = tensor<string, []>("op_43604_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_43604_end_0 = const()[name = tensor<string, []>("op_43604_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_43604_end_mask_0 = const()[name = tensor<string, []>("op_43604_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43604_cast = slice_by_index(begin = var_43604_begin_0, end = var_43604_end_0, end_mask = var_43604_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43604_cast")];
            tensor<int32, [4]> var_43608_begin_0 = const()[name = tensor<string, []>("op_43608_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_43608_end_0 = const()[name = tensor<string, []>("op_43608_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_43608_end_mask_0 = const()[name = tensor<string, []>("op_43608_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43608_cast = slice_by_index(begin = var_43608_begin_0, end = var_43608_end_0, end_mask = var_43608_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43608_cast")];
            tensor<int32, [4]> var_43612_begin_0 = const()[name = tensor<string, []>("op_43612_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_43612_end_0 = const()[name = tensor<string, []>("op_43612_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_43612_end_mask_0 = const()[name = tensor<string, []>("op_43612_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43612_cast = slice_by_index(begin = var_43612_begin_0, end = var_43612_end_0, end_mask = var_43612_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43612_cast")];
            tensor<int32, [4]> var_43616_begin_0 = const()[name = tensor<string, []>("op_43616_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_43616_end_0 = const()[name = tensor<string, []>("op_43616_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_43616_end_mask_0 = const()[name = tensor<string, []>("op_43616_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43616_cast = slice_by_index(begin = var_43616_begin_0, end = var_43616_end_0, end_mask = var_43616_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43616_cast")];
            tensor<int32, [4]> var_43620_begin_0 = const()[name = tensor<string, []>("op_43620_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_43620_end_0 = const()[name = tensor<string, []>("op_43620_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_43620_end_mask_0 = const()[name = tensor<string, []>("op_43620_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43620_cast = slice_by_index(begin = var_43620_begin_0, end = var_43620_end_0, end_mask = var_43620_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43620_cast")];
            tensor<int32, [4]> var_43624_begin_0 = const()[name = tensor<string, []>("op_43624_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_43624_end_0 = const()[name = tensor<string, []>("op_43624_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_43624_end_mask_0 = const()[name = tensor<string, []>("op_43624_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43624_cast = slice_by_index(begin = var_43624_begin_0, end = var_43624_end_0, end_mask = var_43624_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43624_cast")];
            tensor<int32, [4]> var_43628_begin_0 = const()[name = tensor<string, []>("op_43628_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_43628_end_0 = const()[name = tensor<string, []>("op_43628_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_43628_end_mask_0 = const()[name = tensor<string, []>("op_43628_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43628_cast = slice_by_index(begin = var_43628_begin_0, end = var_43628_end_0, end_mask = var_43628_end_mask_0, x = v_193_cast)[name = tensor<string, []>("op_43628_cast")];
            tensor<string, []> var_43632_equation_0 = const()[name = tensor<string, []>("op_43632_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43632_cast = einsum(equation = var_43632_equation_0, values = (var_43474_cast, var_43391_cast))[name = tensor<string, []>("op_43632_cast")];
            tensor<fp16, []> var_43633_to_fp16 = const()[name = tensor<string, []>("op_43633_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3681_cast = mul(x = var_43632_cast, y = var_43633_to_fp16)[name = tensor<string, []>("aw_3681_cast")];
            tensor<string, []> var_43636_equation_0 = const()[name = tensor<string, []>("op_43636_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43636_cast = einsum(equation = var_43636_equation_0, values = (var_43478_cast, var_43395_cast))[name = tensor<string, []>("op_43636_cast")];
            tensor<fp16, []> var_43637_to_fp16 = const()[name = tensor<string, []>("op_43637_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3683_cast = mul(x = var_43636_cast, y = var_43637_to_fp16)[name = tensor<string, []>("aw_3683_cast")];
            tensor<string, []> var_43640_equation_0 = const()[name = tensor<string, []>("op_43640_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43640_cast = einsum(equation = var_43640_equation_0, values = (var_43482_cast, var_43399_cast))[name = tensor<string, []>("op_43640_cast")];
            tensor<fp16, []> var_43641_to_fp16 = const()[name = tensor<string, []>("op_43641_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3685_cast = mul(x = var_43640_cast, y = var_43641_to_fp16)[name = tensor<string, []>("aw_3685_cast")];
            tensor<string, []> var_43644_equation_0 = const()[name = tensor<string, []>("op_43644_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43644_cast = einsum(equation = var_43644_equation_0, values = (var_43486_cast, var_43403_cast))[name = tensor<string, []>("op_43644_cast")];
            tensor<fp16, []> var_43645_to_fp16 = const()[name = tensor<string, []>("op_43645_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3687_cast = mul(x = var_43644_cast, y = var_43645_to_fp16)[name = tensor<string, []>("aw_3687_cast")];
            tensor<string, []> var_43648_equation_0 = const()[name = tensor<string, []>("op_43648_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43648_cast = einsum(equation = var_43648_equation_0, values = (var_43490_cast, var_43407_cast))[name = tensor<string, []>("op_43648_cast")];
            tensor<fp16, []> var_43649_to_fp16 = const()[name = tensor<string, []>("op_43649_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3689_cast = mul(x = var_43648_cast, y = var_43649_to_fp16)[name = tensor<string, []>("aw_3689_cast")];
            tensor<string, []> var_43652_equation_0 = const()[name = tensor<string, []>("op_43652_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43652_cast = einsum(equation = var_43652_equation_0, values = (var_43494_cast, var_43411_cast))[name = tensor<string, []>("op_43652_cast")];
            tensor<fp16, []> var_43653_to_fp16 = const()[name = tensor<string, []>("op_43653_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3691_cast = mul(x = var_43652_cast, y = var_43653_to_fp16)[name = tensor<string, []>("aw_3691_cast")];
            tensor<string, []> var_43656_equation_0 = const()[name = tensor<string, []>("op_43656_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43656_cast = einsum(equation = var_43656_equation_0, values = (var_43498_cast, var_43415_cast))[name = tensor<string, []>("op_43656_cast")];
            tensor<fp16, []> var_43657_to_fp16 = const()[name = tensor<string, []>("op_43657_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3693_cast = mul(x = var_43656_cast, y = var_43657_to_fp16)[name = tensor<string, []>("aw_3693_cast")];
            tensor<string, []> var_43660_equation_0 = const()[name = tensor<string, []>("op_43660_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43660_cast = einsum(equation = var_43660_equation_0, values = (var_43502_cast, var_43419_cast))[name = tensor<string, []>("op_43660_cast")];
            tensor<fp16, []> var_43661_to_fp16 = const()[name = tensor<string, []>("op_43661_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3695_cast = mul(x = var_43660_cast, y = var_43661_to_fp16)[name = tensor<string, []>("aw_3695_cast")];
            tensor<string, []> var_43664_equation_0 = const()[name = tensor<string, []>("op_43664_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43664_cast = einsum(equation = var_43664_equation_0, values = (var_43506_cast, var_43423_cast))[name = tensor<string, []>("op_43664_cast")];
            tensor<fp16, []> var_43665_to_fp16 = const()[name = tensor<string, []>("op_43665_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3697_cast = mul(x = var_43664_cast, y = var_43665_to_fp16)[name = tensor<string, []>("aw_3697_cast")];
            tensor<string, []> var_43668_equation_0 = const()[name = tensor<string, []>("op_43668_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43668_cast = einsum(equation = var_43668_equation_0, values = (var_43510_cast, var_43427_cast))[name = tensor<string, []>("op_43668_cast")];
            tensor<fp16, []> var_43669_to_fp16 = const()[name = tensor<string, []>("op_43669_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3699_cast = mul(x = var_43668_cast, y = var_43669_to_fp16)[name = tensor<string, []>("aw_3699_cast")];
            tensor<string, []> var_43672_equation_0 = const()[name = tensor<string, []>("op_43672_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43672_cast = einsum(equation = var_43672_equation_0, values = (var_43514_cast, var_43431_cast))[name = tensor<string, []>("op_43672_cast")];
            tensor<fp16, []> var_43673_to_fp16 = const()[name = tensor<string, []>("op_43673_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3701_cast = mul(x = var_43672_cast, y = var_43673_to_fp16)[name = tensor<string, []>("aw_3701_cast")];
            tensor<string, []> var_43676_equation_0 = const()[name = tensor<string, []>("op_43676_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43676_cast = einsum(equation = var_43676_equation_0, values = (var_43518_cast, var_43435_cast))[name = tensor<string, []>("op_43676_cast")];
            tensor<fp16, []> var_43677_to_fp16 = const()[name = tensor<string, []>("op_43677_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3703_cast = mul(x = var_43676_cast, y = var_43677_to_fp16)[name = tensor<string, []>("aw_3703_cast")];
            tensor<string, []> var_43680_equation_0 = const()[name = tensor<string, []>("op_43680_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43680_cast = einsum(equation = var_43680_equation_0, values = (var_43522_cast, var_43439_cast))[name = tensor<string, []>("op_43680_cast")];
            tensor<fp16, []> var_43681_to_fp16 = const()[name = tensor<string, []>("op_43681_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3705_cast = mul(x = var_43680_cast, y = var_43681_to_fp16)[name = tensor<string, []>("aw_3705_cast")];
            tensor<string, []> var_43684_equation_0 = const()[name = tensor<string, []>("op_43684_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43684_cast = einsum(equation = var_43684_equation_0, values = (var_43526_cast, var_43443_cast))[name = tensor<string, []>("op_43684_cast")];
            tensor<fp16, []> var_43685_to_fp16 = const()[name = tensor<string, []>("op_43685_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3707_cast = mul(x = var_43684_cast, y = var_43685_to_fp16)[name = tensor<string, []>("aw_3707_cast")];
            tensor<string, []> var_43688_equation_0 = const()[name = tensor<string, []>("op_43688_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43688_cast = einsum(equation = var_43688_equation_0, values = (var_43530_cast, var_43447_cast))[name = tensor<string, []>("op_43688_cast")];
            tensor<fp16, []> var_43689_to_fp16 = const()[name = tensor<string, []>("op_43689_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3709_cast = mul(x = var_43688_cast, y = var_43689_to_fp16)[name = tensor<string, []>("aw_3709_cast")];
            tensor<string, []> var_43692_equation_0 = const()[name = tensor<string, []>("op_43692_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43692_cast = einsum(equation = var_43692_equation_0, values = (var_43534_cast, var_43451_cast))[name = tensor<string, []>("op_43692_cast")];
            tensor<fp16, []> var_43693_to_fp16 = const()[name = tensor<string, []>("op_43693_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3711_cast = mul(x = var_43692_cast, y = var_43693_to_fp16)[name = tensor<string, []>("aw_3711_cast")];
            tensor<string, []> var_43696_equation_0 = const()[name = tensor<string, []>("op_43696_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43696_cast = einsum(equation = var_43696_equation_0, values = (var_43538_cast, var_43455_cast))[name = tensor<string, []>("op_43696_cast")];
            tensor<fp16, []> var_43697_to_fp16 = const()[name = tensor<string, []>("op_43697_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3713_cast = mul(x = var_43696_cast, y = var_43697_to_fp16)[name = tensor<string, []>("aw_3713_cast")];
            tensor<string, []> var_43700_equation_0 = const()[name = tensor<string, []>("op_43700_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43700_cast = einsum(equation = var_43700_equation_0, values = (var_43542_cast, var_43459_cast))[name = tensor<string, []>("op_43700_cast")];
            tensor<fp16, []> var_43701_to_fp16 = const()[name = tensor<string, []>("op_43701_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3715_cast = mul(x = var_43700_cast, y = var_43701_to_fp16)[name = tensor<string, []>("aw_3715_cast")];
            tensor<string, []> var_43704_equation_0 = const()[name = tensor<string, []>("op_43704_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43704_cast = einsum(equation = var_43704_equation_0, values = (var_43546_cast, var_43463_cast))[name = tensor<string, []>("op_43704_cast")];
            tensor<fp16, []> var_43705_to_fp16 = const()[name = tensor<string, []>("op_43705_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3717_cast = mul(x = var_43704_cast, y = var_43705_to_fp16)[name = tensor<string, []>("aw_3717_cast")];
            tensor<string, []> var_43708_equation_0 = const()[name = tensor<string, []>("op_43708_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43708_cast = einsum(equation = var_43708_equation_0, values = (var_43550_cast, var_43467_cast))[name = tensor<string, []>("op_43708_cast")];
            tensor<fp16, []> var_43709_to_fp16 = const()[name = tensor<string, []>("op_43709_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3719_cast = mul(x = var_43708_cast, y = var_43709_to_fp16)[name = tensor<string, []>("aw_3719_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43711_cast = softmax(axis = var_30385, x = aw_3681_cast)[name = tensor<string, []>("op_43711_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43712_cast = softmax(axis = var_30385, x = aw_3683_cast)[name = tensor<string, []>("op_43712_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43713_cast = softmax(axis = var_30385, x = aw_3685_cast)[name = tensor<string, []>("op_43713_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43714_cast = softmax(axis = var_30385, x = aw_3687_cast)[name = tensor<string, []>("op_43714_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43715_cast = softmax(axis = var_30385, x = aw_3689_cast)[name = tensor<string, []>("op_43715_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43716_cast = softmax(axis = var_30385, x = aw_3691_cast)[name = tensor<string, []>("op_43716_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43717_cast = softmax(axis = var_30385, x = aw_3693_cast)[name = tensor<string, []>("op_43717_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43718_cast = softmax(axis = var_30385, x = aw_3695_cast)[name = tensor<string, []>("op_43718_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43719_cast = softmax(axis = var_30385, x = aw_3697_cast)[name = tensor<string, []>("op_43719_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43720_cast = softmax(axis = var_30385, x = aw_3699_cast)[name = tensor<string, []>("op_43720_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43721_cast = softmax(axis = var_30385, x = aw_3701_cast)[name = tensor<string, []>("op_43721_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43722_cast = softmax(axis = var_30385, x = aw_3703_cast)[name = tensor<string, []>("op_43722_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43723_cast = softmax(axis = var_30385, x = aw_3705_cast)[name = tensor<string, []>("op_43723_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43724_cast = softmax(axis = var_30385, x = aw_3707_cast)[name = tensor<string, []>("op_43724_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43725_cast = softmax(axis = var_30385, x = aw_3709_cast)[name = tensor<string, []>("op_43725_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43726_cast = softmax(axis = var_30385, x = aw_3711_cast)[name = tensor<string, []>("op_43726_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43727_cast = softmax(axis = var_30385, x = aw_3713_cast)[name = tensor<string, []>("op_43727_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43728_cast = softmax(axis = var_30385, x = aw_3715_cast)[name = tensor<string, []>("op_43728_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43729_cast = softmax(axis = var_30385, x = aw_3717_cast)[name = tensor<string, []>("op_43729_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_43730_cast = softmax(axis = var_30385, x = aw_3719_cast)[name = tensor<string, []>("op_43730_cast")];
            tensor<string, []> var_43732_equation_0 = const()[name = tensor<string, []>("op_43732_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43732_cast = einsum(equation = var_43732_equation_0, values = (var_43552_cast, var_43711_cast))[name = tensor<string, []>("op_43732_cast")];
            tensor<string, []> var_43734_equation_0 = const()[name = tensor<string, []>("op_43734_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43734_cast = einsum(equation = var_43734_equation_0, values = (var_43556_cast, var_43712_cast))[name = tensor<string, []>("op_43734_cast")];
            tensor<string, []> var_43736_equation_0 = const()[name = tensor<string, []>("op_43736_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43736_cast = einsum(equation = var_43736_equation_0, values = (var_43560_cast, var_43713_cast))[name = tensor<string, []>("op_43736_cast")];
            tensor<string, []> var_43738_equation_0 = const()[name = tensor<string, []>("op_43738_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43738_cast = einsum(equation = var_43738_equation_0, values = (var_43564_cast, var_43714_cast))[name = tensor<string, []>("op_43738_cast")];
            tensor<string, []> var_43740_equation_0 = const()[name = tensor<string, []>("op_43740_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43740_cast = einsum(equation = var_43740_equation_0, values = (var_43568_cast, var_43715_cast))[name = tensor<string, []>("op_43740_cast")];
            tensor<string, []> var_43742_equation_0 = const()[name = tensor<string, []>("op_43742_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43742_cast = einsum(equation = var_43742_equation_0, values = (var_43572_cast, var_43716_cast))[name = tensor<string, []>("op_43742_cast")];
            tensor<string, []> var_43744_equation_0 = const()[name = tensor<string, []>("op_43744_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43744_cast = einsum(equation = var_43744_equation_0, values = (var_43576_cast, var_43717_cast))[name = tensor<string, []>("op_43744_cast")];
            tensor<string, []> var_43746_equation_0 = const()[name = tensor<string, []>("op_43746_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43746_cast = einsum(equation = var_43746_equation_0, values = (var_43580_cast, var_43718_cast))[name = tensor<string, []>("op_43746_cast")];
            tensor<string, []> var_43748_equation_0 = const()[name = tensor<string, []>("op_43748_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43748_cast = einsum(equation = var_43748_equation_0, values = (var_43584_cast, var_43719_cast))[name = tensor<string, []>("op_43748_cast")];
            tensor<string, []> var_43750_equation_0 = const()[name = tensor<string, []>("op_43750_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43750_cast = einsum(equation = var_43750_equation_0, values = (var_43588_cast, var_43720_cast))[name = tensor<string, []>("op_43750_cast")];
            tensor<string, []> var_43752_equation_0 = const()[name = tensor<string, []>("op_43752_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43752_cast = einsum(equation = var_43752_equation_0, values = (var_43592_cast, var_43721_cast))[name = tensor<string, []>("op_43752_cast")];
            tensor<string, []> var_43754_equation_0 = const()[name = tensor<string, []>("op_43754_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43754_cast = einsum(equation = var_43754_equation_0, values = (var_43596_cast, var_43722_cast))[name = tensor<string, []>("op_43754_cast")];
            tensor<string, []> var_43756_equation_0 = const()[name = tensor<string, []>("op_43756_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43756_cast = einsum(equation = var_43756_equation_0, values = (var_43600_cast, var_43723_cast))[name = tensor<string, []>("op_43756_cast")];
            tensor<string, []> var_43758_equation_0 = const()[name = tensor<string, []>("op_43758_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43758_cast = einsum(equation = var_43758_equation_0, values = (var_43604_cast, var_43724_cast))[name = tensor<string, []>("op_43758_cast")];
            tensor<string, []> var_43760_equation_0 = const()[name = tensor<string, []>("op_43760_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43760_cast = einsum(equation = var_43760_equation_0, values = (var_43608_cast, var_43725_cast))[name = tensor<string, []>("op_43760_cast")];
            tensor<string, []> var_43762_equation_0 = const()[name = tensor<string, []>("op_43762_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43762_cast = einsum(equation = var_43762_equation_0, values = (var_43612_cast, var_43726_cast))[name = tensor<string, []>("op_43762_cast")];
            tensor<string, []> var_43764_equation_0 = const()[name = tensor<string, []>("op_43764_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43764_cast = einsum(equation = var_43764_equation_0, values = (var_43616_cast, var_43727_cast))[name = tensor<string, []>("op_43764_cast")];
            tensor<string, []> var_43766_equation_0 = const()[name = tensor<string, []>("op_43766_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43766_cast = einsum(equation = var_43766_equation_0, values = (var_43620_cast, var_43728_cast))[name = tensor<string, []>("op_43766_cast")];
            tensor<string, []> var_43768_equation_0 = const()[name = tensor<string, []>("op_43768_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43768_cast = einsum(equation = var_43768_equation_0, values = (var_43624_cast, var_43729_cast))[name = tensor<string, []>("op_43768_cast")];
            tensor<string, []> var_43770_equation_0 = const()[name = tensor<string, []>("op_43770_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_43770_cast = einsum(equation = var_43770_equation_0, values = (var_43628_cast, var_43730_cast))[name = tensor<string, []>("op_43770_cast")];
            tensor<bool, []> input_571_interleave_0 = const()[name = tensor<string, []>("input_571_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_571_cast = concat(axis = var_30385, interleave = input_571_interleave_0, values = (var_43732_cast, var_43734_cast, var_43736_cast, var_43738_cast, var_43740_cast, var_43742_cast, var_43744_cast, var_43746_cast, var_43748_cast, var_43750_cast, var_43752_cast, var_43754_cast, var_43756_cast, var_43758_cast, var_43760_cast, var_43762_cast, var_43764_cast, var_43766_cast, var_43768_cast, var_43770_cast))[name = tensor<string, []>("input_571_cast")];
            tensor<int32, [2]> var_43776 = const()[name = tensor<string, []>("op_43776"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_43778 = const()[name = tensor<string, []>("op_43778"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_43780_pad_type_0 = const()[name = tensor<string, []>("op_43780_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_43780_pad_0 = const()[name = tensor<string, []>("op_43780_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3676333248)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3679610112)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_43780_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_out_0_bias_to_fp16, dilations = var_43778, groups = var_30385, pad = var_43780_pad_0, pad_type = var_43780_pad_type_0, strides = var_43776, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn1_to_out_0_weight_to_fp16, x = input_571_cast)[name = tensor<string, []>("op_43780_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_291_cast = add(x = var_43780_cast, y = inputs_289_cast)[name = tensor<string, []>("inputs_291_cast")];
            tensor<int32, [1]> var_43784 = const()[name = tensor<string, []>("op_43784"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_291_cast = reduce_mean(axes = var_43784, keep_dims = var_30380, x = inputs_291_cast)[name = tensor<string, []>("channels_mean_291_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_291_cast = sub(x = inputs_291_cast, y = channels_mean_291_cast)[name = tensor<string, []>("zero_mean_291_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_291_cast = mul(x = zero_mean_291_cast, y = zero_mean_291_cast)[name = tensor<string, []>("zero_mean_sq_291_cast")];
            tensor<int32, [1]> var_43788 = const()[name = tensor<string, []>("op_43788"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_43789_cast = reduce_mean(axes = var_43788, keep_dims = var_30380, x = zero_mean_sq_291_cast)[name = tensor<string, []>("op_43789_cast")];
            tensor<fp16, []> var_43790_to_fp16 = const()[name = tensor<string, []>("op_43790_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_43791_cast = add(x = var_43789_cast, y = var_43790_to_fp16)[name = tensor<string, []>("op_43791_cast")];
            tensor<fp16, []> denom_291_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_291_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_291_cast = rsqrt(epsilon = denom_291_epsilon_0_to_fp16, x = var_43791_cast)[name = tensor<string, []>("denom_291_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_291_cast = mul(x = zero_mean_291_cast, y = denom_291_cast)[name = tensor<string, []>("out_291_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_43795_to_fp16 = const()[name = tensor<string, []>("op_43795_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3679612736)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_43796_cast = add(x = out_291_cast, y = var_43795_to_fp16)[name = tensor<string, []>("op_43796_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_43798_to_fp16 = const()[name = tensor<string, []>("op_43798_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3679615360)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_387_cast = mul(x = var_43796_cast, y = var_43798_to_fp16)[name = tensor<string, []>("hidden_states_387_cast")];
            tensor<int32, [2]> var_43805 = const()[name = tensor<string, []>("op_43805"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_43807 = const()[name = tensor<string, []>("op_43807"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_195_pad_type_0 = const()[name = tensor<string, []>("q_195_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_195_pad_0 = const()[name = tensor<string, []>("q_195_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3679617984)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_195_cast = conv(dilations = var_43807, groups = var_30385, pad = q_195_pad_0, pad_type = q_195_pad_type_0, strides = var_43805, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_q_weight_to_fp16, x = hidden_states_387_cast)[name = tensor<string, []>("q_195_cast")];
            tensor<int32, [2]> var_43811 = const()[name = tensor<string, []>("op_43811"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_43813 = const()[name = tensor<string, []>("op_43813"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_389_pad_type_0 = const()[name = tensor<string, []>("k_389_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_389_pad_0 = const()[name = tensor<string, []>("k_389_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3682894848)))];
            tensor<fp16, [2, 1280, 1, 77]> k_389_cast = conv(dilations = var_43813, groups = var_30385, pad = k_389_pad_0, pad_type = k_389_pad_type_0, strides = var_43811, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_389_cast")];
            tensor<int32, [2]> var_43817 = const()[name = tensor<string, []>("op_43817"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_43819 = const()[name = tensor<string, []>("op_43819"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_195_pad_type_0 = const()[name = tensor<string, []>("v_195_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_195_pad_0 = const()[name = tensor<string, []>("v_195_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3688137792)))];
            tensor<fp16, [2, 1280, 1, 77]> v_195_cast = conv(dilations = var_43819, groups = var_30385, pad = v_195_pad_0, pad_type = v_195_pad_type_0, strides = var_43817, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_195_cast")];
            tensor<int32, [4]> var_43823_begin_0 = const()[name = tensor<string, []>("op_43823_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_43823_end_0 = const()[name = tensor<string, []>("op_43823_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_43823_end_mask_0 = const()[name = tensor<string, []>("op_43823_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43823_cast = slice_by_index(begin = var_43823_begin_0, end = var_43823_end_0, end_mask = var_43823_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43823_cast")];
            tensor<int32, [4]> var_43827_begin_0 = const()[name = tensor<string, []>("op_43827_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_43827_end_0 = const()[name = tensor<string, []>("op_43827_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_43827_end_mask_0 = const()[name = tensor<string, []>("op_43827_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43827_cast = slice_by_index(begin = var_43827_begin_0, end = var_43827_end_0, end_mask = var_43827_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43827_cast")];
            tensor<int32, [4]> var_43831_begin_0 = const()[name = tensor<string, []>("op_43831_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_43831_end_0 = const()[name = tensor<string, []>("op_43831_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_43831_end_mask_0 = const()[name = tensor<string, []>("op_43831_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43831_cast = slice_by_index(begin = var_43831_begin_0, end = var_43831_end_0, end_mask = var_43831_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43831_cast")];
            tensor<int32, [4]> var_43835_begin_0 = const()[name = tensor<string, []>("op_43835_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_43835_end_0 = const()[name = tensor<string, []>("op_43835_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_43835_end_mask_0 = const()[name = tensor<string, []>("op_43835_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43835_cast = slice_by_index(begin = var_43835_begin_0, end = var_43835_end_0, end_mask = var_43835_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43835_cast")];
            tensor<int32, [4]> var_43839_begin_0 = const()[name = tensor<string, []>("op_43839_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_43839_end_0 = const()[name = tensor<string, []>("op_43839_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_43839_end_mask_0 = const()[name = tensor<string, []>("op_43839_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43839_cast = slice_by_index(begin = var_43839_begin_0, end = var_43839_end_0, end_mask = var_43839_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43839_cast")];
            tensor<int32, [4]> var_43843_begin_0 = const()[name = tensor<string, []>("op_43843_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_43843_end_0 = const()[name = tensor<string, []>("op_43843_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_43843_end_mask_0 = const()[name = tensor<string, []>("op_43843_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43843_cast = slice_by_index(begin = var_43843_begin_0, end = var_43843_end_0, end_mask = var_43843_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43843_cast")];
            tensor<int32, [4]> var_43847_begin_0 = const()[name = tensor<string, []>("op_43847_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_43847_end_0 = const()[name = tensor<string, []>("op_43847_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_43847_end_mask_0 = const()[name = tensor<string, []>("op_43847_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43847_cast = slice_by_index(begin = var_43847_begin_0, end = var_43847_end_0, end_mask = var_43847_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43847_cast")];
            tensor<int32, [4]> var_43851_begin_0 = const()[name = tensor<string, []>("op_43851_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_43851_end_0 = const()[name = tensor<string, []>("op_43851_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_43851_end_mask_0 = const()[name = tensor<string, []>("op_43851_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43851_cast = slice_by_index(begin = var_43851_begin_0, end = var_43851_end_0, end_mask = var_43851_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43851_cast")];
            tensor<int32, [4]> var_43855_begin_0 = const()[name = tensor<string, []>("op_43855_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_43855_end_0 = const()[name = tensor<string, []>("op_43855_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_43855_end_mask_0 = const()[name = tensor<string, []>("op_43855_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43855_cast = slice_by_index(begin = var_43855_begin_0, end = var_43855_end_0, end_mask = var_43855_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43855_cast")];
            tensor<int32, [4]> var_43859_begin_0 = const()[name = tensor<string, []>("op_43859_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_43859_end_0 = const()[name = tensor<string, []>("op_43859_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_43859_end_mask_0 = const()[name = tensor<string, []>("op_43859_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43859_cast = slice_by_index(begin = var_43859_begin_0, end = var_43859_end_0, end_mask = var_43859_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43859_cast")];
            tensor<int32, [4]> var_43863_begin_0 = const()[name = tensor<string, []>("op_43863_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_43863_end_0 = const()[name = tensor<string, []>("op_43863_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_43863_end_mask_0 = const()[name = tensor<string, []>("op_43863_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43863_cast = slice_by_index(begin = var_43863_begin_0, end = var_43863_end_0, end_mask = var_43863_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43863_cast")];
            tensor<int32, [4]> var_43867_begin_0 = const()[name = tensor<string, []>("op_43867_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_43867_end_0 = const()[name = tensor<string, []>("op_43867_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_43867_end_mask_0 = const()[name = tensor<string, []>("op_43867_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43867_cast = slice_by_index(begin = var_43867_begin_0, end = var_43867_end_0, end_mask = var_43867_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43867_cast")];
            tensor<int32, [4]> var_43871_begin_0 = const()[name = tensor<string, []>("op_43871_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_43871_end_0 = const()[name = tensor<string, []>("op_43871_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_43871_end_mask_0 = const()[name = tensor<string, []>("op_43871_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43871_cast = slice_by_index(begin = var_43871_begin_0, end = var_43871_end_0, end_mask = var_43871_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43871_cast")];
            tensor<int32, [4]> var_43875_begin_0 = const()[name = tensor<string, []>("op_43875_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_43875_end_0 = const()[name = tensor<string, []>("op_43875_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_43875_end_mask_0 = const()[name = tensor<string, []>("op_43875_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43875_cast = slice_by_index(begin = var_43875_begin_0, end = var_43875_end_0, end_mask = var_43875_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43875_cast")];
            tensor<int32, [4]> var_43879_begin_0 = const()[name = tensor<string, []>("op_43879_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_43879_end_0 = const()[name = tensor<string, []>("op_43879_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_43879_end_mask_0 = const()[name = tensor<string, []>("op_43879_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43879_cast = slice_by_index(begin = var_43879_begin_0, end = var_43879_end_0, end_mask = var_43879_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43879_cast")];
            tensor<int32, [4]> var_43883_begin_0 = const()[name = tensor<string, []>("op_43883_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_43883_end_0 = const()[name = tensor<string, []>("op_43883_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_43883_end_mask_0 = const()[name = tensor<string, []>("op_43883_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43883_cast = slice_by_index(begin = var_43883_begin_0, end = var_43883_end_0, end_mask = var_43883_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43883_cast")];
            tensor<int32, [4]> var_43887_begin_0 = const()[name = tensor<string, []>("op_43887_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_43887_end_0 = const()[name = tensor<string, []>("op_43887_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_43887_end_mask_0 = const()[name = tensor<string, []>("op_43887_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43887_cast = slice_by_index(begin = var_43887_begin_0, end = var_43887_end_0, end_mask = var_43887_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43887_cast")];
            tensor<int32, [4]> var_43891_begin_0 = const()[name = tensor<string, []>("op_43891_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_43891_end_0 = const()[name = tensor<string, []>("op_43891_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_43891_end_mask_0 = const()[name = tensor<string, []>("op_43891_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43891_cast = slice_by_index(begin = var_43891_begin_0, end = var_43891_end_0, end_mask = var_43891_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43891_cast")];
            tensor<int32, [4]> var_43895_begin_0 = const()[name = tensor<string, []>("op_43895_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_43895_end_0 = const()[name = tensor<string, []>("op_43895_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_43895_end_mask_0 = const()[name = tensor<string, []>("op_43895_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43895_cast = slice_by_index(begin = var_43895_begin_0, end = var_43895_end_0, end_mask = var_43895_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43895_cast")];
            tensor<int32, [4]> var_43899_begin_0 = const()[name = tensor<string, []>("op_43899_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_43899_end_0 = const()[name = tensor<string, []>("op_43899_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_43899_end_mask_0 = const()[name = tensor<string, []>("op_43899_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_43899_cast = slice_by_index(begin = var_43899_begin_0, end = var_43899_end_0, end_mask = var_43899_end_mask_0, x = q_195_cast)[name = tensor<string, []>("op_43899_cast")];
            tensor<int32, [4]> k_391_perm_0 = const()[name = tensor<string, []>("k_391_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_43906_begin_0 = const()[name = tensor<string, []>("op_43906_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_43906_end_0 = const()[name = tensor<string, []>("op_43906_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_43906_end_mask_0 = const()[name = tensor<string, []>("op_43906_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_42 = transpose(perm = k_391_perm_0, x = k_389_cast)[name = tensor<string, []>("transpose_42")];
            tensor<fp16, [2, 77, 1, 64]> var_43906_cast = slice_by_index(begin = var_43906_begin_0, end = var_43906_end_0, end_mask = var_43906_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43906_cast")];
            tensor<int32, [4]> var_43910_begin_0 = const()[name = tensor<string, []>("op_43910_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_43910_end_0 = const()[name = tensor<string, []>("op_43910_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_43910_end_mask_0 = const()[name = tensor<string, []>("op_43910_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43910_cast = slice_by_index(begin = var_43910_begin_0, end = var_43910_end_0, end_mask = var_43910_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43910_cast")];
            tensor<int32, [4]> var_43914_begin_0 = const()[name = tensor<string, []>("op_43914_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_43914_end_0 = const()[name = tensor<string, []>("op_43914_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_43914_end_mask_0 = const()[name = tensor<string, []>("op_43914_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43914_cast = slice_by_index(begin = var_43914_begin_0, end = var_43914_end_0, end_mask = var_43914_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43914_cast")];
            tensor<int32, [4]> var_43918_begin_0 = const()[name = tensor<string, []>("op_43918_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_43918_end_0 = const()[name = tensor<string, []>("op_43918_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_43918_end_mask_0 = const()[name = tensor<string, []>("op_43918_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43918_cast = slice_by_index(begin = var_43918_begin_0, end = var_43918_end_0, end_mask = var_43918_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43918_cast")];
            tensor<int32, [4]> var_43922_begin_0 = const()[name = tensor<string, []>("op_43922_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_43922_end_0 = const()[name = tensor<string, []>("op_43922_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_43922_end_mask_0 = const()[name = tensor<string, []>("op_43922_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43922_cast = slice_by_index(begin = var_43922_begin_0, end = var_43922_end_0, end_mask = var_43922_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43922_cast")];
            tensor<int32, [4]> var_43926_begin_0 = const()[name = tensor<string, []>("op_43926_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_43926_end_0 = const()[name = tensor<string, []>("op_43926_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_43926_end_mask_0 = const()[name = tensor<string, []>("op_43926_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43926_cast = slice_by_index(begin = var_43926_begin_0, end = var_43926_end_0, end_mask = var_43926_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43926_cast")];
            tensor<int32, [4]> var_43930_begin_0 = const()[name = tensor<string, []>("op_43930_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_43930_end_0 = const()[name = tensor<string, []>("op_43930_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_43930_end_mask_0 = const()[name = tensor<string, []>("op_43930_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43930_cast = slice_by_index(begin = var_43930_begin_0, end = var_43930_end_0, end_mask = var_43930_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43930_cast")];
            tensor<int32, [4]> var_43934_begin_0 = const()[name = tensor<string, []>("op_43934_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_43934_end_0 = const()[name = tensor<string, []>("op_43934_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_43934_end_mask_0 = const()[name = tensor<string, []>("op_43934_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43934_cast = slice_by_index(begin = var_43934_begin_0, end = var_43934_end_0, end_mask = var_43934_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43934_cast")];
            tensor<int32, [4]> var_43938_begin_0 = const()[name = tensor<string, []>("op_43938_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_43938_end_0 = const()[name = tensor<string, []>("op_43938_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_43938_end_mask_0 = const()[name = tensor<string, []>("op_43938_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43938_cast = slice_by_index(begin = var_43938_begin_0, end = var_43938_end_0, end_mask = var_43938_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43938_cast")];
            tensor<int32, [4]> var_43942_begin_0 = const()[name = tensor<string, []>("op_43942_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_43942_end_0 = const()[name = tensor<string, []>("op_43942_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_43942_end_mask_0 = const()[name = tensor<string, []>("op_43942_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43942_cast = slice_by_index(begin = var_43942_begin_0, end = var_43942_end_0, end_mask = var_43942_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43942_cast")];
            tensor<int32, [4]> var_43946_begin_0 = const()[name = tensor<string, []>("op_43946_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_43946_end_0 = const()[name = tensor<string, []>("op_43946_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_43946_end_mask_0 = const()[name = tensor<string, []>("op_43946_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43946_cast = slice_by_index(begin = var_43946_begin_0, end = var_43946_end_0, end_mask = var_43946_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43946_cast")];
            tensor<int32, [4]> var_43950_begin_0 = const()[name = tensor<string, []>("op_43950_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_43950_end_0 = const()[name = tensor<string, []>("op_43950_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_43950_end_mask_0 = const()[name = tensor<string, []>("op_43950_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43950_cast = slice_by_index(begin = var_43950_begin_0, end = var_43950_end_0, end_mask = var_43950_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43950_cast")];
            tensor<int32, [4]> var_43954_begin_0 = const()[name = tensor<string, []>("op_43954_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_43954_end_0 = const()[name = tensor<string, []>("op_43954_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_43954_end_mask_0 = const()[name = tensor<string, []>("op_43954_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43954_cast = slice_by_index(begin = var_43954_begin_0, end = var_43954_end_0, end_mask = var_43954_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43954_cast")];
            tensor<int32, [4]> var_43958_begin_0 = const()[name = tensor<string, []>("op_43958_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_43958_end_0 = const()[name = tensor<string, []>("op_43958_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_43958_end_mask_0 = const()[name = tensor<string, []>("op_43958_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43958_cast = slice_by_index(begin = var_43958_begin_0, end = var_43958_end_0, end_mask = var_43958_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43958_cast")];
            tensor<int32, [4]> var_43962_begin_0 = const()[name = tensor<string, []>("op_43962_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_43962_end_0 = const()[name = tensor<string, []>("op_43962_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_43962_end_mask_0 = const()[name = tensor<string, []>("op_43962_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43962_cast = slice_by_index(begin = var_43962_begin_0, end = var_43962_end_0, end_mask = var_43962_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43962_cast")];
            tensor<int32, [4]> var_43966_begin_0 = const()[name = tensor<string, []>("op_43966_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_43966_end_0 = const()[name = tensor<string, []>("op_43966_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_43966_end_mask_0 = const()[name = tensor<string, []>("op_43966_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43966_cast = slice_by_index(begin = var_43966_begin_0, end = var_43966_end_0, end_mask = var_43966_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43966_cast")];
            tensor<int32, [4]> var_43970_begin_0 = const()[name = tensor<string, []>("op_43970_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_43970_end_0 = const()[name = tensor<string, []>("op_43970_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_43970_end_mask_0 = const()[name = tensor<string, []>("op_43970_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43970_cast = slice_by_index(begin = var_43970_begin_0, end = var_43970_end_0, end_mask = var_43970_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43970_cast")];
            tensor<int32, [4]> var_43974_begin_0 = const()[name = tensor<string, []>("op_43974_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_43974_end_0 = const()[name = tensor<string, []>("op_43974_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_43974_end_mask_0 = const()[name = tensor<string, []>("op_43974_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43974_cast = slice_by_index(begin = var_43974_begin_0, end = var_43974_end_0, end_mask = var_43974_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43974_cast")];
            tensor<int32, [4]> var_43978_begin_0 = const()[name = tensor<string, []>("op_43978_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_43978_end_0 = const()[name = tensor<string, []>("op_43978_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_43978_end_mask_0 = const()[name = tensor<string, []>("op_43978_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43978_cast = slice_by_index(begin = var_43978_begin_0, end = var_43978_end_0, end_mask = var_43978_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43978_cast")];
            tensor<int32, [4]> var_43982_begin_0 = const()[name = tensor<string, []>("op_43982_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_43982_end_0 = const()[name = tensor<string, []>("op_43982_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_43982_end_mask_0 = const()[name = tensor<string, []>("op_43982_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_43982_cast = slice_by_index(begin = var_43982_begin_0, end = var_43982_end_0, end_mask = var_43982_end_mask_0, x = transpose_42)[name = tensor<string, []>("op_43982_cast")];
            tensor<int32, [4]> var_43984_begin_0 = const()[name = tensor<string, []>("op_43984_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_43984_end_0 = const()[name = tensor<string, []>("op_43984_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_43984_end_mask_0 = const()[name = tensor<string, []>("op_43984_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43984_cast = slice_by_index(begin = var_43984_begin_0, end = var_43984_end_0, end_mask = var_43984_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_43984_cast")];
            tensor<int32, [4]> var_43988_begin_0 = const()[name = tensor<string, []>("op_43988_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_43988_end_0 = const()[name = tensor<string, []>("op_43988_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_43988_end_mask_0 = const()[name = tensor<string, []>("op_43988_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43988_cast = slice_by_index(begin = var_43988_begin_0, end = var_43988_end_0, end_mask = var_43988_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_43988_cast")];
            tensor<int32, [4]> var_43992_begin_0 = const()[name = tensor<string, []>("op_43992_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_43992_end_0 = const()[name = tensor<string, []>("op_43992_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_43992_end_mask_0 = const()[name = tensor<string, []>("op_43992_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43992_cast = slice_by_index(begin = var_43992_begin_0, end = var_43992_end_0, end_mask = var_43992_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_43992_cast")];
            tensor<int32, [4]> var_43996_begin_0 = const()[name = tensor<string, []>("op_43996_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_43996_end_0 = const()[name = tensor<string, []>("op_43996_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_43996_end_mask_0 = const()[name = tensor<string, []>("op_43996_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_43996_cast = slice_by_index(begin = var_43996_begin_0, end = var_43996_end_0, end_mask = var_43996_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_43996_cast")];
            tensor<int32, [4]> var_44000_begin_0 = const()[name = tensor<string, []>("op_44000_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_44000_end_0 = const()[name = tensor<string, []>("op_44000_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_44000_end_mask_0 = const()[name = tensor<string, []>("op_44000_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44000_cast = slice_by_index(begin = var_44000_begin_0, end = var_44000_end_0, end_mask = var_44000_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_44000_cast")];
            tensor<int32, [4]> var_44004_begin_0 = const()[name = tensor<string, []>("op_44004_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_44004_end_0 = const()[name = tensor<string, []>("op_44004_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_44004_end_mask_0 = const()[name = tensor<string, []>("op_44004_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44004_cast = slice_by_index(begin = var_44004_begin_0, end = var_44004_end_0, end_mask = var_44004_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_44004_cast")];
            tensor<int32, [4]> var_44008_begin_0 = const()[name = tensor<string, []>("op_44008_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_44008_end_0 = const()[name = tensor<string, []>("op_44008_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_44008_end_mask_0 = const()[name = tensor<string, []>("op_44008_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44008_cast = slice_by_index(begin = var_44008_begin_0, end = var_44008_end_0, end_mask = var_44008_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_44008_cast")];
            tensor<int32, [4]> var_44012_begin_0 = const()[name = tensor<string, []>("op_44012_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_44012_end_0 = const()[name = tensor<string, []>("op_44012_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_44012_end_mask_0 = const()[name = tensor<string, []>("op_44012_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44012_cast = slice_by_index(begin = var_44012_begin_0, end = var_44012_end_0, end_mask = var_44012_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_44012_cast")];
            tensor<int32, [4]> var_44016_begin_0 = const()[name = tensor<string, []>("op_44016_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_44016_end_0 = const()[name = tensor<string, []>("op_44016_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_44016_end_mask_0 = const()[name = tensor<string, []>("op_44016_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44016_cast = slice_by_index(begin = var_44016_begin_0, end = var_44016_end_0, end_mask = var_44016_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_44016_cast")];
            tensor<int32, [4]> var_44020_begin_0 = const()[name = tensor<string, []>("op_44020_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_44020_end_0 = const()[name = tensor<string, []>("op_44020_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_44020_end_mask_0 = const()[name = tensor<string, []>("op_44020_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44020_cast = slice_by_index(begin = var_44020_begin_0, end = var_44020_end_0, end_mask = var_44020_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_44020_cast")];
            tensor<int32, [4]> var_44024_begin_0 = const()[name = tensor<string, []>("op_44024_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_44024_end_0 = const()[name = tensor<string, []>("op_44024_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_44024_end_mask_0 = const()[name = tensor<string, []>("op_44024_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44024_cast = slice_by_index(begin = var_44024_begin_0, end = var_44024_end_0, end_mask = var_44024_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_44024_cast")];
            tensor<int32, [4]> var_44028_begin_0 = const()[name = tensor<string, []>("op_44028_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_44028_end_0 = const()[name = tensor<string, []>("op_44028_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_44028_end_mask_0 = const()[name = tensor<string, []>("op_44028_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44028_cast = slice_by_index(begin = var_44028_begin_0, end = var_44028_end_0, end_mask = var_44028_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_44028_cast")];
            tensor<int32, [4]> var_44032_begin_0 = const()[name = tensor<string, []>("op_44032_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_44032_end_0 = const()[name = tensor<string, []>("op_44032_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_44032_end_mask_0 = const()[name = tensor<string, []>("op_44032_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44032_cast = slice_by_index(begin = var_44032_begin_0, end = var_44032_end_0, end_mask = var_44032_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_44032_cast")];
            tensor<int32, [4]> var_44036_begin_0 = const()[name = tensor<string, []>("op_44036_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_44036_end_0 = const()[name = tensor<string, []>("op_44036_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_44036_end_mask_0 = const()[name = tensor<string, []>("op_44036_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44036_cast = slice_by_index(begin = var_44036_begin_0, end = var_44036_end_0, end_mask = var_44036_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_44036_cast")];
            tensor<int32, [4]> var_44040_begin_0 = const()[name = tensor<string, []>("op_44040_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_44040_end_0 = const()[name = tensor<string, []>("op_44040_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_44040_end_mask_0 = const()[name = tensor<string, []>("op_44040_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44040_cast = slice_by_index(begin = var_44040_begin_0, end = var_44040_end_0, end_mask = var_44040_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_44040_cast")];
            tensor<int32, [4]> var_44044_begin_0 = const()[name = tensor<string, []>("op_44044_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_44044_end_0 = const()[name = tensor<string, []>("op_44044_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_44044_end_mask_0 = const()[name = tensor<string, []>("op_44044_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44044_cast = slice_by_index(begin = var_44044_begin_0, end = var_44044_end_0, end_mask = var_44044_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_44044_cast")];
            tensor<int32, [4]> var_44048_begin_0 = const()[name = tensor<string, []>("op_44048_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_44048_end_0 = const()[name = tensor<string, []>("op_44048_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_44048_end_mask_0 = const()[name = tensor<string, []>("op_44048_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44048_cast = slice_by_index(begin = var_44048_begin_0, end = var_44048_end_0, end_mask = var_44048_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_44048_cast")];
            tensor<int32, [4]> var_44052_begin_0 = const()[name = tensor<string, []>("op_44052_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_44052_end_0 = const()[name = tensor<string, []>("op_44052_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_44052_end_mask_0 = const()[name = tensor<string, []>("op_44052_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44052_cast = slice_by_index(begin = var_44052_begin_0, end = var_44052_end_0, end_mask = var_44052_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_44052_cast")];
            tensor<int32, [4]> var_44056_begin_0 = const()[name = tensor<string, []>("op_44056_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_44056_end_0 = const()[name = tensor<string, []>("op_44056_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_44056_end_mask_0 = const()[name = tensor<string, []>("op_44056_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44056_cast = slice_by_index(begin = var_44056_begin_0, end = var_44056_end_0, end_mask = var_44056_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_44056_cast")];
            tensor<int32, [4]> var_44060_begin_0 = const()[name = tensor<string, []>("op_44060_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_44060_end_0 = const()[name = tensor<string, []>("op_44060_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_44060_end_mask_0 = const()[name = tensor<string, []>("op_44060_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44060_cast = slice_by_index(begin = var_44060_begin_0, end = var_44060_end_0, end_mask = var_44060_end_mask_0, x = v_195_cast)[name = tensor<string, []>("op_44060_cast")];
            tensor<string, []> var_44064_equation_0 = const()[name = tensor<string, []>("op_44064_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44064_cast = einsum(equation = var_44064_equation_0, values = (var_43906_cast, var_43823_cast))[name = tensor<string, []>("op_44064_cast")];
            tensor<fp16, []> var_44065_to_fp16 = const()[name = tensor<string, []>("op_44065_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3721_cast = mul(x = var_44064_cast, y = var_44065_to_fp16)[name = tensor<string, []>("aw_3721_cast")];
            tensor<string, []> var_44068_equation_0 = const()[name = tensor<string, []>("op_44068_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44068_cast = einsum(equation = var_44068_equation_0, values = (var_43910_cast, var_43827_cast))[name = tensor<string, []>("op_44068_cast")];
            tensor<fp16, []> var_44069_to_fp16 = const()[name = tensor<string, []>("op_44069_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3723_cast = mul(x = var_44068_cast, y = var_44069_to_fp16)[name = tensor<string, []>("aw_3723_cast")];
            tensor<string, []> var_44072_equation_0 = const()[name = tensor<string, []>("op_44072_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44072_cast = einsum(equation = var_44072_equation_0, values = (var_43914_cast, var_43831_cast))[name = tensor<string, []>("op_44072_cast")];
            tensor<fp16, []> var_44073_to_fp16 = const()[name = tensor<string, []>("op_44073_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3725_cast = mul(x = var_44072_cast, y = var_44073_to_fp16)[name = tensor<string, []>("aw_3725_cast")];
            tensor<string, []> var_44076_equation_0 = const()[name = tensor<string, []>("op_44076_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44076_cast = einsum(equation = var_44076_equation_0, values = (var_43918_cast, var_43835_cast))[name = tensor<string, []>("op_44076_cast")];
            tensor<fp16, []> var_44077_to_fp16 = const()[name = tensor<string, []>("op_44077_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3727_cast = mul(x = var_44076_cast, y = var_44077_to_fp16)[name = tensor<string, []>("aw_3727_cast")];
            tensor<string, []> var_44080_equation_0 = const()[name = tensor<string, []>("op_44080_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44080_cast = einsum(equation = var_44080_equation_0, values = (var_43922_cast, var_43839_cast))[name = tensor<string, []>("op_44080_cast")];
            tensor<fp16, []> var_44081_to_fp16 = const()[name = tensor<string, []>("op_44081_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3729_cast = mul(x = var_44080_cast, y = var_44081_to_fp16)[name = tensor<string, []>("aw_3729_cast")];
            tensor<string, []> var_44084_equation_0 = const()[name = tensor<string, []>("op_44084_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44084_cast = einsum(equation = var_44084_equation_0, values = (var_43926_cast, var_43843_cast))[name = tensor<string, []>("op_44084_cast")];
            tensor<fp16, []> var_44085_to_fp16 = const()[name = tensor<string, []>("op_44085_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3731_cast = mul(x = var_44084_cast, y = var_44085_to_fp16)[name = tensor<string, []>("aw_3731_cast")];
            tensor<string, []> var_44088_equation_0 = const()[name = tensor<string, []>("op_44088_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44088_cast = einsum(equation = var_44088_equation_0, values = (var_43930_cast, var_43847_cast))[name = tensor<string, []>("op_44088_cast")];
            tensor<fp16, []> var_44089_to_fp16 = const()[name = tensor<string, []>("op_44089_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3733_cast = mul(x = var_44088_cast, y = var_44089_to_fp16)[name = tensor<string, []>("aw_3733_cast")];
            tensor<string, []> var_44092_equation_0 = const()[name = tensor<string, []>("op_44092_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44092_cast = einsum(equation = var_44092_equation_0, values = (var_43934_cast, var_43851_cast))[name = tensor<string, []>("op_44092_cast")];
            tensor<fp16, []> var_44093_to_fp16 = const()[name = tensor<string, []>("op_44093_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3735_cast = mul(x = var_44092_cast, y = var_44093_to_fp16)[name = tensor<string, []>("aw_3735_cast")];
            tensor<string, []> var_44096_equation_0 = const()[name = tensor<string, []>("op_44096_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44096_cast = einsum(equation = var_44096_equation_0, values = (var_43938_cast, var_43855_cast))[name = tensor<string, []>("op_44096_cast")];
            tensor<fp16, []> var_44097_to_fp16 = const()[name = tensor<string, []>("op_44097_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3737_cast = mul(x = var_44096_cast, y = var_44097_to_fp16)[name = tensor<string, []>("aw_3737_cast")];
            tensor<string, []> var_44100_equation_0 = const()[name = tensor<string, []>("op_44100_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44100_cast = einsum(equation = var_44100_equation_0, values = (var_43942_cast, var_43859_cast))[name = tensor<string, []>("op_44100_cast")];
            tensor<fp16, []> var_44101_to_fp16 = const()[name = tensor<string, []>("op_44101_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3739_cast = mul(x = var_44100_cast, y = var_44101_to_fp16)[name = tensor<string, []>("aw_3739_cast")];
            tensor<string, []> var_44104_equation_0 = const()[name = tensor<string, []>("op_44104_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44104_cast = einsum(equation = var_44104_equation_0, values = (var_43946_cast, var_43863_cast))[name = tensor<string, []>("op_44104_cast")];
            tensor<fp16, []> var_44105_to_fp16 = const()[name = tensor<string, []>("op_44105_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3741_cast = mul(x = var_44104_cast, y = var_44105_to_fp16)[name = tensor<string, []>("aw_3741_cast")];
            tensor<string, []> var_44108_equation_0 = const()[name = tensor<string, []>("op_44108_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44108_cast = einsum(equation = var_44108_equation_0, values = (var_43950_cast, var_43867_cast))[name = tensor<string, []>("op_44108_cast")];
            tensor<fp16, []> var_44109_to_fp16 = const()[name = tensor<string, []>("op_44109_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3743_cast = mul(x = var_44108_cast, y = var_44109_to_fp16)[name = tensor<string, []>("aw_3743_cast")];
            tensor<string, []> var_44112_equation_0 = const()[name = tensor<string, []>("op_44112_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44112_cast = einsum(equation = var_44112_equation_0, values = (var_43954_cast, var_43871_cast))[name = tensor<string, []>("op_44112_cast")];
            tensor<fp16, []> var_44113_to_fp16 = const()[name = tensor<string, []>("op_44113_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3745_cast = mul(x = var_44112_cast, y = var_44113_to_fp16)[name = tensor<string, []>("aw_3745_cast")];
            tensor<string, []> var_44116_equation_0 = const()[name = tensor<string, []>("op_44116_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44116_cast = einsum(equation = var_44116_equation_0, values = (var_43958_cast, var_43875_cast))[name = tensor<string, []>("op_44116_cast")];
            tensor<fp16, []> var_44117_to_fp16 = const()[name = tensor<string, []>("op_44117_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3747_cast = mul(x = var_44116_cast, y = var_44117_to_fp16)[name = tensor<string, []>("aw_3747_cast")];
            tensor<string, []> var_44120_equation_0 = const()[name = tensor<string, []>("op_44120_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44120_cast = einsum(equation = var_44120_equation_0, values = (var_43962_cast, var_43879_cast))[name = tensor<string, []>("op_44120_cast")];
            tensor<fp16, []> var_44121_to_fp16 = const()[name = tensor<string, []>("op_44121_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3749_cast = mul(x = var_44120_cast, y = var_44121_to_fp16)[name = tensor<string, []>("aw_3749_cast")];
            tensor<string, []> var_44124_equation_0 = const()[name = tensor<string, []>("op_44124_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44124_cast = einsum(equation = var_44124_equation_0, values = (var_43966_cast, var_43883_cast))[name = tensor<string, []>("op_44124_cast")];
            tensor<fp16, []> var_44125_to_fp16 = const()[name = tensor<string, []>("op_44125_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3751_cast = mul(x = var_44124_cast, y = var_44125_to_fp16)[name = tensor<string, []>("aw_3751_cast")];
            tensor<string, []> var_44128_equation_0 = const()[name = tensor<string, []>("op_44128_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44128_cast = einsum(equation = var_44128_equation_0, values = (var_43970_cast, var_43887_cast))[name = tensor<string, []>("op_44128_cast")];
            tensor<fp16, []> var_44129_to_fp16 = const()[name = tensor<string, []>("op_44129_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3753_cast = mul(x = var_44128_cast, y = var_44129_to_fp16)[name = tensor<string, []>("aw_3753_cast")];
            tensor<string, []> var_44132_equation_0 = const()[name = tensor<string, []>("op_44132_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44132_cast = einsum(equation = var_44132_equation_0, values = (var_43974_cast, var_43891_cast))[name = tensor<string, []>("op_44132_cast")];
            tensor<fp16, []> var_44133_to_fp16 = const()[name = tensor<string, []>("op_44133_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3755_cast = mul(x = var_44132_cast, y = var_44133_to_fp16)[name = tensor<string, []>("aw_3755_cast")];
            tensor<string, []> var_44136_equation_0 = const()[name = tensor<string, []>("op_44136_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44136_cast = einsum(equation = var_44136_equation_0, values = (var_43978_cast, var_43895_cast))[name = tensor<string, []>("op_44136_cast")];
            tensor<fp16, []> var_44137_to_fp16 = const()[name = tensor<string, []>("op_44137_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3757_cast = mul(x = var_44136_cast, y = var_44137_to_fp16)[name = tensor<string, []>("aw_3757_cast")];
            tensor<string, []> var_44140_equation_0 = const()[name = tensor<string, []>("op_44140_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44140_cast = einsum(equation = var_44140_equation_0, values = (var_43982_cast, var_43899_cast))[name = tensor<string, []>("op_44140_cast")];
            tensor<fp16, []> var_44141_to_fp16 = const()[name = tensor<string, []>("op_44141_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3759_cast = mul(x = var_44140_cast, y = var_44141_to_fp16)[name = tensor<string, []>("aw_3759_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44143_cast = softmax(axis = var_30385, x = aw_3721_cast)[name = tensor<string, []>("op_44143_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44144_cast = softmax(axis = var_30385, x = aw_3723_cast)[name = tensor<string, []>("op_44144_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44145_cast = softmax(axis = var_30385, x = aw_3725_cast)[name = tensor<string, []>("op_44145_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44146_cast = softmax(axis = var_30385, x = aw_3727_cast)[name = tensor<string, []>("op_44146_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44147_cast = softmax(axis = var_30385, x = aw_3729_cast)[name = tensor<string, []>("op_44147_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44148_cast = softmax(axis = var_30385, x = aw_3731_cast)[name = tensor<string, []>("op_44148_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44149_cast = softmax(axis = var_30385, x = aw_3733_cast)[name = tensor<string, []>("op_44149_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44150_cast = softmax(axis = var_30385, x = aw_3735_cast)[name = tensor<string, []>("op_44150_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44151_cast = softmax(axis = var_30385, x = aw_3737_cast)[name = tensor<string, []>("op_44151_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44152_cast = softmax(axis = var_30385, x = aw_3739_cast)[name = tensor<string, []>("op_44152_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44153_cast = softmax(axis = var_30385, x = aw_3741_cast)[name = tensor<string, []>("op_44153_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44154_cast = softmax(axis = var_30385, x = aw_3743_cast)[name = tensor<string, []>("op_44154_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44155_cast = softmax(axis = var_30385, x = aw_3745_cast)[name = tensor<string, []>("op_44155_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44156_cast = softmax(axis = var_30385, x = aw_3747_cast)[name = tensor<string, []>("op_44156_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44157_cast = softmax(axis = var_30385, x = aw_3749_cast)[name = tensor<string, []>("op_44157_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44158_cast = softmax(axis = var_30385, x = aw_3751_cast)[name = tensor<string, []>("op_44158_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44159_cast = softmax(axis = var_30385, x = aw_3753_cast)[name = tensor<string, []>("op_44159_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44160_cast = softmax(axis = var_30385, x = aw_3755_cast)[name = tensor<string, []>("op_44160_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44161_cast = softmax(axis = var_30385, x = aw_3757_cast)[name = tensor<string, []>("op_44161_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_44162_cast = softmax(axis = var_30385, x = aw_3759_cast)[name = tensor<string, []>("op_44162_cast")];
            tensor<string, []> var_44164_equation_0 = const()[name = tensor<string, []>("op_44164_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44164_cast = einsum(equation = var_44164_equation_0, values = (var_43984_cast, var_44143_cast))[name = tensor<string, []>("op_44164_cast")];
            tensor<string, []> var_44166_equation_0 = const()[name = tensor<string, []>("op_44166_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44166_cast = einsum(equation = var_44166_equation_0, values = (var_43988_cast, var_44144_cast))[name = tensor<string, []>("op_44166_cast")];
            tensor<string, []> var_44168_equation_0 = const()[name = tensor<string, []>("op_44168_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44168_cast = einsum(equation = var_44168_equation_0, values = (var_43992_cast, var_44145_cast))[name = tensor<string, []>("op_44168_cast")];
            tensor<string, []> var_44170_equation_0 = const()[name = tensor<string, []>("op_44170_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44170_cast = einsum(equation = var_44170_equation_0, values = (var_43996_cast, var_44146_cast))[name = tensor<string, []>("op_44170_cast")];
            tensor<string, []> var_44172_equation_0 = const()[name = tensor<string, []>("op_44172_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44172_cast = einsum(equation = var_44172_equation_0, values = (var_44000_cast, var_44147_cast))[name = tensor<string, []>("op_44172_cast")];
            tensor<string, []> var_44174_equation_0 = const()[name = tensor<string, []>("op_44174_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44174_cast = einsum(equation = var_44174_equation_0, values = (var_44004_cast, var_44148_cast))[name = tensor<string, []>("op_44174_cast")];
            tensor<string, []> var_44176_equation_0 = const()[name = tensor<string, []>("op_44176_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44176_cast = einsum(equation = var_44176_equation_0, values = (var_44008_cast, var_44149_cast))[name = tensor<string, []>("op_44176_cast")];
            tensor<string, []> var_44178_equation_0 = const()[name = tensor<string, []>("op_44178_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44178_cast = einsum(equation = var_44178_equation_0, values = (var_44012_cast, var_44150_cast))[name = tensor<string, []>("op_44178_cast")];
            tensor<string, []> var_44180_equation_0 = const()[name = tensor<string, []>("op_44180_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44180_cast = einsum(equation = var_44180_equation_0, values = (var_44016_cast, var_44151_cast))[name = tensor<string, []>("op_44180_cast")];
            tensor<string, []> var_44182_equation_0 = const()[name = tensor<string, []>("op_44182_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44182_cast = einsum(equation = var_44182_equation_0, values = (var_44020_cast, var_44152_cast))[name = tensor<string, []>("op_44182_cast")];
            tensor<string, []> var_44184_equation_0 = const()[name = tensor<string, []>("op_44184_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44184_cast = einsum(equation = var_44184_equation_0, values = (var_44024_cast, var_44153_cast))[name = tensor<string, []>("op_44184_cast")];
            tensor<string, []> var_44186_equation_0 = const()[name = tensor<string, []>("op_44186_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44186_cast = einsum(equation = var_44186_equation_0, values = (var_44028_cast, var_44154_cast))[name = tensor<string, []>("op_44186_cast")];
            tensor<string, []> var_44188_equation_0 = const()[name = tensor<string, []>("op_44188_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44188_cast = einsum(equation = var_44188_equation_0, values = (var_44032_cast, var_44155_cast))[name = tensor<string, []>("op_44188_cast")];
            tensor<string, []> var_44190_equation_0 = const()[name = tensor<string, []>("op_44190_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44190_cast = einsum(equation = var_44190_equation_0, values = (var_44036_cast, var_44156_cast))[name = tensor<string, []>("op_44190_cast")];
            tensor<string, []> var_44192_equation_0 = const()[name = tensor<string, []>("op_44192_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44192_cast = einsum(equation = var_44192_equation_0, values = (var_44040_cast, var_44157_cast))[name = tensor<string, []>("op_44192_cast")];
            tensor<string, []> var_44194_equation_0 = const()[name = tensor<string, []>("op_44194_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44194_cast = einsum(equation = var_44194_equation_0, values = (var_44044_cast, var_44158_cast))[name = tensor<string, []>("op_44194_cast")];
            tensor<string, []> var_44196_equation_0 = const()[name = tensor<string, []>("op_44196_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44196_cast = einsum(equation = var_44196_equation_0, values = (var_44048_cast, var_44159_cast))[name = tensor<string, []>("op_44196_cast")];
            tensor<string, []> var_44198_equation_0 = const()[name = tensor<string, []>("op_44198_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44198_cast = einsum(equation = var_44198_equation_0, values = (var_44052_cast, var_44160_cast))[name = tensor<string, []>("op_44198_cast")];
            tensor<string, []> var_44200_equation_0 = const()[name = tensor<string, []>("op_44200_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44200_cast = einsum(equation = var_44200_equation_0, values = (var_44056_cast, var_44161_cast))[name = tensor<string, []>("op_44200_cast")];
            tensor<string, []> var_44202_equation_0 = const()[name = tensor<string, []>("op_44202_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44202_cast = einsum(equation = var_44202_equation_0, values = (var_44060_cast, var_44162_cast))[name = tensor<string, []>("op_44202_cast")];
            tensor<bool, []> input_573_interleave_0 = const()[name = tensor<string, []>("input_573_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_573_cast = concat(axis = var_30385, interleave = input_573_interleave_0, values = (var_44164_cast, var_44166_cast, var_44168_cast, var_44170_cast, var_44172_cast, var_44174_cast, var_44176_cast, var_44178_cast, var_44180_cast, var_44182_cast, var_44184_cast, var_44186_cast, var_44188_cast, var_44190_cast, var_44192_cast, var_44194_cast, var_44196_cast, var_44198_cast, var_44200_cast, var_44202_cast))[name = tensor<string, []>("input_573_cast")];
            tensor<int32, [2]> var_44208 = const()[name = tensor<string, []>("op_44208"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_44210 = const()[name = tensor<string, []>("op_44210"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_44212_pad_type_0 = const()[name = tensor<string, []>("op_44212_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_44212_pad_0 = const()[name = tensor<string, []>("op_44212_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3693380736)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3696657600)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_44212_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_out_0_bias_to_fp16, dilations = var_44210, groups = var_30385, pad = var_44212_pad_0, pad_type = var_44212_pad_type_0, strides = var_44208, weight = up_blocks_0_attentions_1_transformer_blocks_4_attn2_to_out_0_weight_to_fp16, x = input_573_cast)[name = tensor<string, []>("op_44212_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_293_cast = add(x = var_44212_cast, y = inputs_291_cast)[name = tensor<string, []>("inputs_293_cast")];
            tensor<int32, [1]> var_44216 = const()[name = tensor<string, []>("op_44216"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_293_cast = reduce_mean(axes = var_44216, keep_dims = var_30380, x = inputs_293_cast)[name = tensor<string, []>("channels_mean_293_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_293_cast = sub(x = inputs_293_cast, y = channels_mean_293_cast)[name = tensor<string, []>("zero_mean_293_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_293_cast = mul(x = zero_mean_293_cast, y = zero_mean_293_cast)[name = tensor<string, []>("zero_mean_sq_293_cast")];
            tensor<int32, [1]> var_44220 = const()[name = tensor<string, []>("op_44220"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_44221_cast = reduce_mean(axes = var_44220, keep_dims = var_30380, x = zero_mean_sq_293_cast)[name = tensor<string, []>("op_44221_cast")];
            tensor<fp16, []> var_44222_to_fp16 = const()[name = tensor<string, []>("op_44222_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_44223_cast = add(x = var_44221_cast, y = var_44222_to_fp16)[name = tensor<string, []>("op_44223_cast")];
            tensor<fp16, []> denom_293_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_293_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_293_cast = rsqrt(epsilon = denom_293_epsilon_0_to_fp16, x = var_44223_cast)[name = tensor<string, []>("denom_293_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_293_cast = mul(x = zero_mean_293_cast, y = denom_293_cast)[name = tensor<string, []>("out_293_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_44227_to_fp16 = const()[name = tensor<string, []>("op_44227_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3696660224)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_44228_cast = add(x = out_293_cast, y = var_44227_to_fp16)[name = tensor<string, []>("op_44228_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_44230_to_fp16 = const()[name = tensor<string, []>("op_44230_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3696662848)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_575_cast = mul(x = var_44228_cast, y = var_44230_to_fp16)[name = tensor<string, []>("input_575_cast")];
            tensor<int32, [2]> var_44238 = const()[name = tensor<string, []>("op_44238"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_44240 = const()[name = tensor<string, []>("op_44240"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_44242_pad_type_0 = const()[name = tensor<string, []>("op_44242_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_44242_pad_0 = const()[name = tensor<string, []>("op_44242_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3696665472)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_1_transformer_blocks_4_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3722879936)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_44242_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_4_ff_net_0_proj_bias_to_fp16, dilations = var_44240, groups = var_30385, pad = var_44242_pad_0, pad_type = var_44242_pad_type_0, strides = var_44238, weight = up_blocks_0_attentions_1_transformer_blocks_4_ff_net_0_proj_weight_to_fp16, x = input_575_cast)[name = tensor<string, []>("op_44242_cast")];
            tensor<int32, [2]> var_44243_split_sizes_0 = const()[name = tensor<string, []>("op_44243_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_44243_axis_0 = const()[name = tensor<string, []>("op_44243_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_44243_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_44243_cast_1 = split(axis = var_44243_axis_0, split_sizes = var_44243_split_sizes_0, x = var_44242_cast)[name = tensor<string, []>("op_44243_cast")];
            tensor<string, []> var_44245_mode_0 = const()[name = tensor<string, []>("op_44245_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_44245_cast = gelu(mode = var_44245_mode_0, x = var_44243_cast_1)[name = tensor<string, []>("op_44245_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_577_cast = mul(x = var_44243_cast_0, y = var_44245_cast)[name = tensor<string, []>("input_577_cast")];
            tensor<int32, [2]> var_44249 = const()[name = tensor<string, []>("op_44249"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_44251 = const()[name = tensor<string, []>("op_44251"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_44253_pad_type_0 = const()[name = tensor<string, []>("op_44253_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_44253_pad_0 = const()[name = tensor<string, []>("op_44253_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_4_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3722900480)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_4_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_4_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3736007744)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_44253_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_4_ff_net_2_bias_to_fp16, dilations = var_44251, groups = var_30385, pad = var_44253_pad_0, pad_type = var_44253_pad_type_0, strides = var_44249, weight = up_blocks_0_attentions_1_transformer_blocks_4_ff_net_2_weight_to_fp16, x = input_577_cast)[name = tensor<string, []>("op_44253_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_295_cast = add(x = var_44253_cast, y = inputs_293_cast)[name = tensor<string, []>("inputs_295_cast")];
            tensor<int32, [1]> var_44263 = const()[name = tensor<string, []>("op_44263"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_295_cast = reduce_mean(axes = var_44263, keep_dims = var_30380, x = inputs_295_cast)[name = tensor<string, []>("channels_mean_295_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_295_cast = sub(x = inputs_295_cast, y = channels_mean_295_cast)[name = tensor<string, []>("zero_mean_295_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_295_cast = mul(x = zero_mean_295_cast, y = zero_mean_295_cast)[name = tensor<string, []>("zero_mean_sq_295_cast")];
            tensor<int32, [1]> var_44267 = const()[name = tensor<string, []>("op_44267"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_44268_cast = reduce_mean(axes = var_44267, keep_dims = var_30380, x = zero_mean_sq_295_cast)[name = tensor<string, []>("op_44268_cast")];
            tensor<fp16, []> var_44269_to_fp16 = const()[name = tensor<string, []>("op_44269_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_44270_cast = add(x = var_44268_cast, y = var_44269_to_fp16)[name = tensor<string, []>("op_44270_cast")];
            tensor<fp16, []> denom_295_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_295_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_295_cast = rsqrt(epsilon = denom_295_epsilon_0_to_fp16, x = var_44270_cast)[name = tensor<string, []>("denom_295_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_295_cast = mul(x = zero_mean_295_cast, y = denom_295_cast)[name = tensor<string, []>("out_295_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_44274_to_fp16 = const()[name = tensor<string, []>("op_44274_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3736010368)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_44275_cast = add(x = out_295_cast, y = var_44274_to_fp16)[name = tensor<string, []>("op_44275_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_44277_to_fp16 = const()[name = tensor<string, []>("op_44277_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3736012992)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_391_cast = mul(x = var_44275_cast, y = var_44277_to_fp16)[name = tensor<string, []>("hidden_states_391_cast")];
            tensor<int32, [2]> var_44284 = const()[name = tensor<string, []>("op_44284"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_44286 = const()[name = tensor<string, []>("op_44286"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_197_pad_type_0 = const()[name = tensor<string, []>("q_197_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_197_pad_0 = const()[name = tensor<string, []>("q_197_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3736015616)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_197_cast = conv(dilations = var_44286, groups = var_30385, pad = q_197_pad_0, pad_type = q_197_pad_type_0, strides = var_44284, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_q_weight_to_fp16, x = hidden_states_391_cast)[name = tensor<string, []>("q_197_cast")];
            tensor<int32, [2]> var_44290 = const()[name = tensor<string, []>("op_44290"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_44292 = const()[name = tensor<string, []>("op_44292"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_393_pad_type_0 = const()[name = tensor<string, []>("k_393_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_393_pad_0 = const()[name = tensor<string, []>("k_393_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3739292480)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_393_cast = conv(dilations = var_44292, groups = var_30385, pad = k_393_pad_0, pad_type = k_393_pad_type_0, strides = var_44290, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_k_weight_to_fp16, x = hidden_states_391_cast)[name = tensor<string, []>("k_393_cast")];
            tensor<int32, [2]> var_44296 = const()[name = tensor<string, []>("op_44296"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_44298 = const()[name = tensor<string, []>("op_44298"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_197_pad_type_0 = const()[name = tensor<string, []>("v_197_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_197_pad_0 = const()[name = tensor<string, []>("v_197_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3742569344)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_197_cast = conv(dilations = var_44298, groups = var_30385, pad = v_197_pad_0, pad_type = v_197_pad_type_0, strides = var_44296, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_v_weight_to_fp16, x = hidden_states_391_cast)[name = tensor<string, []>("v_197_cast")];
            tensor<int32, [4]> var_44302_begin_0 = const()[name = tensor<string, []>("op_44302_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_44302_end_0 = const()[name = tensor<string, []>("op_44302_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_44302_end_mask_0 = const()[name = tensor<string, []>("op_44302_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44302_cast = slice_by_index(begin = var_44302_begin_0, end = var_44302_end_0, end_mask = var_44302_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44302_cast")];
            tensor<int32, [4]> var_44306_begin_0 = const()[name = tensor<string, []>("op_44306_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_44306_end_0 = const()[name = tensor<string, []>("op_44306_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_44306_end_mask_0 = const()[name = tensor<string, []>("op_44306_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44306_cast = slice_by_index(begin = var_44306_begin_0, end = var_44306_end_0, end_mask = var_44306_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44306_cast")];
            tensor<int32, [4]> var_44310_begin_0 = const()[name = tensor<string, []>("op_44310_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_44310_end_0 = const()[name = tensor<string, []>("op_44310_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_44310_end_mask_0 = const()[name = tensor<string, []>("op_44310_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44310_cast = slice_by_index(begin = var_44310_begin_0, end = var_44310_end_0, end_mask = var_44310_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44310_cast")];
            tensor<int32, [4]> var_44314_begin_0 = const()[name = tensor<string, []>("op_44314_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_44314_end_0 = const()[name = tensor<string, []>("op_44314_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_44314_end_mask_0 = const()[name = tensor<string, []>("op_44314_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44314_cast = slice_by_index(begin = var_44314_begin_0, end = var_44314_end_0, end_mask = var_44314_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44314_cast")];
            tensor<int32, [4]> var_44318_begin_0 = const()[name = tensor<string, []>("op_44318_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_44318_end_0 = const()[name = tensor<string, []>("op_44318_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_44318_end_mask_0 = const()[name = tensor<string, []>("op_44318_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44318_cast = slice_by_index(begin = var_44318_begin_0, end = var_44318_end_0, end_mask = var_44318_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44318_cast")];
            tensor<int32, [4]> var_44322_begin_0 = const()[name = tensor<string, []>("op_44322_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_44322_end_0 = const()[name = tensor<string, []>("op_44322_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_44322_end_mask_0 = const()[name = tensor<string, []>("op_44322_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44322_cast = slice_by_index(begin = var_44322_begin_0, end = var_44322_end_0, end_mask = var_44322_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44322_cast")];
            tensor<int32, [4]> var_44326_begin_0 = const()[name = tensor<string, []>("op_44326_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_44326_end_0 = const()[name = tensor<string, []>("op_44326_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_44326_end_mask_0 = const()[name = tensor<string, []>("op_44326_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44326_cast = slice_by_index(begin = var_44326_begin_0, end = var_44326_end_0, end_mask = var_44326_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44326_cast")];
            tensor<int32, [4]> var_44330_begin_0 = const()[name = tensor<string, []>("op_44330_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_44330_end_0 = const()[name = tensor<string, []>("op_44330_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_44330_end_mask_0 = const()[name = tensor<string, []>("op_44330_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44330_cast = slice_by_index(begin = var_44330_begin_0, end = var_44330_end_0, end_mask = var_44330_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44330_cast")];
            tensor<int32, [4]> var_44334_begin_0 = const()[name = tensor<string, []>("op_44334_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_44334_end_0 = const()[name = tensor<string, []>("op_44334_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_44334_end_mask_0 = const()[name = tensor<string, []>("op_44334_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44334_cast = slice_by_index(begin = var_44334_begin_0, end = var_44334_end_0, end_mask = var_44334_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44334_cast")];
            tensor<int32, [4]> var_44338_begin_0 = const()[name = tensor<string, []>("op_44338_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_44338_end_0 = const()[name = tensor<string, []>("op_44338_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_44338_end_mask_0 = const()[name = tensor<string, []>("op_44338_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44338_cast = slice_by_index(begin = var_44338_begin_0, end = var_44338_end_0, end_mask = var_44338_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44338_cast")];
            tensor<int32, [4]> var_44342_begin_0 = const()[name = tensor<string, []>("op_44342_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_44342_end_0 = const()[name = tensor<string, []>("op_44342_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_44342_end_mask_0 = const()[name = tensor<string, []>("op_44342_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44342_cast = slice_by_index(begin = var_44342_begin_0, end = var_44342_end_0, end_mask = var_44342_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44342_cast")];
            tensor<int32, [4]> var_44346_begin_0 = const()[name = tensor<string, []>("op_44346_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_44346_end_0 = const()[name = tensor<string, []>("op_44346_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_44346_end_mask_0 = const()[name = tensor<string, []>("op_44346_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44346_cast = slice_by_index(begin = var_44346_begin_0, end = var_44346_end_0, end_mask = var_44346_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44346_cast")];
            tensor<int32, [4]> var_44350_begin_0 = const()[name = tensor<string, []>("op_44350_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_44350_end_0 = const()[name = tensor<string, []>("op_44350_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_44350_end_mask_0 = const()[name = tensor<string, []>("op_44350_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44350_cast = slice_by_index(begin = var_44350_begin_0, end = var_44350_end_0, end_mask = var_44350_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44350_cast")];
            tensor<int32, [4]> var_44354_begin_0 = const()[name = tensor<string, []>("op_44354_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_44354_end_0 = const()[name = tensor<string, []>("op_44354_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_44354_end_mask_0 = const()[name = tensor<string, []>("op_44354_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44354_cast = slice_by_index(begin = var_44354_begin_0, end = var_44354_end_0, end_mask = var_44354_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44354_cast")];
            tensor<int32, [4]> var_44358_begin_0 = const()[name = tensor<string, []>("op_44358_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_44358_end_0 = const()[name = tensor<string, []>("op_44358_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_44358_end_mask_0 = const()[name = tensor<string, []>("op_44358_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44358_cast = slice_by_index(begin = var_44358_begin_0, end = var_44358_end_0, end_mask = var_44358_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44358_cast")];
            tensor<int32, [4]> var_44362_begin_0 = const()[name = tensor<string, []>("op_44362_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_44362_end_0 = const()[name = tensor<string, []>("op_44362_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_44362_end_mask_0 = const()[name = tensor<string, []>("op_44362_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44362_cast = slice_by_index(begin = var_44362_begin_0, end = var_44362_end_0, end_mask = var_44362_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44362_cast")];
            tensor<int32, [4]> var_44366_begin_0 = const()[name = tensor<string, []>("op_44366_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_44366_end_0 = const()[name = tensor<string, []>("op_44366_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_44366_end_mask_0 = const()[name = tensor<string, []>("op_44366_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44366_cast = slice_by_index(begin = var_44366_begin_0, end = var_44366_end_0, end_mask = var_44366_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44366_cast")];
            tensor<int32, [4]> var_44370_begin_0 = const()[name = tensor<string, []>("op_44370_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_44370_end_0 = const()[name = tensor<string, []>("op_44370_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_44370_end_mask_0 = const()[name = tensor<string, []>("op_44370_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44370_cast = slice_by_index(begin = var_44370_begin_0, end = var_44370_end_0, end_mask = var_44370_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44370_cast")];
            tensor<int32, [4]> var_44374_begin_0 = const()[name = tensor<string, []>("op_44374_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_44374_end_0 = const()[name = tensor<string, []>("op_44374_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_44374_end_mask_0 = const()[name = tensor<string, []>("op_44374_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44374_cast = slice_by_index(begin = var_44374_begin_0, end = var_44374_end_0, end_mask = var_44374_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44374_cast")];
            tensor<int32, [4]> var_44378_begin_0 = const()[name = tensor<string, []>("op_44378_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_44378_end_0 = const()[name = tensor<string, []>("op_44378_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_44378_end_mask_0 = const()[name = tensor<string, []>("op_44378_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44378_cast = slice_by_index(begin = var_44378_begin_0, end = var_44378_end_0, end_mask = var_44378_end_mask_0, x = q_197_cast)[name = tensor<string, []>("op_44378_cast")];
            tensor<int32, [4]> k_395_perm_0 = const()[name = tensor<string, []>("k_395_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_44385_begin_0 = const()[name = tensor<string, []>("op_44385_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_44385_end_0 = const()[name = tensor<string, []>("op_44385_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_44385_end_mask_0 = const()[name = tensor<string, []>("op_44385_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_41 = transpose(perm = k_395_perm_0, x = k_393_cast)[name = tensor<string, []>("transpose_41")];
            tensor<fp16, [2, 1024, 1, 64]> var_44385_cast = slice_by_index(begin = var_44385_begin_0, end = var_44385_end_0, end_mask = var_44385_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44385_cast")];
            tensor<int32, [4]> var_44389_begin_0 = const()[name = tensor<string, []>("op_44389_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_44389_end_0 = const()[name = tensor<string, []>("op_44389_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_44389_end_mask_0 = const()[name = tensor<string, []>("op_44389_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44389_cast = slice_by_index(begin = var_44389_begin_0, end = var_44389_end_0, end_mask = var_44389_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44389_cast")];
            tensor<int32, [4]> var_44393_begin_0 = const()[name = tensor<string, []>("op_44393_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_44393_end_0 = const()[name = tensor<string, []>("op_44393_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_44393_end_mask_0 = const()[name = tensor<string, []>("op_44393_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44393_cast = slice_by_index(begin = var_44393_begin_0, end = var_44393_end_0, end_mask = var_44393_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44393_cast")];
            tensor<int32, [4]> var_44397_begin_0 = const()[name = tensor<string, []>("op_44397_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_44397_end_0 = const()[name = tensor<string, []>("op_44397_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_44397_end_mask_0 = const()[name = tensor<string, []>("op_44397_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44397_cast = slice_by_index(begin = var_44397_begin_0, end = var_44397_end_0, end_mask = var_44397_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44397_cast")];
            tensor<int32, [4]> var_44401_begin_0 = const()[name = tensor<string, []>("op_44401_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_44401_end_0 = const()[name = tensor<string, []>("op_44401_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_44401_end_mask_0 = const()[name = tensor<string, []>("op_44401_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44401_cast = slice_by_index(begin = var_44401_begin_0, end = var_44401_end_0, end_mask = var_44401_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44401_cast")];
            tensor<int32, [4]> var_44405_begin_0 = const()[name = tensor<string, []>("op_44405_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_44405_end_0 = const()[name = tensor<string, []>("op_44405_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_44405_end_mask_0 = const()[name = tensor<string, []>("op_44405_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44405_cast = slice_by_index(begin = var_44405_begin_0, end = var_44405_end_0, end_mask = var_44405_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44405_cast")];
            tensor<int32, [4]> var_44409_begin_0 = const()[name = tensor<string, []>("op_44409_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_44409_end_0 = const()[name = tensor<string, []>("op_44409_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_44409_end_mask_0 = const()[name = tensor<string, []>("op_44409_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44409_cast = slice_by_index(begin = var_44409_begin_0, end = var_44409_end_0, end_mask = var_44409_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44409_cast")];
            tensor<int32, [4]> var_44413_begin_0 = const()[name = tensor<string, []>("op_44413_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_44413_end_0 = const()[name = tensor<string, []>("op_44413_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_44413_end_mask_0 = const()[name = tensor<string, []>("op_44413_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44413_cast = slice_by_index(begin = var_44413_begin_0, end = var_44413_end_0, end_mask = var_44413_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44413_cast")];
            tensor<int32, [4]> var_44417_begin_0 = const()[name = tensor<string, []>("op_44417_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_44417_end_0 = const()[name = tensor<string, []>("op_44417_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_44417_end_mask_0 = const()[name = tensor<string, []>("op_44417_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44417_cast = slice_by_index(begin = var_44417_begin_0, end = var_44417_end_0, end_mask = var_44417_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44417_cast")];
            tensor<int32, [4]> var_44421_begin_0 = const()[name = tensor<string, []>("op_44421_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_44421_end_0 = const()[name = tensor<string, []>("op_44421_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_44421_end_mask_0 = const()[name = tensor<string, []>("op_44421_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44421_cast = slice_by_index(begin = var_44421_begin_0, end = var_44421_end_0, end_mask = var_44421_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44421_cast")];
            tensor<int32, [4]> var_44425_begin_0 = const()[name = tensor<string, []>("op_44425_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_44425_end_0 = const()[name = tensor<string, []>("op_44425_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_44425_end_mask_0 = const()[name = tensor<string, []>("op_44425_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44425_cast = slice_by_index(begin = var_44425_begin_0, end = var_44425_end_0, end_mask = var_44425_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44425_cast")];
            tensor<int32, [4]> var_44429_begin_0 = const()[name = tensor<string, []>("op_44429_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_44429_end_0 = const()[name = tensor<string, []>("op_44429_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_44429_end_mask_0 = const()[name = tensor<string, []>("op_44429_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44429_cast = slice_by_index(begin = var_44429_begin_0, end = var_44429_end_0, end_mask = var_44429_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44429_cast")];
            tensor<int32, [4]> var_44433_begin_0 = const()[name = tensor<string, []>("op_44433_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_44433_end_0 = const()[name = tensor<string, []>("op_44433_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_44433_end_mask_0 = const()[name = tensor<string, []>("op_44433_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44433_cast = slice_by_index(begin = var_44433_begin_0, end = var_44433_end_0, end_mask = var_44433_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44433_cast")];
            tensor<int32, [4]> var_44437_begin_0 = const()[name = tensor<string, []>("op_44437_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_44437_end_0 = const()[name = tensor<string, []>("op_44437_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_44437_end_mask_0 = const()[name = tensor<string, []>("op_44437_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44437_cast = slice_by_index(begin = var_44437_begin_0, end = var_44437_end_0, end_mask = var_44437_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44437_cast")];
            tensor<int32, [4]> var_44441_begin_0 = const()[name = tensor<string, []>("op_44441_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_44441_end_0 = const()[name = tensor<string, []>("op_44441_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_44441_end_mask_0 = const()[name = tensor<string, []>("op_44441_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44441_cast = slice_by_index(begin = var_44441_begin_0, end = var_44441_end_0, end_mask = var_44441_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44441_cast")];
            tensor<int32, [4]> var_44445_begin_0 = const()[name = tensor<string, []>("op_44445_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_44445_end_0 = const()[name = tensor<string, []>("op_44445_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_44445_end_mask_0 = const()[name = tensor<string, []>("op_44445_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44445_cast = slice_by_index(begin = var_44445_begin_0, end = var_44445_end_0, end_mask = var_44445_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44445_cast")];
            tensor<int32, [4]> var_44449_begin_0 = const()[name = tensor<string, []>("op_44449_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_44449_end_0 = const()[name = tensor<string, []>("op_44449_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_44449_end_mask_0 = const()[name = tensor<string, []>("op_44449_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44449_cast = slice_by_index(begin = var_44449_begin_0, end = var_44449_end_0, end_mask = var_44449_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44449_cast")];
            tensor<int32, [4]> var_44453_begin_0 = const()[name = tensor<string, []>("op_44453_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_44453_end_0 = const()[name = tensor<string, []>("op_44453_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_44453_end_mask_0 = const()[name = tensor<string, []>("op_44453_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44453_cast = slice_by_index(begin = var_44453_begin_0, end = var_44453_end_0, end_mask = var_44453_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44453_cast")];
            tensor<int32, [4]> var_44457_begin_0 = const()[name = tensor<string, []>("op_44457_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_44457_end_0 = const()[name = tensor<string, []>("op_44457_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_44457_end_mask_0 = const()[name = tensor<string, []>("op_44457_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44457_cast = slice_by_index(begin = var_44457_begin_0, end = var_44457_end_0, end_mask = var_44457_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44457_cast")];
            tensor<int32, [4]> var_44461_begin_0 = const()[name = tensor<string, []>("op_44461_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_44461_end_0 = const()[name = tensor<string, []>("op_44461_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_44461_end_mask_0 = const()[name = tensor<string, []>("op_44461_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_44461_cast = slice_by_index(begin = var_44461_begin_0, end = var_44461_end_0, end_mask = var_44461_end_mask_0, x = transpose_41)[name = tensor<string, []>("op_44461_cast")];
            tensor<int32, [4]> var_44463_begin_0 = const()[name = tensor<string, []>("op_44463_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_44463_end_0 = const()[name = tensor<string, []>("op_44463_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_44463_end_mask_0 = const()[name = tensor<string, []>("op_44463_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44463_cast = slice_by_index(begin = var_44463_begin_0, end = var_44463_end_0, end_mask = var_44463_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44463_cast")];
            tensor<int32, [4]> var_44467_begin_0 = const()[name = tensor<string, []>("op_44467_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_44467_end_0 = const()[name = tensor<string, []>("op_44467_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_44467_end_mask_0 = const()[name = tensor<string, []>("op_44467_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44467_cast = slice_by_index(begin = var_44467_begin_0, end = var_44467_end_0, end_mask = var_44467_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44467_cast")];
            tensor<int32, [4]> var_44471_begin_0 = const()[name = tensor<string, []>("op_44471_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_44471_end_0 = const()[name = tensor<string, []>("op_44471_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_44471_end_mask_0 = const()[name = tensor<string, []>("op_44471_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44471_cast = slice_by_index(begin = var_44471_begin_0, end = var_44471_end_0, end_mask = var_44471_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44471_cast")];
            tensor<int32, [4]> var_44475_begin_0 = const()[name = tensor<string, []>("op_44475_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_44475_end_0 = const()[name = tensor<string, []>("op_44475_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_44475_end_mask_0 = const()[name = tensor<string, []>("op_44475_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44475_cast = slice_by_index(begin = var_44475_begin_0, end = var_44475_end_0, end_mask = var_44475_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44475_cast")];
            tensor<int32, [4]> var_44479_begin_0 = const()[name = tensor<string, []>("op_44479_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_44479_end_0 = const()[name = tensor<string, []>("op_44479_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_44479_end_mask_0 = const()[name = tensor<string, []>("op_44479_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44479_cast = slice_by_index(begin = var_44479_begin_0, end = var_44479_end_0, end_mask = var_44479_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44479_cast")];
            tensor<int32, [4]> var_44483_begin_0 = const()[name = tensor<string, []>("op_44483_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_44483_end_0 = const()[name = tensor<string, []>("op_44483_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_44483_end_mask_0 = const()[name = tensor<string, []>("op_44483_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44483_cast = slice_by_index(begin = var_44483_begin_0, end = var_44483_end_0, end_mask = var_44483_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44483_cast")];
            tensor<int32, [4]> var_44487_begin_0 = const()[name = tensor<string, []>("op_44487_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_44487_end_0 = const()[name = tensor<string, []>("op_44487_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_44487_end_mask_0 = const()[name = tensor<string, []>("op_44487_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44487_cast = slice_by_index(begin = var_44487_begin_0, end = var_44487_end_0, end_mask = var_44487_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44487_cast")];
            tensor<int32, [4]> var_44491_begin_0 = const()[name = tensor<string, []>("op_44491_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_44491_end_0 = const()[name = tensor<string, []>("op_44491_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_44491_end_mask_0 = const()[name = tensor<string, []>("op_44491_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44491_cast = slice_by_index(begin = var_44491_begin_0, end = var_44491_end_0, end_mask = var_44491_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44491_cast")];
            tensor<int32, [4]> var_44495_begin_0 = const()[name = tensor<string, []>("op_44495_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_44495_end_0 = const()[name = tensor<string, []>("op_44495_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_44495_end_mask_0 = const()[name = tensor<string, []>("op_44495_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44495_cast = slice_by_index(begin = var_44495_begin_0, end = var_44495_end_0, end_mask = var_44495_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44495_cast")];
            tensor<int32, [4]> var_44499_begin_0 = const()[name = tensor<string, []>("op_44499_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_44499_end_0 = const()[name = tensor<string, []>("op_44499_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_44499_end_mask_0 = const()[name = tensor<string, []>("op_44499_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44499_cast = slice_by_index(begin = var_44499_begin_0, end = var_44499_end_0, end_mask = var_44499_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44499_cast")];
            tensor<int32, [4]> var_44503_begin_0 = const()[name = tensor<string, []>("op_44503_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_44503_end_0 = const()[name = tensor<string, []>("op_44503_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_44503_end_mask_0 = const()[name = tensor<string, []>("op_44503_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44503_cast = slice_by_index(begin = var_44503_begin_0, end = var_44503_end_0, end_mask = var_44503_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44503_cast")];
            tensor<int32, [4]> var_44507_begin_0 = const()[name = tensor<string, []>("op_44507_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_44507_end_0 = const()[name = tensor<string, []>("op_44507_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_44507_end_mask_0 = const()[name = tensor<string, []>("op_44507_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44507_cast = slice_by_index(begin = var_44507_begin_0, end = var_44507_end_0, end_mask = var_44507_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44507_cast")];
            tensor<int32, [4]> var_44511_begin_0 = const()[name = tensor<string, []>("op_44511_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_44511_end_0 = const()[name = tensor<string, []>("op_44511_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_44511_end_mask_0 = const()[name = tensor<string, []>("op_44511_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44511_cast = slice_by_index(begin = var_44511_begin_0, end = var_44511_end_0, end_mask = var_44511_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44511_cast")];
            tensor<int32, [4]> var_44515_begin_0 = const()[name = tensor<string, []>("op_44515_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_44515_end_0 = const()[name = tensor<string, []>("op_44515_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_44515_end_mask_0 = const()[name = tensor<string, []>("op_44515_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44515_cast = slice_by_index(begin = var_44515_begin_0, end = var_44515_end_0, end_mask = var_44515_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44515_cast")];
            tensor<int32, [4]> var_44519_begin_0 = const()[name = tensor<string, []>("op_44519_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_44519_end_0 = const()[name = tensor<string, []>("op_44519_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_44519_end_mask_0 = const()[name = tensor<string, []>("op_44519_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44519_cast = slice_by_index(begin = var_44519_begin_0, end = var_44519_end_0, end_mask = var_44519_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44519_cast")];
            tensor<int32, [4]> var_44523_begin_0 = const()[name = tensor<string, []>("op_44523_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_44523_end_0 = const()[name = tensor<string, []>("op_44523_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_44523_end_mask_0 = const()[name = tensor<string, []>("op_44523_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44523_cast = slice_by_index(begin = var_44523_begin_0, end = var_44523_end_0, end_mask = var_44523_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44523_cast")];
            tensor<int32, [4]> var_44527_begin_0 = const()[name = tensor<string, []>("op_44527_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_44527_end_0 = const()[name = tensor<string, []>("op_44527_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_44527_end_mask_0 = const()[name = tensor<string, []>("op_44527_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44527_cast = slice_by_index(begin = var_44527_begin_0, end = var_44527_end_0, end_mask = var_44527_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44527_cast")];
            tensor<int32, [4]> var_44531_begin_0 = const()[name = tensor<string, []>("op_44531_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_44531_end_0 = const()[name = tensor<string, []>("op_44531_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_44531_end_mask_0 = const()[name = tensor<string, []>("op_44531_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44531_cast = slice_by_index(begin = var_44531_begin_0, end = var_44531_end_0, end_mask = var_44531_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44531_cast")];
            tensor<int32, [4]> var_44535_begin_0 = const()[name = tensor<string, []>("op_44535_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_44535_end_0 = const()[name = tensor<string, []>("op_44535_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_44535_end_mask_0 = const()[name = tensor<string, []>("op_44535_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44535_cast = slice_by_index(begin = var_44535_begin_0, end = var_44535_end_0, end_mask = var_44535_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44535_cast")];
            tensor<int32, [4]> var_44539_begin_0 = const()[name = tensor<string, []>("op_44539_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_44539_end_0 = const()[name = tensor<string, []>("op_44539_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_44539_end_mask_0 = const()[name = tensor<string, []>("op_44539_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44539_cast = slice_by_index(begin = var_44539_begin_0, end = var_44539_end_0, end_mask = var_44539_end_mask_0, x = v_197_cast)[name = tensor<string, []>("op_44539_cast")];
            tensor<string, []> var_44543_equation_0 = const()[name = tensor<string, []>("op_44543_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44543_cast = einsum(equation = var_44543_equation_0, values = (var_44385_cast, var_44302_cast))[name = tensor<string, []>("op_44543_cast")];
            tensor<fp16, []> var_44544_to_fp16 = const()[name = tensor<string, []>("op_44544_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3761_cast = mul(x = var_44543_cast, y = var_44544_to_fp16)[name = tensor<string, []>("aw_3761_cast")];
            tensor<string, []> var_44547_equation_0 = const()[name = tensor<string, []>("op_44547_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44547_cast = einsum(equation = var_44547_equation_0, values = (var_44389_cast, var_44306_cast))[name = tensor<string, []>("op_44547_cast")];
            tensor<fp16, []> var_44548_to_fp16 = const()[name = tensor<string, []>("op_44548_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3763_cast = mul(x = var_44547_cast, y = var_44548_to_fp16)[name = tensor<string, []>("aw_3763_cast")];
            tensor<string, []> var_44551_equation_0 = const()[name = tensor<string, []>("op_44551_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44551_cast = einsum(equation = var_44551_equation_0, values = (var_44393_cast, var_44310_cast))[name = tensor<string, []>("op_44551_cast")];
            tensor<fp16, []> var_44552_to_fp16 = const()[name = tensor<string, []>("op_44552_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3765_cast = mul(x = var_44551_cast, y = var_44552_to_fp16)[name = tensor<string, []>("aw_3765_cast")];
            tensor<string, []> var_44555_equation_0 = const()[name = tensor<string, []>("op_44555_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44555_cast = einsum(equation = var_44555_equation_0, values = (var_44397_cast, var_44314_cast))[name = tensor<string, []>("op_44555_cast")];
            tensor<fp16, []> var_44556_to_fp16 = const()[name = tensor<string, []>("op_44556_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3767_cast = mul(x = var_44555_cast, y = var_44556_to_fp16)[name = tensor<string, []>("aw_3767_cast")];
            tensor<string, []> var_44559_equation_0 = const()[name = tensor<string, []>("op_44559_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44559_cast = einsum(equation = var_44559_equation_0, values = (var_44401_cast, var_44318_cast))[name = tensor<string, []>("op_44559_cast")];
            tensor<fp16, []> var_44560_to_fp16 = const()[name = tensor<string, []>("op_44560_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3769_cast = mul(x = var_44559_cast, y = var_44560_to_fp16)[name = tensor<string, []>("aw_3769_cast")];
            tensor<string, []> var_44563_equation_0 = const()[name = tensor<string, []>("op_44563_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44563_cast = einsum(equation = var_44563_equation_0, values = (var_44405_cast, var_44322_cast))[name = tensor<string, []>("op_44563_cast")];
            tensor<fp16, []> var_44564_to_fp16 = const()[name = tensor<string, []>("op_44564_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3771_cast = mul(x = var_44563_cast, y = var_44564_to_fp16)[name = tensor<string, []>("aw_3771_cast")];
            tensor<string, []> var_44567_equation_0 = const()[name = tensor<string, []>("op_44567_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44567_cast = einsum(equation = var_44567_equation_0, values = (var_44409_cast, var_44326_cast))[name = tensor<string, []>("op_44567_cast")];
            tensor<fp16, []> var_44568_to_fp16 = const()[name = tensor<string, []>("op_44568_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3773_cast = mul(x = var_44567_cast, y = var_44568_to_fp16)[name = tensor<string, []>("aw_3773_cast")];
            tensor<string, []> var_44571_equation_0 = const()[name = tensor<string, []>("op_44571_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44571_cast = einsum(equation = var_44571_equation_0, values = (var_44413_cast, var_44330_cast))[name = tensor<string, []>("op_44571_cast")];
            tensor<fp16, []> var_44572_to_fp16 = const()[name = tensor<string, []>("op_44572_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3775_cast = mul(x = var_44571_cast, y = var_44572_to_fp16)[name = tensor<string, []>("aw_3775_cast")];
            tensor<string, []> var_44575_equation_0 = const()[name = tensor<string, []>("op_44575_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44575_cast = einsum(equation = var_44575_equation_0, values = (var_44417_cast, var_44334_cast))[name = tensor<string, []>("op_44575_cast")];
            tensor<fp16, []> var_44576_to_fp16 = const()[name = tensor<string, []>("op_44576_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3777_cast = mul(x = var_44575_cast, y = var_44576_to_fp16)[name = tensor<string, []>("aw_3777_cast")];
            tensor<string, []> var_44579_equation_0 = const()[name = tensor<string, []>("op_44579_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44579_cast = einsum(equation = var_44579_equation_0, values = (var_44421_cast, var_44338_cast))[name = tensor<string, []>("op_44579_cast")];
            tensor<fp16, []> var_44580_to_fp16 = const()[name = tensor<string, []>("op_44580_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3779_cast = mul(x = var_44579_cast, y = var_44580_to_fp16)[name = tensor<string, []>("aw_3779_cast")];
            tensor<string, []> var_44583_equation_0 = const()[name = tensor<string, []>("op_44583_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44583_cast = einsum(equation = var_44583_equation_0, values = (var_44425_cast, var_44342_cast))[name = tensor<string, []>("op_44583_cast")];
            tensor<fp16, []> var_44584_to_fp16 = const()[name = tensor<string, []>("op_44584_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3781_cast = mul(x = var_44583_cast, y = var_44584_to_fp16)[name = tensor<string, []>("aw_3781_cast")];
            tensor<string, []> var_44587_equation_0 = const()[name = tensor<string, []>("op_44587_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44587_cast = einsum(equation = var_44587_equation_0, values = (var_44429_cast, var_44346_cast))[name = tensor<string, []>("op_44587_cast")];
            tensor<fp16, []> var_44588_to_fp16 = const()[name = tensor<string, []>("op_44588_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3783_cast = mul(x = var_44587_cast, y = var_44588_to_fp16)[name = tensor<string, []>("aw_3783_cast")];
            tensor<string, []> var_44591_equation_0 = const()[name = tensor<string, []>("op_44591_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44591_cast = einsum(equation = var_44591_equation_0, values = (var_44433_cast, var_44350_cast))[name = tensor<string, []>("op_44591_cast")];
            tensor<fp16, []> var_44592_to_fp16 = const()[name = tensor<string, []>("op_44592_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3785_cast = mul(x = var_44591_cast, y = var_44592_to_fp16)[name = tensor<string, []>("aw_3785_cast")];
            tensor<string, []> var_44595_equation_0 = const()[name = tensor<string, []>("op_44595_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44595_cast = einsum(equation = var_44595_equation_0, values = (var_44437_cast, var_44354_cast))[name = tensor<string, []>("op_44595_cast")];
            tensor<fp16, []> var_44596_to_fp16 = const()[name = tensor<string, []>("op_44596_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3787_cast = mul(x = var_44595_cast, y = var_44596_to_fp16)[name = tensor<string, []>("aw_3787_cast")];
            tensor<string, []> var_44599_equation_0 = const()[name = tensor<string, []>("op_44599_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44599_cast = einsum(equation = var_44599_equation_0, values = (var_44441_cast, var_44358_cast))[name = tensor<string, []>("op_44599_cast")];
            tensor<fp16, []> var_44600_to_fp16 = const()[name = tensor<string, []>("op_44600_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3789_cast = mul(x = var_44599_cast, y = var_44600_to_fp16)[name = tensor<string, []>("aw_3789_cast")];
            tensor<string, []> var_44603_equation_0 = const()[name = tensor<string, []>("op_44603_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44603_cast = einsum(equation = var_44603_equation_0, values = (var_44445_cast, var_44362_cast))[name = tensor<string, []>("op_44603_cast")];
            tensor<fp16, []> var_44604_to_fp16 = const()[name = tensor<string, []>("op_44604_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3791_cast = mul(x = var_44603_cast, y = var_44604_to_fp16)[name = tensor<string, []>("aw_3791_cast")];
            tensor<string, []> var_44607_equation_0 = const()[name = tensor<string, []>("op_44607_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44607_cast = einsum(equation = var_44607_equation_0, values = (var_44449_cast, var_44366_cast))[name = tensor<string, []>("op_44607_cast")];
            tensor<fp16, []> var_44608_to_fp16 = const()[name = tensor<string, []>("op_44608_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3793_cast = mul(x = var_44607_cast, y = var_44608_to_fp16)[name = tensor<string, []>("aw_3793_cast")];
            tensor<string, []> var_44611_equation_0 = const()[name = tensor<string, []>("op_44611_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44611_cast = einsum(equation = var_44611_equation_0, values = (var_44453_cast, var_44370_cast))[name = tensor<string, []>("op_44611_cast")];
            tensor<fp16, []> var_44612_to_fp16 = const()[name = tensor<string, []>("op_44612_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3795_cast = mul(x = var_44611_cast, y = var_44612_to_fp16)[name = tensor<string, []>("aw_3795_cast")];
            tensor<string, []> var_44615_equation_0 = const()[name = tensor<string, []>("op_44615_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44615_cast = einsum(equation = var_44615_equation_0, values = (var_44457_cast, var_44374_cast))[name = tensor<string, []>("op_44615_cast")];
            tensor<fp16, []> var_44616_to_fp16 = const()[name = tensor<string, []>("op_44616_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3797_cast = mul(x = var_44615_cast, y = var_44616_to_fp16)[name = tensor<string, []>("aw_3797_cast")];
            tensor<string, []> var_44619_equation_0 = const()[name = tensor<string, []>("op_44619_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44619_cast = einsum(equation = var_44619_equation_0, values = (var_44461_cast, var_44378_cast))[name = tensor<string, []>("op_44619_cast")];
            tensor<fp16, []> var_44620_to_fp16 = const()[name = tensor<string, []>("op_44620_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3799_cast = mul(x = var_44619_cast, y = var_44620_to_fp16)[name = tensor<string, []>("aw_3799_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44622_cast = softmax(axis = var_30385, x = aw_3761_cast)[name = tensor<string, []>("op_44622_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44623_cast = softmax(axis = var_30385, x = aw_3763_cast)[name = tensor<string, []>("op_44623_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44624_cast = softmax(axis = var_30385, x = aw_3765_cast)[name = tensor<string, []>("op_44624_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44625_cast = softmax(axis = var_30385, x = aw_3767_cast)[name = tensor<string, []>("op_44625_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44626_cast = softmax(axis = var_30385, x = aw_3769_cast)[name = tensor<string, []>("op_44626_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44627_cast = softmax(axis = var_30385, x = aw_3771_cast)[name = tensor<string, []>("op_44627_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44628_cast = softmax(axis = var_30385, x = aw_3773_cast)[name = tensor<string, []>("op_44628_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44629_cast = softmax(axis = var_30385, x = aw_3775_cast)[name = tensor<string, []>("op_44629_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44630_cast = softmax(axis = var_30385, x = aw_3777_cast)[name = tensor<string, []>("op_44630_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44631_cast = softmax(axis = var_30385, x = aw_3779_cast)[name = tensor<string, []>("op_44631_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44632_cast = softmax(axis = var_30385, x = aw_3781_cast)[name = tensor<string, []>("op_44632_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44633_cast = softmax(axis = var_30385, x = aw_3783_cast)[name = tensor<string, []>("op_44633_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44634_cast = softmax(axis = var_30385, x = aw_3785_cast)[name = tensor<string, []>("op_44634_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44635_cast = softmax(axis = var_30385, x = aw_3787_cast)[name = tensor<string, []>("op_44635_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44636_cast = softmax(axis = var_30385, x = aw_3789_cast)[name = tensor<string, []>("op_44636_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44637_cast = softmax(axis = var_30385, x = aw_3791_cast)[name = tensor<string, []>("op_44637_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44638_cast = softmax(axis = var_30385, x = aw_3793_cast)[name = tensor<string, []>("op_44638_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44639_cast = softmax(axis = var_30385, x = aw_3795_cast)[name = tensor<string, []>("op_44639_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44640_cast = softmax(axis = var_30385, x = aw_3797_cast)[name = tensor<string, []>("op_44640_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_44641_cast = softmax(axis = var_30385, x = aw_3799_cast)[name = tensor<string, []>("op_44641_cast")];
            tensor<string, []> var_44643_equation_0 = const()[name = tensor<string, []>("op_44643_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44643_cast = einsum(equation = var_44643_equation_0, values = (var_44463_cast, var_44622_cast))[name = tensor<string, []>("op_44643_cast")];
            tensor<string, []> var_44645_equation_0 = const()[name = tensor<string, []>("op_44645_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44645_cast = einsum(equation = var_44645_equation_0, values = (var_44467_cast, var_44623_cast))[name = tensor<string, []>("op_44645_cast")];
            tensor<string, []> var_44647_equation_0 = const()[name = tensor<string, []>("op_44647_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44647_cast = einsum(equation = var_44647_equation_0, values = (var_44471_cast, var_44624_cast))[name = tensor<string, []>("op_44647_cast")];
            tensor<string, []> var_44649_equation_0 = const()[name = tensor<string, []>("op_44649_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44649_cast = einsum(equation = var_44649_equation_0, values = (var_44475_cast, var_44625_cast))[name = tensor<string, []>("op_44649_cast")];
            tensor<string, []> var_44651_equation_0 = const()[name = tensor<string, []>("op_44651_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44651_cast = einsum(equation = var_44651_equation_0, values = (var_44479_cast, var_44626_cast))[name = tensor<string, []>("op_44651_cast")];
            tensor<string, []> var_44653_equation_0 = const()[name = tensor<string, []>("op_44653_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44653_cast = einsum(equation = var_44653_equation_0, values = (var_44483_cast, var_44627_cast))[name = tensor<string, []>("op_44653_cast")];
            tensor<string, []> var_44655_equation_0 = const()[name = tensor<string, []>("op_44655_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44655_cast = einsum(equation = var_44655_equation_0, values = (var_44487_cast, var_44628_cast))[name = tensor<string, []>("op_44655_cast")];
            tensor<string, []> var_44657_equation_0 = const()[name = tensor<string, []>("op_44657_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44657_cast = einsum(equation = var_44657_equation_0, values = (var_44491_cast, var_44629_cast))[name = tensor<string, []>("op_44657_cast")];
            tensor<string, []> var_44659_equation_0 = const()[name = tensor<string, []>("op_44659_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44659_cast = einsum(equation = var_44659_equation_0, values = (var_44495_cast, var_44630_cast))[name = tensor<string, []>("op_44659_cast")];
            tensor<string, []> var_44661_equation_0 = const()[name = tensor<string, []>("op_44661_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44661_cast = einsum(equation = var_44661_equation_0, values = (var_44499_cast, var_44631_cast))[name = tensor<string, []>("op_44661_cast")];
            tensor<string, []> var_44663_equation_0 = const()[name = tensor<string, []>("op_44663_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44663_cast = einsum(equation = var_44663_equation_0, values = (var_44503_cast, var_44632_cast))[name = tensor<string, []>("op_44663_cast")];
            tensor<string, []> var_44665_equation_0 = const()[name = tensor<string, []>("op_44665_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44665_cast = einsum(equation = var_44665_equation_0, values = (var_44507_cast, var_44633_cast))[name = tensor<string, []>("op_44665_cast")];
            tensor<string, []> var_44667_equation_0 = const()[name = tensor<string, []>("op_44667_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44667_cast = einsum(equation = var_44667_equation_0, values = (var_44511_cast, var_44634_cast))[name = tensor<string, []>("op_44667_cast")];
            tensor<string, []> var_44669_equation_0 = const()[name = tensor<string, []>("op_44669_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44669_cast = einsum(equation = var_44669_equation_0, values = (var_44515_cast, var_44635_cast))[name = tensor<string, []>("op_44669_cast")];
            tensor<string, []> var_44671_equation_0 = const()[name = tensor<string, []>("op_44671_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44671_cast = einsum(equation = var_44671_equation_0, values = (var_44519_cast, var_44636_cast))[name = tensor<string, []>("op_44671_cast")];
            tensor<string, []> var_44673_equation_0 = const()[name = tensor<string, []>("op_44673_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44673_cast = einsum(equation = var_44673_equation_0, values = (var_44523_cast, var_44637_cast))[name = tensor<string, []>("op_44673_cast")];
            tensor<string, []> var_44675_equation_0 = const()[name = tensor<string, []>("op_44675_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44675_cast = einsum(equation = var_44675_equation_0, values = (var_44527_cast, var_44638_cast))[name = tensor<string, []>("op_44675_cast")];
            tensor<string, []> var_44677_equation_0 = const()[name = tensor<string, []>("op_44677_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44677_cast = einsum(equation = var_44677_equation_0, values = (var_44531_cast, var_44639_cast))[name = tensor<string, []>("op_44677_cast")];
            tensor<string, []> var_44679_equation_0 = const()[name = tensor<string, []>("op_44679_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44679_cast = einsum(equation = var_44679_equation_0, values = (var_44535_cast, var_44640_cast))[name = tensor<string, []>("op_44679_cast")];
            tensor<string, []> var_44681_equation_0 = const()[name = tensor<string, []>("op_44681_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_44681_cast = einsum(equation = var_44681_equation_0, values = (var_44539_cast, var_44641_cast))[name = tensor<string, []>("op_44681_cast")];
            tensor<bool, []> input_579_interleave_0 = const()[name = tensor<string, []>("input_579_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_579_cast = concat(axis = var_30385, interleave = input_579_interleave_0, values = (var_44643_cast, var_44645_cast, var_44647_cast, var_44649_cast, var_44651_cast, var_44653_cast, var_44655_cast, var_44657_cast, var_44659_cast, var_44661_cast, var_44663_cast, var_44665_cast, var_44667_cast, var_44669_cast, var_44671_cast, var_44673_cast, var_44675_cast, var_44677_cast, var_44679_cast, var_44681_cast))[name = tensor<string, []>("input_579_cast")];
            tensor<int32, [2]> var_44687 = const()[name = tensor<string, []>("op_44687"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_44689 = const()[name = tensor<string, []>("op_44689"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_44691_pad_type_0 = const()[name = tensor<string, []>("op_44691_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_44691_pad_0 = const()[name = tensor<string, []>("op_44691_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3745846208)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3749123072)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_44691_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_out_0_bias_to_fp16, dilations = var_44689, groups = var_30385, pad = var_44691_pad_0, pad_type = var_44691_pad_type_0, strides = var_44687, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn1_to_out_0_weight_to_fp16, x = input_579_cast)[name = tensor<string, []>("op_44691_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_297_cast = add(x = var_44691_cast, y = inputs_295_cast)[name = tensor<string, []>("inputs_297_cast")];
            tensor<int32, [1]> var_44695 = const()[name = tensor<string, []>("op_44695"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_297_cast = reduce_mean(axes = var_44695, keep_dims = var_30380, x = inputs_297_cast)[name = tensor<string, []>("channels_mean_297_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_297_cast = sub(x = inputs_297_cast, y = channels_mean_297_cast)[name = tensor<string, []>("zero_mean_297_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_297_cast = mul(x = zero_mean_297_cast, y = zero_mean_297_cast)[name = tensor<string, []>("zero_mean_sq_297_cast")];
            tensor<int32, [1]> var_44699 = const()[name = tensor<string, []>("op_44699"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_44700_cast = reduce_mean(axes = var_44699, keep_dims = var_30380, x = zero_mean_sq_297_cast)[name = tensor<string, []>("op_44700_cast")];
            tensor<fp16, []> var_44701_to_fp16 = const()[name = tensor<string, []>("op_44701_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_44702_cast = add(x = var_44700_cast, y = var_44701_to_fp16)[name = tensor<string, []>("op_44702_cast")];
            tensor<fp16, []> denom_297_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_297_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_297_cast = rsqrt(epsilon = denom_297_epsilon_0_to_fp16, x = var_44702_cast)[name = tensor<string, []>("denom_297_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_297_cast = mul(x = zero_mean_297_cast, y = denom_297_cast)[name = tensor<string, []>("out_297_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_44706_to_fp16 = const()[name = tensor<string, []>("op_44706_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3749125696)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_44707_cast = add(x = out_297_cast, y = var_44706_to_fp16)[name = tensor<string, []>("op_44707_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_44709_to_fp16 = const()[name = tensor<string, []>("op_44709_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3749128320)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_393_cast = mul(x = var_44707_cast, y = var_44709_to_fp16)[name = tensor<string, []>("hidden_states_393_cast")];
            tensor<int32, [2]> var_44716 = const()[name = tensor<string, []>("op_44716"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_44718 = const()[name = tensor<string, []>("op_44718"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_199_pad_type_0 = const()[name = tensor<string, []>("q_199_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_199_pad_0 = const()[name = tensor<string, []>("q_199_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3749130944)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_199_cast = conv(dilations = var_44718, groups = var_30385, pad = q_199_pad_0, pad_type = q_199_pad_type_0, strides = var_44716, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_q_weight_to_fp16, x = hidden_states_393_cast)[name = tensor<string, []>("q_199_cast")];
            tensor<int32, [2]> var_44722 = const()[name = tensor<string, []>("op_44722"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_44724 = const()[name = tensor<string, []>("op_44724"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_397_pad_type_0 = const()[name = tensor<string, []>("k_397_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_397_pad_0 = const()[name = tensor<string, []>("k_397_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3752407808)))];
            tensor<fp16, [2, 1280, 1, 77]> k_397_cast = conv(dilations = var_44724, groups = var_30385, pad = k_397_pad_0, pad_type = k_397_pad_type_0, strides = var_44722, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_397_cast")];
            tensor<int32, [2]> var_44728 = const()[name = tensor<string, []>("op_44728"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_44730 = const()[name = tensor<string, []>("op_44730"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_199_pad_type_0 = const()[name = tensor<string, []>("v_199_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_199_pad_0 = const()[name = tensor<string, []>("v_199_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3757650752)))];
            tensor<fp16, [2, 1280, 1, 77]> v_199_cast = conv(dilations = var_44730, groups = var_30385, pad = v_199_pad_0, pad_type = v_199_pad_type_0, strides = var_44728, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_199_cast")];
            tensor<int32, [4]> var_44734_begin_0 = const()[name = tensor<string, []>("op_44734_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_44734_end_0 = const()[name = tensor<string, []>("op_44734_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_44734_end_mask_0 = const()[name = tensor<string, []>("op_44734_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44734_cast = slice_by_index(begin = var_44734_begin_0, end = var_44734_end_0, end_mask = var_44734_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44734_cast")];
            tensor<int32, [4]> var_44738_begin_0 = const()[name = tensor<string, []>("op_44738_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_44738_end_0 = const()[name = tensor<string, []>("op_44738_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_44738_end_mask_0 = const()[name = tensor<string, []>("op_44738_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44738_cast = slice_by_index(begin = var_44738_begin_0, end = var_44738_end_0, end_mask = var_44738_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44738_cast")];
            tensor<int32, [4]> var_44742_begin_0 = const()[name = tensor<string, []>("op_44742_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_44742_end_0 = const()[name = tensor<string, []>("op_44742_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_44742_end_mask_0 = const()[name = tensor<string, []>("op_44742_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44742_cast = slice_by_index(begin = var_44742_begin_0, end = var_44742_end_0, end_mask = var_44742_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44742_cast")];
            tensor<int32, [4]> var_44746_begin_0 = const()[name = tensor<string, []>("op_44746_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_44746_end_0 = const()[name = tensor<string, []>("op_44746_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_44746_end_mask_0 = const()[name = tensor<string, []>("op_44746_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44746_cast = slice_by_index(begin = var_44746_begin_0, end = var_44746_end_0, end_mask = var_44746_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44746_cast")];
            tensor<int32, [4]> var_44750_begin_0 = const()[name = tensor<string, []>("op_44750_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_44750_end_0 = const()[name = tensor<string, []>("op_44750_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_44750_end_mask_0 = const()[name = tensor<string, []>("op_44750_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44750_cast = slice_by_index(begin = var_44750_begin_0, end = var_44750_end_0, end_mask = var_44750_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44750_cast")];
            tensor<int32, [4]> var_44754_begin_0 = const()[name = tensor<string, []>("op_44754_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_44754_end_0 = const()[name = tensor<string, []>("op_44754_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_44754_end_mask_0 = const()[name = tensor<string, []>("op_44754_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44754_cast = slice_by_index(begin = var_44754_begin_0, end = var_44754_end_0, end_mask = var_44754_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44754_cast")];
            tensor<int32, [4]> var_44758_begin_0 = const()[name = tensor<string, []>("op_44758_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_44758_end_0 = const()[name = tensor<string, []>("op_44758_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_44758_end_mask_0 = const()[name = tensor<string, []>("op_44758_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44758_cast = slice_by_index(begin = var_44758_begin_0, end = var_44758_end_0, end_mask = var_44758_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44758_cast")];
            tensor<int32, [4]> var_44762_begin_0 = const()[name = tensor<string, []>("op_44762_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_44762_end_0 = const()[name = tensor<string, []>("op_44762_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_44762_end_mask_0 = const()[name = tensor<string, []>("op_44762_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44762_cast = slice_by_index(begin = var_44762_begin_0, end = var_44762_end_0, end_mask = var_44762_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44762_cast")];
            tensor<int32, [4]> var_44766_begin_0 = const()[name = tensor<string, []>("op_44766_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_44766_end_0 = const()[name = tensor<string, []>("op_44766_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_44766_end_mask_0 = const()[name = tensor<string, []>("op_44766_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44766_cast = slice_by_index(begin = var_44766_begin_0, end = var_44766_end_0, end_mask = var_44766_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44766_cast")];
            tensor<int32, [4]> var_44770_begin_0 = const()[name = tensor<string, []>("op_44770_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_44770_end_0 = const()[name = tensor<string, []>("op_44770_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_44770_end_mask_0 = const()[name = tensor<string, []>("op_44770_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44770_cast = slice_by_index(begin = var_44770_begin_0, end = var_44770_end_0, end_mask = var_44770_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44770_cast")];
            tensor<int32, [4]> var_44774_begin_0 = const()[name = tensor<string, []>("op_44774_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_44774_end_0 = const()[name = tensor<string, []>("op_44774_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_44774_end_mask_0 = const()[name = tensor<string, []>("op_44774_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44774_cast = slice_by_index(begin = var_44774_begin_0, end = var_44774_end_0, end_mask = var_44774_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44774_cast")];
            tensor<int32, [4]> var_44778_begin_0 = const()[name = tensor<string, []>("op_44778_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_44778_end_0 = const()[name = tensor<string, []>("op_44778_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_44778_end_mask_0 = const()[name = tensor<string, []>("op_44778_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44778_cast = slice_by_index(begin = var_44778_begin_0, end = var_44778_end_0, end_mask = var_44778_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44778_cast")];
            tensor<int32, [4]> var_44782_begin_0 = const()[name = tensor<string, []>("op_44782_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_44782_end_0 = const()[name = tensor<string, []>("op_44782_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_44782_end_mask_0 = const()[name = tensor<string, []>("op_44782_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44782_cast = slice_by_index(begin = var_44782_begin_0, end = var_44782_end_0, end_mask = var_44782_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44782_cast")];
            tensor<int32, [4]> var_44786_begin_0 = const()[name = tensor<string, []>("op_44786_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_44786_end_0 = const()[name = tensor<string, []>("op_44786_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_44786_end_mask_0 = const()[name = tensor<string, []>("op_44786_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44786_cast = slice_by_index(begin = var_44786_begin_0, end = var_44786_end_0, end_mask = var_44786_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44786_cast")];
            tensor<int32, [4]> var_44790_begin_0 = const()[name = tensor<string, []>("op_44790_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_44790_end_0 = const()[name = tensor<string, []>("op_44790_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_44790_end_mask_0 = const()[name = tensor<string, []>("op_44790_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44790_cast = slice_by_index(begin = var_44790_begin_0, end = var_44790_end_0, end_mask = var_44790_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44790_cast")];
            tensor<int32, [4]> var_44794_begin_0 = const()[name = tensor<string, []>("op_44794_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_44794_end_0 = const()[name = tensor<string, []>("op_44794_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_44794_end_mask_0 = const()[name = tensor<string, []>("op_44794_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44794_cast = slice_by_index(begin = var_44794_begin_0, end = var_44794_end_0, end_mask = var_44794_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44794_cast")];
            tensor<int32, [4]> var_44798_begin_0 = const()[name = tensor<string, []>("op_44798_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_44798_end_0 = const()[name = tensor<string, []>("op_44798_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_44798_end_mask_0 = const()[name = tensor<string, []>("op_44798_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44798_cast = slice_by_index(begin = var_44798_begin_0, end = var_44798_end_0, end_mask = var_44798_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44798_cast")];
            tensor<int32, [4]> var_44802_begin_0 = const()[name = tensor<string, []>("op_44802_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_44802_end_0 = const()[name = tensor<string, []>("op_44802_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_44802_end_mask_0 = const()[name = tensor<string, []>("op_44802_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44802_cast = slice_by_index(begin = var_44802_begin_0, end = var_44802_end_0, end_mask = var_44802_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44802_cast")];
            tensor<int32, [4]> var_44806_begin_0 = const()[name = tensor<string, []>("op_44806_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_44806_end_0 = const()[name = tensor<string, []>("op_44806_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_44806_end_mask_0 = const()[name = tensor<string, []>("op_44806_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44806_cast = slice_by_index(begin = var_44806_begin_0, end = var_44806_end_0, end_mask = var_44806_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44806_cast")];
            tensor<int32, [4]> var_44810_begin_0 = const()[name = tensor<string, []>("op_44810_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_44810_end_0 = const()[name = tensor<string, []>("op_44810_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_44810_end_mask_0 = const()[name = tensor<string, []>("op_44810_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_44810_cast = slice_by_index(begin = var_44810_begin_0, end = var_44810_end_0, end_mask = var_44810_end_mask_0, x = q_199_cast)[name = tensor<string, []>("op_44810_cast")];
            tensor<int32, [4]> k_399_perm_0 = const()[name = tensor<string, []>("k_399_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_44817_begin_0 = const()[name = tensor<string, []>("op_44817_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_44817_end_0 = const()[name = tensor<string, []>("op_44817_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_44817_end_mask_0 = const()[name = tensor<string, []>("op_44817_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_40 = transpose(perm = k_399_perm_0, x = k_397_cast)[name = tensor<string, []>("transpose_40")];
            tensor<fp16, [2, 77, 1, 64]> var_44817_cast = slice_by_index(begin = var_44817_begin_0, end = var_44817_end_0, end_mask = var_44817_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44817_cast")];
            tensor<int32, [4]> var_44821_begin_0 = const()[name = tensor<string, []>("op_44821_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_44821_end_0 = const()[name = tensor<string, []>("op_44821_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_44821_end_mask_0 = const()[name = tensor<string, []>("op_44821_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44821_cast = slice_by_index(begin = var_44821_begin_0, end = var_44821_end_0, end_mask = var_44821_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44821_cast")];
            tensor<int32, [4]> var_44825_begin_0 = const()[name = tensor<string, []>("op_44825_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_44825_end_0 = const()[name = tensor<string, []>("op_44825_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_44825_end_mask_0 = const()[name = tensor<string, []>("op_44825_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44825_cast = slice_by_index(begin = var_44825_begin_0, end = var_44825_end_0, end_mask = var_44825_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44825_cast")];
            tensor<int32, [4]> var_44829_begin_0 = const()[name = tensor<string, []>("op_44829_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_44829_end_0 = const()[name = tensor<string, []>("op_44829_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_44829_end_mask_0 = const()[name = tensor<string, []>("op_44829_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44829_cast = slice_by_index(begin = var_44829_begin_0, end = var_44829_end_0, end_mask = var_44829_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44829_cast")];
            tensor<int32, [4]> var_44833_begin_0 = const()[name = tensor<string, []>("op_44833_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_44833_end_0 = const()[name = tensor<string, []>("op_44833_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_44833_end_mask_0 = const()[name = tensor<string, []>("op_44833_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44833_cast = slice_by_index(begin = var_44833_begin_0, end = var_44833_end_0, end_mask = var_44833_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44833_cast")];
            tensor<int32, [4]> var_44837_begin_0 = const()[name = tensor<string, []>("op_44837_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_44837_end_0 = const()[name = tensor<string, []>("op_44837_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_44837_end_mask_0 = const()[name = tensor<string, []>("op_44837_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44837_cast = slice_by_index(begin = var_44837_begin_0, end = var_44837_end_0, end_mask = var_44837_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44837_cast")];
            tensor<int32, [4]> var_44841_begin_0 = const()[name = tensor<string, []>("op_44841_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_44841_end_0 = const()[name = tensor<string, []>("op_44841_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_44841_end_mask_0 = const()[name = tensor<string, []>("op_44841_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44841_cast = slice_by_index(begin = var_44841_begin_0, end = var_44841_end_0, end_mask = var_44841_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44841_cast")];
            tensor<int32, [4]> var_44845_begin_0 = const()[name = tensor<string, []>("op_44845_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_44845_end_0 = const()[name = tensor<string, []>("op_44845_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_44845_end_mask_0 = const()[name = tensor<string, []>("op_44845_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44845_cast = slice_by_index(begin = var_44845_begin_0, end = var_44845_end_0, end_mask = var_44845_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44845_cast")];
            tensor<int32, [4]> var_44849_begin_0 = const()[name = tensor<string, []>("op_44849_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_44849_end_0 = const()[name = tensor<string, []>("op_44849_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_44849_end_mask_0 = const()[name = tensor<string, []>("op_44849_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44849_cast = slice_by_index(begin = var_44849_begin_0, end = var_44849_end_0, end_mask = var_44849_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44849_cast")];
            tensor<int32, [4]> var_44853_begin_0 = const()[name = tensor<string, []>("op_44853_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_44853_end_0 = const()[name = tensor<string, []>("op_44853_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_44853_end_mask_0 = const()[name = tensor<string, []>("op_44853_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44853_cast = slice_by_index(begin = var_44853_begin_0, end = var_44853_end_0, end_mask = var_44853_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44853_cast")];
            tensor<int32, [4]> var_44857_begin_0 = const()[name = tensor<string, []>("op_44857_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_44857_end_0 = const()[name = tensor<string, []>("op_44857_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_44857_end_mask_0 = const()[name = tensor<string, []>("op_44857_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44857_cast = slice_by_index(begin = var_44857_begin_0, end = var_44857_end_0, end_mask = var_44857_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44857_cast")];
            tensor<int32, [4]> var_44861_begin_0 = const()[name = tensor<string, []>("op_44861_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_44861_end_0 = const()[name = tensor<string, []>("op_44861_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_44861_end_mask_0 = const()[name = tensor<string, []>("op_44861_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44861_cast = slice_by_index(begin = var_44861_begin_0, end = var_44861_end_0, end_mask = var_44861_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44861_cast")];
            tensor<int32, [4]> var_44865_begin_0 = const()[name = tensor<string, []>("op_44865_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_44865_end_0 = const()[name = tensor<string, []>("op_44865_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_44865_end_mask_0 = const()[name = tensor<string, []>("op_44865_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44865_cast = slice_by_index(begin = var_44865_begin_0, end = var_44865_end_0, end_mask = var_44865_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44865_cast")];
            tensor<int32, [4]> var_44869_begin_0 = const()[name = tensor<string, []>("op_44869_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_44869_end_0 = const()[name = tensor<string, []>("op_44869_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_44869_end_mask_0 = const()[name = tensor<string, []>("op_44869_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44869_cast = slice_by_index(begin = var_44869_begin_0, end = var_44869_end_0, end_mask = var_44869_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44869_cast")];
            tensor<int32, [4]> var_44873_begin_0 = const()[name = tensor<string, []>("op_44873_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_44873_end_0 = const()[name = tensor<string, []>("op_44873_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_44873_end_mask_0 = const()[name = tensor<string, []>("op_44873_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44873_cast = slice_by_index(begin = var_44873_begin_0, end = var_44873_end_0, end_mask = var_44873_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44873_cast")];
            tensor<int32, [4]> var_44877_begin_0 = const()[name = tensor<string, []>("op_44877_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_44877_end_0 = const()[name = tensor<string, []>("op_44877_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_44877_end_mask_0 = const()[name = tensor<string, []>("op_44877_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44877_cast = slice_by_index(begin = var_44877_begin_0, end = var_44877_end_0, end_mask = var_44877_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44877_cast")];
            tensor<int32, [4]> var_44881_begin_0 = const()[name = tensor<string, []>("op_44881_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_44881_end_0 = const()[name = tensor<string, []>("op_44881_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_44881_end_mask_0 = const()[name = tensor<string, []>("op_44881_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44881_cast = slice_by_index(begin = var_44881_begin_0, end = var_44881_end_0, end_mask = var_44881_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44881_cast")];
            tensor<int32, [4]> var_44885_begin_0 = const()[name = tensor<string, []>("op_44885_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_44885_end_0 = const()[name = tensor<string, []>("op_44885_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_44885_end_mask_0 = const()[name = tensor<string, []>("op_44885_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44885_cast = slice_by_index(begin = var_44885_begin_0, end = var_44885_end_0, end_mask = var_44885_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44885_cast")];
            tensor<int32, [4]> var_44889_begin_0 = const()[name = tensor<string, []>("op_44889_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_44889_end_0 = const()[name = tensor<string, []>("op_44889_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_44889_end_mask_0 = const()[name = tensor<string, []>("op_44889_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44889_cast = slice_by_index(begin = var_44889_begin_0, end = var_44889_end_0, end_mask = var_44889_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44889_cast")];
            tensor<int32, [4]> var_44893_begin_0 = const()[name = tensor<string, []>("op_44893_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_44893_end_0 = const()[name = tensor<string, []>("op_44893_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_44893_end_mask_0 = const()[name = tensor<string, []>("op_44893_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_44893_cast = slice_by_index(begin = var_44893_begin_0, end = var_44893_end_0, end_mask = var_44893_end_mask_0, x = transpose_40)[name = tensor<string, []>("op_44893_cast")];
            tensor<int32, [4]> var_44895_begin_0 = const()[name = tensor<string, []>("op_44895_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_44895_end_0 = const()[name = tensor<string, []>("op_44895_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_44895_end_mask_0 = const()[name = tensor<string, []>("op_44895_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44895_cast = slice_by_index(begin = var_44895_begin_0, end = var_44895_end_0, end_mask = var_44895_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44895_cast")];
            tensor<int32, [4]> var_44899_begin_0 = const()[name = tensor<string, []>("op_44899_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_44899_end_0 = const()[name = tensor<string, []>("op_44899_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_44899_end_mask_0 = const()[name = tensor<string, []>("op_44899_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44899_cast = slice_by_index(begin = var_44899_begin_0, end = var_44899_end_0, end_mask = var_44899_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44899_cast")];
            tensor<int32, [4]> var_44903_begin_0 = const()[name = tensor<string, []>("op_44903_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_44903_end_0 = const()[name = tensor<string, []>("op_44903_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_44903_end_mask_0 = const()[name = tensor<string, []>("op_44903_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44903_cast = slice_by_index(begin = var_44903_begin_0, end = var_44903_end_0, end_mask = var_44903_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44903_cast")];
            tensor<int32, [4]> var_44907_begin_0 = const()[name = tensor<string, []>("op_44907_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_44907_end_0 = const()[name = tensor<string, []>("op_44907_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_44907_end_mask_0 = const()[name = tensor<string, []>("op_44907_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44907_cast = slice_by_index(begin = var_44907_begin_0, end = var_44907_end_0, end_mask = var_44907_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44907_cast")];
            tensor<int32, [4]> var_44911_begin_0 = const()[name = tensor<string, []>("op_44911_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_44911_end_0 = const()[name = tensor<string, []>("op_44911_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_44911_end_mask_0 = const()[name = tensor<string, []>("op_44911_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44911_cast = slice_by_index(begin = var_44911_begin_0, end = var_44911_end_0, end_mask = var_44911_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44911_cast")];
            tensor<int32, [4]> var_44915_begin_0 = const()[name = tensor<string, []>("op_44915_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_44915_end_0 = const()[name = tensor<string, []>("op_44915_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_44915_end_mask_0 = const()[name = tensor<string, []>("op_44915_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44915_cast = slice_by_index(begin = var_44915_begin_0, end = var_44915_end_0, end_mask = var_44915_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44915_cast")];
            tensor<int32, [4]> var_44919_begin_0 = const()[name = tensor<string, []>("op_44919_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_44919_end_0 = const()[name = tensor<string, []>("op_44919_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_44919_end_mask_0 = const()[name = tensor<string, []>("op_44919_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44919_cast = slice_by_index(begin = var_44919_begin_0, end = var_44919_end_0, end_mask = var_44919_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44919_cast")];
            tensor<int32, [4]> var_44923_begin_0 = const()[name = tensor<string, []>("op_44923_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_44923_end_0 = const()[name = tensor<string, []>("op_44923_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_44923_end_mask_0 = const()[name = tensor<string, []>("op_44923_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44923_cast = slice_by_index(begin = var_44923_begin_0, end = var_44923_end_0, end_mask = var_44923_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44923_cast")];
            tensor<int32, [4]> var_44927_begin_0 = const()[name = tensor<string, []>("op_44927_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_44927_end_0 = const()[name = tensor<string, []>("op_44927_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_44927_end_mask_0 = const()[name = tensor<string, []>("op_44927_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44927_cast = slice_by_index(begin = var_44927_begin_0, end = var_44927_end_0, end_mask = var_44927_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44927_cast")];
            tensor<int32, [4]> var_44931_begin_0 = const()[name = tensor<string, []>("op_44931_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_44931_end_0 = const()[name = tensor<string, []>("op_44931_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_44931_end_mask_0 = const()[name = tensor<string, []>("op_44931_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44931_cast = slice_by_index(begin = var_44931_begin_0, end = var_44931_end_0, end_mask = var_44931_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44931_cast")];
            tensor<int32, [4]> var_44935_begin_0 = const()[name = tensor<string, []>("op_44935_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_44935_end_0 = const()[name = tensor<string, []>("op_44935_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_44935_end_mask_0 = const()[name = tensor<string, []>("op_44935_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44935_cast = slice_by_index(begin = var_44935_begin_0, end = var_44935_end_0, end_mask = var_44935_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44935_cast")];
            tensor<int32, [4]> var_44939_begin_0 = const()[name = tensor<string, []>("op_44939_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_44939_end_0 = const()[name = tensor<string, []>("op_44939_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_44939_end_mask_0 = const()[name = tensor<string, []>("op_44939_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44939_cast = slice_by_index(begin = var_44939_begin_0, end = var_44939_end_0, end_mask = var_44939_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44939_cast")];
            tensor<int32, [4]> var_44943_begin_0 = const()[name = tensor<string, []>("op_44943_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_44943_end_0 = const()[name = tensor<string, []>("op_44943_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_44943_end_mask_0 = const()[name = tensor<string, []>("op_44943_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44943_cast = slice_by_index(begin = var_44943_begin_0, end = var_44943_end_0, end_mask = var_44943_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44943_cast")];
            tensor<int32, [4]> var_44947_begin_0 = const()[name = tensor<string, []>("op_44947_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_44947_end_0 = const()[name = tensor<string, []>("op_44947_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_44947_end_mask_0 = const()[name = tensor<string, []>("op_44947_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44947_cast = slice_by_index(begin = var_44947_begin_0, end = var_44947_end_0, end_mask = var_44947_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44947_cast")];
            tensor<int32, [4]> var_44951_begin_0 = const()[name = tensor<string, []>("op_44951_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_44951_end_0 = const()[name = tensor<string, []>("op_44951_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_44951_end_mask_0 = const()[name = tensor<string, []>("op_44951_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44951_cast = slice_by_index(begin = var_44951_begin_0, end = var_44951_end_0, end_mask = var_44951_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44951_cast")];
            tensor<int32, [4]> var_44955_begin_0 = const()[name = tensor<string, []>("op_44955_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_44955_end_0 = const()[name = tensor<string, []>("op_44955_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_44955_end_mask_0 = const()[name = tensor<string, []>("op_44955_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44955_cast = slice_by_index(begin = var_44955_begin_0, end = var_44955_end_0, end_mask = var_44955_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44955_cast")];
            tensor<int32, [4]> var_44959_begin_0 = const()[name = tensor<string, []>("op_44959_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_44959_end_0 = const()[name = tensor<string, []>("op_44959_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_44959_end_mask_0 = const()[name = tensor<string, []>("op_44959_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44959_cast = slice_by_index(begin = var_44959_begin_0, end = var_44959_end_0, end_mask = var_44959_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44959_cast")];
            tensor<int32, [4]> var_44963_begin_0 = const()[name = tensor<string, []>("op_44963_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_44963_end_0 = const()[name = tensor<string, []>("op_44963_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_44963_end_mask_0 = const()[name = tensor<string, []>("op_44963_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44963_cast = slice_by_index(begin = var_44963_begin_0, end = var_44963_end_0, end_mask = var_44963_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44963_cast")];
            tensor<int32, [4]> var_44967_begin_0 = const()[name = tensor<string, []>("op_44967_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_44967_end_0 = const()[name = tensor<string, []>("op_44967_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_44967_end_mask_0 = const()[name = tensor<string, []>("op_44967_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44967_cast = slice_by_index(begin = var_44967_begin_0, end = var_44967_end_0, end_mask = var_44967_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44967_cast")];
            tensor<int32, [4]> var_44971_begin_0 = const()[name = tensor<string, []>("op_44971_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_44971_end_0 = const()[name = tensor<string, []>("op_44971_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_44971_end_mask_0 = const()[name = tensor<string, []>("op_44971_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_44971_cast = slice_by_index(begin = var_44971_begin_0, end = var_44971_end_0, end_mask = var_44971_end_mask_0, x = v_199_cast)[name = tensor<string, []>("op_44971_cast")];
            tensor<string, []> var_44975_equation_0 = const()[name = tensor<string, []>("op_44975_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44975_cast = einsum(equation = var_44975_equation_0, values = (var_44817_cast, var_44734_cast))[name = tensor<string, []>("op_44975_cast")];
            tensor<fp16, []> var_44976_to_fp16 = const()[name = tensor<string, []>("op_44976_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3801_cast = mul(x = var_44975_cast, y = var_44976_to_fp16)[name = tensor<string, []>("aw_3801_cast")];
            tensor<string, []> var_44979_equation_0 = const()[name = tensor<string, []>("op_44979_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44979_cast = einsum(equation = var_44979_equation_0, values = (var_44821_cast, var_44738_cast))[name = tensor<string, []>("op_44979_cast")];
            tensor<fp16, []> var_44980_to_fp16 = const()[name = tensor<string, []>("op_44980_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3803_cast = mul(x = var_44979_cast, y = var_44980_to_fp16)[name = tensor<string, []>("aw_3803_cast")];
            tensor<string, []> var_44983_equation_0 = const()[name = tensor<string, []>("op_44983_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44983_cast = einsum(equation = var_44983_equation_0, values = (var_44825_cast, var_44742_cast))[name = tensor<string, []>("op_44983_cast")];
            tensor<fp16, []> var_44984_to_fp16 = const()[name = tensor<string, []>("op_44984_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3805_cast = mul(x = var_44983_cast, y = var_44984_to_fp16)[name = tensor<string, []>("aw_3805_cast")];
            tensor<string, []> var_44987_equation_0 = const()[name = tensor<string, []>("op_44987_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44987_cast = einsum(equation = var_44987_equation_0, values = (var_44829_cast, var_44746_cast))[name = tensor<string, []>("op_44987_cast")];
            tensor<fp16, []> var_44988_to_fp16 = const()[name = tensor<string, []>("op_44988_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3807_cast = mul(x = var_44987_cast, y = var_44988_to_fp16)[name = tensor<string, []>("aw_3807_cast")];
            tensor<string, []> var_44991_equation_0 = const()[name = tensor<string, []>("op_44991_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44991_cast = einsum(equation = var_44991_equation_0, values = (var_44833_cast, var_44750_cast))[name = tensor<string, []>("op_44991_cast")];
            tensor<fp16, []> var_44992_to_fp16 = const()[name = tensor<string, []>("op_44992_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3809_cast = mul(x = var_44991_cast, y = var_44992_to_fp16)[name = tensor<string, []>("aw_3809_cast")];
            tensor<string, []> var_44995_equation_0 = const()[name = tensor<string, []>("op_44995_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44995_cast = einsum(equation = var_44995_equation_0, values = (var_44837_cast, var_44754_cast))[name = tensor<string, []>("op_44995_cast")];
            tensor<fp16, []> var_44996_to_fp16 = const()[name = tensor<string, []>("op_44996_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3811_cast = mul(x = var_44995_cast, y = var_44996_to_fp16)[name = tensor<string, []>("aw_3811_cast")];
            tensor<string, []> var_44999_equation_0 = const()[name = tensor<string, []>("op_44999_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_44999_cast = einsum(equation = var_44999_equation_0, values = (var_44841_cast, var_44758_cast))[name = tensor<string, []>("op_44999_cast")];
            tensor<fp16, []> var_45000_to_fp16 = const()[name = tensor<string, []>("op_45000_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3813_cast = mul(x = var_44999_cast, y = var_45000_to_fp16)[name = tensor<string, []>("aw_3813_cast")];
            tensor<string, []> var_45003_equation_0 = const()[name = tensor<string, []>("op_45003_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45003_cast = einsum(equation = var_45003_equation_0, values = (var_44845_cast, var_44762_cast))[name = tensor<string, []>("op_45003_cast")];
            tensor<fp16, []> var_45004_to_fp16 = const()[name = tensor<string, []>("op_45004_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3815_cast = mul(x = var_45003_cast, y = var_45004_to_fp16)[name = tensor<string, []>("aw_3815_cast")];
            tensor<string, []> var_45007_equation_0 = const()[name = tensor<string, []>("op_45007_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45007_cast = einsum(equation = var_45007_equation_0, values = (var_44849_cast, var_44766_cast))[name = tensor<string, []>("op_45007_cast")];
            tensor<fp16, []> var_45008_to_fp16 = const()[name = tensor<string, []>("op_45008_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3817_cast = mul(x = var_45007_cast, y = var_45008_to_fp16)[name = tensor<string, []>("aw_3817_cast")];
            tensor<string, []> var_45011_equation_0 = const()[name = tensor<string, []>("op_45011_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45011_cast = einsum(equation = var_45011_equation_0, values = (var_44853_cast, var_44770_cast))[name = tensor<string, []>("op_45011_cast")];
            tensor<fp16, []> var_45012_to_fp16 = const()[name = tensor<string, []>("op_45012_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3819_cast = mul(x = var_45011_cast, y = var_45012_to_fp16)[name = tensor<string, []>("aw_3819_cast")];
            tensor<string, []> var_45015_equation_0 = const()[name = tensor<string, []>("op_45015_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45015_cast = einsum(equation = var_45015_equation_0, values = (var_44857_cast, var_44774_cast))[name = tensor<string, []>("op_45015_cast")];
            tensor<fp16, []> var_45016_to_fp16 = const()[name = tensor<string, []>("op_45016_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3821_cast = mul(x = var_45015_cast, y = var_45016_to_fp16)[name = tensor<string, []>("aw_3821_cast")];
            tensor<string, []> var_45019_equation_0 = const()[name = tensor<string, []>("op_45019_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45019_cast = einsum(equation = var_45019_equation_0, values = (var_44861_cast, var_44778_cast))[name = tensor<string, []>("op_45019_cast")];
            tensor<fp16, []> var_45020_to_fp16 = const()[name = tensor<string, []>("op_45020_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3823_cast = mul(x = var_45019_cast, y = var_45020_to_fp16)[name = tensor<string, []>("aw_3823_cast")];
            tensor<string, []> var_45023_equation_0 = const()[name = tensor<string, []>("op_45023_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45023_cast = einsum(equation = var_45023_equation_0, values = (var_44865_cast, var_44782_cast))[name = tensor<string, []>("op_45023_cast")];
            tensor<fp16, []> var_45024_to_fp16 = const()[name = tensor<string, []>("op_45024_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3825_cast = mul(x = var_45023_cast, y = var_45024_to_fp16)[name = tensor<string, []>("aw_3825_cast")];
            tensor<string, []> var_45027_equation_0 = const()[name = tensor<string, []>("op_45027_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45027_cast = einsum(equation = var_45027_equation_0, values = (var_44869_cast, var_44786_cast))[name = tensor<string, []>("op_45027_cast")];
            tensor<fp16, []> var_45028_to_fp16 = const()[name = tensor<string, []>("op_45028_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3827_cast = mul(x = var_45027_cast, y = var_45028_to_fp16)[name = tensor<string, []>("aw_3827_cast")];
            tensor<string, []> var_45031_equation_0 = const()[name = tensor<string, []>("op_45031_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45031_cast = einsum(equation = var_45031_equation_0, values = (var_44873_cast, var_44790_cast))[name = tensor<string, []>("op_45031_cast")];
            tensor<fp16, []> var_45032_to_fp16 = const()[name = tensor<string, []>("op_45032_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3829_cast = mul(x = var_45031_cast, y = var_45032_to_fp16)[name = tensor<string, []>("aw_3829_cast")];
            tensor<string, []> var_45035_equation_0 = const()[name = tensor<string, []>("op_45035_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45035_cast = einsum(equation = var_45035_equation_0, values = (var_44877_cast, var_44794_cast))[name = tensor<string, []>("op_45035_cast")];
            tensor<fp16, []> var_45036_to_fp16 = const()[name = tensor<string, []>("op_45036_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3831_cast = mul(x = var_45035_cast, y = var_45036_to_fp16)[name = tensor<string, []>("aw_3831_cast")];
            tensor<string, []> var_45039_equation_0 = const()[name = tensor<string, []>("op_45039_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45039_cast = einsum(equation = var_45039_equation_0, values = (var_44881_cast, var_44798_cast))[name = tensor<string, []>("op_45039_cast")];
            tensor<fp16, []> var_45040_to_fp16 = const()[name = tensor<string, []>("op_45040_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3833_cast = mul(x = var_45039_cast, y = var_45040_to_fp16)[name = tensor<string, []>("aw_3833_cast")];
            tensor<string, []> var_45043_equation_0 = const()[name = tensor<string, []>("op_45043_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45043_cast = einsum(equation = var_45043_equation_0, values = (var_44885_cast, var_44802_cast))[name = tensor<string, []>("op_45043_cast")];
            tensor<fp16, []> var_45044_to_fp16 = const()[name = tensor<string, []>("op_45044_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3835_cast = mul(x = var_45043_cast, y = var_45044_to_fp16)[name = tensor<string, []>("aw_3835_cast")];
            tensor<string, []> var_45047_equation_0 = const()[name = tensor<string, []>("op_45047_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45047_cast = einsum(equation = var_45047_equation_0, values = (var_44889_cast, var_44806_cast))[name = tensor<string, []>("op_45047_cast")];
            tensor<fp16, []> var_45048_to_fp16 = const()[name = tensor<string, []>("op_45048_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3837_cast = mul(x = var_45047_cast, y = var_45048_to_fp16)[name = tensor<string, []>("aw_3837_cast")];
            tensor<string, []> var_45051_equation_0 = const()[name = tensor<string, []>("op_45051_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45051_cast = einsum(equation = var_45051_equation_0, values = (var_44893_cast, var_44810_cast))[name = tensor<string, []>("op_45051_cast")];
            tensor<fp16, []> var_45052_to_fp16 = const()[name = tensor<string, []>("op_45052_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3839_cast = mul(x = var_45051_cast, y = var_45052_to_fp16)[name = tensor<string, []>("aw_3839_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45054_cast = softmax(axis = var_30385, x = aw_3801_cast)[name = tensor<string, []>("op_45054_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45055_cast = softmax(axis = var_30385, x = aw_3803_cast)[name = tensor<string, []>("op_45055_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45056_cast = softmax(axis = var_30385, x = aw_3805_cast)[name = tensor<string, []>("op_45056_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45057_cast = softmax(axis = var_30385, x = aw_3807_cast)[name = tensor<string, []>("op_45057_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45058_cast = softmax(axis = var_30385, x = aw_3809_cast)[name = tensor<string, []>("op_45058_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45059_cast = softmax(axis = var_30385, x = aw_3811_cast)[name = tensor<string, []>("op_45059_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45060_cast = softmax(axis = var_30385, x = aw_3813_cast)[name = tensor<string, []>("op_45060_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45061_cast = softmax(axis = var_30385, x = aw_3815_cast)[name = tensor<string, []>("op_45061_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45062_cast = softmax(axis = var_30385, x = aw_3817_cast)[name = tensor<string, []>("op_45062_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45063_cast = softmax(axis = var_30385, x = aw_3819_cast)[name = tensor<string, []>("op_45063_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45064_cast = softmax(axis = var_30385, x = aw_3821_cast)[name = tensor<string, []>("op_45064_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45065_cast = softmax(axis = var_30385, x = aw_3823_cast)[name = tensor<string, []>("op_45065_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45066_cast = softmax(axis = var_30385, x = aw_3825_cast)[name = tensor<string, []>("op_45066_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45067_cast = softmax(axis = var_30385, x = aw_3827_cast)[name = tensor<string, []>("op_45067_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45068_cast = softmax(axis = var_30385, x = aw_3829_cast)[name = tensor<string, []>("op_45068_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45069_cast = softmax(axis = var_30385, x = aw_3831_cast)[name = tensor<string, []>("op_45069_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45070_cast = softmax(axis = var_30385, x = aw_3833_cast)[name = tensor<string, []>("op_45070_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45071_cast = softmax(axis = var_30385, x = aw_3835_cast)[name = tensor<string, []>("op_45071_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45072_cast = softmax(axis = var_30385, x = aw_3837_cast)[name = tensor<string, []>("op_45072_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45073_cast = softmax(axis = var_30385, x = aw_3839_cast)[name = tensor<string, []>("op_45073_cast")];
            tensor<string, []> var_45075_equation_0 = const()[name = tensor<string, []>("op_45075_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45075_cast = einsum(equation = var_45075_equation_0, values = (var_44895_cast, var_45054_cast))[name = tensor<string, []>("op_45075_cast")];
            tensor<string, []> var_45077_equation_0 = const()[name = tensor<string, []>("op_45077_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45077_cast = einsum(equation = var_45077_equation_0, values = (var_44899_cast, var_45055_cast))[name = tensor<string, []>("op_45077_cast")];
            tensor<string, []> var_45079_equation_0 = const()[name = tensor<string, []>("op_45079_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45079_cast = einsum(equation = var_45079_equation_0, values = (var_44903_cast, var_45056_cast))[name = tensor<string, []>("op_45079_cast")];
            tensor<string, []> var_45081_equation_0 = const()[name = tensor<string, []>("op_45081_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45081_cast = einsum(equation = var_45081_equation_0, values = (var_44907_cast, var_45057_cast))[name = tensor<string, []>("op_45081_cast")];
            tensor<string, []> var_45083_equation_0 = const()[name = tensor<string, []>("op_45083_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45083_cast = einsum(equation = var_45083_equation_0, values = (var_44911_cast, var_45058_cast))[name = tensor<string, []>("op_45083_cast")];
            tensor<string, []> var_45085_equation_0 = const()[name = tensor<string, []>("op_45085_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45085_cast = einsum(equation = var_45085_equation_0, values = (var_44915_cast, var_45059_cast))[name = tensor<string, []>("op_45085_cast")];
            tensor<string, []> var_45087_equation_0 = const()[name = tensor<string, []>("op_45087_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45087_cast = einsum(equation = var_45087_equation_0, values = (var_44919_cast, var_45060_cast))[name = tensor<string, []>("op_45087_cast")];
            tensor<string, []> var_45089_equation_0 = const()[name = tensor<string, []>("op_45089_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45089_cast = einsum(equation = var_45089_equation_0, values = (var_44923_cast, var_45061_cast))[name = tensor<string, []>("op_45089_cast")];
            tensor<string, []> var_45091_equation_0 = const()[name = tensor<string, []>("op_45091_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45091_cast = einsum(equation = var_45091_equation_0, values = (var_44927_cast, var_45062_cast))[name = tensor<string, []>("op_45091_cast")];
            tensor<string, []> var_45093_equation_0 = const()[name = tensor<string, []>("op_45093_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45093_cast = einsum(equation = var_45093_equation_0, values = (var_44931_cast, var_45063_cast))[name = tensor<string, []>("op_45093_cast")];
            tensor<string, []> var_45095_equation_0 = const()[name = tensor<string, []>("op_45095_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45095_cast = einsum(equation = var_45095_equation_0, values = (var_44935_cast, var_45064_cast))[name = tensor<string, []>("op_45095_cast")];
            tensor<string, []> var_45097_equation_0 = const()[name = tensor<string, []>("op_45097_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45097_cast = einsum(equation = var_45097_equation_0, values = (var_44939_cast, var_45065_cast))[name = tensor<string, []>("op_45097_cast")];
            tensor<string, []> var_45099_equation_0 = const()[name = tensor<string, []>("op_45099_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45099_cast = einsum(equation = var_45099_equation_0, values = (var_44943_cast, var_45066_cast))[name = tensor<string, []>("op_45099_cast")];
            tensor<string, []> var_45101_equation_0 = const()[name = tensor<string, []>("op_45101_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45101_cast = einsum(equation = var_45101_equation_0, values = (var_44947_cast, var_45067_cast))[name = tensor<string, []>("op_45101_cast")];
            tensor<string, []> var_45103_equation_0 = const()[name = tensor<string, []>("op_45103_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45103_cast = einsum(equation = var_45103_equation_0, values = (var_44951_cast, var_45068_cast))[name = tensor<string, []>("op_45103_cast")];
            tensor<string, []> var_45105_equation_0 = const()[name = tensor<string, []>("op_45105_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45105_cast = einsum(equation = var_45105_equation_0, values = (var_44955_cast, var_45069_cast))[name = tensor<string, []>("op_45105_cast")];
            tensor<string, []> var_45107_equation_0 = const()[name = tensor<string, []>("op_45107_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45107_cast = einsum(equation = var_45107_equation_0, values = (var_44959_cast, var_45070_cast))[name = tensor<string, []>("op_45107_cast")];
            tensor<string, []> var_45109_equation_0 = const()[name = tensor<string, []>("op_45109_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45109_cast = einsum(equation = var_45109_equation_0, values = (var_44963_cast, var_45071_cast))[name = tensor<string, []>("op_45109_cast")];
            tensor<string, []> var_45111_equation_0 = const()[name = tensor<string, []>("op_45111_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45111_cast = einsum(equation = var_45111_equation_0, values = (var_44967_cast, var_45072_cast))[name = tensor<string, []>("op_45111_cast")];
            tensor<string, []> var_45113_equation_0 = const()[name = tensor<string, []>("op_45113_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45113_cast = einsum(equation = var_45113_equation_0, values = (var_44971_cast, var_45073_cast))[name = tensor<string, []>("op_45113_cast")];
            tensor<bool, []> input_581_interleave_0 = const()[name = tensor<string, []>("input_581_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_581_cast = concat(axis = var_30385, interleave = input_581_interleave_0, values = (var_45075_cast, var_45077_cast, var_45079_cast, var_45081_cast, var_45083_cast, var_45085_cast, var_45087_cast, var_45089_cast, var_45091_cast, var_45093_cast, var_45095_cast, var_45097_cast, var_45099_cast, var_45101_cast, var_45103_cast, var_45105_cast, var_45107_cast, var_45109_cast, var_45111_cast, var_45113_cast))[name = tensor<string, []>("input_581_cast")];
            tensor<int32, [2]> var_45119 = const()[name = tensor<string, []>("op_45119"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_45121 = const()[name = tensor<string, []>("op_45121"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_45123_pad_type_0 = const()[name = tensor<string, []>("op_45123_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_45123_pad_0 = const()[name = tensor<string, []>("op_45123_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3762893696)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3766170560)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_45123_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_out_0_bias_to_fp16, dilations = var_45121, groups = var_30385, pad = var_45123_pad_0, pad_type = var_45123_pad_type_0, strides = var_45119, weight = up_blocks_0_attentions_1_transformer_blocks_5_attn2_to_out_0_weight_to_fp16, x = input_581_cast)[name = tensor<string, []>("op_45123_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_299_cast = add(x = var_45123_cast, y = inputs_297_cast)[name = tensor<string, []>("inputs_299_cast")];
            tensor<int32, [1]> var_45127 = const()[name = tensor<string, []>("op_45127"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_299_cast = reduce_mean(axes = var_45127, keep_dims = var_30380, x = inputs_299_cast)[name = tensor<string, []>("channels_mean_299_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_299_cast = sub(x = inputs_299_cast, y = channels_mean_299_cast)[name = tensor<string, []>("zero_mean_299_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_299_cast = mul(x = zero_mean_299_cast, y = zero_mean_299_cast)[name = tensor<string, []>("zero_mean_sq_299_cast")];
            tensor<int32, [1]> var_45131 = const()[name = tensor<string, []>("op_45131"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_45132_cast = reduce_mean(axes = var_45131, keep_dims = var_30380, x = zero_mean_sq_299_cast)[name = tensor<string, []>("op_45132_cast")];
            tensor<fp16, []> var_45133_to_fp16 = const()[name = tensor<string, []>("op_45133_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_45134_cast = add(x = var_45132_cast, y = var_45133_to_fp16)[name = tensor<string, []>("op_45134_cast")];
            tensor<fp16, []> denom_299_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_299_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_299_cast = rsqrt(epsilon = denom_299_epsilon_0_to_fp16, x = var_45134_cast)[name = tensor<string, []>("denom_299_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_299_cast = mul(x = zero_mean_299_cast, y = denom_299_cast)[name = tensor<string, []>("out_299_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_45138_to_fp16 = const()[name = tensor<string, []>("op_45138_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3766173184)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_45139_cast = add(x = out_299_cast, y = var_45138_to_fp16)[name = tensor<string, []>("op_45139_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_45141_to_fp16 = const()[name = tensor<string, []>("op_45141_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3766175808)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_583_cast = mul(x = var_45139_cast, y = var_45141_to_fp16)[name = tensor<string, []>("input_583_cast")];
            tensor<int32, [2]> var_45149 = const()[name = tensor<string, []>("op_45149"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_45151 = const()[name = tensor<string, []>("op_45151"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_45153_pad_type_0 = const()[name = tensor<string, []>("op_45153_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_45153_pad_0 = const()[name = tensor<string, []>("op_45153_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3766178432)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_1_transformer_blocks_5_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3792392896)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_45153_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_5_ff_net_0_proj_bias_to_fp16, dilations = var_45151, groups = var_30385, pad = var_45153_pad_0, pad_type = var_45153_pad_type_0, strides = var_45149, weight = up_blocks_0_attentions_1_transformer_blocks_5_ff_net_0_proj_weight_to_fp16, x = input_583_cast)[name = tensor<string, []>("op_45153_cast")];
            tensor<int32, [2]> var_45154_split_sizes_0 = const()[name = tensor<string, []>("op_45154_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_45154_axis_0 = const()[name = tensor<string, []>("op_45154_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_45154_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_45154_cast_1 = split(axis = var_45154_axis_0, split_sizes = var_45154_split_sizes_0, x = var_45153_cast)[name = tensor<string, []>("op_45154_cast")];
            tensor<string, []> var_45156_mode_0 = const()[name = tensor<string, []>("op_45156_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_45156_cast = gelu(mode = var_45156_mode_0, x = var_45154_cast_1)[name = tensor<string, []>("op_45156_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_585_cast = mul(x = var_45154_cast_0, y = var_45156_cast)[name = tensor<string, []>("input_585_cast")];
            tensor<int32, [2]> var_45160 = const()[name = tensor<string, []>("op_45160"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_45162 = const()[name = tensor<string, []>("op_45162"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_45164_pad_type_0 = const()[name = tensor<string, []>("op_45164_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_45164_pad_0 = const()[name = tensor<string, []>("op_45164_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_5_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3792413440)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_5_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_5_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3805520704)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_45164_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_5_ff_net_2_bias_to_fp16, dilations = var_45162, groups = var_30385, pad = var_45164_pad_0, pad_type = var_45164_pad_type_0, strides = var_45160, weight = up_blocks_0_attentions_1_transformer_blocks_5_ff_net_2_weight_to_fp16, x = input_585_cast)[name = tensor<string, []>("op_45164_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_301_cast = add(x = var_45164_cast, y = inputs_299_cast)[name = tensor<string, []>("inputs_301_cast")];
            tensor<int32, [1]> var_45174 = const()[name = tensor<string, []>("op_45174"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_301_cast = reduce_mean(axes = var_45174, keep_dims = var_30380, x = inputs_301_cast)[name = tensor<string, []>("channels_mean_301_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_301_cast = sub(x = inputs_301_cast, y = channels_mean_301_cast)[name = tensor<string, []>("zero_mean_301_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_301_cast = mul(x = zero_mean_301_cast, y = zero_mean_301_cast)[name = tensor<string, []>("zero_mean_sq_301_cast")];
            tensor<int32, [1]> var_45178 = const()[name = tensor<string, []>("op_45178"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_45179_cast = reduce_mean(axes = var_45178, keep_dims = var_30380, x = zero_mean_sq_301_cast)[name = tensor<string, []>("op_45179_cast")];
            tensor<fp16, []> var_45180_to_fp16 = const()[name = tensor<string, []>("op_45180_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_45181_cast = add(x = var_45179_cast, y = var_45180_to_fp16)[name = tensor<string, []>("op_45181_cast")];
            tensor<fp16, []> denom_301_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_301_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_301_cast = rsqrt(epsilon = denom_301_epsilon_0_to_fp16, x = var_45181_cast)[name = tensor<string, []>("denom_301_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_301_cast = mul(x = zero_mean_301_cast, y = denom_301_cast)[name = tensor<string, []>("out_301_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_45185_to_fp16 = const()[name = tensor<string, []>("op_45185_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3805523328)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_45186_cast = add(x = out_301_cast, y = var_45185_to_fp16)[name = tensor<string, []>("op_45186_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_45188_to_fp16 = const()[name = tensor<string, []>("op_45188_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3805525952)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_397_cast = mul(x = var_45186_cast, y = var_45188_to_fp16)[name = tensor<string, []>("hidden_states_397_cast")];
            tensor<int32, [2]> var_45195 = const()[name = tensor<string, []>("op_45195"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_45197 = const()[name = tensor<string, []>("op_45197"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_201_pad_type_0 = const()[name = tensor<string, []>("q_201_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_201_pad_0 = const()[name = tensor<string, []>("q_201_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3805528576)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_201_cast = conv(dilations = var_45197, groups = var_30385, pad = q_201_pad_0, pad_type = q_201_pad_type_0, strides = var_45195, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_q_weight_to_fp16, x = hidden_states_397_cast)[name = tensor<string, []>("q_201_cast")];
            tensor<int32, [2]> var_45201 = const()[name = tensor<string, []>("op_45201"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_45203 = const()[name = tensor<string, []>("op_45203"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_401_pad_type_0 = const()[name = tensor<string, []>("k_401_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_401_pad_0 = const()[name = tensor<string, []>("k_401_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3808805440)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_401_cast = conv(dilations = var_45203, groups = var_30385, pad = k_401_pad_0, pad_type = k_401_pad_type_0, strides = var_45201, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_k_weight_to_fp16, x = hidden_states_397_cast)[name = tensor<string, []>("k_401_cast")];
            tensor<int32, [2]> var_45207 = const()[name = tensor<string, []>("op_45207"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_45209 = const()[name = tensor<string, []>("op_45209"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_201_pad_type_0 = const()[name = tensor<string, []>("v_201_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_201_pad_0 = const()[name = tensor<string, []>("v_201_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3812082304)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_201_cast = conv(dilations = var_45209, groups = var_30385, pad = v_201_pad_0, pad_type = v_201_pad_type_0, strides = var_45207, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_v_weight_to_fp16, x = hidden_states_397_cast)[name = tensor<string, []>("v_201_cast")];
            tensor<int32, [4]> var_45213_begin_0 = const()[name = tensor<string, []>("op_45213_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_45213_end_0 = const()[name = tensor<string, []>("op_45213_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_45213_end_mask_0 = const()[name = tensor<string, []>("op_45213_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45213_cast = slice_by_index(begin = var_45213_begin_0, end = var_45213_end_0, end_mask = var_45213_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45213_cast")];
            tensor<int32, [4]> var_45217_begin_0 = const()[name = tensor<string, []>("op_45217_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_45217_end_0 = const()[name = tensor<string, []>("op_45217_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_45217_end_mask_0 = const()[name = tensor<string, []>("op_45217_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45217_cast = slice_by_index(begin = var_45217_begin_0, end = var_45217_end_0, end_mask = var_45217_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45217_cast")];
            tensor<int32, [4]> var_45221_begin_0 = const()[name = tensor<string, []>("op_45221_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_45221_end_0 = const()[name = tensor<string, []>("op_45221_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_45221_end_mask_0 = const()[name = tensor<string, []>("op_45221_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45221_cast = slice_by_index(begin = var_45221_begin_0, end = var_45221_end_0, end_mask = var_45221_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45221_cast")];
            tensor<int32, [4]> var_45225_begin_0 = const()[name = tensor<string, []>("op_45225_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_45225_end_0 = const()[name = tensor<string, []>("op_45225_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_45225_end_mask_0 = const()[name = tensor<string, []>("op_45225_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45225_cast = slice_by_index(begin = var_45225_begin_0, end = var_45225_end_0, end_mask = var_45225_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45225_cast")];
            tensor<int32, [4]> var_45229_begin_0 = const()[name = tensor<string, []>("op_45229_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_45229_end_0 = const()[name = tensor<string, []>("op_45229_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_45229_end_mask_0 = const()[name = tensor<string, []>("op_45229_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45229_cast = slice_by_index(begin = var_45229_begin_0, end = var_45229_end_0, end_mask = var_45229_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45229_cast")];
            tensor<int32, [4]> var_45233_begin_0 = const()[name = tensor<string, []>("op_45233_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_45233_end_0 = const()[name = tensor<string, []>("op_45233_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_45233_end_mask_0 = const()[name = tensor<string, []>("op_45233_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45233_cast = slice_by_index(begin = var_45233_begin_0, end = var_45233_end_0, end_mask = var_45233_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45233_cast")];
            tensor<int32, [4]> var_45237_begin_0 = const()[name = tensor<string, []>("op_45237_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_45237_end_0 = const()[name = tensor<string, []>("op_45237_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_45237_end_mask_0 = const()[name = tensor<string, []>("op_45237_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45237_cast = slice_by_index(begin = var_45237_begin_0, end = var_45237_end_0, end_mask = var_45237_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45237_cast")];
            tensor<int32, [4]> var_45241_begin_0 = const()[name = tensor<string, []>("op_45241_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_45241_end_0 = const()[name = tensor<string, []>("op_45241_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_45241_end_mask_0 = const()[name = tensor<string, []>("op_45241_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45241_cast = slice_by_index(begin = var_45241_begin_0, end = var_45241_end_0, end_mask = var_45241_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45241_cast")];
            tensor<int32, [4]> var_45245_begin_0 = const()[name = tensor<string, []>("op_45245_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_45245_end_0 = const()[name = tensor<string, []>("op_45245_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_45245_end_mask_0 = const()[name = tensor<string, []>("op_45245_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45245_cast = slice_by_index(begin = var_45245_begin_0, end = var_45245_end_0, end_mask = var_45245_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45245_cast")];
            tensor<int32, [4]> var_45249_begin_0 = const()[name = tensor<string, []>("op_45249_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_45249_end_0 = const()[name = tensor<string, []>("op_45249_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_45249_end_mask_0 = const()[name = tensor<string, []>("op_45249_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45249_cast = slice_by_index(begin = var_45249_begin_0, end = var_45249_end_0, end_mask = var_45249_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45249_cast")];
            tensor<int32, [4]> var_45253_begin_0 = const()[name = tensor<string, []>("op_45253_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_45253_end_0 = const()[name = tensor<string, []>("op_45253_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_45253_end_mask_0 = const()[name = tensor<string, []>("op_45253_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45253_cast = slice_by_index(begin = var_45253_begin_0, end = var_45253_end_0, end_mask = var_45253_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45253_cast")];
            tensor<int32, [4]> var_45257_begin_0 = const()[name = tensor<string, []>("op_45257_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_45257_end_0 = const()[name = tensor<string, []>("op_45257_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_45257_end_mask_0 = const()[name = tensor<string, []>("op_45257_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45257_cast = slice_by_index(begin = var_45257_begin_0, end = var_45257_end_0, end_mask = var_45257_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45257_cast")];
            tensor<int32, [4]> var_45261_begin_0 = const()[name = tensor<string, []>("op_45261_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_45261_end_0 = const()[name = tensor<string, []>("op_45261_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_45261_end_mask_0 = const()[name = tensor<string, []>("op_45261_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45261_cast = slice_by_index(begin = var_45261_begin_0, end = var_45261_end_0, end_mask = var_45261_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45261_cast")];
            tensor<int32, [4]> var_45265_begin_0 = const()[name = tensor<string, []>("op_45265_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_45265_end_0 = const()[name = tensor<string, []>("op_45265_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_45265_end_mask_0 = const()[name = tensor<string, []>("op_45265_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45265_cast = slice_by_index(begin = var_45265_begin_0, end = var_45265_end_0, end_mask = var_45265_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45265_cast")];
            tensor<int32, [4]> var_45269_begin_0 = const()[name = tensor<string, []>("op_45269_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_45269_end_0 = const()[name = tensor<string, []>("op_45269_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_45269_end_mask_0 = const()[name = tensor<string, []>("op_45269_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45269_cast = slice_by_index(begin = var_45269_begin_0, end = var_45269_end_0, end_mask = var_45269_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45269_cast")];
            tensor<int32, [4]> var_45273_begin_0 = const()[name = tensor<string, []>("op_45273_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_45273_end_0 = const()[name = tensor<string, []>("op_45273_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_45273_end_mask_0 = const()[name = tensor<string, []>("op_45273_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45273_cast = slice_by_index(begin = var_45273_begin_0, end = var_45273_end_0, end_mask = var_45273_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45273_cast")];
            tensor<int32, [4]> var_45277_begin_0 = const()[name = tensor<string, []>("op_45277_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_45277_end_0 = const()[name = tensor<string, []>("op_45277_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_45277_end_mask_0 = const()[name = tensor<string, []>("op_45277_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45277_cast = slice_by_index(begin = var_45277_begin_0, end = var_45277_end_0, end_mask = var_45277_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45277_cast")];
            tensor<int32, [4]> var_45281_begin_0 = const()[name = tensor<string, []>("op_45281_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_45281_end_0 = const()[name = tensor<string, []>("op_45281_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_45281_end_mask_0 = const()[name = tensor<string, []>("op_45281_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45281_cast = slice_by_index(begin = var_45281_begin_0, end = var_45281_end_0, end_mask = var_45281_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45281_cast")];
            tensor<int32, [4]> var_45285_begin_0 = const()[name = tensor<string, []>("op_45285_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_45285_end_0 = const()[name = tensor<string, []>("op_45285_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_45285_end_mask_0 = const()[name = tensor<string, []>("op_45285_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45285_cast = slice_by_index(begin = var_45285_begin_0, end = var_45285_end_0, end_mask = var_45285_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45285_cast")];
            tensor<int32, [4]> var_45289_begin_0 = const()[name = tensor<string, []>("op_45289_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_45289_end_0 = const()[name = tensor<string, []>("op_45289_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_45289_end_mask_0 = const()[name = tensor<string, []>("op_45289_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45289_cast = slice_by_index(begin = var_45289_begin_0, end = var_45289_end_0, end_mask = var_45289_end_mask_0, x = q_201_cast)[name = tensor<string, []>("op_45289_cast")];
            tensor<int32, [4]> k_403_perm_0 = const()[name = tensor<string, []>("k_403_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_45296_begin_0 = const()[name = tensor<string, []>("op_45296_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_45296_end_0 = const()[name = tensor<string, []>("op_45296_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_45296_end_mask_0 = const()[name = tensor<string, []>("op_45296_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_39 = transpose(perm = k_403_perm_0, x = k_401_cast)[name = tensor<string, []>("transpose_39")];
            tensor<fp16, [2, 1024, 1, 64]> var_45296_cast = slice_by_index(begin = var_45296_begin_0, end = var_45296_end_0, end_mask = var_45296_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45296_cast")];
            tensor<int32, [4]> var_45300_begin_0 = const()[name = tensor<string, []>("op_45300_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_45300_end_0 = const()[name = tensor<string, []>("op_45300_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_45300_end_mask_0 = const()[name = tensor<string, []>("op_45300_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45300_cast = slice_by_index(begin = var_45300_begin_0, end = var_45300_end_0, end_mask = var_45300_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45300_cast")];
            tensor<int32, [4]> var_45304_begin_0 = const()[name = tensor<string, []>("op_45304_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_45304_end_0 = const()[name = tensor<string, []>("op_45304_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_45304_end_mask_0 = const()[name = tensor<string, []>("op_45304_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45304_cast = slice_by_index(begin = var_45304_begin_0, end = var_45304_end_0, end_mask = var_45304_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45304_cast")];
            tensor<int32, [4]> var_45308_begin_0 = const()[name = tensor<string, []>("op_45308_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_45308_end_0 = const()[name = tensor<string, []>("op_45308_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_45308_end_mask_0 = const()[name = tensor<string, []>("op_45308_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45308_cast = slice_by_index(begin = var_45308_begin_0, end = var_45308_end_0, end_mask = var_45308_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45308_cast")];
            tensor<int32, [4]> var_45312_begin_0 = const()[name = tensor<string, []>("op_45312_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_45312_end_0 = const()[name = tensor<string, []>("op_45312_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_45312_end_mask_0 = const()[name = tensor<string, []>("op_45312_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45312_cast = slice_by_index(begin = var_45312_begin_0, end = var_45312_end_0, end_mask = var_45312_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45312_cast")];
            tensor<int32, [4]> var_45316_begin_0 = const()[name = tensor<string, []>("op_45316_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_45316_end_0 = const()[name = tensor<string, []>("op_45316_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_45316_end_mask_0 = const()[name = tensor<string, []>("op_45316_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45316_cast = slice_by_index(begin = var_45316_begin_0, end = var_45316_end_0, end_mask = var_45316_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45316_cast")];
            tensor<int32, [4]> var_45320_begin_0 = const()[name = tensor<string, []>("op_45320_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_45320_end_0 = const()[name = tensor<string, []>("op_45320_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_45320_end_mask_0 = const()[name = tensor<string, []>("op_45320_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45320_cast = slice_by_index(begin = var_45320_begin_0, end = var_45320_end_0, end_mask = var_45320_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45320_cast")];
            tensor<int32, [4]> var_45324_begin_0 = const()[name = tensor<string, []>("op_45324_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_45324_end_0 = const()[name = tensor<string, []>("op_45324_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_45324_end_mask_0 = const()[name = tensor<string, []>("op_45324_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45324_cast = slice_by_index(begin = var_45324_begin_0, end = var_45324_end_0, end_mask = var_45324_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45324_cast")];
            tensor<int32, [4]> var_45328_begin_0 = const()[name = tensor<string, []>("op_45328_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_45328_end_0 = const()[name = tensor<string, []>("op_45328_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_45328_end_mask_0 = const()[name = tensor<string, []>("op_45328_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45328_cast = slice_by_index(begin = var_45328_begin_0, end = var_45328_end_0, end_mask = var_45328_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45328_cast")];
            tensor<int32, [4]> var_45332_begin_0 = const()[name = tensor<string, []>("op_45332_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_45332_end_0 = const()[name = tensor<string, []>("op_45332_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_45332_end_mask_0 = const()[name = tensor<string, []>("op_45332_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45332_cast = slice_by_index(begin = var_45332_begin_0, end = var_45332_end_0, end_mask = var_45332_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45332_cast")];
            tensor<int32, [4]> var_45336_begin_0 = const()[name = tensor<string, []>("op_45336_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_45336_end_0 = const()[name = tensor<string, []>("op_45336_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_45336_end_mask_0 = const()[name = tensor<string, []>("op_45336_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45336_cast = slice_by_index(begin = var_45336_begin_0, end = var_45336_end_0, end_mask = var_45336_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45336_cast")];
            tensor<int32, [4]> var_45340_begin_0 = const()[name = tensor<string, []>("op_45340_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_45340_end_0 = const()[name = tensor<string, []>("op_45340_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_45340_end_mask_0 = const()[name = tensor<string, []>("op_45340_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45340_cast = slice_by_index(begin = var_45340_begin_0, end = var_45340_end_0, end_mask = var_45340_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45340_cast")];
            tensor<int32, [4]> var_45344_begin_0 = const()[name = tensor<string, []>("op_45344_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_45344_end_0 = const()[name = tensor<string, []>("op_45344_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_45344_end_mask_0 = const()[name = tensor<string, []>("op_45344_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45344_cast = slice_by_index(begin = var_45344_begin_0, end = var_45344_end_0, end_mask = var_45344_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45344_cast")];
            tensor<int32, [4]> var_45348_begin_0 = const()[name = tensor<string, []>("op_45348_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_45348_end_0 = const()[name = tensor<string, []>("op_45348_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_45348_end_mask_0 = const()[name = tensor<string, []>("op_45348_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45348_cast = slice_by_index(begin = var_45348_begin_0, end = var_45348_end_0, end_mask = var_45348_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45348_cast")];
            tensor<int32, [4]> var_45352_begin_0 = const()[name = tensor<string, []>("op_45352_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_45352_end_0 = const()[name = tensor<string, []>("op_45352_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_45352_end_mask_0 = const()[name = tensor<string, []>("op_45352_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45352_cast = slice_by_index(begin = var_45352_begin_0, end = var_45352_end_0, end_mask = var_45352_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45352_cast")];
            tensor<int32, [4]> var_45356_begin_0 = const()[name = tensor<string, []>("op_45356_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_45356_end_0 = const()[name = tensor<string, []>("op_45356_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_45356_end_mask_0 = const()[name = tensor<string, []>("op_45356_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45356_cast = slice_by_index(begin = var_45356_begin_0, end = var_45356_end_0, end_mask = var_45356_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45356_cast")];
            tensor<int32, [4]> var_45360_begin_0 = const()[name = tensor<string, []>("op_45360_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_45360_end_0 = const()[name = tensor<string, []>("op_45360_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_45360_end_mask_0 = const()[name = tensor<string, []>("op_45360_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45360_cast = slice_by_index(begin = var_45360_begin_0, end = var_45360_end_0, end_mask = var_45360_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45360_cast")];
            tensor<int32, [4]> var_45364_begin_0 = const()[name = tensor<string, []>("op_45364_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_45364_end_0 = const()[name = tensor<string, []>("op_45364_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_45364_end_mask_0 = const()[name = tensor<string, []>("op_45364_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45364_cast = slice_by_index(begin = var_45364_begin_0, end = var_45364_end_0, end_mask = var_45364_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45364_cast")];
            tensor<int32, [4]> var_45368_begin_0 = const()[name = tensor<string, []>("op_45368_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_45368_end_0 = const()[name = tensor<string, []>("op_45368_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_45368_end_mask_0 = const()[name = tensor<string, []>("op_45368_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45368_cast = slice_by_index(begin = var_45368_begin_0, end = var_45368_end_0, end_mask = var_45368_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45368_cast")];
            tensor<int32, [4]> var_45372_begin_0 = const()[name = tensor<string, []>("op_45372_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_45372_end_0 = const()[name = tensor<string, []>("op_45372_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_45372_end_mask_0 = const()[name = tensor<string, []>("op_45372_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_45372_cast = slice_by_index(begin = var_45372_begin_0, end = var_45372_end_0, end_mask = var_45372_end_mask_0, x = transpose_39)[name = tensor<string, []>("op_45372_cast")];
            tensor<int32, [4]> var_45374_begin_0 = const()[name = tensor<string, []>("op_45374_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_45374_end_0 = const()[name = tensor<string, []>("op_45374_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_45374_end_mask_0 = const()[name = tensor<string, []>("op_45374_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45374_cast = slice_by_index(begin = var_45374_begin_0, end = var_45374_end_0, end_mask = var_45374_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45374_cast")];
            tensor<int32, [4]> var_45378_begin_0 = const()[name = tensor<string, []>("op_45378_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_45378_end_0 = const()[name = tensor<string, []>("op_45378_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_45378_end_mask_0 = const()[name = tensor<string, []>("op_45378_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45378_cast = slice_by_index(begin = var_45378_begin_0, end = var_45378_end_0, end_mask = var_45378_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45378_cast")];
            tensor<int32, [4]> var_45382_begin_0 = const()[name = tensor<string, []>("op_45382_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_45382_end_0 = const()[name = tensor<string, []>("op_45382_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_45382_end_mask_0 = const()[name = tensor<string, []>("op_45382_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45382_cast = slice_by_index(begin = var_45382_begin_0, end = var_45382_end_0, end_mask = var_45382_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45382_cast")];
            tensor<int32, [4]> var_45386_begin_0 = const()[name = tensor<string, []>("op_45386_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_45386_end_0 = const()[name = tensor<string, []>("op_45386_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_45386_end_mask_0 = const()[name = tensor<string, []>("op_45386_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45386_cast = slice_by_index(begin = var_45386_begin_0, end = var_45386_end_0, end_mask = var_45386_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45386_cast")];
            tensor<int32, [4]> var_45390_begin_0 = const()[name = tensor<string, []>("op_45390_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_45390_end_0 = const()[name = tensor<string, []>("op_45390_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_45390_end_mask_0 = const()[name = tensor<string, []>("op_45390_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45390_cast = slice_by_index(begin = var_45390_begin_0, end = var_45390_end_0, end_mask = var_45390_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45390_cast")];
            tensor<int32, [4]> var_45394_begin_0 = const()[name = tensor<string, []>("op_45394_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_45394_end_0 = const()[name = tensor<string, []>("op_45394_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_45394_end_mask_0 = const()[name = tensor<string, []>("op_45394_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45394_cast = slice_by_index(begin = var_45394_begin_0, end = var_45394_end_0, end_mask = var_45394_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45394_cast")];
            tensor<int32, [4]> var_45398_begin_0 = const()[name = tensor<string, []>("op_45398_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_45398_end_0 = const()[name = tensor<string, []>("op_45398_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_45398_end_mask_0 = const()[name = tensor<string, []>("op_45398_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45398_cast = slice_by_index(begin = var_45398_begin_0, end = var_45398_end_0, end_mask = var_45398_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45398_cast")];
            tensor<int32, [4]> var_45402_begin_0 = const()[name = tensor<string, []>("op_45402_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_45402_end_0 = const()[name = tensor<string, []>("op_45402_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_45402_end_mask_0 = const()[name = tensor<string, []>("op_45402_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45402_cast = slice_by_index(begin = var_45402_begin_0, end = var_45402_end_0, end_mask = var_45402_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45402_cast")];
            tensor<int32, [4]> var_45406_begin_0 = const()[name = tensor<string, []>("op_45406_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_45406_end_0 = const()[name = tensor<string, []>("op_45406_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_45406_end_mask_0 = const()[name = tensor<string, []>("op_45406_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45406_cast = slice_by_index(begin = var_45406_begin_0, end = var_45406_end_0, end_mask = var_45406_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45406_cast")];
            tensor<int32, [4]> var_45410_begin_0 = const()[name = tensor<string, []>("op_45410_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_45410_end_0 = const()[name = tensor<string, []>("op_45410_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_45410_end_mask_0 = const()[name = tensor<string, []>("op_45410_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45410_cast = slice_by_index(begin = var_45410_begin_0, end = var_45410_end_0, end_mask = var_45410_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45410_cast")];
            tensor<int32, [4]> var_45414_begin_0 = const()[name = tensor<string, []>("op_45414_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_45414_end_0 = const()[name = tensor<string, []>("op_45414_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_45414_end_mask_0 = const()[name = tensor<string, []>("op_45414_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45414_cast = slice_by_index(begin = var_45414_begin_0, end = var_45414_end_0, end_mask = var_45414_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45414_cast")];
            tensor<int32, [4]> var_45418_begin_0 = const()[name = tensor<string, []>("op_45418_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_45418_end_0 = const()[name = tensor<string, []>("op_45418_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_45418_end_mask_0 = const()[name = tensor<string, []>("op_45418_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45418_cast = slice_by_index(begin = var_45418_begin_0, end = var_45418_end_0, end_mask = var_45418_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45418_cast")];
            tensor<int32, [4]> var_45422_begin_0 = const()[name = tensor<string, []>("op_45422_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_45422_end_0 = const()[name = tensor<string, []>("op_45422_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_45422_end_mask_0 = const()[name = tensor<string, []>("op_45422_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45422_cast = slice_by_index(begin = var_45422_begin_0, end = var_45422_end_0, end_mask = var_45422_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45422_cast")];
            tensor<int32, [4]> var_45426_begin_0 = const()[name = tensor<string, []>("op_45426_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_45426_end_0 = const()[name = tensor<string, []>("op_45426_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_45426_end_mask_0 = const()[name = tensor<string, []>("op_45426_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45426_cast = slice_by_index(begin = var_45426_begin_0, end = var_45426_end_0, end_mask = var_45426_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45426_cast")];
            tensor<int32, [4]> var_45430_begin_0 = const()[name = tensor<string, []>("op_45430_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_45430_end_0 = const()[name = tensor<string, []>("op_45430_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_45430_end_mask_0 = const()[name = tensor<string, []>("op_45430_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45430_cast = slice_by_index(begin = var_45430_begin_0, end = var_45430_end_0, end_mask = var_45430_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45430_cast")];
            tensor<int32, [4]> var_45434_begin_0 = const()[name = tensor<string, []>("op_45434_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_45434_end_0 = const()[name = tensor<string, []>("op_45434_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_45434_end_mask_0 = const()[name = tensor<string, []>("op_45434_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45434_cast = slice_by_index(begin = var_45434_begin_0, end = var_45434_end_0, end_mask = var_45434_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45434_cast")];
            tensor<int32, [4]> var_45438_begin_0 = const()[name = tensor<string, []>("op_45438_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_45438_end_0 = const()[name = tensor<string, []>("op_45438_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_45438_end_mask_0 = const()[name = tensor<string, []>("op_45438_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45438_cast = slice_by_index(begin = var_45438_begin_0, end = var_45438_end_0, end_mask = var_45438_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45438_cast")];
            tensor<int32, [4]> var_45442_begin_0 = const()[name = tensor<string, []>("op_45442_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_45442_end_0 = const()[name = tensor<string, []>("op_45442_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_45442_end_mask_0 = const()[name = tensor<string, []>("op_45442_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45442_cast = slice_by_index(begin = var_45442_begin_0, end = var_45442_end_0, end_mask = var_45442_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45442_cast")];
            tensor<int32, [4]> var_45446_begin_0 = const()[name = tensor<string, []>("op_45446_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_45446_end_0 = const()[name = tensor<string, []>("op_45446_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_45446_end_mask_0 = const()[name = tensor<string, []>("op_45446_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45446_cast = slice_by_index(begin = var_45446_begin_0, end = var_45446_end_0, end_mask = var_45446_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45446_cast")];
            tensor<int32, [4]> var_45450_begin_0 = const()[name = tensor<string, []>("op_45450_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_45450_end_0 = const()[name = tensor<string, []>("op_45450_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_45450_end_mask_0 = const()[name = tensor<string, []>("op_45450_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45450_cast = slice_by_index(begin = var_45450_begin_0, end = var_45450_end_0, end_mask = var_45450_end_mask_0, x = v_201_cast)[name = tensor<string, []>("op_45450_cast")];
            tensor<string, []> var_45454_equation_0 = const()[name = tensor<string, []>("op_45454_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45454_cast = einsum(equation = var_45454_equation_0, values = (var_45296_cast, var_45213_cast))[name = tensor<string, []>("op_45454_cast")];
            tensor<fp16, []> var_45455_to_fp16 = const()[name = tensor<string, []>("op_45455_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3841_cast = mul(x = var_45454_cast, y = var_45455_to_fp16)[name = tensor<string, []>("aw_3841_cast")];
            tensor<string, []> var_45458_equation_0 = const()[name = tensor<string, []>("op_45458_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45458_cast = einsum(equation = var_45458_equation_0, values = (var_45300_cast, var_45217_cast))[name = tensor<string, []>("op_45458_cast")];
            tensor<fp16, []> var_45459_to_fp16 = const()[name = tensor<string, []>("op_45459_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3843_cast = mul(x = var_45458_cast, y = var_45459_to_fp16)[name = tensor<string, []>("aw_3843_cast")];
            tensor<string, []> var_45462_equation_0 = const()[name = tensor<string, []>("op_45462_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45462_cast = einsum(equation = var_45462_equation_0, values = (var_45304_cast, var_45221_cast))[name = tensor<string, []>("op_45462_cast")];
            tensor<fp16, []> var_45463_to_fp16 = const()[name = tensor<string, []>("op_45463_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3845_cast = mul(x = var_45462_cast, y = var_45463_to_fp16)[name = tensor<string, []>("aw_3845_cast")];
            tensor<string, []> var_45466_equation_0 = const()[name = tensor<string, []>("op_45466_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45466_cast = einsum(equation = var_45466_equation_0, values = (var_45308_cast, var_45225_cast))[name = tensor<string, []>("op_45466_cast")];
            tensor<fp16, []> var_45467_to_fp16 = const()[name = tensor<string, []>("op_45467_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3847_cast = mul(x = var_45466_cast, y = var_45467_to_fp16)[name = tensor<string, []>("aw_3847_cast")];
            tensor<string, []> var_45470_equation_0 = const()[name = tensor<string, []>("op_45470_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45470_cast = einsum(equation = var_45470_equation_0, values = (var_45312_cast, var_45229_cast))[name = tensor<string, []>("op_45470_cast")];
            tensor<fp16, []> var_45471_to_fp16 = const()[name = tensor<string, []>("op_45471_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3849_cast = mul(x = var_45470_cast, y = var_45471_to_fp16)[name = tensor<string, []>("aw_3849_cast")];
            tensor<string, []> var_45474_equation_0 = const()[name = tensor<string, []>("op_45474_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45474_cast = einsum(equation = var_45474_equation_0, values = (var_45316_cast, var_45233_cast))[name = tensor<string, []>("op_45474_cast")];
            tensor<fp16, []> var_45475_to_fp16 = const()[name = tensor<string, []>("op_45475_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3851_cast = mul(x = var_45474_cast, y = var_45475_to_fp16)[name = tensor<string, []>("aw_3851_cast")];
            tensor<string, []> var_45478_equation_0 = const()[name = tensor<string, []>("op_45478_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45478_cast = einsum(equation = var_45478_equation_0, values = (var_45320_cast, var_45237_cast))[name = tensor<string, []>("op_45478_cast")];
            tensor<fp16, []> var_45479_to_fp16 = const()[name = tensor<string, []>("op_45479_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3853_cast = mul(x = var_45478_cast, y = var_45479_to_fp16)[name = tensor<string, []>("aw_3853_cast")];
            tensor<string, []> var_45482_equation_0 = const()[name = tensor<string, []>("op_45482_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45482_cast = einsum(equation = var_45482_equation_0, values = (var_45324_cast, var_45241_cast))[name = tensor<string, []>("op_45482_cast")];
            tensor<fp16, []> var_45483_to_fp16 = const()[name = tensor<string, []>("op_45483_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3855_cast = mul(x = var_45482_cast, y = var_45483_to_fp16)[name = tensor<string, []>("aw_3855_cast")];
            tensor<string, []> var_45486_equation_0 = const()[name = tensor<string, []>("op_45486_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45486_cast = einsum(equation = var_45486_equation_0, values = (var_45328_cast, var_45245_cast))[name = tensor<string, []>("op_45486_cast")];
            tensor<fp16, []> var_45487_to_fp16 = const()[name = tensor<string, []>("op_45487_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3857_cast = mul(x = var_45486_cast, y = var_45487_to_fp16)[name = tensor<string, []>("aw_3857_cast")];
            tensor<string, []> var_45490_equation_0 = const()[name = tensor<string, []>("op_45490_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45490_cast = einsum(equation = var_45490_equation_0, values = (var_45332_cast, var_45249_cast))[name = tensor<string, []>("op_45490_cast")];
            tensor<fp16, []> var_45491_to_fp16 = const()[name = tensor<string, []>("op_45491_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3859_cast = mul(x = var_45490_cast, y = var_45491_to_fp16)[name = tensor<string, []>("aw_3859_cast")];
            tensor<string, []> var_45494_equation_0 = const()[name = tensor<string, []>("op_45494_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45494_cast = einsum(equation = var_45494_equation_0, values = (var_45336_cast, var_45253_cast))[name = tensor<string, []>("op_45494_cast")];
            tensor<fp16, []> var_45495_to_fp16 = const()[name = tensor<string, []>("op_45495_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3861_cast = mul(x = var_45494_cast, y = var_45495_to_fp16)[name = tensor<string, []>("aw_3861_cast")];
            tensor<string, []> var_45498_equation_0 = const()[name = tensor<string, []>("op_45498_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45498_cast = einsum(equation = var_45498_equation_0, values = (var_45340_cast, var_45257_cast))[name = tensor<string, []>("op_45498_cast")];
            tensor<fp16, []> var_45499_to_fp16 = const()[name = tensor<string, []>("op_45499_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3863_cast = mul(x = var_45498_cast, y = var_45499_to_fp16)[name = tensor<string, []>("aw_3863_cast")];
            tensor<string, []> var_45502_equation_0 = const()[name = tensor<string, []>("op_45502_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45502_cast = einsum(equation = var_45502_equation_0, values = (var_45344_cast, var_45261_cast))[name = tensor<string, []>("op_45502_cast")];
            tensor<fp16, []> var_45503_to_fp16 = const()[name = tensor<string, []>("op_45503_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3865_cast = mul(x = var_45502_cast, y = var_45503_to_fp16)[name = tensor<string, []>("aw_3865_cast")];
            tensor<string, []> var_45506_equation_0 = const()[name = tensor<string, []>("op_45506_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45506_cast = einsum(equation = var_45506_equation_0, values = (var_45348_cast, var_45265_cast))[name = tensor<string, []>("op_45506_cast")];
            tensor<fp16, []> var_45507_to_fp16 = const()[name = tensor<string, []>("op_45507_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3867_cast = mul(x = var_45506_cast, y = var_45507_to_fp16)[name = tensor<string, []>("aw_3867_cast")];
            tensor<string, []> var_45510_equation_0 = const()[name = tensor<string, []>("op_45510_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45510_cast = einsum(equation = var_45510_equation_0, values = (var_45352_cast, var_45269_cast))[name = tensor<string, []>("op_45510_cast")];
            tensor<fp16, []> var_45511_to_fp16 = const()[name = tensor<string, []>("op_45511_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3869_cast = mul(x = var_45510_cast, y = var_45511_to_fp16)[name = tensor<string, []>("aw_3869_cast")];
            tensor<string, []> var_45514_equation_0 = const()[name = tensor<string, []>("op_45514_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45514_cast = einsum(equation = var_45514_equation_0, values = (var_45356_cast, var_45273_cast))[name = tensor<string, []>("op_45514_cast")];
            tensor<fp16, []> var_45515_to_fp16 = const()[name = tensor<string, []>("op_45515_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3871_cast = mul(x = var_45514_cast, y = var_45515_to_fp16)[name = tensor<string, []>("aw_3871_cast")];
            tensor<string, []> var_45518_equation_0 = const()[name = tensor<string, []>("op_45518_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45518_cast = einsum(equation = var_45518_equation_0, values = (var_45360_cast, var_45277_cast))[name = tensor<string, []>("op_45518_cast")];
            tensor<fp16, []> var_45519_to_fp16 = const()[name = tensor<string, []>("op_45519_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3873_cast = mul(x = var_45518_cast, y = var_45519_to_fp16)[name = tensor<string, []>("aw_3873_cast")];
            tensor<string, []> var_45522_equation_0 = const()[name = tensor<string, []>("op_45522_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45522_cast = einsum(equation = var_45522_equation_0, values = (var_45364_cast, var_45281_cast))[name = tensor<string, []>("op_45522_cast")];
            tensor<fp16, []> var_45523_to_fp16 = const()[name = tensor<string, []>("op_45523_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3875_cast = mul(x = var_45522_cast, y = var_45523_to_fp16)[name = tensor<string, []>("aw_3875_cast")];
            tensor<string, []> var_45526_equation_0 = const()[name = tensor<string, []>("op_45526_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45526_cast = einsum(equation = var_45526_equation_0, values = (var_45368_cast, var_45285_cast))[name = tensor<string, []>("op_45526_cast")];
            tensor<fp16, []> var_45527_to_fp16 = const()[name = tensor<string, []>("op_45527_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3877_cast = mul(x = var_45526_cast, y = var_45527_to_fp16)[name = tensor<string, []>("aw_3877_cast")];
            tensor<string, []> var_45530_equation_0 = const()[name = tensor<string, []>("op_45530_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45530_cast = einsum(equation = var_45530_equation_0, values = (var_45372_cast, var_45289_cast))[name = tensor<string, []>("op_45530_cast")];
            tensor<fp16, []> var_45531_to_fp16 = const()[name = tensor<string, []>("op_45531_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3879_cast = mul(x = var_45530_cast, y = var_45531_to_fp16)[name = tensor<string, []>("aw_3879_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45533_cast = softmax(axis = var_30385, x = aw_3841_cast)[name = tensor<string, []>("op_45533_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45534_cast = softmax(axis = var_30385, x = aw_3843_cast)[name = tensor<string, []>("op_45534_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45535_cast = softmax(axis = var_30385, x = aw_3845_cast)[name = tensor<string, []>("op_45535_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45536_cast = softmax(axis = var_30385, x = aw_3847_cast)[name = tensor<string, []>("op_45536_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45537_cast = softmax(axis = var_30385, x = aw_3849_cast)[name = tensor<string, []>("op_45537_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45538_cast = softmax(axis = var_30385, x = aw_3851_cast)[name = tensor<string, []>("op_45538_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45539_cast = softmax(axis = var_30385, x = aw_3853_cast)[name = tensor<string, []>("op_45539_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45540_cast = softmax(axis = var_30385, x = aw_3855_cast)[name = tensor<string, []>("op_45540_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45541_cast = softmax(axis = var_30385, x = aw_3857_cast)[name = tensor<string, []>("op_45541_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45542_cast = softmax(axis = var_30385, x = aw_3859_cast)[name = tensor<string, []>("op_45542_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45543_cast = softmax(axis = var_30385, x = aw_3861_cast)[name = tensor<string, []>("op_45543_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45544_cast = softmax(axis = var_30385, x = aw_3863_cast)[name = tensor<string, []>("op_45544_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45545_cast = softmax(axis = var_30385, x = aw_3865_cast)[name = tensor<string, []>("op_45545_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45546_cast = softmax(axis = var_30385, x = aw_3867_cast)[name = tensor<string, []>("op_45546_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45547_cast = softmax(axis = var_30385, x = aw_3869_cast)[name = tensor<string, []>("op_45547_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45548_cast = softmax(axis = var_30385, x = aw_3871_cast)[name = tensor<string, []>("op_45548_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45549_cast = softmax(axis = var_30385, x = aw_3873_cast)[name = tensor<string, []>("op_45549_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45550_cast = softmax(axis = var_30385, x = aw_3875_cast)[name = tensor<string, []>("op_45550_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45551_cast = softmax(axis = var_30385, x = aw_3877_cast)[name = tensor<string, []>("op_45551_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_45552_cast = softmax(axis = var_30385, x = aw_3879_cast)[name = tensor<string, []>("op_45552_cast")];
            tensor<string, []> var_45554_equation_0 = const()[name = tensor<string, []>("op_45554_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45554_cast = einsum(equation = var_45554_equation_0, values = (var_45374_cast, var_45533_cast))[name = tensor<string, []>("op_45554_cast")];
            tensor<string, []> var_45556_equation_0 = const()[name = tensor<string, []>("op_45556_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45556_cast = einsum(equation = var_45556_equation_0, values = (var_45378_cast, var_45534_cast))[name = tensor<string, []>("op_45556_cast")];
            tensor<string, []> var_45558_equation_0 = const()[name = tensor<string, []>("op_45558_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45558_cast = einsum(equation = var_45558_equation_0, values = (var_45382_cast, var_45535_cast))[name = tensor<string, []>("op_45558_cast")];
            tensor<string, []> var_45560_equation_0 = const()[name = tensor<string, []>("op_45560_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45560_cast = einsum(equation = var_45560_equation_0, values = (var_45386_cast, var_45536_cast))[name = tensor<string, []>("op_45560_cast")];
            tensor<string, []> var_45562_equation_0 = const()[name = tensor<string, []>("op_45562_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45562_cast = einsum(equation = var_45562_equation_0, values = (var_45390_cast, var_45537_cast))[name = tensor<string, []>("op_45562_cast")];
            tensor<string, []> var_45564_equation_0 = const()[name = tensor<string, []>("op_45564_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45564_cast = einsum(equation = var_45564_equation_0, values = (var_45394_cast, var_45538_cast))[name = tensor<string, []>("op_45564_cast")];
            tensor<string, []> var_45566_equation_0 = const()[name = tensor<string, []>("op_45566_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45566_cast = einsum(equation = var_45566_equation_0, values = (var_45398_cast, var_45539_cast))[name = tensor<string, []>("op_45566_cast")];
            tensor<string, []> var_45568_equation_0 = const()[name = tensor<string, []>("op_45568_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45568_cast = einsum(equation = var_45568_equation_0, values = (var_45402_cast, var_45540_cast))[name = tensor<string, []>("op_45568_cast")];
            tensor<string, []> var_45570_equation_0 = const()[name = tensor<string, []>("op_45570_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45570_cast = einsum(equation = var_45570_equation_0, values = (var_45406_cast, var_45541_cast))[name = tensor<string, []>("op_45570_cast")];
            tensor<string, []> var_45572_equation_0 = const()[name = tensor<string, []>("op_45572_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45572_cast = einsum(equation = var_45572_equation_0, values = (var_45410_cast, var_45542_cast))[name = tensor<string, []>("op_45572_cast")];
            tensor<string, []> var_45574_equation_0 = const()[name = tensor<string, []>("op_45574_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45574_cast = einsum(equation = var_45574_equation_0, values = (var_45414_cast, var_45543_cast))[name = tensor<string, []>("op_45574_cast")];
            tensor<string, []> var_45576_equation_0 = const()[name = tensor<string, []>("op_45576_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45576_cast = einsum(equation = var_45576_equation_0, values = (var_45418_cast, var_45544_cast))[name = tensor<string, []>("op_45576_cast")];
            tensor<string, []> var_45578_equation_0 = const()[name = tensor<string, []>("op_45578_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45578_cast = einsum(equation = var_45578_equation_0, values = (var_45422_cast, var_45545_cast))[name = tensor<string, []>("op_45578_cast")];
            tensor<string, []> var_45580_equation_0 = const()[name = tensor<string, []>("op_45580_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45580_cast = einsum(equation = var_45580_equation_0, values = (var_45426_cast, var_45546_cast))[name = tensor<string, []>("op_45580_cast")];
            tensor<string, []> var_45582_equation_0 = const()[name = tensor<string, []>("op_45582_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45582_cast = einsum(equation = var_45582_equation_0, values = (var_45430_cast, var_45547_cast))[name = tensor<string, []>("op_45582_cast")];
            tensor<string, []> var_45584_equation_0 = const()[name = tensor<string, []>("op_45584_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45584_cast = einsum(equation = var_45584_equation_0, values = (var_45434_cast, var_45548_cast))[name = tensor<string, []>("op_45584_cast")];
            tensor<string, []> var_45586_equation_0 = const()[name = tensor<string, []>("op_45586_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45586_cast = einsum(equation = var_45586_equation_0, values = (var_45438_cast, var_45549_cast))[name = tensor<string, []>("op_45586_cast")];
            tensor<string, []> var_45588_equation_0 = const()[name = tensor<string, []>("op_45588_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45588_cast = einsum(equation = var_45588_equation_0, values = (var_45442_cast, var_45550_cast))[name = tensor<string, []>("op_45588_cast")];
            tensor<string, []> var_45590_equation_0 = const()[name = tensor<string, []>("op_45590_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45590_cast = einsum(equation = var_45590_equation_0, values = (var_45446_cast, var_45551_cast))[name = tensor<string, []>("op_45590_cast")];
            tensor<string, []> var_45592_equation_0 = const()[name = tensor<string, []>("op_45592_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45592_cast = einsum(equation = var_45592_equation_0, values = (var_45450_cast, var_45552_cast))[name = tensor<string, []>("op_45592_cast")];
            tensor<bool, []> input_587_interleave_0 = const()[name = tensor<string, []>("input_587_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_587_cast = concat(axis = var_30385, interleave = input_587_interleave_0, values = (var_45554_cast, var_45556_cast, var_45558_cast, var_45560_cast, var_45562_cast, var_45564_cast, var_45566_cast, var_45568_cast, var_45570_cast, var_45572_cast, var_45574_cast, var_45576_cast, var_45578_cast, var_45580_cast, var_45582_cast, var_45584_cast, var_45586_cast, var_45588_cast, var_45590_cast, var_45592_cast))[name = tensor<string, []>("input_587_cast")];
            tensor<int32, [2]> var_45598 = const()[name = tensor<string, []>("op_45598"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_45600 = const()[name = tensor<string, []>("op_45600"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_45602_pad_type_0 = const()[name = tensor<string, []>("op_45602_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_45602_pad_0 = const()[name = tensor<string, []>("op_45602_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3815359168)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3818636032)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_45602_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_out_0_bias_to_fp16, dilations = var_45600, groups = var_30385, pad = var_45602_pad_0, pad_type = var_45602_pad_type_0, strides = var_45598, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn1_to_out_0_weight_to_fp16, x = input_587_cast)[name = tensor<string, []>("op_45602_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_303_cast = add(x = var_45602_cast, y = inputs_301_cast)[name = tensor<string, []>("inputs_303_cast")];
            tensor<int32, [1]> var_45606 = const()[name = tensor<string, []>("op_45606"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_303_cast = reduce_mean(axes = var_45606, keep_dims = var_30380, x = inputs_303_cast)[name = tensor<string, []>("channels_mean_303_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_303_cast = sub(x = inputs_303_cast, y = channels_mean_303_cast)[name = tensor<string, []>("zero_mean_303_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_303_cast = mul(x = zero_mean_303_cast, y = zero_mean_303_cast)[name = tensor<string, []>("zero_mean_sq_303_cast")];
            tensor<int32, [1]> var_45610 = const()[name = tensor<string, []>("op_45610"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_45611_cast = reduce_mean(axes = var_45610, keep_dims = var_30380, x = zero_mean_sq_303_cast)[name = tensor<string, []>("op_45611_cast")];
            tensor<fp16, []> var_45612_to_fp16 = const()[name = tensor<string, []>("op_45612_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_45613_cast = add(x = var_45611_cast, y = var_45612_to_fp16)[name = tensor<string, []>("op_45613_cast")];
            tensor<fp16, []> denom_303_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_303_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_303_cast = rsqrt(epsilon = denom_303_epsilon_0_to_fp16, x = var_45613_cast)[name = tensor<string, []>("denom_303_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_303_cast = mul(x = zero_mean_303_cast, y = denom_303_cast)[name = tensor<string, []>("out_303_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_45617_to_fp16 = const()[name = tensor<string, []>("op_45617_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3818638656)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_45618_cast = add(x = out_303_cast, y = var_45617_to_fp16)[name = tensor<string, []>("op_45618_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_45620_to_fp16 = const()[name = tensor<string, []>("op_45620_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3818641280)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_399_cast = mul(x = var_45618_cast, y = var_45620_to_fp16)[name = tensor<string, []>("hidden_states_399_cast")];
            tensor<int32, [2]> var_45627 = const()[name = tensor<string, []>("op_45627"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_45629 = const()[name = tensor<string, []>("op_45629"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_203_pad_type_0 = const()[name = tensor<string, []>("q_203_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_203_pad_0 = const()[name = tensor<string, []>("q_203_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3818643904)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_203_cast = conv(dilations = var_45629, groups = var_30385, pad = q_203_pad_0, pad_type = q_203_pad_type_0, strides = var_45627, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_q_weight_to_fp16, x = hidden_states_399_cast)[name = tensor<string, []>("q_203_cast")];
            tensor<int32, [2]> var_45633 = const()[name = tensor<string, []>("op_45633"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_45635 = const()[name = tensor<string, []>("op_45635"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_405_pad_type_0 = const()[name = tensor<string, []>("k_405_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_405_pad_0 = const()[name = tensor<string, []>("k_405_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3821920768)))];
            tensor<fp16, [2, 1280, 1, 77]> k_405_cast = conv(dilations = var_45635, groups = var_30385, pad = k_405_pad_0, pad_type = k_405_pad_type_0, strides = var_45633, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_405_cast")];
            tensor<int32, [2]> var_45639 = const()[name = tensor<string, []>("op_45639"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_45641 = const()[name = tensor<string, []>("op_45641"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_203_pad_type_0 = const()[name = tensor<string, []>("v_203_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_203_pad_0 = const()[name = tensor<string, []>("v_203_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3827163712)))];
            tensor<fp16, [2, 1280, 1, 77]> v_203_cast = conv(dilations = var_45641, groups = var_30385, pad = v_203_pad_0, pad_type = v_203_pad_type_0, strides = var_45639, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_203_cast")];
            tensor<int32, [4]> var_45645_begin_0 = const()[name = tensor<string, []>("op_45645_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_45645_end_0 = const()[name = tensor<string, []>("op_45645_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_45645_end_mask_0 = const()[name = tensor<string, []>("op_45645_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45645_cast = slice_by_index(begin = var_45645_begin_0, end = var_45645_end_0, end_mask = var_45645_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45645_cast")];
            tensor<int32, [4]> var_45649_begin_0 = const()[name = tensor<string, []>("op_45649_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_45649_end_0 = const()[name = tensor<string, []>("op_45649_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_45649_end_mask_0 = const()[name = tensor<string, []>("op_45649_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45649_cast = slice_by_index(begin = var_45649_begin_0, end = var_45649_end_0, end_mask = var_45649_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45649_cast")];
            tensor<int32, [4]> var_45653_begin_0 = const()[name = tensor<string, []>("op_45653_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_45653_end_0 = const()[name = tensor<string, []>("op_45653_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_45653_end_mask_0 = const()[name = tensor<string, []>("op_45653_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45653_cast = slice_by_index(begin = var_45653_begin_0, end = var_45653_end_0, end_mask = var_45653_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45653_cast")];
            tensor<int32, [4]> var_45657_begin_0 = const()[name = tensor<string, []>("op_45657_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_45657_end_0 = const()[name = tensor<string, []>("op_45657_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_45657_end_mask_0 = const()[name = tensor<string, []>("op_45657_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45657_cast = slice_by_index(begin = var_45657_begin_0, end = var_45657_end_0, end_mask = var_45657_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45657_cast")];
            tensor<int32, [4]> var_45661_begin_0 = const()[name = tensor<string, []>("op_45661_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_45661_end_0 = const()[name = tensor<string, []>("op_45661_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_45661_end_mask_0 = const()[name = tensor<string, []>("op_45661_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45661_cast = slice_by_index(begin = var_45661_begin_0, end = var_45661_end_0, end_mask = var_45661_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45661_cast")];
            tensor<int32, [4]> var_45665_begin_0 = const()[name = tensor<string, []>("op_45665_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_45665_end_0 = const()[name = tensor<string, []>("op_45665_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_45665_end_mask_0 = const()[name = tensor<string, []>("op_45665_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45665_cast = slice_by_index(begin = var_45665_begin_0, end = var_45665_end_0, end_mask = var_45665_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45665_cast")];
            tensor<int32, [4]> var_45669_begin_0 = const()[name = tensor<string, []>("op_45669_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_45669_end_0 = const()[name = tensor<string, []>("op_45669_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_45669_end_mask_0 = const()[name = tensor<string, []>("op_45669_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45669_cast = slice_by_index(begin = var_45669_begin_0, end = var_45669_end_0, end_mask = var_45669_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45669_cast")];
            tensor<int32, [4]> var_45673_begin_0 = const()[name = tensor<string, []>("op_45673_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_45673_end_0 = const()[name = tensor<string, []>("op_45673_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_45673_end_mask_0 = const()[name = tensor<string, []>("op_45673_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45673_cast = slice_by_index(begin = var_45673_begin_0, end = var_45673_end_0, end_mask = var_45673_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45673_cast")];
            tensor<int32, [4]> var_45677_begin_0 = const()[name = tensor<string, []>("op_45677_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_45677_end_0 = const()[name = tensor<string, []>("op_45677_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_45677_end_mask_0 = const()[name = tensor<string, []>("op_45677_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45677_cast = slice_by_index(begin = var_45677_begin_0, end = var_45677_end_0, end_mask = var_45677_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45677_cast")];
            tensor<int32, [4]> var_45681_begin_0 = const()[name = tensor<string, []>("op_45681_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_45681_end_0 = const()[name = tensor<string, []>("op_45681_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_45681_end_mask_0 = const()[name = tensor<string, []>("op_45681_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45681_cast = slice_by_index(begin = var_45681_begin_0, end = var_45681_end_0, end_mask = var_45681_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45681_cast")];
            tensor<int32, [4]> var_45685_begin_0 = const()[name = tensor<string, []>("op_45685_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_45685_end_0 = const()[name = tensor<string, []>("op_45685_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_45685_end_mask_0 = const()[name = tensor<string, []>("op_45685_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45685_cast = slice_by_index(begin = var_45685_begin_0, end = var_45685_end_0, end_mask = var_45685_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45685_cast")];
            tensor<int32, [4]> var_45689_begin_0 = const()[name = tensor<string, []>("op_45689_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_45689_end_0 = const()[name = tensor<string, []>("op_45689_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_45689_end_mask_0 = const()[name = tensor<string, []>("op_45689_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45689_cast = slice_by_index(begin = var_45689_begin_0, end = var_45689_end_0, end_mask = var_45689_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45689_cast")];
            tensor<int32, [4]> var_45693_begin_0 = const()[name = tensor<string, []>("op_45693_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_45693_end_0 = const()[name = tensor<string, []>("op_45693_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_45693_end_mask_0 = const()[name = tensor<string, []>("op_45693_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45693_cast = slice_by_index(begin = var_45693_begin_0, end = var_45693_end_0, end_mask = var_45693_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45693_cast")];
            tensor<int32, [4]> var_45697_begin_0 = const()[name = tensor<string, []>("op_45697_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_45697_end_0 = const()[name = tensor<string, []>("op_45697_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_45697_end_mask_0 = const()[name = tensor<string, []>("op_45697_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45697_cast = slice_by_index(begin = var_45697_begin_0, end = var_45697_end_0, end_mask = var_45697_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45697_cast")];
            tensor<int32, [4]> var_45701_begin_0 = const()[name = tensor<string, []>("op_45701_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_45701_end_0 = const()[name = tensor<string, []>("op_45701_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_45701_end_mask_0 = const()[name = tensor<string, []>("op_45701_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45701_cast = slice_by_index(begin = var_45701_begin_0, end = var_45701_end_0, end_mask = var_45701_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45701_cast")];
            tensor<int32, [4]> var_45705_begin_0 = const()[name = tensor<string, []>("op_45705_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_45705_end_0 = const()[name = tensor<string, []>("op_45705_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_45705_end_mask_0 = const()[name = tensor<string, []>("op_45705_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45705_cast = slice_by_index(begin = var_45705_begin_0, end = var_45705_end_0, end_mask = var_45705_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45705_cast")];
            tensor<int32, [4]> var_45709_begin_0 = const()[name = tensor<string, []>("op_45709_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_45709_end_0 = const()[name = tensor<string, []>("op_45709_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_45709_end_mask_0 = const()[name = tensor<string, []>("op_45709_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45709_cast = slice_by_index(begin = var_45709_begin_0, end = var_45709_end_0, end_mask = var_45709_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45709_cast")];
            tensor<int32, [4]> var_45713_begin_0 = const()[name = tensor<string, []>("op_45713_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_45713_end_0 = const()[name = tensor<string, []>("op_45713_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_45713_end_mask_0 = const()[name = tensor<string, []>("op_45713_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45713_cast = slice_by_index(begin = var_45713_begin_0, end = var_45713_end_0, end_mask = var_45713_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45713_cast")];
            tensor<int32, [4]> var_45717_begin_0 = const()[name = tensor<string, []>("op_45717_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_45717_end_0 = const()[name = tensor<string, []>("op_45717_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_45717_end_mask_0 = const()[name = tensor<string, []>("op_45717_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45717_cast = slice_by_index(begin = var_45717_begin_0, end = var_45717_end_0, end_mask = var_45717_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45717_cast")];
            tensor<int32, [4]> var_45721_begin_0 = const()[name = tensor<string, []>("op_45721_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_45721_end_0 = const()[name = tensor<string, []>("op_45721_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_45721_end_mask_0 = const()[name = tensor<string, []>("op_45721_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_45721_cast = slice_by_index(begin = var_45721_begin_0, end = var_45721_end_0, end_mask = var_45721_end_mask_0, x = q_203_cast)[name = tensor<string, []>("op_45721_cast")];
            tensor<int32, [4]> k_407_perm_0 = const()[name = tensor<string, []>("k_407_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_45728_begin_0 = const()[name = tensor<string, []>("op_45728_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_45728_end_0 = const()[name = tensor<string, []>("op_45728_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_45728_end_mask_0 = const()[name = tensor<string, []>("op_45728_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_38 = transpose(perm = k_407_perm_0, x = k_405_cast)[name = tensor<string, []>("transpose_38")];
            tensor<fp16, [2, 77, 1, 64]> var_45728_cast = slice_by_index(begin = var_45728_begin_0, end = var_45728_end_0, end_mask = var_45728_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45728_cast")];
            tensor<int32, [4]> var_45732_begin_0 = const()[name = tensor<string, []>("op_45732_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_45732_end_0 = const()[name = tensor<string, []>("op_45732_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_45732_end_mask_0 = const()[name = tensor<string, []>("op_45732_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45732_cast = slice_by_index(begin = var_45732_begin_0, end = var_45732_end_0, end_mask = var_45732_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45732_cast")];
            tensor<int32, [4]> var_45736_begin_0 = const()[name = tensor<string, []>("op_45736_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_45736_end_0 = const()[name = tensor<string, []>("op_45736_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_45736_end_mask_0 = const()[name = tensor<string, []>("op_45736_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45736_cast = slice_by_index(begin = var_45736_begin_0, end = var_45736_end_0, end_mask = var_45736_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45736_cast")];
            tensor<int32, [4]> var_45740_begin_0 = const()[name = tensor<string, []>("op_45740_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_45740_end_0 = const()[name = tensor<string, []>("op_45740_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_45740_end_mask_0 = const()[name = tensor<string, []>("op_45740_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45740_cast = slice_by_index(begin = var_45740_begin_0, end = var_45740_end_0, end_mask = var_45740_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45740_cast")];
            tensor<int32, [4]> var_45744_begin_0 = const()[name = tensor<string, []>("op_45744_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_45744_end_0 = const()[name = tensor<string, []>("op_45744_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_45744_end_mask_0 = const()[name = tensor<string, []>("op_45744_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45744_cast = slice_by_index(begin = var_45744_begin_0, end = var_45744_end_0, end_mask = var_45744_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45744_cast")];
            tensor<int32, [4]> var_45748_begin_0 = const()[name = tensor<string, []>("op_45748_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_45748_end_0 = const()[name = tensor<string, []>("op_45748_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_45748_end_mask_0 = const()[name = tensor<string, []>("op_45748_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45748_cast = slice_by_index(begin = var_45748_begin_0, end = var_45748_end_0, end_mask = var_45748_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45748_cast")];
            tensor<int32, [4]> var_45752_begin_0 = const()[name = tensor<string, []>("op_45752_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_45752_end_0 = const()[name = tensor<string, []>("op_45752_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_45752_end_mask_0 = const()[name = tensor<string, []>("op_45752_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45752_cast = slice_by_index(begin = var_45752_begin_0, end = var_45752_end_0, end_mask = var_45752_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45752_cast")];
            tensor<int32, [4]> var_45756_begin_0 = const()[name = tensor<string, []>("op_45756_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_45756_end_0 = const()[name = tensor<string, []>("op_45756_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_45756_end_mask_0 = const()[name = tensor<string, []>("op_45756_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45756_cast = slice_by_index(begin = var_45756_begin_0, end = var_45756_end_0, end_mask = var_45756_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45756_cast")];
            tensor<int32, [4]> var_45760_begin_0 = const()[name = tensor<string, []>("op_45760_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_45760_end_0 = const()[name = tensor<string, []>("op_45760_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_45760_end_mask_0 = const()[name = tensor<string, []>("op_45760_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45760_cast = slice_by_index(begin = var_45760_begin_0, end = var_45760_end_0, end_mask = var_45760_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45760_cast")];
            tensor<int32, [4]> var_45764_begin_0 = const()[name = tensor<string, []>("op_45764_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_45764_end_0 = const()[name = tensor<string, []>("op_45764_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_45764_end_mask_0 = const()[name = tensor<string, []>("op_45764_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45764_cast = slice_by_index(begin = var_45764_begin_0, end = var_45764_end_0, end_mask = var_45764_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45764_cast")];
            tensor<int32, [4]> var_45768_begin_0 = const()[name = tensor<string, []>("op_45768_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_45768_end_0 = const()[name = tensor<string, []>("op_45768_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_45768_end_mask_0 = const()[name = tensor<string, []>("op_45768_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45768_cast = slice_by_index(begin = var_45768_begin_0, end = var_45768_end_0, end_mask = var_45768_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45768_cast")];
            tensor<int32, [4]> var_45772_begin_0 = const()[name = tensor<string, []>("op_45772_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_45772_end_0 = const()[name = tensor<string, []>("op_45772_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_45772_end_mask_0 = const()[name = tensor<string, []>("op_45772_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45772_cast = slice_by_index(begin = var_45772_begin_0, end = var_45772_end_0, end_mask = var_45772_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45772_cast")];
            tensor<int32, [4]> var_45776_begin_0 = const()[name = tensor<string, []>("op_45776_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_45776_end_0 = const()[name = tensor<string, []>("op_45776_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_45776_end_mask_0 = const()[name = tensor<string, []>("op_45776_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45776_cast = slice_by_index(begin = var_45776_begin_0, end = var_45776_end_0, end_mask = var_45776_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45776_cast")];
            tensor<int32, [4]> var_45780_begin_0 = const()[name = tensor<string, []>("op_45780_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_45780_end_0 = const()[name = tensor<string, []>("op_45780_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_45780_end_mask_0 = const()[name = tensor<string, []>("op_45780_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45780_cast = slice_by_index(begin = var_45780_begin_0, end = var_45780_end_0, end_mask = var_45780_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45780_cast")];
            tensor<int32, [4]> var_45784_begin_0 = const()[name = tensor<string, []>("op_45784_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_45784_end_0 = const()[name = tensor<string, []>("op_45784_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_45784_end_mask_0 = const()[name = tensor<string, []>("op_45784_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45784_cast = slice_by_index(begin = var_45784_begin_0, end = var_45784_end_0, end_mask = var_45784_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45784_cast")];
            tensor<int32, [4]> var_45788_begin_0 = const()[name = tensor<string, []>("op_45788_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_45788_end_0 = const()[name = tensor<string, []>("op_45788_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_45788_end_mask_0 = const()[name = tensor<string, []>("op_45788_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45788_cast = slice_by_index(begin = var_45788_begin_0, end = var_45788_end_0, end_mask = var_45788_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45788_cast")];
            tensor<int32, [4]> var_45792_begin_0 = const()[name = tensor<string, []>("op_45792_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_45792_end_0 = const()[name = tensor<string, []>("op_45792_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_45792_end_mask_0 = const()[name = tensor<string, []>("op_45792_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45792_cast = slice_by_index(begin = var_45792_begin_0, end = var_45792_end_0, end_mask = var_45792_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45792_cast")];
            tensor<int32, [4]> var_45796_begin_0 = const()[name = tensor<string, []>("op_45796_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_45796_end_0 = const()[name = tensor<string, []>("op_45796_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_45796_end_mask_0 = const()[name = tensor<string, []>("op_45796_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45796_cast = slice_by_index(begin = var_45796_begin_0, end = var_45796_end_0, end_mask = var_45796_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45796_cast")];
            tensor<int32, [4]> var_45800_begin_0 = const()[name = tensor<string, []>("op_45800_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_45800_end_0 = const()[name = tensor<string, []>("op_45800_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_45800_end_mask_0 = const()[name = tensor<string, []>("op_45800_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45800_cast = slice_by_index(begin = var_45800_begin_0, end = var_45800_end_0, end_mask = var_45800_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45800_cast")];
            tensor<int32, [4]> var_45804_begin_0 = const()[name = tensor<string, []>("op_45804_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_45804_end_0 = const()[name = tensor<string, []>("op_45804_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_45804_end_mask_0 = const()[name = tensor<string, []>("op_45804_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_45804_cast = slice_by_index(begin = var_45804_begin_0, end = var_45804_end_0, end_mask = var_45804_end_mask_0, x = transpose_38)[name = tensor<string, []>("op_45804_cast")];
            tensor<int32, [4]> var_45806_begin_0 = const()[name = tensor<string, []>("op_45806_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_45806_end_0 = const()[name = tensor<string, []>("op_45806_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_45806_end_mask_0 = const()[name = tensor<string, []>("op_45806_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45806_cast = slice_by_index(begin = var_45806_begin_0, end = var_45806_end_0, end_mask = var_45806_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45806_cast")];
            tensor<int32, [4]> var_45810_begin_0 = const()[name = tensor<string, []>("op_45810_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_45810_end_0 = const()[name = tensor<string, []>("op_45810_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_45810_end_mask_0 = const()[name = tensor<string, []>("op_45810_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45810_cast = slice_by_index(begin = var_45810_begin_0, end = var_45810_end_0, end_mask = var_45810_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45810_cast")];
            tensor<int32, [4]> var_45814_begin_0 = const()[name = tensor<string, []>("op_45814_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_45814_end_0 = const()[name = tensor<string, []>("op_45814_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_45814_end_mask_0 = const()[name = tensor<string, []>("op_45814_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45814_cast = slice_by_index(begin = var_45814_begin_0, end = var_45814_end_0, end_mask = var_45814_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45814_cast")];
            tensor<int32, [4]> var_45818_begin_0 = const()[name = tensor<string, []>("op_45818_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_45818_end_0 = const()[name = tensor<string, []>("op_45818_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_45818_end_mask_0 = const()[name = tensor<string, []>("op_45818_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45818_cast = slice_by_index(begin = var_45818_begin_0, end = var_45818_end_0, end_mask = var_45818_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45818_cast")];
            tensor<int32, [4]> var_45822_begin_0 = const()[name = tensor<string, []>("op_45822_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_45822_end_0 = const()[name = tensor<string, []>("op_45822_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_45822_end_mask_0 = const()[name = tensor<string, []>("op_45822_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45822_cast = slice_by_index(begin = var_45822_begin_0, end = var_45822_end_0, end_mask = var_45822_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45822_cast")];
            tensor<int32, [4]> var_45826_begin_0 = const()[name = tensor<string, []>("op_45826_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_45826_end_0 = const()[name = tensor<string, []>("op_45826_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_45826_end_mask_0 = const()[name = tensor<string, []>("op_45826_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45826_cast = slice_by_index(begin = var_45826_begin_0, end = var_45826_end_0, end_mask = var_45826_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45826_cast")];
            tensor<int32, [4]> var_45830_begin_0 = const()[name = tensor<string, []>("op_45830_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_45830_end_0 = const()[name = tensor<string, []>("op_45830_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_45830_end_mask_0 = const()[name = tensor<string, []>("op_45830_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45830_cast = slice_by_index(begin = var_45830_begin_0, end = var_45830_end_0, end_mask = var_45830_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45830_cast")];
            tensor<int32, [4]> var_45834_begin_0 = const()[name = tensor<string, []>("op_45834_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_45834_end_0 = const()[name = tensor<string, []>("op_45834_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_45834_end_mask_0 = const()[name = tensor<string, []>("op_45834_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45834_cast = slice_by_index(begin = var_45834_begin_0, end = var_45834_end_0, end_mask = var_45834_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45834_cast")];
            tensor<int32, [4]> var_45838_begin_0 = const()[name = tensor<string, []>("op_45838_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_45838_end_0 = const()[name = tensor<string, []>("op_45838_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_45838_end_mask_0 = const()[name = tensor<string, []>("op_45838_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45838_cast = slice_by_index(begin = var_45838_begin_0, end = var_45838_end_0, end_mask = var_45838_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45838_cast")];
            tensor<int32, [4]> var_45842_begin_0 = const()[name = tensor<string, []>("op_45842_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_45842_end_0 = const()[name = tensor<string, []>("op_45842_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_45842_end_mask_0 = const()[name = tensor<string, []>("op_45842_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45842_cast = slice_by_index(begin = var_45842_begin_0, end = var_45842_end_0, end_mask = var_45842_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45842_cast")];
            tensor<int32, [4]> var_45846_begin_0 = const()[name = tensor<string, []>("op_45846_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_45846_end_0 = const()[name = tensor<string, []>("op_45846_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_45846_end_mask_0 = const()[name = tensor<string, []>("op_45846_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45846_cast = slice_by_index(begin = var_45846_begin_0, end = var_45846_end_0, end_mask = var_45846_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45846_cast")];
            tensor<int32, [4]> var_45850_begin_0 = const()[name = tensor<string, []>("op_45850_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_45850_end_0 = const()[name = tensor<string, []>("op_45850_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_45850_end_mask_0 = const()[name = tensor<string, []>("op_45850_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45850_cast = slice_by_index(begin = var_45850_begin_0, end = var_45850_end_0, end_mask = var_45850_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45850_cast")];
            tensor<int32, [4]> var_45854_begin_0 = const()[name = tensor<string, []>("op_45854_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_45854_end_0 = const()[name = tensor<string, []>("op_45854_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_45854_end_mask_0 = const()[name = tensor<string, []>("op_45854_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45854_cast = slice_by_index(begin = var_45854_begin_0, end = var_45854_end_0, end_mask = var_45854_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45854_cast")];
            tensor<int32, [4]> var_45858_begin_0 = const()[name = tensor<string, []>("op_45858_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_45858_end_0 = const()[name = tensor<string, []>("op_45858_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_45858_end_mask_0 = const()[name = tensor<string, []>("op_45858_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45858_cast = slice_by_index(begin = var_45858_begin_0, end = var_45858_end_0, end_mask = var_45858_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45858_cast")];
            tensor<int32, [4]> var_45862_begin_0 = const()[name = tensor<string, []>("op_45862_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_45862_end_0 = const()[name = tensor<string, []>("op_45862_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_45862_end_mask_0 = const()[name = tensor<string, []>("op_45862_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45862_cast = slice_by_index(begin = var_45862_begin_0, end = var_45862_end_0, end_mask = var_45862_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45862_cast")];
            tensor<int32, [4]> var_45866_begin_0 = const()[name = tensor<string, []>("op_45866_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_45866_end_0 = const()[name = tensor<string, []>("op_45866_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_45866_end_mask_0 = const()[name = tensor<string, []>("op_45866_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45866_cast = slice_by_index(begin = var_45866_begin_0, end = var_45866_end_0, end_mask = var_45866_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45866_cast")];
            tensor<int32, [4]> var_45870_begin_0 = const()[name = tensor<string, []>("op_45870_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_45870_end_0 = const()[name = tensor<string, []>("op_45870_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_45870_end_mask_0 = const()[name = tensor<string, []>("op_45870_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45870_cast = slice_by_index(begin = var_45870_begin_0, end = var_45870_end_0, end_mask = var_45870_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45870_cast")];
            tensor<int32, [4]> var_45874_begin_0 = const()[name = tensor<string, []>("op_45874_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_45874_end_0 = const()[name = tensor<string, []>("op_45874_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_45874_end_mask_0 = const()[name = tensor<string, []>("op_45874_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45874_cast = slice_by_index(begin = var_45874_begin_0, end = var_45874_end_0, end_mask = var_45874_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45874_cast")];
            tensor<int32, [4]> var_45878_begin_0 = const()[name = tensor<string, []>("op_45878_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_45878_end_0 = const()[name = tensor<string, []>("op_45878_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_45878_end_mask_0 = const()[name = tensor<string, []>("op_45878_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45878_cast = slice_by_index(begin = var_45878_begin_0, end = var_45878_end_0, end_mask = var_45878_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45878_cast")];
            tensor<int32, [4]> var_45882_begin_0 = const()[name = tensor<string, []>("op_45882_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_45882_end_0 = const()[name = tensor<string, []>("op_45882_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_45882_end_mask_0 = const()[name = tensor<string, []>("op_45882_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_45882_cast = slice_by_index(begin = var_45882_begin_0, end = var_45882_end_0, end_mask = var_45882_end_mask_0, x = v_203_cast)[name = tensor<string, []>("op_45882_cast")];
            tensor<string, []> var_45886_equation_0 = const()[name = tensor<string, []>("op_45886_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45886_cast = einsum(equation = var_45886_equation_0, values = (var_45728_cast, var_45645_cast))[name = tensor<string, []>("op_45886_cast")];
            tensor<fp16, []> var_45887_to_fp16 = const()[name = tensor<string, []>("op_45887_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3881_cast = mul(x = var_45886_cast, y = var_45887_to_fp16)[name = tensor<string, []>("aw_3881_cast")];
            tensor<string, []> var_45890_equation_0 = const()[name = tensor<string, []>("op_45890_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45890_cast = einsum(equation = var_45890_equation_0, values = (var_45732_cast, var_45649_cast))[name = tensor<string, []>("op_45890_cast")];
            tensor<fp16, []> var_45891_to_fp16 = const()[name = tensor<string, []>("op_45891_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3883_cast = mul(x = var_45890_cast, y = var_45891_to_fp16)[name = tensor<string, []>("aw_3883_cast")];
            tensor<string, []> var_45894_equation_0 = const()[name = tensor<string, []>("op_45894_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45894_cast = einsum(equation = var_45894_equation_0, values = (var_45736_cast, var_45653_cast))[name = tensor<string, []>("op_45894_cast")];
            tensor<fp16, []> var_45895_to_fp16 = const()[name = tensor<string, []>("op_45895_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3885_cast = mul(x = var_45894_cast, y = var_45895_to_fp16)[name = tensor<string, []>("aw_3885_cast")];
            tensor<string, []> var_45898_equation_0 = const()[name = tensor<string, []>("op_45898_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45898_cast = einsum(equation = var_45898_equation_0, values = (var_45740_cast, var_45657_cast))[name = tensor<string, []>("op_45898_cast")];
            tensor<fp16, []> var_45899_to_fp16 = const()[name = tensor<string, []>("op_45899_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3887_cast = mul(x = var_45898_cast, y = var_45899_to_fp16)[name = tensor<string, []>("aw_3887_cast")];
            tensor<string, []> var_45902_equation_0 = const()[name = tensor<string, []>("op_45902_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45902_cast = einsum(equation = var_45902_equation_0, values = (var_45744_cast, var_45661_cast))[name = tensor<string, []>("op_45902_cast")];
            tensor<fp16, []> var_45903_to_fp16 = const()[name = tensor<string, []>("op_45903_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3889_cast = mul(x = var_45902_cast, y = var_45903_to_fp16)[name = tensor<string, []>("aw_3889_cast")];
            tensor<string, []> var_45906_equation_0 = const()[name = tensor<string, []>("op_45906_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45906_cast = einsum(equation = var_45906_equation_0, values = (var_45748_cast, var_45665_cast))[name = tensor<string, []>("op_45906_cast")];
            tensor<fp16, []> var_45907_to_fp16 = const()[name = tensor<string, []>("op_45907_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3891_cast = mul(x = var_45906_cast, y = var_45907_to_fp16)[name = tensor<string, []>("aw_3891_cast")];
            tensor<string, []> var_45910_equation_0 = const()[name = tensor<string, []>("op_45910_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45910_cast = einsum(equation = var_45910_equation_0, values = (var_45752_cast, var_45669_cast))[name = tensor<string, []>("op_45910_cast")];
            tensor<fp16, []> var_45911_to_fp16 = const()[name = tensor<string, []>("op_45911_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3893_cast = mul(x = var_45910_cast, y = var_45911_to_fp16)[name = tensor<string, []>("aw_3893_cast")];
            tensor<string, []> var_45914_equation_0 = const()[name = tensor<string, []>("op_45914_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45914_cast = einsum(equation = var_45914_equation_0, values = (var_45756_cast, var_45673_cast))[name = tensor<string, []>("op_45914_cast")];
            tensor<fp16, []> var_45915_to_fp16 = const()[name = tensor<string, []>("op_45915_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3895_cast = mul(x = var_45914_cast, y = var_45915_to_fp16)[name = tensor<string, []>("aw_3895_cast")];
            tensor<string, []> var_45918_equation_0 = const()[name = tensor<string, []>("op_45918_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45918_cast = einsum(equation = var_45918_equation_0, values = (var_45760_cast, var_45677_cast))[name = tensor<string, []>("op_45918_cast")];
            tensor<fp16, []> var_45919_to_fp16 = const()[name = tensor<string, []>("op_45919_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3897_cast = mul(x = var_45918_cast, y = var_45919_to_fp16)[name = tensor<string, []>("aw_3897_cast")];
            tensor<string, []> var_45922_equation_0 = const()[name = tensor<string, []>("op_45922_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45922_cast = einsum(equation = var_45922_equation_0, values = (var_45764_cast, var_45681_cast))[name = tensor<string, []>("op_45922_cast")];
            tensor<fp16, []> var_45923_to_fp16 = const()[name = tensor<string, []>("op_45923_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3899_cast = mul(x = var_45922_cast, y = var_45923_to_fp16)[name = tensor<string, []>("aw_3899_cast")];
            tensor<string, []> var_45926_equation_0 = const()[name = tensor<string, []>("op_45926_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45926_cast = einsum(equation = var_45926_equation_0, values = (var_45768_cast, var_45685_cast))[name = tensor<string, []>("op_45926_cast")];
            tensor<fp16, []> var_45927_to_fp16 = const()[name = tensor<string, []>("op_45927_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3901_cast = mul(x = var_45926_cast, y = var_45927_to_fp16)[name = tensor<string, []>("aw_3901_cast")];
            tensor<string, []> var_45930_equation_0 = const()[name = tensor<string, []>("op_45930_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45930_cast = einsum(equation = var_45930_equation_0, values = (var_45772_cast, var_45689_cast))[name = tensor<string, []>("op_45930_cast")];
            tensor<fp16, []> var_45931_to_fp16 = const()[name = tensor<string, []>("op_45931_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3903_cast = mul(x = var_45930_cast, y = var_45931_to_fp16)[name = tensor<string, []>("aw_3903_cast")];
            tensor<string, []> var_45934_equation_0 = const()[name = tensor<string, []>("op_45934_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45934_cast = einsum(equation = var_45934_equation_0, values = (var_45776_cast, var_45693_cast))[name = tensor<string, []>("op_45934_cast")];
            tensor<fp16, []> var_45935_to_fp16 = const()[name = tensor<string, []>("op_45935_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3905_cast = mul(x = var_45934_cast, y = var_45935_to_fp16)[name = tensor<string, []>("aw_3905_cast")];
            tensor<string, []> var_45938_equation_0 = const()[name = tensor<string, []>("op_45938_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45938_cast = einsum(equation = var_45938_equation_0, values = (var_45780_cast, var_45697_cast))[name = tensor<string, []>("op_45938_cast")];
            tensor<fp16, []> var_45939_to_fp16 = const()[name = tensor<string, []>("op_45939_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3907_cast = mul(x = var_45938_cast, y = var_45939_to_fp16)[name = tensor<string, []>("aw_3907_cast")];
            tensor<string, []> var_45942_equation_0 = const()[name = tensor<string, []>("op_45942_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45942_cast = einsum(equation = var_45942_equation_0, values = (var_45784_cast, var_45701_cast))[name = tensor<string, []>("op_45942_cast")];
            tensor<fp16, []> var_45943_to_fp16 = const()[name = tensor<string, []>("op_45943_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3909_cast = mul(x = var_45942_cast, y = var_45943_to_fp16)[name = tensor<string, []>("aw_3909_cast")];
            tensor<string, []> var_45946_equation_0 = const()[name = tensor<string, []>("op_45946_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45946_cast = einsum(equation = var_45946_equation_0, values = (var_45788_cast, var_45705_cast))[name = tensor<string, []>("op_45946_cast")];
            tensor<fp16, []> var_45947_to_fp16 = const()[name = tensor<string, []>("op_45947_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3911_cast = mul(x = var_45946_cast, y = var_45947_to_fp16)[name = tensor<string, []>("aw_3911_cast")];
            tensor<string, []> var_45950_equation_0 = const()[name = tensor<string, []>("op_45950_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45950_cast = einsum(equation = var_45950_equation_0, values = (var_45792_cast, var_45709_cast))[name = tensor<string, []>("op_45950_cast")];
            tensor<fp16, []> var_45951_to_fp16 = const()[name = tensor<string, []>("op_45951_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3913_cast = mul(x = var_45950_cast, y = var_45951_to_fp16)[name = tensor<string, []>("aw_3913_cast")];
            tensor<string, []> var_45954_equation_0 = const()[name = tensor<string, []>("op_45954_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45954_cast = einsum(equation = var_45954_equation_0, values = (var_45796_cast, var_45713_cast))[name = tensor<string, []>("op_45954_cast")];
            tensor<fp16, []> var_45955_to_fp16 = const()[name = tensor<string, []>("op_45955_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3915_cast = mul(x = var_45954_cast, y = var_45955_to_fp16)[name = tensor<string, []>("aw_3915_cast")];
            tensor<string, []> var_45958_equation_0 = const()[name = tensor<string, []>("op_45958_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45958_cast = einsum(equation = var_45958_equation_0, values = (var_45800_cast, var_45717_cast))[name = tensor<string, []>("op_45958_cast")];
            tensor<fp16, []> var_45959_to_fp16 = const()[name = tensor<string, []>("op_45959_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3917_cast = mul(x = var_45958_cast, y = var_45959_to_fp16)[name = tensor<string, []>("aw_3917_cast")];
            tensor<string, []> var_45962_equation_0 = const()[name = tensor<string, []>("op_45962_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_45962_cast = einsum(equation = var_45962_equation_0, values = (var_45804_cast, var_45721_cast))[name = tensor<string, []>("op_45962_cast")];
            tensor<fp16, []> var_45963_to_fp16 = const()[name = tensor<string, []>("op_45963_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3919_cast = mul(x = var_45962_cast, y = var_45963_to_fp16)[name = tensor<string, []>("aw_3919_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45965_cast = softmax(axis = var_30385, x = aw_3881_cast)[name = tensor<string, []>("op_45965_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45966_cast = softmax(axis = var_30385, x = aw_3883_cast)[name = tensor<string, []>("op_45966_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45967_cast = softmax(axis = var_30385, x = aw_3885_cast)[name = tensor<string, []>("op_45967_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45968_cast = softmax(axis = var_30385, x = aw_3887_cast)[name = tensor<string, []>("op_45968_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45969_cast = softmax(axis = var_30385, x = aw_3889_cast)[name = tensor<string, []>("op_45969_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45970_cast = softmax(axis = var_30385, x = aw_3891_cast)[name = tensor<string, []>("op_45970_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45971_cast = softmax(axis = var_30385, x = aw_3893_cast)[name = tensor<string, []>("op_45971_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45972_cast = softmax(axis = var_30385, x = aw_3895_cast)[name = tensor<string, []>("op_45972_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45973_cast = softmax(axis = var_30385, x = aw_3897_cast)[name = tensor<string, []>("op_45973_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45974_cast = softmax(axis = var_30385, x = aw_3899_cast)[name = tensor<string, []>("op_45974_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45975_cast = softmax(axis = var_30385, x = aw_3901_cast)[name = tensor<string, []>("op_45975_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45976_cast = softmax(axis = var_30385, x = aw_3903_cast)[name = tensor<string, []>("op_45976_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45977_cast = softmax(axis = var_30385, x = aw_3905_cast)[name = tensor<string, []>("op_45977_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45978_cast = softmax(axis = var_30385, x = aw_3907_cast)[name = tensor<string, []>("op_45978_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45979_cast = softmax(axis = var_30385, x = aw_3909_cast)[name = tensor<string, []>("op_45979_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45980_cast = softmax(axis = var_30385, x = aw_3911_cast)[name = tensor<string, []>("op_45980_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45981_cast = softmax(axis = var_30385, x = aw_3913_cast)[name = tensor<string, []>("op_45981_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45982_cast = softmax(axis = var_30385, x = aw_3915_cast)[name = tensor<string, []>("op_45982_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45983_cast = softmax(axis = var_30385, x = aw_3917_cast)[name = tensor<string, []>("op_45983_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_45984_cast = softmax(axis = var_30385, x = aw_3919_cast)[name = tensor<string, []>("op_45984_cast")];
            tensor<string, []> var_45986_equation_0 = const()[name = tensor<string, []>("op_45986_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45986_cast = einsum(equation = var_45986_equation_0, values = (var_45806_cast, var_45965_cast))[name = tensor<string, []>("op_45986_cast")];
            tensor<string, []> var_45988_equation_0 = const()[name = tensor<string, []>("op_45988_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45988_cast = einsum(equation = var_45988_equation_0, values = (var_45810_cast, var_45966_cast))[name = tensor<string, []>("op_45988_cast")];
            tensor<string, []> var_45990_equation_0 = const()[name = tensor<string, []>("op_45990_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45990_cast = einsum(equation = var_45990_equation_0, values = (var_45814_cast, var_45967_cast))[name = tensor<string, []>("op_45990_cast")];
            tensor<string, []> var_45992_equation_0 = const()[name = tensor<string, []>("op_45992_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45992_cast = einsum(equation = var_45992_equation_0, values = (var_45818_cast, var_45968_cast))[name = tensor<string, []>("op_45992_cast")];
            tensor<string, []> var_45994_equation_0 = const()[name = tensor<string, []>("op_45994_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45994_cast = einsum(equation = var_45994_equation_0, values = (var_45822_cast, var_45969_cast))[name = tensor<string, []>("op_45994_cast")];
            tensor<string, []> var_45996_equation_0 = const()[name = tensor<string, []>("op_45996_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45996_cast = einsum(equation = var_45996_equation_0, values = (var_45826_cast, var_45970_cast))[name = tensor<string, []>("op_45996_cast")];
            tensor<string, []> var_45998_equation_0 = const()[name = tensor<string, []>("op_45998_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_45998_cast = einsum(equation = var_45998_equation_0, values = (var_45830_cast, var_45971_cast))[name = tensor<string, []>("op_45998_cast")];
            tensor<string, []> var_46000_equation_0 = const()[name = tensor<string, []>("op_46000_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46000_cast = einsum(equation = var_46000_equation_0, values = (var_45834_cast, var_45972_cast))[name = tensor<string, []>("op_46000_cast")];
            tensor<string, []> var_46002_equation_0 = const()[name = tensor<string, []>("op_46002_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46002_cast = einsum(equation = var_46002_equation_0, values = (var_45838_cast, var_45973_cast))[name = tensor<string, []>("op_46002_cast")];
            tensor<string, []> var_46004_equation_0 = const()[name = tensor<string, []>("op_46004_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46004_cast = einsum(equation = var_46004_equation_0, values = (var_45842_cast, var_45974_cast))[name = tensor<string, []>("op_46004_cast")];
            tensor<string, []> var_46006_equation_0 = const()[name = tensor<string, []>("op_46006_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46006_cast = einsum(equation = var_46006_equation_0, values = (var_45846_cast, var_45975_cast))[name = tensor<string, []>("op_46006_cast")];
            tensor<string, []> var_46008_equation_0 = const()[name = tensor<string, []>("op_46008_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46008_cast = einsum(equation = var_46008_equation_0, values = (var_45850_cast, var_45976_cast))[name = tensor<string, []>("op_46008_cast")];
            tensor<string, []> var_46010_equation_0 = const()[name = tensor<string, []>("op_46010_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46010_cast = einsum(equation = var_46010_equation_0, values = (var_45854_cast, var_45977_cast))[name = tensor<string, []>("op_46010_cast")];
            tensor<string, []> var_46012_equation_0 = const()[name = tensor<string, []>("op_46012_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46012_cast = einsum(equation = var_46012_equation_0, values = (var_45858_cast, var_45978_cast))[name = tensor<string, []>("op_46012_cast")];
            tensor<string, []> var_46014_equation_0 = const()[name = tensor<string, []>("op_46014_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46014_cast = einsum(equation = var_46014_equation_0, values = (var_45862_cast, var_45979_cast))[name = tensor<string, []>("op_46014_cast")];
            tensor<string, []> var_46016_equation_0 = const()[name = tensor<string, []>("op_46016_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46016_cast = einsum(equation = var_46016_equation_0, values = (var_45866_cast, var_45980_cast))[name = tensor<string, []>("op_46016_cast")];
            tensor<string, []> var_46018_equation_0 = const()[name = tensor<string, []>("op_46018_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46018_cast = einsum(equation = var_46018_equation_0, values = (var_45870_cast, var_45981_cast))[name = tensor<string, []>("op_46018_cast")];
            tensor<string, []> var_46020_equation_0 = const()[name = tensor<string, []>("op_46020_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46020_cast = einsum(equation = var_46020_equation_0, values = (var_45874_cast, var_45982_cast))[name = tensor<string, []>("op_46020_cast")];
            tensor<string, []> var_46022_equation_0 = const()[name = tensor<string, []>("op_46022_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46022_cast = einsum(equation = var_46022_equation_0, values = (var_45878_cast, var_45983_cast))[name = tensor<string, []>("op_46022_cast")];
            tensor<string, []> var_46024_equation_0 = const()[name = tensor<string, []>("op_46024_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46024_cast = einsum(equation = var_46024_equation_0, values = (var_45882_cast, var_45984_cast))[name = tensor<string, []>("op_46024_cast")];
            tensor<bool, []> input_589_interleave_0 = const()[name = tensor<string, []>("input_589_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_589_cast = concat(axis = var_30385, interleave = input_589_interleave_0, values = (var_45986_cast, var_45988_cast, var_45990_cast, var_45992_cast, var_45994_cast, var_45996_cast, var_45998_cast, var_46000_cast, var_46002_cast, var_46004_cast, var_46006_cast, var_46008_cast, var_46010_cast, var_46012_cast, var_46014_cast, var_46016_cast, var_46018_cast, var_46020_cast, var_46022_cast, var_46024_cast))[name = tensor<string, []>("input_589_cast")];
            tensor<int32, [2]> var_46030 = const()[name = tensor<string, []>("op_46030"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_46032 = const()[name = tensor<string, []>("op_46032"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_46034_pad_type_0 = const()[name = tensor<string, []>("op_46034_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_46034_pad_0 = const()[name = tensor<string, []>("op_46034_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3832406656)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3835683520)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_46034_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_out_0_bias_to_fp16, dilations = var_46032, groups = var_30385, pad = var_46034_pad_0, pad_type = var_46034_pad_type_0, strides = var_46030, weight = up_blocks_0_attentions_1_transformer_blocks_6_attn2_to_out_0_weight_to_fp16, x = input_589_cast)[name = tensor<string, []>("op_46034_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_305_cast = add(x = var_46034_cast, y = inputs_303_cast)[name = tensor<string, []>("inputs_305_cast")];
            tensor<int32, [1]> var_46038 = const()[name = tensor<string, []>("op_46038"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_305_cast = reduce_mean(axes = var_46038, keep_dims = var_30380, x = inputs_305_cast)[name = tensor<string, []>("channels_mean_305_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_305_cast = sub(x = inputs_305_cast, y = channels_mean_305_cast)[name = tensor<string, []>("zero_mean_305_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_305_cast = mul(x = zero_mean_305_cast, y = zero_mean_305_cast)[name = tensor<string, []>("zero_mean_sq_305_cast")];
            tensor<int32, [1]> var_46042 = const()[name = tensor<string, []>("op_46042"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_46043_cast = reduce_mean(axes = var_46042, keep_dims = var_30380, x = zero_mean_sq_305_cast)[name = tensor<string, []>("op_46043_cast")];
            tensor<fp16, []> var_46044_to_fp16 = const()[name = tensor<string, []>("op_46044_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_46045_cast = add(x = var_46043_cast, y = var_46044_to_fp16)[name = tensor<string, []>("op_46045_cast")];
            tensor<fp16, []> denom_305_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_305_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_305_cast = rsqrt(epsilon = denom_305_epsilon_0_to_fp16, x = var_46045_cast)[name = tensor<string, []>("denom_305_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_305_cast = mul(x = zero_mean_305_cast, y = denom_305_cast)[name = tensor<string, []>("out_305_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_46049_to_fp16 = const()[name = tensor<string, []>("op_46049_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3835686144)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_46050_cast = add(x = out_305_cast, y = var_46049_to_fp16)[name = tensor<string, []>("op_46050_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_46052_to_fp16 = const()[name = tensor<string, []>("op_46052_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3835688768)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_591_cast = mul(x = var_46050_cast, y = var_46052_to_fp16)[name = tensor<string, []>("input_591_cast")];
            tensor<int32, [2]> var_46060 = const()[name = tensor<string, []>("op_46060"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_46062 = const()[name = tensor<string, []>("op_46062"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_46064_pad_type_0 = const()[name = tensor<string, []>("op_46064_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_46064_pad_0 = const()[name = tensor<string, []>("op_46064_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3835691392)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_1_transformer_blocks_6_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3861905856)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_46064_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_6_ff_net_0_proj_bias_to_fp16, dilations = var_46062, groups = var_30385, pad = var_46064_pad_0, pad_type = var_46064_pad_type_0, strides = var_46060, weight = up_blocks_0_attentions_1_transformer_blocks_6_ff_net_0_proj_weight_to_fp16, x = input_591_cast)[name = tensor<string, []>("op_46064_cast")];
            tensor<int32, [2]> var_46065_split_sizes_0 = const()[name = tensor<string, []>("op_46065_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_46065_axis_0 = const()[name = tensor<string, []>("op_46065_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_46065_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_46065_cast_1 = split(axis = var_46065_axis_0, split_sizes = var_46065_split_sizes_0, x = var_46064_cast)[name = tensor<string, []>("op_46065_cast")];
            tensor<string, []> var_46067_mode_0 = const()[name = tensor<string, []>("op_46067_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_46067_cast = gelu(mode = var_46067_mode_0, x = var_46065_cast_1)[name = tensor<string, []>("op_46067_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_593_cast = mul(x = var_46065_cast_0, y = var_46067_cast)[name = tensor<string, []>("input_593_cast")];
            tensor<int32, [2]> var_46071 = const()[name = tensor<string, []>("op_46071"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_46073 = const()[name = tensor<string, []>("op_46073"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_46075_pad_type_0 = const()[name = tensor<string, []>("op_46075_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_46075_pad_0 = const()[name = tensor<string, []>("op_46075_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_6_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3861926400)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_6_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_6_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3875033664)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_46075_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_6_ff_net_2_bias_to_fp16, dilations = var_46073, groups = var_30385, pad = var_46075_pad_0, pad_type = var_46075_pad_type_0, strides = var_46071, weight = up_blocks_0_attentions_1_transformer_blocks_6_ff_net_2_weight_to_fp16, x = input_593_cast)[name = tensor<string, []>("op_46075_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_307_cast = add(x = var_46075_cast, y = inputs_305_cast)[name = tensor<string, []>("inputs_307_cast")];
            tensor<int32, [1]> var_46085 = const()[name = tensor<string, []>("op_46085"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_307_cast = reduce_mean(axes = var_46085, keep_dims = var_30380, x = inputs_307_cast)[name = tensor<string, []>("channels_mean_307_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_307_cast = sub(x = inputs_307_cast, y = channels_mean_307_cast)[name = tensor<string, []>("zero_mean_307_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_307_cast = mul(x = zero_mean_307_cast, y = zero_mean_307_cast)[name = tensor<string, []>("zero_mean_sq_307_cast")];
            tensor<int32, [1]> var_46089 = const()[name = tensor<string, []>("op_46089"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_46090_cast = reduce_mean(axes = var_46089, keep_dims = var_30380, x = zero_mean_sq_307_cast)[name = tensor<string, []>("op_46090_cast")];
            tensor<fp16, []> var_46091_to_fp16 = const()[name = tensor<string, []>("op_46091_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_46092_cast = add(x = var_46090_cast, y = var_46091_to_fp16)[name = tensor<string, []>("op_46092_cast")];
            tensor<fp16, []> denom_307_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_307_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_307_cast = rsqrt(epsilon = denom_307_epsilon_0_to_fp16, x = var_46092_cast)[name = tensor<string, []>("denom_307_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_307_cast = mul(x = zero_mean_307_cast, y = denom_307_cast)[name = tensor<string, []>("out_307_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_46096_to_fp16 = const()[name = tensor<string, []>("op_46096_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3875036288)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_46097_cast = add(x = out_307_cast, y = var_46096_to_fp16)[name = tensor<string, []>("op_46097_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_46099_to_fp16 = const()[name = tensor<string, []>("op_46099_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3875038912)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_403_cast = mul(x = var_46097_cast, y = var_46099_to_fp16)[name = tensor<string, []>("hidden_states_403_cast")];
            tensor<int32, [2]> var_46106 = const()[name = tensor<string, []>("op_46106"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_46108 = const()[name = tensor<string, []>("op_46108"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_205_pad_type_0 = const()[name = tensor<string, []>("q_205_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_205_pad_0 = const()[name = tensor<string, []>("q_205_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3875041536)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_205_cast = conv(dilations = var_46108, groups = var_30385, pad = q_205_pad_0, pad_type = q_205_pad_type_0, strides = var_46106, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_q_weight_to_fp16, x = hidden_states_403_cast)[name = tensor<string, []>("q_205_cast")];
            tensor<int32, [2]> var_46112 = const()[name = tensor<string, []>("op_46112"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_46114 = const()[name = tensor<string, []>("op_46114"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_409_pad_type_0 = const()[name = tensor<string, []>("k_409_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_409_pad_0 = const()[name = tensor<string, []>("k_409_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3878318400)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_409_cast = conv(dilations = var_46114, groups = var_30385, pad = k_409_pad_0, pad_type = k_409_pad_type_0, strides = var_46112, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_k_weight_to_fp16, x = hidden_states_403_cast)[name = tensor<string, []>("k_409_cast")];
            tensor<int32, [2]> var_46118 = const()[name = tensor<string, []>("op_46118"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_46120 = const()[name = tensor<string, []>("op_46120"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_205_pad_type_0 = const()[name = tensor<string, []>("v_205_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_205_pad_0 = const()[name = tensor<string, []>("v_205_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3881595264)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_205_cast = conv(dilations = var_46120, groups = var_30385, pad = v_205_pad_0, pad_type = v_205_pad_type_0, strides = var_46118, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_v_weight_to_fp16, x = hidden_states_403_cast)[name = tensor<string, []>("v_205_cast")];
            tensor<int32, [4]> var_46124_begin_0 = const()[name = tensor<string, []>("op_46124_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_46124_end_0 = const()[name = tensor<string, []>("op_46124_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_46124_end_mask_0 = const()[name = tensor<string, []>("op_46124_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46124_cast = slice_by_index(begin = var_46124_begin_0, end = var_46124_end_0, end_mask = var_46124_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46124_cast")];
            tensor<int32, [4]> var_46128_begin_0 = const()[name = tensor<string, []>("op_46128_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_46128_end_0 = const()[name = tensor<string, []>("op_46128_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_46128_end_mask_0 = const()[name = tensor<string, []>("op_46128_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46128_cast = slice_by_index(begin = var_46128_begin_0, end = var_46128_end_0, end_mask = var_46128_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46128_cast")];
            tensor<int32, [4]> var_46132_begin_0 = const()[name = tensor<string, []>("op_46132_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_46132_end_0 = const()[name = tensor<string, []>("op_46132_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_46132_end_mask_0 = const()[name = tensor<string, []>("op_46132_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46132_cast = slice_by_index(begin = var_46132_begin_0, end = var_46132_end_0, end_mask = var_46132_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46132_cast")];
            tensor<int32, [4]> var_46136_begin_0 = const()[name = tensor<string, []>("op_46136_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_46136_end_0 = const()[name = tensor<string, []>("op_46136_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_46136_end_mask_0 = const()[name = tensor<string, []>("op_46136_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46136_cast = slice_by_index(begin = var_46136_begin_0, end = var_46136_end_0, end_mask = var_46136_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46136_cast")];
            tensor<int32, [4]> var_46140_begin_0 = const()[name = tensor<string, []>("op_46140_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_46140_end_0 = const()[name = tensor<string, []>("op_46140_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_46140_end_mask_0 = const()[name = tensor<string, []>("op_46140_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46140_cast = slice_by_index(begin = var_46140_begin_0, end = var_46140_end_0, end_mask = var_46140_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46140_cast")];
            tensor<int32, [4]> var_46144_begin_0 = const()[name = tensor<string, []>("op_46144_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_46144_end_0 = const()[name = tensor<string, []>("op_46144_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_46144_end_mask_0 = const()[name = tensor<string, []>("op_46144_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46144_cast = slice_by_index(begin = var_46144_begin_0, end = var_46144_end_0, end_mask = var_46144_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46144_cast")];
            tensor<int32, [4]> var_46148_begin_0 = const()[name = tensor<string, []>("op_46148_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_46148_end_0 = const()[name = tensor<string, []>("op_46148_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_46148_end_mask_0 = const()[name = tensor<string, []>("op_46148_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46148_cast = slice_by_index(begin = var_46148_begin_0, end = var_46148_end_0, end_mask = var_46148_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46148_cast")];
            tensor<int32, [4]> var_46152_begin_0 = const()[name = tensor<string, []>("op_46152_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_46152_end_0 = const()[name = tensor<string, []>("op_46152_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_46152_end_mask_0 = const()[name = tensor<string, []>("op_46152_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46152_cast = slice_by_index(begin = var_46152_begin_0, end = var_46152_end_0, end_mask = var_46152_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46152_cast")];
            tensor<int32, [4]> var_46156_begin_0 = const()[name = tensor<string, []>("op_46156_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_46156_end_0 = const()[name = tensor<string, []>("op_46156_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_46156_end_mask_0 = const()[name = tensor<string, []>("op_46156_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46156_cast = slice_by_index(begin = var_46156_begin_0, end = var_46156_end_0, end_mask = var_46156_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46156_cast")];
            tensor<int32, [4]> var_46160_begin_0 = const()[name = tensor<string, []>("op_46160_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_46160_end_0 = const()[name = tensor<string, []>("op_46160_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_46160_end_mask_0 = const()[name = tensor<string, []>("op_46160_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46160_cast = slice_by_index(begin = var_46160_begin_0, end = var_46160_end_0, end_mask = var_46160_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46160_cast")];
            tensor<int32, [4]> var_46164_begin_0 = const()[name = tensor<string, []>("op_46164_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_46164_end_0 = const()[name = tensor<string, []>("op_46164_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_46164_end_mask_0 = const()[name = tensor<string, []>("op_46164_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46164_cast = slice_by_index(begin = var_46164_begin_0, end = var_46164_end_0, end_mask = var_46164_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46164_cast")];
            tensor<int32, [4]> var_46168_begin_0 = const()[name = tensor<string, []>("op_46168_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_46168_end_0 = const()[name = tensor<string, []>("op_46168_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_46168_end_mask_0 = const()[name = tensor<string, []>("op_46168_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46168_cast = slice_by_index(begin = var_46168_begin_0, end = var_46168_end_0, end_mask = var_46168_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46168_cast")];
            tensor<int32, [4]> var_46172_begin_0 = const()[name = tensor<string, []>("op_46172_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_46172_end_0 = const()[name = tensor<string, []>("op_46172_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_46172_end_mask_0 = const()[name = tensor<string, []>("op_46172_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46172_cast = slice_by_index(begin = var_46172_begin_0, end = var_46172_end_0, end_mask = var_46172_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46172_cast")];
            tensor<int32, [4]> var_46176_begin_0 = const()[name = tensor<string, []>("op_46176_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_46176_end_0 = const()[name = tensor<string, []>("op_46176_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_46176_end_mask_0 = const()[name = tensor<string, []>("op_46176_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46176_cast = slice_by_index(begin = var_46176_begin_0, end = var_46176_end_0, end_mask = var_46176_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46176_cast")];
            tensor<int32, [4]> var_46180_begin_0 = const()[name = tensor<string, []>("op_46180_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_46180_end_0 = const()[name = tensor<string, []>("op_46180_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_46180_end_mask_0 = const()[name = tensor<string, []>("op_46180_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46180_cast = slice_by_index(begin = var_46180_begin_0, end = var_46180_end_0, end_mask = var_46180_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46180_cast")];
            tensor<int32, [4]> var_46184_begin_0 = const()[name = tensor<string, []>("op_46184_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_46184_end_0 = const()[name = tensor<string, []>("op_46184_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_46184_end_mask_0 = const()[name = tensor<string, []>("op_46184_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46184_cast = slice_by_index(begin = var_46184_begin_0, end = var_46184_end_0, end_mask = var_46184_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46184_cast")];
            tensor<int32, [4]> var_46188_begin_0 = const()[name = tensor<string, []>("op_46188_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_46188_end_0 = const()[name = tensor<string, []>("op_46188_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_46188_end_mask_0 = const()[name = tensor<string, []>("op_46188_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46188_cast = slice_by_index(begin = var_46188_begin_0, end = var_46188_end_0, end_mask = var_46188_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46188_cast")];
            tensor<int32, [4]> var_46192_begin_0 = const()[name = tensor<string, []>("op_46192_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_46192_end_0 = const()[name = tensor<string, []>("op_46192_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_46192_end_mask_0 = const()[name = tensor<string, []>("op_46192_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46192_cast = slice_by_index(begin = var_46192_begin_0, end = var_46192_end_0, end_mask = var_46192_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46192_cast")];
            tensor<int32, [4]> var_46196_begin_0 = const()[name = tensor<string, []>("op_46196_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_46196_end_0 = const()[name = tensor<string, []>("op_46196_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_46196_end_mask_0 = const()[name = tensor<string, []>("op_46196_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46196_cast = slice_by_index(begin = var_46196_begin_0, end = var_46196_end_0, end_mask = var_46196_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46196_cast")];
            tensor<int32, [4]> var_46200_begin_0 = const()[name = tensor<string, []>("op_46200_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_46200_end_0 = const()[name = tensor<string, []>("op_46200_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_46200_end_mask_0 = const()[name = tensor<string, []>("op_46200_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46200_cast = slice_by_index(begin = var_46200_begin_0, end = var_46200_end_0, end_mask = var_46200_end_mask_0, x = q_205_cast)[name = tensor<string, []>("op_46200_cast")];
            tensor<int32, [4]> k_411_perm_0 = const()[name = tensor<string, []>("k_411_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_46207_begin_0 = const()[name = tensor<string, []>("op_46207_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_46207_end_0 = const()[name = tensor<string, []>("op_46207_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_46207_end_mask_0 = const()[name = tensor<string, []>("op_46207_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_37 = transpose(perm = k_411_perm_0, x = k_409_cast)[name = tensor<string, []>("transpose_37")];
            tensor<fp16, [2, 1024, 1, 64]> var_46207_cast = slice_by_index(begin = var_46207_begin_0, end = var_46207_end_0, end_mask = var_46207_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46207_cast")];
            tensor<int32, [4]> var_46211_begin_0 = const()[name = tensor<string, []>("op_46211_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_46211_end_0 = const()[name = tensor<string, []>("op_46211_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_46211_end_mask_0 = const()[name = tensor<string, []>("op_46211_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46211_cast = slice_by_index(begin = var_46211_begin_0, end = var_46211_end_0, end_mask = var_46211_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46211_cast")];
            tensor<int32, [4]> var_46215_begin_0 = const()[name = tensor<string, []>("op_46215_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_46215_end_0 = const()[name = tensor<string, []>("op_46215_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_46215_end_mask_0 = const()[name = tensor<string, []>("op_46215_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46215_cast = slice_by_index(begin = var_46215_begin_0, end = var_46215_end_0, end_mask = var_46215_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46215_cast")];
            tensor<int32, [4]> var_46219_begin_0 = const()[name = tensor<string, []>("op_46219_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_46219_end_0 = const()[name = tensor<string, []>("op_46219_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_46219_end_mask_0 = const()[name = tensor<string, []>("op_46219_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46219_cast = slice_by_index(begin = var_46219_begin_0, end = var_46219_end_0, end_mask = var_46219_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46219_cast")];
            tensor<int32, [4]> var_46223_begin_0 = const()[name = tensor<string, []>("op_46223_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_46223_end_0 = const()[name = tensor<string, []>("op_46223_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_46223_end_mask_0 = const()[name = tensor<string, []>("op_46223_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46223_cast = slice_by_index(begin = var_46223_begin_0, end = var_46223_end_0, end_mask = var_46223_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46223_cast")];
            tensor<int32, [4]> var_46227_begin_0 = const()[name = tensor<string, []>("op_46227_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_46227_end_0 = const()[name = tensor<string, []>("op_46227_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_46227_end_mask_0 = const()[name = tensor<string, []>("op_46227_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46227_cast = slice_by_index(begin = var_46227_begin_0, end = var_46227_end_0, end_mask = var_46227_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46227_cast")];
            tensor<int32, [4]> var_46231_begin_0 = const()[name = tensor<string, []>("op_46231_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_46231_end_0 = const()[name = tensor<string, []>("op_46231_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_46231_end_mask_0 = const()[name = tensor<string, []>("op_46231_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46231_cast = slice_by_index(begin = var_46231_begin_0, end = var_46231_end_0, end_mask = var_46231_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46231_cast")];
            tensor<int32, [4]> var_46235_begin_0 = const()[name = tensor<string, []>("op_46235_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_46235_end_0 = const()[name = tensor<string, []>("op_46235_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_46235_end_mask_0 = const()[name = tensor<string, []>("op_46235_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46235_cast = slice_by_index(begin = var_46235_begin_0, end = var_46235_end_0, end_mask = var_46235_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46235_cast")];
            tensor<int32, [4]> var_46239_begin_0 = const()[name = tensor<string, []>("op_46239_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_46239_end_0 = const()[name = tensor<string, []>("op_46239_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_46239_end_mask_0 = const()[name = tensor<string, []>("op_46239_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46239_cast = slice_by_index(begin = var_46239_begin_0, end = var_46239_end_0, end_mask = var_46239_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46239_cast")];
            tensor<int32, [4]> var_46243_begin_0 = const()[name = tensor<string, []>("op_46243_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_46243_end_0 = const()[name = tensor<string, []>("op_46243_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_46243_end_mask_0 = const()[name = tensor<string, []>("op_46243_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46243_cast = slice_by_index(begin = var_46243_begin_0, end = var_46243_end_0, end_mask = var_46243_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46243_cast")];
            tensor<int32, [4]> var_46247_begin_0 = const()[name = tensor<string, []>("op_46247_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_46247_end_0 = const()[name = tensor<string, []>("op_46247_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_46247_end_mask_0 = const()[name = tensor<string, []>("op_46247_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46247_cast = slice_by_index(begin = var_46247_begin_0, end = var_46247_end_0, end_mask = var_46247_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46247_cast")];
            tensor<int32, [4]> var_46251_begin_0 = const()[name = tensor<string, []>("op_46251_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_46251_end_0 = const()[name = tensor<string, []>("op_46251_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_46251_end_mask_0 = const()[name = tensor<string, []>("op_46251_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46251_cast = slice_by_index(begin = var_46251_begin_0, end = var_46251_end_0, end_mask = var_46251_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46251_cast")];
            tensor<int32, [4]> var_46255_begin_0 = const()[name = tensor<string, []>("op_46255_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_46255_end_0 = const()[name = tensor<string, []>("op_46255_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_46255_end_mask_0 = const()[name = tensor<string, []>("op_46255_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46255_cast = slice_by_index(begin = var_46255_begin_0, end = var_46255_end_0, end_mask = var_46255_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46255_cast")];
            tensor<int32, [4]> var_46259_begin_0 = const()[name = tensor<string, []>("op_46259_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_46259_end_0 = const()[name = tensor<string, []>("op_46259_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_46259_end_mask_0 = const()[name = tensor<string, []>("op_46259_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46259_cast = slice_by_index(begin = var_46259_begin_0, end = var_46259_end_0, end_mask = var_46259_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46259_cast")];
            tensor<int32, [4]> var_46263_begin_0 = const()[name = tensor<string, []>("op_46263_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_46263_end_0 = const()[name = tensor<string, []>("op_46263_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_46263_end_mask_0 = const()[name = tensor<string, []>("op_46263_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46263_cast = slice_by_index(begin = var_46263_begin_0, end = var_46263_end_0, end_mask = var_46263_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46263_cast")];
            tensor<int32, [4]> var_46267_begin_0 = const()[name = tensor<string, []>("op_46267_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_46267_end_0 = const()[name = tensor<string, []>("op_46267_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_46267_end_mask_0 = const()[name = tensor<string, []>("op_46267_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46267_cast = slice_by_index(begin = var_46267_begin_0, end = var_46267_end_0, end_mask = var_46267_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46267_cast")];
            tensor<int32, [4]> var_46271_begin_0 = const()[name = tensor<string, []>("op_46271_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_46271_end_0 = const()[name = tensor<string, []>("op_46271_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_46271_end_mask_0 = const()[name = tensor<string, []>("op_46271_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46271_cast = slice_by_index(begin = var_46271_begin_0, end = var_46271_end_0, end_mask = var_46271_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46271_cast")];
            tensor<int32, [4]> var_46275_begin_0 = const()[name = tensor<string, []>("op_46275_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_46275_end_0 = const()[name = tensor<string, []>("op_46275_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_46275_end_mask_0 = const()[name = tensor<string, []>("op_46275_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46275_cast = slice_by_index(begin = var_46275_begin_0, end = var_46275_end_0, end_mask = var_46275_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46275_cast")];
            tensor<int32, [4]> var_46279_begin_0 = const()[name = tensor<string, []>("op_46279_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_46279_end_0 = const()[name = tensor<string, []>("op_46279_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_46279_end_mask_0 = const()[name = tensor<string, []>("op_46279_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46279_cast = slice_by_index(begin = var_46279_begin_0, end = var_46279_end_0, end_mask = var_46279_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46279_cast")];
            tensor<int32, [4]> var_46283_begin_0 = const()[name = tensor<string, []>("op_46283_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_46283_end_0 = const()[name = tensor<string, []>("op_46283_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_46283_end_mask_0 = const()[name = tensor<string, []>("op_46283_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_46283_cast = slice_by_index(begin = var_46283_begin_0, end = var_46283_end_0, end_mask = var_46283_end_mask_0, x = transpose_37)[name = tensor<string, []>("op_46283_cast")];
            tensor<int32, [4]> var_46285_begin_0 = const()[name = tensor<string, []>("op_46285_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_46285_end_0 = const()[name = tensor<string, []>("op_46285_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_46285_end_mask_0 = const()[name = tensor<string, []>("op_46285_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46285_cast = slice_by_index(begin = var_46285_begin_0, end = var_46285_end_0, end_mask = var_46285_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46285_cast")];
            tensor<int32, [4]> var_46289_begin_0 = const()[name = tensor<string, []>("op_46289_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_46289_end_0 = const()[name = tensor<string, []>("op_46289_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_46289_end_mask_0 = const()[name = tensor<string, []>("op_46289_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46289_cast = slice_by_index(begin = var_46289_begin_0, end = var_46289_end_0, end_mask = var_46289_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46289_cast")];
            tensor<int32, [4]> var_46293_begin_0 = const()[name = tensor<string, []>("op_46293_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_46293_end_0 = const()[name = tensor<string, []>("op_46293_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_46293_end_mask_0 = const()[name = tensor<string, []>("op_46293_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46293_cast = slice_by_index(begin = var_46293_begin_0, end = var_46293_end_0, end_mask = var_46293_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46293_cast")];
            tensor<int32, [4]> var_46297_begin_0 = const()[name = tensor<string, []>("op_46297_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_46297_end_0 = const()[name = tensor<string, []>("op_46297_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_46297_end_mask_0 = const()[name = tensor<string, []>("op_46297_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46297_cast = slice_by_index(begin = var_46297_begin_0, end = var_46297_end_0, end_mask = var_46297_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46297_cast")];
            tensor<int32, [4]> var_46301_begin_0 = const()[name = tensor<string, []>("op_46301_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_46301_end_0 = const()[name = tensor<string, []>("op_46301_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_46301_end_mask_0 = const()[name = tensor<string, []>("op_46301_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46301_cast = slice_by_index(begin = var_46301_begin_0, end = var_46301_end_0, end_mask = var_46301_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46301_cast")];
            tensor<int32, [4]> var_46305_begin_0 = const()[name = tensor<string, []>("op_46305_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_46305_end_0 = const()[name = tensor<string, []>("op_46305_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_46305_end_mask_0 = const()[name = tensor<string, []>("op_46305_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46305_cast = slice_by_index(begin = var_46305_begin_0, end = var_46305_end_0, end_mask = var_46305_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46305_cast")];
            tensor<int32, [4]> var_46309_begin_0 = const()[name = tensor<string, []>("op_46309_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_46309_end_0 = const()[name = tensor<string, []>("op_46309_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_46309_end_mask_0 = const()[name = tensor<string, []>("op_46309_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46309_cast = slice_by_index(begin = var_46309_begin_0, end = var_46309_end_0, end_mask = var_46309_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46309_cast")];
            tensor<int32, [4]> var_46313_begin_0 = const()[name = tensor<string, []>("op_46313_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_46313_end_0 = const()[name = tensor<string, []>("op_46313_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_46313_end_mask_0 = const()[name = tensor<string, []>("op_46313_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46313_cast = slice_by_index(begin = var_46313_begin_0, end = var_46313_end_0, end_mask = var_46313_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46313_cast")];
            tensor<int32, [4]> var_46317_begin_0 = const()[name = tensor<string, []>("op_46317_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_46317_end_0 = const()[name = tensor<string, []>("op_46317_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_46317_end_mask_0 = const()[name = tensor<string, []>("op_46317_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46317_cast = slice_by_index(begin = var_46317_begin_0, end = var_46317_end_0, end_mask = var_46317_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46317_cast")];
            tensor<int32, [4]> var_46321_begin_0 = const()[name = tensor<string, []>("op_46321_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_46321_end_0 = const()[name = tensor<string, []>("op_46321_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_46321_end_mask_0 = const()[name = tensor<string, []>("op_46321_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46321_cast = slice_by_index(begin = var_46321_begin_0, end = var_46321_end_0, end_mask = var_46321_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46321_cast")];
            tensor<int32, [4]> var_46325_begin_0 = const()[name = tensor<string, []>("op_46325_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_46325_end_0 = const()[name = tensor<string, []>("op_46325_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_46325_end_mask_0 = const()[name = tensor<string, []>("op_46325_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46325_cast = slice_by_index(begin = var_46325_begin_0, end = var_46325_end_0, end_mask = var_46325_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46325_cast")];
            tensor<int32, [4]> var_46329_begin_0 = const()[name = tensor<string, []>("op_46329_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_46329_end_0 = const()[name = tensor<string, []>("op_46329_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_46329_end_mask_0 = const()[name = tensor<string, []>("op_46329_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46329_cast = slice_by_index(begin = var_46329_begin_0, end = var_46329_end_0, end_mask = var_46329_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46329_cast")];
            tensor<int32, [4]> var_46333_begin_0 = const()[name = tensor<string, []>("op_46333_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_46333_end_0 = const()[name = tensor<string, []>("op_46333_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_46333_end_mask_0 = const()[name = tensor<string, []>("op_46333_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46333_cast = slice_by_index(begin = var_46333_begin_0, end = var_46333_end_0, end_mask = var_46333_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46333_cast")];
            tensor<int32, [4]> var_46337_begin_0 = const()[name = tensor<string, []>("op_46337_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_46337_end_0 = const()[name = tensor<string, []>("op_46337_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_46337_end_mask_0 = const()[name = tensor<string, []>("op_46337_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46337_cast = slice_by_index(begin = var_46337_begin_0, end = var_46337_end_0, end_mask = var_46337_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46337_cast")];
            tensor<int32, [4]> var_46341_begin_0 = const()[name = tensor<string, []>("op_46341_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_46341_end_0 = const()[name = tensor<string, []>("op_46341_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_46341_end_mask_0 = const()[name = tensor<string, []>("op_46341_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46341_cast = slice_by_index(begin = var_46341_begin_0, end = var_46341_end_0, end_mask = var_46341_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46341_cast")];
            tensor<int32, [4]> var_46345_begin_0 = const()[name = tensor<string, []>("op_46345_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_46345_end_0 = const()[name = tensor<string, []>("op_46345_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_46345_end_mask_0 = const()[name = tensor<string, []>("op_46345_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46345_cast = slice_by_index(begin = var_46345_begin_0, end = var_46345_end_0, end_mask = var_46345_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46345_cast")];
            tensor<int32, [4]> var_46349_begin_0 = const()[name = tensor<string, []>("op_46349_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_46349_end_0 = const()[name = tensor<string, []>("op_46349_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_46349_end_mask_0 = const()[name = tensor<string, []>("op_46349_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46349_cast = slice_by_index(begin = var_46349_begin_0, end = var_46349_end_0, end_mask = var_46349_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46349_cast")];
            tensor<int32, [4]> var_46353_begin_0 = const()[name = tensor<string, []>("op_46353_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_46353_end_0 = const()[name = tensor<string, []>("op_46353_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_46353_end_mask_0 = const()[name = tensor<string, []>("op_46353_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46353_cast = slice_by_index(begin = var_46353_begin_0, end = var_46353_end_0, end_mask = var_46353_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46353_cast")];
            tensor<int32, [4]> var_46357_begin_0 = const()[name = tensor<string, []>("op_46357_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_46357_end_0 = const()[name = tensor<string, []>("op_46357_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_46357_end_mask_0 = const()[name = tensor<string, []>("op_46357_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46357_cast = slice_by_index(begin = var_46357_begin_0, end = var_46357_end_0, end_mask = var_46357_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46357_cast")];
            tensor<int32, [4]> var_46361_begin_0 = const()[name = tensor<string, []>("op_46361_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_46361_end_0 = const()[name = tensor<string, []>("op_46361_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_46361_end_mask_0 = const()[name = tensor<string, []>("op_46361_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46361_cast = slice_by_index(begin = var_46361_begin_0, end = var_46361_end_0, end_mask = var_46361_end_mask_0, x = v_205_cast)[name = tensor<string, []>("op_46361_cast")];
            tensor<string, []> var_46365_equation_0 = const()[name = tensor<string, []>("op_46365_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46365_cast = einsum(equation = var_46365_equation_0, values = (var_46207_cast, var_46124_cast))[name = tensor<string, []>("op_46365_cast")];
            tensor<fp16, []> var_46366_to_fp16 = const()[name = tensor<string, []>("op_46366_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3921_cast = mul(x = var_46365_cast, y = var_46366_to_fp16)[name = tensor<string, []>("aw_3921_cast")];
            tensor<string, []> var_46369_equation_0 = const()[name = tensor<string, []>("op_46369_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46369_cast = einsum(equation = var_46369_equation_0, values = (var_46211_cast, var_46128_cast))[name = tensor<string, []>("op_46369_cast")];
            tensor<fp16, []> var_46370_to_fp16 = const()[name = tensor<string, []>("op_46370_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3923_cast = mul(x = var_46369_cast, y = var_46370_to_fp16)[name = tensor<string, []>("aw_3923_cast")];
            tensor<string, []> var_46373_equation_0 = const()[name = tensor<string, []>("op_46373_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46373_cast = einsum(equation = var_46373_equation_0, values = (var_46215_cast, var_46132_cast))[name = tensor<string, []>("op_46373_cast")];
            tensor<fp16, []> var_46374_to_fp16 = const()[name = tensor<string, []>("op_46374_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3925_cast = mul(x = var_46373_cast, y = var_46374_to_fp16)[name = tensor<string, []>("aw_3925_cast")];
            tensor<string, []> var_46377_equation_0 = const()[name = tensor<string, []>("op_46377_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46377_cast = einsum(equation = var_46377_equation_0, values = (var_46219_cast, var_46136_cast))[name = tensor<string, []>("op_46377_cast")];
            tensor<fp16, []> var_46378_to_fp16 = const()[name = tensor<string, []>("op_46378_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3927_cast = mul(x = var_46377_cast, y = var_46378_to_fp16)[name = tensor<string, []>("aw_3927_cast")];
            tensor<string, []> var_46381_equation_0 = const()[name = tensor<string, []>("op_46381_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46381_cast = einsum(equation = var_46381_equation_0, values = (var_46223_cast, var_46140_cast))[name = tensor<string, []>("op_46381_cast")];
            tensor<fp16, []> var_46382_to_fp16 = const()[name = tensor<string, []>("op_46382_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3929_cast = mul(x = var_46381_cast, y = var_46382_to_fp16)[name = tensor<string, []>("aw_3929_cast")];
            tensor<string, []> var_46385_equation_0 = const()[name = tensor<string, []>("op_46385_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46385_cast = einsum(equation = var_46385_equation_0, values = (var_46227_cast, var_46144_cast))[name = tensor<string, []>("op_46385_cast")];
            tensor<fp16, []> var_46386_to_fp16 = const()[name = tensor<string, []>("op_46386_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3931_cast = mul(x = var_46385_cast, y = var_46386_to_fp16)[name = tensor<string, []>("aw_3931_cast")];
            tensor<string, []> var_46389_equation_0 = const()[name = tensor<string, []>("op_46389_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46389_cast = einsum(equation = var_46389_equation_0, values = (var_46231_cast, var_46148_cast))[name = tensor<string, []>("op_46389_cast")];
            tensor<fp16, []> var_46390_to_fp16 = const()[name = tensor<string, []>("op_46390_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3933_cast = mul(x = var_46389_cast, y = var_46390_to_fp16)[name = tensor<string, []>("aw_3933_cast")];
            tensor<string, []> var_46393_equation_0 = const()[name = tensor<string, []>("op_46393_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46393_cast = einsum(equation = var_46393_equation_0, values = (var_46235_cast, var_46152_cast))[name = tensor<string, []>("op_46393_cast")];
            tensor<fp16, []> var_46394_to_fp16 = const()[name = tensor<string, []>("op_46394_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3935_cast = mul(x = var_46393_cast, y = var_46394_to_fp16)[name = tensor<string, []>("aw_3935_cast")];
            tensor<string, []> var_46397_equation_0 = const()[name = tensor<string, []>("op_46397_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46397_cast = einsum(equation = var_46397_equation_0, values = (var_46239_cast, var_46156_cast))[name = tensor<string, []>("op_46397_cast")];
            tensor<fp16, []> var_46398_to_fp16 = const()[name = tensor<string, []>("op_46398_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3937_cast = mul(x = var_46397_cast, y = var_46398_to_fp16)[name = tensor<string, []>("aw_3937_cast")];
            tensor<string, []> var_46401_equation_0 = const()[name = tensor<string, []>("op_46401_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46401_cast = einsum(equation = var_46401_equation_0, values = (var_46243_cast, var_46160_cast))[name = tensor<string, []>("op_46401_cast")];
            tensor<fp16, []> var_46402_to_fp16 = const()[name = tensor<string, []>("op_46402_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3939_cast = mul(x = var_46401_cast, y = var_46402_to_fp16)[name = tensor<string, []>("aw_3939_cast")];
            tensor<string, []> var_46405_equation_0 = const()[name = tensor<string, []>("op_46405_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46405_cast = einsum(equation = var_46405_equation_0, values = (var_46247_cast, var_46164_cast))[name = tensor<string, []>("op_46405_cast")];
            tensor<fp16, []> var_46406_to_fp16 = const()[name = tensor<string, []>("op_46406_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3941_cast = mul(x = var_46405_cast, y = var_46406_to_fp16)[name = tensor<string, []>("aw_3941_cast")];
            tensor<string, []> var_46409_equation_0 = const()[name = tensor<string, []>("op_46409_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46409_cast = einsum(equation = var_46409_equation_0, values = (var_46251_cast, var_46168_cast))[name = tensor<string, []>("op_46409_cast")];
            tensor<fp16, []> var_46410_to_fp16 = const()[name = tensor<string, []>("op_46410_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3943_cast = mul(x = var_46409_cast, y = var_46410_to_fp16)[name = tensor<string, []>("aw_3943_cast")];
            tensor<string, []> var_46413_equation_0 = const()[name = tensor<string, []>("op_46413_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46413_cast = einsum(equation = var_46413_equation_0, values = (var_46255_cast, var_46172_cast))[name = tensor<string, []>("op_46413_cast")];
            tensor<fp16, []> var_46414_to_fp16 = const()[name = tensor<string, []>("op_46414_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3945_cast = mul(x = var_46413_cast, y = var_46414_to_fp16)[name = tensor<string, []>("aw_3945_cast")];
            tensor<string, []> var_46417_equation_0 = const()[name = tensor<string, []>("op_46417_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46417_cast = einsum(equation = var_46417_equation_0, values = (var_46259_cast, var_46176_cast))[name = tensor<string, []>("op_46417_cast")];
            tensor<fp16, []> var_46418_to_fp16 = const()[name = tensor<string, []>("op_46418_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3947_cast = mul(x = var_46417_cast, y = var_46418_to_fp16)[name = tensor<string, []>("aw_3947_cast")];
            tensor<string, []> var_46421_equation_0 = const()[name = tensor<string, []>("op_46421_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46421_cast = einsum(equation = var_46421_equation_0, values = (var_46263_cast, var_46180_cast))[name = tensor<string, []>("op_46421_cast")];
            tensor<fp16, []> var_46422_to_fp16 = const()[name = tensor<string, []>("op_46422_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3949_cast = mul(x = var_46421_cast, y = var_46422_to_fp16)[name = tensor<string, []>("aw_3949_cast")];
            tensor<string, []> var_46425_equation_0 = const()[name = tensor<string, []>("op_46425_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46425_cast = einsum(equation = var_46425_equation_0, values = (var_46267_cast, var_46184_cast))[name = tensor<string, []>("op_46425_cast")];
            tensor<fp16, []> var_46426_to_fp16 = const()[name = tensor<string, []>("op_46426_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3951_cast = mul(x = var_46425_cast, y = var_46426_to_fp16)[name = tensor<string, []>("aw_3951_cast")];
            tensor<string, []> var_46429_equation_0 = const()[name = tensor<string, []>("op_46429_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46429_cast = einsum(equation = var_46429_equation_0, values = (var_46271_cast, var_46188_cast))[name = tensor<string, []>("op_46429_cast")];
            tensor<fp16, []> var_46430_to_fp16 = const()[name = tensor<string, []>("op_46430_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3953_cast = mul(x = var_46429_cast, y = var_46430_to_fp16)[name = tensor<string, []>("aw_3953_cast")];
            tensor<string, []> var_46433_equation_0 = const()[name = tensor<string, []>("op_46433_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46433_cast = einsum(equation = var_46433_equation_0, values = (var_46275_cast, var_46192_cast))[name = tensor<string, []>("op_46433_cast")];
            tensor<fp16, []> var_46434_to_fp16 = const()[name = tensor<string, []>("op_46434_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3955_cast = mul(x = var_46433_cast, y = var_46434_to_fp16)[name = tensor<string, []>("aw_3955_cast")];
            tensor<string, []> var_46437_equation_0 = const()[name = tensor<string, []>("op_46437_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46437_cast = einsum(equation = var_46437_equation_0, values = (var_46279_cast, var_46196_cast))[name = tensor<string, []>("op_46437_cast")];
            tensor<fp16, []> var_46438_to_fp16 = const()[name = tensor<string, []>("op_46438_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3957_cast = mul(x = var_46437_cast, y = var_46438_to_fp16)[name = tensor<string, []>("aw_3957_cast")];
            tensor<string, []> var_46441_equation_0 = const()[name = tensor<string, []>("op_46441_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46441_cast = einsum(equation = var_46441_equation_0, values = (var_46283_cast, var_46200_cast))[name = tensor<string, []>("op_46441_cast")];
            tensor<fp16, []> var_46442_to_fp16 = const()[name = tensor<string, []>("op_46442_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_3959_cast = mul(x = var_46441_cast, y = var_46442_to_fp16)[name = tensor<string, []>("aw_3959_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46444_cast = softmax(axis = var_30385, x = aw_3921_cast)[name = tensor<string, []>("op_46444_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46445_cast = softmax(axis = var_30385, x = aw_3923_cast)[name = tensor<string, []>("op_46445_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46446_cast = softmax(axis = var_30385, x = aw_3925_cast)[name = tensor<string, []>("op_46446_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46447_cast = softmax(axis = var_30385, x = aw_3927_cast)[name = tensor<string, []>("op_46447_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46448_cast = softmax(axis = var_30385, x = aw_3929_cast)[name = tensor<string, []>("op_46448_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46449_cast = softmax(axis = var_30385, x = aw_3931_cast)[name = tensor<string, []>("op_46449_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46450_cast = softmax(axis = var_30385, x = aw_3933_cast)[name = tensor<string, []>("op_46450_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46451_cast = softmax(axis = var_30385, x = aw_3935_cast)[name = tensor<string, []>("op_46451_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46452_cast = softmax(axis = var_30385, x = aw_3937_cast)[name = tensor<string, []>("op_46452_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46453_cast = softmax(axis = var_30385, x = aw_3939_cast)[name = tensor<string, []>("op_46453_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46454_cast = softmax(axis = var_30385, x = aw_3941_cast)[name = tensor<string, []>("op_46454_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46455_cast = softmax(axis = var_30385, x = aw_3943_cast)[name = tensor<string, []>("op_46455_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46456_cast = softmax(axis = var_30385, x = aw_3945_cast)[name = tensor<string, []>("op_46456_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46457_cast = softmax(axis = var_30385, x = aw_3947_cast)[name = tensor<string, []>("op_46457_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46458_cast = softmax(axis = var_30385, x = aw_3949_cast)[name = tensor<string, []>("op_46458_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46459_cast = softmax(axis = var_30385, x = aw_3951_cast)[name = tensor<string, []>("op_46459_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46460_cast = softmax(axis = var_30385, x = aw_3953_cast)[name = tensor<string, []>("op_46460_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46461_cast = softmax(axis = var_30385, x = aw_3955_cast)[name = tensor<string, []>("op_46461_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46462_cast = softmax(axis = var_30385, x = aw_3957_cast)[name = tensor<string, []>("op_46462_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_46463_cast = softmax(axis = var_30385, x = aw_3959_cast)[name = tensor<string, []>("op_46463_cast")];
            tensor<string, []> var_46465_equation_0 = const()[name = tensor<string, []>("op_46465_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46465_cast = einsum(equation = var_46465_equation_0, values = (var_46285_cast, var_46444_cast))[name = tensor<string, []>("op_46465_cast")];
            tensor<string, []> var_46467_equation_0 = const()[name = tensor<string, []>("op_46467_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46467_cast = einsum(equation = var_46467_equation_0, values = (var_46289_cast, var_46445_cast))[name = tensor<string, []>("op_46467_cast")];
            tensor<string, []> var_46469_equation_0 = const()[name = tensor<string, []>("op_46469_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46469_cast = einsum(equation = var_46469_equation_0, values = (var_46293_cast, var_46446_cast))[name = tensor<string, []>("op_46469_cast")];
            tensor<string, []> var_46471_equation_0 = const()[name = tensor<string, []>("op_46471_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46471_cast = einsum(equation = var_46471_equation_0, values = (var_46297_cast, var_46447_cast))[name = tensor<string, []>("op_46471_cast")];
            tensor<string, []> var_46473_equation_0 = const()[name = tensor<string, []>("op_46473_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46473_cast = einsum(equation = var_46473_equation_0, values = (var_46301_cast, var_46448_cast))[name = tensor<string, []>("op_46473_cast")];
            tensor<string, []> var_46475_equation_0 = const()[name = tensor<string, []>("op_46475_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46475_cast = einsum(equation = var_46475_equation_0, values = (var_46305_cast, var_46449_cast))[name = tensor<string, []>("op_46475_cast")];
            tensor<string, []> var_46477_equation_0 = const()[name = tensor<string, []>("op_46477_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46477_cast = einsum(equation = var_46477_equation_0, values = (var_46309_cast, var_46450_cast))[name = tensor<string, []>("op_46477_cast")];
            tensor<string, []> var_46479_equation_0 = const()[name = tensor<string, []>("op_46479_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46479_cast = einsum(equation = var_46479_equation_0, values = (var_46313_cast, var_46451_cast))[name = tensor<string, []>("op_46479_cast")];
            tensor<string, []> var_46481_equation_0 = const()[name = tensor<string, []>("op_46481_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46481_cast = einsum(equation = var_46481_equation_0, values = (var_46317_cast, var_46452_cast))[name = tensor<string, []>("op_46481_cast")];
            tensor<string, []> var_46483_equation_0 = const()[name = tensor<string, []>("op_46483_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46483_cast = einsum(equation = var_46483_equation_0, values = (var_46321_cast, var_46453_cast))[name = tensor<string, []>("op_46483_cast")];
            tensor<string, []> var_46485_equation_0 = const()[name = tensor<string, []>("op_46485_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46485_cast = einsum(equation = var_46485_equation_0, values = (var_46325_cast, var_46454_cast))[name = tensor<string, []>("op_46485_cast")];
            tensor<string, []> var_46487_equation_0 = const()[name = tensor<string, []>("op_46487_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46487_cast = einsum(equation = var_46487_equation_0, values = (var_46329_cast, var_46455_cast))[name = tensor<string, []>("op_46487_cast")];
            tensor<string, []> var_46489_equation_0 = const()[name = tensor<string, []>("op_46489_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46489_cast = einsum(equation = var_46489_equation_0, values = (var_46333_cast, var_46456_cast))[name = tensor<string, []>("op_46489_cast")];
            tensor<string, []> var_46491_equation_0 = const()[name = tensor<string, []>("op_46491_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46491_cast = einsum(equation = var_46491_equation_0, values = (var_46337_cast, var_46457_cast))[name = tensor<string, []>("op_46491_cast")];
            tensor<string, []> var_46493_equation_0 = const()[name = tensor<string, []>("op_46493_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46493_cast = einsum(equation = var_46493_equation_0, values = (var_46341_cast, var_46458_cast))[name = tensor<string, []>("op_46493_cast")];
            tensor<string, []> var_46495_equation_0 = const()[name = tensor<string, []>("op_46495_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46495_cast = einsum(equation = var_46495_equation_0, values = (var_46345_cast, var_46459_cast))[name = tensor<string, []>("op_46495_cast")];
            tensor<string, []> var_46497_equation_0 = const()[name = tensor<string, []>("op_46497_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46497_cast = einsum(equation = var_46497_equation_0, values = (var_46349_cast, var_46460_cast))[name = tensor<string, []>("op_46497_cast")];
            tensor<string, []> var_46499_equation_0 = const()[name = tensor<string, []>("op_46499_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46499_cast = einsum(equation = var_46499_equation_0, values = (var_46353_cast, var_46461_cast))[name = tensor<string, []>("op_46499_cast")];
            tensor<string, []> var_46501_equation_0 = const()[name = tensor<string, []>("op_46501_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46501_cast = einsum(equation = var_46501_equation_0, values = (var_46357_cast, var_46462_cast))[name = tensor<string, []>("op_46501_cast")];
            tensor<string, []> var_46503_equation_0 = const()[name = tensor<string, []>("op_46503_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46503_cast = einsum(equation = var_46503_equation_0, values = (var_46361_cast, var_46463_cast))[name = tensor<string, []>("op_46503_cast")];
            tensor<bool, []> input_595_interleave_0 = const()[name = tensor<string, []>("input_595_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_595_cast = concat(axis = var_30385, interleave = input_595_interleave_0, values = (var_46465_cast, var_46467_cast, var_46469_cast, var_46471_cast, var_46473_cast, var_46475_cast, var_46477_cast, var_46479_cast, var_46481_cast, var_46483_cast, var_46485_cast, var_46487_cast, var_46489_cast, var_46491_cast, var_46493_cast, var_46495_cast, var_46497_cast, var_46499_cast, var_46501_cast, var_46503_cast))[name = tensor<string, []>("input_595_cast")];
            tensor<int32, [2]> var_46509 = const()[name = tensor<string, []>("op_46509"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_46511 = const()[name = tensor<string, []>("op_46511"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_46513_pad_type_0 = const()[name = tensor<string, []>("op_46513_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_46513_pad_0 = const()[name = tensor<string, []>("op_46513_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3884872128)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3888148992)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_46513_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_out_0_bias_to_fp16, dilations = var_46511, groups = var_30385, pad = var_46513_pad_0, pad_type = var_46513_pad_type_0, strides = var_46509, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn1_to_out_0_weight_to_fp16, x = input_595_cast)[name = tensor<string, []>("op_46513_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_309_cast = add(x = var_46513_cast, y = inputs_307_cast)[name = tensor<string, []>("inputs_309_cast")];
            tensor<int32, [1]> var_46517 = const()[name = tensor<string, []>("op_46517"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_309_cast = reduce_mean(axes = var_46517, keep_dims = var_30380, x = inputs_309_cast)[name = tensor<string, []>("channels_mean_309_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_309_cast = sub(x = inputs_309_cast, y = channels_mean_309_cast)[name = tensor<string, []>("zero_mean_309_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_309_cast = mul(x = zero_mean_309_cast, y = zero_mean_309_cast)[name = tensor<string, []>("zero_mean_sq_309_cast")];
            tensor<int32, [1]> var_46521 = const()[name = tensor<string, []>("op_46521"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_46522_cast = reduce_mean(axes = var_46521, keep_dims = var_30380, x = zero_mean_sq_309_cast)[name = tensor<string, []>("op_46522_cast")];
            tensor<fp16, []> var_46523_to_fp16 = const()[name = tensor<string, []>("op_46523_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_46524_cast = add(x = var_46522_cast, y = var_46523_to_fp16)[name = tensor<string, []>("op_46524_cast")];
            tensor<fp16, []> denom_309_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_309_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_309_cast = rsqrt(epsilon = denom_309_epsilon_0_to_fp16, x = var_46524_cast)[name = tensor<string, []>("denom_309_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_309_cast = mul(x = zero_mean_309_cast, y = denom_309_cast)[name = tensor<string, []>("out_309_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_46528_to_fp16 = const()[name = tensor<string, []>("op_46528_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3888151616)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_46529_cast = add(x = out_309_cast, y = var_46528_to_fp16)[name = tensor<string, []>("op_46529_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_46531_to_fp16 = const()[name = tensor<string, []>("op_46531_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3888154240)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_405_cast = mul(x = var_46529_cast, y = var_46531_to_fp16)[name = tensor<string, []>("hidden_states_405_cast")];
            tensor<int32, [2]> var_46538 = const()[name = tensor<string, []>("op_46538"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_46540 = const()[name = tensor<string, []>("op_46540"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_207_pad_type_0 = const()[name = tensor<string, []>("q_207_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_207_pad_0 = const()[name = tensor<string, []>("q_207_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3888156864)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_207_cast = conv(dilations = var_46540, groups = var_30385, pad = q_207_pad_0, pad_type = q_207_pad_type_0, strides = var_46538, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_q_weight_to_fp16, x = hidden_states_405_cast)[name = tensor<string, []>("q_207_cast")];
            tensor<int32, [2]> var_46544 = const()[name = tensor<string, []>("op_46544"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_46546 = const()[name = tensor<string, []>("op_46546"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_413_pad_type_0 = const()[name = tensor<string, []>("k_413_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_413_pad_0 = const()[name = tensor<string, []>("k_413_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3891433728)))];
            tensor<fp16, [2, 1280, 1, 77]> k_413_cast = conv(dilations = var_46546, groups = var_30385, pad = k_413_pad_0, pad_type = k_413_pad_type_0, strides = var_46544, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_413_cast")];
            tensor<int32, [2]> var_46550 = const()[name = tensor<string, []>("op_46550"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_46552 = const()[name = tensor<string, []>("op_46552"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_207_pad_type_0 = const()[name = tensor<string, []>("v_207_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_207_pad_0 = const()[name = tensor<string, []>("v_207_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3896676672)))];
            tensor<fp16, [2, 1280, 1, 77]> v_207_cast = conv(dilations = var_46552, groups = var_30385, pad = v_207_pad_0, pad_type = v_207_pad_type_0, strides = var_46550, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_207_cast")];
            tensor<int32, [4]> var_46556_begin_0 = const()[name = tensor<string, []>("op_46556_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_46556_end_0 = const()[name = tensor<string, []>("op_46556_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_46556_end_mask_0 = const()[name = tensor<string, []>("op_46556_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46556_cast = slice_by_index(begin = var_46556_begin_0, end = var_46556_end_0, end_mask = var_46556_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46556_cast")];
            tensor<int32, [4]> var_46560_begin_0 = const()[name = tensor<string, []>("op_46560_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_46560_end_0 = const()[name = tensor<string, []>("op_46560_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_46560_end_mask_0 = const()[name = tensor<string, []>("op_46560_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46560_cast = slice_by_index(begin = var_46560_begin_0, end = var_46560_end_0, end_mask = var_46560_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46560_cast")];
            tensor<int32, [4]> var_46564_begin_0 = const()[name = tensor<string, []>("op_46564_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_46564_end_0 = const()[name = tensor<string, []>("op_46564_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_46564_end_mask_0 = const()[name = tensor<string, []>("op_46564_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46564_cast = slice_by_index(begin = var_46564_begin_0, end = var_46564_end_0, end_mask = var_46564_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46564_cast")];
            tensor<int32, [4]> var_46568_begin_0 = const()[name = tensor<string, []>("op_46568_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_46568_end_0 = const()[name = tensor<string, []>("op_46568_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_46568_end_mask_0 = const()[name = tensor<string, []>("op_46568_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46568_cast = slice_by_index(begin = var_46568_begin_0, end = var_46568_end_0, end_mask = var_46568_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46568_cast")];
            tensor<int32, [4]> var_46572_begin_0 = const()[name = tensor<string, []>("op_46572_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_46572_end_0 = const()[name = tensor<string, []>("op_46572_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_46572_end_mask_0 = const()[name = tensor<string, []>("op_46572_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46572_cast = slice_by_index(begin = var_46572_begin_0, end = var_46572_end_0, end_mask = var_46572_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46572_cast")];
            tensor<int32, [4]> var_46576_begin_0 = const()[name = tensor<string, []>("op_46576_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_46576_end_0 = const()[name = tensor<string, []>("op_46576_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_46576_end_mask_0 = const()[name = tensor<string, []>("op_46576_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46576_cast = slice_by_index(begin = var_46576_begin_0, end = var_46576_end_0, end_mask = var_46576_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46576_cast")];
            tensor<int32, [4]> var_46580_begin_0 = const()[name = tensor<string, []>("op_46580_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_46580_end_0 = const()[name = tensor<string, []>("op_46580_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_46580_end_mask_0 = const()[name = tensor<string, []>("op_46580_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46580_cast = slice_by_index(begin = var_46580_begin_0, end = var_46580_end_0, end_mask = var_46580_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46580_cast")];
            tensor<int32, [4]> var_46584_begin_0 = const()[name = tensor<string, []>("op_46584_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_46584_end_0 = const()[name = tensor<string, []>("op_46584_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_46584_end_mask_0 = const()[name = tensor<string, []>("op_46584_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46584_cast = slice_by_index(begin = var_46584_begin_0, end = var_46584_end_0, end_mask = var_46584_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46584_cast")];
            tensor<int32, [4]> var_46588_begin_0 = const()[name = tensor<string, []>("op_46588_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_46588_end_0 = const()[name = tensor<string, []>("op_46588_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_46588_end_mask_0 = const()[name = tensor<string, []>("op_46588_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46588_cast = slice_by_index(begin = var_46588_begin_0, end = var_46588_end_0, end_mask = var_46588_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46588_cast")];
            tensor<int32, [4]> var_46592_begin_0 = const()[name = tensor<string, []>("op_46592_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_46592_end_0 = const()[name = tensor<string, []>("op_46592_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_46592_end_mask_0 = const()[name = tensor<string, []>("op_46592_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46592_cast = slice_by_index(begin = var_46592_begin_0, end = var_46592_end_0, end_mask = var_46592_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46592_cast")];
            tensor<int32, [4]> var_46596_begin_0 = const()[name = tensor<string, []>("op_46596_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_46596_end_0 = const()[name = tensor<string, []>("op_46596_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_46596_end_mask_0 = const()[name = tensor<string, []>("op_46596_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46596_cast = slice_by_index(begin = var_46596_begin_0, end = var_46596_end_0, end_mask = var_46596_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46596_cast")];
            tensor<int32, [4]> var_46600_begin_0 = const()[name = tensor<string, []>("op_46600_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_46600_end_0 = const()[name = tensor<string, []>("op_46600_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_46600_end_mask_0 = const()[name = tensor<string, []>("op_46600_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46600_cast = slice_by_index(begin = var_46600_begin_0, end = var_46600_end_0, end_mask = var_46600_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46600_cast")];
            tensor<int32, [4]> var_46604_begin_0 = const()[name = tensor<string, []>("op_46604_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_46604_end_0 = const()[name = tensor<string, []>("op_46604_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_46604_end_mask_0 = const()[name = tensor<string, []>("op_46604_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46604_cast = slice_by_index(begin = var_46604_begin_0, end = var_46604_end_0, end_mask = var_46604_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46604_cast")];
            tensor<int32, [4]> var_46608_begin_0 = const()[name = tensor<string, []>("op_46608_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_46608_end_0 = const()[name = tensor<string, []>("op_46608_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_46608_end_mask_0 = const()[name = tensor<string, []>("op_46608_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46608_cast = slice_by_index(begin = var_46608_begin_0, end = var_46608_end_0, end_mask = var_46608_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46608_cast")];
            tensor<int32, [4]> var_46612_begin_0 = const()[name = tensor<string, []>("op_46612_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_46612_end_0 = const()[name = tensor<string, []>("op_46612_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_46612_end_mask_0 = const()[name = tensor<string, []>("op_46612_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46612_cast = slice_by_index(begin = var_46612_begin_0, end = var_46612_end_0, end_mask = var_46612_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46612_cast")];
            tensor<int32, [4]> var_46616_begin_0 = const()[name = tensor<string, []>("op_46616_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_46616_end_0 = const()[name = tensor<string, []>("op_46616_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_46616_end_mask_0 = const()[name = tensor<string, []>("op_46616_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46616_cast = slice_by_index(begin = var_46616_begin_0, end = var_46616_end_0, end_mask = var_46616_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46616_cast")];
            tensor<int32, [4]> var_46620_begin_0 = const()[name = tensor<string, []>("op_46620_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_46620_end_0 = const()[name = tensor<string, []>("op_46620_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_46620_end_mask_0 = const()[name = tensor<string, []>("op_46620_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46620_cast = slice_by_index(begin = var_46620_begin_0, end = var_46620_end_0, end_mask = var_46620_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46620_cast")];
            tensor<int32, [4]> var_46624_begin_0 = const()[name = tensor<string, []>("op_46624_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_46624_end_0 = const()[name = tensor<string, []>("op_46624_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_46624_end_mask_0 = const()[name = tensor<string, []>("op_46624_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46624_cast = slice_by_index(begin = var_46624_begin_0, end = var_46624_end_0, end_mask = var_46624_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46624_cast")];
            tensor<int32, [4]> var_46628_begin_0 = const()[name = tensor<string, []>("op_46628_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_46628_end_0 = const()[name = tensor<string, []>("op_46628_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_46628_end_mask_0 = const()[name = tensor<string, []>("op_46628_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46628_cast = slice_by_index(begin = var_46628_begin_0, end = var_46628_end_0, end_mask = var_46628_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46628_cast")];
            tensor<int32, [4]> var_46632_begin_0 = const()[name = tensor<string, []>("op_46632_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_46632_end_0 = const()[name = tensor<string, []>("op_46632_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_46632_end_mask_0 = const()[name = tensor<string, []>("op_46632_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_46632_cast = slice_by_index(begin = var_46632_begin_0, end = var_46632_end_0, end_mask = var_46632_end_mask_0, x = q_207_cast)[name = tensor<string, []>("op_46632_cast")];
            tensor<int32, [4]> k_415_perm_0 = const()[name = tensor<string, []>("k_415_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_46639_begin_0 = const()[name = tensor<string, []>("op_46639_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_46639_end_0 = const()[name = tensor<string, []>("op_46639_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_46639_end_mask_0 = const()[name = tensor<string, []>("op_46639_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_36 = transpose(perm = k_415_perm_0, x = k_413_cast)[name = tensor<string, []>("transpose_36")];
            tensor<fp16, [2, 77, 1, 64]> var_46639_cast = slice_by_index(begin = var_46639_begin_0, end = var_46639_end_0, end_mask = var_46639_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46639_cast")];
            tensor<int32, [4]> var_46643_begin_0 = const()[name = tensor<string, []>("op_46643_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_46643_end_0 = const()[name = tensor<string, []>("op_46643_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_46643_end_mask_0 = const()[name = tensor<string, []>("op_46643_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46643_cast = slice_by_index(begin = var_46643_begin_0, end = var_46643_end_0, end_mask = var_46643_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46643_cast")];
            tensor<int32, [4]> var_46647_begin_0 = const()[name = tensor<string, []>("op_46647_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_46647_end_0 = const()[name = tensor<string, []>("op_46647_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_46647_end_mask_0 = const()[name = tensor<string, []>("op_46647_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46647_cast = slice_by_index(begin = var_46647_begin_0, end = var_46647_end_0, end_mask = var_46647_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46647_cast")];
            tensor<int32, [4]> var_46651_begin_0 = const()[name = tensor<string, []>("op_46651_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_46651_end_0 = const()[name = tensor<string, []>("op_46651_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_46651_end_mask_0 = const()[name = tensor<string, []>("op_46651_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46651_cast = slice_by_index(begin = var_46651_begin_0, end = var_46651_end_0, end_mask = var_46651_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46651_cast")];
            tensor<int32, [4]> var_46655_begin_0 = const()[name = tensor<string, []>("op_46655_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_46655_end_0 = const()[name = tensor<string, []>("op_46655_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_46655_end_mask_0 = const()[name = tensor<string, []>("op_46655_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46655_cast = slice_by_index(begin = var_46655_begin_0, end = var_46655_end_0, end_mask = var_46655_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46655_cast")];
            tensor<int32, [4]> var_46659_begin_0 = const()[name = tensor<string, []>("op_46659_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_46659_end_0 = const()[name = tensor<string, []>("op_46659_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_46659_end_mask_0 = const()[name = tensor<string, []>("op_46659_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46659_cast = slice_by_index(begin = var_46659_begin_0, end = var_46659_end_0, end_mask = var_46659_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46659_cast")];
            tensor<int32, [4]> var_46663_begin_0 = const()[name = tensor<string, []>("op_46663_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_46663_end_0 = const()[name = tensor<string, []>("op_46663_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_46663_end_mask_0 = const()[name = tensor<string, []>("op_46663_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46663_cast = slice_by_index(begin = var_46663_begin_0, end = var_46663_end_0, end_mask = var_46663_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46663_cast")];
            tensor<int32, [4]> var_46667_begin_0 = const()[name = tensor<string, []>("op_46667_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_46667_end_0 = const()[name = tensor<string, []>("op_46667_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_46667_end_mask_0 = const()[name = tensor<string, []>("op_46667_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46667_cast = slice_by_index(begin = var_46667_begin_0, end = var_46667_end_0, end_mask = var_46667_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46667_cast")];
            tensor<int32, [4]> var_46671_begin_0 = const()[name = tensor<string, []>("op_46671_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_46671_end_0 = const()[name = tensor<string, []>("op_46671_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_46671_end_mask_0 = const()[name = tensor<string, []>("op_46671_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46671_cast = slice_by_index(begin = var_46671_begin_0, end = var_46671_end_0, end_mask = var_46671_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46671_cast")];
            tensor<int32, [4]> var_46675_begin_0 = const()[name = tensor<string, []>("op_46675_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_46675_end_0 = const()[name = tensor<string, []>("op_46675_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_46675_end_mask_0 = const()[name = tensor<string, []>("op_46675_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46675_cast = slice_by_index(begin = var_46675_begin_0, end = var_46675_end_0, end_mask = var_46675_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46675_cast")];
            tensor<int32, [4]> var_46679_begin_0 = const()[name = tensor<string, []>("op_46679_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_46679_end_0 = const()[name = tensor<string, []>("op_46679_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_46679_end_mask_0 = const()[name = tensor<string, []>("op_46679_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46679_cast = slice_by_index(begin = var_46679_begin_0, end = var_46679_end_0, end_mask = var_46679_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46679_cast")];
            tensor<int32, [4]> var_46683_begin_0 = const()[name = tensor<string, []>("op_46683_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_46683_end_0 = const()[name = tensor<string, []>("op_46683_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_46683_end_mask_0 = const()[name = tensor<string, []>("op_46683_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46683_cast = slice_by_index(begin = var_46683_begin_0, end = var_46683_end_0, end_mask = var_46683_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46683_cast")];
            tensor<int32, [4]> var_46687_begin_0 = const()[name = tensor<string, []>("op_46687_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_46687_end_0 = const()[name = tensor<string, []>("op_46687_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_46687_end_mask_0 = const()[name = tensor<string, []>("op_46687_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46687_cast = slice_by_index(begin = var_46687_begin_0, end = var_46687_end_0, end_mask = var_46687_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46687_cast")];
            tensor<int32, [4]> var_46691_begin_0 = const()[name = tensor<string, []>("op_46691_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_46691_end_0 = const()[name = tensor<string, []>("op_46691_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_46691_end_mask_0 = const()[name = tensor<string, []>("op_46691_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46691_cast = slice_by_index(begin = var_46691_begin_0, end = var_46691_end_0, end_mask = var_46691_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46691_cast")];
            tensor<int32, [4]> var_46695_begin_0 = const()[name = tensor<string, []>("op_46695_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_46695_end_0 = const()[name = tensor<string, []>("op_46695_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_46695_end_mask_0 = const()[name = tensor<string, []>("op_46695_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46695_cast = slice_by_index(begin = var_46695_begin_0, end = var_46695_end_0, end_mask = var_46695_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46695_cast")];
            tensor<int32, [4]> var_46699_begin_0 = const()[name = tensor<string, []>("op_46699_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_46699_end_0 = const()[name = tensor<string, []>("op_46699_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_46699_end_mask_0 = const()[name = tensor<string, []>("op_46699_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46699_cast = slice_by_index(begin = var_46699_begin_0, end = var_46699_end_0, end_mask = var_46699_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46699_cast")];
            tensor<int32, [4]> var_46703_begin_0 = const()[name = tensor<string, []>("op_46703_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_46703_end_0 = const()[name = tensor<string, []>("op_46703_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_46703_end_mask_0 = const()[name = tensor<string, []>("op_46703_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46703_cast = slice_by_index(begin = var_46703_begin_0, end = var_46703_end_0, end_mask = var_46703_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46703_cast")];
            tensor<int32, [4]> var_46707_begin_0 = const()[name = tensor<string, []>("op_46707_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_46707_end_0 = const()[name = tensor<string, []>("op_46707_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_46707_end_mask_0 = const()[name = tensor<string, []>("op_46707_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46707_cast = slice_by_index(begin = var_46707_begin_0, end = var_46707_end_0, end_mask = var_46707_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46707_cast")];
            tensor<int32, [4]> var_46711_begin_0 = const()[name = tensor<string, []>("op_46711_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_46711_end_0 = const()[name = tensor<string, []>("op_46711_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_46711_end_mask_0 = const()[name = tensor<string, []>("op_46711_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46711_cast = slice_by_index(begin = var_46711_begin_0, end = var_46711_end_0, end_mask = var_46711_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46711_cast")];
            tensor<int32, [4]> var_46715_begin_0 = const()[name = tensor<string, []>("op_46715_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_46715_end_0 = const()[name = tensor<string, []>("op_46715_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_46715_end_mask_0 = const()[name = tensor<string, []>("op_46715_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_46715_cast = slice_by_index(begin = var_46715_begin_0, end = var_46715_end_0, end_mask = var_46715_end_mask_0, x = transpose_36)[name = tensor<string, []>("op_46715_cast")];
            tensor<int32, [4]> var_46717_begin_0 = const()[name = tensor<string, []>("op_46717_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_46717_end_0 = const()[name = tensor<string, []>("op_46717_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_46717_end_mask_0 = const()[name = tensor<string, []>("op_46717_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46717_cast = slice_by_index(begin = var_46717_begin_0, end = var_46717_end_0, end_mask = var_46717_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46717_cast")];
            tensor<int32, [4]> var_46721_begin_0 = const()[name = tensor<string, []>("op_46721_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_46721_end_0 = const()[name = tensor<string, []>("op_46721_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_46721_end_mask_0 = const()[name = tensor<string, []>("op_46721_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46721_cast = slice_by_index(begin = var_46721_begin_0, end = var_46721_end_0, end_mask = var_46721_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46721_cast")];
            tensor<int32, [4]> var_46725_begin_0 = const()[name = tensor<string, []>("op_46725_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_46725_end_0 = const()[name = tensor<string, []>("op_46725_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_46725_end_mask_0 = const()[name = tensor<string, []>("op_46725_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46725_cast = slice_by_index(begin = var_46725_begin_0, end = var_46725_end_0, end_mask = var_46725_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46725_cast")];
            tensor<int32, [4]> var_46729_begin_0 = const()[name = tensor<string, []>("op_46729_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_46729_end_0 = const()[name = tensor<string, []>("op_46729_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_46729_end_mask_0 = const()[name = tensor<string, []>("op_46729_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46729_cast = slice_by_index(begin = var_46729_begin_0, end = var_46729_end_0, end_mask = var_46729_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46729_cast")];
            tensor<int32, [4]> var_46733_begin_0 = const()[name = tensor<string, []>("op_46733_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_46733_end_0 = const()[name = tensor<string, []>("op_46733_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_46733_end_mask_0 = const()[name = tensor<string, []>("op_46733_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46733_cast = slice_by_index(begin = var_46733_begin_0, end = var_46733_end_0, end_mask = var_46733_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46733_cast")];
            tensor<int32, [4]> var_46737_begin_0 = const()[name = tensor<string, []>("op_46737_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_46737_end_0 = const()[name = tensor<string, []>("op_46737_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_46737_end_mask_0 = const()[name = tensor<string, []>("op_46737_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46737_cast = slice_by_index(begin = var_46737_begin_0, end = var_46737_end_0, end_mask = var_46737_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46737_cast")];
            tensor<int32, [4]> var_46741_begin_0 = const()[name = tensor<string, []>("op_46741_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_46741_end_0 = const()[name = tensor<string, []>("op_46741_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_46741_end_mask_0 = const()[name = tensor<string, []>("op_46741_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46741_cast = slice_by_index(begin = var_46741_begin_0, end = var_46741_end_0, end_mask = var_46741_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46741_cast")];
            tensor<int32, [4]> var_46745_begin_0 = const()[name = tensor<string, []>("op_46745_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_46745_end_0 = const()[name = tensor<string, []>("op_46745_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_46745_end_mask_0 = const()[name = tensor<string, []>("op_46745_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46745_cast = slice_by_index(begin = var_46745_begin_0, end = var_46745_end_0, end_mask = var_46745_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46745_cast")];
            tensor<int32, [4]> var_46749_begin_0 = const()[name = tensor<string, []>("op_46749_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_46749_end_0 = const()[name = tensor<string, []>("op_46749_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_46749_end_mask_0 = const()[name = tensor<string, []>("op_46749_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46749_cast = slice_by_index(begin = var_46749_begin_0, end = var_46749_end_0, end_mask = var_46749_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46749_cast")];
            tensor<int32, [4]> var_46753_begin_0 = const()[name = tensor<string, []>("op_46753_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_46753_end_0 = const()[name = tensor<string, []>("op_46753_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_46753_end_mask_0 = const()[name = tensor<string, []>("op_46753_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46753_cast = slice_by_index(begin = var_46753_begin_0, end = var_46753_end_0, end_mask = var_46753_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46753_cast")];
            tensor<int32, [4]> var_46757_begin_0 = const()[name = tensor<string, []>("op_46757_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_46757_end_0 = const()[name = tensor<string, []>("op_46757_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_46757_end_mask_0 = const()[name = tensor<string, []>("op_46757_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46757_cast = slice_by_index(begin = var_46757_begin_0, end = var_46757_end_0, end_mask = var_46757_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46757_cast")];
            tensor<int32, [4]> var_46761_begin_0 = const()[name = tensor<string, []>("op_46761_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_46761_end_0 = const()[name = tensor<string, []>("op_46761_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_46761_end_mask_0 = const()[name = tensor<string, []>("op_46761_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46761_cast = slice_by_index(begin = var_46761_begin_0, end = var_46761_end_0, end_mask = var_46761_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46761_cast")];
            tensor<int32, [4]> var_46765_begin_0 = const()[name = tensor<string, []>("op_46765_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_46765_end_0 = const()[name = tensor<string, []>("op_46765_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_46765_end_mask_0 = const()[name = tensor<string, []>("op_46765_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46765_cast = slice_by_index(begin = var_46765_begin_0, end = var_46765_end_0, end_mask = var_46765_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46765_cast")];
            tensor<int32, [4]> var_46769_begin_0 = const()[name = tensor<string, []>("op_46769_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_46769_end_0 = const()[name = tensor<string, []>("op_46769_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_46769_end_mask_0 = const()[name = tensor<string, []>("op_46769_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46769_cast = slice_by_index(begin = var_46769_begin_0, end = var_46769_end_0, end_mask = var_46769_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46769_cast")];
            tensor<int32, [4]> var_46773_begin_0 = const()[name = tensor<string, []>("op_46773_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_46773_end_0 = const()[name = tensor<string, []>("op_46773_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_46773_end_mask_0 = const()[name = tensor<string, []>("op_46773_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46773_cast = slice_by_index(begin = var_46773_begin_0, end = var_46773_end_0, end_mask = var_46773_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46773_cast")];
            tensor<int32, [4]> var_46777_begin_0 = const()[name = tensor<string, []>("op_46777_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_46777_end_0 = const()[name = tensor<string, []>("op_46777_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_46777_end_mask_0 = const()[name = tensor<string, []>("op_46777_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46777_cast = slice_by_index(begin = var_46777_begin_0, end = var_46777_end_0, end_mask = var_46777_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46777_cast")];
            tensor<int32, [4]> var_46781_begin_0 = const()[name = tensor<string, []>("op_46781_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_46781_end_0 = const()[name = tensor<string, []>("op_46781_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_46781_end_mask_0 = const()[name = tensor<string, []>("op_46781_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46781_cast = slice_by_index(begin = var_46781_begin_0, end = var_46781_end_0, end_mask = var_46781_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46781_cast")];
            tensor<int32, [4]> var_46785_begin_0 = const()[name = tensor<string, []>("op_46785_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_46785_end_0 = const()[name = tensor<string, []>("op_46785_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_46785_end_mask_0 = const()[name = tensor<string, []>("op_46785_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46785_cast = slice_by_index(begin = var_46785_begin_0, end = var_46785_end_0, end_mask = var_46785_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46785_cast")];
            tensor<int32, [4]> var_46789_begin_0 = const()[name = tensor<string, []>("op_46789_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_46789_end_0 = const()[name = tensor<string, []>("op_46789_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_46789_end_mask_0 = const()[name = tensor<string, []>("op_46789_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46789_cast = slice_by_index(begin = var_46789_begin_0, end = var_46789_end_0, end_mask = var_46789_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46789_cast")];
            tensor<int32, [4]> var_46793_begin_0 = const()[name = tensor<string, []>("op_46793_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_46793_end_0 = const()[name = tensor<string, []>("op_46793_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_46793_end_mask_0 = const()[name = tensor<string, []>("op_46793_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_46793_cast = slice_by_index(begin = var_46793_begin_0, end = var_46793_end_0, end_mask = var_46793_end_mask_0, x = v_207_cast)[name = tensor<string, []>("op_46793_cast")];
            tensor<string, []> var_46797_equation_0 = const()[name = tensor<string, []>("op_46797_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46797_cast = einsum(equation = var_46797_equation_0, values = (var_46639_cast, var_46556_cast))[name = tensor<string, []>("op_46797_cast")];
            tensor<fp16, []> var_46798_to_fp16 = const()[name = tensor<string, []>("op_46798_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3961_cast = mul(x = var_46797_cast, y = var_46798_to_fp16)[name = tensor<string, []>("aw_3961_cast")];
            tensor<string, []> var_46801_equation_0 = const()[name = tensor<string, []>("op_46801_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46801_cast = einsum(equation = var_46801_equation_0, values = (var_46643_cast, var_46560_cast))[name = tensor<string, []>("op_46801_cast")];
            tensor<fp16, []> var_46802_to_fp16 = const()[name = tensor<string, []>("op_46802_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3963_cast = mul(x = var_46801_cast, y = var_46802_to_fp16)[name = tensor<string, []>("aw_3963_cast")];
            tensor<string, []> var_46805_equation_0 = const()[name = tensor<string, []>("op_46805_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46805_cast = einsum(equation = var_46805_equation_0, values = (var_46647_cast, var_46564_cast))[name = tensor<string, []>("op_46805_cast")];
            tensor<fp16, []> var_46806_to_fp16 = const()[name = tensor<string, []>("op_46806_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3965_cast = mul(x = var_46805_cast, y = var_46806_to_fp16)[name = tensor<string, []>("aw_3965_cast")];
            tensor<string, []> var_46809_equation_0 = const()[name = tensor<string, []>("op_46809_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46809_cast = einsum(equation = var_46809_equation_0, values = (var_46651_cast, var_46568_cast))[name = tensor<string, []>("op_46809_cast")];
            tensor<fp16, []> var_46810_to_fp16 = const()[name = tensor<string, []>("op_46810_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3967_cast = mul(x = var_46809_cast, y = var_46810_to_fp16)[name = tensor<string, []>("aw_3967_cast")];
            tensor<string, []> var_46813_equation_0 = const()[name = tensor<string, []>("op_46813_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46813_cast = einsum(equation = var_46813_equation_0, values = (var_46655_cast, var_46572_cast))[name = tensor<string, []>("op_46813_cast")];
            tensor<fp16, []> var_46814_to_fp16 = const()[name = tensor<string, []>("op_46814_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3969_cast = mul(x = var_46813_cast, y = var_46814_to_fp16)[name = tensor<string, []>("aw_3969_cast")];
            tensor<string, []> var_46817_equation_0 = const()[name = tensor<string, []>("op_46817_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46817_cast = einsum(equation = var_46817_equation_0, values = (var_46659_cast, var_46576_cast))[name = tensor<string, []>("op_46817_cast")];
            tensor<fp16, []> var_46818_to_fp16 = const()[name = tensor<string, []>("op_46818_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3971_cast = mul(x = var_46817_cast, y = var_46818_to_fp16)[name = tensor<string, []>("aw_3971_cast")];
            tensor<string, []> var_46821_equation_0 = const()[name = tensor<string, []>("op_46821_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46821_cast = einsum(equation = var_46821_equation_0, values = (var_46663_cast, var_46580_cast))[name = tensor<string, []>("op_46821_cast")];
            tensor<fp16, []> var_46822_to_fp16 = const()[name = tensor<string, []>("op_46822_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3973_cast = mul(x = var_46821_cast, y = var_46822_to_fp16)[name = tensor<string, []>("aw_3973_cast")];
            tensor<string, []> var_46825_equation_0 = const()[name = tensor<string, []>("op_46825_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46825_cast = einsum(equation = var_46825_equation_0, values = (var_46667_cast, var_46584_cast))[name = tensor<string, []>("op_46825_cast")];
            tensor<fp16, []> var_46826_to_fp16 = const()[name = tensor<string, []>("op_46826_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3975_cast = mul(x = var_46825_cast, y = var_46826_to_fp16)[name = tensor<string, []>("aw_3975_cast")];
            tensor<string, []> var_46829_equation_0 = const()[name = tensor<string, []>("op_46829_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46829_cast = einsum(equation = var_46829_equation_0, values = (var_46671_cast, var_46588_cast))[name = tensor<string, []>("op_46829_cast")];
            tensor<fp16, []> var_46830_to_fp16 = const()[name = tensor<string, []>("op_46830_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3977_cast = mul(x = var_46829_cast, y = var_46830_to_fp16)[name = tensor<string, []>("aw_3977_cast")];
            tensor<string, []> var_46833_equation_0 = const()[name = tensor<string, []>("op_46833_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46833_cast = einsum(equation = var_46833_equation_0, values = (var_46675_cast, var_46592_cast))[name = tensor<string, []>("op_46833_cast")];
            tensor<fp16, []> var_46834_to_fp16 = const()[name = tensor<string, []>("op_46834_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3979_cast = mul(x = var_46833_cast, y = var_46834_to_fp16)[name = tensor<string, []>("aw_3979_cast")];
            tensor<string, []> var_46837_equation_0 = const()[name = tensor<string, []>("op_46837_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46837_cast = einsum(equation = var_46837_equation_0, values = (var_46679_cast, var_46596_cast))[name = tensor<string, []>("op_46837_cast")];
            tensor<fp16, []> var_46838_to_fp16 = const()[name = tensor<string, []>("op_46838_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3981_cast = mul(x = var_46837_cast, y = var_46838_to_fp16)[name = tensor<string, []>("aw_3981_cast")];
            tensor<string, []> var_46841_equation_0 = const()[name = tensor<string, []>("op_46841_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46841_cast = einsum(equation = var_46841_equation_0, values = (var_46683_cast, var_46600_cast))[name = tensor<string, []>("op_46841_cast")];
            tensor<fp16, []> var_46842_to_fp16 = const()[name = tensor<string, []>("op_46842_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3983_cast = mul(x = var_46841_cast, y = var_46842_to_fp16)[name = tensor<string, []>("aw_3983_cast")];
            tensor<string, []> var_46845_equation_0 = const()[name = tensor<string, []>("op_46845_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46845_cast = einsum(equation = var_46845_equation_0, values = (var_46687_cast, var_46604_cast))[name = tensor<string, []>("op_46845_cast")];
            tensor<fp16, []> var_46846_to_fp16 = const()[name = tensor<string, []>("op_46846_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3985_cast = mul(x = var_46845_cast, y = var_46846_to_fp16)[name = tensor<string, []>("aw_3985_cast")];
            tensor<string, []> var_46849_equation_0 = const()[name = tensor<string, []>("op_46849_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46849_cast = einsum(equation = var_46849_equation_0, values = (var_46691_cast, var_46608_cast))[name = tensor<string, []>("op_46849_cast")];
            tensor<fp16, []> var_46850_to_fp16 = const()[name = tensor<string, []>("op_46850_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3987_cast = mul(x = var_46849_cast, y = var_46850_to_fp16)[name = tensor<string, []>("aw_3987_cast")];
            tensor<string, []> var_46853_equation_0 = const()[name = tensor<string, []>("op_46853_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46853_cast = einsum(equation = var_46853_equation_0, values = (var_46695_cast, var_46612_cast))[name = tensor<string, []>("op_46853_cast")];
            tensor<fp16, []> var_46854_to_fp16 = const()[name = tensor<string, []>("op_46854_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3989_cast = mul(x = var_46853_cast, y = var_46854_to_fp16)[name = tensor<string, []>("aw_3989_cast")];
            tensor<string, []> var_46857_equation_0 = const()[name = tensor<string, []>("op_46857_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46857_cast = einsum(equation = var_46857_equation_0, values = (var_46699_cast, var_46616_cast))[name = tensor<string, []>("op_46857_cast")];
            tensor<fp16, []> var_46858_to_fp16 = const()[name = tensor<string, []>("op_46858_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3991_cast = mul(x = var_46857_cast, y = var_46858_to_fp16)[name = tensor<string, []>("aw_3991_cast")];
            tensor<string, []> var_46861_equation_0 = const()[name = tensor<string, []>("op_46861_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46861_cast = einsum(equation = var_46861_equation_0, values = (var_46703_cast, var_46620_cast))[name = tensor<string, []>("op_46861_cast")];
            tensor<fp16, []> var_46862_to_fp16 = const()[name = tensor<string, []>("op_46862_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3993_cast = mul(x = var_46861_cast, y = var_46862_to_fp16)[name = tensor<string, []>("aw_3993_cast")];
            tensor<string, []> var_46865_equation_0 = const()[name = tensor<string, []>("op_46865_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46865_cast = einsum(equation = var_46865_equation_0, values = (var_46707_cast, var_46624_cast))[name = tensor<string, []>("op_46865_cast")];
            tensor<fp16, []> var_46866_to_fp16 = const()[name = tensor<string, []>("op_46866_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3995_cast = mul(x = var_46865_cast, y = var_46866_to_fp16)[name = tensor<string, []>("aw_3995_cast")];
            tensor<string, []> var_46869_equation_0 = const()[name = tensor<string, []>("op_46869_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46869_cast = einsum(equation = var_46869_equation_0, values = (var_46711_cast, var_46628_cast))[name = tensor<string, []>("op_46869_cast")];
            tensor<fp16, []> var_46870_to_fp16 = const()[name = tensor<string, []>("op_46870_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3997_cast = mul(x = var_46869_cast, y = var_46870_to_fp16)[name = tensor<string, []>("aw_3997_cast")];
            tensor<string, []> var_46873_equation_0 = const()[name = tensor<string, []>("op_46873_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_46873_cast = einsum(equation = var_46873_equation_0, values = (var_46715_cast, var_46632_cast))[name = tensor<string, []>("op_46873_cast")];
            tensor<fp16, []> var_46874_to_fp16 = const()[name = tensor<string, []>("op_46874_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_3999_cast = mul(x = var_46873_cast, y = var_46874_to_fp16)[name = tensor<string, []>("aw_3999_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46876_cast = softmax(axis = var_30385, x = aw_3961_cast)[name = tensor<string, []>("op_46876_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46877_cast = softmax(axis = var_30385, x = aw_3963_cast)[name = tensor<string, []>("op_46877_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46878_cast = softmax(axis = var_30385, x = aw_3965_cast)[name = tensor<string, []>("op_46878_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46879_cast = softmax(axis = var_30385, x = aw_3967_cast)[name = tensor<string, []>("op_46879_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46880_cast = softmax(axis = var_30385, x = aw_3969_cast)[name = tensor<string, []>("op_46880_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46881_cast = softmax(axis = var_30385, x = aw_3971_cast)[name = tensor<string, []>("op_46881_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46882_cast = softmax(axis = var_30385, x = aw_3973_cast)[name = tensor<string, []>("op_46882_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46883_cast = softmax(axis = var_30385, x = aw_3975_cast)[name = tensor<string, []>("op_46883_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46884_cast = softmax(axis = var_30385, x = aw_3977_cast)[name = tensor<string, []>("op_46884_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46885_cast = softmax(axis = var_30385, x = aw_3979_cast)[name = tensor<string, []>("op_46885_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46886_cast = softmax(axis = var_30385, x = aw_3981_cast)[name = tensor<string, []>("op_46886_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46887_cast = softmax(axis = var_30385, x = aw_3983_cast)[name = tensor<string, []>("op_46887_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46888_cast = softmax(axis = var_30385, x = aw_3985_cast)[name = tensor<string, []>("op_46888_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46889_cast = softmax(axis = var_30385, x = aw_3987_cast)[name = tensor<string, []>("op_46889_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46890_cast = softmax(axis = var_30385, x = aw_3989_cast)[name = tensor<string, []>("op_46890_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46891_cast = softmax(axis = var_30385, x = aw_3991_cast)[name = tensor<string, []>("op_46891_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46892_cast = softmax(axis = var_30385, x = aw_3993_cast)[name = tensor<string, []>("op_46892_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46893_cast = softmax(axis = var_30385, x = aw_3995_cast)[name = tensor<string, []>("op_46893_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46894_cast = softmax(axis = var_30385, x = aw_3997_cast)[name = tensor<string, []>("op_46894_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_46895_cast = softmax(axis = var_30385, x = aw_3999_cast)[name = tensor<string, []>("op_46895_cast")];
            tensor<string, []> var_46897_equation_0 = const()[name = tensor<string, []>("op_46897_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46897_cast = einsum(equation = var_46897_equation_0, values = (var_46717_cast, var_46876_cast))[name = tensor<string, []>("op_46897_cast")];
            tensor<string, []> var_46899_equation_0 = const()[name = tensor<string, []>("op_46899_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46899_cast = einsum(equation = var_46899_equation_0, values = (var_46721_cast, var_46877_cast))[name = tensor<string, []>("op_46899_cast")];
            tensor<string, []> var_46901_equation_0 = const()[name = tensor<string, []>("op_46901_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46901_cast = einsum(equation = var_46901_equation_0, values = (var_46725_cast, var_46878_cast))[name = tensor<string, []>("op_46901_cast")];
            tensor<string, []> var_46903_equation_0 = const()[name = tensor<string, []>("op_46903_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46903_cast = einsum(equation = var_46903_equation_0, values = (var_46729_cast, var_46879_cast))[name = tensor<string, []>("op_46903_cast")];
            tensor<string, []> var_46905_equation_0 = const()[name = tensor<string, []>("op_46905_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46905_cast = einsum(equation = var_46905_equation_0, values = (var_46733_cast, var_46880_cast))[name = tensor<string, []>("op_46905_cast")];
            tensor<string, []> var_46907_equation_0 = const()[name = tensor<string, []>("op_46907_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46907_cast = einsum(equation = var_46907_equation_0, values = (var_46737_cast, var_46881_cast))[name = tensor<string, []>("op_46907_cast")];
            tensor<string, []> var_46909_equation_0 = const()[name = tensor<string, []>("op_46909_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46909_cast = einsum(equation = var_46909_equation_0, values = (var_46741_cast, var_46882_cast))[name = tensor<string, []>("op_46909_cast")];
            tensor<string, []> var_46911_equation_0 = const()[name = tensor<string, []>("op_46911_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46911_cast = einsum(equation = var_46911_equation_0, values = (var_46745_cast, var_46883_cast))[name = tensor<string, []>("op_46911_cast")];
            tensor<string, []> var_46913_equation_0 = const()[name = tensor<string, []>("op_46913_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46913_cast = einsum(equation = var_46913_equation_0, values = (var_46749_cast, var_46884_cast))[name = tensor<string, []>("op_46913_cast")];
            tensor<string, []> var_46915_equation_0 = const()[name = tensor<string, []>("op_46915_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46915_cast = einsum(equation = var_46915_equation_0, values = (var_46753_cast, var_46885_cast))[name = tensor<string, []>("op_46915_cast")];
            tensor<string, []> var_46917_equation_0 = const()[name = tensor<string, []>("op_46917_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46917_cast = einsum(equation = var_46917_equation_0, values = (var_46757_cast, var_46886_cast))[name = tensor<string, []>("op_46917_cast")];
            tensor<string, []> var_46919_equation_0 = const()[name = tensor<string, []>("op_46919_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46919_cast = einsum(equation = var_46919_equation_0, values = (var_46761_cast, var_46887_cast))[name = tensor<string, []>("op_46919_cast")];
            tensor<string, []> var_46921_equation_0 = const()[name = tensor<string, []>("op_46921_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46921_cast = einsum(equation = var_46921_equation_0, values = (var_46765_cast, var_46888_cast))[name = tensor<string, []>("op_46921_cast")];
            tensor<string, []> var_46923_equation_0 = const()[name = tensor<string, []>("op_46923_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46923_cast = einsum(equation = var_46923_equation_0, values = (var_46769_cast, var_46889_cast))[name = tensor<string, []>("op_46923_cast")];
            tensor<string, []> var_46925_equation_0 = const()[name = tensor<string, []>("op_46925_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46925_cast = einsum(equation = var_46925_equation_0, values = (var_46773_cast, var_46890_cast))[name = tensor<string, []>("op_46925_cast")];
            tensor<string, []> var_46927_equation_0 = const()[name = tensor<string, []>("op_46927_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46927_cast = einsum(equation = var_46927_equation_0, values = (var_46777_cast, var_46891_cast))[name = tensor<string, []>("op_46927_cast")];
            tensor<string, []> var_46929_equation_0 = const()[name = tensor<string, []>("op_46929_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46929_cast = einsum(equation = var_46929_equation_0, values = (var_46781_cast, var_46892_cast))[name = tensor<string, []>("op_46929_cast")];
            tensor<string, []> var_46931_equation_0 = const()[name = tensor<string, []>("op_46931_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46931_cast = einsum(equation = var_46931_equation_0, values = (var_46785_cast, var_46893_cast))[name = tensor<string, []>("op_46931_cast")];
            tensor<string, []> var_46933_equation_0 = const()[name = tensor<string, []>("op_46933_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46933_cast = einsum(equation = var_46933_equation_0, values = (var_46789_cast, var_46894_cast))[name = tensor<string, []>("op_46933_cast")];
            tensor<string, []> var_46935_equation_0 = const()[name = tensor<string, []>("op_46935_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_46935_cast = einsum(equation = var_46935_equation_0, values = (var_46793_cast, var_46895_cast))[name = tensor<string, []>("op_46935_cast")];
            tensor<bool, []> input_597_interleave_0 = const()[name = tensor<string, []>("input_597_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_597_cast = concat(axis = var_30385, interleave = input_597_interleave_0, values = (var_46897_cast, var_46899_cast, var_46901_cast, var_46903_cast, var_46905_cast, var_46907_cast, var_46909_cast, var_46911_cast, var_46913_cast, var_46915_cast, var_46917_cast, var_46919_cast, var_46921_cast, var_46923_cast, var_46925_cast, var_46927_cast, var_46929_cast, var_46931_cast, var_46933_cast, var_46935_cast))[name = tensor<string, []>("input_597_cast")];
            tensor<int32, [2]> var_46941 = const()[name = tensor<string, []>("op_46941"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_46943 = const()[name = tensor<string, []>("op_46943"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_46945_pad_type_0 = const()[name = tensor<string, []>("op_46945_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_46945_pad_0 = const()[name = tensor<string, []>("op_46945_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3901919616)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3905196480)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_46945_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_out_0_bias_to_fp16, dilations = var_46943, groups = var_30385, pad = var_46945_pad_0, pad_type = var_46945_pad_type_0, strides = var_46941, weight = up_blocks_0_attentions_1_transformer_blocks_7_attn2_to_out_0_weight_to_fp16, x = input_597_cast)[name = tensor<string, []>("op_46945_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_311_cast = add(x = var_46945_cast, y = inputs_309_cast)[name = tensor<string, []>("inputs_311_cast")];
            tensor<int32, [1]> var_46949 = const()[name = tensor<string, []>("op_46949"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_311_cast = reduce_mean(axes = var_46949, keep_dims = var_30380, x = inputs_311_cast)[name = tensor<string, []>("channels_mean_311_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_311_cast = sub(x = inputs_311_cast, y = channels_mean_311_cast)[name = tensor<string, []>("zero_mean_311_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_311_cast = mul(x = zero_mean_311_cast, y = zero_mean_311_cast)[name = tensor<string, []>("zero_mean_sq_311_cast")];
            tensor<int32, [1]> var_46953 = const()[name = tensor<string, []>("op_46953"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_46954_cast = reduce_mean(axes = var_46953, keep_dims = var_30380, x = zero_mean_sq_311_cast)[name = tensor<string, []>("op_46954_cast")];
            tensor<fp16, []> var_46955_to_fp16 = const()[name = tensor<string, []>("op_46955_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_46956_cast = add(x = var_46954_cast, y = var_46955_to_fp16)[name = tensor<string, []>("op_46956_cast")];
            tensor<fp16, []> denom_311_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_311_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_311_cast = rsqrt(epsilon = denom_311_epsilon_0_to_fp16, x = var_46956_cast)[name = tensor<string, []>("denom_311_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_311_cast = mul(x = zero_mean_311_cast, y = denom_311_cast)[name = tensor<string, []>("out_311_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_46960_to_fp16 = const()[name = tensor<string, []>("op_46960_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3905199104)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_46961_cast = add(x = out_311_cast, y = var_46960_to_fp16)[name = tensor<string, []>("op_46961_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_46963_to_fp16 = const()[name = tensor<string, []>("op_46963_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3905201728)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_599_cast = mul(x = var_46961_cast, y = var_46963_to_fp16)[name = tensor<string, []>("input_599_cast")];
            tensor<int32, [2]> var_46971 = const()[name = tensor<string, []>("op_46971"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_46973 = const()[name = tensor<string, []>("op_46973"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_46975_pad_type_0 = const()[name = tensor<string, []>("op_46975_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_46975_pad_0 = const()[name = tensor<string, []>("op_46975_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3905204352)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_1_transformer_blocks_7_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3931418816)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_46975_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_7_ff_net_0_proj_bias_to_fp16, dilations = var_46973, groups = var_30385, pad = var_46975_pad_0, pad_type = var_46975_pad_type_0, strides = var_46971, weight = up_blocks_0_attentions_1_transformer_blocks_7_ff_net_0_proj_weight_to_fp16, x = input_599_cast)[name = tensor<string, []>("op_46975_cast")];
            tensor<int32, [2]> var_46976_split_sizes_0 = const()[name = tensor<string, []>("op_46976_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_46976_axis_0 = const()[name = tensor<string, []>("op_46976_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_46976_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_46976_cast_1 = split(axis = var_46976_axis_0, split_sizes = var_46976_split_sizes_0, x = var_46975_cast)[name = tensor<string, []>("op_46976_cast")];
            tensor<string, []> var_46978_mode_0 = const()[name = tensor<string, []>("op_46978_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_46978_cast = gelu(mode = var_46978_mode_0, x = var_46976_cast_1)[name = tensor<string, []>("op_46978_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_601_cast = mul(x = var_46976_cast_0, y = var_46978_cast)[name = tensor<string, []>("input_601_cast")];
            tensor<int32, [2]> var_46982 = const()[name = tensor<string, []>("op_46982"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_46984 = const()[name = tensor<string, []>("op_46984"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_46986_pad_type_0 = const()[name = tensor<string, []>("op_46986_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_46986_pad_0 = const()[name = tensor<string, []>("op_46986_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_7_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3931439360)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_7_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_7_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3944546624)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_46986_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_7_ff_net_2_bias_to_fp16, dilations = var_46984, groups = var_30385, pad = var_46986_pad_0, pad_type = var_46986_pad_type_0, strides = var_46982, weight = up_blocks_0_attentions_1_transformer_blocks_7_ff_net_2_weight_to_fp16, x = input_601_cast)[name = tensor<string, []>("op_46986_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_313_cast = add(x = var_46986_cast, y = inputs_311_cast)[name = tensor<string, []>("inputs_313_cast")];
            tensor<int32, [1]> var_46996 = const()[name = tensor<string, []>("op_46996"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_313_cast = reduce_mean(axes = var_46996, keep_dims = var_30380, x = inputs_313_cast)[name = tensor<string, []>("channels_mean_313_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_313_cast = sub(x = inputs_313_cast, y = channels_mean_313_cast)[name = tensor<string, []>("zero_mean_313_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_313_cast = mul(x = zero_mean_313_cast, y = zero_mean_313_cast)[name = tensor<string, []>("zero_mean_sq_313_cast")];
            tensor<int32, [1]> var_47000 = const()[name = tensor<string, []>("op_47000"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_47001_cast = reduce_mean(axes = var_47000, keep_dims = var_30380, x = zero_mean_sq_313_cast)[name = tensor<string, []>("op_47001_cast")];
            tensor<fp16, []> var_47002_to_fp16 = const()[name = tensor<string, []>("op_47002_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_47003_cast = add(x = var_47001_cast, y = var_47002_to_fp16)[name = tensor<string, []>("op_47003_cast")];
            tensor<fp16, []> denom_313_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_313_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_313_cast = rsqrt(epsilon = denom_313_epsilon_0_to_fp16, x = var_47003_cast)[name = tensor<string, []>("denom_313_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_313_cast = mul(x = zero_mean_313_cast, y = denom_313_cast)[name = tensor<string, []>("out_313_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_47007_to_fp16 = const()[name = tensor<string, []>("op_47007_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3944549248)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_47008_cast = add(x = out_313_cast, y = var_47007_to_fp16)[name = tensor<string, []>("op_47008_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_47010_to_fp16 = const()[name = tensor<string, []>("op_47010_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3944551872)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_409_cast = mul(x = var_47008_cast, y = var_47010_to_fp16)[name = tensor<string, []>("hidden_states_409_cast")];
            tensor<int32, [2]> var_47017 = const()[name = tensor<string, []>("op_47017"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_47019 = const()[name = tensor<string, []>("op_47019"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_209_pad_type_0 = const()[name = tensor<string, []>("q_209_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_209_pad_0 = const()[name = tensor<string, []>("q_209_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3944554496)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_209_cast = conv(dilations = var_47019, groups = var_30385, pad = q_209_pad_0, pad_type = q_209_pad_type_0, strides = var_47017, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_q_weight_to_fp16, x = hidden_states_409_cast)[name = tensor<string, []>("q_209_cast")];
            tensor<int32, [2]> var_47023 = const()[name = tensor<string, []>("op_47023"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_47025 = const()[name = tensor<string, []>("op_47025"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_417_pad_type_0 = const()[name = tensor<string, []>("k_417_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_417_pad_0 = const()[name = tensor<string, []>("k_417_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3947831360)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_417_cast = conv(dilations = var_47025, groups = var_30385, pad = k_417_pad_0, pad_type = k_417_pad_type_0, strides = var_47023, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_k_weight_to_fp16, x = hidden_states_409_cast)[name = tensor<string, []>("k_417_cast")];
            tensor<int32, [2]> var_47029 = const()[name = tensor<string, []>("op_47029"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_47031 = const()[name = tensor<string, []>("op_47031"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_209_pad_type_0 = const()[name = tensor<string, []>("v_209_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_209_pad_0 = const()[name = tensor<string, []>("v_209_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3951108224)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_209_cast = conv(dilations = var_47031, groups = var_30385, pad = v_209_pad_0, pad_type = v_209_pad_type_0, strides = var_47029, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_v_weight_to_fp16, x = hidden_states_409_cast)[name = tensor<string, []>("v_209_cast")];
            tensor<int32, [4]> var_47035_begin_0 = const()[name = tensor<string, []>("op_47035_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_47035_end_0 = const()[name = tensor<string, []>("op_47035_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_47035_end_mask_0 = const()[name = tensor<string, []>("op_47035_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47035_cast = slice_by_index(begin = var_47035_begin_0, end = var_47035_end_0, end_mask = var_47035_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47035_cast")];
            tensor<int32, [4]> var_47039_begin_0 = const()[name = tensor<string, []>("op_47039_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_47039_end_0 = const()[name = tensor<string, []>("op_47039_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_47039_end_mask_0 = const()[name = tensor<string, []>("op_47039_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47039_cast = slice_by_index(begin = var_47039_begin_0, end = var_47039_end_0, end_mask = var_47039_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47039_cast")];
            tensor<int32, [4]> var_47043_begin_0 = const()[name = tensor<string, []>("op_47043_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_47043_end_0 = const()[name = tensor<string, []>("op_47043_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_47043_end_mask_0 = const()[name = tensor<string, []>("op_47043_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47043_cast = slice_by_index(begin = var_47043_begin_0, end = var_47043_end_0, end_mask = var_47043_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47043_cast")];
            tensor<int32, [4]> var_47047_begin_0 = const()[name = tensor<string, []>("op_47047_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_47047_end_0 = const()[name = tensor<string, []>("op_47047_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_47047_end_mask_0 = const()[name = tensor<string, []>("op_47047_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47047_cast = slice_by_index(begin = var_47047_begin_0, end = var_47047_end_0, end_mask = var_47047_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47047_cast")];
            tensor<int32, [4]> var_47051_begin_0 = const()[name = tensor<string, []>("op_47051_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_47051_end_0 = const()[name = tensor<string, []>("op_47051_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_47051_end_mask_0 = const()[name = tensor<string, []>("op_47051_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47051_cast = slice_by_index(begin = var_47051_begin_0, end = var_47051_end_0, end_mask = var_47051_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47051_cast")];
            tensor<int32, [4]> var_47055_begin_0 = const()[name = tensor<string, []>("op_47055_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_47055_end_0 = const()[name = tensor<string, []>("op_47055_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_47055_end_mask_0 = const()[name = tensor<string, []>("op_47055_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47055_cast = slice_by_index(begin = var_47055_begin_0, end = var_47055_end_0, end_mask = var_47055_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47055_cast")];
            tensor<int32, [4]> var_47059_begin_0 = const()[name = tensor<string, []>("op_47059_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_47059_end_0 = const()[name = tensor<string, []>("op_47059_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_47059_end_mask_0 = const()[name = tensor<string, []>("op_47059_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47059_cast = slice_by_index(begin = var_47059_begin_0, end = var_47059_end_0, end_mask = var_47059_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47059_cast")];
            tensor<int32, [4]> var_47063_begin_0 = const()[name = tensor<string, []>("op_47063_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_47063_end_0 = const()[name = tensor<string, []>("op_47063_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_47063_end_mask_0 = const()[name = tensor<string, []>("op_47063_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47063_cast = slice_by_index(begin = var_47063_begin_0, end = var_47063_end_0, end_mask = var_47063_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47063_cast")];
            tensor<int32, [4]> var_47067_begin_0 = const()[name = tensor<string, []>("op_47067_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_47067_end_0 = const()[name = tensor<string, []>("op_47067_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_47067_end_mask_0 = const()[name = tensor<string, []>("op_47067_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47067_cast = slice_by_index(begin = var_47067_begin_0, end = var_47067_end_0, end_mask = var_47067_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47067_cast")];
            tensor<int32, [4]> var_47071_begin_0 = const()[name = tensor<string, []>("op_47071_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_47071_end_0 = const()[name = tensor<string, []>("op_47071_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_47071_end_mask_0 = const()[name = tensor<string, []>("op_47071_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47071_cast = slice_by_index(begin = var_47071_begin_0, end = var_47071_end_0, end_mask = var_47071_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47071_cast")];
            tensor<int32, [4]> var_47075_begin_0 = const()[name = tensor<string, []>("op_47075_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_47075_end_0 = const()[name = tensor<string, []>("op_47075_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_47075_end_mask_0 = const()[name = tensor<string, []>("op_47075_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47075_cast = slice_by_index(begin = var_47075_begin_0, end = var_47075_end_0, end_mask = var_47075_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47075_cast")];
            tensor<int32, [4]> var_47079_begin_0 = const()[name = tensor<string, []>("op_47079_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_47079_end_0 = const()[name = tensor<string, []>("op_47079_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_47079_end_mask_0 = const()[name = tensor<string, []>("op_47079_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47079_cast = slice_by_index(begin = var_47079_begin_0, end = var_47079_end_0, end_mask = var_47079_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47079_cast")];
            tensor<int32, [4]> var_47083_begin_0 = const()[name = tensor<string, []>("op_47083_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_47083_end_0 = const()[name = tensor<string, []>("op_47083_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_47083_end_mask_0 = const()[name = tensor<string, []>("op_47083_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47083_cast = slice_by_index(begin = var_47083_begin_0, end = var_47083_end_0, end_mask = var_47083_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47083_cast")];
            tensor<int32, [4]> var_47087_begin_0 = const()[name = tensor<string, []>("op_47087_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_47087_end_0 = const()[name = tensor<string, []>("op_47087_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_47087_end_mask_0 = const()[name = tensor<string, []>("op_47087_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47087_cast = slice_by_index(begin = var_47087_begin_0, end = var_47087_end_0, end_mask = var_47087_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47087_cast")];
            tensor<int32, [4]> var_47091_begin_0 = const()[name = tensor<string, []>("op_47091_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_47091_end_0 = const()[name = tensor<string, []>("op_47091_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_47091_end_mask_0 = const()[name = tensor<string, []>("op_47091_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47091_cast = slice_by_index(begin = var_47091_begin_0, end = var_47091_end_0, end_mask = var_47091_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47091_cast")];
            tensor<int32, [4]> var_47095_begin_0 = const()[name = tensor<string, []>("op_47095_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_47095_end_0 = const()[name = tensor<string, []>("op_47095_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_47095_end_mask_0 = const()[name = tensor<string, []>("op_47095_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47095_cast = slice_by_index(begin = var_47095_begin_0, end = var_47095_end_0, end_mask = var_47095_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47095_cast")];
            tensor<int32, [4]> var_47099_begin_0 = const()[name = tensor<string, []>("op_47099_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_47099_end_0 = const()[name = tensor<string, []>("op_47099_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_47099_end_mask_0 = const()[name = tensor<string, []>("op_47099_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47099_cast = slice_by_index(begin = var_47099_begin_0, end = var_47099_end_0, end_mask = var_47099_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47099_cast")];
            tensor<int32, [4]> var_47103_begin_0 = const()[name = tensor<string, []>("op_47103_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_47103_end_0 = const()[name = tensor<string, []>("op_47103_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_47103_end_mask_0 = const()[name = tensor<string, []>("op_47103_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47103_cast = slice_by_index(begin = var_47103_begin_0, end = var_47103_end_0, end_mask = var_47103_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47103_cast")];
            tensor<int32, [4]> var_47107_begin_0 = const()[name = tensor<string, []>("op_47107_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_47107_end_0 = const()[name = tensor<string, []>("op_47107_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_47107_end_mask_0 = const()[name = tensor<string, []>("op_47107_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47107_cast = slice_by_index(begin = var_47107_begin_0, end = var_47107_end_0, end_mask = var_47107_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47107_cast")];
            tensor<int32, [4]> var_47111_begin_0 = const()[name = tensor<string, []>("op_47111_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_47111_end_0 = const()[name = tensor<string, []>("op_47111_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_47111_end_mask_0 = const()[name = tensor<string, []>("op_47111_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47111_cast = slice_by_index(begin = var_47111_begin_0, end = var_47111_end_0, end_mask = var_47111_end_mask_0, x = q_209_cast)[name = tensor<string, []>("op_47111_cast")];
            tensor<int32, [4]> k_419_perm_0 = const()[name = tensor<string, []>("k_419_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_47118_begin_0 = const()[name = tensor<string, []>("op_47118_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_47118_end_0 = const()[name = tensor<string, []>("op_47118_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_47118_end_mask_0 = const()[name = tensor<string, []>("op_47118_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_35 = transpose(perm = k_419_perm_0, x = k_417_cast)[name = tensor<string, []>("transpose_35")];
            tensor<fp16, [2, 1024, 1, 64]> var_47118_cast = slice_by_index(begin = var_47118_begin_0, end = var_47118_end_0, end_mask = var_47118_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47118_cast")];
            tensor<int32, [4]> var_47122_begin_0 = const()[name = tensor<string, []>("op_47122_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_47122_end_0 = const()[name = tensor<string, []>("op_47122_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_47122_end_mask_0 = const()[name = tensor<string, []>("op_47122_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47122_cast = slice_by_index(begin = var_47122_begin_0, end = var_47122_end_0, end_mask = var_47122_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47122_cast")];
            tensor<int32, [4]> var_47126_begin_0 = const()[name = tensor<string, []>("op_47126_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_47126_end_0 = const()[name = tensor<string, []>("op_47126_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_47126_end_mask_0 = const()[name = tensor<string, []>("op_47126_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47126_cast = slice_by_index(begin = var_47126_begin_0, end = var_47126_end_0, end_mask = var_47126_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47126_cast")];
            tensor<int32, [4]> var_47130_begin_0 = const()[name = tensor<string, []>("op_47130_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_47130_end_0 = const()[name = tensor<string, []>("op_47130_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_47130_end_mask_0 = const()[name = tensor<string, []>("op_47130_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47130_cast = slice_by_index(begin = var_47130_begin_0, end = var_47130_end_0, end_mask = var_47130_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47130_cast")];
            tensor<int32, [4]> var_47134_begin_0 = const()[name = tensor<string, []>("op_47134_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_47134_end_0 = const()[name = tensor<string, []>("op_47134_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_47134_end_mask_0 = const()[name = tensor<string, []>("op_47134_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47134_cast = slice_by_index(begin = var_47134_begin_0, end = var_47134_end_0, end_mask = var_47134_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47134_cast")];
            tensor<int32, [4]> var_47138_begin_0 = const()[name = tensor<string, []>("op_47138_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_47138_end_0 = const()[name = tensor<string, []>("op_47138_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_47138_end_mask_0 = const()[name = tensor<string, []>("op_47138_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47138_cast = slice_by_index(begin = var_47138_begin_0, end = var_47138_end_0, end_mask = var_47138_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47138_cast")];
            tensor<int32, [4]> var_47142_begin_0 = const()[name = tensor<string, []>("op_47142_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_47142_end_0 = const()[name = tensor<string, []>("op_47142_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_47142_end_mask_0 = const()[name = tensor<string, []>("op_47142_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47142_cast = slice_by_index(begin = var_47142_begin_0, end = var_47142_end_0, end_mask = var_47142_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47142_cast")];
            tensor<int32, [4]> var_47146_begin_0 = const()[name = tensor<string, []>("op_47146_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_47146_end_0 = const()[name = tensor<string, []>("op_47146_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_47146_end_mask_0 = const()[name = tensor<string, []>("op_47146_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47146_cast = slice_by_index(begin = var_47146_begin_0, end = var_47146_end_0, end_mask = var_47146_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47146_cast")];
            tensor<int32, [4]> var_47150_begin_0 = const()[name = tensor<string, []>("op_47150_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_47150_end_0 = const()[name = tensor<string, []>("op_47150_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_47150_end_mask_0 = const()[name = tensor<string, []>("op_47150_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47150_cast = slice_by_index(begin = var_47150_begin_0, end = var_47150_end_0, end_mask = var_47150_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47150_cast")];
            tensor<int32, [4]> var_47154_begin_0 = const()[name = tensor<string, []>("op_47154_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_47154_end_0 = const()[name = tensor<string, []>("op_47154_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_47154_end_mask_0 = const()[name = tensor<string, []>("op_47154_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47154_cast = slice_by_index(begin = var_47154_begin_0, end = var_47154_end_0, end_mask = var_47154_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47154_cast")];
            tensor<int32, [4]> var_47158_begin_0 = const()[name = tensor<string, []>("op_47158_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_47158_end_0 = const()[name = tensor<string, []>("op_47158_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_47158_end_mask_0 = const()[name = tensor<string, []>("op_47158_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47158_cast = slice_by_index(begin = var_47158_begin_0, end = var_47158_end_0, end_mask = var_47158_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47158_cast")];
            tensor<int32, [4]> var_47162_begin_0 = const()[name = tensor<string, []>("op_47162_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_47162_end_0 = const()[name = tensor<string, []>("op_47162_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_47162_end_mask_0 = const()[name = tensor<string, []>("op_47162_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47162_cast = slice_by_index(begin = var_47162_begin_0, end = var_47162_end_0, end_mask = var_47162_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47162_cast")];
            tensor<int32, [4]> var_47166_begin_0 = const()[name = tensor<string, []>("op_47166_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_47166_end_0 = const()[name = tensor<string, []>("op_47166_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_47166_end_mask_0 = const()[name = tensor<string, []>("op_47166_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47166_cast = slice_by_index(begin = var_47166_begin_0, end = var_47166_end_0, end_mask = var_47166_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47166_cast")];
            tensor<int32, [4]> var_47170_begin_0 = const()[name = tensor<string, []>("op_47170_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_47170_end_0 = const()[name = tensor<string, []>("op_47170_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_47170_end_mask_0 = const()[name = tensor<string, []>("op_47170_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47170_cast = slice_by_index(begin = var_47170_begin_0, end = var_47170_end_0, end_mask = var_47170_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47170_cast")];
            tensor<int32, [4]> var_47174_begin_0 = const()[name = tensor<string, []>("op_47174_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_47174_end_0 = const()[name = tensor<string, []>("op_47174_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_47174_end_mask_0 = const()[name = tensor<string, []>("op_47174_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47174_cast = slice_by_index(begin = var_47174_begin_0, end = var_47174_end_0, end_mask = var_47174_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47174_cast")];
            tensor<int32, [4]> var_47178_begin_0 = const()[name = tensor<string, []>("op_47178_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_47178_end_0 = const()[name = tensor<string, []>("op_47178_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_47178_end_mask_0 = const()[name = tensor<string, []>("op_47178_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47178_cast = slice_by_index(begin = var_47178_begin_0, end = var_47178_end_0, end_mask = var_47178_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47178_cast")];
            tensor<int32, [4]> var_47182_begin_0 = const()[name = tensor<string, []>("op_47182_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_47182_end_0 = const()[name = tensor<string, []>("op_47182_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_47182_end_mask_0 = const()[name = tensor<string, []>("op_47182_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47182_cast = slice_by_index(begin = var_47182_begin_0, end = var_47182_end_0, end_mask = var_47182_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47182_cast")];
            tensor<int32, [4]> var_47186_begin_0 = const()[name = tensor<string, []>("op_47186_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_47186_end_0 = const()[name = tensor<string, []>("op_47186_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_47186_end_mask_0 = const()[name = tensor<string, []>("op_47186_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47186_cast = slice_by_index(begin = var_47186_begin_0, end = var_47186_end_0, end_mask = var_47186_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47186_cast")];
            tensor<int32, [4]> var_47190_begin_0 = const()[name = tensor<string, []>("op_47190_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_47190_end_0 = const()[name = tensor<string, []>("op_47190_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_47190_end_mask_0 = const()[name = tensor<string, []>("op_47190_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47190_cast = slice_by_index(begin = var_47190_begin_0, end = var_47190_end_0, end_mask = var_47190_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47190_cast")];
            tensor<int32, [4]> var_47194_begin_0 = const()[name = tensor<string, []>("op_47194_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_47194_end_0 = const()[name = tensor<string, []>("op_47194_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_47194_end_mask_0 = const()[name = tensor<string, []>("op_47194_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_47194_cast = slice_by_index(begin = var_47194_begin_0, end = var_47194_end_0, end_mask = var_47194_end_mask_0, x = transpose_35)[name = tensor<string, []>("op_47194_cast")];
            tensor<int32, [4]> var_47196_begin_0 = const()[name = tensor<string, []>("op_47196_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_47196_end_0 = const()[name = tensor<string, []>("op_47196_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_47196_end_mask_0 = const()[name = tensor<string, []>("op_47196_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47196_cast = slice_by_index(begin = var_47196_begin_0, end = var_47196_end_0, end_mask = var_47196_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47196_cast")];
            tensor<int32, [4]> var_47200_begin_0 = const()[name = tensor<string, []>("op_47200_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_47200_end_0 = const()[name = tensor<string, []>("op_47200_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_47200_end_mask_0 = const()[name = tensor<string, []>("op_47200_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47200_cast = slice_by_index(begin = var_47200_begin_0, end = var_47200_end_0, end_mask = var_47200_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47200_cast")];
            tensor<int32, [4]> var_47204_begin_0 = const()[name = tensor<string, []>("op_47204_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_47204_end_0 = const()[name = tensor<string, []>("op_47204_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_47204_end_mask_0 = const()[name = tensor<string, []>("op_47204_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47204_cast = slice_by_index(begin = var_47204_begin_0, end = var_47204_end_0, end_mask = var_47204_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47204_cast")];
            tensor<int32, [4]> var_47208_begin_0 = const()[name = tensor<string, []>("op_47208_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_47208_end_0 = const()[name = tensor<string, []>("op_47208_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_47208_end_mask_0 = const()[name = tensor<string, []>("op_47208_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47208_cast = slice_by_index(begin = var_47208_begin_0, end = var_47208_end_0, end_mask = var_47208_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47208_cast")];
            tensor<int32, [4]> var_47212_begin_0 = const()[name = tensor<string, []>("op_47212_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_47212_end_0 = const()[name = tensor<string, []>("op_47212_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_47212_end_mask_0 = const()[name = tensor<string, []>("op_47212_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47212_cast = slice_by_index(begin = var_47212_begin_0, end = var_47212_end_0, end_mask = var_47212_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47212_cast")];
            tensor<int32, [4]> var_47216_begin_0 = const()[name = tensor<string, []>("op_47216_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_47216_end_0 = const()[name = tensor<string, []>("op_47216_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_47216_end_mask_0 = const()[name = tensor<string, []>("op_47216_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47216_cast = slice_by_index(begin = var_47216_begin_0, end = var_47216_end_0, end_mask = var_47216_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47216_cast")];
            tensor<int32, [4]> var_47220_begin_0 = const()[name = tensor<string, []>("op_47220_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_47220_end_0 = const()[name = tensor<string, []>("op_47220_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_47220_end_mask_0 = const()[name = tensor<string, []>("op_47220_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47220_cast = slice_by_index(begin = var_47220_begin_0, end = var_47220_end_0, end_mask = var_47220_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47220_cast")];
            tensor<int32, [4]> var_47224_begin_0 = const()[name = tensor<string, []>("op_47224_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_47224_end_0 = const()[name = tensor<string, []>("op_47224_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_47224_end_mask_0 = const()[name = tensor<string, []>("op_47224_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47224_cast = slice_by_index(begin = var_47224_begin_0, end = var_47224_end_0, end_mask = var_47224_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47224_cast")];
            tensor<int32, [4]> var_47228_begin_0 = const()[name = tensor<string, []>("op_47228_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_47228_end_0 = const()[name = tensor<string, []>("op_47228_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_47228_end_mask_0 = const()[name = tensor<string, []>("op_47228_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47228_cast = slice_by_index(begin = var_47228_begin_0, end = var_47228_end_0, end_mask = var_47228_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47228_cast")];
            tensor<int32, [4]> var_47232_begin_0 = const()[name = tensor<string, []>("op_47232_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_47232_end_0 = const()[name = tensor<string, []>("op_47232_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_47232_end_mask_0 = const()[name = tensor<string, []>("op_47232_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47232_cast = slice_by_index(begin = var_47232_begin_0, end = var_47232_end_0, end_mask = var_47232_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47232_cast")];
            tensor<int32, [4]> var_47236_begin_0 = const()[name = tensor<string, []>("op_47236_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_47236_end_0 = const()[name = tensor<string, []>("op_47236_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_47236_end_mask_0 = const()[name = tensor<string, []>("op_47236_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47236_cast = slice_by_index(begin = var_47236_begin_0, end = var_47236_end_0, end_mask = var_47236_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47236_cast")];
            tensor<int32, [4]> var_47240_begin_0 = const()[name = tensor<string, []>("op_47240_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_47240_end_0 = const()[name = tensor<string, []>("op_47240_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_47240_end_mask_0 = const()[name = tensor<string, []>("op_47240_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47240_cast = slice_by_index(begin = var_47240_begin_0, end = var_47240_end_0, end_mask = var_47240_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47240_cast")];
            tensor<int32, [4]> var_47244_begin_0 = const()[name = tensor<string, []>("op_47244_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_47244_end_0 = const()[name = tensor<string, []>("op_47244_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_47244_end_mask_0 = const()[name = tensor<string, []>("op_47244_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47244_cast = slice_by_index(begin = var_47244_begin_0, end = var_47244_end_0, end_mask = var_47244_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47244_cast")];
            tensor<int32, [4]> var_47248_begin_0 = const()[name = tensor<string, []>("op_47248_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_47248_end_0 = const()[name = tensor<string, []>("op_47248_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_47248_end_mask_0 = const()[name = tensor<string, []>("op_47248_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47248_cast = slice_by_index(begin = var_47248_begin_0, end = var_47248_end_0, end_mask = var_47248_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47248_cast")];
            tensor<int32, [4]> var_47252_begin_0 = const()[name = tensor<string, []>("op_47252_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_47252_end_0 = const()[name = tensor<string, []>("op_47252_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_47252_end_mask_0 = const()[name = tensor<string, []>("op_47252_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47252_cast = slice_by_index(begin = var_47252_begin_0, end = var_47252_end_0, end_mask = var_47252_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47252_cast")];
            tensor<int32, [4]> var_47256_begin_0 = const()[name = tensor<string, []>("op_47256_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_47256_end_0 = const()[name = tensor<string, []>("op_47256_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_47256_end_mask_0 = const()[name = tensor<string, []>("op_47256_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47256_cast = slice_by_index(begin = var_47256_begin_0, end = var_47256_end_0, end_mask = var_47256_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47256_cast")];
            tensor<int32, [4]> var_47260_begin_0 = const()[name = tensor<string, []>("op_47260_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_47260_end_0 = const()[name = tensor<string, []>("op_47260_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_47260_end_mask_0 = const()[name = tensor<string, []>("op_47260_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47260_cast = slice_by_index(begin = var_47260_begin_0, end = var_47260_end_0, end_mask = var_47260_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47260_cast")];
            tensor<int32, [4]> var_47264_begin_0 = const()[name = tensor<string, []>("op_47264_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_47264_end_0 = const()[name = tensor<string, []>("op_47264_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_47264_end_mask_0 = const()[name = tensor<string, []>("op_47264_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47264_cast = slice_by_index(begin = var_47264_begin_0, end = var_47264_end_0, end_mask = var_47264_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47264_cast")];
            tensor<int32, [4]> var_47268_begin_0 = const()[name = tensor<string, []>("op_47268_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_47268_end_0 = const()[name = tensor<string, []>("op_47268_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_47268_end_mask_0 = const()[name = tensor<string, []>("op_47268_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47268_cast = slice_by_index(begin = var_47268_begin_0, end = var_47268_end_0, end_mask = var_47268_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47268_cast")];
            tensor<int32, [4]> var_47272_begin_0 = const()[name = tensor<string, []>("op_47272_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_47272_end_0 = const()[name = tensor<string, []>("op_47272_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_47272_end_mask_0 = const()[name = tensor<string, []>("op_47272_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47272_cast = slice_by_index(begin = var_47272_begin_0, end = var_47272_end_0, end_mask = var_47272_end_mask_0, x = v_209_cast)[name = tensor<string, []>("op_47272_cast")];
            tensor<string, []> var_47276_equation_0 = const()[name = tensor<string, []>("op_47276_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47276_cast = einsum(equation = var_47276_equation_0, values = (var_47118_cast, var_47035_cast))[name = tensor<string, []>("op_47276_cast")];
            tensor<fp16, []> var_47277_to_fp16 = const()[name = tensor<string, []>("op_47277_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4001_cast = mul(x = var_47276_cast, y = var_47277_to_fp16)[name = tensor<string, []>("aw_4001_cast")];
            tensor<string, []> var_47280_equation_0 = const()[name = tensor<string, []>("op_47280_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47280_cast = einsum(equation = var_47280_equation_0, values = (var_47122_cast, var_47039_cast))[name = tensor<string, []>("op_47280_cast")];
            tensor<fp16, []> var_47281_to_fp16 = const()[name = tensor<string, []>("op_47281_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4003_cast = mul(x = var_47280_cast, y = var_47281_to_fp16)[name = tensor<string, []>("aw_4003_cast")];
            tensor<string, []> var_47284_equation_0 = const()[name = tensor<string, []>("op_47284_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47284_cast = einsum(equation = var_47284_equation_0, values = (var_47126_cast, var_47043_cast))[name = tensor<string, []>("op_47284_cast")];
            tensor<fp16, []> var_47285_to_fp16 = const()[name = tensor<string, []>("op_47285_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4005_cast = mul(x = var_47284_cast, y = var_47285_to_fp16)[name = tensor<string, []>("aw_4005_cast")];
            tensor<string, []> var_47288_equation_0 = const()[name = tensor<string, []>("op_47288_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47288_cast = einsum(equation = var_47288_equation_0, values = (var_47130_cast, var_47047_cast))[name = tensor<string, []>("op_47288_cast")];
            tensor<fp16, []> var_47289_to_fp16 = const()[name = tensor<string, []>("op_47289_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4007_cast = mul(x = var_47288_cast, y = var_47289_to_fp16)[name = tensor<string, []>("aw_4007_cast")];
            tensor<string, []> var_47292_equation_0 = const()[name = tensor<string, []>("op_47292_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47292_cast = einsum(equation = var_47292_equation_0, values = (var_47134_cast, var_47051_cast))[name = tensor<string, []>("op_47292_cast")];
            tensor<fp16, []> var_47293_to_fp16 = const()[name = tensor<string, []>("op_47293_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4009_cast = mul(x = var_47292_cast, y = var_47293_to_fp16)[name = tensor<string, []>("aw_4009_cast")];
            tensor<string, []> var_47296_equation_0 = const()[name = tensor<string, []>("op_47296_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47296_cast = einsum(equation = var_47296_equation_0, values = (var_47138_cast, var_47055_cast))[name = tensor<string, []>("op_47296_cast")];
            tensor<fp16, []> var_47297_to_fp16 = const()[name = tensor<string, []>("op_47297_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4011_cast = mul(x = var_47296_cast, y = var_47297_to_fp16)[name = tensor<string, []>("aw_4011_cast")];
            tensor<string, []> var_47300_equation_0 = const()[name = tensor<string, []>("op_47300_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47300_cast = einsum(equation = var_47300_equation_0, values = (var_47142_cast, var_47059_cast))[name = tensor<string, []>("op_47300_cast")];
            tensor<fp16, []> var_47301_to_fp16 = const()[name = tensor<string, []>("op_47301_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4013_cast = mul(x = var_47300_cast, y = var_47301_to_fp16)[name = tensor<string, []>("aw_4013_cast")];
            tensor<string, []> var_47304_equation_0 = const()[name = tensor<string, []>("op_47304_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47304_cast = einsum(equation = var_47304_equation_0, values = (var_47146_cast, var_47063_cast))[name = tensor<string, []>("op_47304_cast")];
            tensor<fp16, []> var_47305_to_fp16 = const()[name = tensor<string, []>("op_47305_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4015_cast = mul(x = var_47304_cast, y = var_47305_to_fp16)[name = tensor<string, []>("aw_4015_cast")];
            tensor<string, []> var_47308_equation_0 = const()[name = tensor<string, []>("op_47308_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47308_cast = einsum(equation = var_47308_equation_0, values = (var_47150_cast, var_47067_cast))[name = tensor<string, []>("op_47308_cast")];
            tensor<fp16, []> var_47309_to_fp16 = const()[name = tensor<string, []>("op_47309_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4017_cast = mul(x = var_47308_cast, y = var_47309_to_fp16)[name = tensor<string, []>("aw_4017_cast")];
            tensor<string, []> var_47312_equation_0 = const()[name = tensor<string, []>("op_47312_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47312_cast = einsum(equation = var_47312_equation_0, values = (var_47154_cast, var_47071_cast))[name = tensor<string, []>("op_47312_cast")];
            tensor<fp16, []> var_47313_to_fp16 = const()[name = tensor<string, []>("op_47313_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4019_cast = mul(x = var_47312_cast, y = var_47313_to_fp16)[name = tensor<string, []>("aw_4019_cast")];
            tensor<string, []> var_47316_equation_0 = const()[name = tensor<string, []>("op_47316_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47316_cast = einsum(equation = var_47316_equation_0, values = (var_47158_cast, var_47075_cast))[name = tensor<string, []>("op_47316_cast")];
            tensor<fp16, []> var_47317_to_fp16 = const()[name = tensor<string, []>("op_47317_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4021_cast = mul(x = var_47316_cast, y = var_47317_to_fp16)[name = tensor<string, []>("aw_4021_cast")];
            tensor<string, []> var_47320_equation_0 = const()[name = tensor<string, []>("op_47320_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47320_cast = einsum(equation = var_47320_equation_0, values = (var_47162_cast, var_47079_cast))[name = tensor<string, []>("op_47320_cast")];
            tensor<fp16, []> var_47321_to_fp16 = const()[name = tensor<string, []>("op_47321_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4023_cast = mul(x = var_47320_cast, y = var_47321_to_fp16)[name = tensor<string, []>("aw_4023_cast")];
            tensor<string, []> var_47324_equation_0 = const()[name = tensor<string, []>("op_47324_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47324_cast = einsum(equation = var_47324_equation_0, values = (var_47166_cast, var_47083_cast))[name = tensor<string, []>("op_47324_cast")];
            tensor<fp16, []> var_47325_to_fp16 = const()[name = tensor<string, []>("op_47325_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4025_cast = mul(x = var_47324_cast, y = var_47325_to_fp16)[name = tensor<string, []>("aw_4025_cast")];
            tensor<string, []> var_47328_equation_0 = const()[name = tensor<string, []>("op_47328_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47328_cast = einsum(equation = var_47328_equation_0, values = (var_47170_cast, var_47087_cast))[name = tensor<string, []>("op_47328_cast")];
            tensor<fp16, []> var_47329_to_fp16 = const()[name = tensor<string, []>("op_47329_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4027_cast = mul(x = var_47328_cast, y = var_47329_to_fp16)[name = tensor<string, []>("aw_4027_cast")];
            tensor<string, []> var_47332_equation_0 = const()[name = tensor<string, []>("op_47332_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47332_cast = einsum(equation = var_47332_equation_0, values = (var_47174_cast, var_47091_cast))[name = tensor<string, []>("op_47332_cast")];
            tensor<fp16, []> var_47333_to_fp16 = const()[name = tensor<string, []>("op_47333_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4029_cast = mul(x = var_47332_cast, y = var_47333_to_fp16)[name = tensor<string, []>("aw_4029_cast")];
            tensor<string, []> var_47336_equation_0 = const()[name = tensor<string, []>("op_47336_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47336_cast = einsum(equation = var_47336_equation_0, values = (var_47178_cast, var_47095_cast))[name = tensor<string, []>("op_47336_cast")];
            tensor<fp16, []> var_47337_to_fp16 = const()[name = tensor<string, []>("op_47337_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4031_cast = mul(x = var_47336_cast, y = var_47337_to_fp16)[name = tensor<string, []>("aw_4031_cast")];
            tensor<string, []> var_47340_equation_0 = const()[name = tensor<string, []>("op_47340_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47340_cast = einsum(equation = var_47340_equation_0, values = (var_47182_cast, var_47099_cast))[name = tensor<string, []>("op_47340_cast")];
            tensor<fp16, []> var_47341_to_fp16 = const()[name = tensor<string, []>("op_47341_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4033_cast = mul(x = var_47340_cast, y = var_47341_to_fp16)[name = tensor<string, []>("aw_4033_cast")];
            tensor<string, []> var_47344_equation_0 = const()[name = tensor<string, []>("op_47344_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47344_cast = einsum(equation = var_47344_equation_0, values = (var_47186_cast, var_47103_cast))[name = tensor<string, []>("op_47344_cast")];
            tensor<fp16, []> var_47345_to_fp16 = const()[name = tensor<string, []>("op_47345_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4035_cast = mul(x = var_47344_cast, y = var_47345_to_fp16)[name = tensor<string, []>("aw_4035_cast")];
            tensor<string, []> var_47348_equation_0 = const()[name = tensor<string, []>("op_47348_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47348_cast = einsum(equation = var_47348_equation_0, values = (var_47190_cast, var_47107_cast))[name = tensor<string, []>("op_47348_cast")];
            tensor<fp16, []> var_47349_to_fp16 = const()[name = tensor<string, []>("op_47349_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4037_cast = mul(x = var_47348_cast, y = var_47349_to_fp16)[name = tensor<string, []>("aw_4037_cast")];
            tensor<string, []> var_47352_equation_0 = const()[name = tensor<string, []>("op_47352_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47352_cast = einsum(equation = var_47352_equation_0, values = (var_47194_cast, var_47111_cast))[name = tensor<string, []>("op_47352_cast")];
            tensor<fp16, []> var_47353_to_fp16 = const()[name = tensor<string, []>("op_47353_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4039_cast = mul(x = var_47352_cast, y = var_47353_to_fp16)[name = tensor<string, []>("aw_4039_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47355_cast = softmax(axis = var_30385, x = aw_4001_cast)[name = tensor<string, []>("op_47355_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47356_cast = softmax(axis = var_30385, x = aw_4003_cast)[name = tensor<string, []>("op_47356_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47357_cast = softmax(axis = var_30385, x = aw_4005_cast)[name = tensor<string, []>("op_47357_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47358_cast = softmax(axis = var_30385, x = aw_4007_cast)[name = tensor<string, []>("op_47358_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47359_cast = softmax(axis = var_30385, x = aw_4009_cast)[name = tensor<string, []>("op_47359_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47360_cast = softmax(axis = var_30385, x = aw_4011_cast)[name = tensor<string, []>("op_47360_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47361_cast = softmax(axis = var_30385, x = aw_4013_cast)[name = tensor<string, []>("op_47361_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47362_cast = softmax(axis = var_30385, x = aw_4015_cast)[name = tensor<string, []>("op_47362_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47363_cast = softmax(axis = var_30385, x = aw_4017_cast)[name = tensor<string, []>("op_47363_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47364_cast = softmax(axis = var_30385, x = aw_4019_cast)[name = tensor<string, []>("op_47364_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47365_cast = softmax(axis = var_30385, x = aw_4021_cast)[name = tensor<string, []>("op_47365_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47366_cast = softmax(axis = var_30385, x = aw_4023_cast)[name = tensor<string, []>("op_47366_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47367_cast = softmax(axis = var_30385, x = aw_4025_cast)[name = tensor<string, []>("op_47367_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47368_cast = softmax(axis = var_30385, x = aw_4027_cast)[name = tensor<string, []>("op_47368_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47369_cast = softmax(axis = var_30385, x = aw_4029_cast)[name = tensor<string, []>("op_47369_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47370_cast = softmax(axis = var_30385, x = aw_4031_cast)[name = tensor<string, []>("op_47370_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47371_cast = softmax(axis = var_30385, x = aw_4033_cast)[name = tensor<string, []>("op_47371_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47372_cast = softmax(axis = var_30385, x = aw_4035_cast)[name = tensor<string, []>("op_47372_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47373_cast = softmax(axis = var_30385, x = aw_4037_cast)[name = tensor<string, []>("op_47373_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_47374_cast = softmax(axis = var_30385, x = aw_4039_cast)[name = tensor<string, []>("op_47374_cast")];
            tensor<string, []> var_47376_equation_0 = const()[name = tensor<string, []>("op_47376_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47376_cast = einsum(equation = var_47376_equation_0, values = (var_47196_cast, var_47355_cast))[name = tensor<string, []>("op_47376_cast")];
            tensor<string, []> var_47378_equation_0 = const()[name = tensor<string, []>("op_47378_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47378_cast = einsum(equation = var_47378_equation_0, values = (var_47200_cast, var_47356_cast))[name = tensor<string, []>("op_47378_cast")];
            tensor<string, []> var_47380_equation_0 = const()[name = tensor<string, []>("op_47380_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47380_cast = einsum(equation = var_47380_equation_0, values = (var_47204_cast, var_47357_cast))[name = tensor<string, []>("op_47380_cast")];
            tensor<string, []> var_47382_equation_0 = const()[name = tensor<string, []>("op_47382_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47382_cast = einsum(equation = var_47382_equation_0, values = (var_47208_cast, var_47358_cast))[name = tensor<string, []>("op_47382_cast")];
            tensor<string, []> var_47384_equation_0 = const()[name = tensor<string, []>("op_47384_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47384_cast = einsum(equation = var_47384_equation_0, values = (var_47212_cast, var_47359_cast))[name = tensor<string, []>("op_47384_cast")];
            tensor<string, []> var_47386_equation_0 = const()[name = tensor<string, []>("op_47386_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47386_cast = einsum(equation = var_47386_equation_0, values = (var_47216_cast, var_47360_cast))[name = tensor<string, []>("op_47386_cast")];
            tensor<string, []> var_47388_equation_0 = const()[name = tensor<string, []>("op_47388_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47388_cast = einsum(equation = var_47388_equation_0, values = (var_47220_cast, var_47361_cast))[name = tensor<string, []>("op_47388_cast")];
            tensor<string, []> var_47390_equation_0 = const()[name = tensor<string, []>("op_47390_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47390_cast = einsum(equation = var_47390_equation_0, values = (var_47224_cast, var_47362_cast))[name = tensor<string, []>("op_47390_cast")];
            tensor<string, []> var_47392_equation_0 = const()[name = tensor<string, []>("op_47392_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47392_cast = einsum(equation = var_47392_equation_0, values = (var_47228_cast, var_47363_cast))[name = tensor<string, []>("op_47392_cast")];
            tensor<string, []> var_47394_equation_0 = const()[name = tensor<string, []>("op_47394_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47394_cast = einsum(equation = var_47394_equation_0, values = (var_47232_cast, var_47364_cast))[name = tensor<string, []>("op_47394_cast")];
            tensor<string, []> var_47396_equation_0 = const()[name = tensor<string, []>("op_47396_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47396_cast = einsum(equation = var_47396_equation_0, values = (var_47236_cast, var_47365_cast))[name = tensor<string, []>("op_47396_cast")];
            tensor<string, []> var_47398_equation_0 = const()[name = tensor<string, []>("op_47398_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47398_cast = einsum(equation = var_47398_equation_0, values = (var_47240_cast, var_47366_cast))[name = tensor<string, []>("op_47398_cast")];
            tensor<string, []> var_47400_equation_0 = const()[name = tensor<string, []>("op_47400_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47400_cast = einsum(equation = var_47400_equation_0, values = (var_47244_cast, var_47367_cast))[name = tensor<string, []>("op_47400_cast")];
            tensor<string, []> var_47402_equation_0 = const()[name = tensor<string, []>("op_47402_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47402_cast = einsum(equation = var_47402_equation_0, values = (var_47248_cast, var_47368_cast))[name = tensor<string, []>("op_47402_cast")];
            tensor<string, []> var_47404_equation_0 = const()[name = tensor<string, []>("op_47404_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47404_cast = einsum(equation = var_47404_equation_0, values = (var_47252_cast, var_47369_cast))[name = tensor<string, []>("op_47404_cast")];
            tensor<string, []> var_47406_equation_0 = const()[name = tensor<string, []>("op_47406_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47406_cast = einsum(equation = var_47406_equation_0, values = (var_47256_cast, var_47370_cast))[name = tensor<string, []>("op_47406_cast")];
            tensor<string, []> var_47408_equation_0 = const()[name = tensor<string, []>("op_47408_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47408_cast = einsum(equation = var_47408_equation_0, values = (var_47260_cast, var_47371_cast))[name = tensor<string, []>("op_47408_cast")];
            tensor<string, []> var_47410_equation_0 = const()[name = tensor<string, []>("op_47410_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47410_cast = einsum(equation = var_47410_equation_0, values = (var_47264_cast, var_47372_cast))[name = tensor<string, []>("op_47410_cast")];
            tensor<string, []> var_47412_equation_0 = const()[name = tensor<string, []>("op_47412_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47412_cast = einsum(equation = var_47412_equation_0, values = (var_47268_cast, var_47373_cast))[name = tensor<string, []>("op_47412_cast")];
            tensor<string, []> var_47414_equation_0 = const()[name = tensor<string, []>("op_47414_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47414_cast = einsum(equation = var_47414_equation_0, values = (var_47272_cast, var_47374_cast))[name = tensor<string, []>("op_47414_cast")];
            tensor<bool, []> input_603_interleave_0 = const()[name = tensor<string, []>("input_603_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_603_cast = concat(axis = var_30385, interleave = input_603_interleave_0, values = (var_47376_cast, var_47378_cast, var_47380_cast, var_47382_cast, var_47384_cast, var_47386_cast, var_47388_cast, var_47390_cast, var_47392_cast, var_47394_cast, var_47396_cast, var_47398_cast, var_47400_cast, var_47402_cast, var_47404_cast, var_47406_cast, var_47408_cast, var_47410_cast, var_47412_cast, var_47414_cast))[name = tensor<string, []>("input_603_cast")];
            tensor<int32, [2]> var_47420 = const()[name = tensor<string, []>("op_47420"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_47422 = const()[name = tensor<string, []>("op_47422"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_47424_pad_type_0 = const()[name = tensor<string, []>("op_47424_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_47424_pad_0 = const()[name = tensor<string, []>("op_47424_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3954385088)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3957661952)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_47424_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_out_0_bias_to_fp16, dilations = var_47422, groups = var_30385, pad = var_47424_pad_0, pad_type = var_47424_pad_type_0, strides = var_47420, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn1_to_out_0_weight_to_fp16, x = input_603_cast)[name = tensor<string, []>("op_47424_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_315_cast = add(x = var_47424_cast, y = inputs_313_cast)[name = tensor<string, []>("inputs_315_cast")];
            tensor<int32, [1]> var_47428 = const()[name = tensor<string, []>("op_47428"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_315_cast = reduce_mean(axes = var_47428, keep_dims = var_30380, x = inputs_315_cast)[name = tensor<string, []>("channels_mean_315_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_315_cast = sub(x = inputs_315_cast, y = channels_mean_315_cast)[name = tensor<string, []>("zero_mean_315_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_315_cast = mul(x = zero_mean_315_cast, y = zero_mean_315_cast)[name = tensor<string, []>("zero_mean_sq_315_cast")];
            tensor<int32, [1]> var_47432 = const()[name = tensor<string, []>("op_47432"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_47433_cast = reduce_mean(axes = var_47432, keep_dims = var_30380, x = zero_mean_sq_315_cast)[name = tensor<string, []>("op_47433_cast")];
            tensor<fp16, []> var_47434_to_fp16 = const()[name = tensor<string, []>("op_47434_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_47435_cast = add(x = var_47433_cast, y = var_47434_to_fp16)[name = tensor<string, []>("op_47435_cast")];
            tensor<fp16, []> denom_315_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_315_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_315_cast = rsqrt(epsilon = denom_315_epsilon_0_to_fp16, x = var_47435_cast)[name = tensor<string, []>("denom_315_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_315_cast = mul(x = zero_mean_315_cast, y = denom_315_cast)[name = tensor<string, []>("out_315_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_47439_to_fp16 = const()[name = tensor<string, []>("op_47439_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3957664576)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_47440_cast = add(x = out_315_cast, y = var_47439_to_fp16)[name = tensor<string, []>("op_47440_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_47442_to_fp16 = const()[name = tensor<string, []>("op_47442_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3957667200)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_411_cast = mul(x = var_47440_cast, y = var_47442_to_fp16)[name = tensor<string, []>("hidden_states_411_cast")];
            tensor<int32, [2]> var_47449 = const()[name = tensor<string, []>("op_47449"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_47451 = const()[name = tensor<string, []>("op_47451"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_211_pad_type_0 = const()[name = tensor<string, []>("q_211_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_211_pad_0 = const()[name = tensor<string, []>("q_211_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3957669824)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_211_cast = conv(dilations = var_47451, groups = var_30385, pad = q_211_pad_0, pad_type = q_211_pad_type_0, strides = var_47449, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_q_weight_to_fp16, x = hidden_states_411_cast)[name = tensor<string, []>("q_211_cast")];
            tensor<int32, [2]> var_47455 = const()[name = tensor<string, []>("op_47455"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_47457 = const()[name = tensor<string, []>("op_47457"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_421_pad_type_0 = const()[name = tensor<string, []>("k_421_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_421_pad_0 = const()[name = tensor<string, []>("k_421_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3960946688)))];
            tensor<fp16, [2, 1280, 1, 77]> k_421_cast = conv(dilations = var_47457, groups = var_30385, pad = k_421_pad_0, pad_type = k_421_pad_type_0, strides = var_47455, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_421_cast")];
            tensor<int32, [2]> var_47461 = const()[name = tensor<string, []>("op_47461"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_47463 = const()[name = tensor<string, []>("op_47463"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_211_pad_type_0 = const()[name = tensor<string, []>("v_211_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_211_pad_0 = const()[name = tensor<string, []>("v_211_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3966189632)))];
            tensor<fp16, [2, 1280, 1, 77]> v_211_cast = conv(dilations = var_47463, groups = var_30385, pad = v_211_pad_0, pad_type = v_211_pad_type_0, strides = var_47461, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_211_cast")];
            tensor<int32, [4]> var_47467_begin_0 = const()[name = tensor<string, []>("op_47467_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_47467_end_0 = const()[name = tensor<string, []>("op_47467_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_47467_end_mask_0 = const()[name = tensor<string, []>("op_47467_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47467_cast = slice_by_index(begin = var_47467_begin_0, end = var_47467_end_0, end_mask = var_47467_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47467_cast")];
            tensor<int32, [4]> var_47471_begin_0 = const()[name = tensor<string, []>("op_47471_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_47471_end_0 = const()[name = tensor<string, []>("op_47471_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_47471_end_mask_0 = const()[name = tensor<string, []>("op_47471_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47471_cast = slice_by_index(begin = var_47471_begin_0, end = var_47471_end_0, end_mask = var_47471_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47471_cast")];
            tensor<int32, [4]> var_47475_begin_0 = const()[name = tensor<string, []>("op_47475_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_47475_end_0 = const()[name = tensor<string, []>("op_47475_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_47475_end_mask_0 = const()[name = tensor<string, []>("op_47475_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47475_cast = slice_by_index(begin = var_47475_begin_0, end = var_47475_end_0, end_mask = var_47475_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47475_cast")];
            tensor<int32, [4]> var_47479_begin_0 = const()[name = tensor<string, []>("op_47479_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_47479_end_0 = const()[name = tensor<string, []>("op_47479_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_47479_end_mask_0 = const()[name = tensor<string, []>("op_47479_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47479_cast = slice_by_index(begin = var_47479_begin_0, end = var_47479_end_0, end_mask = var_47479_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47479_cast")];
            tensor<int32, [4]> var_47483_begin_0 = const()[name = tensor<string, []>("op_47483_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_47483_end_0 = const()[name = tensor<string, []>("op_47483_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_47483_end_mask_0 = const()[name = tensor<string, []>("op_47483_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47483_cast = slice_by_index(begin = var_47483_begin_0, end = var_47483_end_0, end_mask = var_47483_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47483_cast")];
            tensor<int32, [4]> var_47487_begin_0 = const()[name = tensor<string, []>("op_47487_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_47487_end_0 = const()[name = tensor<string, []>("op_47487_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_47487_end_mask_0 = const()[name = tensor<string, []>("op_47487_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47487_cast = slice_by_index(begin = var_47487_begin_0, end = var_47487_end_0, end_mask = var_47487_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47487_cast")];
            tensor<int32, [4]> var_47491_begin_0 = const()[name = tensor<string, []>("op_47491_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_47491_end_0 = const()[name = tensor<string, []>("op_47491_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_47491_end_mask_0 = const()[name = tensor<string, []>("op_47491_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47491_cast = slice_by_index(begin = var_47491_begin_0, end = var_47491_end_0, end_mask = var_47491_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47491_cast")];
            tensor<int32, [4]> var_47495_begin_0 = const()[name = tensor<string, []>("op_47495_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_47495_end_0 = const()[name = tensor<string, []>("op_47495_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_47495_end_mask_0 = const()[name = tensor<string, []>("op_47495_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47495_cast = slice_by_index(begin = var_47495_begin_0, end = var_47495_end_0, end_mask = var_47495_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47495_cast")];
            tensor<int32, [4]> var_47499_begin_0 = const()[name = tensor<string, []>("op_47499_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_47499_end_0 = const()[name = tensor<string, []>("op_47499_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_47499_end_mask_0 = const()[name = tensor<string, []>("op_47499_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47499_cast = slice_by_index(begin = var_47499_begin_0, end = var_47499_end_0, end_mask = var_47499_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47499_cast")];
            tensor<int32, [4]> var_47503_begin_0 = const()[name = tensor<string, []>("op_47503_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_47503_end_0 = const()[name = tensor<string, []>("op_47503_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_47503_end_mask_0 = const()[name = tensor<string, []>("op_47503_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47503_cast = slice_by_index(begin = var_47503_begin_0, end = var_47503_end_0, end_mask = var_47503_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47503_cast")];
            tensor<int32, [4]> var_47507_begin_0 = const()[name = tensor<string, []>("op_47507_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_47507_end_0 = const()[name = tensor<string, []>("op_47507_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_47507_end_mask_0 = const()[name = tensor<string, []>("op_47507_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47507_cast = slice_by_index(begin = var_47507_begin_0, end = var_47507_end_0, end_mask = var_47507_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47507_cast")];
            tensor<int32, [4]> var_47511_begin_0 = const()[name = tensor<string, []>("op_47511_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_47511_end_0 = const()[name = tensor<string, []>("op_47511_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_47511_end_mask_0 = const()[name = tensor<string, []>("op_47511_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47511_cast = slice_by_index(begin = var_47511_begin_0, end = var_47511_end_0, end_mask = var_47511_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47511_cast")];
            tensor<int32, [4]> var_47515_begin_0 = const()[name = tensor<string, []>("op_47515_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_47515_end_0 = const()[name = tensor<string, []>("op_47515_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_47515_end_mask_0 = const()[name = tensor<string, []>("op_47515_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47515_cast = slice_by_index(begin = var_47515_begin_0, end = var_47515_end_0, end_mask = var_47515_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47515_cast")];
            tensor<int32, [4]> var_47519_begin_0 = const()[name = tensor<string, []>("op_47519_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_47519_end_0 = const()[name = tensor<string, []>("op_47519_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_47519_end_mask_0 = const()[name = tensor<string, []>("op_47519_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47519_cast = slice_by_index(begin = var_47519_begin_0, end = var_47519_end_0, end_mask = var_47519_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47519_cast")];
            tensor<int32, [4]> var_47523_begin_0 = const()[name = tensor<string, []>("op_47523_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_47523_end_0 = const()[name = tensor<string, []>("op_47523_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_47523_end_mask_0 = const()[name = tensor<string, []>("op_47523_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47523_cast = slice_by_index(begin = var_47523_begin_0, end = var_47523_end_0, end_mask = var_47523_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47523_cast")];
            tensor<int32, [4]> var_47527_begin_0 = const()[name = tensor<string, []>("op_47527_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_47527_end_0 = const()[name = tensor<string, []>("op_47527_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_47527_end_mask_0 = const()[name = tensor<string, []>("op_47527_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47527_cast = slice_by_index(begin = var_47527_begin_0, end = var_47527_end_0, end_mask = var_47527_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47527_cast")];
            tensor<int32, [4]> var_47531_begin_0 = const()[name = tensor<string, []>("op_47531_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_47531_end_0 = const()[name = tensor<string, []>("op_47531_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_47531_end_mask_0 = const()[name = tensor<string, []>("op_47531_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47531_cast = slice_by_index(begin = var_47531_begin_0, end = var_47531_end_0, end_mask = var_47531_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47531_cast")];
            tensor<int32, [4]> var_47535_begin_0 = const()[name = tensor<string, []>("op_47535_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_47535_end_0 = const()[name = tensor<string, []>("op_47535_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_47535_end_mask_0 = const()[name = tensor<string, []>("op_47535_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47535_cast = slice_by_index(begin = var_47535_begin_0, end = var_47535_end_0, end_mask = var_47535_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47535_cast")];
            tensor<int32, [4]> var_47539_begin_0 = const()[name = tensor<string, []>("op_47539_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_47539_end_0 = const()[name = tensor<string, []>("op_47539_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_47539_end_mask_0 = const()[name = tensor<string, []>("op_47539_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47539_cast = slice_by_index(begin = var_47539_begin_0, end = var_47539_end_0, end_mask = var_47539_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47539_cast")];
            tensor<int32, [4]> var_47543_begin_0 = const()[name = tensor<string, []>("op_47543_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_47543_end_0 = const()[name = tensor<string, []>("op_47543_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_47543_end_mask_0 = const()[name = tensor<string, []>("op_47543_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47543_cast = slice_by_index(begin = var_47543_begin_0, end = var_47543_end_0, end_mask = var_47543_end_mask_0, x = q_211_cast)[name = tensor<string, []>("op_47543_cast")];
            tensor<int32, [4]> k_423_perm_0 = const()[name = tensor<string, []>("k_423_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_47550_begin_0 = const()[name = tensor<string, []>("op_47550_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_47550_end_0 = const()[name = tensor<string, []>("op_47550_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_47550_end_mask_0 = const()[name = tensor<string, []>("op_47550_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_34 = transpose(perm = k_423_perm_0, x = k_421_cast)[name = tensor<string, []>("transpose_34")];
            tensor<fp16, [2, 77, 1, 64]> var_47550_cast = slice_by_index(begin = var_47550_begin_0, end = var_47550_end_0, end_mask = var_47550_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47550_cast")];
            tensor<int32, [4]> var_47554_begin_0 = const()[name = tensor<string, []>("op_47554_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_47554_end_0 = const()[name = tensor<string, []>("op_47554_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_47554_end_mask_0 = const()[name = tensor<string, []>("op_47554_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47554_cast = slice_by_index(begin = var_47554_begin_0, end = var_47554_end_0, end_mask = var_47554_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47554_cast")];
            tensor<int32, [4]> var_47558_begin_0 = const()[name = tensor<string, []>("op_47558_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_47558_end_0 = const()[name = tensor<string, []>("op_47558_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_47558_end_mask_0 = const()[name = tensor<string, []>("op_47558_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47558_cast = slice_by_index(begin = var_47558_begin_0, end = var_47558_end_0, end_mask = var_47558_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47558_cast")];
            tensor<int32, [4]> var_47562_begin_0 = const()[name = tensor<string, []>("op_47562_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_47562_end_0 = const()[name = tensor<string, []>("op_47562_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_47562_end_mask_0 = const()[name = tensor<string, []>("op_47562_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47562_cast = slice_by_index(begin = var_47562_begin_0, end = var_47562_end_0, end_mask = var_47562_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47562_cast")];
            tensor<int32, [4]> var_47566_begin_0 = const()[name = tensor<string, []>("op_47566_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_47566_end_0 = const()[name = tensor<string, []>("op_47566_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_47566_end_mask_0 = const()[name = tensor<string, []>("op_47566_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47566_cast = slice_by_index(begin = var_47566_begin_0, end = var_47566_end_0, end_mask = var_47566_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47566_cast")];
            tensor<int32, [4]> var_47570_begin_0 = const()[name = tensor<string, []>("op_47570_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_47570_end_0 = const()[name = tensor<string, []>("op_47570_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_47570_end_mask_0 = const()[name = tensor<string, []>("op_47570_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47570_cast = slice_by_index(begin = var_47570_begin_0, end = var_47570_end_0, end_mask = var_47570_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47570_cast")];
            tensor<int32, [4]> var_47574_begin_0 = const()[name = tensor<string, []>("op_47574_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_47574_end_0 = const()[name = tensor<string, []>("op_47574_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_47574_end_mask_0 = const()[name = tensor<string, []>("op_47574_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47574_cast = slice_by_index(begin = var_47574_begin_0, end = var_47574_end_0, end_mask = var_47574_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47574_cast")];
            tensor<int32, [4]> var_47578_begin_0 = const()[name = tensor<string, []>("op_47578_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_47578_end_0 = const()[name = tensor<string, []>("op_47578_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_47578_end_mask_0 = const()[name = tensor<string, []>("op_47578_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47578_cast = slice_by_index(begin = var_47578_begin_0, end = var_47578_end_0, end_mask = var_47578_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47578_cast")];
            tensor<int32, [4]> var_47582_begin_0 = const()[name = tensor<string, []>("op_47582_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_47582_end_0 = const()[name = tensor<string, []>("op_47582_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_47582_end_mask_0 = const()[name = tensor<string, []>("op_47582_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47582_cast = slice_by_index(begin = var_47582_begin_0, end = var_47582_end_0, end_mask = var_47582_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47582_cast")];
            tensor<int32, [4]> var_47586_begin_0 = const()[name = tensor<string, []>("op_47586_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_47586_end_0 = const()[name = tensor<string, []>("op_47586_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_47586_end_mask_0 = const()[name = tensor<string, []>("op_47586_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47586_cast = slice_by_index(begin = var_47586_begin_0, end = var_47586_end_0, end_mask = var_47586_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47586_cast")];
            tensor<int32, [4]> var_47590_begin_0 = const()[name = tensor<string, []>("op_47590_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_47590_end_0 = const()[name = tensor<string, []>("op_47590_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_47590_end_mask_0 = const()[name = tensor<string, []>("op_47590_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47590_cast = slice_by_index(begin = var_47590_begin_0, end = var_47590_end_0, end_mask = var_47590_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47590_cast")];
            tensor<int32, [4]> var_47594_begin_0 = const()[name = tensor<string, []>("op_47594_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_47594_end_0 = const()[name = tensor<string, []>("op_47594_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_47594_end_mask_0 = const()[name = tensor<string, []>("op_47594_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47594_cast = slice_by_index(begin = var_47594_begin_0, end = var_47594_end_0, end_mask = var_47594_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47594_cast")];
            tensor<int32, [4]> var_47598_begin_0 = const()[name = tensor<string, []>("op_47598_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_47598_end_0 = const()[name = tensor<string, []>("op_47598_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_47598_end_mask_0 = const()[name = tensor<string, []>("op_47598_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47598_cast = slice_by_index(begin = var_47598_begin_0, end = var_47598_end_0, end_mask = var_47598_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47598_cast")];
            tensor<int32, [4]> var_47602_begin_0 = const()[name = tensor<string, []>("op_47602_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_47602_end_0 = const()[name = tensor<string, []>("op_47602_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_47602_end_mask_0 = const()[name = tensor<string, []>("op_47602_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47602_cast = slice_by_index(begin = var_47602_begin_0, end = var_47602_end_0, end_mask = var_47602_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47602_cast")];
            tensor<int32, [4]> var_47606_begin_0 = const()[name = tensor<string, []>("op_47606_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_47606_end_0 = const()[name = tensor<string, []>("op_47606_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_47606_end_mask_0 = const()[name = tensor<string, []>("op_47606_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47606_cast = slice_by_index(begin = var_47606_begin_0, end = var_47606_end_0, end_mask = var_47606_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47606_cast")];
            tensor<int32, [4]> var_47610_begin_0 = const()[name = tensor<string, []>("op_47610_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_47610_end_0 = const()[name = tensor<string, []>("op_47610_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_47610_end_mask_0 = const()[name = tensor<string, []>("op_47610_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47610_cast = slice_by_index(begin = var_47610_begin_0, end = var_47610_end_0, end_mask = var_47610_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47610_cast")];
            tensor<int32, [4]> var_47614_begin_0 = const()[name = tensor<string, []>("op_47614_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_47614_end_0 = const()[name = tensor<string, []>("op_47614_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_47614_end_mask_0 = const()[name = tensor<string, []>("op_47614_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47614_cast = slice_by_index(begin = var_47614_begin_0, end = var_47614_end_0, end_mask = var_47614_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47614_cast")];
            tensor<int32, [4]> var_47618_begin_0 = const()[name = tensor<string, []>("op_47618_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_47618_end_0 = const()[name = tensor<string, []>("op_47618_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_47618_end_mask_0 = const()[name = tensor<string, []>("op_47618_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47618_cast = slice_by_index(begin = var_47618_begin_0, end = var_47618_end_0, end_mask = var_47618_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47618_cast")];
            tensor<int32, [4]> var_47622_begin_0 = const()[name = tensor<string, []>("op_47622_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_47622_end_0 = const()[name = tensor<string, []>("op_47622_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_47622_end_mask_0 = const()[name = tensor<string, []>("op_47622_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47622_cast = slice_by_index(begin = var_47622_begin_0, end = var_47622_end_0, end_mask = var_47622_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47622_cast")];
            tensor<int32, [4]> var_47626_begin_0 = const()[name = tensor<string, []>("op_47626_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_47626_end_0 = const()[name = tensor<string, []>("op_47626_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_47626_end_mask_0 = const()[name = tensor<string, []>("op_47626_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_47626_cast = slice_by_index(begin = var_47626_begin_0, end = var_47626_end_0, end_mask = var_47626_end_mask_0, x = transpose_34)[name = tensor<string, []>("op_47626_cast")];
            tensor<int32, [4]> var_47628_begin_0 = const()[name = tensor<string, []>("op_47628_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_47628_end_0 = const()[name = tensor<string, []>("op_47628_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_47628_end_mask_0 = const()[name = tensor<string, []>("op_47628_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47628_cast = slice_by_index(begin = var_47628_begin_0, end = var_47628_end_0, end_mask = var_47628_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47628_cast")];
            tensor<int32, [4]> var_47632_begin_0 = const()[name = tensor<string, []>("op_47632_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_47632_end_0 = const()[name = tensor<string, []>("op_47632_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_47632_end_mask_0 = const()[name = tensor<string, []>("op_47632_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47632_cast = slice_by_index(begin = var_47632_begin_0, end = var_47632_end_0, end_mask = var_47632_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47632_cast")];
            tensor<int32, [4]> var_47636_begin_0 = const()[name = tensor<string, []>("op_47636_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_47636_end_0 = const()[name = tensor<string, []>("op_47636_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_47636_end_mask_0 = const()[name = tensor<string, []>("op_47636_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47636_cast = slice_by_index(begin = var_47636_begin_0, end = var_47636_end_0, end_mask = var_47636_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47636_cast")];
            tensor<int32, [4]> var_47640_begin_0 = const()[name = tensor<string, []>("op_47640_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_47640_end_0 = const()[name = tensor<string, []>("op_47640_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_47640_end_mask_0 = const()[name = tensor<string, []>("op_47640_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47640_cast = slice_by_index(begin = var_47640_begin_0, end = var_47640_end_0, end_mask = var_47640_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47640_cast")];
            tensor<int32, [4]> var_47644_begin_0 = const()[name = tensor<string, []>("op_47644_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_47644_end_0 = const()[name = tensor<string, []>("op_47644_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_47644_end_mask_0 = const()[name = tensor<string, []>("op_47644_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47644_cast = slice_by_index(begin = var_47644_begin_0, end = var_47644_end_0, end_mask = var_47644_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47644_cast")];
            tensor<int32, [4]> var_47648_begin_0 = const()[name = tensor<string, []>("op_47648_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_47648_end_0 = const()[name = tensor<string, []>("op_47648_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_47648_end_mask_0 = const()[name = tensor<string, []>("op_47648_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47648_cast = slice_by_index(begin = var_47648_begin_0, end = var_47648_end_0, end_mask = var_47648_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47648_cast")];
            tensor<int32, [4]> var_47652_begin_0 = const()[name = tensor<string, []>("op_47652_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_47652_end_0 = const()[name = tensor<string, []>("op_47652_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_47652_end_mask_0 = const()[name = tensor<string, []>("op_47652_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47652_cast = slice_by_index(begin = var_47652_begin_0, end = var_47652_end_0, end_mask = var_47652_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47652_cast")];
            tensor<int32, [4]> var_47656_begin_0 = const()[name = tensor<string, []>("op_47656_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_47656_end_0 = const()[name = tensor<string, []>("op_47656_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_47656_end_mask_0 = const()[name = tensor<string, []>("op_47656_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47656_cast = slice_by_index(begin = var_47656_begin_0, end = var_47656_end_0, end_mask = var_47656_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47656_cast")];
            tensor<int32, [4]> var_47660_begin_0 = const()[name = tensor<string, []>("op_47660_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_47660_end_0 = const()[name = tensor<string, []>("op_47660_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_47660_end_mask_0 = const()[name = tensor<string, []>("op_47660_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47660_cast = slice_by_index(begin = var_47660_begin_0, end = var_47660_end_0, end_mask = var_47660_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47660_cast")];
            tensor<int32, [4]> var_47664_begin_0 = const()[name = tensor<string, []>("op_47664_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_47664_end_0 = const()[name = tensor<string, []>("op_47664_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_47664_end_mask_0 = const()[name = tensor<string, []>("op_47664_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47664_cast = slice_by_index(begin = var_47664_begin_0, end = var_47664_end_0, end_mask = var_47664_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47664_cast")];
            tensor<int32, [4]> var_47668_begin_0 = const()[name = tensor<string, []>("op_47668_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_47668_end_0 = const()[name = tensor<string, []>("op_47668_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_47668_end_mask_0 = const()[name = tensor<string, []>("op_47668_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47668_cast = slice_by_index(begin = var_47668_begin_0, end = var_47668_end_0, end_mask = var_47668_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47668_cast")];
            tensor<int32, [4]> var_47672_begin_0 = const()[name = tensor<string, []>("op_47672_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_47672_end_0 = const()[name = tensor<string, []>("op_47672_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_47672_end_mask_0 = const()[name = tensor<string, []>("op_47672_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47672_cast = slice_by_index(begin = var_47672_begin_0, end = var_47672_end_0, end_mask = var_47672_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47672_cast")];
            tensor<int32, [4]> var_47676_begin_0 = const()[name = tensor<string, []>("op_47676_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_47676_end_0 = const()[name = tensor<string, []>("op_47676_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_47676_end_mask_0 = const()[name = tensor<string, []>("op_47676_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47676_cast = slice_by_index(begin = var_47676_begin_0, end = var_47676_end_0, end_mask = var_47676_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47676_cast")];
            tensor<int32, [4]> var_47680_begin_0 = const()[name = tensor<string, []>("op_47680_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_47680_end_0 = const()[name = tensor<string, []>("op_47680_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_47680_end_mask_0 = const()[name = tensor<string, []>("op_47680_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47680_cast = slice_by_index(begin = var_47680_begin_0, end = var_47680_end_0, end_mask = var_47680_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47680_cast")];
            tensor<int32, [4]> var_47684_begin_0 = const()[name = tensor<string, []>("op_47684_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_47684_end_0 = const()[name = tensor<string, []>("op_47684_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_47684_end_mask_0 = const()[name = tensor<string, []>("op_47684_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47684_cast = slice_by_index(begin = var_47684_begin_0, end = var_47684_end_0, end_mask = var_47684_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47684_cast")];
            tensor<int32, [4]> var_47688_begin_0 = const()[name = tensor<string, []>("op_47688_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_47688_end_0 = const()[name = tensor<string, []>("op_47688_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_47688_end_mask_0 = const()[name = tensor<string, []>("op_47688_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47688_cast = slice_by_index(begin = var_47688_begin_0, end = var_47688_end_0, end_mask = var_47688_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47688_cast")];
            tensor<int32, [4]> var_47692_begin_0 = const()[name = tensor<string, []>("op_47692_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_47692_end_0 = const()[name = tensor<string, []>("op_47692_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_47692_end_mask_0 = const()[name = tensor<string, []>("op_47692_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47692_cast = slice_by_index(begin = var_47692_begin_0, end = var_47692_end_0, end_mask = var_47692_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47692_cast")];
            tensor<int32, [4]> var_47696_begin_0 = const()[name = tensor<string, []>("op_47696_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_47696_end_0 = const()[name = tensor<string, []>("op_47696_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_47696_end_mask_0 = const()[name = tensor<string, []>("op_47696_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47696_cast = slice_by_index(begin = var_47696_begin_0, end = var_47696_end_0, end_mask = var_47696_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47696_cast")];
            tensor<int32, [4]> var_47700_begin_0 = const()[name = tensor<string, []>("op_47700_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_47700_end_0 = const()[name = tensor<string, []>("op_47700_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_47700_end_mask_0 = const()[name = tensor<string, []>("op_47700_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47700_cast = slice_by_index(begin = var_47700_begin_0, end = var_47700_end_0, end_mask = var_47700_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47700_cast")];
            tensor<int32, [4]> var_47704_begin_0 = const()[name = tensor<string, []>("op_47704_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_47704_end_0 = const()[name = tensor<string, []>("op_47704_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_47704_end_mask_0 = const()[name = tensor<string, []>("op_47704_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_47704_cast = slice_by_index(begin = var_47704_begin_0, end = var_47704_end_0, end_mask = var_47704_end_mask_0, x = v_211_cast)[name = tensor<string, []>("op_47704_cast")];
            tensor<string, []> var_47708_equation_0 = const()[name = tensor<string, []>("op_47708_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47708_cast = einsum(equation = var_47708_equation_0, values = (var_47550_cast, var_47467_cast))[name = tensor<string, []>("op_47708_cast")];
            tensor<fp16, []> var_47709_to_fp16 = const()[name = tensor<string, []>("op_47709_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4041_cast = mul(x = var_47708_cast, y = var_47709_to_fp16)[name = tensor<string, []>("aw_4041_cast")];
            tensor<string, []> var_47712_equation_0 = const()[name = tensor<string, []>("op_47712_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47712_cast = einsum(equation = var_47712_equation_0, values = (var_47554_cast, var_47471_cast))[name = tensor<string, []>("op_47712_cast")];
            tensor<fp16, []> var_47713_to_fp16 = const()[name = tensor<string, []>("op_47713_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4043_cast = mul(x = var_47712_cast, y = var_47713_to_fp16)[name = tensor<string, []>("aw_4043_cast")];
            tensor<string, []> var_47716_equation_0 = const()[name = tensor<string, []>("op_47716_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47716_cast = einsum(equation = var_47716_equation_0, values = (var_47558_cast, var_47475_cast))[name = tensor<string, []>("op_47716_cast")];
            tensor<fp16, []> var_47717_to_fp16 = const()[name = tensor<string, []>("op_47717_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4045_cast = mul(x = var_47716_cast, y = var_47717_to_fp16)[name = tensor<string, []>("aw_4045_cast")];
            tensor<string, []> var_47720_equation_0 = const()[name = tensor<string, []>("op_47720_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47720_cast = einsum(equation = var_47720_equation_0, values = (var_47562_cast, var_47479_cast))[name = tensor<string, []>("op_47720_cast")];
            tensor<fp16, []> var_47721_to_fp16 = const()[name = tensor<string, []>("op_47721_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4047_cast = mul(x = var_47720_cast, y = var_47721_to_fp16)[name = tensor<string, []>("aw_4047_cast")];
            tensor<string, []> var_47724_equation_0 = const()[name = tensor<string, []>("op_47724_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47724_cast = einsum(equation = var_47724_equation_0, values = (var_47566_cast, var_47483_cast))[name = tensor<string, []>("op_47724_cast")];
            tensor<fp16, []> var_47725_to_fp16 = const()[name = tensor<string, []>("op_47725_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4049_cast = mul(x = var_47724_cast, y = var_47725_to_fp16)[name = tensor<string, []>("aw_4049_cast")];
            tensor<string, []> var_47728_equation_0 = const()[name = tensor<string, []>("op_47728_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47728_cast = einsum(equation = var_47728_equation_0, values = (var_47570_cast, var_47487_cast))[name = tensor<string, []>("op_47728_cast")];
            tensor<fp16, []> var_47729_to_fp16 = const()[name = tensor<string, []>("op_47729_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4051_cast = mul(x = var_47728_cast, y = var_47729_to_fp16)[name = tensor<string, []>("aw_4051_cast")];
            tensor<string, []> var_47732_equation_0 = const()[name = tensor<string, []>("op_47732_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47732_cast = einsum(equation = var_47732_equation_0, values = (var_47574_cast, var_47491_cast))[name = tensor<string, []>("op_47732_cast")];
            tensor<fp16, []> var_47733_to_fp16 = const()[name = tensor<string, []>("op_47733_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4053_cast = mul(x = var_47732_cast, y = var_47733_to_fp16)[name = tensor<string, []>("aw_4053_cast")];
            tensor<string, []> var_47736_equation_0 = const()[name = tensor<string, []>("op_47736_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47736_cast = einsum(equation = var_47736_equation_0, values = (var_47578_cast, var_47495_cast))[name = tensor<string, []>("op_47736_cast")];
            tensor<fp16, []> var_47737_to_fp16 = const()[name = tensor<string, []>("op_47737_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4055_cast = mul(x = var_47736_cast, y = var_47737_to_fp16)[name = tensor<string, []>("aw_4055_cast")];
            tensor<string, []> var_47740_equation_0 = const()[name = tensor<string, []>("op_47740_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47740_cast = einsum(equation = var_47740_equation_0, values = (var_47582_cast, var_47499_cast))[name = tensor<string, []>("op_47740_cast")];
            tensor<fp16, []> var_47741_to_fp16 = const()[name = tensor<string, []>("op_47741_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4057_cast = mul(x = var_47740_cast, y = var_47741_to_fp16)[name = tensor<string, []>("aw_4057_cast")];
            tensor<string, []> var_47744_equation_0 = const()[name = tensor<string, []>("op_47744_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47744_cast = einsum(equation = var_47744_equation_0, values = (var_47586_cast, var_47503_cast))[name = tensor<string, []>("op_47744_cast")];
            tensor<fp16, []> var_47745_to_fp16 = const()[name = tensor<string, []>("op_47745_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4059_cast = mul(x = var_47744_cast, y = var_47745_to_fp16)[name = tensor<string, []>("aw_4059_cast")];
            tensor<string, []> var_47748_equation_0 = const()[name = tensor<string, []>("op_47748_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47748_cast = einsum(equation = var_47748_equation_0, values = (var_47590_cast, var_47507_cast))[name = tensor<string, []>("op_47748_cast")];
            tensor<fp16, []> var_47749_to_fp16 = const()[name = tensor<string, []>("op_47749_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4061_cast = mul(x = var_47748_cast, y = var_47749_to_fp16)[name = tensor<string, []>("aw_4061_cast")];
            tensor<string, []> var_47752_equation_0 = const()[name = tensor<string, []>("op_47752_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47752_cast = einsum(equation = var_47752_equation_0, values = (var_47594_cast, var_47511_cast))[name = tensor<string, []>("op_47752_cast")];
            tensor<fp16, []> var_47753_to_fp16 = const()[name = tensor<string, []>("op_47753_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4063_cast = mul(x = var_47752_cast, y = var_47753_to_fp16)[name = tensor<string, []>("aw_4063_cast")];
            tensor<string, []> var_47756_equation_0 = const()[name = tensor<string, []>("op_47756_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47756_cast = einsum(equation = var_47756_equation_0, values = (var_47598_cast, var_47515_cast))[name = tensor<string, []>("op_47756_cast")];
            tensor<fp16, []> var_47757_to_fp16 = const()[name = tensor<string, []>("op_47757_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4065_cast = mul(x = var_47756_cast, y = var_47757_to_fp16)[name = tensor<string, []>("aw_4065_cast")];
            tensor<string, []> var_47760_equation_0 = const()[name = tensor<string, []>("op_47760_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47760_cast = einsum(equation = var_47760_equation_0, values = (var_47602_cast, var_47519_cast))[name = tensor<string, []>("op_47760_cast")];
            tensor<fp16, []> var_47761_to_fp16 = const()[name = tensor<string, []>("op_47761_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4067_cast = mul(x = var_47760_cast, y = var_47761_to_fp16)[name = tensor<string, []>("aw_4067_cast")];
            tensor<string, []> var_47764_equation_0 = const()[name = tensor<string, []>("op_47764_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47764_cast = einsum(equation = var_47764_equation_0, values = (var_47606_cast, var_47523_cast))[name = tensor<string, []>("op_47764_cast")];
            tensor<fp16, []> var_47765_to_fp16 = const()[name = tensor<string, []>("op_47765_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4069_cast = mul(x = var_47764_cast, y = var_47765_to_fp16)[name = tensor<string, []>("aw_4069_cast")];
            tensor<string, []> var_47768_equation_0 = const()[name = tensor<string, []>("op_47768_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47768_cast = einsum(equation = var_47768_equation_0, values = (var_47610_cast, var_47527_cast))[name = tensor<string, []>("op_47768_cast")];
            tensor<fp16, []> var_47769_to_fp16 = const()[name = tensor<string, []>("op_47769_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4071_cast = mul(x = var_47768_cast, y = var_47769_to_fp16)[name = tensor<string, []>("aw_4071_cast")];
            tensor<string, []> var_47772_equation_0 = const()[name = tensor<string, []>("op_47772_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47772_cast = einsum(equation = var_47772_equation_0, values = (var_47614_cast, var_47531_cast))[name = tensor<string, []>("op_47772_cast")];
            tensor<fp16, []> var_47773_to_fp16 = const()[name = tensor<string, []>("op_47773_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4073_cast = mul(x = var_47772_cast, y = var_47773_to_fp16)[name = tensor<string, []>("aw_4073_cast")];
            tensor<string, []> var_47776_equation_0 = const()[name = tensor<string, []>("op_47776_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47776_cast = einsum(equation = var_47776_equation_0, values = (var_47618_cast, var_47535_cast))[name = tensor<string, []>("op_47776_cast")];
            tensor<fp16, []> var_47777_to_fp16 = const()[name = tensor<string, []>("op_47777_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4075_cast = mul(x = var_47776_cast, y = var_47777_to_fp16)[name = tensor<string, []>("aw_4075_cast")];
            tensor<string, []> var_47780_equation_0 = const()[name = tensor<string, []>("op_47780_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47780_cast = einsum(equation = var_47780_equation_0, values = (var_47622_cast, var_47539_cast))[name = tensor<string, []>("op_47780_cast")];
            tensor<fp16, []> var_47781_to_fp16 = const()[name = tensor<string, []>("op_47781_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4077_cast = mul(x = var_47780_cast, y = var_47781_to_fp16)[name = tensor<string, []>("aw_4077_cast")];
            tensor<string, []> var_47784_equation_0 = const()[name = tensor<string, []>("op_47784_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_47784_cast = einsum(equation = var_47784_equation_0, values = (var_47626_cast, var_47543_cast))[name = tensor<string, []>("op_47784_cast")];
            tensor<fp16, []> var_47785_to_fp16 = const()[name = tensor<string, []>("op_47785_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4079_cast = mul(x = var_47784_cast, y = var_47785_to_fp16)[name = tensor<string, []>("aw_4079_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47787_cast = softmax(axis = var_30385, x = aw_4041_cast)[name = tensor<string, []>("op_47787_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47788_cast = softmax(axis = var_30385, x = aw_4043_cast)[name = tensor<string, []>("op_47788_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47789_cast = softmax(axis = var_30385, x = aw_4045_cast)[name = tensor<string, []>("op_47789_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47790_cast = softmax(axis = var_30385, x = aw_4047_cast)[name = tensor<string, []>("op_47790_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47791_cast = softmax(axis = var_30385, x = aw_4049_cast)[name = tensor<string, []>("op_47791_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47792_cast = softmax(axis = var_30385, x = aw_4051_cast)[name = tensor<string, []>("op_47792_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47793_cast = softmax(axis = var_30385, x = aw_4053_cast)[name = tensor<string, []>("op_47793_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47794_cast = softmax(axis = var_30385, x = aw_4055_cast)[name = tensor<string, []>("op_47794_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47795_cast = softmax(axis = var_30385, x = aw_4057_cast)[name = tensor<string, []>("op_47795_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47796_cast = softmax(axis = var_30385, x = aw_4059_cast)[name = tensor<string, []>("op_47796_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47797_cast = softmax(axis = var_30385, x = aw_4061_cast)[name = tensor<string, []>("op_47797_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47798_cast = softmax(axis = var_30385, x = aw_4063_cast)[name = tensor<string, []>("op_47798_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47799_cast = softmax(axis = var_30385, x = aw_4065_cast)[name = tensor<string, []>("op_47799_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47800_cast = softmax(axis = var_30385, x = aw_4067_cast)[name = tensor<string, []>("op_47800_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47801_cast = softmax(axis = var_30385, x = aw_4069_cast)[name = tensor<string, []>("op_47801_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47802_cast = softmax(axis = var_30385, x = aw_4071_cast)[name = tensor<string, []>("op_47802_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47803_cast = softmax(axis = var_30385, x = aw_4073_cast)[name = tensor<string, []>("op_47803_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47804_cast = softmax(axis = var_30385, x = aw_4075_cast)[name = tensor<string, []>("op_47804_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47805_cast = softmax(axis = var_30385, x = aw_4077_cast)[name = tensor<string, []>("op_47805_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_47806_cast = softmax(axis = var_30385, x = aw_4079_cast)[name = tensor<string, []>("op_47806_cast")];
            tensor<string, []> var_47808_equation_0 = const()[name = tensor<string, []>("op_47808_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47808_cast = einsum(equation = var_47808_equation_0, values = (var_47628_cast, var_47787_cast))[name = tensor<string, []>("op_47808_cast")];
            tensor<string, []> var_47810_equation_0 = const()[name = tensor<string, []>("op_47810_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47810_cast = einsum(equation = var_47810_equation_0, values = (var_47632_cast, var_47788_cast))[name = tensor<string, []>("op_47810_cast")];
            tensor<string, []> var_47812_equation_0 = const()[name = tensor<string, []>("op_47812_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47812_cast = einsum(equation = var_47812_equation_0, values = (var_47636_cast, var_47789_cast))[name = tensor<string, []>("op_47812_cast")];
            tensor<string, []> var_47814_equation_0 = const()[name = tensor<string, []>("op_47814_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47814_cast = einsum(equation = var_47814_equation_0, values = (var_47640_cast, var_47790_cast))[name = tensor<string, []>("op_47814_cast")];
            tensor<string, []> var_47816_equation_0 = const()[name = tensor<string, []>("op_47816_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47816_cast = einsum(equation = var_47816_equation_0, values = (var_47644_cast, var_47791_cast))[name = tensor<string, []>("op_47816_cast")];
            tensor<string, []> var_47818_equation_0 = const()[name = tensor<string, []>("op_47818_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47818_cast = einsum(equation = var_47818_equation_0, values = (var_47648_cast, var_47792_cast))[name = tensor<string, []>("op_47818_cast")];
            tensor<string, []> var_47820_equation_0 = const()[name = tensor<string, []>("op_47820_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47820_cast = einsum(equation = var_47820_equation_0, values = (var_47652_cast, var_47793_cast))[name = tensor<string, []>("op_47820_cast")];
            tensor<string, []> var_47822_equation_0 = const()[name = tensor<string, []>("op_47822_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47822_cast = einsum(equation = var_47822_equation_0, values = (var_47656_cast, var_47794_cast))[name = tensor<string, []>("op_47822_cast")];
            tensor<string, []> var_47824_equation_0 = const()[name = tensor<string, []>("op_47824_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47824_cast = einsum(equation = var_47824_equation_0, values = (var_47660_cast, var_47795_cast))[name = tensor<string, []>("op_47824_cast")];
            tensor<string, []> var_47826_equation_0 = const()[name = tensor<string, []>("op_47826_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47826_cast = einsum(equation = var_47826_equation_0, values = (var_47664_cast, var_47796_cast))[name = tensor<string, []>("op_47826_cast")];
            tensor<string, []> var_47828_equation_0 = const()[name = tensor<string, []>("op_47828_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47828_cast = einsum(equation = var_47828_equation_0, values = (var_47668_cast, var_47797_cast))[name = tensor<string, []>("op_47828_cast")];
            tensor<string, []> var_47830_equation_0 = const()[name = tensor<string, []>("op_47830_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47830_cast = einsum(equation = var_47830_equation_0, values = (var_47672_cast, var_47798_cast))[name = tensor<string, []>("op_47830_cast")];
            tensor<string, []> var_47832_equation_0 = const()[name = tensor<string, []>("op_47832_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47832_cast = einsum(equation = var_47832_equation_0, values = (var_47676_cast, var_47799_cast))[name = tensor<string, []>("op_47832_cast")];
            tensor<string, []> var_47834_equation_0 = const()[name = tensor<string, []>("op_47834_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47834_cast = einsum(equation = var_47834_equation_0, values = (var_47680_cast, var_47800_cast))[name = tensor<string, []>("op_47834_cast")];
            tensor<string, []> var_47836_equation_0 = const()[name = tensor<string, []>("op_47836_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47836_cast = einsum(equation = var_47836_equation_0, values = (var_47684_cast, var_47801_cast))[name = tensor<string, []>("op_47836_cast")];
            tensor<string, []> var_47838_equation_0 = const()[name = tensor<string, []>("op_47838_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47838_cast = einsum(equation = var_47838_equation_0, values = (var_47688_cast, var_47802_cast))[name = tensor<string, []>("op_47838_cast")];
            tensor<string, []> var_47840_equation_0 = const()[name = tensor<string, []>("op_47840_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47840_cast = einsum(equation = var_47840_equation_0, values = (var_47692_cast, var_47803_cast))[name = tensor<string, []>("op_47840_cast")];
            tensor<string, []> var_47842_equation_0 = const()[name = tensor<string, []>("op_47842_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47842_cast = einsum(equation = var_47842_equation_0, values = (var_47696_cast, var_47804_cast))[name = tensor<string, []>("op_47842_cast")];
            tensor<string, []> var_47844_equation_0 = const()[name = tensor<string, []>("op_47844_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47844_cast = einsum(equation = var_47844_equation_0, values = (var_47700_cast, var_47805_cast))[name = tensor<string, []>("op_47844_cast")];
            tensor<string, []> var_47846_equation_0 = const()[name = tensor<string, []>("op_47846_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_47846_cast = einsum(equation = var_47846_equation_0, values = (var_47704_cast, var_47806_cast))[name = tensor<string, []>("op_47846_cast")];
            tensor<bool, []> input_605_interleave_0 = const()[name = tensor<string, []>("input_605_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_605_cast = concat(axis = var_30385, interleave = input_605_interleave_0, values = (var_47808_cast, var_47810_cast, var_47812_cast, var_47814_cast, var_47816_cast, var_47818_cast, var_47820_cast, var_47822_cast, var_47824_cast, var_47826_cast, var_47828_cast, var_47830_cast, var_47832_cast, var_47834_cast, var_47836_cast, var_47838_cast, var_47840_cast, var_47842_cast, var_47844_cast, var_47846_cast))[name = tensor<string, []>("input_605_cast")];
            tensor<int32, [2]> var_47852 = const()[name = tensor<string, []>("op_47852"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_47854 = const()[name = tensor<string, []>("op_47854"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_47856_pad_type_0 = const()[name = tensor<string, []>("op_47856_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_47856_pad_0 = const()[name = tensor<string, []>("op_47856_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3971432576)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3974709440)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_47856_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_out_0_bias_to_fp16, dilations = var_47854, groups = var_30385, pad = var_47856_pad_0, pad_type = var_47856_pad_type_0, strides = var_47852, weight = up_blocks_0_attentions_1_transformer_blocks_8_attn2_to_out_0_weight_to_fp16, x = input_605_cast)[name = tensor<string, []>("op_47856_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_317_cast = add(x = var_47856_cast, y = inputs_315_cast)[name = tensor<string, []>("inputs_317_cast")];
            tensor<int32, [1]> var_47860 = const()[name = tensor<string, []>("op_47860"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_317_cast = reduce_mean(axes = var_47860, keep_dims = var_30380, x = inputs_317_cast)[name = tensor<string, []>("channels_mean_317_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_317_cast = sub(x = inputs_317_cast, y = channels_mean_317_cast)[name = tensor<string, []>("zero_mean_317_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_317_cast = mul(x = zero_mean_317_cast, y = zero_mean_317_cast)[name = tensor<string, []>("zero_mean_sq_317_cast")];
            tensor<int32, [1]> var_47864 = const()[name = tensor<string, []>("op_47864"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_47865_cast = reduce_mean(axes = var_47864, keep_dims = var_30380, x = zero_mean_sq_317_cast)[name = tensor<string, []>("op_47865_cast")];
            tensor<fp16, []> var_47866_to_fp16 = const()[name = tensor<string, []>("op_47866_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_47867_cast = add(x = var_47865_cast, y = var_47866_to_fp16)[name = tensor<string, []>("op_47867_cast")];
            tensor<fp16, []> denom_317_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_317_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_317_cast = rsqrt(epsilon = denom_317_epsilon_0_to_fp16, x = var_47867_cast)[name = tensor<string, []>("denom_317_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_317_cast = mul(x = zero_mean_317_cast, y = denom_317_cast)[name = tensor<string, []>("out_317_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_47871_to_fp16 = const()[name = tensor<string, []>("op_47871_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3974712064)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_47872_cast = add(x = out_317_cast, y = var_47871_to_fp16)[name = tensor<string, []>("op_47872_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_47874_to_fp16 = const()[name = tensor<string, []>("op_47874_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3974714688)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_607_cast = mul(x = var_47872_cast, y = var_47874_to_fp16)[name = tensor<string, []>("input_607_cast")];
            tensor<int32, [2]> var_47882 = const()[name = tensor<string, []>("op_47882"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_47884 = const()[name = tensor<string, []>("op_47884"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_47886_pad_type_0 = const()[name = tensor<string, []>("op_47886_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_47886_pad_0 = const()[name = tensor<string, []>("op_47886_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3974717312)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_1_transformer_blocks_8_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4000931776)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_47886_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_8_ff_net_0_proj_bias_to_fp16, dilations = var_47884, groups = var_30385, pad = var_47886_pad_0, pad_type = var_47886_pad_type_0, strides = var_47882, weight = up_blocks_0_attentions_1_transformer_blocks_8_ff_net_0_proj_weight_to_fp16, x = input_607_cast)[name = tensor<string, []>("op_47886_cast")];
            tensor<int32, [2]> var_47887_split_sizes_0 = const()[name = tensor<string, []>("op_47887_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_47887_axis_0 = const()[name = tensor<string, []>("op_47887_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_47887_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_47887_cast_1 = split(axis = var_47887_axis_0, split_sizes = var_47887_split_sizes_0, x = var_47886_cast)[name = tensor<string, []>("op_47887_cast")];
            tensor<string, []> var_47889_mode_0 = const()[name = tensor<string, []>("op_47889_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_47889_cast = gelu(mode = var_47889_mode_0, x = var_47887_cast_1)[name = tensor<string, []>("op_47889_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_609_cast = mul(x = var_47887_cast_0, y = var_47889_cast)[name = tensor<string, []>("input_609_cast")];
            tensor<int32, [2]> var_47893 = const()[name = tensor<string, []>("op_47893"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_47895 = const()[name = tensor<string, []>("op_47895"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_47897_pad_type_0 = const()[name = tensor<string, []>("op_47897_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_47897_pad_0 = const()[name = tensor<string, []>("op_47897_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_8_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4000952320)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_8_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_8_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4014059584)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_47897_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_8_ff_net_2_bias_to_fp16, dilations = var_47895, groups = var_30385, pad = var_47897_pad_0, pad_type = var_47897_pad_type_0, strides = var_47893, weight = up_blocks_0_attentions_1_transformer_blocks_8_ff_net_2_weight_to_fp16, x = input_609_cast)[name = tensor<string, []>("op_47897_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_319_cast = add(x = var_47897_cast, y = inputs_317_cast)[name = tensor<string, []>("inputs_319_cast")];
            tensor<int32, [1]> var_47907 = const()[name = tensor<string, []>("op_47907"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_319_cast = reduce_mean(axes = var_47907, keep_dims = var_30380, x = inputs_319_cast)[name = tensor<string, []>("channels_mean_319_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_319_cast = sub(x = inputs_319_cast, y = channels_mean_319_cast)[name = tensor<string, []>("zero_mean_319_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_319_cast = mul(x = zero_mean_319_cast, y = zero_mean_319_cast)[name = tensor<string, []>("zero_mean_sq_319_cast")];
            tensor<int32, [1]> var_47911 = const()[name = tensor<string, []>("op_47911"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_47912_cast = reduce_mean(axes = var_47911, keep_dims = var_30380, x = zero_mean_sq_319_cast)[name = tensor<string, []>("op_47912_cast")];
            tensor<fp16, []> var_47913_to_fp16 = const()[name = tensor<string, []>("op_47913_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_47914_cast = add(x = var_47912_cast, y = var_47913_to_fp16)[name = tensor<string, []>("op_47914_cast")];
            tensor<fp16, []> denom_319_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_319_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_319_cast = rsqrt(epsilon = denom_319_epsilon_0_to_fp16, x = var_47914_cast)[name = tensor<string, []>("denom_319_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_319_cast = mul(x = zero_mean_319_cast, y = denom_319_cast)[name = tensor<string, []>("out_319_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_47918_to_fp16 = const()[name = tensor<string, []>("op_47918_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4014062208)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_47919_cast = add(x = out_319_cast, y = var_47918_to_fp16)[name = tensor<string, []>("op_47919_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_47921_to_fp16 = const()[name = tensor<string, []>("op_47921_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4014064832)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_415_cast = mul(x = var_47919_cast, y = var_47921_to_fp16)[name = tensor<string, []>("hidden_states_415_cast")];
            tensor<int32, [2]> var_47928 = const()[name = tensor<string, []>("op_47928"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_47930 = const()[name = tensor<string, []>("op_47930"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_213_pad_type_0 = const()[name = tensor<string, []>("q_213_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_213_pad_0 = const()[name = tensor<string, []>("q_213_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4014067456)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_213_cast = conv(dilations = var_47930, groups = var_30385, pad = q_213_pad_0, pad_type = q_213_pad_type_0, strides = var_47928, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_q_weight_to_fp16, x = hidden_states_415_cast)[name = tensor<string, []>("q_213_cast")];
            tensor<int32, [2]> var_47934 = const()[name = tensor<string, []>("op_47934"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_47936 = const()[name = tensor<string, []>("op_47936"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_425_pad_type_0 = const()[name = tensor<string, []>("k_425_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_425_pad_0 = const()[name = tensor<string, []>("k_425_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4017344320)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_425_cast = conv(dilations = var_47936, groups = var_30385, pad = k_425_pad_0, pad_type = k_425_pad_type_0, strides = var_47934, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_k_weight_to_fp16, x = hidden_states_415_cast)[name = tensor<string, []>("k_425_cast")];
            tensor<int32, [2]> var_47940 = const()[name = tensor<string, []>("op_47940"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_47942 = const()[name = tensor<string, []>("op_47942"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_213_pad_type_0 = const()[name = tensor<string, []>("v_213_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_213_pad_0 = const()[name = tensor<string, []>("v_213_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4020621184)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_213_cast = conv(dilations = var_47942, groups = var_30385, pad = v_213_pad_0, pad_type = v_213_pad_type_0, strides = var_47940, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_v_weight_to_fp16, x = hidden_states_415_cast)[name = tensor<string, []>("v_213_cast")];
            tensor<int32, [4]> var_47946_begin_0 = const()[name = tensor<string, []>("op_47946_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_47946_end_0 = const()[name = tensor<string, []>("op_47946_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_47946_end_mask_0 = const()[name = tensor<string, []>("op_47946_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47946_cast = slice_by_index(begin = var_47946_begin_0, end = var_47946_end_0, end_mask = var_47946_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_47946_cast")];
            tensor<int32, [4]> var_47950_begin_0 = const()[name = tensor<string, []>("op_47950_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_47950_end_0 = const()[name = tensor<string, []>("op_47950_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_47950_end_mask_0 = const()[name = tensor<string, []>("op_47950_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47950_cast = slice_by_index(begin = var_47950_begin_0, end = var_47950_end_0, end_mask = var_47950_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_47950_cast")];
            tensor<int32, [4]> var_47954_begin_0 = const()[name = tensor<string, []>("op_47954_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_47954_end_0 = const()[name = tensor<string, []>("op_47954_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_47954_end_mask_0 = const()[name = tensor<string, []>("op_47954_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47954_cast = slice_by_index(begin = var_47954_begin_0, end = var_47954_end_0, end_mask = var_47954_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_47954_cast")];
            tensor<int32, [4]> var_47958_begin_0 = const()[name = tensor<string, []>("op_47958_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_47958_end_0 = const()[name = tensor<string, []>("op_47958_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_47958_end_mask_0 = const()[name = tensor<string, []>("op_47958_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47958_cast = slice_by_index(begin = var_47958_begin_0, end = var_47958_end_0, end_mask = var_47958_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_47958_cast")];
            tensor<int32, [4]> var_47962_begin_0 = const()[name = tensor<string, []>("op_47962_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_47962_end_0 = const()[name = tensor<string, []>("op_47962_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_47962_end_mask_0 = const()[name = tensor<string, []>("op_47962_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47962_cast = slice_by_index(begin = var_47962_begin_0, end = var_47962_end_0, end_mask = var_47962_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_47962_cast")];
            tensor<int32, [4]> var_47966_begin_0 = const()[name = tensor<string, []>("op_47966_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_47966_end_0 = const()[name = tensor<string, []>("op_47966_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_47966_end_mask_0 = const()[name = tensor<string, []>("op_47966_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47966_cast = slice_by_index(begin = var_47966_begin_0, end = var_47966_end_0, end_mask = var_47966_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_47966_cast")];
            tensor<int32, [4]> var_47970_begin_0 = const()[name = tensor<string, []>("op_47970_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_47970_end_0 = const()[name = tensor<string, []>("op_47970_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_47970_end_mask_0 = const()[name = tensor<string, []>("op_47970_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47970_cast = slice_by_index(begin = var_47970_begin_0, end = var_47970_end_0, end_mask = var_47970_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_47970_cast")];
            tensor<int32, [4]> var_47974_begin_0 = const()[name = tensor<string, []>("op_47974_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_47974_end_0 = const()[name = tensor<string, []>("op_47974_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_47974_end_mask_0 = const()[name = tensor<string, []>("op_47974_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47974_cast = slice_by_index(begin = var_47974_begin_0, end = var_47974_end_0, end_mask = var_47974_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_47974_cast")];
            tensor<int32, [4]> var_47978_begin_0 = const()[name = tensor<string, []>("op_47978_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_47978_end_0 = const()[name = tensor<string, []>("op_47978_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_47978_end_mask_0 = const()[name = tensor<string, []>("op_47978_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47978_cast = slice_by_index(begin = var_47978_begin_0, end = var_47978_end_0, end_mask = var_47978_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_47978_cast")];
            tensor<int32, [4]> var_47982_begin_0 = const()[name = tensor<string, []>("op_47982_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_47982_end_0 = const()[name = tensor<string, []>("op_47982_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_47982_end_mask_0 = const()[name = tensor<string, []>("op_47982_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47982_cast = slice_by_index(begin = var_47982_begin_0, end = var_47982_end_0, end_mask = var_47982_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_47982_cast")];
            tensor<int32, [4]> var_47986_begin_0 = const()[name = tensor<string, []>("op_47986_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_47986_end_0 = const()[name = tensor<string, []>("op_47986_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_47986_end_mask_0 = const()[name = tensor<string, []>("op_47986_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47986_cast = slice_by_index(begin = var_47986_begin_0, end = var_47986_end_0, end_mask = var_47986_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_47986_cast")];
            tensor<int32, [4]> var_47990_begin_0 = const()[name = tensor<string, []>("op_47990_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_47990_end_0 = const()[name = tensor<string, []>("op_47990_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_47990_end_mask_0 = const()[name = tensor<string, []>("op_47990_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47990_cast = slice_by_index(begin = var_47990_begin_0, end = var_47990_end_0, end_mask = var_47990_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_47990_cast")];
            tensor<int32, [4]> var_47994_begin_0 = const()[name = tensor<string, []>("op_47994_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_47994_end_0 = const()[name = tensor<string, []>("op_47994_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_47994_end_mask_0 = const()[name = tensor<string, []>("op_47994_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47994_cast = slice_by_index(begin = var_47994_begin_0, end = var_47994_end_0, end_mask = var_47994_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_47994_cast")];
            tensor<int32, [4]> var_47998_begin_0 = const()[name = tensor<string, []>("op_47998_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_47998_end_0 = const()[name = tensor<string, []>("op_47998_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_47998_end_mask_0 = const()[name = tensor<string, []>("op_47998_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_47998_cast = slice_by_index(begin = var_47998_begin_0, end = var_47998_end_0, end_mask = var_47998_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_47998_cast")];
            tensor<int32, [4]> var_48002_begin_0 = const()[name = tensor<string, []>("op_48002_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_48002_end_0 = const()[name = tensor<string, []>("op_48002_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_48002_end_mask_0 = const()[name = tensor<string, []>("op_48002_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48002_cast = slice_by_index(begin = var_48002_begin_0, end = var_48002_end_0, end_mask = var_48002_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_48002_cast")];
            tensor<int32, [4]> var_48006_begin_0 = const()[name = tensor<string, []>("op_48006_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_48006_end_0 = const()[name = tensor<string, []>("op_48006_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_48006_end_mask_0 = const()[name = tensor<string, []>("op_48006_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48006_cast = slice_by_index(begin = var_48006_begin_0, end = var_48006_end_0, end_mask = var_48006_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_48006_cast")];
            tensor<int32, [4]> var_48010_begin_0 = const()[name = tensor<string, []>("op_48010_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_48010_end_0 = const()[name = tensor<string, []>("op_48010_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_48010_end_mask_0 = const()[name = tensor<string, []>("op_48010_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48010_cast = slice_by_index(begin = var_48010_begin_0, end = var_48010_end_0, end_mask = var_48010_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_48010_cast")];
            tensor<int32, [4]> var_48014_begin_0 = const()[name = tensor<string, []>("op_48014_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_48014_end_0 = const()[name = tensor<string, []>("op_48014_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_48014_end_mask_0 = const()[name = tensor<string, []>("op_48014_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48014_cast = slice_by_index(begin = var_48014_begin_0, end = var_48014_end_0, end_mask = var_48014_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_48014_cast")];
            tensor<int32, [4]> var_48018_begin_0 = const()[name = tensor<string, []>("op_48018_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_48018_end_0 = const()[name = tensor<string, []>("op_48018_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_48018_end_mask_0 = const()[name = tensor<string, []>("op_48018_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48018_cast = slice_by_index(begin = var_48018_begin_0, end = var_48018_end_0, end_mask = var_48018_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_48018_cast")];
            tensor<int32, [4]> var_48022_begin_0 = const()[name = tensor<string, []>("op_48022_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_48022_end_0 = const()[name = tensor<string, []>("op_48022_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_48022_end_mask_0 = const()[name = tensor<string, []>("op_48022_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48022_cast = slice_by_index(begin = var_48022_begin_0, end = var_48022_end_0, end_mask = var_48022_end_mask_0, x = q_213_cast)[name = tensor<string, []>("op_48022_cast")];
            tensor<int32, [4]> k_427_perm_0 = const()[name = tensor<string, []>("k_427_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_48029_begin_0 = const()[name = tensor<string, []>("op_48029_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_48029_end_0 = const()[name = tensor<string, []>("op_48029_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_48029_end_mask_0 = const()[name = tensor<string, []>("op_48029_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_33 = transpose(perm = k_427_perm_0, x = k_425_cast)[name = tensor<string, []>("transpose_33")];
            tensor<fp16, [2, 1024, 1, 64]> var_48029_cast = slice_by_index(begin = var_48029_begin_0, end = var_48029_end_0, end_mask = var_48029_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48029_cast")];
            tensor<int32, [4]> var_48033_begin_0 = const()[name = tensor<string, []>("op_48033_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_48033_end_0 = const()[name = tensor<string, []>("op_48033_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_48033_end_mask_0 = const()[name = tensor<string, []>("op_48033_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48033_cast = slice_by_index(begin = var_48033_begin_0, end = var_48033_end_0, end_mask = var_48033_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48033_cast")];
            tensor<int32, [4]> var_48037_begin_0 = const()[name = tensor<string, []>("op_48037_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_48037_end_0 = const()[name = tensor<string, []>("op_48037_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_48037_end_mask_0 = const()[name = tensor<string, []>("op_48037_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48037_cast = slice_by_index(begin = var_48037_begin_0, end = var_48037_end_0, end_mask = var_48037_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48037_cast")];
            tensor<int32, [4]> var_48041_begin_0 = const()[name = tensor<string, []>("op_48041_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_48041_end_0 = const()[name = tensor<string, []>("op_48041_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_48041_end_mask_0 = const()[name = tensor<string, []>("op_48041_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48041_cast = slice_by_index(begin = var_48041_begin_0, end = var_48041_end_0, end_mask = var_48041_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48041_cast")];
            tensor<int32, [4]> var_48045_begin_0 = const()[name = tensor<string, []>("op_48045_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_48045_end_0 = const()[name = tensor<string, []>("op_48045_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_48045_end_mask_0 = const()[name = tensor<string, []>("op_48045_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48045_cast = slice_by_index(begin = var_48045_begin_0, end = var_48045_end_0, end_mask = var_48045_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48045_cast")];
            tensor<int32, [4]> var_48049_begin_0 = const()[name = tensor<string, []>("op_48049_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_48049_end_0 = const()[name = tensor<string, []>("op_48049_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_48049_end_mask_0 = const()[name = tensor<string, []>("op_48049_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48049_cast = slice_by_index(begin = var_48049_begin_0, end = var_48049_end_0, end_mask = var_48049_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48049_cast")];
            tensor<int32, [4]> var_48053_begin_0 = const()[name = tensor<string, []>("op_48053_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_48053_end_0 = const()[name = tensor<string, []>("op_48053_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_48053_end_mask_0 = const()[name = tensor<string, []>("op_48053_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48053_cast = slice_by_index(begin = var_48053_begin_0, end = var_48053_end_0, end_mask = var_48053_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48053_cast")];
            tensor<int32, [4]> var_48057_begin_0 = const()[name = tensor<string, []>("op_48057_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_48057_end_0 = const()[name = tensor<string, []>("op_48057_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_48057_end_mask_0 = const()[name = tensor<string, []>("op_48057_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48057_cast = slice_by_index(begin = var_48057_begin_0, end = var_48057_end_0, end_mask = var_48057_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48057_cast")];
            tensor<int32, [4]> var_48061_begin_0 = const()[name = tensor<string, []>("op_48061_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_48061_end_0 = const()[name = tensor<string, []>("op_48061_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_48061_end_mask_0 = const()[name = tensor<string, []>("op_48061_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48061_cast = slice_by_index(begin = var_48061_begin_0, end = var_48061_end_0, end_mask = var_48061_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48061_cast")];
            tensor<int32, [4]> var_48065_begin_0 = const()[name = tensor<string, []>("op_48065_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_48065_end_0 = const()[name = tensor<string, []>("op_48065_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_48065_end_mask_0 = const()[name = tensor<string, []>("op_48065_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48065_cast = slice_by_index(begin = var_48065_begin_0, end = var_48065_end_0, end_mask = var_48065_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48065_cast")];
            tensor<int32, [4]> var_48069_begin_0 = const()[name = tensor<string, []>("op_48069_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_48069_end_0 = const()[name = tensor<string, []>("op_48069_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_48069_end_mask_0 = const()[name = tensor<string, []>("op_48069_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48069_cast = slice_by_index(begin = var_48069_begin_0, end = var_48069_end_0, end_mask = var_48069_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48069_cast")];
            tensor<int32, [4]> var_48073_begin_0 = const()[name = tensor<string, []>("op_48073_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_48073_end_0 = const()[name = tensor<string, []>("op_48073_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_48073_end_mask_0 = const()[name = tensor<string, []>("op_48073_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48073_cast = slice_by_index(begin = var_48073_begin_0, end = var_48073_end_0, end_mask = var_48073_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48073_cast")];
            tensor<int32, [4]> var_48077_begin_0 = const()[name = tensor<string, []>("op_48077_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_48077_end_0 = const()[name = tensor<string, []>("op_48077_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_48077_end_mask_0 = const()[name = tensor<string, []>("op_48077_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48077_cast = slice_by_index(begin = var_48077_begin_0, end = var_48077_end_0, end_mask = var_48077_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48077_cast")];
            tensor<int32, [4]> var_48081_begin_0 = const()[name = tensor<string, []>("op_48081_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_48081_end_0 = const()[name = tensor<string, []>("op_48081_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_48081_end_mask_0 = const()[name = tensor<string, []>("op_48081_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48081_cast = slice_by_index(begin = var_48081_begin_0, end = var_48081_end_0, end_mask = var_48081_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48081_cast")];
            tensor<int32, [4]> var_48085_begin_0 = const()[name = tensor<string, []>("op_48085_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_48085_end_0 = const()[name = tensor<string, []>("op_48085_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_48085_end_mask_0 = const()[name = tensor<string, []>("op_48085_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48085_cast = slice_by_index(begin = var_48085_begin_0, end = var_48085_end_0, end_mask = var_48085_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48085_cast")];
            tensor<int32, [4]> var_48089_begin_0 = const()[name = tensor<string, []>("op_48089_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_48089_end_0 = const()[name = tensor<string, []>("op_48089_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_48089_end_mask_0 = const()[name = tensor<string, []>("op_48089_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48089_cast = slice_by_index(begin = var_48089_begin_0, end = var_48089_end_0, end_mask = var_48089_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48089_cast")];
            tensor<int32, [4]> var_48093_begin_0 = const()[name = tensor<string, []>("op_48093_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_48093_end_0 = const()[name = tensor<string, []>("op_48093_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_48093_end_mask_0 = const()[name = tensor<string, []>("op_48093_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48093_cast = slice_by_index(begin = var_48093_begin_0, end = var_48093_end_0, end_mask = var_48093_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48093_cast")];
            tensor<int32, [4]> var_48097_begin_0 = const()[name = tensor<string, []>("op_48097_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_48097_end_0 = const()[name = tensor<string, []>("op_48097_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_48097_end_mask_0 = const()[name = tensor<string, []>("op_48097_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48097_cast = slice_by_index(begin = var_48097_begin_0, end = var_48097_end_0, end_mask = var_48097_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48097_cast")];
            tensor<int32, [4]> var_48101_begin_0 = const()[name = tensor<string, []>("op_48101_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_48101_end_0 = const()[name = tensor<string, []>("op_48101_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_48101_end_mask_0 = const()[name = tensor<string, []>("op_48101_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48101_cast = slice_by_index(begin = var_48101_begin_0, end = var_48101_end_0, end_mask = var_48101_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48101_cast")];
            tensor<int32, [4]> var_48105_begin_0 = const()[name = tensor<string, []>("op_48105_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_48105_end_0 = const()[name = tensor<string, []>("op_48105_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_48105_end_mask_0 = const()[name = tensor<string, []>("op_48105_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_48105_cast = slice_by_index(begin = var_48105_begin_0, end = var_48105_end_0, end_mask = var_48105_end_mask_0, x = transpose_33)[name = tensor<string, []>("op_48105_cast")];
            tensor<int32, [4]> var_48107_begin_0 = const()[name = tensor<string, []>("op_48107_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_48107_end_0 = const()[name = tensor<string, []>("op_48107_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_48107_end_mask_0 = const()[name = tensor<string, []>("op_48107_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48107_cast = slice_by_index(begin = var_48107_begin_0, end = var_48107_end_0, end_mask = var_48107_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48107_cast")];
            tensor<int32, [4]> var_48111_begin_0 = const()[name = tensor<string, []>("op_48111_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_48111_end_0 = const()[name = tensor<string, []>("op_48111_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_48111_end_mask_0 = const()[name = tensor<string, []>("op_48111_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48111_cast = slice_by_index(begin = var_48111_begin_0, end = var_48111_end_0, end_mask = var_48111_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48111_cast")];
            tensor<int32, [4]> var_48115_begin_0 = const()[name = tensor<string, []>("op_48115_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_48115_end_0 = const()[name = tensor<string, []>("op_48115_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_48115_end_mask_0 = const()[name = tensor<string, []>("op_48115_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48115_cast = slice_by_index(begin = var_48115_begin_0, end = var_48115_end_0, end_mask = var_48115_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48115_cast")];
            tensor<int32, [4]> var_48119_begin_0 = const()[name = tensor<string, []>("op_48119_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_48119_end_0 = const()[name = tensor<string, []>("op_48119_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_48119_end_mask_0 = const()[name = tensor<string, []>("op_48119_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48119_cast = slice_by_index(begin = var_48119_begin_0, end = var_48119_end_0, end_mask = var_48119_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48119_cast")];
            tensor<int32, [4]> var_48123_begin_0 = const()[name = tensor<string, []>("op_48123_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_48123_end_0 = const()[name = tensor<string, []>("op_48123_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_48123_end_mask_0 = const()[name = tensor<string, []>("op_48123_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48123_cast = slice_by_index(begin = var_48123_begin_0, end = var_48123_end_0, end_mask = var_48123_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48123_cast")];
            tensor<int32, [4]> var_48127_begin_0 = const()[name = tensor<string, []>("op_48127_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_48127_end_0 = const()[name = tensor<string, []>("op_48127_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_48127_end_mask_0 = const()[name = tensor<string, []>("op_48127_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48127_cast = slice_by_index(begin = var_48127_begin_0, end = var_48127_end_0, end_mask = var_48127_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48127_cast")];
            tensor<int32, [4]> var_48131_begin_0 = const()[name = tensor<string, []>("op_48131_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_48131_end_0 = const()[name = tensor<string, []>("op_48131_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_48131_end_mask_0 = const()[name = tensor<string, []>("op_48131_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48131_cast = slice_by_index(begin = var_48131_begin_0, end = var_48131_end_0, end_mask = var_48131_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48131_cast")];
            tensor<int32, [4]> var_48135_begin_0 = const()[name = tensor<string, []>("op_48135_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_48135_end_0 = const()[name = tensor<string, []>("op_48135_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_48135_end_mask_0 = const()[name = tensor<string, []>("op_48135_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48135_cast = slice_by_index(begin = var_48135_begin_0, end = var_48135_end_0, end_mask = var_48135_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48135_cast")];
            tensor<int32, [4]> var_48139_begin_0 = const()[name = tensor<string, []>("op_48139_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_48139_end_0 = const()[name = tensor<string, []>("op_48139_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_48139_end_mask_0 = const()[name = tensor<string, []>("op_48139_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48139_cast = slice_by_index(begin = var_48139_begin_0, end = var_48139_end_0, end_mask = var_48139_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48139_cast")];
            tensor<int32, [4]> var_48143_begin_0 = const()[name = tensor<string, []>("op_48143_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_48143_end_0 = const()[name = tensor<string, []>("op_48143_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_48143_end_mask_0 = const()[name = tensor<string, []>("op_48143_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48143_cast = slice_by_index(begin = var_48143_begin_0, end = var_48143_end_0, end_mask = var_48143_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48143_cast")];
            tensor<int32, [4]> var_48147_begin_0 = const()[name = tensor<string, []>("op_48147_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_48147_end_0 = const()[name = tensor<string, []>("op_48147_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_48147_end_mask_0 = const()[name = tensor<string, []>("op_48147_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48147_cast = slice_by_index(begin = var_48147_begin_0, end = var_48147_end_0, end_mask = var_48147_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48147_cast")];
            tensor<int32, [4]> var_48151_begin_0 = const()[name = tensor<string, []>("op_48151_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_48151_end_0 = const()[name = tensor<string, []>("op_48151_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_48151_end_mask_0 = const()[name = tensor<string, []>("op_48151_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48151_cast = slice_by_index(begin = var_48151_begin_0, end = var_48151_end_0, end_mask = var_48151_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48151_cast")];
            tensor<int32, [4]> var_48155_begin_0 = const()[name = tensor<string, []>("op_48155_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_48155_end_0 = const()[name = tensor<string, []>("op_48155_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_48155_end_mask_0 = const()[name = tensor<string, []>("op_48155_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48155_cast = slice_by_index(begin = var_48155_begin_0, end = var_48155_end_0, end_mask = var_48155_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48155_cast")];
            tensor<int32, [4]> var_48159_begin_0 = const()[name = tensor<string, []>("op_48159_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_48159_end_0 = const()[name = tensor<string, []>("op_48159_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_48159_end_mask_0 = const()[name = tensor<string, []>("op_48159_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48159_cast = slice_by_index(begin = var_48159_begin_0, end = var_48159_end_0, end_mask = var_48159_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48159_cast")];
            tensor<int32, [4]> var_48163_begin_0 = const()[name = tensor<string, []>("op_48163_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_48163_end_0 = const()[name = tensor<string, []>("op_48163_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_48163_end_mask_0 = const()[name = tensor<string, []>("op_48163_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48163_cast = slice_by_index(begin = var_48163_begin_0, end = var_48163_end_0, end_mask = var_48163_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48163_cast")];
            tensor<int32, [4]> var_48167_begin_0 = const()[name = tensor<string, []>("op_48167_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_48167_end_0 = const()[name = tensor<string, []>("op_48167_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_48167_end_mask_0 = const()[name = tensor<string, []>("op_48167_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48167_cast = slice_by_index(begin = var_48167_begin_0, end = var_48167_end_0, end_mask = var_48167_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48167_cast")];
            tensor<int32, [4]> var_48171_begin_0 = const()[name = tensor<string, []>("op_48171_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_48171_end_0 = const()[name = tensor<string, []>("op_48171_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_48171_end_mask_0 = const()[name = tensor<string, []>("op_48171_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48171_cast = slice_by_index(begin = var_48171_begin_0, end = var_48171_end_0, end_mask = var_48171_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48171_cast")];
            tensor<int32, [4]> var_48175_begin_0 = const()[name = tensor<string, []>("op_48175_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_48175_end_0 = const()[name = tensor<string, []>("op_48175_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_48175_end_mask_0 = const()[name = tensor<string, []>("op_48175_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48175_cast = slice_by_index(begin = var_48175_begin_0, end = var_48175_end_0, end_mask = var_48175_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48175_cast")];
            tensor<int32, [4]> var_48179_begin_0 = const()[name = tensor<string, []>("op_48179_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_48179_end_0 = const()[name = tensor<string, []>("op_48179_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_48179_end_mask_0 = const()[name = tensor<string, []>("op_48179_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48179_cast = slice_by_index(begin = var_48179_begin_0, end = var_48179_end_0, end_mask = var_48179_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48179_cast")];
            tensor<int32, [4]> var_48183_begin_0 = const()[name = tensor<string, []>("op_48183_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_48183_end_0 = const()[name = tensor<string, []>("op_48183_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_48183_end_mask_0 = const()[name = tensor<string, []>("op_48183_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48183_cast = slice_by_index(begin = var_48183_begin_0, end = var_48183_end_0, end_mask = var_48183_end_mask_0, x = v_213_cast)[name = tensor<string, []>("op_48183_cast")];
            tensor<string, []> var_48187_equation_0 = const()[name = tensor<string, []>("op_48187_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48187_cast = einsum(equation = var_48187_equation_0, values = (var_48029_cast, var_47946_cast))[name = tensor<string, []>("op_48187_cast")];
            tensor<fp16, []> var_48188_to_fp16 = const()[name = tensor<string, []>("op_48188_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4081_cast = mul(x = var_48187_cast, y = var_48188_to_fp16)[name = tensor<string, []>("aw_4081_cast")];
            tensor<string, []> var_48191_equation_0 = const()[name = tensor<string, []>("op_48191_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48191_cast = einsum(equation = var_48191_equation_0, values = (var_48033_cast, var_47950_cast))[name = tensor<string, []>("op_48191_cast")];
            tensor<fp16, []> var_48192_to_fp16 = const()[name = tensor<string, []>("op_48192_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4083_cast = mul(x = var_48191_cast, y = var_48192_to_fp16)[name = tensor<string, []>("aw_4083_cast")];
            tensor<string, []> var_48195_equation_0 = const()[name = tensor<string, []>("op_48195_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48195_cast = einsum(equation = var_48195_equation_0, values = (var_48037_cast, var_47954_cast))[name = tensor<string, []>("op_48195_cast")];
            tensor<fp16, []> var_48196_to_fp16 = const()[name = tensor<string, []>("op_48196_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4085_cast = mul(x = var_48195_cast, y = var_48196_to_fp16)[name = tensor<string, []>("aw_4085_cast")];
            tensor<string, []> var_48199_equation_0 = const()[name = tensor<string, []>("op_48199_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48199_cast = einsum(equation = var_48199_equation_0, values = (var_48041_cast, var_47958_cast))[name = tensor<string, []>("op_48199_cast")];
            tensor<fp16, []> var_48200_to_fp16 = const()[name = tensor<string, []>("op_48200_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4087_cast = mul(x = var_48199_cast, y = var_48200_to_fp16)[name = tensor<string, []>("aw_4087_cast")];
            tensor<string, []> var_48203_equation_0 = const()[name = tensor<string, []>("op_48203_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48203_cast = einsum(equation = var_48203_equation_0, values = (var_48045_cast, var_47962_cast))[name = tensor<string, []>("op_48203_cast")];
            tensor<fp16, []> var_48204_to_fp16 = const()[name = tensor<string, []>("op_48204_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4089_cast = mul(x = var_48203_cast, y = var_48204_to_fp16)[name = tensor<string, []>("aw_4089_cast")];
            tensor<string, []> var_48207_equation_0 = const()[name = tensor<string, []>("op_48207_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48207_cast = einsum(equation = var_48207_equation_0, values = (var_48049_cast, var_47966_cast))[name = tensor<string, []>("op_48207_cast")];
            tensor<fp16, []> var_48208_to_fp16 = const()[name = tensor<string, []>("op_48208_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4091_cast = mul(x = var_48207_cast, y = var_48208_to_fp16)[name = tensor<string, []>("aw_4091_cast")];
            tensor<string, []> var_48211_equation_0 = const()[name = tensor<string, []>("op_48211_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48211_cast = einsum(equation = var_48211_equation_0, values = (var_48053_cast, var_47970_cast))[name = tensor<string, []>("op_48211_cast")];
            tensor<fp16, []> var_48212_to_fp16 = const()[name = tensor<string, []>("op_48212_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4093_cast = mul(x = var_48211_cast, y = var_48212_to_fp16)[name = tensor<string, []>("aw_4093_cast")];
            tensor<string, []> var_48215_equation_0 = const()[name = tensor<string, []>("op_48215_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48215_cast = einsum(equation = var_48215_equation_0, values = (var_48057_cast, var_47974_cast))[name = tensor<string, []>("op_48215_cast")];
            tensor<fp16, []> var_48216_to_fp16 = const()[name = tensor<string, []>("op_48216_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4095_cast = mul(x = var_48215_cast, y = var_48216_to_fp16)[name = tensor<string, []>("aw_4095_cast")];
            tensor<string, []> var_48219_equation_0 = const()[name = tensor<string, []>("op_48219_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48219_cast = einsum(equation = var_48219_equation_0, values = (var_48061_cast, var_47978_cast))[name = tensor<string, []>("op_48219_cast")];
            tensor<fp16, []> var_48220_to_fp16 = const()[name = tensor<string, []>("op_48220_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4097_cast = mul(x = var_48219_cast, y = var_48220_to_fp16)[name = tensor<string, []>("aw_4097_cast")];
            tensor<string, []> var_48223_equation_0 = const()[name = tensor<string, []>("op_48223_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48223_cast = einsum(equation = var_48223_equation_0, values = (var_48065_cast, var_47982_cast))[name = tensor<string, []>("op_48223_cast")];
            tensor<fp16, []> var_48224_to_fp16 = const()[name = tensor<string, []>("op_48224_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4099_cast = mul(x = var_48223_cast, y = var_48224_to_fp16)[name = tensor<string, []>("aw_4099_cast")];
            tensor<string, []> var_48227_equation_0 = const()[name = tensor<string, []>("op_48227_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48227_cast = einsum(equation = var_48227_equation_0, values = (var_48069_cast, var_47986_cast))[name = tensor<string, []>("op_48227_cast")];
            tensor<fp16, []> var_48228_to_fp16 = const()[name = tensor<string, []>("op_48228_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4101_cast = mul(x = var_48227_cast, y = var_48228_to_fp16)[name = tensor<string, []>("aw_4101_cast")];
            tensor<string, []> var_48231_equation_0 = const()[name = tensor<string, []>("op_48231_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48231_cast = einsum(equation = var_48231_equation_0, values = (var_48073_cast, var_47990_cast))[name = tensor<string, []>("op_48231_cast")];
            tensor<fp16, []> var_48232_to_fp16 = const()[name = tensor<string, []>("op_48232_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4103_cast = mul(x = var_48231_cast, y = var_48232_to_fp16)[name = tensor<string, []>("aw_4103_cast")];
            tensor<string, []> var_48235_equation_0 = const()[name = tensor<string, []>("op_48235_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48235_cast = einsum(equation = var_48235_equation_0, values = (var_48077_cast, var_47994_cast))[name = tensor<string, []>("op_48235_cast")];
            tensor<fp16, []> var_48236_to_fp16 = const()[name = tensor<string, []>("op_48236_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4105_cast = mul(x = var_48235_cast, y = var_48236_to_fp16)[name = tensor<string, []>("aw_4105_cast")];
            tensor<string, []> var_48239_equation_0 = const()[name = tensor<string, []>("op_48239_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48239_cast = einsum(equation = var_48239_equation_0, values = (var_48081_cast, var_47998_cast))[name = tensor<string, []>("op_48239_cast")];
            tensor<fp16, []> var_48240_to_fp16 = const()[name = tensor<string, []>("op_48240_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4107_cast = mul(x = var_48239_cast, y = var_48240_to_fp16)[name = tensor<string, []>("aw_4107_cast")];
            tensor<string, []> var_48243_equation_0 = const()[name = tensor<string, []>("op_48243_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48243_cast = einsum(equation = var_48243_equation_0, values = (var_48085_cast, var_48002_cast))[name = tensor<string, []>("op_48243_cast")];
            tensor<fp16, []> var_48244_to_fp16 = const()[name = tensor<string, []>("op_48244_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4109_cast = mul(x = var_48243_cast, y = var_48244_to_fp16)[name = tensor<string, []>("aw_4109_cast")];
            tensor<string, []> var_48247_equation_0 = const()[name = tensor<string, []>("op_48247_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48247_cast = einsum(equation = var_48247_equation_0, values = (var_48089_cast, var_48006_cast))[name = tensor<string, []>("op_48247_cast")];
            tensor<fp16, []> var_48248_to_fp16 = const()[name = tensor<string, []>("op_48248_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4111_cast = mul(x = var_48247_cast, y = var_48248_to_fp16)[name = tensor<string, []>("aw_4111_cast")];
            tensor<string, []> var_48251_equation_0 = const()[name = tensor<string, []>("op_48251_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48251_cast = einsum(equation = var_48251_equation_0, values = (var_48093_cast, var_48010_cast))[name = tensor<string, []>("op_48251_cast")];
            tensor<fp16, []> var_48252_to_fp16 = const()[name = tensor<string, []>("op_48252_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4113_cast = mul(x = var_48251_cast, y = var_48252_to_fp16)[name = tensor<string, []>("aw_4113_cast")];
            tensor<string, []> var_48255_equation_0 = const()[name = tensor<string, []>("op_48255_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48255_cast = einsum(equation = var_48255_equation_0, values = (var_48097_cast, var_48014_cast))[name = tensor<string, []>("op_48255_cast")];
            tensor<fp16, []> var_48256_to_fp16 = const()[name = tensor<string, []>("op_48256_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4115_cast = mul(x = var_48255_cast, y = var_48256_to_fp16)[name = tensor<string, []>("aw_4115_cast")];
            tensor<string, []> var_48259_equation_0 = const()[name = tensor<string, []>("op_48259_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48259_cast = einsum(equation = var_48259_equation_0, values = (var_48101_cast, var_48018_cast))[name = tensor<string, []>("op_48259_cast")];
            tensor<fp16, []> var_48260_to_fp16 = const()[name = tensor<string, []>("op_48260_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4117_cast = mul(x = var_48259_cast, y = var_48260_to_fp16)[name = tensor<string, []>("aw_4117_cast")];
            tensor<string, []> var_48263_equation_0 = const()[name = tensor<string, []>("op_48263_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48263_cast = einsum(equation = var_48263_equation_0, values = (var_48105_cast, var_48022_cast))[name = tensor<string, []>("op_48263_cast")];
            tensor<fp16, []> var_48264_to_fp16 = const()[name = tensor<string, []>("op_48264_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4119_cast = mul(x = var_48263_cast, y = var_48264_to_fp16)[name = tensor<string, []>("aw_4119_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48266_cast = softmax(axis = var_30385, x = aw_4081_cast)[name = tensor<string, []>("op_48266_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48267_cast = softmax(axis = var_30385, x = aw_4083_cast)[name = tensor<string, []>("op_48267_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48268_cast = softmax(axis = var_30385, x = aw_4085_cast)[name = tensor<string, []>("op_48268_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48269_cast = softmax(axis = var_30385, x = aw_4087_cast)[name = tensor<string, []>("op_48269_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48270_cast = softmax(axis = var_30385, x = aw_4089_cast)[name = tensor<string, []>("op_48270_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48271_cast = softmax(axis = var_30385, x = aw_4091_cast)[name = tensor<string, []>("op_48271_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48272_cast = softmax(axis = var_30385, x = aw_4093_cast)[name = tensor<string, []>("op_48272_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48273_cast = softmax(axis = var_30385, x = aw_4095_cast)[name = tensor<string, []>("op_48273_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48274_cast = softmax(axis = var_30385, x = aw_4097_cast)[name = tensor<string, []>("op_48274_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48275_cast = softmax(axis = var_30385, x = aw_4099_cast)[name = tensor<string, []>("op_48275_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48276_cast = softmax(axis = var_30385, x = aw_4101_cast)[name = tensor<string, []>("op_48276_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48277_cast = softmax(axis = var_30385, x = aw_4103_cast)[name = tensor<string, []>("op_48277_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48278_cast = softmax(axis = var_30385, x = aw_4105_cast)[name = tensor<string, []>("op_48278_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48279_cast = softmax(axis = var_30385, x = aw_4107_cast)[name = tensor<string, []>("op_48279_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48280_cast = softmax(axis = var_30385, x = aw_4109_cast)[name = tensor<string, []>("op_48280_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48281_cast = softmax(axis = var_30385, x = aw_4111_cast)[name = tensor<string, []>("op_48281_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48282_cast = softmax(axis = var_30385, x = aw_4113_cast)[name = tensor<string, []>("op_48282_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48283_cast = softmax(axis = var_30385, x = aw_4115_cast)[name = tensor<string, []>("op_48283_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48284_cast = softmax(axis = var_30385, x = aw_4117_cast)[name = tensor<string, []>("op_48284_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_48285_cast = softmax(axis = var_30385, x = aw_4119_cast)[name = tensor<string, []>("op_48285_cast")];
            tensor<string, []> var_48287_equation_0 = const()[name = tensor<string, []>("op_48287_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48287_cast = einsum(equation = var_48287_equation_0, values = (var_48107_cast, var_48266_cast))[name = tensor<string, []>("op_48287_cast")];
            tensor<string, []> var_48289_equation_0 = const()[name = tensor<string, []>("op_48289_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48289_cast = einsum(equation = var_48289_equation_0, values = (var_48111_cast, var_48267_cast))[name = tensor<string, []>("op_48289_cast")];
            tensor<string, []> var_48291_equation_0 = const()[name = tensor<string, []>("op_48291_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48291_cast = einsum(equation = var_48291_equation_0, values = (var_48115_cast, var_48268_cast))[name = tensor<string, []>("op_48291_cast")];
            tensor<string, []> var_48293_equation_0 = const()[name = tensor<string, []>("op_48293_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48293_cast = einsum(equation = var_48293_equation_0, values = (var_48119_cast, var_48269_cast))[name = tensor<string, []>("op_48293_cast")];
            tensor<string, []> var_48295_equation_0 = const()[name = tensor<string, []>("op_48295_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48295_cast = einsum(equation = var_48295_equation_0, values = (var_48123_cast, var_48270_cast))[name = tensor<string, []>("op_48295_cast")];
            tensor<string, []> var_48297_equation_0 = const()[name = tensor<string, []>("op_48297_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48297_cast = einsum(equation = var_48297_equation_0, values = (var_48127_cast, var_48271_cast))[name = tensor<string, []>("op_48297_cast")];
            tensor<string, []> var_48299_equation_0 = const()[name = tensor<string, []>("op_48299_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48299_cast = einsum(equation = var_48299_equation_0, values = (var_48131_cast, var_48272_cast))[name = tensor<string, []>("op_48299_cast")];
            tensor<string, []> var_48301_equation_0 = const()[name = tensor<string, []>("op_48301_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48301_cast = einsum(equation = var_48301_equation_0, values = (var_48135_cast, var_48273_cast))[name = tensor<string, []>("op_48301_cast")];
            tensor<string, []> var_48303_equation_0 = const()[name = tensor<string, []>("op_48303_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48303_cast = einsum(equation = var_48303_equation_0, values = (var_48139_cast, var_48274_cast))[name = tensor<string, []>("op_48303_cast")];
            tensor<string, []> var_48305_equation_0 = const()[name = tensor<string, []>("op_48305_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48305_cast = einsum(equation = var_48305_equation_0, values = (var_48143_cast, var_48275_cast))[name = tensor<string, []>("op_48305_cast")];
            tensor<string, []> var_48307_equation_0 = const()[name = tensor<string, []>("op_48307_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48307_cast = einsum(equation = var_48307_equation_0, values = (var_48147_cast, var_48276_cast))[name = tensor<string, []>("op_48307_cast")];
            tensor<string, []> var_48309_equation_0 = const()[name = tensor<string, []>("op_48309_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48309_cast = einsum(equation = var_48309_equation_0, values = (var_48151_cast, var_48277_cast))[name = tensor<string, []>("op_48309_cast")];
            tensor<string, []> var_48311_equation_0 = const()[name = tensor<string, []>("op_48311_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48311_cast = einsum(equation = var_48311_equation_0, values = (var_48155_cast, var_48278_cast))[name = tensor<string, []>("op_48311_cast")];
            tensor<string, []> var_48313_equation_0 = const()[name = tensor<string, []>("op_48313_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48313_cast = einsum(equation = var_48313_equation_0, values = (var_48159_cast, var_48279_cast))[name = tensor<string, []>("op_48313_cast")];
            tensor<string, []> var_48315_equation_0 = const()[name = tensor<string, []>("op_48315_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48315_cast = einsum(equation = var_48315_equation_0, values = (var_48163_cast, var_48280_cast))[name = tensor<string, []>("op_48315_cast")];
            tensor<string, []> var_48317_equation_0 = const()[name = tensor<string, []>("op_48317_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48317_cast = einsum(equation = var_48317_equation_0, values = (var_48167_cast, var_48281_cast))[name = tensor<string, []>("op_48317_cast")];
            tensor<string, []> var_48319_equation_0 = const()[name = tensor<string, []>("op_48319_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48319_cast = einsum(equation = var_48319_equation_0, values = (var_48171_cast, var_48282_cast))[name = tensor<string, []>("op_48319_cast")];
            tensor<string, []> var_48321_equation_0 = const()[name = tensor<string, []>("op_48321_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48321_cast = einsum(equation = var_48321_equation_0, values = (var_48175_cast, var_48283_cast))[name = tensor<string, []>("op_48321_cast")];
            tensor<string, []> var_48323_equation_0 = const()[name = tensor<string, []>("op_48323_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48323_cast = einsum(equation = var_48323_equation_0, values = (var_48179_cast, var_48284_cast))[name = tensor<string, []>("op_48323_cast")];
            tensor<string, []> var_48325_equation_0 = const()[name = tensor<string, []>("op_48325_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48325_cast = einsum(equation = var_48325_equation_0, values = (var_48183_cast, var_48285_cast))[name = tensor<string, []>("op_48325_cast")];
            tensor<bool, []> input_611_interleave_0 = const()[name = tensor<string, []>("input_611_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_611_cast = concat(axis = var_30385, interleave = input_611_interleave_0, values = (var_48287_cast, var_48289_cast, var_48291_cast, var_48293_cast, var_48295_cast, var_48297_cast, var_48299_cast, var_48301_cast, var_48303_cast, var_48305_cast, var_48307_cast, var_48309_cast, var_48311_cast, var_48313_cast, var_48315_cast, var_48317_cast, var_48319_cast, var_48321_cast, var_48323_cast, var_48325_cast))[name = tensor<string, []>("input_611_cast")];
            tensor<int32, [2]> var_48331 = const()[name = tensor<string, []>("op_48331"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_48333 = const()[name = tensor<string, []>("op_48333"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_48335_pad_type_0 = const()[name = tensor<string, []>("op_48335_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_48335_pad_0 = const()[name = tensor<string, []>("op_48335_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4023898048)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4027174912)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_48335_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_out_0_bias_to_fp16, dilations = var_48333, groups = var_30385, pad = var_48335_pad_0, pad_type = var_48335_pad_type_0, strides = var_48331, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn1_to_out_0_weight_to_fp16, x = input_611_cast)[name = tensor<string, []>("op_48335_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_321_cast = add(x = var_48335_cast, y = inputs_319_cast)[name = tensor<string, []>("inputs_321_cast")];
            tensor<int32, [1]> var_48339 = const()[name = tensor<string, []>("op_48339"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_321_cast = reduce_mean(axes = var_48339, keep_dims = var_30380, x = inputs_321_cast)[name = tensor<string, []>("channels_mean_321_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_321_cast = sub(x = inputs_321_cast, y = channels_mean_321_cast)[name = tensor<string, []>("zero_mean_321_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_321_cast = mul(x = zero_mean_321_cast, y = zero_mean_321_cast)[name = tensor<string, []>("zero_mean_sq_321_cast")];
            tensor<int32, [1]> var_48343 = const()[name = tensor<string, []>("op_48343"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_48344_cast = reduce_mean(axes = var_48343, keep_dims = var_30380, x = zero_mean_sq_321_cast)[name = tensor<string, []>("op_48344_cast")];
            tensor<fp16, []> var_48345_to_fp16 = const()[name = tensor<string, []>("op_48345_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_48346_cast = add(x = var_48344_cast, y = var_48345_to_fp16)[name = tensor<string, []>("op_48346_cast")];
            tensor<fp16, []> denom_321_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_321_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_321_cast = rsqrt(epsilon = denom_321_epsilon_0_to_fp16, x = var_48346_cast)[name = tensor<string, []>("denom_321_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_321_cast = mul(x = zero_mean_321_cast, y = denom_321_cast)[name = tensor<string, []>("out_321_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_48350_to_fp16 = const()[name = tensor<string, []>("op_48350_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4027177536)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_48351_cast = add(x = out_321_cast, y = var_48350_to_fp16)[name = tensor<string, []>("op_48351_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_48353_to_fp16 = const()[name = tensor<string, []>("op_48353_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4027180160)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_417_cast = mul(x = var_48351_cast, y = var_48353_to_fp16)[name = tensor<string, []>("hidden_states_417_cast")];
            tensor<int32, [2]> var_48360 = const()[name = tensor<string, []>("op_48360"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_48362 = const()[name = tensor<string, []>("op_48362"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_215_pad_type_0 = const()[name = tensor<string, []>("q_215_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_215_pad_0 = const()[name = tensor<string, []>("q_215_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4027182784)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_215_cast = conv(dilations = var_48362, groups = var_30385, pad = q_215_pad_0, pad_type = q_215_pad_type_0, strides = var_48360, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_q_weight_to_fp16, x = hidden_states_417_cast)[name = tensor<string, []>("q_215_cast")];
            tensor<int32, [2]> var_48366 = const()[name = tensor<string, []>("op_48366"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_48368 = const()[name = tensor<string, []>("op_48368"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_429_pad_type_0 = const()[name = tensor<string, []>("k_429_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_429_pad_0 = const()[name = tensor<string, []>("k_429_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4030459648)))];
            tensor<fp16, [2, 1280, 1, 77]> k_429_cast = conv(dilations = var_48368, groups = var_30385, pad = k_429_pad_0, pad_type = k_429_pad_type_0, strides = var_48366, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_429_cast")];
            tensor<int32, [2]> var_48372 = const()[name = tensor<string, []>("op_48372"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_48374 = const()[name = tensor<string, []>("op_48374"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_215_pad_type_0 = const()[name = tensor<string, []>("v_215_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_215_pad_0 = const()[name = tensor<string, []>("v_215_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4035702592)))];
            tensor<fp16, [2, 1280, 1, 77]> v_215_cast = conv(dilations = var_48374, groups = var_30385, pad = v_215_pad_0, pad_type = v_215_pad_type_0, strides = var_48372, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_215_cast")];
            tensor<int32, [4]> var_48378_begin_0 = const()[name = tensor<string, []>("op_48378_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_48378_end_0 = const()[name = tensor<string, []>("op_48378_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_48378_end_mask_0 = const()[name = tensor<string, []>("op_48378_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48378_cast = slice_by_index(begin = var_48378_begin_0, end = var_48378_end_0, end_mask = var_48378_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48378_cast")];
            tensor<int32, [4]> var_48382_begin_0 = const()[name = tensor<string, []>("op_48382_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_48382_end_0 = const()[name = tensor<string, []>("op_48382_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_48382_end_mask_0 = const()[name = tensor<string, []>("op_48382_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48382_cast = slice_by_index(begin = var_48382_begin_0, end = var_48382_end_0, end_mask = var_48382_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48382_cast")];
            tensor<int32, [4]> var_48386_begin_0 = const()[name = tensor<string, []>("op_48386_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_48386_end_0 = const()[name = tensor<string, []>("op_48386_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_48386_end_mask_0 = const()[name = tensor<string, []>("op_48386_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48386_cast = slice_by_index(begin = var_48386_begin_0, end = var_48386_end_0, end_mask = var_48386_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48386_cast")];
            tensor<int32, [4]> var_48390_begin_0 = const()[name = tensor<string, []>("op_48390_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_48390_end_0 = const()[name = tensor<string, []>("op_48390_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_48390_end_mask_0 = const()[name = tensor<string, []>("op_48390_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48390_cast = slice_by_index(begin = var_48390_begin_0, end = var_48390_end_0, end_mask = var_48390_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48390_cast")];
            tensor<int32, [4]> var_48394_begin_0 = const()[name = tensor<string, []>("op_48394_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_48394_end_0 = const()[name = tensor<string, []>("op_48394_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_48394_end_mask_0 = const()[name = tensor<string, []>("op_48394_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48394_cast = slice_by_index(begin = var_48394_begin_0, end = var_48394_end_0, end_mask = var_48394_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48394_cast")];
            tensor<int32, [4]> var_48398_begin_0 = const()[name = tensor<string, []>("op_48398_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_48398_end_0 = const()[name = tensor<string, []>("op_48398_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_48398_end_mask_0 = const()[name = tensor<string, []>("op_48398_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48398_cast = slice_by_index(begin = var_48398_begin_0, end = var_48398_end_0, end_mask = var_48398_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48398_cast")];
            tensor<int32, [4]> var_48402_begin_0 = const()[name = tensor<string, []>("op_48402_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_48402_end_0 = const()[name = tensor<string, []>("op_48402_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_48402_end_mask_0 = const()[name = tensor<string, []>("op_48402_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48402_cast = slice_by_index(begin = var_48402_begin_0, end = var_48402_end_0, end_mask = var_48402_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48402_cast")];
            tensor<int32, [4]> var_48406_begin_0 = const()[name = tensor<string, []>("op_48406_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_48406_end_0 = const()[name = tensor<string, []>("op_48406_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_48406_end_mask_0 = const()[name = tensor<string, []>("op_48406_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48406_cast = slice_by_index(begin = var_48406_begin_0, end = var_48406_end_0, end_mask = var_48406_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48406_cast")];
            tensor<int32, [4]> var_48410_begin_0 = const()[name = tensor<string, []>("op_48410_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_48410_end_0 = const()[name = tensor<string, []>("op_48410_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_48410_end_mask_0 = const()[name = tensor<string, []>("op_48410_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48410_cast = slice_by_index(begin = var_48410_begin_0, end = var_48410_end_0, end_mask = var_48410_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48410_cast")];
            tensor<int32, [4]> var_48414_begin_0 = const()[name = tensor<string, []>("op_48414_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_48414_end_0 = const()[name = tensor<string, []>("op_48414_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_48414_end_mask_0 = const()[name = tensor<string, []>("op_48414_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48414_cast = slice_by_index(begin = var_48414_begin_0, end = var_48414_end_0, end_mask = var_48414_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48414_cast")];
            tensor<int32, [4]> var_48418_begin_0 = const()[name = tensor<string, []>("op_48418_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_48418_end_0 = const()[name = tensor<string, []>("op_48418_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_48418_end_mask_0 = const()[name = tensor<string, []>("op_48418_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48418_cast = slice_by_index(begin = var_48418_begin_0, end = var_48418_end_0, end_mask = var_48418_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48418_cast")];
            tensor<int32, [4]> var_48422_begin_0 = const()[name = tensor<string, []>("op_48422_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_48422_end_0 = const()[name = tensor<string, []>("op_48422_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_48422_end_mask_0 = const()[name = tensor<string, []>("op_48422_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48422_cast = slice_by_index(begin = var_48422_begin_0, end = var_48422_end_0, end_mask = var_48422_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48422_cast")];
            tensor<int32, [4]> var_48426_begin_0 = const()[name = tensor<string, []>("op_48426_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_48426_end_0 = const()[name = tensor<string, []>("op_48426_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_48426_end_mask_0 = const()[name = tensor<string, []>("op_48426_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48426_cast = slice_by_index(begin = var_48426_begin_0, end = var_48426_end_0, end_mask = var_48426_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48426_cast")];
            tensor<int32, [4]> var_48430_begin_0 = const()[name = tensor<string, []>("op_48430_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_48430_end_0 = const()[name = tensor<string, []>("op_48430_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_48430_end_mask_0 = const()[name = tensor<string, []>("op_48430_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48430_cast = slice_by_index(begin = var_48430_begin_0, end = var_48430_end_0, end_mask = var_48430_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48430_cast")];
            tensor<int32, [4]> var_48434_begin_0 = const()[name = tensor<string, []>("op_48434_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_48434_end_0 = const()[name = tensor<string, []>("op_48434_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_48434_end_mask_0 = const()[name = tensor<string, []>("op_48434_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48434_cast = slice_by_index(begin = var_48434_begin_0, end = var_48434_end_0, end_mask = var_48434_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48434_cast")];
            tensor<int32, [4]> var_48438_begin_0 = const()[name = tensor<string, []>("op_48438_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_48438_end_0 = const()[name = tensor<string, []>("op_48438_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_48438_end_mask_0 = const()[name = tensor<string, []>("op_48438_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48438_cast = slice_by_index(begin = var_48438_begin_0, end = var_48438_end_0, end_mask = var_48438_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48438_cast")];
            tensor<int32, [4]> var_48442_begin_0 = const()[name = tensor<string, []>("op_48442_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_48442_end_0 = const()[name = tensor<string, []>("op_48442_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_48442_end_mask_0 = const()[name = tensor<string, []>("op_48442_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48442_cast = slice_by_index(begin = var_48442_begin_0, end = var_48442_end_0, end_mask = var_48442_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48442_cast")];
            tensor<int32, [4]> var_48446_begin_0 = const()[name = tensor<string, []>("op_48446_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_48446_end_0 = const()[name = tensor<string, []>("op_48446_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_48446_end_mask_0 = const()[name = tensor<string, []>("op_48446_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48446_cast = slice_by_index(begin = var_48446_begin_0, end = var_48446_end_0, end_mask = var_48446_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48446_cast")];
            tensor<int32, [4]> var_48450_begin_0 = const()[name = tensor<string, []>("op_48450_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_48450_end_0 = const()[name = tensor<string, []>("op_48450_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_48450_end_mask_0 = const()[name = tensor<string, []>("op_48450_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48450_cast = slice_by_index(begin = var_48450_begin_0, end = var_48450_end_0, end_mask = var_48450_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48450_cast")];
            tensor<int32, [4]> var_48454_begin_0 = const()[name = tensor<string, []>("op_48454_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_48454_end_0 = const()[name = tensor<string, []>("op_48454_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_48454_end_mask_0 = const()[name = tensor<string, []>("op_48454_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48454_cast = slice_by_index(begin = var_48454_begin_0, end = var_48454_end_0, end_mask = var_48454_end_mask_0, x = q_215_cast)[name = tensor<string, []>("op_48454_cast")];
            tensor<int32, [4]> k_431_perm_0 = const()[name = tensor<string, []>("k_431_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_48461_begin_0 = const()[name = tensor<string, []>("op_48461_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_48461_end_0 = const()[name = tensor<string, []>("op_48461_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_48461_end_mask_0 = const()[name = tensor<string, []>("op_48461_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_32 = transpose(perm = k_431_perm_0, x = k_429_cast)[name = tensor<string, []>("transpose_32")];
            tensor<fp16, [2, 77, 1, 64]> var_48461_cast = slice_by_index(begin = var_48461_begin_0, end = var_48461_end_0, end_mask = var_48461_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48461_cast")];
            tensor<int32, [4]> var_48465_begin_0 = const()[name = tensor<string, []>("op_48465_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_48465_end_0 = const()[name = tensor<string, []>("op_48465_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_48465_end_mask_0 = const()[name = tensor<string, []>("op_48465_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48465_cast = slice_by_index(begin = var_48465_begin_0, end = var_48465_end_0, end_mask = var_48465_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48465_cast")];
            tensor<int32, [4]> var_48469_begin_0 = const()[name = tensor<string, []>("op_48469_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_48469_end_0 = const()[name = tensor<string, []>("op_48469_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_48469_end_mask_0 = const()[name = tensor<string, []>("op_48469_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48469_cast = slice_by_index(begin = var_48469_begin_0, end = var_48469_end_0, end_mask = var_48469_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48469_cast")];
            tensor<int32, [4]> var_48473_begin_0 = const()[name = tensor<string, []>("op_48473_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_48473_end_0 = const()[name = tensor<string, []>("op_48473_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_48473_end_mask_0 = const()[name = tensor<string, []>("op_48473_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48473_cast = slice_by_index(begin = var_48473_begin_0, end = var_48473_end_0, end_mask = var_48473_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48473_cast")];
            tensor<int32, [4]> var_48477_begin_0 = const()[name = tensor<string, []>("op_48477_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_48477_end_0 = const()[name = tensor<string, []>("op_48477_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_48477_end_mask_0 = const()[name = tensor<string, []>("op_48477_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48477_cast = slice_by_index(begin = var_48477_begin_0, end = var_48477_end_0, end_mask = var_48477_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48477_cast")];
            tensor<int32, [4]> var_48481_begin_0 = const()[name = tensor<string, []>("op_48481_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_48481_end_0 = const()[name = tensor<string, []>("op_48481_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_48481_end_mask_0 = const()[name = tensor<string, []>("op_48481_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48481_cast = slice_by_index(begin = var_48481_begin_0, end = var_48481_end_0, end_mask = var_48481_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48481_cast")];
            tensor<int32, [4]> var_48485_begin_0 = const()[name = tensor<string, []>("op_48485_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_48485_end_0 = const()[name = tensor<string, []>("op_48485_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_48485_end_mask_0 = const()[name = tensor<string, []>("op_48485_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48485_cast = slice_by_index(begin = var_48485_begin_0, end = var_48485_end_0, end_mask = var_48485_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48485_cast")];
            tensor<int32, [4]> var_48489_begin_0 = const()[name = tensor<string, []>("op_48489_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_48489_end_0 = const()[name = tensor<string, []>("op_48489_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_48489_end_mask_0 = const()[name = tensor<string, []>("op_48489_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48489_cast = slice_by_index(begin = var_48489_begin_0, end = var_48489_end_0, end_mask = var_48489_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48489_cast")];
            tensor<int32, [4]> var_48493_begin_0 = const()[name = tensor<string, []>("op_48493_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_48493_end_0 = const()[name = tensor<string, []>("op_48493_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_48493_end_mask_0 = const()[name = tensor<string, []>("op_48493_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48493_cast = slice_by_index(begin = var_48493_begin_0, end = var_48493_end_0, end_mask = var_48493_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48493_cast")];
            tensor<int32, [4]> var_48497_begin_0 = const()[name = tensor<string, []>("op_48497_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_48497_end_0 = const()[name = tensor<string, []>("op_48497_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_48497_end_mask_0 = const()[name = tensor<string, []>("op_48497_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48497_cast = slice_by_index(begin = var_48497_begin_0, end = var_48497_end_0, end_mask = var_48497_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48497_cast")];
            tensor<int32, [4]> var_48501_begin_0 = const()[name = tensor<string, []>("op_48501_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_48501_end_0 = const()[name = tensor<string, []>("op_48501_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_48501_end_mask_0 = const()[name = tensor<string, []>("op_48501_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48501_cast = slice_by_index(begin = var_48501_begin_0, end = var_48501_end_0, end_mask = var_48501_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48501_cast")];
            tensor<int32, [4]> var_48505_begin_0 = const()[name = tensor<string, []>("op_48505_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_48505_end_0 = const()[name = tensor<string, []>("op_48505_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_48505_end_mask_0 = const()[name = tensor<string, []>("op_48505_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48505_cast = slice_by_index(begin = var_48505_begin_0, end = var_48505_end_0, end_mask = var_48505_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48505_cast")];
            tensor<int32, [4]> var_48509_begin_0 = const()[name = tensor<string, []>("op_48509_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_48509_end_0 = const()[name = tensor<string, []>("op_48509_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_48509_end_mask_0 = const()[name = tensor<string, []>("op_48509_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48509_cast = slice_by_index(begin = var_48509_begin_0, end = var_48509_end_0, end_mask = var_48509_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48509_cast")];
            tensor<int32, [4]> var_48513_begin_0 = const()[name = tensor<string, []>("op_48513_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_48513_end_0 = const()[name = tensor<string, []>("op_48513_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_48513_end_mask_0 = const()[name = tensor<string, []>("op_48513_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48513_cast = slice_by_index(begin = var_48513_begin_0, end = var_48513_end_0, end_mask = var_48513_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48513_cast")];
            tensor<int32, [4]> var_48517_begin_0 = const()[name = tensor<string, []>("op_48517_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_48517_end_0 = const()[name = tensor<string, []>("op_48517_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_48517_end_mask_0 = const()[name = tensor<string, []>("op_48517_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48517_cast = slice_by_index(begin = var_48517_begin_0, end = var_48517_end_0, end_mask = var_48517_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48517_cast")];
            tensor<int32, [4]> var_48521_begin_0 = const()[name = tensor<string, []>("op_48521_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_48521_end_0 = const()[name = tensor<string, []>("op_48521_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_48521_end_mask_0 = const()[name = tensor<string, []>("op_48521_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48521_cast = slice_by_index(begin = var_48521_begin_0, end = var_48521_end_0, end_mask = var_48521_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48521_cast")];
            tensor<int32, [4]> var_48525_begin_0 = const()[name = tensor<string, []>("op_48525_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_48525_end_0 = const()[name = tensor<string, []>("op_48525_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_48525_end_mask_0 = const()[name = tensor<string, []>("op_48525_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48525_cast = slice_by_index(begin = var_48525_begin_0, end = var_48525_end_0, end_mask = var_48525_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48525_cast")];
            tensor<int32, [4]> var_48529_begin_0 = const()[name = tensor<string, []>("op_48529_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_48529_end_0 = const()[name = tensor<string, []>("op_48529_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_48529_end_mask_0 = const()[name = tensor<string, []>("op_48529_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48529_cast = slice_by_index(begin = var_48529_begin_0, end = var_48529_end_0, end_mask = var_48529_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48529_cast")];
            tensor<int32, [4]> var_48533_begin_0 = const()[name = tensor<string, []>("op_48533_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_48533_end_0 = const()[name = tensor<string, []>("op_48533_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_48533_end_mask_0 = const()[name = tensor<string, []>("op_48533_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48533_cast = slice_by_index(begin = var_48533_begin_0, end = var_48533_end_0, end_mask = var_48533_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48533_cast")];
            tensor<int32, [4]> var_48537_begin_0 = const()[name = tensor<string, []>("op_48537_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_48537_end_0 = const()[name = tensor<string, []>("op_48537_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_48537_end_mask_0 = const()[name = tensor<string, []>("op_48537_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_48537_cast = slice_by_index(begin = var_48537_begin_0, end = var_48537_end_0, end_mask = var_48537_end_mask_0, x = transpose_32)[name = tensor<string, []>("op_48537_cast")];
            tensor<int32, [4]> var_48539_begin_0 = const()[name = tensor<string, []>("op_48539_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_48539_end_0 = const()[name = tensor<string, []>("op_48539_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_48539_end_mask_0 = const()[name = tensor<string, []>("op_48539_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48539_cast = slice_by_index(begin = var_48539_begin_0, end = var_48539_end_0, end_mask = var_48539_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48539_cast")];
            tensor<int32, [4]> var_48543_begin_0 = const()[name = tensor<string, []>("op_48543_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_48543_end_0 = const()[name = tensor<string, []>("op_48543_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_48543_end_mask_0 = const()[name = tensor<string, []>("op_48543_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48543_cast = slice_by_index(begin = var_48543_begin_0, end = var_48543_end_0, end_mask = var_48543_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48543_cast")];
            tensor<int32, [4]> var_48547_begin_0 = const()[name = tensor<string, []>("op_48547_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_48547_end_0 = const()[name = tensor<string, []>("op_48547_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_48547_end_mask_0 = const()[name = tensor<string, []>("op_48547_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48547_cast = slice_by_index(begin = var_48547_begin_0, end = var_48547_end_0, end_mask = var_48547_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48547_cast")];
            tensor<int32, [4]> var_48551_begin_0 = const()[name = tensor<string, []>("op_48551_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_48551_end_0 = const()[name = tensor<string, []>("op_48551_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_48551_end_mask_0 = const()[name = tensor<string, []>("op_48551_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48551_cast = slice_by_index(begin = var_48551_begin_0, end = var_48551_end_0, end_mask = var_48551_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48551_cast")];
            tensor<int32, [4]> var_48555_begin_0 = const()[name = tensor<string, []>("op_48555_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_48555_end_0 = const()[name = tensor<string, []>("op_48555_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_48555_end_mask_0 = const()[name = tensor<string, []>("op_48555_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48555_cast = slice_by_index(begin = var_48555_begin_0, end = var_48555_end_0, end_mask = var_48555_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48555_cast")];
            tensor<int32, [4]> var_48559_begin_0 = const()[name = tensor<string, []>("op_48559_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_48559_end_0 = const()[name = tensor<string, []>("op_48559_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_48559_end_mask_0 = const()[name = tensor<string, []>("op_48559_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48559_cast = slice_by_index(begin = var_48559_begin_0, end = var_48559_end_0, end_mask = var_48559_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48559_cast")];
            tensor<int32, [4]> var_48563_begin_0 = const()[name = tensor<string, []>("op_48563_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_48563_end_0 = const()[name = tensor<string, []>("op_48563_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_48563_end_mask_0 = const()[name = tensor<string, []>("op_48563_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48563_cast = slice_by_index(begin = var_48563_begin_0, end = var_48563_end_0, end_mask = var_48563_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48563_cast")];
            tensor<int32, [4]> var_48567_begin_0 = const()[name = tensor<string, []>("op_48567_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_48567_end_0 = const()[name = tensor<string, []>("op_48567_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_48567_end_mask_0 = const()[name = tensor<string, []>("op_48567_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48567_cast = slice_by_index(begin = var_48567_begin_0, end = var_48567_end_0, end_mask = var_48567_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48567_cast")];
            tensor<int32, [4]> var_48571_begin_0 = const()[name = tensor<string, []>("op_48571_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_48571_end_0 = const()[name = tensor<string, []>("op_48571_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_48571_end_mask_0 = const()[name = tensor<string, []>("op_48571_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48571_cast = slice_by_index(begin = var_48571_begin_0, end = var_48571_end_0, end_mask = var_48571_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48571_cast")];
            tensor<int32, [4]> var_48575_begin_0 = const()[name = tensor<string, []>("op_48575_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_48575_end_0 = const()[name = tensor<string, []>("op_48575_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_48575_end_mask_0 = const()[name = tensor<string, []>("op_48575_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48575_cast = slice_by_index(begin = var_48575_begin_0, end = var_48575_end_0, end_mask = var_48575_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48575_cast")];
            tensor<int32, [4]> var_48579_begin_0 = const()[name = tensor<string, []>("op_48579_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_48579_end_0 = const()[name = tensor<string, []>("op_48579_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_48579_end_mask_0 = const()[name = tensor<string, []>("op_48579_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48579_cast = slice_by_index(begin = var_48579_begin_0, end = var_48579_end_0, end_mask = var_48579_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48579_cast")];
            tensor<int32, [4]> var_48583_begin_0 = const()[name = tensor<string, []>("op_48583_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_48583_end_0 = const()[name = tensor<string, []>("op_48583_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_48583_end_mask_0 = const()[name = tensor<string, []>("op_48583_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48583_cast = slice_by_index(begin = var_48583_begin_0, end = var_48583_end_0, end_mask = var_48583_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48583_cast")];
            tensor<int32, [4]> var_48587_begin_0 = const()[name = tensor<string, []>("op_48587_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_48587_end_0 = const()[name = tensor<string, []>("op_48587_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_48587_end_mask_0 = const()[name = tensor<string, []>("op_48587_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48587_cast = slice_by_index(begin = var_48587_begin_0, end = var_48587_end_0, end_mask = var_48587_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48587_cast")];
            tensor<int32, [4]> var_48591_begin_0 = const()[name = tensor<string, []>("op_48591_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_48591_end_0 = const()[name = tensor<string, []>("op_48591_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_48591_end_mask_0 = const()[name = tensor<string, []>("op_48591_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48591_cast = slice_by_index(begin = var_48591_begin_0, end = var_48591_end_0, end_mask = var_48591_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48591_cast")];
            tensor<int32, [4]> var_48595_begin_0 = const()[name = tensor<string, []>("op_48595_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_48595_end_0 = const()[name = tensor<string, []>("op_48595_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_48595_end_mask_0 = const()[name = tensor<string, []>("op_48595_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48595_cast = slice_by_index(begin = var_48595_begin_0, end = var_48595_end_0, end_mask = var_48595_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48595_cast")];
            tensor<int32, [4]> var_48599_begin_0 = const()[name = tensor<string, []>("op_48599_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_48599_end_0 = const()[name = tensor<string, []>("op_48599_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_48599_end_mask_0 = const()[name = tensor<string, []>("op_48599_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48599_cast = slice_by_index(begin = var_48599_begin_0, end = var_48599_end_0, end_mask = var_48599_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48599_cast")];
            tensor<int32, [4]> var_48603_begin_0 = const()[name = tensor<string, []>("op_48603_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_48603_end_0 = const()[name = tensor<string, []>("op_48603_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_48603_end_mask_0 = const()[name = tensor<string, []>("op_48603_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48603_cast = slice_by_index(begin = var_48603_begin_0, end = var_48603_end_0, end_mask = var_48603_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48603_cast")];
            tensor<int32, [4]> var_48607_begin_0 = const()[name = tensor<string, []>("op_48607_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_48607_end_0 = const()[name = tensor<string, []>("op_48607_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_48607_end_mask_0 = const()[name = tensor<string, []>("op_48607_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48607_cast = slice_by_index(begin = var_48607_begin_0, end = var_48607_end_0, end_mask = var_48607_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48607_cast")];
            tensor<int32, [4]> var_48611_begin_0 = const()[name = tensor<string, []>("op_48611_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_48611_end_0 = const()[name = tensor<string, []>("op_48611_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_48611_end_mask_0 = const()[name = tensor<string, []>("op_48611_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48611_cast = slice_by_index(begin = var_48611_begin_0, end = var_48611_end_0, end_mask = var_48611_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48611_cast")];
            tensor<int32, [4]> var_48615_begin_0 = const()[name = tensor<string, []>("op_48615_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_48615_end_0 = const()[name = tensor<string, []>("op_48615_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_48615_end_mask_0 = const()[name = tensor<string, []>("op_48615_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_48615_cast = slice_by_index(begin = var_48615_begin_0, end = var_48615_end_0, end_mask = var_48615_end_mask_0, x = v_215_cast)[name = tensor<string, []>("op_48615_cast")];
            tensor<string, []> var_48619_equation_0 = const()[name = tensor<string, []>("op_48619_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48619_cast = einsum(equation = var_48619_equation_0, values = (var_48461_cast, var_48378_cast))[name = tensor<string, []>("op_48619_cast")];
            tensor<fp16, []> var_48620_to_fp16 = const()[name = tensor<string, []>("op_48620_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4121_cast = mul(x = var_48619_cast, y = var_48620_to_fp16)[name = tensor<string, []>("aw_4121_cast")];
            tensor<string, []> var_48623_equation_0 = const()[name = tensor<string, []>("op_48623_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48623_cast = einsum(equation = var_48623_equation_0, values = (var_48465_cast, var_48382_cast))[name = tensor<string, []>("op_48623_cast")];
            tensor<fp16, []> var_48624_to_fp16 = const()[name = tensor<string, []>("op_48624_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4123_cast = mul(x = var_48623_cast, y = var_48624_to_fp16)[name = tensor<string, []>("aw_4123_cast")];
            tensor<string, []> var_48627_equation_0 = const()[name = tensor<string, []>("op_48627_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48627_cast = einsum(equation = var_48627_equation_0, values = (var_48469_cast, var_48386_cast))[name = tensor<string, []>("op_48627_cast")];
            tensor<fp16, []> var_48628_to_fp16 = const()[name = tensor<string, []>("op_48628_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4125_cast = mul(x = var_48627_cast, y = var_48628_to_fp16)[name = tensor<string, []>("aw_4125_cast")];
            tensor<string, []> var_48631_equation_0 = const()[name = tensor<string, []>("op_48631_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48631_cast = einsum(equation = var_48631_equation_0, values = (var_48473_cast, var_48390_cast))[name = tensor<string, []>("op_48631_cast")];
            tensor<fp16, []> var_48632_to_fp16 = const()[name = tensor<string, []>("op_48632_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4127_cast = mul(x = var_48631_cast, y = var_48632_to_fp16)[name = tensor<string, []>("aw_4127_cast")];
            tensor<string, []> var_48635_equation_0 = const()[name = tensor<string, []>("op_48635_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48635_cast = einsum(equation = var_48635_equation_0, values = (var_48477_cast, var_48394_cast))[name = tensor<string, []>("op_48635_cast")];
            tensor<fp16, []> var_48636_to_fp16 = const()[name = tensor<string, []>("op_48636_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4129_cast = mul(x = var_48635_cast, y = var_48636_to_fp16)[name = tensor<string, []>("aw_4129_cast")];
            tensor<string, []> var_48639_equation_0 = const()[name = tensor<string, []>("op_48639_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48639_cast = einsum(equation = var_48639_equation_0, values = (var_48481_cast, var_48398_cast))[name = tensor<string, []>("op_48639_cast")];
            tensor<fp16, []> var_48640_to_fp16 = const()[name = tensor<string, []>("op_48640_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4131_cast = mul(x = var_48639_cast, y = var_48640_to_fp16)[name = tensor<string, []>("aw_4131_cast")];
            tensor<string, []> var_48643_equation_0 = const()[name = tensor<string, []>("op_48643_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48643_cast = einsum(equation = var_48643_equation_0, values = (var_48485_cast, var_48402_cast))[name = tensor<string, []>("op_48643_cast")];
            tensor<fp16, []> var_48644_to_fp16 = const()[name = tensor<string, []>("op_48644_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4133_cast = mul(x = var_48643_cast, y = var_48644_to_fp16)[name = tensor<string, []>("aw_4133_cast")];
            tensor<string, []> var_48647_equation_0 = const()[name = tensor<string, []>("op_48647_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48647_cast = einsum(equation = var_48647_equation_0, values = (var_48489_cast, var_48406_cast))[name = tensor<string, []>("op_48647_cast")];
            tensor<fp16, []> var_48648_to_fp16 = const()[name = tensor<string, []>("op_48648_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4135_cast = mul(x = var_48647_cast, y = var_48648_to_fp16)[name = tensor<string, []>("aw_4135_cast")];
            tensor<string, []> var_48651_equation_0 = const()[name = tensor<string, []>("op_48651_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48651_cast = einsum(equation = var_48651_equation_0, values = (var_48493_cast, var_48410_cast))[name = tensor<string, []>("op_48651_cast")];
            tensor<fp16, []> var_48652_to_fp16 = const()[name = tensor<string, []>("op_48652_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4137_cast = mul(x = var_48651_cast, y = var_48652_to_fp16)[name = tensor<string, []>("aw_4137_cast")];
            tensor<string, []> var_48655_equation_0 = const()[name = tensor<string, []>("op_48655_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48655_cast = einsum(equation = var_48655_equation_0, values = (var_48497_cast, var_48414_cast))[name = tensor<string, []>("op_48655_cast")];
            tensor<fp16, []> var_48656_to_fp16 = const()[name = tensor<string, []>("op_48656_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4139_cast = mul(x = var_48655_cast, y = var_48656_to_fp16)[name = tensor<string, []>("aw_4139_cast")];
            tensor<string, []> var_48659_equation_0 = const()[name = tensor<string, []>("op_48659_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48659_cast = einsum(equation = var_48659_equation_0, values = (var_48501_cast, var_48418_cast))[name = tensor<string, []>("op_48659_cast")];
            tensor<fp16, []> var_48660_to_fp16 = const()[name = tensor<string, []>("op_48660_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4141_cast = mul(x = var_48659_cast, y = var_48660_to_fp16)[name = tensor<string, []>("aw_4141_cast")];
            tensor<string, []> var_48663_equation_0 = const()[name = tensor<string, []>("op_48663_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48663_cast = einsum(equation = var_48663_equation_0, values = (var_48505_cast, var_48422_cast))[name = tensor<string, []>("op_48663_cast")];
            tensor<fp16, []> var_48664_to_fp16 = const()[name = tensor<string, []>("op_48664_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4143_cast = mul(x = var_48663_cast, y = var_48664_to_fp16)[name = tensor<string, []>("aw_4143_cast")];
            tensor<string, []> var_48667_equation_0 = const()[name = tensor<string, []>("op_48667_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48667_cast = einsum(equation = var_48667_equation_0, values = (var_48509_cast, var_48426_cast))[name = tensor<string, []>("op_48667_cast")];
            tensor<fp16, []> var_48668_to_fp16 = const()[name = tensor<string, []>("op_48668_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4145_cast = mul(x = var_48667_cast, y = var_48668_to_fp16)[name = tensor<string, []>("aw_4145_cast")];
            tensor<string, []> var_48671_equation_0 = const()[name = tensor<string, []>("op_48671_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48671_cast = einsum(equation = var_48671_equation_0, values = (var_48513_cast, var_48430_cast))[name = tensor<string, []>("op_48671_cast")];
            tensor<fp16, []> var_48672_to_fp16 = const()[name = tensor<string, []>("op_48672_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4147_cast = mul(x = var_48671_cast, y = var_48672_to_fp16)[name = tensor<string, []>("aw_4147_cast")];
            tensor<string, []> var_48675_equation_0 = const()[name = tensor<string, []>("op_48675_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48675_cast = einsum(equation = var_48675_equation_0, values = (var_48517_cast, var_48434_cast))[name = tensor<string, []>("op_48675_cast")];
            tensor<fp16, []> var_48676_to_fp16 = const()[name = tensor<string, []>("op_48676_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4149_cast = mul(x = var_48675_cast, y = var_48676_to_fp16)[name = tensor<string, []>("aw_4149_cast")];
            tensor<string, []> var_48679_equation_0 = const()[name = tensor<string, []>("op_48679_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48679_cast = einsum(equation = var_48679_equation_0, values = (var_48521_cast, var_48438_cast))[name = tensor<string, []>("op_48679_cast")];
            tensor<fp16, []> var_48680_to_fp16 = const()[name = tensor<string, []>("op_48680_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4151_cast = mul(x = var_48679_cast, y = var_48680_to_fp16)[name = tensor<string, []>("aw_4151_cast")];
            tensor<string, []> var_48683_equation_0 = const()[name = tensor<string, []>("op_48683_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48683_cast = einsum(equation = var_48683_equation_0, values = (var_48525_cast, var_48442_cast))[name = tensor<string, []>("op_48683_cast")];
            tensor<fp16, []> var_48684_to_fp16 = const()[name = tensor<string, []>("op_48684_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4153_cast = mul(x = var_48683_cast, y = var_48684_to_fp16)[name = tensor<string, []>("aw_4153_cast")];
            tensor<string, []> var_48687_equation_0 = const()[name = tensor<string, []>("op_48687_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48687_cast = einsum(equation = var_48687_equation_0, values = (var_48529_cast, var_48446_cast))[name = tensor<string, []>("op_48687_cast")];
            tensor<fp16, []> var_48688_to_fp16 = const()[name = tensor<string, []>("op_48688_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4155_cast = mul(x = var_48687_cast, y = var_48688_to_fp16)[name = tensor<string, []>("aw_4155_cast")];
            tensor<string, []> var_48691_equation_0 = const()[name = tensor<string, []>("op_48691_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48691_cast = einsum(equation = var_48691_equation_0, values = (var_48533_cast, var_48450_cast))[name = tensor<string, []>("op_48691_cast")];
            tensor<fp16, []> var_48692_to_fp16 = const()[name = tensor<string, []>("op_48692_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4157_cast = mul(x = var_48691_cast, y = var_48692_to_fp16)[name = tensor<string, []>("aw_4157_cast")];
            tensor<string, []> var_48695_equation_0 = const()[name = tensor<string, []>("op_48695_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_48695_cast = einsum(equation = var_48695_equation_0, values = (var_48537_cast, var_48454_cast))[name = tensor<string, []>("op_48695_cast")];
            tensor<fp16, []> var_48696_to_fp16 = const()[name = tensor<string, []>("op_48696_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4159_cast = mul(x = var_48695_cast, y = var_48696_to_fp16)[name = tensor<string, []>("aw_4159_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48698_cast = softmax(axis = var_30385, x = aw_4121_cast)[name = tensor<string, []>("op_48698_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48699_cast = softmax(axis = var_30385, x = aw_4123_cast)[name = tensor<string, []>("op_48699_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48700_cast = softmax(axis = var_30385, x = aw_4125_cast)[name = tensor<string, []>("op_48700_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48701_cast = softmax(axis = var_30385, x = aw_4127_cast)[name = tensor<string, []>("op_48701_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48702_cast = softmax(axis = var_30385, x = aw_4129_cast)[name = tensor<string, []>("op_48702_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48703_cast = softmax(axis = var_30385, x = aw_4131_cast)[name = tensor<string, []>("op_48703_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48704_cast = softmax(axis = var_30385, x = aw_4133_cast)[name = tensor<string, []>("op_48704_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48705_cast = softmax(axis = var_30385, x = aw_4135_cast)[name = tensor<string, []>("op_48705_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48706_cast = softmax(axis = var_30385, x = aw_4137_cast)[name = tensor<string, []>("op_48706_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48707_cast = softmax(axis = var_30385, x = aw_4139_cast)[name = tensor<string, []>("op_48707_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48708_cast = softmax(axis = var_30385, x = aw_4141_cast)[name = tensor<string, []>("op_48708_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48709_cast = softmax(axis = var_30385, x = aw_4143_cast)[name = tensor<string, []>("op_48709_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48710_cast = softmax(axis = var_30385, x = aw_4145_cast)[name = tensor<string, []>("op_48710_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48711_cast = softmax(axis = var_30385, x = aw_4147_cast)[name = tensor<string, []>("op_48711_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48712_cast = softmax(axis = var_30385, x = aw_4149_cast)[name = tensor<string, []>("op_48712_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48713_cast = softmax(axis = var_30385, x = aw_4151_cast)[name = tensor<string, []>("op_48713_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48714_cast = softmax(axis = var_30385, x = aw_4153_cast)[name = tensor<string, []>("op_48714_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48715_cast = softmax(axis = var_30385, x = aw_4155_cast)[name = tensor<string, []>("op_48715_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48716_cast = softmax(axis = var_30385, x = aw_4157_cast)[name = tensor<string, []>("op_48716_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_48717_cast = softmax(axis = var_30385, x = aw_4159_cast)[name = tensor<string, []>("op_48717_cast")];
            tensor<string, []> var_48719_equation_0 = const()[name = tensor<string, []>("op_48719_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48719_cast = einsum(equation = var_48719_equation_0, values = (var_48539_cast, var_48698_cast))[name = tensor<string, []>("op_48719_cast")];
            tensor<string, []> var_48721_equation_0 = const()[name = tensor<string, []>("op_48721_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48721_cast = einsum(equation = var_48721_equation_0, values = (var_48543_cast, var_48699_cast))[name = tensor<string, []>("op_48721_cast")];
            tensor<string, []> var_48723_equation_0 = const()[name = tensor<string, []>("op_48723_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48723_cast = einsum(equation = var_48723_equation_0, values = (var_48547_cast, var_48700_cast))[name = tensor<string, []>("op_48723_cast")];
            tensor<string, []> var_48725_equation_0 = const()[name = tensor<string, []>("op_48725_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48725_cast = einsum(equation = var_48725_equation_0, values = (var_48551_cast, var_48701_cast))[name = tensor<string, []>("op_48725_cast")];
            tensor<string, []> var_48727_equation_0 = const()[name = tensor<string, []>("op_48727_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48727_cast = einsum(equation = var_48727_equation_0, values = (var_48555_cast, var_48702_cast))[name = tensor<string, []>("op_48727_cast")];
            tensor<string, []> var_48729_equation_0 = const()[name = tensor<string, []>("op_48729_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48729_cast = einsum(equation = var_48729_equation_0, values = (var_48559_cast, var_48703_cast))[name = tensor<string, []>("op_48729_cast")];
            tensor<string, []> var_48731_equation_0 = const()[name = tensor<string, []>("op_48731_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48731_cast = einsum(equation = var_48731_equation_0, values = (var_48563_cast, var_48704_cast))[name = tensor<string, []>("op_48731_cast")];
            tensor<string, []> var_48733_equation_0 = const()[name = tensor<string, []>("op_48733_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48733_cast = einsum(equation = var_48733_equation_0, values = (var_48567_cast, var_48705_cast))[name = tensor<string, []>("op_48733_cast")];
            tensor<string, []> var_48735_equation_0 = const()[name = tensor<string, []>("op_48735_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48735_cast = einsum(equation = var_48735_equation_0, values = (var_48571_cast, var_48706_cast))[name = tensor<string, []>("op_48735_cast")];
            tensor<string, []> var_48737_equation_0 = const()[name = tensor<string, []>("op_48737_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48737_cast = einsum(equation = var_48737_equation_0, values = (var_48575_cast, var_48707_cast))[name = tensor<string, []>("op_48737_cast")];
            tensor<string, []> var_48739_equation_0 = const()[name = tensor<string, []>("op_48739_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48739_cast = einsum(equation = var_48739_equation_0, values = (var_48579_cast, var_48708_cast))[name = tensor<string, []>("op_48739_cast")];
            tensor<string, []> var_48741_equation_0 = const()[name = tensor<string, []>("op_48741_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48741_cast = einsum(equation = var_48741_equation_0, values = (var_48583_cast, var_48709_cast))[name = tensor<string, []>("op_48741_cast")];
            tensor<string, []> var_48743_equation_0 = const()[name = tensor<string, []>("op_48743_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48743_cast = einsum(equation = var_48743_equation_0, values = (var_48587_cast, var_48710_cast))[name = tensor<string, []>("op_48743_cast")];
            tensor<string, []> var_48745_equation_0 = const()[name = tensor<string, []>("op_48745_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48745_cast = einsum(equation = var_48745_equation_0, values = (var_48591_cast, var_48711_cast))[name = tensor<string, []>("op_48745_cast")];
            tensor<string, []> var_48747_equation_0 = const()[name = tensor<string, []>("op_48747_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48747_cast = einsum(equation = var_48747_equation_0, values = (var_48595_cast, var_48712_cast))[name = tensor<string, []>("op_48747_cast")];
            tensor<string, []> var_48749_equation_0 = const()[name = tensor<string, []>("op_48749_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48749_cast = einsum(equation = var_48749_equation_0, values = (var_48599_cast, var_48713_cast))[name = tensor<string, []>("op_48749_cast")];
            tensor<string, []> var_48751_equation_0 = const()[name = tensor<string, []>("op_48751_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48751_cast = einsum(equation = var_48751_equation_0, values = (var_48603_cast, var_48714_cast))[name = tensor<string, []>("op_48751_cast")];
            tensor<string, []> var_48753_equation_0 = const()[name = tensor<string, []>("op_48753_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48753_cast = einsum(equation = var_48753_equation_0, values = (var_48607_cast, var_48715_cast))[name = tensor<string, []>("op_48753_cast")];
            tensor<string, []> var_48755_equation_0 = const()[name = tensor<string, []>("op_48755_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48755_cast = einsum(equation = var_48755_equation_0, values = (var_48611_cast, var_48716_cast))[name = tensor<string, []>("op_48755_cast")];
            tensor<string, []> var_48757_equation_0 = const()[name = tensor<string, []>("op_48757_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_48757_cast = einsum(equation = var_48757_equation_0, values = (var_48615_cast, var_48717_cast))[name = tensor<string, []>("op_48757_cast")];
            tensor<bool, []> input_613_interleave_0 = const()[name = tensor<string, []>("input_613_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_613_cast = concat(axis = var_30385, interleave = input_613_interleave_0, values = (var_48719_cast, var_48721_cast, var_48723_cast, var_48725_cast, var_48727_cast, var_48729_cast, var_48731_cast, var_48733_cast, var_48735_cast, var_48737_cast, var_48739_cast, var_48741_cast, var_48743_cast, var_48745_cast, var_48747_cast, var_48749_cast, var_48751_cast, var_48753_cast, var_48755_cast, var_48757_cast))[name = tensor<string, []>("input_613_cast")];
            tensor<int32, [2]> var_48763 = const()[name = tensor<string, []>("op_48763"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_48765 = const()[name = tensor<string, []>("op_48765"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_48767_pad_type_0 = const()[name = tensor<string, []>("op_48767_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_48767_pad_0 = const()[name = tensor<string, []>("op_48767_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4040945536)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4044222400)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_48767_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_out_0_bias_to_fp16, dilations = var_48765, groups = var_30385, pad = var_48767_pad_0, pad_type = var_48767_pad_type_0, strides = var_48763, weight = up_blocks_0_attentions_1_transformer_blocks_9_attn2_to_out_0_weight_to_fp16, x = input_613_cast)[name = tensor<string, []>("op_48767_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_323_cast = add(x = var_48767_cast, y = inputs_321_cast)[name = tensor<string, []>("inputs_323_cast")];
            tensor<int32, [1]> var_48771 = const()[name = tensor<string, []>("op_48771"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_323_cast = reduce_mean(axes = var_48771, keep_dims = var_30380, x = inputs_323_cast)[name = tensor<string, []>("channels_mean_323_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_323_cast = sub(x = inputs_323_cast, y = channels_mean_323_cast)[name = tensor<string, []>("zero_mean_323_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_323_cast = mul(x = zero_mean_323_cast, y = zero_mean_323_cast)[name = tensor<string, []>("zero_mean_sq_323_cast")];
            tensor<int32, [1]> var_48775 = const()[name = tensor<string, []>("op_48775"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_48776_cast = reduce_mean(axes = var_48775, keep_dims = var_30380, x = zero_mean_sq_323_cast)[name = tensor<string, []>("op_48776_cast")];
            tensor<fp16, []> var_48777_to_fp16 = const()[name = tensor<string, []>("op_48777_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_48778_cast = add(x = var_48776_cast, y = var_48777_to_fp16)[name = tensor<string, []>("op_48778_cast")];
            tensor<fp16, []> denom_323_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_323_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_323_cast = rsqrt(epsilon = denom_323_epsilon_0_to_fp16, x = var_48778_cast)[name = tensor<string, []>("denom_323_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_323_cast = mul(x = zero_mean_323_cast, y = denom_323_cast)[name = tensor<string, []>("out_323_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_48782_to_fp16 = const()[name = tensor<string, []>("op_48782_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4044225024)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_48783_cast = add(x = out_323_cast, y = var_48782_to_fp16)[name = tensor<string, []>("op_48783_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_48785_to_fp16 = const()[name = tensor<string, []>("op_48785_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4044227648)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_615_cast = mul(x = var_48783_cast, y = var_48785_to_fp16)[name = tensor<string, []>("input_615_cast")];
            tensor<int32, [2]> var_48793 = const()[name = tensor<string, []>("op_48793"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_48795 = const()[name = tensor<string, []>("op_48795"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_48797_pad_type_0 = const()[name = tensor<string, []>("op_48797_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_48797_pad_0 = const()[name = tensor<string, []>("op_48797_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4044230272)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_1_transformer_blocks_9_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4070444736)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_48797_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_9_ff_net_0_proj_bias_to_fp16, dilations = var_48795, groups = var_30385, pad = var_48797_pad_0, pad_type = var_48797_pad_type_0, strides = var_48793, weight = up_blocks_0_attentions_1_transformer_blocks_9_ff_net_0_proj_weight_to_fp16, x = input_615_cast)[name = tensor<string, []>("op_48797_cast")];
            tensor<int32, [2]> var_48798_split_sizes_0 = const()[name = tensor<string, []>("op_48798_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_48798_axis_0 = const()[name = tensor<string, []>("op_48798_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_48798_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_48798_cast_1 = split(axis = var_48798_axis_0, split_sizes = var_48798_split_sizes_0, x = var_48797_cast)[name = tensor<string, []>("op_48798_cast")];
            tensor<string, []> var_48800_mode_0 = const()[name = tensor<string, []>("op_48800_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_48800_cast = gelu(mode = var_48800_mode_0, x = var_48798_cast_1)[name = tensor<string, []>("op_48800_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_617_cast = mul(x = var_48798_cast_0, y = var_48800_cast)[name = tensor<string, []>("input_617_cast")];
            tensor<int32, [2]> var_48804 = const()[name = tensor<string, []>("op_48804"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_48806 = const()[name = tensor<string, []>("op_48806"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_48808_pad_type_0 = const()[name = tensor<string, []>("op_48808_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_48808_pad_0 = const()[name = tensor<string, []>("op_48808_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_1_transformer_blocks_9_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4070465280)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_transformer_blocks_9_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_transformer_blocks_9_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4083572544)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_48808_cast = conv(bias = up_blocks_0_attentions_1_transformer_blocks_9_ff_net_2_bias_to_fp16, dilations = var_48806, groups = var_30385, pad = var_48808_pad_0, pad_type = var_48808_pad_type_0, strides = var_48804, weight = up_blocks_0_attentions_1_transformer_blocks_9_ff_net_2_weight_to_fp16, x = input_617_cast)[name = tensor<string, []>("op_48808_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_421_cast = add(x = var_48808_cast, y = inputs_323_cast)[name = tensor<string, []>("hidden_states_421_cast")];
            tensor<int32, [4]> var_48810 = const()[name = tensor<string, []>("op_48810"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> input_619_cast = reshape(shape = var_48810, x = hidden_states_421_cast)[name = tensor<string, []>("input_619_cast")];
            tensor<int32, [2]> var_48814 = const()[name = tensor<string, []>("op_48814"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_48816 = const()[name = tensor<string, []>("op_48816"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_423_pad_type_0 = const()[name = tensor<string, []>("hidden_states_423_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_423_pad_0 = const()[name = tensor<string, []>("hidden_states_423_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_proj_out_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4083575168)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_1_proj_out_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4086852032)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_423_cast = conv(bias = up_blocks_0_attentions_1_proj_out_bias_to_fp16, dilations = var_48816, groups = var_30385, pad = hidden_states_423_pad_0, pad_type = hidden_states_423_pad_type_0, strides = var_48814, weight = up_blocks_0_attentions_1_proj_out_weight_to_fp16, x = input_619_cast)[name = tensor<string, []>("hidden_states_423_cast")];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_425_cast = add(x = hidden_states_423_cast, y = hidden_states_357_cast)[name = tensor<string, []>("hidden_states_425_cast")];
            tensor<bool, []> input_621_interleave_0 = const()[name = tensor<string, []>("input_621_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1920, 32, 32]> input_621_cast = concat(axis = var_30385, interleave = input_621_interleave_0, values = (hidden_states_425_cast, res_hidden_states_5_cast))[name = tensor<string, []>("input_621_cast")];
            tensor<int32, [5]> reshape_108_shape_0 = const()[name = tensor<string, []>("reshape_108_shape_0"), val = tensor<int32, [5]>([2, 32, 60, 32, 32])];
            tensor<fp16, [2, 32, 60, 32, 32]> reshape_108_cast = reshape(shape = reshape_108_shape_0, x = input_621_cast)[name = tensor<string, []>("reshape_108_cast")];
            tensor<int32, [3]> reduce_mean_81_axes_0 = const()[name = tensor<string, []>("reduce_mean_81_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_81_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_81_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_81_cast = reduce_mean(axes = reduce_mean_81_axes_0, keep_dims = reduce_mean_81_keep_dims_0, x = reshape_108_cast)[name = tensor<string, []>("reduce_mean_81_cast")];
            tensor<fp16, [2, 32, 60, 32, 32]> sub_54_cast = sub(x = reshape_108_cast, y = reduce_mean_81_cast)[name = tensor<string, []>("sub_54_cast")];
            tensor<fp16, [2, 32, 60, 32, 32]> square_27_cast = square(x = sub_54_cast)[name = tensor<string, []>("square_27_cast")];
            tensor<int32, [3]> reduce_mean_83_axes_0 = const()[name = tensor<string, []>("reduce_mean_83_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_83_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_83_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_83_cast = reduce_mean(axes = reduce_mean_83_axes_0, keep_dims = reduce_mean_83_keep_dims_0, x = square_27_cast)[name = tensor<string, []>("reduce_mean_83_cast")];
            tensor<fp16, []> add_54_y_0_to_fp16 = const()[name = tensor<string, []>("add_54_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_54_cast = add(x = reduce_mean_83_cast, y = add_54_y_0_to_fp16)[name = tensor<string, []>("add_54_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_27_cast = sqrt(x = add_54_cast)[name = tensor<string, []>("sqrt_27_cast")];
            tensor<fp16, [2, 32, 60, 32, 32]> real_div_27_cast = real_div(x = sub_54_cast, y = sqrt_27_cast)[name = tensor<string, []>("real_div_27_cast")];
            tensor<int32, [4]> reshape_109_shape_0 = const()[name = tensor<string, []>("reshape_109_shape_0"), val = tensor<int32, [4]>([2, 1920, 32, 32])];
            tensor<fp16, [2, 1920, 32, 32]> reshape_109_cast = reshape(shape = reshape_109_shape_0, x = real_div_27_cast)[name = tensor<string, []>("reshape_109_cast")];
            tensor<fp16, [1920]> add_55_mean_0_to_fp16 = const()[name = tensor<string, []>("add_55_mean_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4086854656)))];
            tensor<fp16, [1920]> add_55_variance_0_to_fp16 = const()[name = tensor<string, []>("add_55_variance_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4086858560)))];
            tensor<fp16, [1920]> add_55_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_55_gamma_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4086862464)))];
            tensor<fp16, [1920]> add_55_beta_0_to_fp16 = const()[name = tensor<string, []>("add_55_beta_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4086866368)))];
            tensor<fp16, []> add_55_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_55_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1920, 32, 32]> add_55_cast = batch_norm(beta = add_55_beta_0_to_fp16, epsilon = add_55_epsilon_0_to_fp16, gamma = add_55_gamma_0_to_fp16, mean = add_55_mean_0_to_fp16, variance = add_55_variance_0_to_fp16, x = reshape_109_cast)[name = tensor<string, []>("add_55_cast")];
            tensor<fp16, [2, 1920, 32, 32]> input_625_cast = silu(x = add_55_cast)[name = tensor<string, []>("input_625_cast")];
            tensor<int32, [2]> var_48834 = const()[name = tensor<string, []>("op_48834"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_48836 = const()[name = tensor<string, []>("op_48836"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_427_pad_type_0 = const()[name = tensor<string, []>("hidden_states_427_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_427_pad_0 = const()[name = tensor<string, []>("hidden_states_427_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1920, 3, 3]> up_blocks_0_resnets_2_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_2_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 1920, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4086870272)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_2_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_2_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4131107136)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_427_cast = conv(bias = up_blocks_0_resnets_2_conv1_bias_to_fp16, dilations = var_48836, groups = var_30385, pad = hidden_states_427_pad_0, pad_type = hidden_states_427_pad_type_0, strides = var_48834, weight = up_blocks_0_resnets_2_conv1_weight_to_fp16, x = input_625_cast)[name = tensor<string, []>("hidden_states_427_cast")];
            tensor<int32, [2]> var_48842 = const()[name = tensor<string, []>("op_48842"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_48844 = const()[name = tensor<string, []>("op_48844"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_21_pad_type_0 = const()[name = tensor<string, []>("temb_21_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_21_pad_0 = const()[name = tensor<string, []>("temb_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4131109760)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4134386624)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_21_cast = conv(bias = up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_48844, groups = var_30385, pad = temb_21_pad_0, pad_type = temb_21_pad_type_0, strides = var_48842, weight = up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16, x = input_23_cast)[name = tensor<string, []>("temb_21_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_629_cast = add(x = hidden_states_427_cast, y = temb_21_cast)[name = tensor<string, []>("input_629_cast")];
            tensor<int32, [5]> reshape_112_shape_0 = const()[name = tensor<string, []>("reshape_112_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_112_cast = reshape(shape = reshape_112_shape_0, x = input_629_cast)[name = tensor<string, []>("reshape_112_cast")];
            tensor<int32, [3]> reduce_mean_84_axes_0 = const()[name = tensor<string, []>("reduce_mean_84_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_84_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_84_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_84_cast = reduce_mean(axes = reduce_mean_84_axes_0, keep_dims = reduce_mean_84_keep_dims_0, x = reshape_112_cast)[name = tensor<string, []>("reduce_mean_84_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_56_cast = sub(x = reshape_112_cast, y = reduce_mean_84_cast)[name = tensor<string, []>("sub_56_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_28_cast = square(x = sub_56_cast)[name = tensor<string, []>("square_28_cast")];
            tensor<int32, [3]> reduce_mean_86_axes_0 = const()[name = tensor<string, []>("reduce_mean_86_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_86_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_86_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_86_cast = reduce_mean(axes = reduce_mean_86_axes_0, keep_dims = reduce_mean_86_keep_dims_0, x = square_28_cast)[name = tensor<string, []>("reduce_mean_86_cast")];
            tensor<fp16, []> add_56_y_0_to_fp16 = const()[name = tensor<string, []>("add_56_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_56_cast = add(x = reduce_mean_86_cast, y = add_56_y_0_to_fp16)[name = tensor<string, []>("add_56_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_28_cast = sqrt(x = add_56_cast)[name = tensor<string, []>("sqrt_28_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_28_cast = real_div(x = sub_56_cast, y = sqrt_28_cast)[name = tensor<string, []>("real_div_28_cast")];
            tensor<int32, [4]> reshape_113_shape_0 = const()[name = tensor<string, []>("reshape_113_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_113_cast = reshape(shape = reshape_113_shape_0, x = real_div_28_cast)[name = tensor<string, []>("reshape_113_cast")];
            tensor<fp16, [1280]> add_57_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_57_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4134389248)))];
            tensor<fp16, [1280]> add_57_beta_0_to_fp16 = const()[name = tensor<string, []>("add_57_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4134391872)))];
            tensor<fp16, []> add_57_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_57_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 32, 32]> add_57_cast = batch_norm(beta = add_57_beta_0_to_fp16, epsilon = add_57_epsilon_0_to_fp16, gamma = add_57_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_113_cast)[name = tensor<string, []>("add_57_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_633_cast = silu(x = add_57_cast)[name = tensor<string, []>("input_633_cast")];
            tensor<int32, [2]> var_48854 = const()[name = tensor<string, []>("op_48854"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_48856 = const()[name = tensor<string, []>("op_48856"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_429_pad_type_0 = const()[name = tensor<string, []>("hidden_states_429_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_429_pad_0 = const()[name = tensor<string, []>("hidden_states_429_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> up_blocks_0_resnets_2_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_2_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4134394496)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_2_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_2_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4163885760)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_429_cast = conv(bias = up_blocks_0_resnets_2_conv2_bias_to_fp16, dilations = var_48856, groups = var_30385, pad = hidden_states_429_pad_0, pad_type = hidden_states_429_pad_type_0, strides = var_48854, weight = up_blocks_0_resnets_2_conv2_weight_to_fp16, x = input_633_cast)[name = tensor<string, []>("hidden_states_429_cast")];
            tensor<int32, [2]> var_48861 = const()[name = tensor<string, []>("op_48861"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_48863 = const()[name = tensor<string, []>("op_48863"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_9_pad_type_0 = const()[name = tensor<string, []>("x_9_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_9_pad_0 = const()[name = tensor<string, []>("x_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1920, 1, 1]> up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [1280, 1920, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4163888384)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4168803648)))];
            tensor<fp16, [2, 1280, 32, 32]> x_9_cast = conv(bias = up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_48863, groups = var_30385, pad = x_9_pad_0, pad_type = x_9_pad_type_0, strides = var_48861, weight = up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16, x = input_621_cast)[name = tensor<string, []>("x_9_cast")];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_431_cast = add(x = x_9_cast, y = hidden_states_429_cast)[name = tensor<string, []>("hidden_states_431_cast")];
            tensor<int32, [5]> reshape_116_shape_0 = const()[name = tensor<string, []>("reshape_116_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_116_cast = reshape(shape = reshape_116_shape_0, x = hidden_states_431_cast)[name = tensor<string, []>("reshape_116_cast")];
            tensor<int32, [3]> reduce_mean_87_axes_0 = const()[name = tensor<string, []>("reduce_mean_87_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_87_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_87_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_87_cast = reduce_mean(axes = reduce_mean_87_axes_0, keep_dims = reduce_mean_87_keep_dims_0, x = reshape_116_cast)[name = tensor<string, []>("reduce_mean_87_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_58_cast = sub(x = reshape_116_cast, y = reduce_mean_87_cast)[name = tensor<string, []>("sub_58_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_29_cast = square(x = sub_58_cast)[name = tensor<string, []>("square_29_cast")];
            tensor<int32, [3]> reduce_mean_89_axes_0 = const()[name = tensor<string, []>("reduce_mean_89_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_89_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_89_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_89_cast = reduce_mean(axes = reduce_mean_89_axes_0, keep_dims = reduce_mean_89_keep_dims_0, x = square_29_cast)[name = tensor<string, []>("reduce_mean_89_cast")];
            tensor<fp16, []> add_58_y_0_to_fp16 = const()[name = tensor<string, []>("add_58_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_58_cast = add(x = reduce_mean_89_cast, y = add_58_y_0_to_fp16)[name = tensor<string, []>("add_58_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_29_cast = sqrt(x = add_58_cast)[name = tensor<string, []>("sqrt_29_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_29_cast = real_div(x = sub_58_cast, y = sqrt_29_cast)[name = tensor<string, []>("real_div_29_cast")];
            tensor<int32, [4]> reshape_117_shape_0 = const()[name = tensor<string, []>("reshape_117_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_117_cast = reshape(shape = reshape_117_shape_0, x = real_div_29_cast)[name = tensor<string, []>("reshape_117_cast")];
            tensor<fp16, [1280]> add_59_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_59_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4168806272)))];
            tensor<fp16, [1280]> add_59_beta_0_to_fp16 = const()[name = tensor<string, []>("add_59_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4168808896)))];
            tensor<fp16, []> add_59_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_59_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 32, 32]> add_59_cast = batch_norm(beta = add_59_beta_0_to_fp16, epsilon = add_59_epsilon_0_to_fp16, gamma = add_59_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_117_cast)[name = tensor<string, []>("add_59_cast")];
            tensor<int32, [2]> var_48901 = const()[name = tensor<string, []>("op_48901"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_48903 = const()[name = tensor<string, []>("op_48903"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_433_pad_type_0 = const()[name = tensor<string, []>("hidden_states_433_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_433_pad_0 = const()[name = tensor<string, []>("hidden_states_433_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_proj_in_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4168811520)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_proj_in_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4172088384)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_433_cast = conv(bias = up_blocks_0_attentions_2_proj_in_bias_to_fp16, dilations = var_48903, groups = var_30385, pad = hidden_states_433_pad_0, pad_type = hidden_states_433_pad_type_0, strides = var_48901, weight = up_blocks_0_attentions_2_proj_in_weight_to_fp16, x = add_59_cast)[name = tensor<string, []>("hidden_states_433_cast")];
            tensor<int32, [4]> var_48908 = const()[name = tensor<string, []>("op_48908"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_325_cast = reshape(shape = var_48908, x = hidden_states_433_cast)[name = tensor<string, []>("inputs_325_cast")];
            tensor<int32, [1]> var_48918 = const()[name = tensor<string, []>("op_48918"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_325_cast = reduce_mean(axes = var_48918, keep_dims = var_30380, x = inputs_325_cast)[name = tensor<string, []>("channels_mean_325_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_325_cast = sub(x = inputs_325_cast, y = channels_mean_325_cast)[name = tensor<string, []>("zero_mean_325_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_325_cast = mul(x = zero_mean_325_cast, y = zero_mean_325_cast)[name = tensor<string, []>("zero_mean_sq_325_cast")];
            tensor<int32, [1]> var_48922 = const()[name = tensor<string, []>("op_48922"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_48923_cast = reduce_mean(axes = var_48922, keep_dims = var_30380, x = zero_mean_sq_325_cast)[name = tensor<string, []>("op_48923_cast")];
            tensor<fp16, []> var_48924_to_fp16 = const()[name = tensor<string, []>("op_48924_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_48925_cast = add(x = var_48923_cast, y = var_48924_to_fp16)[name = tensor<string, []>("op_48925_cast")];
            tensor<fp16, []> denom_325_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_325_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_325_cast = rsqrt(epsilon = denom_325_epsilon_0_to_fp16, x = var_48925_cast)[name = tensor<string, []>("denom_325_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_325_cast = mul(x = zero_mean_325_cast, y = denom_325_cast)[name = tensor<string, []>("out_325_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_48929_to_fp16 = const()[name = tensor<string, []>("op_48929_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4172091008)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_48930_cast = add(x = out_325_cast, y = var_48929_to_fp16)[name = tensor<string, []>("op_48930_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_48932_to_fp16 = const()[name = tensor<string, []>("op_48932_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4172093632)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_435_cast = mul(x = var_48930_cast, y = var_48932_to_fp16)[name = tensor<string, []>("hidden_states_435_cast")];
            tensor<int32, [2]> var_48939 = const()[name = tensor<string, []>("op_48939"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_48941 = const()[name = tensor<string, []>("op_48941"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_217_pad_type_0 = const()[name = tensor<string, []>("q_217_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_217_pad_0 = const()[name = tensor<string, []>("q_217_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4172096256)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_217_cast = conv(dilations = var_48941, groups = var_30385, pad = q_217_pad_0, pad_type = q_217_pad_type_0, strides = var_48939, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_435_cast)[name = tensor<string, []>("q_217_cast")];
            tensor<int32, [2]> var_48945 = const()[name = tensor<string, []>("op_48945"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_48947 = const()[name = tensor<string, []>("op_48947"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_433_pad_type_0 = const()[name = tensor<string, []>("k_433_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_433_pad_0 = const()[name = tensor<string, []>("k_433_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4175373120)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_433_cast = conv(dilations = var_48947, groups = var_30385, pad = k_433_pad_0, pad_type = k_433_pad_type_0, strides = var_48945, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_435_cast)[name = tensor<string, []>("k_433_cast")];
            tensor<int32, [2]> var_48951 = const()[name = tensor<string, []>("op_48951"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_48953 = const()[name = tensor<string, []>("op_48953"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_217_pad_type_0 = const()[name = tensor<string, []>("v_217_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_217_pad_0 = const()[name = tensor<string, []>("v_217_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4178649984)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_217_cast = conv(dilations = var_48953, groups = var_30385, pad = v_217_pad_0, pad_type = v_217_pad_type_0, strides = var_48951, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_435_cast)[name = tensor<string, []>("v_217_cast")];
            tensor<int32, [4]> var_48957_begin_0 = const()[name = tensor<string, []>("op_48957_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_48957_end_0 = const()[name = tensor<string, []>("op_48957_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_48957_end_mask_0 = const()[name = tensor<string, []>("op_48957_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48957_cast = slice_by_index(begin = var_48957_begin_0, end = var_48957_end_0, end_mask = var_48957_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_48957_cast")];
            tensor<int32, [4]> var_48961_begin_0 = const()[name = tensor<string, []>("op_48961_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_48961_end_0 = const()[name = tensor<string, []>("op_48961_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_48961_end_mask_0 = const()[name = tensor<string, []>("op_48961_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48961_cast = slice_by_index(begin = var_48961_begin_0, end = var_48961_end_0, end_mask = var_48961_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_48961_cast")];
            tensor<int32, [4]> var_48965_begin_0 = const()[name = tensor<string, []>("op_48965_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_48965_end_0 = const()[name = tensor<string, []>("op_48965_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_48965_end_mask_0 = const()[name = tensor<string, []>("op_48965_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48965_cast = slice_by_index(begin = var_48965_begin_0, end = var_48965_end_0, end_mask = var_48965_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_48965_cast")];
            tensor<int32, [4]> var_48969_begin_0 = const()[name = tensor<string, []>("op_48969_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_48969_end_0 = const()[name = tensor<string, []>("op_48969_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_48969_end_mask_0 = const()[name = tensor<string, []>("op_48969_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48969_cast = slice_by_index(begin = var_48969_begin_0, end = var_48969_end_0, end_mask = var_48969_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_48969_cast")];
            tensor<int32, [4]> var_48973_begin_0 = const()[name = tensor<string, []>("op_48973_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_48973_end_0 = const()[name = tensor<string, []>("op_48973_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_48973_end_mask_0 = const()[name = tensor<string, []>("op_48973_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48973_cast = slice_by_index(begin = var_48973_begin_0, end = var_48973_end_0, end_mask = var_48973_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_48973_cast")];
            tensor<int32, [4]> var_48977_begin_0 = const()[name = tensor<string, []>("op_48977_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_48977_end_0 = const()[name = tensor<string, []>("op_48977_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_48977_end_mask_0 = const()[name = tensor<string, []>("op_48977_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48977_cast = slice_by_index(begin = var_48977_begin_0, end = var_48977_end_0, end_mask = var_48977_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_48977_cast")];
            tensor<int32, [4]> var_48981_begin_0 = const()[name = tensor<string, []>("op_48981_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_48981_end_0 = const()[name = tensor<string, []>("op_48981_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_48981_end_mask_0 = const()[name = tensor<string, []>("op_48981_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48981_cast = slice_by_index(begin = var_48981_begin_0, end = var_48981_end_0, end_mask = var_48981_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_48981_cast")];
            tensor<int32, [4]> var_48985_begin_0 = const()[name = tensor<string, []>("op_48985_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_48985_end_0 = const()[name = tensor<string, []>("op_48985_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_48985_end_mask_0 = const()[name = tensor<string, []>("op_48985_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48985_cast = slice_by_index(begin = var_48985_begin_0, end = var_48985_end_0, end_mask = var_48985_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_48985_cast")];
            tensor<int32, [4]> var_48989_begin_0 = const()[name = tensor<string, []>("op_48989_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_48989_end_0 = const()[name = tensor<string, []>("op_48989_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_48989_end_mask_0 = const()[name = tensor<string, []>("op_48989_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48989_cast = slice_by_index(begin = var_48989_begin_0, end = var_48989_end_0, end_mask = var_48989_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_48989_cast")];
            tensor<int32, [4]> var_48993_begin_0 = const()[name = tensor<string, []>("op_48993_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_48993_end_0 = const()[name = tensor<string, []>("op_48993_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_48993_end_mask_0 = const()[name = tensor<string, []>("op_48993_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48993_cast = slice_by_index(begin = var_48993_begin_0, end = var_48993_end_0, end_mask = var_48993_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_48993_cast")];
            tensor<int32, [4]> var_48997_begin_0 = const()[name = tensor<string, []>("op_48997_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_48997_end_0 = const()[name = tensor<string, []>("op_48997_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_48997_end_mask_0 = const()[name = tensor<string, []>("op_48997_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_48997_cast = slice_by_index(begin = var_48997_begin_0, end = var_48997_end_0, end_mask = var_48997_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_48997_cast")];
            tensor<int32, [4]> var_49001_begin_0 = const()[name = tensor<string, []>("op_49001_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_49001_end_0 = const()[name = tensor<string, []>("op_49001_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_49001_end_mask_0 = const()[name = tensor<string, []>("op_49001_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49001_cast = slice_by_index(begin = var_49001_begin_0, end = var_49001_end_0, end_mask = var_49001_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_49001_cast")];
            tensor<int32, [4]> var_49005_begin_0 = const()[name = tensor<string, []>("op_49005_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_49005_end_0 = const()[name = tensor<string, []>("op_49005_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_49005_end_mask_0 = const()[name = tensor<string, []>("op_49005_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49005_cast = slice_by_index(begin = var_49005_begin_0, end = var_49005_end_0, end_mask = var_49005_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_49005_cast")];
            tensor<int32, [4]> var_49009_begin_0 = const()[name = tensor<string, []>("op_49009_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_49009_end_0 = const()[name = tensor<string, []>("op_49009_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_49009_end_mask_0 = const()[name = tensor<string, []>("op_49009_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49009_cast = slice_by_index(begin = var_49009_begin_0, end = var_49009_end_0, end_mask = var_49009_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_49009_cast")];
            tensor<int32, [4]> var_49013_begin_0 = const()[name = tensor<string, []>("op_49013_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_49013_end_0 = const()[name = tensor<string, []>("op_49013_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_49013_end_mask_0 = const()[name = tensor<string, []>("op_49013_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49013_cast = slice_by_index(begin = var_49013_begin_0, end = var_49013_end_0, end_mask = var_49013_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_49013_cast")];
            tensor<int32, [4]> var_49017_begin_0 = const()[name = tensor<string, []>("op_49017_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_49017_end_0 = const()[name = tensor<string, []>("op_49017_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_49017_end_mask_0 = const()[name = tensor<string, []>("op_49017_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49017_cast = slice_by_index(begin = var_49017_begin_0, end = var_49017_end_0, end_mask = var_49017_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_49017_cast")];
            tensor<int32, [4]> var_49021_begin_0 = const()[name = tensor<string, []>("op_49021_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_49021_end_0 = const()[name = tensor<string, []>("op_49021_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_49021_end_mask_0 = const()[name = tensor<string, []>("op_49021_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49021_cast = slice_by_index(begin = var_49021_begin_0, end = var_49021_end_0, end_mask = var_49021_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_49021_cast")];
            tensor<int32, [4]> var_49025_begin_0 = const()[name = tensor<string, []>("op_49025_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_49025_end_0 = const()[name = tensor<string, []>("op_49025_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_49025_end_mask_0 = const()[name = tensor<string, []>("op_49025_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49025_cast = slice_by_index(begin = var_49025_begin_0, end = var_49025_end_0, end_mask = var_49025_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_49025_cast")];
            tensor<int32, [4]> var_49029_begin_0 = const()[name = tensor<string, []>("op_49029_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_49029_end_0 = const()[name = tensor<string, []>("op_49029_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_49029_end_mask_0 = const()[name = tensor<string, []>("op_49029_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49029_cast = slice_by_index(begin = var_49029_begin_0, end = var_49029_end_0, end_mask = var_49029_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_49029_cast")];
            tensor<int32, [4]> var_49033_begin_0 = const()[name = tensor<string, []>("op_49033_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_49033_end_0 = const()[name = tensor<string, []>("op_49033_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_49033_end_mask_0 = const()[name = tensor<string, []>("op_49033_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49033_cast = slice_by_index(begin = var_49033_begin_0, end = var_49033_end_0, end_mask = var_49033_end_mask_0, x = q_217_cast)[name = tensor<string, []>("op_49033_cast")];
            tensor<int32, [4]> k_435_perm_0 = const()[name = tensor<string, []>("k_435_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_49040_begin_0 = const()[name = tensor<string, []>("op_49040_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_49040_end_0 = const()[name = tensor<string, []>("op_49040_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_49040_end_mask_0 = const()[name = tensor<string, []>("op_49040_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_31 = transpose(perm = k_435_perm_0, x = k_433_cast)[name = tensor<string, []>("transpose_31")];
            tensor<fp16, [2, 1024, 1, 64]> var_49040_cast = slice_by_index(begin = var_49040_begin_0, end = var_49040_end_0, end_mask = var_49040_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49040_cast")];
            tensor<int32, [4]> var_49044_begin_0 = const()[name = tensor<string, []>("op_49044_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_49044_end_0 = const()[name = tensor<string, []>("op_49044_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_49044_end_mask_0 = const()[name = tensor<string, []>("op_49044_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49044_cast = slice_by_index(begin = var_49044_begin_0, end = var_49044_end_0, end_mask = var_49044_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49044_cast")];
            tensor<int32, [4]> var_49048_begin_0 = const()[name = tensor<string, []>("op_49048_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_49048_end_0 = const()[name = tensor<string, []>("op_49048_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_49048_end_mask_0 = const()[name = tensor<string, []>("op_49048_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49048_cast = slice_by_index(begin = var_49048_begin_0, end = var_49048_end_0, end_mask = var_49048_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49048_cast")];
            tensor<int32, [4]> var_49052_begin_0 = const()[name = tensor<string, []>("op_49052_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_49052_end_0 = const()[name = tensor<string, []>("op_49052_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_49052_end_mask_0 = const()[name = tensor<string, []>("op_49052_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49052_cast = slice_by_index(begin = var_49052_begin_0, end = var_49052_end_0, end_mask = var_49052_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49052_cast")];
            tensor<int32, [4]> var_49056_begin_0 = const()[name = tensor<string, []>("op_49056_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_49056_end_0 = const()[name = tensor<string, []>("op_49056_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_49056_end_mask_0 = const()[name = tensor<string, []>("op_49056_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49056_cast = slice_by_index(begin = var_49056_begin_0, end = var_49056_end_0, end_mask = var_49056_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49056_cast")];
            tensor<int32, [4]> var_49060_begin_0 = const()[name = tensor<string, []>("op_49060_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_49060_end_0 = const()[name = tensor<string, []>("op_49060_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_49060_end_mask_0 = const()[name = tensor<string, []>("op_49060_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49060_cast = slice_by_index(begin = var_49060_begin_0, end = var_49060_end_0, end_mask = var_49060_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49060_cast")];
            tensor<int32, [4]> var_49064_begin_0 = const()[name = tensor<string, []>("op_49064_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_49064_end_0 = const()[name = tensor<string, []>("op_49064_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_49064_end_mask_0 = const()[name = tensor<string, []>("op_49064_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49064_cast = slice_by_index(begin = var_49064_begin_0, end = var_49064_end_0, end_mask = var_49064_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49064_cast")];
            tensor<int32, [4]> var_49068_begin_0 = const()[name = tensor<string, []>("op_49068_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_49068_end_0 = const()[name = tensor<string, []>("op_49068_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_49068_end_mask_0 = const()[name = tensor<string, []>("op_49068_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49068_cast = slice_by_index(begin = var_49068_begin_0, end = var_49068_end_0, end_mask = var_49068_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49068_cast")];
            tensor<int32, [4]> var_49072_begin_0 = const()[name = tensor<string, []>("op_49072_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_49072_end_0 = const()[name = tensor<string, []>("op_49072_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_49072_end_mask_0 = const()[name = tensor<string, []>("op_49072_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49072_cast = slice_by_index(begin = var_49072_begin_0, end = var_49072_end_0, end_mask = var_49072_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49072_cast")];
            tensor<int32, [4]> var_49076_begin_0 = const()[name = tensor<string, []>("op_49076_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_49076_end_0 = const()[name = tensor<string, []>("op_49076_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_49076_end_mask_0 = const()[name = tensor<string, []>("op_49076_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49076_cast = slice_by_index(begin = var_49076_begin_0, end = var_49076_end_0, end_mask = var_49076_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49076_cast")];
            tensor<int32, [4]> var_49080_begin_0 = const()[name = tensor<string, []>("op_49080_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_49080_end_0 = const()[name = tensor<string, []>("op_49080_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_49080_end_mask_0 = const()[name = tensor<string, []>("op_49080_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49080_cast = slice_by_index(begin = var_49080_begin_0, end = var_49080_end_0, end_mask = var_49080_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49080_cast")];
            tensor<int32, [4]> var_49084_begin_0 = const()[name = tensor<string, []>("op_49084_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_49084_end_0 = const()[name = tensor<string, []>("op_49084_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_49084_end_mask_0 = const()[name = tensor<string, []>("op_49084_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49084_cast = slice_by_index(begin = var_49084_begin_0, end = var_49084_end_0, end_mask = var_49084_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49084_cast")];
            tensor<int32, [4]> var_49088_begin_0 = const()[name = tensor<string, []>("op_49088_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_49088_end_0 = const()[name = tensor<string, []>("op_49088_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_49088_end_mask_0 = const()[name = tensor<string, []>("op_49088_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49088_cast = slice_by_index(begin = var_49088_begin_0, end = var_49088_end_0, end_mask = var_49088_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49088_cast")];
            tensor<int32, [4]> var_49092_begin_0 = const()[name = tensor<string, []>("op_49092_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_49092_end_0 = const()[name = tensor<string, []>("op_49092_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_49092_end_mask_0 = const()[name = tensor<string, []>("op_49092_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49092_cast = slice_by_index(begin = var_49092_begin_0, end = var_49092_end_0, end_mask = var_49092_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49092_cast")];
            tensor<int32, [4]> var_49096_begin_0 = const()[name = tensor<string, []>("op_49096_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_49096_end_0 = const()[name = tensor<string, []>("op_49096_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_49096_end_mask_0 = const()[name = tensor<string, []>("op_49096_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49096_cast = slice_by_index(begin = var_49096_begin_0, end = var_49096_end_0, end_mask = var_49096_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49096_cast")];
            tensor<int32, [4]> var_49100_begin_0 = const()[name = tensor<string, []>("op_49100_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_49100_end_0 = const()[name = tensor<string, []>("op_49100_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_49100_end_mask_0 = const()[name = tensor<string, []>("op_49100_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49100_cast = slice_by_index(begin = var_49100_begin_0, end = var_49100_end_0, end_mask = var_49100_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49100_cast")];
            tensor<int32, [4]> var_49104_begin_0 = const()[name = tensor<string, []>("op_49104_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_49104_end_0 = const()[name = tensor<string, []>("op_49104_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_49104_end_mask_0 = const()[name = tensor<string, []>("op_49104_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49104_cast = slice_by_index(begin = var_49104_begin_0, end = var_49104_end_0, end_mask = var_49104_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49104_cast")];
            tensor<int32, [4]> var_49108_begin_0 = const()[name = tensor<string, []>("op_49108_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_49108_end_0 = const()[name = tensor<string, []>("op_49108_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_49108_end_mask_0 = const()[name = tensor<string, []>("op_49108_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49108_cast = slice_by_index(begin = var_49108_begin_0, end = var_49108_end_0, end_mask = var_49108_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49108_cast")];
            tensor<int32, [4]> var_49112_begin_0 = const()[name = tensor<string, []>("op_49112_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_49112_end_0 = const()[name = tensor<string, []>("op_49112_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_49112_end_mask_0 = const()[name = tensor<string, []>("op_49112_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49112_cast = slice_by_index(begin = var_49112_begin_0, end = var_49112_end_0, end_mask = var_49112_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49112_cast")];
            tensor<int32, [4]> var_49116_begin_0 = const()[name = tensor<string, []>("op_49116_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_49116_end_0 = const()[name = tensor<string, []>("op_49116_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_49116_end_mask_0 = const()[name = tensor<string, []>("op_49116_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49116_cast = slice_by_index(begin = var_49116_begin_0, end = var_49116_end_0, end_mask = var_49116_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_49116_cast")];
            tensor<int32, [4]> var_49118_begin_0 = const()[name = tensor<string, []>("op_49118_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_49118_end_0 = const()[name = tensor<string, []>("op_49118_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_49118_end_mask_0 = const()[name = tensor<string, []>("op_49118_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49118_cast = slice_by_index(begin = var_49118_begin_0, end = var_49118_end_0, end_mask = var_49118_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49118_cast")];
            tensor<int32, [4]> var_49122_begin_0 = const()[name = tensor<string, []>("op_49122_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_49122_end_0 = const()[name = tensor<string, []>("op_49122_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_49122_end_mask_0 = const()[name = tensor<string, []>("op_49122_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49122_cast = slice_by_index(begin = var_49122_begin_0, end = var_49122_end_0, end_mask = var_49122_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49122_cast")];
            tensor<int32, [4]> var_49126_begin_0 = const()[name = tensor<string, []>("op_49126_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_49126_end_0 = const()[name = tensor<string, []>("op_49126_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_49126_end_mask_0 = const()[name = tensor<string, []>("op_49126_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49126_cast = slice_by_index(begin = var_49126_begin_0, end = var_49126_end_0, end_mask = var_49126_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49126_cast")];
            tensor<int32, [4]> var_49130_begin_0 = const()[name = tensor<string, []>("op_49130_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_49130_end_0 = const()[name = tensor<string, []>("op_49130_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_49130_end_mask_0 = const()[name = tensor<string, []>("op_49130_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49130_cast = slice_by_index(begin = var_49130_begin_0, end = var_49130_end_0, end_mask = var_49130_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49130_cast")];
            tensor<int32, [4]> var_49134_begin_0 = const()[name = tensor<string, []>("op_49134_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_49134_end_0 = const()[name = tensor<string, []>("op_49134_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_49134_end_mask_0 = const()[name = tensor<string, []>("op_49134_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49134_cast = slice_by_index(begin = var_49134_begin_0, end = var_49134_end_0, end_mask = var_49134_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49134_cast")];
            tensor<int32, [4]> var_49138_begin_0 = const()[name = tensor<string, []>("op_49138_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_49138_end_0 = const()[name = tensor<string, []>("op_49138_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_49138_end_mask_0 = const()[name = tensor<string, []>("op_49138_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49138_cast = slice_by_index(begin = var_49138_begin_0, end = var_49138_end_0, end_mask = var_49138_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49138_cast")];
            tensor<int32, [4]> var_49142_begin_0 = const()[name = tensor<string, []>("op_49142_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_49142_end_0 = const()[name = tensor<string, []>("op_49142_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_49142_end_mask_0 = const()[name = tensor<string, []>("op_49142_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49142_cast = slice_by_index(begin = var_49142_begin_0, end = var_49142_end_0, end_mask = var_49142_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49142_cast")];
            tensor<int32, [4]> var_49146_begin_0 = const()[name = tensor<string, []>("op_49146_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_49146_end_0 = const()[name = tensor<string, []>("op_49146_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_49146_end_mask_0 = const()[name = tensor<string, []>("op_49146_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49146_cast = slice_by_index(begin = var_49146_begin_0, end = var_49146_end_0, end_mask = var_49146_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49146_cast")];
            tensor<int32, [4]> var_49150_begin_0 = const()[name = tensor<string, []>("op_49150_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_49150_end_0 = const()[name = tensor<string, []>("op_49150_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_49150_end_mask_0 = const()[name = tensor<string, []>("op_49150_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49150_cast = slice_by_index(begin = var_49150_begin_0, end = var_49150_end_0, end_mask = var_49150_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49150_cast")];
            tensor<int32, [4]> var_49154_begin_0 = const()[name = tensor<string, []>("op_49154_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_49154_end_0 = const()[name = tensor<string, []>("op_49154_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_49154_end_mask_0 = const()[name = tensor<string, []>("op_49154_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49154_cast = slice_by_index(begin = var_49154_begin_0, end = var_49154_end_0, end_mask = var_49154_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49154_cast")];
            tensor<int32, [4]> var_49158_begin_0 = const()[name = tensor<string, []>("op_49158_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_49158_end_0 = const()[name = tensor<string, []>("op_49158_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_49158_end_mask_0 = const()[name = tensor<string, []>("op_49158_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49158_cast = slice_by_index(begin = var_49158_begin_0, end = var_49158_end_0, end_mask = var_49158_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49158_cast")];
            tensor<int32, [4]> var_49162_begin_0 = const()[name = tensor<string, []>("op_49162_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_49162_end_0 = const()[name = tensor<string, []>("op_49162_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_49162_end_mask_0 = const()[name = tensor<string, []>("op_49162_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49162_cast = slice_by_index(begin = var_49162_begin_0, end = var_49162_end_0, end_mask = var_49162_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49162_cast")];
            tensor<int32, [4]> var_49166_begin_0 = const()[name = tensor<string, []>("op_49166_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_49166_end_0 = const()[name = tensor<string, []>("op_49166_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_49166_end_mask_0 = const()[name = tensor<string, []>("op_49166_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49166_cast = slice_by_index(begin = var_49166_begin_0, end = var_49166_end_0, end_mask = var_49166_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49166_cast")];
            tensor<int32, [4]> var_49170_begin_0 = const()[name = tensor<string, []>("op_49170_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_49170_end_0 = const()[name = tensor<string, []>("op_49170_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_49170_end_mask_0 = const()[name = tensor<string, []>("op_49170_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49170_cast = slice_by_index(begin = var_49170_begin_0, end = var_49170_end_0, end_mask = var_49170_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49170_cast")];
            tensor<int32, [4]> var_49174_begin_0 = const()[name = tensor<string, []>("op_49174_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_49174_end_0 = const()[name = tensor<string, []>("op_49174_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_49174_end_mask_0 = const()[name = tensor<string, []>("op_49174_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49174_cast = slice_by_index(begin = var_49174_begin_0, end = var_49174_end_0, end_mask = var_49174_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49174_cast")];
            tensor<int32, [4]> var_49178_begin_0 = const()[name = tensor<string, []>("op_49178_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_49178_end_0 = const()[name = tensor<string, []>("op_49178_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_49178_end_mask_0 = const()[name = tensor<string, []>("op_49178_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49178_cast = slice_by_index(begin = var_49178_begin_0, end = var_49178_end_0, end_mask = var_49178_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49178_cast")];
            tensor<int32, [4]> var_49182_begin_0 = const()[name = tensor<string, []>("op_49182_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_49182_end_0 = const()[name = tensor<string, []>("op_49182_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_49182_end_mask_0 = const()[name = tensor<string, []>("op_49182_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49182_cast = slice_by_index(begin = var_49182_begin_0, end = var_49182_end_0, end_mask = var_49182_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49182_cast")];
            tensor<int32, [4]> var_49186_begin_0 = const()[name = tensor<string, []>("op_49186_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_49186_end_0 = const()[name = tensor<string, []>("op_49186_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_49186_end_mask_0 = const()[name = tensor<string, []>("op_49186_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49186_cast = slice_by_index(begin = var_49186_begin_0, end = var_49186_end_0, end_mask = var_49186_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49186_cast")];
            tensor<int32, [4]> var_49190_begin_0 = const()[name = tensor<string, []>("op_49190_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_49190_end_0 = const()[name = tensor<string, []>("op_49190_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_49190_end_mask_0 = const()[name = tensor<string, []>("op_49190_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49190_cast = slice_by_index(begin = var_49190_begin_0, end = var_49190_end_0, end_mask = var_49190_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49190_cast")];
            tensor<int32, [4]> var_49194_begin_0 = const()[name = tensor<string, []>("op_49194_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_49194_end_0 = const()[name = tensor<string, []>("op_49194_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_49194_end_mask_0 = const()[name = tensor<string, []>("op_49194_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49194_cast = slice_by_index(begin = var_49194_begin_0, end = var_49194_end_0, end_mask = var_49194_end_mask_0, x = v_217_cast)[name = tensor<string, []>("op_49194_cast")];
            tensor<string, []> var_49198_equation_0 = const()[name = tensor<string, []>("op_49198_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49198_cast = einsum(equation = var_49198_equation_0, values = (var_49040_cast, var_48957_cast))[name = tensor<string, []>("op_49198_cast")];
            tensor<fp16, []> var_49199_to_fp16 = const()[name = tensor<string, []>("op_49199_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4161_cast = mul(x = var_49198_cast, y = var_49199_to_fp16)[name = tensor<string, []>("aw_4161_cast")];
            tensor<string, []> var_49202_equation_0 = const()[name = tensor<string, []>("op_49202_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49202_cast = einsum(equation = var_49202_equation_0, values = (var_49044_cast, var_48961_cast))[name = tensor<string, []>("op_49202_cast")];
            tensor<fp16, []> var_49203_to_fp16 = const()[name = tensor<string, []>("op_49203_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4163_cast = mul(x = var_49202_cast, y = var_49203_to_fp16)[name = tensor<string, []>("aw_4163_cast")];
            tensor<string, []> var_49206_equation_0 = const()[name = tensor<string, []>("op_49206_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49206_cast = einsum(equation = var_49206_equation_0, values = (var_49048_cast, var_48965_cast))[name = tensor<string, []>("op_49206_cast")];
            tensor<fp16, []> var_49207_to_fp16 = const()[name = tensor<string, []>("op_49207_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4165_cast = mul(x = var_49206_cast, y = var_49207_to_fp16)[name = tensor<string, []>("aw_4165_cast")];
            tensor<string, []> var_49210_equation_0 = const()[name = tensor<string, []>("op_49210_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49210_cast = einsum(equation = var_49210_equation_0, values = (var_49052_cast, var_48969_cast))[name = tensor<string, []>("op_49210_cast")];
            tensor<fp16, []> var_49211_to_fp16 = const()[name = tensor<string, []>("op_49211_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4167_cast = mul(x = var_49210_cast, y = var_49211_to_fp16)[name = tensor<string, []>("aw_4167_cast")];
            tensor<string, []> var_49214_equation_0 = const()[name = tensor<string, []>("op_49214_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49214_cast = einsum(equation = var_49214_equation_0, values = (var_49056_cast, var_48973_cast))[name = tensor<string, []>("op_49214_cast")];
            tensor<fp16, []> var_49215_to_fp16 = const()[name = tensor<string, []>("op_49215_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4169_cast = mul(x = var_49214_cast, y = var_49215_to_fp16)[name = tensor<string, []>("aw_4169_cast")];
            tensor<string, []> var_49218_equation_0 = const()[name = tensor<string, []>("op_49218_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49218_cast = einsum(equation = var_49218_equation_0, values = (var_49060_cast, var_48977_cast))[name = tensor<string, []>("op_49218_cast")];
            tensor<fp16, []> var_49219_to_fp16 = const()[name = tensor<string, []>("op_49219_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4171_cast = mul(x = var_49218_cast, y = var_49219_to_fp16)[name = tensor<string, []>("aw_4171_cast")];
            tensor<string, []> var_49222_equation_0 = const()[name = tensor<string, []>("op_49222_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49222_cast = einsum(equation = var_49222_equation_0, values = (var_49064_cast, var_48981_cast))[name = tensor<string, []>("op_49222_cast")];
            tensor<fp16, []> var_49223_to_fp16 = const()[name = tensor<string, []>("op_49223_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4173_cast = mul(x = var_49222_cast, y = var_49223_to_fp16)[name = tensor<string, []>("aw_4173_cast")];
            tensor<string, []> var_49226_equation_0 = const()[name = tensor<string, []>("op_49226_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49226_cast = einsum(equation = var_49226_equation_0, values = (var_49068_cast, var_48985_cast))[name = tensor<string, []>("op_49226_cast")];
            tensor<fp16, []> var_49227_to_fp16 = const()[name = tensor<string, []>("op_49227_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4175_cast = mul(x = var_49226_cast, y = var_49227_to_fp16)[name = tensor<string, []>("aw_4175_cast")];
            tensor<string, []> var_49230_equation_0 = const()[name = tensor<string, []>("op_49230_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49230_cast = einsum(equation = var_49230_equation_0, values = (var_49072_cast, var_48989_cast))[name = tensor<string, []>("op_49230_cast")];
            tensor<fp16, []> var_49231_to_fp16 = const()[name = tensor<string, []>("op_49231_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4177_cast = mul(x = var_49230_cast, y = var_49231_to_fp16)[name = tensor<string, []>("aw_4177_cast")];
            tensor<string, []> var_49234_equation_0 = const()[name = tensor<string, []>("op_49234_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49234_cast = einsum(equation = var_49234_equation_0, values = (var_49076_cast, var_48993_cast))[name = tensor<string, []>("op_49234_cast")];
            tensor<fp16, []> var_49235_to_fp16 = const()[name = tensor<string, []>("op_49235_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4179_cast = mul(x = var_49234_cast, y = var_49235_to_fp16)[name = tensor<string, []>("aw_4179_cast")];
            tensor<string, []> var_49238_equation_0 = const()[name = tensor<string, []>("op_49238_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49238_cast = einsum(equation = var_49238_equation_0, values = (var_49080_cast, var_48997_cast))[name = tensor<string, []>("op_49238_cast")];
            tensor<fp16, []> var_49239_to_fp16 = const()[name = tensor<string, []>("op_49239_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4181_cast = mul(x = var_49238_cast, y = var_49239_to_fp16)[name = tensor<string, []>("aw_4181_cast")];
            tensor<string, []> var_49242_equation_0 = const()[name = tensor<string, []>("op_49242_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49242_cast = einsum(equation = var_49242_equation_0, values = (var_49084_cast, var_49001_cast))[name = tensor<string, []>("op_49242_cast")];
            tensor<fp16, []> var_49243_to_fp16 = const()[name = tensor<string, []>("op_49243_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4183_cast = mul(x = var_49242_cast, y = var_49243_to_fp16)[name = tensor<string, []>("aw_4183_cast")];
            tensor<string, []> var_49246_equation_0 = const()[name = tensor<string, []>("op_49246_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49246_cast = einsum(equation = var_49246_equation_0, values = (var_49088_cast, var_49005_cast))[name = tensor<string, []>("op_49246_cast")];
            tensor<fp16, []> var_49247_to_fp16 = const()[name = tensor<string, []>("op_49247_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4185_cast = mul(x = var_49246_cast, y = var_49247_to_fp16)[name = tensor<string, []>("aw_4185_cast")];
            tensor<string, []> var_49250_equation_0 = const()[name = tensor<string, []>("op_49250_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49250_cast = einsum(equation = var_49250_equation_0, values = (var_49092_cast, var_49009_cast))[name = tensor<string, []>("op_49250_cast")];
            tensor<fp16, []> var_49251_to_fp16 = const()[name = tensor<string, []>("op_49251_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4187_cast = mul(x = var_49250_cast, y = var_49251_to_fp16)[name = tensor<string, []>("aw_4187_cast")];
            tensor<string, []> var_49254_equation_0 = const()[name = tensor<string, []>("op_49254_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49254_cast = einsum(equation = var_49254_equation_0, values = (var_49096_cast, var_49013_cast))[name = tensor<string, []>("op_49254_cast")];
            tensor<fp16, []> var_49255_to_fp16 = const()[name = tensor<string, []>("op_49255_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4189_cast = mul(x = var_49254_cast, y = var_49255_to_fp16)[name = tensor<string, []>("aw_4189_cast")];
            tensor<string, []> var_49258_equation_0 = const()[name = tensor<string, []>("op_49258_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49258_cast = einsum(equation = var_49258_equation_0, values = (var_49100_cast, var_49017_cast))[name = tensor<string, []>("op_49258_cast")];
            tensor<fp16, []> var_49259_to_fp16 = const()[name = tensor<string, []>("op_49259_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4191_cast = mul(x = var_49258_cast, y = var_49259_to_fp16)[name = tensor<string, []>("aw_4191_cast")];
            tensor<string, []> var_49262_equation_0 = const()[name = tensor<string, []>("op_49262_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49262_cast = einsum(equation = var_49262_equation_0, values = (var_49104_cast, var_49021_cast))[name = tensor<string, []>("op_49262_cast")];
            tensor<fp16, []> var_49263_to_fp16 = const()[name = tensor<string, []>("op_49263_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4193_cast = mul(x = var_49262_cast, y = var_49263_to_fp16)[name = tensor<string, []>("aw_4193_cast")];
            tensor<string, []> var_49266_equation_0 = const()[name = tensor<string, []>("op_49266_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49266_cast = einsum(equation = var_49266_equation_0, values = (var_49108_cast, var_49025_cast))[name = tensor<string, []>("op_49266_cast")];
            tensor<fp16, []> var_49267_to_fp16 = const()[name = tensor<string, []>("op_49267_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4195_cast = mul(x = var_49266_cast, y = var_49267_to_fp16)[name = tensor<string, []>("aw_4195_cast")];
            tensor<string, []> var_49270_equation_0 = const()[name = tensor<string, []>("op_49270_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49270_cast = einsum(equation = var_49270_equation_0, values = (var_49112_cast, var_49029_cast))[name = tensor<string, []>("op_49270_cast")];
            tensor<fp16, []> var_49271_to_fp16 = const()[name = tensor<string, []>("op_49271_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4197_cast = mul(x = var_49270_cast, y = var_49271_to_fp16)[name = tensor<string, []>("aw_4197_cast")];
            tensor<string, []> var_49274_equation_0 = const()[name = tensor<string, []>("op_49274_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49274_cast = einsum(equation = var_49274_equation_0, values = (var_49116_cast, var_49033_cast))[name = tensor<string, []>("op_49274_cast")];
            tensor<fp16, []> var_49275_to_fp16 = const()[name = tensor<string, []>("op_49275_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4199_cast = mul(x = var_49274_cast, y = var_49275_to_fp16)[name = tensor<string, []>("aw_4199_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49277_cast = softmax(axis = var_30385, x = aw_4161_cast)[name = tensor<string, []>("op_49277_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49278_cast = softmax(axis = var_30385, x = aw_4163_cast)[name = tensor<string, []>("op_49278_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49279_cast = softmax(axis = var_30385, x = aw_4165_cast)[name = tensor<string, []>("op_49279_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49280_cast = softmax(axis = var_30385, x = aw_4167_cast)[name = tensor<string, []>("op_49280_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49281_cast = softmax(axis = var_30385, x = aw_4169_cast)[name = tensor<string, []>("op_49281_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49282_cast = softmax(axis = var_30385, x = aw_4171_cast)[name = tensor<string, []>("op_49282_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49283_cast = softmax(axis = var_30385, x = aw_4173_cast)[name = tensor<string, []>("op_49283_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49284_cast = softmax(axis = var_30385, x = aw_4175_cast)[name = tensor<string, []>("op_49284_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49285_cast = softmax(axis = var_30385, x = aw_4177_cast)[name = tensor<string, []>("op_49285_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49286_cast = softmax(axis = var_30385, x = aw_4179_cast)[name = tensor<string, []>("op_49286_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49287_cast = softmax(axis = var_30385, x = aw_4181_cast)[name = tensor<string, []>("op_49287_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49288_cast = softmax(axis = var_30385, x = aw_4183_cast)[name = tensor<string, []>("op_49288_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49289_cast = softmax(axis = var_30385, x = aw_4185_cast)[name = tensor<string, []>("op_49289_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49290_cast = softmax(axis = var_30385, x = aw_4187_cast)[name = tensor<string, []>("op_49290_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49291_cast = softmax(axis = var_30385, x = aw_4189_cast)[name = tensor<string, []>("op_49291_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49292_cast = softmax(axis = var_30385, x = aw_4191_cast)[name = tensor<string, []>("op_49292_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49293_cast = softmax(axis = var_30385, x = aw_4193_cast)[name = tensor<string, []>("op_49293_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49294_cast = softmax(axis = var_30385, x = aw_4195_cast)[name = tensor<string, []>("op_49294_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49295_cast = softmax(axis = var_30385, x = aw_4197_cast)[name = tensor<string, []>("op_49295_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_49296_cast = softmax(axis = var_30385, x = aw_4199_cast)[name = tensor<string, []>("op_49296_cast")];
            tensor<string, []> var_49298_equation_0 = const()[name = tensor<string, []>("op_49298_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49298_cast = einsum(equation = var_49298_equation_0, values = (var_49118_cast, var_49277_cast))[name = tensor<string, []>("op_49298_cast")];
            tensor<string, []> var_49300_equation_0 = const()[name = tensor<string, []>("op_49300_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49300_cast = einsum(equation = var_49300_equation_0, values = (var_49122_cast, var_49278_cast))[name = tensor<string, []>("op_49300_cast")];
            tensor<string, []> var_49302_equation_0 = const()[name = tensor<string, []>("op_49302_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49302_cast = einsum(equation = var_49302_equation_0, values = (var_49126_cast, var_49279_cast))[name = tensor<string, []>("op_49302_cast")];
            tensor<string, []> var_49304_equation_0 = const()[name = tensor<string, []>("op_49304_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49304_cast = einsum(equation = var_49304_equation_0, values = (var_49130_cast, var_49280_cast))[name = tensor<string, []>("op_49304_cast")];
            tensor<string, []> var_49306_equation_0 = const()[name = tensor<string, []>("op_49306_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49306_cast = einsum(equation = var_49306_equation_0, values = (var_49134_cast, var_49281_cast))[name = tensor<string, []>("op_49306_cast")];
            tensor<string, []> var_49308_equation_0 = const()[name = tensor<string, []>("op_49308_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49308_cast = einsum(equation = var_49308_equation_0, values = (var_49138_cast, var_49282_cast))[name = tensor<string, []>("op_49308_cast")];
            tensor<string, []> var_49310_equation_0 = const()[name = tensor<string, []>("op_49310_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49310_cast = einsum(equation = var_49310_equation_0, values = (var_49142_cast, var_49283_cast))[name = tensor<string, []>("op_49310_cast")];
            tensor<string, []> var_49312_equation_0 = const()[name = tensor<string, []>("op_49312_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49312_cast = einsum(equation = var_49312_equation_0, values = (var_49146_cast, var_49284_cast))[name = tensor<string, []>("op_49312_cast")];
            tensor<string, []> var_49314_equation_0 = const()[name = tensor<string, []>("op_49314_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49314_cast = einsum(equation = var_49314_equation_0, values = (var_49150_cast, var_49285_cast))[name = tensor<string, []>("op_49314_cast")];
            tensor<string, []> var_49316_equation_0 = const()[name = tensor<string, []>("op_49316_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49316_cast = einsum(equation = var_49316_equation_0, values = (var_49154_cast, var_49286_cast))[name = tensor<string, []>("op_49316_cast")];
            tensor<string, []> var_49318_equation_0 = const()[name = tensor<string, []>("op_49318_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49318_cast = einsum(equation = var_49318_equation_0, values = (var_49158_cast, var_49287_cast))[name = tensor<string, []>("op_49318_cast")];
            tensor<string, []> var_49320_equation_0 = const()[name = tensor<string, []>("op_49320_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49320_cast = einsum(equation = var_49320_equation_0, values = (var_49162_cast, var_49288_cast))[name = tensor<string, []>("op_49320_cast")];
            tensor<string, []> var_49322_equation_0 = const()[name = tensor<string, []>("op_49322_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49322_cast = einsum(equation = var_49322_equation_0, values = (var_49166_cast, var_49289_cast))[name = tensor<string, []>("op_49322_cast")];
            tensor<string, []> var_49324_equation_0 = const()[name = tensor<string, []>("op_49324_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49324_cast = einsum(equation = var_49324_equation_0, values = (var_49170_cast, var_49290_cast))[name = tensor<string, []>("op_49324_cast")];
            tensor<string, []> var_49326_equation_0 = const()[name = tensor<string, []>("op_49326_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49326_cast = einsum(equation = var_49326_equation_0, values = (var_49174_cast, var_49291_cast))[name = tensor<string, []>("op_49326_cast")];
            tensor<string, []> var_49328_equation_0 = const()[name = tensor<string, []>("op_49328_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49328_cast = einsum(equation = var_49328_equation_0, values = (var_49178_cast, var_49292_cast))[name = tensor<string, []>("op_49328_cast")];
            tensor<string, []> var_49330_equation_0 = const()[name = tensor<string, []>("op_49330_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49330_cast = einsum(equation = var_49330_equation_0, values = (var_49182_cast, var_49293_cast))[name = tensor<string, []>("op_49330_cast")];
            tensor<string, []> var_49332_equation_0 = const()[name = tensor<string, []>("op_49332_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49332_cast = einsum(equation = var_49332_equation_0, values = (var_49186_cast, var_49294_cast))[name = tensor<string, []>("op_49332_cast")];
            tensor<string, []> var_49334_equation_0 = const()[name = tensor<string, []>("op_49334_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49334_cast = einsum(equation = var_49334_equation_0, values = (var_49190_cast, var_49295_cast))[name = tensor<string, []>("op_49334_cast")];
            tensor<string, []> var_49336_equation_0 = const()[name = tensor<string, []>("op_49336_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49336_cast = einsum(equation = var_49336_equation_0, values = (var_49194_cast, var_49296_cast))[name = tensor<string, []>("op_49336_cast")];
            tensor<bool, []> input_637_interleave_0 = const()[name = tensor<string, []>("input_637_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_637_cast = concat(axis = var_30385, interleave = input_637_interleave_0, values = (var_49298_cast, var_49300_cast, var_49302_cast, var_49304_cast, var_49306_cast, var_49308_cast, var_49310_cast, var_49312_cast, var_49314_cast, var_49316_cast, var_49318_cast, var_49320_cast, var_49322_cast, var_49324_cast, var_49326_cast, var_49328_cast, var_49330_cast, var_49332_cast, var_49334_cast, var_49336_cast))[name = tensor<string, []>("input_637_cast")];
            tensor<int32, [2]> var_49342 = const()[name = tensor<string, []>("op_49342"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_49344 = const()[name = tensor<string, []>("op_49344"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_49346_pad_type_0 = const()[name = tensor<string, []>("op_49346_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_49346_pad_0 = const()[name = tensor<string, []>("op_49346_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4181926848)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4185203712)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_49346_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_49344, groups = var_30385, pad = var_49346_pad_0, pad_type = var_49346_pad_type_0, strides = var_49342, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_637_cast)[name = tensor<string, []>("op_49346_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_327_cast = add(x = var_49346_cast, y = inputs_325_cast)[name = tensor<string, []>("inputs_327_cast")];
            tensor<int32, [1]> var_49350 = const()[name = tensor<string, []>("op_49350"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_327_cast = reduce_mean(axes = var_49350, keep_dims = var_30380, x = inputs_327_cast)[name = tensor<string, []>("channels_mean_327_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_327_cast = sub(x = inputs_327_cast, y = channels_mean_327_cast)[name = tensor<string, []>("zero_mean_327_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_327_cast = mul(x = zero_mean_327_cast, y = zero_mean_327_cast)[name = tensor<string, []>("zero_mean_sq_327_cast")];
            tensor<int32, [1]> var_49354 = const()[name = tensor<string, []>("op_49354"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_49355_cast = reduce_mean(axes = var_49354, keep_dims = var_30380, x = zero_mean_sq_327_cast)[name = tensor<string, []>("op_49355_cast")];
            tensor<fp16, []> var_49356_to_fp16 = const()[name = tensor<string, []>("op_49356_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_49357_cast = add(x = var_49355_cast, y = var_49356_to_fp16)[name = tensor<string, []>("op_49357_cast")];
            tensor<fp16, []> denom_327_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_327_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_327_cast = rsqrt(epsilon = denom_327_epsilon_0_to_fp16, x = var_49357_cast)[name = tensor<string, []>("denom_327_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_327_cast = mul(x = zero_mean_327_cast, y = denom_327_cast)[name = tensor<string, []>("out_327_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_49361_to_fp16 = const()[name = tensor<string, []>("op_49361_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4185206336)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_49362_cast = add(x = out_327_cast, y = var_49361_to_fp16)[name = tensor<string, []>("op_49362_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_49364_to_fp16 = const()[name = tensor<string, []>("op_49364_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4185208960)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_437_cast = mul(x = var_49362_cast, y = var_49364_to_fp16)[name = tensor<string, []>("hidden_states_437_cast")];
            tensor<int32, [2]> var_49371 = const()[name = tensor<string, []>("op_49371"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_49373 = const()[name = tensor<string, []>("op_49373"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_219_pad_type_0 = const()[name = tensor<string, []>("q_219_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_219_pad_0 = const()[name = tensor<string, []>("q_219_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4185211584)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_219_cast = conv(dilations = var_49373, groups = var_30385, pad = q_219_pad_0, pad_type = q_219_pad_type_0, strides = var_49371, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_437_cast)[name = tensor<string, []>("q_219_cast")];
            tensor<int32, [2]> var_49377 = const()[name = tensor<string, []>("op_49377"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_49379 = const()[name = tensor<string, []>("op_49379"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_437_pad_type_0 = const()[name = tensor<string, []>("k_437_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_437_pad_0 = const()[name = tensor<string, []>("k_437_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4188488448)))];
            tensor<fp16, [2, 1280, 1, 77]> k_437_cast = conv(dilations = var_49379, groups = var_30385, pad = k_437_pad_0, pad_type = k_437_pad_type_0, strides = var_49377, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_437_cast")];
            tensor<int32, [2]> var_49383 = const()[name = tensor<string, []>("op_49383"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_49385 = const()[name = tensor<string, []>("op_49385"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_219_pad_type_0 = const()[name = tensor<string, []>("v_219_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_219_pad_0 = const()[name = tensor<string, []>("v_219_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4193731392)))];
            tensor<fp16, [2, 1280, 1, 77]> v_219_cast = conv(dilations = var_49385, groups = var_30385, pad = v_219_pad_0, pad_type = v_219_pad_type_0, strides = var_49383, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_219_cast")];
            tensor<int32, [4]> var_49389_begin_0 = const()[name = tensor<string, []>("op_49389_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_49389_end_0 = const()[name = tensor<string, []>("op_49389_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_49389_end_mask_0 = const()[name = tensor<string, []>("op_49389_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49389_cast = slice_by_index(begin = var_49389_begin_0, end = var_49389_end_0, end_mask = var_49389_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49389_cast")];
            tensor<int32, [4]> var_49393_begin_0 = const()[name = tensor<string, []>("op_49393_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_49393_end_0 = const()[name = tensor<string, []>("op_49393_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_49393_end_mask_0 = const()[name = tensor<string, []>("op_49393_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49393_cast = slice_by_index(begin = var_49393_begin_0, end = var_49393_end_0, end_mask = var_49393_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49393_cast")];
            tensor<int32, [4]> var_49397_begin_0 = const()[name = tensor<string, []>("op_49397_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_49397_end_0 = const()[name = tensor<string, []>("op_49397_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_49397_end_mask_0 = const()[name = tensor<string, []>("op_49397_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49397_cast = slice_by_index(begin = var_49397_begin_0, end = var_49397_end_0, end_mask = var_49397_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49397_cast")];
            tensor<int32, [4]> var_49401_begin_0 = const()[name = tensor<string, []>("op_49401_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_49401_end_0 = const()[name = tensor<string, []>("op_49401_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_49401_end_mask_0 = const()[name = tensor<string, []>("op_49401_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49401_cast = slice_by_index(begin = var_49401_begin_0, end = var_49401_end_0, end_mask = var_49401_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49401_cast")];
            tensor<int32, [4]> var_49405_begin_0 = const()[name = tensor<string, []>("op_49405_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_49405_end_0 = const()[name = tensor<string, []>("op_49405_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_49405_end_mask_0 = const()[name = tensor<string, []>("op_49405_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49405_cast = slice_by_index(begin = var_49405_begin_0, end = var_49405_end_0, end_mask = var_49405_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49405_cast")];
            tensor<int32, [4]> var_49409_begin_0 = const()[name = tensor<string, []>("op_49409_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_49409_end_0 = const()[name = tensor<string, []>("op_49409_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_49409_end_mask_0 = const()[name = tensor<string, []>("op_49409_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49409_cast = slice_by_index(begin = var_49409_begin_0, end = var_49409_end_0, end_mask = var_49409_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49409_cast")];
            tensor<int32, [4]> var_49413_begin_0 = const()[name = tensor<string, []>("op_49413_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_49413_end_0 = const()[name = tensor<string, []>("op_49413_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_49413_end_mask_0 = const()[name = tensor<string, []>("op_49413_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49413_cast = slice_by_index(begin = var_49413_begin_0, end = var_49413_end_0, end_mask = var_49413_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49413_cast")];
            tensor<int32, [4]> var_49417_begin_0 = const()[name = tensor<string, []>("op_49417_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_49417_end_0 = const()[name = tensor<string, []>("op_49417_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_49417_end_mask_0 = const()[name = tensor<string, []>("op_49417_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49417_cast = slice_by_index(begin = var_49417_begin_0, end = var_49417_end_0, end_mask = var_49417_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49417_cast")];
            tensor<int32, [4]> var_49421_begin_0 = const()[name = tensor<string, []>("op_49421_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_49421_end_0 = const()[name = tensor<string, []>("op_49421_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_49421_end_mask_0 = const()[name = tensor<string, []>("op_49421_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49421_cast = slice_by_index(begin = var_49421_begin_0, end = var_49421_end_0, end_mask = var_49421_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49421_cast")];
            tensor<int32, [4]> var_49425_begin_0 = const()[name = tensor<string, []>("op_49425_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_49425_end_0 = const()[name = tensor<string, []>("op_49425_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_49425_end_mask_0 = const()[name = tensor<string, []>("op_49425_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49425_cast = slice_by_index(begin = var_49425_begin_0, end = var_49425_end_0, end_mask = var_49425_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49425_cast")];
            tensor<int32, [4]> var_49429_begin_0 = const()[name = tensor<string, []>("op_49429_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_49429_end_0 = const()[name = tensor<string, []>("op_49429_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_49429_end_mask_0 = const()[name = tensor<string, []>("op_49429_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49429_cast = slice_by_index(begin = var_49429_begin_0, end = var_49429_end_0, end_mask = var_49429_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49429_cast")];
            tensor<int32, [4]> var_49433_begin_0 = const()[name = tensor<string, []>("op_49433_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_49433_end_0 = const()[name = tensor<string, []>("op_49433_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_49433_end_mask_0 = const()[name = tensor<string, []>("op_49433_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49433_cast = slice_by_index(begin = var_49433_begin_0, end = var_49433_end_0, end_mask = var_49433_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49433_cast")];
            tensor<int32, [4]> var_49437_begin_0 = const()[name = tensor<string, []>("op_49437_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_49437_end_0 = const()[name = tensor<string, []>("op_49437_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_49437_end_mask_0 = const()[name = tensor<string, []>("op_49437_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49437_cast = slice_by_index(begin = var_49437_begin_0, end = var_49437_end_0, end_mask = var_49437_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49437_cast")];
            tensor<int32, [4]> var_49441_begin_0 = const()[name = tensor<string, []>("op_49441_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_49441_end_0 = const()[name = tensor<string, []>("op_49441_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_49441_end_mask_0 = const()[name = tensor<string, []>("op_49441_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49441_cast = slice_by_index(begin = var_49441_begin_0, end = var_49441_end_0, end_mask = var_49441_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49441_cast")];
            tensor<int32, [4]> var_49445_begin_0 = const()[name = tensor<string, []>("op_49445_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_49445_end_0 = const()[name = tensor<string, []>("op_49445_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_49445_end_mask_0 = const()[name = tensor<string, []>("op_49445_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49445_cast = slice_by_index(begin = var_49445_begin_0, end = var_49445_end_0, end_mask = var_49445_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49445_cast")];
            tensor<int32, [4]> var_49449_begin_0 = const()[name = tensor<string, []>("op_49449_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_49449_end_0 = const()[name = tensor<string, []>("op_49449_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_49449_end_mask_0 = const()[name = tensor<string, []>("op_49449_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49449_cast = slice_by_index(begin = var_49449_begin_0, end = var_49449_end_0, end_mask = var_49449_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49449_cast")];
            tensor<int32, [4]> var_49453_begin_0 = const()[name = tensor<string, []>("op_49453_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_49453_end_0 = const()[name = tensor<string, []>("op_49453_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_49453_end_mask_0 = const()[name = tensor<string, []>("op_49453_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49453_cast = slice_by_index(begin = var_49453_begin_0, end = var_49453_end_0, end_mask = var_49453_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49453_cast")];
            tensor<int32, [4]> var_49457_begin_0 = const()[name = tensor<string, []>("op_49457_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_49457_end_0 = const()[name = tensor<string, []>("op_49457_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_49457_end_mask_0 = const()[name = tensor<string, []>("op_49457_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49457_cast = slice_by_index(begin = var_49457_begin_0, end = var_49457_end_0, end_mask = var_49457_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49457_cast")];
            tensor<int32, [4]> var_49461_begin_0 = const()[name = tensor<string, []>("op_49461_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_49461_end_0 = const()[name = tensor<string, []>("op_49461_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_49461_end_mask_0 = const()[name = tensor<string, []>("op_49461_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49461_cast = slice_by_index(begin = var_49461_begin_0, end = var_49461_end_0, end_mask = var_49461_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49461_cast")];
            tensor<int32, [4]> var_49465_begin_0 = const()[name = tensor<string, []>("op_49465_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_49465_end_0 = const()[name = tensor<string, []>("op_49465_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_49465_end_mask_0 = const()[name = tensor<string, []>("op_49465_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49465_cast = slice_by_index(begin = var_49465_begin_0, end = var_49465_end_0, end_mask = var_49465_end_mask_0, x = q_219_cast)[name = tensor<string, []>("op_49465_cast")];
            tensor<int32, [4]> k_439_perm_0 = const()[name = tensor<string, []>("k_439_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_49472_begin_0 = const()[name = tensor<string, []>("op_49472_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_49472_end_0 = const()[name = tensor<string, []>("op_49472_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_49472_end_mask_0 = const()[name = tensor<string, []>("op_49472_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_30 = transpose(perm = k_439_perm_0, x = k_437_cast)[name = tensor<string, []>("transpose_30")];
            tensor<fp16, [2, 77, 1, 64]> var_49472_cast = slice_by_index(begin = var_49472_begin_0, end = var_49472_end_0, end_mask = var_49472_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49472_cast")];
            tensor<int32, [4]> var_49476_begin_0 = const()[name = tensor<string, []>("op_49476_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_49476_end_0 = const()[name = tensor<string, []>("op_49476_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_49476_end_mask_0 = const()[name = tensor<string, []>("op_49476_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49476_cast = slice_by_index(begin = var_49476_begin_0, end = var_49476_end_0, end_mask = var_49476_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49476_cast")];
            tensor<int32, [4]> var_49480_begin_0 = const()[name = tensor<string, []>("op_49480_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_49480_end_0 = const()[name = tensor<string, []>("op_49480_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_49480_end_mask_0 = const()[name = tensor<string, []>("op_49480_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49480_cast = slice_by_index(begin = var_49480_begin_0, end = var_49480_end_0, end_mask = var_49480_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49480_cast")];
            tensor<int32, [4]> var_49484_begin_0 = const()[name = tensor<string, []>("op_49484_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_49484_end_0 = const()[name = tensor<string, []>("op_49484_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_49484_end_mask_0 = const()[name = tensor<string, []>("op_49484_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49484_cast = slice_by_index(begin = var_49484_begin_0, end = var_49484_end_0, end_mask = var_49484_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49484_cast")];
            tensor<int32, [4]> var_49488_begin_0 = const()[name = tensor<string, []>("op_49488_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_49488_end_0 = const()[name = tensor<string, []>("op_49488_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_49488_end_mask_0 = const()[name = tensor<string, []>("op_49488_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49488_cast = slice_by_index(begin = var_49488_begin_0, end = var_49488_end_0, end_mask = var_49488_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49488_cast")];
            tensor<int32, [4]> var_49492_begin_0 = const()[name = tensor<string, []>("op_49492_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_49492_end_0 = const()[name = tensor<string, []>("op_49492_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_49492_end_mask_0 = const()[name = tensor<string, []>("op_49492_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49492_cast = slice_by_index(begin = var_49492_begin_0, end = var_49492_end_0, end_mask = var_49492_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49492_cast")];
            tensor<int32, [4]> var_49496_begin_0 = const()[name = tensor<string, []>("op_49496_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_49496_end_0 = const()[name = tensor<string, []>("op_49496_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_49496_end_mask_0 = const()[name = tensor<string, []>("op_49496_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49496_cast = slice_by_index(begin = var_49496_begin_0, end = var_49496_end_0, end_mask = var_49496_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49496_cast")];
            tensor<int32, [4]> var_49500_begin_0 = const()[name = tensor<string, []>("op_49500_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_49500_end_0 = const()[name = tensor<string, []>("op_49500_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_49500_end_mask_0 = const()[name = tensor<string, []>("op_49500_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49500_cast = slice_by_index(begin = var_49500_begin_0, end = var_49500_end_0, end_mask = var_49500_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49500_cast")];
            tensor<int32, [4]> var_49504_begin_0 = const()[name = tensor<string, []>("op_49504_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_49504_end_0 = const()[name = tensor<string, []>("op_49504_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_49504_end_mask_0 = const()[name = tensor<string, []>("op_49504_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49504_cast = slice_by_index(begin = var_49504_begin_0, end = var_49504_end_0, end_mask = var_49504_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49504_cast")];
            tensor<int32, [4]> var_49508_begin_0 = const()[name = tensor<string, []>("op_49508_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_49508_end_0 = const()[name = tensor<string, []>("op_49508_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_49508_end_mask_0 = const()[name = tensor<string, []>("op_49508_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49508_cast = slice_by_index(begin = var_49508_begin_0, end = var_49508_end_0, end_mask = var_49508_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49508_cast")];
            tensor<int32, [4]> var_49512_begin_0 = const()[name = tensor<string, []>("op_49512_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_49512_end_0 = const()[name = tensor<string, []>("op_49512_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_49512_end_mask_0 = const()[name = tensor<string, []>("op_49512_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49512_cast = slice_by_index(begin = var_49512_begin_0, end = var_49512_end_0, end_mask = var_49512_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49512_cast")];
            tensor<int32, [4]> var_49516_begin_0 = const()[name = tensor<string, []>("op_49516_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_49516_end_0 = const()[name = tensor<string, []>("op_49516_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_49516_end_mask_0 = const()[name = tensor<string, []>("op_49516_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49516_cast = slice_by_index(begin = var_49516_begin_0, end = var_49516_end_0, end_mask = var_49516_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49516_cast")];
            tensor<int32, [4]> var_49520_begin_0 = const()[name = tensor<string, []>("op_49520_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_49520_end_0 = const()[name = tensor<string, []>("op_49520_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_49520_end_mask_0 = const()[name = tensor<string, []>("op_49520_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49520_cast = slice_by_index(begin = var_49520_begin_0, end = var_49520_end_0, end_mask = var_49520_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49520_cast")];
            tensor<int32, [4]> var_49524_begin_0 = const()[name = tensor<string, []>("op_49524_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_49524_end_0 = const()[name = tensor<string, []>("op_49524_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_49524_end_mask_0 = const()[name = tensor<string, []>("op_49524_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49524_cast = slice_by_index(begin = var_49524_begin_0, end = var_49524_end_0, end_mask = var_49524_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49524_cast")];
            tensor<int32, [4]> var_49528_begin_0 = const()[name = tensor<string, []>("op_49528_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_49528_end_0 = const()[name = tensor<string, []>("op_49528_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_49528_end_mask_0 = const()[name = tensor<string, []>("op_49528_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49528_cast = slice_by_index(begin = var_49528_begin_0, end = var_49528_end_0, end_mask = var_49528_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49528_cast")];
            tensor<int32, [4]> var_49532_begin_0 = const()[name = tensor<string, []>("op_49532_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_49532_end_0 = const()[name = tensor<string, []>("op_49532_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_49532_end_mask_0 = const()[name = tensor<string, []>("op_49532_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49532_cast = slice_by_index(begin = var_49532_begin_0, end = var_49532_end_0, end_mask = var_49532_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49532_cast")];
            tensor<int32, [4]> var_49536_begin_0 = const()[name = tensor<string, []>("op_49536_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_49536_end_0 = const()[name = tensor<string, []>("op_49536_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_49536_end_mask_0 = const()[name = tensor<string, []>("op_49536_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49536_cast = slice_by_index(begin = var_49536_begin_0, end = var_49536_end_0, end_mask = var_49536_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49536_cast")];
            tensor<int32, [4]> var_49540_begin_0 = const()[name = tensor<string, []>("op_49540_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_49540_end_0 = const()[name = tensor<string, []>("op_49540_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_49540_end_mask_0 = const()[name = tensor<string, []>("op_49540_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49540_cast = slice_by_index(begin = var_49540_begin_0, end = var_49540_end_0, end_mask = var_49540_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49540_cast")];
            tensor<int32, [4]> var_49544_begin_0 = const()[name = tensor<string, []>("op_49544_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_49544_end_0 = const()[name = tensor<string, []>("op_49544_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_49544_end_mask_0 = const()[name = tensor<string, []>("op_49544_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49544_cast = slice_by_index(begin = var_49544_begin_0, end = var_49544_end_0, end_mask = var_49544_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49544_cast")];
            tensor<int32, [4]> var_49548_begin_0 = const()[name = tensor<string, []>("op_49548_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_49548_end_0 = const()[name = tensor<string, []>("op_49548_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_49548_end_mask_0 = const()[name = tensor<string, []>("op_49548_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_49548_cast = slice_by_index(begin = var_49548_begin_0, end = var_49548_end_0, end_mask = var_49548_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_49548_cast")];
            tensor<int32, [4]> var_49550_begin_0 = const()[name = tensor<string, []>("op_49550_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_49550_end_0 = const()[name = tensor<string, []>("op_49550_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_49550_end_mask_0 = const()[name = tensor<string, []>("op_49550_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49550_cast = slice_by_index(begin = var_49550_begin_0, end = var_49550_end_0, end_mask = var_49550_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49550_cast")];
            tensor<int32, [4]> var_49554_begin_0 = const()[name = tensor<string, []>("op_49554_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_49554_end_0 = const()[name = tensor<string, []>("op_49554_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_49554_end_mask_0 = const()[name = tensor<string, []>("op_49554_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49554_cast = slice_by_index(begin = var_49554_begin_0, end = var_49554_end_0, end_mask = var_49554_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49554_cast")];
            tensor<int32, [4]> var_49558_begin_0 = const()[name = tensor<string, []>("op_49558_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_49558_end_0 = const()[name = tensor<string, []>("op_49558_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_49558_end_mask_0 = const()[name = tensor<string, []>("op_49558_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49558_cast = slice_by_index(begin = var_49558_begin_0, end = var_49558_end_0, end_mask = var_49558_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49558_cast")];
            tensor<int32, [4]> var_49562_begin_0 = const()[name = tensor<string, []>("op_49562_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_49562_end_0 = const()[name = tensor<string, []>("op_49562_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_49562_end_mask_0 = const()[name = tensor<string, []>("op_49562_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49562_cast = slice_by_index(begin = var_49562_begin_0, end = var_49562_end_0, end_mask = var_49562_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49562_cast")];
            tensor<int32, [4]> var_49566_begin_0 = const()[name = tensor<string, []>("op_49566_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_49566_end_0 = const()[name = tensor<string, []>("op_49566_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_49566_end_mask_0 = const()[name = tensor<string, []>("op_49566_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49566_cast = slice_by_index(begin = var_49566_begin_0, end = var_49566_end_0, end_mask = var_49566_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49566_cast")];
            tensor<int32, [4]> var_49570_begin_0 = const()[name = tensor<string, []>("op_49570_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_49570_end_0 = const()[name = tensor<string, []>("op_49570_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_49570_end_mask_0 = const()[name = tensor<string, []>("op_49570_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49570_cast = slice_by_index(begin = var_49570_begin_0, end = var_49570_end_0, end_mask = var_49570_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49570_cast")];
            tensor<int32, [4]> var_49574_begin_0 = const()[name = tensor<string, []>("op_49574_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_49574_end_0 = const()[name = tensor<string, []>("op_49574_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_49574_end_mask_0 = const()[name = tensor<string, []>("op_49574_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49574_cast = slice_by_index(begin = var_49574_begin_0, end = var_49574_end_0, end_mask = var_49574_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49574_cast")];
            tensor<int32, [4]> var_49578_begin_0 = const()[name = tensor<string, []>("op_49578_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_49578_end_0 = const()[name = tensor<string, []>("op_49578_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_49578_end_mask_0 = const()[name = tensor<string, []>("op_49578_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49578_cast = slice_by_index(begin = var_49578_begin_0, end = var_49578_end_0, end_mask = var_49578_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49578_cast")];
            tensor<int32, [4]> var_49582_begin_0 = const()[name = tensor<string, []>("op_49582_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_49582_end_0 = const()[name = tensor<string, []>("op_49582_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_49582_end_mask_0 = const()[name = tensor<string, []>("op_49582_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49582_cast = slice_by_index(begin = var_49582_begin_0, end = var_49582_end_0, end_mask = var_49582_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49582_cast")];
            tensor<int32, [4]> var_49586_begin_0 = const()[name = tensor<string, []>("op_49586_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_49586_end_0 = const()[name = tensor<string, []>("op_49586_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_49586_end_mask_0 = const()[name = tensor<string, []>("op_49586_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49586_cast = slice_by_index(begin = var_49586_begin_0, end = var_49586_end_0, end_mask = var_49586_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49586_cast")];
            tensor<int32, [4]> var_49590_begin_0 = const()[name = tensor<string, []>("op_49590_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_49590_end_0 = const()[name = tensor<string, []>("op_49590_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_49590_end_mask_0 = const()[name = tensor<string, []>("op_49590_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49590_cast = slice_by_index(begin = var_49590_begin_0, end = var_49590_end_0, end_mask = var_49590_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49590_cast")];
            tensor<int32, [4]> var_49594_begin_0 = const()[name = tensor<string, []>("op_49594_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_49594_end_0 = const()[name = tensor<string, []>("op_49594_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_49594_end_mask_0 = const()[name = tensor<string, []>("op_49594_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49594_cast = slice_by_index(begin = var_49594_begin_0, end = var_49594_end_0, end_mask = var_49594_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49594_cast")];
            tensor<int32, [4]> var_49598_begin_0 = const()[name = tensor<string, []>("op_49598_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_49598_end_0 = const()[name = tensor<string, []>("op_49598_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_49598_end_mask_0 = const()[name = tensor<string, []>("op_49598_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49598_cast = slice_by_index(begin = var_49598_begin_0, end = var_49598_end_0, end_mask = var_49598_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49598_cast")];
            tensor<int32, [4]> var_49602_begin_0 = const()[name = tensor<string, []>("op_49602_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_49602_end_0 = const()[name = tensor<string, []>("op_49602_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_49602_end_mask_0 = const()[name = tensor<string, []>("op_49602_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49602_cast = slice_by_index(begin = var_49602_begin_0, end = var_49602_end_0, end_mask = var_49602_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49602_cast")];
            tensor<int32, [4]> var_49606_begin_0 = const()[name = tensor<string, []>("op_49606_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_49606_end_0 = const()[name = tensor<string, []>("op_49606_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_49606_end_mask_0 = const()[name = tensor<string, []>("op_49606_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49606_cast = slice_by_index(begin = var_49606_begin_0, end = var_49606_end_0, end_mask = var_49606_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49606_cast")];
            tensor<int32, [4]> var_49610_begin_0 = const()[name = tensor<string, []>("op_49610_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_49610_end_0 = const()[name = tensor<string, []>("op_49610_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_49610_end_mask_0 = const()[name = tensor<string, []>("op_49610_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49610_cast = slice_by_index(begin = var_49610_begin_0, end = var_49610_end_0, end_mask = var_49610_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49610_cast")];
            tensor<int32, [4]> var_49614_begin_0 = const()[name = tensor<string, []>("op_49614_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_49614_end_0 = const()[name = tensor<string, []>("op_49614_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_49614_end_mask_0 = const()[name = tensor<string, []>("op_49614_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49614_cast = slice_by_index(begin = var_49614_begin_0, end = var_49614_end_0, end_mask = var_49614_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49614_cast")];
            tensor<int32, [4]> var_49618_begin_0 = const()[name = tensor<string, []>("op_49618_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_49618_end_0 = const()[name = tensor<string, []>("op_49618_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_49618_end_mask_0 = const()[name = tensor<string, []>("op_49618_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49618_cast = slice_by_index(begin = var_49618_begin_0, end = var_49618_end_0, end_mask = var_49618_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49618_cast")];
            tensor<int32, [4]> var_49622_begin_0 = const()[name = tensor<string, []>("op_49622_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_49622_end_0 = const()[name = tensor<string, []>("op_49622_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_49622_end_mask_0 = const()[name = tensor<string, []>("op_49622_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49622_cast = slice_by_index(begin = var_49622_begin_0, end = var_49622_end_0, end_mask = var_49622_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49622_cast")];
            tensor<int32, [4]> var_49626_begin_0 = const()[name = tensor<string, []>("op_49626_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_49626_end_0 = const()[name = tensor<string, []>("op_49626_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_49626_end_mask_0 = const()[name = tensor<string, []>("op_49626_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_49626_cast = slice_by_index(begin = var_49626_begin_0, end = var_49626_end_0, end_mask = var_49626_end_mask_0, x = v_219_cast)[name = tensor<string, []>("op_49626_cast")];
            tensor<string, []> var_49630_equation_0 = const()[name = tensor<string, []>("op_49630_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49630_cast = einsum(equation = var_49630_equation_0, values = (var_49472_cast, var_49389_cast))[name = tensor<string, []>("op_49630_cast")];
            tensor<fp16, []> var_49631_to_fp16 = const()[name = tensor<string, []>("op_49631_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4201_cast = mul(x = var_49630_cast, y = var_49631_to_fp16)[name = tensor<string, []>("aw_4201_cast")];
            tensor<string, []> var_49634_equation_0 = const()[name = tensor<string, []>("op_49634_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49634_cast = einsum(equation = var_49634_equation_0, values = (var_49476_cast, var_49393_cast))[name = tensor<string, []>("op_49634_cast")];
            tensor<fp16, []> var_49635_to_fp16 = const()[name = tensor<string, []>("op_49635_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4203_cast = mul(x = var_49634_cast, y = var_49635_to_fp16)[name = tensor<string, []>("aw_4203_cast")];
            tensor<string, []> var_49638_equation_0 = const()[name = tensor<string, []>("op_49638_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49638_cast = einsum(equation = var_49638_equation_0, values = (var_49480_cast, var_49397_cast))[name = tensor<string, []>("op_49638_cast")];
            tensor<fp16, []> var_49639_to_fp16 = const()[name = tensor<string, []>("op_49639_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4205_cast = mul(x = var_49638_cast, y = var_49639_to_fp16)[name = tensor<string, []>("aw_4205_cast")];
            tensor<string, []> var_49642_equation_0 = const()[name = tensor<string, []>("op_49642_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49642_cast = einsum(equation = var_49642_equation_0, values = (var_49484_cast, var_49401_cast))[name = tensor<string, []>("op_49642_cast")];
            tensor<fp16, []> var_49643_to_fp16 = const()[name = tensor<string, []>("op_49643_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4207_cast = mul(x = var_49642_cast, y = var_49643_to_fp16)[name = tensor<string, []>("aw_4207_cast")];
            tensor<string, []> var_49646_equation_0 = const()[name = tensor<string, []>("op_49646_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49646_cast = einsum(equation = var_49646_equation_0, values = (var_49488_cast, var_49405_cast))[name = tensor<string, []>("op_49646_cast")];
            tensor<fp16, []> var_49647_to_fp16 = const()[name = tensor<string, []>("op_49647_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4209_cast = mul(x = var_49646_cast, y = var_49647_to_fp16)[name = tensor<string, []>("aw_4209_cast")];
            tensor<string, []> var_49650_equation_0 = const()[name = tensor<string, []>("op_49650_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49650_cast = einsum(equation = var_49650_equation_0, values = (var_49492_cast, var_49409_cast))[name = tensor<string, []>("op_49650_cast")];
            tensor<fp16, []> var_49651_to_fp16 = const()[name = tensor<string, []>("op_49651_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4211_cast = mul(x = var_49650_cast, y = var_49651_to_fp16)[name = tensor<string, []>("aw_4211_cast")];
            tensor<string, []> var_49654_equation_0 = const()[name = tensor<string, []>("op_49654_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49654_cast = einsum(equation = var_49654_equation_0, values = (var_49496_cast, var_49413_cast))[name = tensor<string, []>("op_49654_cast")];
            tensor<fp16, []> var_49655_to_fp16 = const()[name = tensor<string, []>("op_49655_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4213_cast = mul(x = var_49654_cast, y = var_49655_to_fp16)[name = tensor<string, []>("aw_4213_cast")];
            tensor<string, []> var_49658_equation_0 = const()[name = tensor<string, []>("op_49658_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49658_cast = einsum(equation = var_49658_equation_0, values = (var_49500_cast, var_49417_cast))[name = tensor<string, []>("op_49658_cast")];
            tensor<fp16, []> var_49659_to_fp16 = const()[name = tensor<string, []>("op_49659_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4215_cast = mul(x = var_49658_cast, y = var_49659_to_fp16)[name = tensor<string, []>("aw_4215_cast")];
            tensor<string, []> var_49662_equation_0 = const()[name = tensor<string, []>("op_49662_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49662_cast = einsum(equation = var_49662_equation_0, values = (var_49504_cast, var_49421_cast))[name = tensor<string, []>("op_49662_cast")];
            tensor<fp16, []> var_49663_to_fp16 = const()[name = tensor<string, []>("op_49663_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4217_cast = mul(x = var_49662_cast, y = var_49663_to_fp16)[name = tensor<string, []>("aw_4217_cast")];
            tensor<string, []> var_49666_equation_0 = const()[name = tensor<string, []>("op_49666_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49666_cast = einsum(equation = var_49666_equation_0, values = (var_49508_cast, var_49425_cast))[name = tensor<string, []>("op_49666_cast")];
            tensor<fp16, []> var_49667_to_fp16 = const()[name = tensor<string, []>("op_49667_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4219_cast = mul(x = var_49666_cast, y = var_49667_to_fp16)[name = tensor<string, []>("aw_4219_cast")];
            tensor<string, []> var_49670_equation_0 = const()[name = tensor<string, []>("op_49670_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49670_cast = einsum(equation = var_49670_equation_0, values = (var_49512_cast, var_49429_cast))[name = tensor<string, []>("op_49670_cast")];
            tensor<fp16, []> var_49671_to_fp16 = const()[name = tensor<string, []>("op_49671_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4221_cast = mul(x = var_49670_cast, y = var_49671_to_fp16)[name = tensor<string, []>("aw_4221_cast")];
            tensor<string, []> var_49674_equation_0 = const()[name = tensor<string, []>("op_49674_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49674_cast = einsum(equation = var_49674_equation_0, values = (var_49516_cast, var_49433_cast))[name = tensor<string, []>("op_49674_cast")];
            tensor<fp16, []> var_49675_to_fp16 = const()[name = tensor<string, []>("op_49675_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4223_cast = mul(x = var_49674_cast, y = var_49675_to_fp16)[name = tensor<string, []>("aw_4223_cast")];
            tensor<string, []> var_49678_equation_0 = const()[name = tensor<string, []>("op_49678_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49678_cast = einsum(equation = var_49678_equation_0, values = (var_49520_cast, var_49437_cast))[name = tensor<string, []>("op_49678_cast")];
            tensor<fp16, []> var_49679_to_fp16 = const()[name = tensor<string, []>("op_49679_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4225_cast = mul(x = var_49678_cast, y = var_49679_to_fp16)[name = tensor<string, []>("aw_4225_cast")];
            tensor<string, []> var_49682_equation_0 = const()[name = tensor<string, []>("op_49682_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49682_cast = einsum(equation = var_49682_equation_0, values = (var_49524_cast, var_49441_cast))[name = tensor<string, []>("op_49682_cast")];
            tensor<fp16, []> var_49683_to_fp16 = const()[name = tensor<string, []>("op_49683_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4227_cast = mul(x = var_49682_cast, y = var_49683_to_fp16)[name = tensor<string, []>("aw_4227_cast")];
            tensor<string, []> var_49686_equation_0 = const()[name = tensor<string, []>("op_49686_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49686_cast = einsum(equation = var_49686_equation_0, values = (var_49528_cast, var_49445_cast))[name = tensor<string, []>("op_49686_cast")];
            tensor<fp16, []> var_49687_to_fp16 = const()[name = tensor<string, []>("op_49687_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4229_cast = mul(x = var_49686_cast, y = var_49687_to_fp16)[name = tensor<string, []>("aw_4229_cast")];
            tensor<string, []> var_49690_equation_0 = const()[name = tensor<string, []>("op_49690_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49690_cast = einsum(equation = var_49690_equation_0, values = (var_49532_cast, var_49449_cast))[name = tensor<string, []>("op_49690_cast")];
            tensor<fp16, []> var_49691_to_fp16 = const()[name = tensor<string, []>("op_49691_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4231_cast = mul(x = var_49690_cast, y = var_49691_to_fp16)[name = tensor<string, []>("aw_4231_cast")];
            tensor<string, []> var_49694_equation_0 = const()[name = tensor<string, []>("op_49694_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49694_cast = einsum(equation = var_49694_equation_0, values = (var_49536_cast, var_49453_cast))[name = tensor<string, []>("op_49694_cast")];
            tensor<fp16, []> var_49695_to_fp16 = const()[name = tensor<string, []>("op_49695_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4233_cast = mul(x = var_49694_cast, y = var_49695_to_fp16)[name = tensor<string, []>("aw_4233_cast")];
            tensor<string, []> var_49698_equation_0 = const()[name = tensor<string, []>("op_49698_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49698_cast = einsum(equation = var_49698_equation_0, values = (var_49540_cast, var_49457_cast))[name = tensor<string, []>("op_49698_cast")];
            tensor<fp16, []> var_49699_to_fp16 = const()[name = tensor<string, []>("op_49699_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4235_cast = mul(x = var_49698_cast, y = var_49699_to_fp16)[name = tensor<string, []>("aw_4235_cast")];
            tensor<string, []> var_49702_equation_0 = const()[name = tensor<string, []>("op_49702_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49702_cast = einsum(equation = var_49702_equation_0, values = (var_49544_cast, var_49461_cast))[name = tensor<string, []>("op_49702_cast")];
            tensor<fp16, []> var_49703_to_fp16 = const()[name = tensor<string, []>("op_49703_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4237_cast = mul(x = var_49702_cast, y = var_49703_to_fp16)[name = tensor<string, []>("aw_4237_cast")];
            tensor<string, []> var_49706_equation_0 = const()[name = tensor<string, []>("op_49706_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_49706_cast = einsum(equation = var_49706_equation_0, values = (var_49548_cast, var_49465_cast))[name = tensor<string, []>("op_49706_cast")];
            tensor<fp16, []> var_49707_to_fp16 = const()[name = tensor<string, []>("op_49707_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4239_cast = mul(x = var_49706_cast, y = var_49707_to_fp16)[name = tensor<string, []>("aw_4239_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49709_cast = softmax(axis = var_30385, x = aw_4201_cast)[name = tensor<string, []>("op_49709_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49710_cast = softmax(axis = var_30385, x = aw_4203_cast)[name = tensor<string, []>("op_49710_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49711_cast = softmax(axis = var_30385, x = aw_4205_cast)[name = tensor<string, []>("op_49711_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49712_cast = softmax(axis = var_30385, x = aw_4207_cast)[name = tensor<string, []>("op_49712_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49713_cast = softmax(axis = var_30385, x = aw_4209_cast)[name = tensor<string, []>("op_49713_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49714_cast = softmax(axis = var_30385, x = aw_4211_cast)[name = tensor<string, []>("op_49714_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49715_cast = softmax(axis = var_30385, x = aw_4213_cast)[name = tensor<string, []>("op_49715_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49716_cast = softmax(axis = var_30385, x = aw_4215_cast)[name = tensor<string, []>("op_49716_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49717_cast = softmax(axis = var_30385, x = aw_4217_cast)[name = tensor<string, []>("op_49717_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49718_cast = softmax(axis = var_30385, x = aw_4219_cast)[name = tensor<string, []>("op_49718_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49719_cast = softmax(axis = var_30385, x = aw_4221_cast)[name = tensor<string, []>("op_49719_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49720_cast = softmax(axis = var_30385, x = aw_4223_cast)[name = tensor<string, []>("op_49720_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49721_cast = softmax(axis = var_30385, x = aw_4225_cast)[name = tensor<string, []>("op_49721_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49722_cast = softmax(axis = var_30385, x = aw_4227_cast)[name = tensor<string, []>("op_49722_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49723_cast = softmax(axis = var_30385, x = aw_4229_cast)[name = tensor<string, []>("op_49723_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49724_cast = softmax(axis = var_30385, x = aw_4231_cast)[name = tensor<string, []>("op_49724_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49725_cast = softmax(axis = var_30385, x = aw_4233_cast)[name = tensor<string, []>("op_49725_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49726_cast = softmax(axis = var_30385, x = aw_4235_cast)[name = tensor<string, []>("op_49726_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49727_cast = softmax(axis = var_30385, x = aw_4237_cast)[name = tensor<string, []>("op_49727_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_49728_cast = softmax(axis = var_30385, x = aw_4239_cast)[name = tensor<string, []>("op_49728_cast")];
            tensor<string, []> var_49730_equation_0 = const()[name = tensor<string, []>("op_49730_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49730_cast = einsum(equation = var_49730_equation_0, values = (var_49550_cast, var_49709_cast))[name = tensor<string, []>("op_49730_cast")];
            tensor<string, []> var_49732_equation_0 = const()[name = tensor<string, []>("op_49732_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49732_cast = einsum(equation = var_49732_equation_0, values = (var_49554_cast, var_49710_cast))[name = tensor<string, []>("op_49732_cast")];
            tensor<string, []> var_49734_equation_0 = const()[name = tensor<string, []>("op_49734_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49734_cast = einsum(equation = var_49734_equation_0, values = (var_49558_cast, var_49711_cast))[name = tensor<string, []>("op_49734_cast")];
            tensor<string, []> var_49736_equation_0 = const()[name = tensor<string, []>("op_49736_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49736_cast = einsum(equation = var_49736_equation_0, values = (var_49562_cast, var_49712_cast))[name = tensor<string, []>("op_49736_cast")];
            tensor<string, []> var_49738_equation_0 = const()[name = tensor<string, []>("op_49738_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49738_cast = einsum(equation = var_49738_equation_0, values = (var_49566_cast, var_49713_cast))[name = tensor<string, []>("op_49738_cast")];
            tensor<string, []> var_49740_equation_0 = const()[name = tensor<string, []>("op_49740_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49740_cast = einsum(equation = var_49740_equation_0, values = (var_49570_cast, var_49714_cast))[name = tensor<string, []>("op_49740_cast")];
            tensor<string, []> var_49742_equation_0 = const()[name = tensor<string, []>("op_49742_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49742_cast = einsum(equation = var_49742_equation_0, values = (var_49574_cast, var_49715_cast))[name = tensor<string, []>("op_49742_cast")];
            tensor<string, []> var_49744_equation_0 = const()[name = tensor<string, []>("op_49744_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49744_cast = einsum(equation = var_49744_equation_0, values = (var_49578_cast, var_49716_cast))[name = tensor<string, []>("op_49744_cast")];
            tensor<string, []> var_49746_equation_0 = const()[name = tensor<string, []>("op_49746_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49746_cast = einsum(equation = var_49746_equation_0, values = (var_49582_cast, var_49717_cast))[name = tensor<string, []>("op_49746_cast")];
            tensor<string, []> var_49748_equation_0 = const()[name = tensor<string, []>("op_49748_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49748_cast = einsum(equation = var_49748_equation_0, values = (var_49586_cast, var_49718_cast))[name = tensor<string, []>("op_49748_cast")];
            tensor<string, []> var_49750_equation_0 = const()[name = tensor<string, []>("op_49750_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49750_cast = einsum(equation = var_49750_equation_0, values = (var_49590_cast, var_49719_cast))[name = tensor<string, []>("op_49750_cast")];
            tensor<string, []> var_49752_equation_0 = const()[name = tensor<string, []>("op_49752_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49752_cast = einsum(equation = var_49752_equation_0, values = (var_49594_cast, var_49720_cast))[name = tensor<string, []>("op_49752_cast")];
            tensor<string, []> var_49754_equation_0 = const()[name = tensor<string, []>("op_49754_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49754_cast = einsum(equation = var_49754_equation_0, values = (var_49598_cast, var_49721_cast))[name = tensor<string, []>("op_49754_cast")];
            tensor<string, []> var_49756_equation_0 = const()[name = tensor<string, []>("op_49756_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49756_cast = einsum(equation = var_49756_equation_0, values = (var_49602_cast, var_49722_cast))[name = tensor<string, []>("op_49756_cast")];
            tensor<string, []> var_49758_equation_0 = const()[name = tensor<string, []>("op_49758_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49758_cast = einsum(equation = var_49758_equation_0, values = (var_49606_cast, var_49723_cast))[name = tensor<string, []>("op_49758_cast")];
            tensor<string, []> var_49760_equation_0 = const()[name = tensor<string, []>("op_49760_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49760_cast = einsum(equation = var_49760_equation_0, values = (var_49610_cast, var_49724_cast))[name = tensor<string, []>("op_49760_cast")];
            tensor<string, []> var_49762_equation_0 = const()[name = tensor<string, []>("op_49762_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49762_cast = einsum(equation = var_49762_equation_0, values = (var_49614_cast, var_49725_cast))[name = tensor<string, []>("op_49762_cast")];
            tensor<string, []> var_49764_equation_0 = const()[name = tensor<string, []>("op_49764_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49764_cast = einsum(equation = var_49764_equation_0, values = (var_49618_cast, var_49726_cast))[name = tensor<string, []>("op_49764_cast")];
            tensor<string, []> var_49766_equation_0 = const()[name = tensor<string, []>("op_49766_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49766_cast = einsum(equation = var_49766_equation_0, values = (var_49622_cast, var_49727_cast))[name = tensor<string, []>("op_49766_cast")];
            tensor<string, []> var_49768_equation_0 = const()[name = tensor<string, []>("op_49768_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_49768_cast = einsum(equation = var_49768_equation_0, values = (var_49626_cast, var_49728_cast))[name = tensor<string, []>("op_49768_cast")];
            tensor<bool, []> input_639_interleave_0 = const()[name = tensor<string, []>("input_639_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_639_cast = concat(axis = var_30385, interleave = input_639_interleave_0, values = (var_49730_cast, var_49732_cast, var_49734_cast, var_49736_cast, var_49738_cast, var_49740_cast, var_49742_cast, var_49744_cast, var_49746_cast, var_49748_cast, var_49750_cast, var_49752_cast, var_49754_cast, var_49756_cast, var_49758_cast, var_49760_cast, var_49762_cast, var_49764_cast, var_49766_cast, var_49768_cast))[name = tensor<string, []>("input_639_cast")];
            tensor<int32, [2]> var_49774 = const()[name = tensor<string, []>("op_49774"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_49776 = const()[name = tensor<string, []>("op_49776"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_49778_pad_type_0 = const()[name = tensor<string, []>("op_49778_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_49778_pad_0 = const()[name = tensor<string, []>("op_49778_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4198974336)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4202251200)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_49778_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_49776, groups = var_30385, pad = var_49778_pad_0, pad_type = var_49778_pad_type_0, strides = var_49774, weight = up_blocks_0_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_639_cast)[name = tensor<string, []>("op_49778_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_329_cast = add(x = var_49778_cast, y = inputs_327_cast)[name = tensor<string, []>("inputs_329_cast")];
            tensor<int32, [1]> var_49782 = const()[name = tensor<string, []>("op_49782"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_329_cast = reduce_mean(axes = var_49782, keep_dims = var_30380, x = inputs_329_cast)[name = tensor<string, []>("channels_mean_329_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_329_cast = sub(x = inputs_329_cast, y = channels_mean_329_cast)[name = tensor<string, []>("zero_mean_329_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_329_cast = mul(x = zero_mean_329_cast, y = zero_mean_329_cast)[name = tensor<string, []>("zero_mean_sq_329_cast")];
            tensor<int32, [1]> var_49786 = const()[name = tensor<string, []>("op_49786"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_49787_cast = reduce_mean(axes = var_49786, keep_dims = var_30380, x = zero_mean_sq_329_cast)[name = tensor<string, []>("op_49787_cast")];
            tensor<fp16, []> var_49788_to_fp16 = const()[name = tensor<string, []>("op_49788_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_49789_cast = add(x = var_49787_cast, y = var_49788_to_fp16)[name = tensor<string, []>("op_49789_cast")];
            tensor<fp16, []> denom_329_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_329_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_329_cast = rsqrt(epsilon = denom_329_epsilon_0_to_fp16, x = var_49789_cast)[name = tensor<string, []>("denom_329_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_329_cast = mul(x = zero_mean_329_cast, y = denom_329_cast)[name = tensor<string, []>("out_329_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_49793_to_fp16 = const()[name = tensor<string, []>("op_49793_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4202253824)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_49794_cast = add(x = out_329_cast, y = var_49793_to_fp16)[name = tensor<string, []>("op_49794_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_49796_to_fp16 = const()[name = tensor<string, []>("op_49796_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4202256448)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_641_cast = mul(x = var_49794_cast, y = var_49796_to_fp16)[name = tensor<string, []>("input_641_cast")];
            tensor<int32, [2]> var_49804 = const()[name = tensor<string, []>("op_49804"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_49806 = const()[name = tensor<string, []>("op_49806"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_49808_pad_type_0 = const()[name = tensor<string, []>("op_49808_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_49808_pad_0 = const()[name = tensor<string, []>("op_49808_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4202259072)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4228473536)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_49808_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_49806, groups = var_30385, pad = var_49808_pad_0, pad_type = var_49808_pad_type_0, strides = var_49804, weight = up_blocks_0_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_641_cast)[name = tensor<string, []>("op_49808_cast")];
            tensor<int32, [2]> var_49809_split_sizes_0 = const()[name = tensor<string, []>("op_49809_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_49809_axis_0 = const()[name = tensor<string, []>("op_49809_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_49809_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_49809_cast_1 = split(axis = var_49809_axis_0, split_sizes = var_49809_split_sizes_0, x = var_49808_cast)[name = tensor<string, []>("op_49809_cast")];
            tensor<string, []> var_49811_mode_0 = const()[name = tensor<string, []>("op_49811_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_49811_cast = gelu(mode = var_49811_mode_0, x = var_49809_cast_1)[name = tensor<string, []>("op_49811_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_643_cast = mul(x = var_49809_cast_0, y = var_49811_cast)[name = tensor<string, []>("input_643_cast")];
            tensor<int32, [2]> var_49815 = const()[name = tensor<string, []>("op_49815"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_49817 = const()[name = tensor<string, []>("op_49817"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_49819_pad_type_0 = const()[name = tensor<string, []>("op_49819_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_49819_pad_0 = const()[name = tensor<string, []>("op_49819_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4228494080)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4241601344)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_49819_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_49817, groups = var_30385, pad = var_49819_pad_0, pad_type = var_49819_pad_type_0, strides = var_49815, weight = up_blocks_0_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_643_cast)[name = tensor<string, []>("op_49819_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_331_cast = add(x = var_49819_cast, y = inputs_329_cast)[name = tensor<string, []>("inputs_331_cast")];
            tensor<int32, [1]> var_49829 = const()[name = tensor<string, []>("op_49829"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_331_cast = reduce_mean(axes = var_49829, keep_dims = var_30380, x = inputs_331_cast)[name = tensor<string, []>("channels_mean_331_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_331_cast = sub(x = inputs_331_cast, y = channels_mean_331_cast)[name = tensor<string, []>("zero_mean_331_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_331_cast = mul(x = zero_mean_331_cast, y = zero_mean_331_cast)[name = tensor<string, []>("zero_mean_sq_331_cast")];
            tensor<int32, [1]> var_49833 = const()[name = tensor<string, []>("op_49833"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_49834_cast = reduce_mean(axes = var_49833, keep_dims = var_30380, x = zero_mean_sq_331_cast)[name = tensor<string, []>("op_49834_cast")];
            tensor<fp16, []> var_49835_to_fp16 = const()[name = tensor<string, []>("op_49835_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_49836_cast = add(x = var_49834_cast, y = var_49835_to_fp16)[name = tensor<string, []>("op_49836_cast")];
            tensor<fp16, []> denom_331_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_331_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_331_cast = rsqrt(epsilon = denom_331_epsilon_0_to_fp16, x = var_49836_cast)[name = tensor<string, []>("denom_331_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_331_cast = mul(x = zero_mean_331_cast, y = denom_331_cast)[name = tensor<string, []>("out_331_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_49840_to_fp16 = const()[name = tensor<string, []>("op_49840_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4241603968)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_49841_cast = add(x = out_331_cast, y = var_49840_to_fp16)[name = tensor<string, []>("op_49841_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_49843_to_fp16 = const()[name = tensor<string, []>("op_49843_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4241606592)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_441_cast = mul(x = var_49841_cast, y = var_49843_to_fp16)[name = tensor<string, []>("hidden_states_441_cast")];
            tensor<int32, [2]> var_49850 = const()[name = tensor<string, []>("op_49850"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_49852 = const()[name = tensor<string, []>("op_49852"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_221_pad_type_0 = const()[name = tensor<string, []>("q_221_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_221_pad_0 = const()[name = tensor<string, []>("q_221_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4241609216)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_221_cast = conv(dilations = var_49852, groups = var_30385, pad = q_221_pad_0, pad_type = q_221_pad_type_0, strides = var_49850, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_441_cast)[name = tensor<string, []>("q_221_cast")];
            tensor<int32, [2]> var_49856 = const()[name = tensor<string, []>("op_49856"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_49858 = const()[name = tensor<string, []>("op_49858"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_441_pad_type_0 = const()[name = tensor<string, []>("k_441_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_441_pad_0 = const()[name = tensor<string, []>("k_441_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4244886080)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_441_cast = conv(dilations = var_49858, groups = var_30385, pad = k_441_pad_0, pad_type = k_441_pad_type_0, strides = var_49856, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_441_cast)[name = tensor<string, []>("k_441_cast")];
            tensor<int32, [2]> var_49862 = const()[name = tensor<string, []>("op_49862"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_49864 = const()[name = tensor<string, []>("op_49864"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_221_pad_type_0 = const()[name = tensor<string, []>("v_221_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_221_pad_0 = const()[name = tensor<string, []>("v_221_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4248162944)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_221_cast = conv(dilations = var_49864, groups = var_30385, pad = v_221_pad_0, pad_type = v_221_pad_type_0, strides = var_49862, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_441_cast)[name = tensor<string, []>("v_221_cast")];
            tensor<int32, [4]> var_49868_begin_0 = const()[name = tensor<string, []>("op_49868_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_49868_end_0 = const()[name = tensor<string, []>("op_49868_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_49868_end_mask_0 = const()[name = tensor<string, []>("op_49868_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49868_cast = slice_by_index(begin = var_49868_begin_0, end = var_49868_end_0, end_mask = var_49868_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49868_cast")];
            tensor<int32, [4]> var_49872_begin_0 = const()[name = tensor<string, []>("op_49872_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_49872_end_0 = const()[name = tensor<string, []>("op_49872_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_49872_end_mask_0 = const()[name = tensor<string, []>("op_49872_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49872_cast = slice_by_index(begin = var_49872_begin_0, end = var_49872_end_0, end_mask = var_49872_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49872_cast")];
            tensor<int32, [4]> var_49876_begin_0 = const()[name = tensor<string, []>("op_49876_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_49876_end_0 = const()[name = tensor<string, []>("op_49876_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_49876_end_mask_0 = const()[name = tensor<string, []>("op_49876_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49876_cast = slice_by_index(begin = var_49876_begin_0, end = var_49876_end_0, end_mask = var_49876_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49876_cast")];
            tensor<int32, [4]> var_49880_begin_0 = const()[name = tensor<string, []>("op_49880_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_49880_end_0 = const()[name = tensor<string, []>("op_49880_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_49880_end_mask_0 = const()[name = tensor<string, []>("op_49880_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49880_cast = slice_by_index(begin = var_49880_begin_0, end = var_49880_end_0, end_mask = var_49880_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49880_cast")];
            tensor<int32, [4]> var_49884_begin_0 = const()[name = tensor<string, []>("op_49884_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_49884_end_0 = const()[name = tensor<string, []>("op_49884_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_49884_end_mask_0 = const()[name = tensor<string, []>("op_49884_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49884_cast = slice_by_index(begin = var_49884_begin_0, end = var_49884_end_0, end_mask = var_49884_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49884_cast")];
            tensor<int32, [4]> var_49888_begin_0 = const()[name = tensor<string, []>("op_49888_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_49888_end_0 = const()[name = tensor<string, []>("op_49888_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_49888_end_mask_0 = const()[name = tensor<string, []>("op_49888_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49888_cast = slice_by_index(begin = var_49888_begin_0, end = var_49888_end_0, end_mask = var_49888_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49888_cast")];
            tensor<int32, [4]> var_49892_begin_0 = const()[name = tensor<string, []>("op_49892_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_49892_end_0 = const()[name = tensor<string, []>("op_49892_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_49892_end_mask_0 = const()[name = tensor<string, []>("op_49892_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49892_cast = slice_by_index(begin = var_49892_begin_0, end = var_49892_end_0, end_mask = var_49892_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49892_cast")];
            tensor<int32, [4]> var_49896_begin_0 = const()[name = tensor<string, []>("op_49896_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_49896_end_0 = const()[name = tensor<string, []>("op_49896_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_49896_end_mask_0 = const()[name = tensor<string, []>("op_49896_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49896_cast = slice_by_index(begin = var_49896_begin_0, end = var_49896_end_0, end_mask = var_49896_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49896_cast")];
            tensor<int32, [4]> var_49900_begin_0 = const()[name = tensor<string, []>("op_49900_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_49900_end_0 = const()[name = tensor<string, []>("op_49900_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_49900_end_mask_0 = const()[name = tensor<string, []>("op_49900_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49900_cast = slice_by_index(begin = var_49900_begin_0, end = var_49900_end_0, end_mask = var_49900_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49900_cast")];
            tensor<int32, [4]> var_49904_begin_0 = const()[name = tensor<string, []>("op_49904_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_49904_end_0 = const()[name = tensor<string, []>("op_49904_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_49904_end_mask_0 = const()[name = tensor<string, []>("op_49904_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49904_cast = slice_by_index(begin = var_49904_begin_0, end = var_49904_end_0, end_mask = var_49904_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49904_cast")];
            tensor<int32, [4]> var_49908_begin_0 = const()[name = tensor<string, []>("op_49908_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_49908_end_0 = const()[name = tensor<string, []>("op_49908_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_49908_end_mask_0 = const()[name = tensor<string, []>("op_49908_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49908_cast = slice_by_index(begin = var_49908_begin_0, end = var_49908_end_0, end_mask = var_49908_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49908_cast")];
            tensor<int32, [4]> var_49912_begin_0 = const()[name = tensor<string, []>("op_49912_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_49912_end_0 = const()[name = tensor<string, []>("op_49912_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_49912_end_mask_0 = const()[name = tensor<string, []>("op_49912_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49912_cast = slice_by_index(begin = var_49912_begin_0, end = var_49912_end_0, end_mask = var_49912_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49912_cast")];
            tensor<int32, [4]> var_49916_begin_0 = const()[name = tensor<string, []>("op_49916_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_49916_end_0 = const()[name = tensor<string, []>("op_49916_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_49916_end_mask_0 = const()[name = tensor<string, []>("op_49916_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49916_cast = slice_by_index(begin = var_49916_begin_0, end = var_49916_end_0, end_mask = var_49916_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49916_cast")];
            tensor<int32, [4]> var_49920_begin_0 = const()[name = tensor<string, []>("op_49920_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_49920_end_0 = const()[name = tensor<string, []>("op_49920_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_49920_end_mask_0 = const()[name = tensor<string, []>("op_49920_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49920_cast = slice_by_index(begin = var_49920_begin_0, end = var_49920_end_0, end_mask = var_49920_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49920_cast")];
            tensor<int32, [4]> var_49924_begin_0 = const()[name = tensor<string, []>("op_49924_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_49924_end_0 = const()[name = tensor<string, []>("op_49924_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_49924_end_mask_0 = const()[name = tensor<string, []>("op_49924_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49924_cast = slice_by_index(begin = var_49924_begin_0, end = var_49924_end_0, end_mask = var_49924_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49924_cast")];
            tensor<int32, [4]> var_49928_begin_0 = const()[name = tensor<string, []>("op_49928_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_49928_end_0 = const()[name = tensor<string, []>("op_49928_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_49928_end_mask_0 = const()[name = tensor<string, []>("op_49928_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49928_cast = slice_by_index(begin = var_49928_begin_0, end = var_49928_end_0, end_mask = var_49928_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49928_cast")];
            tensor<int32, [4]> var_49932_begin_0 = const()[name = tensor<string, []>("op_49932_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_49932_end_0 = const()[name = tensor<string, []>("op_49932_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_49932_end_mask_0 = const()[name = tensor<string, []>("op_49932_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49932_cast = slice_by_index(begin = var_49932_begin_0, end = var_49932_end_0, end_mask = var_49932_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49932_cast")];
            tensor<int32, [4]> var_49936_begin_0 = const()[name = tensor<string, []>("op_49936_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_49936_end_0 = const()[name = tensor<string, []>("op_49936_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_49936_end_mask_0 = const()[name = tensor<string, []>("op_49936_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49936_cast = slice_by_index(begin = var_49936_begin_0, end = var_49936_end_0, end_mask = var_49936_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49936_cast")];
            tensor<int32, [4]> var_49940_begin_0 = const()[name = tensor<string, []>("op_49940_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_49940_end_0 = const()[name = tensor<string, []>("op_49940_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_49940_end_mask_0 = const()[name = tensor<string, []>("op_49940_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49940_cast = slice_by_index(begin = var_49940_begin_0, end = var_49940_end_0, end_mask = var_49940_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49940_cast")];
            tensor<int32, [4]> var_49944_begin_0 = const()[name = tensor<string, []>("op_49944_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_49944_end_0 = const()[name = tensor<string, []>("op_49944_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_49944_end_mask_0 = const()[name = tensor<string, []>("op_49944_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_49944_cast = slice_by_index(begin = var_49944_begin_0, end = var_49944_end_0, end_mask = var_49944_end_mask_0, x = q_221_cast)[name = tensor<string, []>("op_49944_cast")];
            tensor<int32, [4]> k_443_perm_0 = const()[name = tensor<string, []>("k_443_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_49951_begin_0 = const()[name = tensor<string, []>("op_49951_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_49951_end_0 = const()[name = tensor<string, []>("op_49951_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_49951_end_mask_0 = const()[name = tensor<string, []>("op_49951_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_29 = transpose(perm = k_443_perm_0, x = k_441_cast)[name = tensor<string, []>("transpose_29")];
            tensor<fp16, [2, 1024, 1, 64]> var_49951_cast = slice_by_index(begin = var_49951_begin_0, end = var_49951_end_0, end_mask = var_49951_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_49951_cast")];
            tensor<int32, [4]> var_49955_begin_0 = const()[name = tensor<string, []>("op_49955_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_49955_end_0 = const()[name = tensor<string, []>("op_49955_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_49955_end_mask_0 = const()[name = tensor<string, []>("op_49955_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49955_cast = slice_by_index(begin = var_49955_begin_0, end = var_49955_end_0, end_mask = var_49955_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_49955_cast")];
            tensor<int32, [4]> var_49959_begin_0 = const()[name = tensor<string, []>("op_49959_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_49959_end_0 = const()[name = tensor<string, []>("op_49959_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_49959_end_mask_0 = const()[name = tensor<string, []>("op_49959_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49959_cast = slice_by_index(begin = var_49959_begin_0, end = var_49959_end_0, end_mask = var_49959_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_49959_cast")];
            tensor<int32, [4]> var_49963_begin_0 = const()[name = tensor<string, []>("op_49963_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_49963_end_0 = const()[name = tensor<string, []>("op_49963_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_49963_end_mask_0 = const()[name = tensor<string, []>("op_49963_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49963_cast = slice_by_index(begin = var_49963_begin_0, end = var_49963_end_0, end_mask = var_49963_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_49963_cast")];
            tensor<int32, [4]> var_49967_begin_0 = const()[name = tensor<string, []>("op_49967_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_49967_end_0 = const()[name = tensor<string, []>("op_49967_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_49967_end_mask_0 = const()[name = tensor<string, []>("op_49967_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49967_cast = slice_by_index(begin = var_49967_begin_0, end = var_49967_end_0, end_mask = var_49967_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_49967_cast")];
            tensor<int32, [4]> var_49971_begin_0 = const()[name = tensor<string, []>("op_49971_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_49971_end_0 = const()[name = tensor<string, []>("op_49971_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_49971_end_mask_0 = const()[name = tensor<string, []>("op_49971_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49971_cast = slice_by_index(begin = var_49971_begin_0, end = var_49971_end_0, end_mask = var_49971_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_49971_cast")];
            tensor<int32, [4]> var_49975_begin_0 = const()[name = tensor<string, []>("op_49975_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_49975_end_0 = const()[name = tensor<string, []>("op_49975_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_49975_end_mask_0 = const()[name = tensor<string, []>("op_49975_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49975_cast = slice_by_index(begin = var_49975_begin_0, end = var_49975_end_0, end_mask = var_49975_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_49975_cast")];
            tensor<int32, [4]> var_49979_begin_0 = const()[name = tensor<string, []>("op_49979_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_49979_end_0 = const()[name = tensor<string, []>("op_49979_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_49979_end_mask_0 = const()[name = tensor<string, []>("op_49979_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49979_cast = slice_by_index(begin = var_49979_begin_0, end = var_49979_end_0, end_mask = var_49979_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_49979_cast")];
            tensor<int32, [4]> var_49983_begin_0 = const()[name = tensor<string, []>("op_49983_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_49983_end_0 = const()[name = tensor<string, []>("op_49983_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_49983_end_mask_0 = const()[name = tensor<string, []>("op_49983_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49983_cast = slice_by_index(begin = var_49983_begin_0, end = var_49983_end_0, end_mask = var_49983_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_49983_cast")];
            tensor<int32, [4]> var_49987_begin_0 = const()[name = tensor<string, []>("op_49987_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_49987_end_0 = const()[name = tensor<string, []>("op_49987_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_49987_end_mask_0 = const()[name = tensor<string, []>("op_49987_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49987_cast = slice_by_index(begin = var_49987_begin_0, end = var_49987_end_0, end_mask = var_49987_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_49987_cast")];
            tensor<int32, [4]> var_49991_begin_0 = const()[name = tensor<string, []>("op_49991_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_49991_end_0 = const()[name = tensor<string, []>("op_49991_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_49991_end_mask_0 = const()[name = tensor<string, []>("op_49991_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49991_cast = slice_by_index(begin = var_49991_begin_0, end = var_49991_end_0, end_mask = var_49991_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_49991_cast")];
            tensor<int32, [4]> var_49995_begin_0 = const()[name = tensor<string, []>("op_49995_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_49995_end_0 = const()[name = tensor<string, []>("op_49995_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_49995_end_mask_0 = const()[name = tensor<string, []>("op_49995_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49995_cast = slice_by_index(begin = var_49995_begin_0, end = var_49995_end_0, end_mask = var_49995_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_49995_cast")];
            tensor<int32, [4]> var_49999_begin_0 = const()[name = tensor<string, []>("op_49999_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_49999_end_0 = const()[name = tensor<string, []>("op_49999_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_49999_end_mask_0 = const()[name = tensor<string, []>("op_49999_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_49999_cast = slice_by_index(begin = var_49999_begin_0, end = var_49999_end_0, end_mask = var_49999_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_49999_cast")];
            tensor<int32, [4]> var_50003_begin_0 = const()[name = tensor<string, []>("op_50003_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_50003_end_0 = const()[name = tensor<string, []>("op_50003_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_50003_end_mask_0 = const()[name = tensor<string, []>("op_50003_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50003_cast = slice_by_index(begin = var_50003_begin_0, end = var_50003_end_0, end_mask = var_50003_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_50003_cast")];
            tensor<int32, [4]> var_50007_begin_0 = const()[name = tensor<string, []>("op_50007_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_50007_end_0 = const()[name = tensor<string, []>("op_50007_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_50007_end_mask_0 = const()[name = tensor<string, []>("op_50007_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50007_cast = slice_by_index(begin = var_50007_begin_0, end = var_50007_end_0, end_mask = var_50007_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_50007_cast")];
            tensor<int32, [4]> var_50011_begin_0 = const()[name = tensor<string, []>("op_50011_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_50011_end_0 = const()[name = tensor<string, []>("op_50011_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_50011_end_mask_0 = const()[name = tensor<string, []>("op_50011_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50011_cast = slice_by_index(begin = var_50011_begin_0, end = var_50011_end_0, end_mask = var_50011_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_50011_cast")];
            tensor<int32, [4]> var_50015_begin_0 = const()[name = tensor<string, []>("op_50015_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_50015_end_0 = const()[name = tensor<string, []>("op_50015_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_50015_end_mask_0 = const()[name = tensor<string, []>("op_50015_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50015_cast = slice_by_index(begin = var_50015_begin_0, end = var_50015_end_0, end_mask = var_50015_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_50015_cast")];
            tensor<int32, [4]> var_50019_begin_0 = const()[name = tensor<string, []>("op_50019_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_50019_end_0 = const()[name = tensor<string, []>("op_50019_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_50019_end_mask_0 = const()[name = tensor<string, []>("op_50019_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50019_cast = slice_by_index(begin = var_50019_begin_0, end = var_50019_end_0, end_mask = var_50019_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_50019_cast")];
            tensor<int32, [4]> var_50023_begin_0 = const()[name = tensor<string, []>("op_50023_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_50023_end_0 = const()[name = tensor<string, []>("op_50023_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_50023_end_mask_0 = const()[name = tensor<string, []>("op_50023_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50023_cast = slice_by_index(begin = var_50023_begin_0, end = var_50023_end_0, end_mask = var_50023_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_50023_cast")];
            tensor<int32, [4]> var_50027_begin_0 = const()[name = tensor<string, []>("op_50027_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_50027_end_0 = const()[name = tensor<string, []>("op_50027_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_50027_end_mask_0 = const()[name = tensor<string, []>("op_50027_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50027_cast = slice_by_index(begin = var_50027_begin_0, end = var_50027_end_0, end_mask = var_50027_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_50027_cast")];
            tensor<int32, [4]> var_50029_begin_0 = const()[name = tensor<string, []>("op_50029_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_50029_end_0 = const()[name = tensor<string, []>("op_50029_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_50029_end_mask_0 = const()[name = tensor<string, []>("op_50029_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50029_cast = slice_by_index(begin = var_50029_begin_0, end = var_50029_end_0, end_mask = var_50029_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50029_cast")];
            tensor<int32, [4]> var_50033_begin_0 = const()[name = tensor<string, []>("op_50033_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_50033_end_0 = const()[name = tensor<string, []>("op_50033_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_50033_end_mask_0 = const()[name = tensor<string, []>("op_50033_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50033_cast = slice_by_index(begin = var_50033_begin_0, end = var_50033_end_0, end_mask = var_50033_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50033_cast")];
            tensor<int32, [4]> var_50037_begin_0 = const()[name = tensor<string, []>("op_50037_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_50037_end_0 = const()[name = tensor<string, []>("op_50037_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_50037_end_mask_0 = const()[name = tensor<string, []>("op_50037_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50037_cast = slice_by_index(begin = var_50037_begin_0, end = var_50037_end_0, end_mask = var_50037_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50037_cast")];
            tensor<int32, [4]> var_50041_begin_0 = const()[name = tensor<string, []>("op_50041_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_50041_end_0 = const()[name = tensor<string, []>("op_50041_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_50041_end_mask_0 = const()[name = tensor<string, []>("op_50041_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50041_cast = slice_by_index(begin = var_50041_begin_0, end = var_50041_end_0, end_mask = var_50041_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50041_cast")];
            tensor<int32, [4]> var_50045_begin_0 = const()[name = tensor<string, []>("op_50045_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_50045_end_0 = const()[name = tensor<string, []>("op_50045_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_50045_end_mask_0 = const()[name = tensor<string, []>("op_50045_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50045_cast = slice_by_index(begin = var_50045_begin_0, end = var_50045_end_0, end_mask = var_50045_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50045_cast")];
            tensor<int32, [4]> var_50049_begin_0 = const()[name = tensor<string, []>("op_50049_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_50049_end_0 = const()[name = tensor<string, []>("op_50049_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_50049_end_mask_0 = const()[name = tensor<string, []>("op_50049_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50049_cast = slice_by_index(begin = var_50049_begin_0, end = var_50049_end_0, end_mask = var_50049_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50049_cast")];
            tensor<int32, [4]> var_50053_begin_0 = const()[name = tensor<string, []>("op_50053_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_50053_end_0 = const()[name = tensor<string, []>("op_50053_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_50053_end_mask_0 = const()[name = tensor<string, []>("op_50053_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50053_cast = slice_by_index(begin = var_50053_begin_0, end = var_50053_end_0, end_mask = var_50053_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50053_cast")];
            tensor<int32, [4]> var_50057_begin_0 = const()[name = tensor<string, []>("op_50057_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_50057_end_0 = const()[name = tensor<string, []>("op_50057_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_50057_end_mask_0 = const()[name = tensor<string, []>("op_50057_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50057_cast = slice_by_index(begin = var_50057_begin_0, end = var_50057_end_0, end_mask = var_50057_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50057_cast")];
            tensor<int32, [4]> var_50061_begin_0 = const()[name = tensor<string, []>("op_50061_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_50061_end_0 = const()[name = tensor<string, []>("op_50061_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_50061_end_mask_0 = const()[name = tensor<string, []>("op_50061_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50061_cast = slice_by_index(begin = var_50061_begin_0, end = var_50061_end_0, end_mask = var_50061_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50061_cast")];
            tensor<int32, [4]> var_50065_begin_0 = const()[name = tensor<string, []>("op_50065_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_50065_end_0 = const()[name = tensor<string, []>("op_50065_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_50065_end_mask_0 = const()[name = tensor<string, []>("op_50065_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50065_cast = slice_by_index(begin = var_50065_begin_0, end = var_50065_end_0, end_mask = var_50065_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50065_cast")];
            tensor<int32, [4]> var_50069_begin_0 = const()[name = tensor<string, []>("op_50069_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_50069_end_0 = const()[name = tensor<string, []>("op_50069_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_50069_end_mask_0 = const()[name = tensor<string, []>("op_50069_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50069_cast = slice_by_index(begin = var_50069_begin_0, end = var_50069_end_0, end_mask = var_50069_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50069_cast")];
            tensor<int32, [4]> var_50073_begin_0 = const()[name = tensor<string, []>("op_50073_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_50073_end_0 = const()[name = tensor<string, []>("op_50073_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_50073_end_mask_0 = const()[name = tensor<string, []>("op_50073_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50073_cast = slice_by_index(begin = var_50073_begin_0, end = var_50073_end_0, end_mask = var_50073_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50073_cast")];
            tensor<int32, [4]> var_50077_begin_0 = const()[name = tensor<string, []>("op_50077_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_50077_end_0 = const()[name = tensor<string, []>("op_50077_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_50077_end_mask_0 = const()[name = tensor<string, []>("op_50077_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50077_cast = slice_by_index(begin = var_50077_begin_0, end = var_50077_end_0, end_mask = var_50077_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50077_cast")];
            tensor<int32, [4]> var_50081_begin_0 = const()[name = tensor<string, []>("op_50081_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_50081_end_0 = const()[name = tensor<string, []>("op_50081_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_50081_end_mask_0 = const()[name = tensor<string, []>("op_50081_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50081_cast = slice_by_index(begin = var_50081_begin_0, end = var_50081_end_0, end_mask = var_50081_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50081_cast")];
            tensor<int32, [4]> var_50085_begin_0 = const()[name = tensor<string, []>("op_50085_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_50085_end_0 = const()[name = tensor<string, []>("op_50085_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_50085_end_mask_0 = const()[name = tensor<string, []>("op_50085_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50085_cast = slice_by_index(begin = var_50085_begin_0, end = var_50085_end_0, end_mask = var_50085_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50085_cast")];
            tensor<int32, [4]> var_50089_begin_0 = const()[name = tensor<string, []>("op_50089_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_50089_end_0 = const()[name = tensor<string, []>("op_50089_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_50089_end_mask_0 = const()[name = tensor<string, []>("op_50089_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50089_cast = slice_by_index(begin = var_50089_begin_0, end = var_50089_end_0, end_mask = var_50089_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50089_cast")];
            tensor<int32, [4]> var_50093_begin_0 = const()[name = tensor<string, []>("op_50093_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_50093_end_0 = const()[name = tensor<string, []>("op_50093_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_50093_end_mask_0 = const()[name = tensor<string, []>("op_50093_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50093_cast = slice_by_index(begin = var_50093_begin_0, end = var_50093_end_0, end_mask = var_50093_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50093_cast")];
            tensor<int32, [4]> var_50097_begin_0 = const()[name = tensor<string, []>("op_50097_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_50097_end_0 = const()[name = tensor<string, []>("op_50097_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_50097_end_mask_0 = const()[name = tensor<string, []>("op_50097_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50097_cast = slice_by_index(begin = var_50097_begin_0, end = var_50097_end_0, end_mask = var_50097_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50097_cast")];
            tensor<int32, [4]> var_50101_begin_0 = const()[name = tensor<string, []>("op_50101_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_50101_end_0 = const()[name = tensor<string, []>("op_50101_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_50101_end_mask_0 = const()[name = tensor<string, []>("op_50101_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50101_cast = slice_by_index(begin = var_50101_begin_0, end = var_50101_end_0, end_mask = var_50101_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50101_cast")];
            tensor<int32, [4]> var_50105_begin_0 = const()[name = tensor<string, []>("op_50105_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_50105_end_0 = const()[name = tensor<string, []>("op_50105_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_50105_end_mask_0 = const()[name = tensor<string, []>("op_50105_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50105_cast = slice_by_index(begin = var_50105_begin_0, end = var_50105_end_0, end_mask = var_50105_end_mask_0, x = v_221_cast)[name = tensor<string, []>("op_50105_cast")];
            tensor<string, []> var_50109_equation_0 = const()[name = tensor<string, []>("op_50109_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50109_cast = einsum(equation = var_50109_equation_0, values = (var_49951_cast, var_49868_cast))[name = tensor<string, []>("op_50109_cast")];
            tensor<fp16, []> var_50110_to_fp16 = const()[name = tensor<string, []>("op_50110_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4241_cast = mul(x = var_50109_cast, y = var_50110_to_fp16)[name = tensor<string, []>("aw_4241_cast")];
            tensor<string, []> var_50113_equation_0 = const()[name = tensor<string, []>("op_50113_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50113_cast = einsum(equation = var_50113_equation_0, values = (var_49955_cast, var_49872_cast))[name = tensor<string, []>("op_50113_cast")];
            tensor<fp16, []> var_50114_to_fp16 = const()[name = tensor<string, []>("op_50114_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4243_cast = mul(x = var_50113_cast, y = var_50114_to_fp16)[name = tensor<string, []>("aw_4243_cast")];
            tensor<string, []> var_50117_equation_0 = const()[name = tensor<string, []>("op_50117_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50117_cast = einsum(equation = var_50117_equation_0, values = (var_49959_cast, var_49876_cast))[name = tensor<string, []>("op_50117_cast")];
            tensor<fp16, []> var_50118_to_fp16 = const()[name = tensor<string, []>("op_50118_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4245_cast = mul(x = var_50117_cast, y = var_50118_to_fp16)[name = tensor<string, []>("aw_4245_cast")];
            tensor<string, []> var_50121_equation_0 = const()[name = tensor<string, []>("op_50121_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50121_cast = einsum(equation = var_50121_equation_0, values = (var_49963_cast, var_49880_cast))[name = tensor<string, []>("op_50121_cast")];
            tensor<fp16, []> var_50122_to_fp16 = const()[name = tensor<string, []>("op_50122_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4247_cast = mul(x = var_50121_cast, y = var_50122_to_fp16)[name = tensor<string, []>("aw_4247_cast")];
            tensor<string, []> var_50125_equation_0 = const()[name = tensor<string, []>("op_50125_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50125_cast = einsum(equation = var_50125_equation_0, values = (var_49967_cast, var_49884_cast))[name = tensor<string, []>("op_50125_cast")];
            tensor<fp16, []> var_50126_to_fp16 = const()[name = tensor<string, []>("op_50126_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4249_cast = mul(x = var_50125_cast, y = var_50126_to_fp16)[name = tensor<string, []>("aw_4249_cast")];
            tensor<string, []> var_50129_equation_0 = const()[name = tensor<string, []>("op_50129_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50129_cast = einsum(equation = var_50129_equation_0, values = (var_49971_cast, var_49888_cast))[name = tensor<string, []>("op_50129_cast")];
            tensor<fp16, []> var_50130_to_fp16 = const()[name = tensor<string, []>("op_50130_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4251_cast = mul(x = var_50129_cast, y = var_50130_to_fp16)[name = tensor<string, []>("aw_4251_cast")];
            tensor<string, []> var_50133_equation_0 = const()[name = tensor<string, []>("op_50133_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50133_cast = einsum(equation = var_50133_equation_0, values = (var_49975_cast, var_49892_cast))[name = tensor<string, []>("op_50133_cast")];
            tensor<fp16, []> var_50134_to_fp16 = const()[name = tensor<string, []>("op_50134_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4253_cast = mul(x = var_50133_cast, y = var_50134_to_fp16)[name = tensor<string, []>("aw_4253_cast")];
            tensor<string, []> var_50137_equation_0 = const()[name = tensor<string, []>("op_50137_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50137_cast = einsum(equation = var_50137_equation_0, values = (var_49979_cast, var_49896_cast))[name = tensor<string, []>("op_50137_cast")];
            tensor<fp16, []> var_50138_to_fp16 = const()[name = tensor<string, []>("op_50138_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4255_cast = mul(x = var_50137_cast, y = var_50138_to_fp16)[name = tensor<string, []>("aw_4255_cast")];
            tensor<string, []> var_50141_equation_0 = const()[name = tensor<string, []>("op_50141_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50141_cast = einsum(equation = var_50141_equation_0, values = (var_49983_cast, var_49900_cast))[name = tensor<string, []>("op_50141_cast")];
            tensor<fp16, []> var_50142_to_fp16 = const()[name = tensor<string, []>("op_50142_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4257_cast = mul(x = var_50141_cast, y = var_50142_to_fp16)[name = tensor<string, []>("aw_4257_cast")];
            tensor<string, []> var_50145_equation_0 = const()[name = tensor<string, []>("op_50145_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50145_cast = einsum(equation = var_50145_equation_0, values = (var_49987_cast, var_49904_cast))[name = tensor<string, []>("op_50145_cast")];
            tensor<fp16, []> var_50146_to_fp16 = const()[name = tensor<string, []>("op_50146_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4259_cast = mul(x = var_50145_cast, y = var_50146_to_fp16)[name = tensor<string, []>("aw_4259_cast")];
            tensor<string, []> var_50149_equation_0 = const()[name = tensor<string, []>("op_50149_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50149_cast = einsum(equation = var_50149_equation_0, values = (var_49991_cast, var_49908_cast))[name = tensor<string, []>("op_50149_cast")];
            tensor<fp16, []> var_50150_to_fp16 = const()[name = tensor<string, []>("op_50150_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4261_cast = mul(x = var_50149_cast, y = var_50150_to_fp16)[name = tensor<string, []>("aw_4261_cast")];
            tensor<string, []> var_50153_equation_0 = const()[name = tensor<string, []>("op_50153_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50153_cast = einsum(equation = var_50153_equation_0, values = (var_49995_cast, var_49912_cast))[name = tensor<string, []>("op_50153_cast")];
            tensor<fp16, []> var_50154_to_fp16 = const()[name = tensor<string, []>("op_50154_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4263_cast = mul(x = var_50153_cast, y = var_50154_to_fp16)[name = tensor<string, []>("aw_4263_cast")];
            tensor<string, []> var_50157_equation_0 = const()[name = tensor<string, []>("op_50157_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50157_cast = einsum(equation = var_50157_equation_0, values = (var_49999_cast, var_49916_cast))[name = tensor<string, []>("op_50157_cast")];
            tensor<fp16, []> var_50158_to_fp16 = const()[name = tensor<string, []>("op_50158_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4265_cast = mul(x = var_50157_cast, y = var_50158_to_fp16)[name = tensor<string, []>("aw_4265_cast")];
            tensor<string, []> var_50161_equation_0 = const()[name = tensor<string, []>("op_50161_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50161_cast = einsum(equation = var_50161_equation_0, values = (var_50003_cast, var_49920_cast))[name = tensor<string, []>("op_50161_cast")];
            tensor<fp16, []> var_50162_to_fp16 = const()[name = tensor<string, []>("op_50162_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4267_cast = mul(x = var_50161_cast, y = var_50162_to_fp16)[name = tensor<string, []>("aw_4267_cast")];
            tensor<string, []> var_50165_equation_0 = const()[name = tensor<string, []>("op_50165_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50165_cast = einsum(equation = var_50165_equation_0, values = (var_50007_cast, var_49924_cast))[name = tensor<string, []>("op_50165_cast")];
            tensor<fp16, []> var_50166_to_fp16 = const()[name = tensor<string, []>("op_50166_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4269_cast = mul(x = var_50165_cast, y = var_50166_to_fp16)[name = tensor<string, []>("aw_4269_cast")];
            tensor<string, []> var_50169_equation_0 = const()[name = tensor<string, []>("op_50169_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50169_cast = einsum(equation = var_50169_equation_0, values = (var_50011_cast, var_49928_cast))[name = tensor<string, []>("op_50169_cast")];
            tensor<fp16, []> var_50170_to_fp16 = const()[name = tensor<string, []>("op_50170_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4271_cast = mul(x = var_50169_cast, y = var_50170_to_fp16)[name = tensor<string, []>("aw_4271_cast")];
            tensor<string, []> var_50173_equation_0 = const()[name = tensor<string, []>("op_50173_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50173_cast = einsum(equation = var_50173_equation_0, values = (var_50015_cast, var_49932_cast))[name = tensor<string, []>("op_50173_cast")];
            tensor<fp16, []> var_50174_to_fp16 = const()[name = tensor<string, []>("op_50174_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4273_cast = mul(x = var_50173_cast, y = var_50174_to_fp16)[name = tensor<string, []>("aw_4273_cast")];
            tensor<string, []> var_50177_equation_0 = const()[name = tensor<string, []>("op_50177_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50177_cast = einsum(equation = var_50177_equation_0, values = (var_50019_cast, var_49936_cast))[name = tensor<string, []>("op_50177_cast")];
            tensor<fp16, []> var_50178_to_fp16 = const()[name = tensor<string, []>("op_50178_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4275_cast = mul(x = var_50177_cast, y = var_50178_to_fp16)[name = tensor<string, []>("aw_4275_cast")];
            tensor<string, []> var_50181_equation_0 = const()[name = tensor<string, []>("op_50181_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50181_cast = einsum(equation = var_50181_equation_0, values = (var_50023_cast, var_49940_cast))[name = tensor<string, []>("op_50181_cast")];
            tensor<fp16, []> var_50182_to_fp16 = const()[name = tensor<string, []>("op_50182_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4277_cast = mul(x = var_50181_cast, y = var_50182_to_fp16)[name = tensor<string, []>("aw_4277_cast")];
            tensor<string, []> var_50185_equation_0 = const()[name = tensor<string, []>("op_50185_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50185_cast = einsum(equation = var_50185_equation_0, values = (var_50027_cast, var_49944_cast))[name = tensor<string, []>("op_50185_cast")];
            tensor<fp16, []> var_50186_to_fp16 = const()[name = tensor<string, []>("op_50186_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4279_cast = mul(x = var_50185_cast, y = var_50186_to_fp16)[name = tensor<string, []>("aw_4279_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50188_cast = softmax(axis = var_30385, x = aw_4241_cast)[name = tensor<string, []>("op_50188_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50189_cast = softmax(axis = var_30385, x = aw_4243_cast)[name = tensor<string, []>("op_50189_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50190_cast = softmax(axis = var_30385, x = aw_4245_cast)[name = tensor<string, []>("op_50190_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50191_cast = softmax(axis = var_30385, x = aw_4247_cast)[name = tensor<string, []>("op_50191_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50192_cast = softmax(axis = var_30385, x = aw_4249_cast)[name = tensor<string, []>("op_50192_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50193_cast = softmax(axis = var_30385, x = aw_4251_cast)[name = tensor<string, []>("op_50193_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50194_cast = softmax(axis = var_30385, x = aw_4253_cast)[name = tensor<string, []>("op_50194_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50195_cast = softmax(axis = var_30385, x = aw_4255_cast)[name = tensor<string, []>("op_50195_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50196_cast = softmax(axis = var_30385, x = aw_4257_cast)[name = tensor<string, []>("op_50196_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50197_cast = softmax(axis = var_30385, x = aw_4259_cast)[name = tensor<string, []>("op_50197_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50198_cast = softmax(axis = var_30385, x = aw_4261_cast)[name = tensor<string, []>("op_50198_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50199_cast = softmax(axis = var_30385, x = aw_4263_cast)[name = tensor<string, []>("op_50199_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50200_cast = softmax(axis = var_30385, x = aw_4265_cast)[name = tensor<string, []>("op_50200_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50201_cast = softmax(axis = var_30385, x = aw_4267_cast)[name = tensor<string, []>("op_50201_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50202_cast = softmax(axis = var_30385, x = aw_4269_cast)[name = tensor<string, []>("op_50202_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50203_cast = softmax(axis = var_30385, x = aw_4271_cast)[name = tensor<string, []>("op_50203_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50204_cast = softmax(axis = var_30385, x = aw_4273_cast)[name = tensor<string, []>("op_50204_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50205_cast = softmax(axis = var_30385, x = aw_4275_cast)[name = tensor<string, []>("op_50205_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50206_cast = softmax(axis = var_30385, x = aw_4277_cast)[name = tensor<string, []>("op_50206_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_50207_cast = softmax(axis = var_30385, x = aw_4279_cast)[name = tensor<string, []>("op_50207_cast")];
            tensor<string, []> var_50209_equation_0 = const()[name = tensor<string, []>("op_50209_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50209_cast = einsum(equation = var_50209_equation_0, values = (var_50029_cast, var_50188_cast))[name = tensor<string, []>("op_50209_cast")];
            tensor<string, []> var_50211_equation_0 = const()[name = tensor<string, []>("op_50211_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50211_cast = einsum(equation = var_50211_equation_0, values = (var_50033_cast, var_50189_cast))[name = tensor<string, []>("op_50211_cast")];
            tensor<string, []> var_50213_equation_0 = const()[name = tensor<string, []>("op_50213_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50213_cast = einsum(equation = var_50213_equation_0, values = (var_50037_cast, var_50190_cast))[name = tensor<string, []>("op_50213_cast")];
            tensor<string, []> var_50215_equation_0 = const()[name = tensor<string, []>("op_50215_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50215_cast = einsum(equation = var_50215_equation_0, values = (var_50041_cast, var_50191_cast))[name = tensor<string, []>("op_50215_cast")];
            tensor<string, []> var_50217_equation_0 = const()[name = tensor<string, []>("op_50217_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50217_cast = einsum(equation = var_50217_equation_0, values = (var_50045_cast, var_50192_cast))[name = tensor<string, []>("op_50217_cast")];
            tensor<string, []> var_50219_equation_0 = const()[name = tensor<string, []>("op_50219_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50219_cast = einsum(equation = var_50219_equation_0, values = (var_50049_cast, var_50193_cast))[name = tensor<string, []>("op_50219_cast")];
            tensor<string, []> var_50221_equation_0 = const()[name = tensor<string, []>("op_50221_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50221_cast = einsum(equation = var_50221_equation_0, values = (var_50053_cast, var_50194_cast))[name = tensor<string, []>("op_50221_cast")];
            tensor<string, []> var_50223_equation_0 = const()[name = tensor<string, []>("op_50223_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50223_cast = einsum(equation = var_50223_equation_0, values = (var_50057_cast, var_50195_cast))[name = tensor<string, []>("op_50223_cast")];
            tensor<string, []> var_50225_equation_0 = const()[name = tensor<string, []>("op_50225_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50225_cast = einsum(equation = var_50225_equation_0, values = (var_50061_cast, var_50196_cast))[name = tensor<string, []>("op_50225_cast")];
            tensor<string, []> var_50227_equation_0 = const()[name = tensor<string, []>("op_50227_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50227_cast = einsum(equation = var_50227_equation_0, values = (var_50065_cast, var_50197_cast))[name = tensor<string, []>("op_50227_cast")];
            tensor<string, []> var_50229_equation_0 = const()[name = tensor<string, []>("op_50229_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50229_cast = einsum(equation = var_50229_equation_0, values = (var_50069_cast, var_50198_cast))[name = tensor<string, []>("op_50229_cast")];
            tensor<string, []> var_50231_equation_0 = const()[name = tensor<string, []>("op_50231_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50231_cast = einsum(equation = var_50231_equation_0, values = (var_50073_cast, var_50199_cast))[name = tensor<string, []>("op_50231_cast")];
            tensor<string, []> var_50233_equation_0 = const()[name = tensor<string, []>("op_50233_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50233_cast = einsum(equation = var_50233_equation_0, values = (var_50077_cast, var_50200_cast))[name = tensor<string, []>("op_50233_cast")];
            tensor<string, []> var_50235_equation_0 = const()[name = tensor<string, []>("op_50235_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50235_cast = einsum(equation = var_50235_equation_0, values = (var_50081_cast, var_50201_cast))[name = tensor<string, []>("op_50235_cast")];
            tensor<string, []> var_50237_equation_0 = const()[name = tensor<string, []>("op_50237_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50237_cast = einsum(equation = var_50237_equation_0, values = (var_50085_cast, var_50202_cast))[name = tensor<string, []>("op_50237_cast")];
            tensor<string, []> var_50239_equation_0 = const()[name = tensor<string, []>("op_50239_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50239_cast = einsum(equation = var_50239_equation_0, values = (var_50089_cast, var_50203_cast))[name = tensor<string, []>("op_50239_cast")];
            tensor<string, []> var_50241_equation_0 = const()[name = tensor<string, []>("op_50241_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50241_cast = einsum(equation = var_50241_equation_0, values = (var_50093_cast, var_50204_cast))[name = tensor<string, []>("op_50241_cast")];
            tensor<string, []> var_50243_equation_0 = const()[name = tensor<string, []>("op_50243_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50243_cast = einsum(equation = var_50243_equation_0, values = (var_50097_cast, var_50205_cast))[name = tensor<string, []>("op_50243_cast")];
            tensor<string, []> var_50245_equation_0 = const()[name = tensor<string, []>("op_50245_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50245_cast = einsum(equation = var_50245_equation_0, values = (var_50101_cast, var_50206_cast))[name = tensor<string, []>("op_50245_cast")];
            tensor<string, []> var_50247_equation_0 = const()[name = tensor<string, []>("op_50247_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50247_cast = einsum(equation = var_50247_equation_0, values = (var_50105_cast, var_50207_cast))[name = tensor<string, []>("op_50247_cast")];
            tensor<bool, []> input_645_interleave_0 = const()[name = tensor<string, []>("input_645_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_645_cast = concat(axis = var_30385, interleave = input_645_interleave_0, values = (var_50209_cast, var_50211_cast, var_50213_cast, var_50215_cast, var_50217_cast, var_50219_cast, var_50221_cast, var_50223_cast, var_50225_cast, var_50227_cast, var_50229_cast, var_50231_cast, var_50233_cast, var_50235_cast, var_50237_cast, var_50239_cast, var_50241_cast, var_50243_cast, var_50245_cast, var_50247_cast))[name = tensor<string, []>("input_645_cast")];
            tensor<int32, [2]> var_50253 = const()[name = tensor<string, []>("op_50253"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_50255 = const()[name = tensor<string, []>("op_50255"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_50257_pad_type_0 = const()[name = tensor<string, []>("op_50257_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_50257_pad_0 = const()[name = tensor<string, []>("op_50257_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4251439808)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4254716672)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_50257_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_50255, groups = var_30385, pad = var_50257_pad_0, pad_type = var_50257_pad_type_0, strides = var_50253, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_645_cast)[name = tensor<string, []>("op_50257_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_333_cast = add(x = var_50257_cast, y = inputs_331_cast)[name = tensor<string, []>("inputs_333_cast")];
            tensor<int32, [1]> var_50261 = const()[name = tensor<string, []>("op_50261"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_333_cast = reduce_mean(axes = var_50261, keep_dims = var_30380, x = inputs_333_cast)[name = tensor<string, []>("channels_mean_333_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_333_cast = sub(x = inputs_333_cast, y = channels_mean_333_cast)[name = tensor<string, []>("zero_mean_333_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_333_cast = mul(x = zero_mean_333_cast, y = zero_mean_333_cast)[name = tensor<string, []>("zero_mean_sq_333_cast")];
            tensor<int32, [1]> var_50265 = const()[name = tensor<string, []>("op_50265"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_50266_cast = reduce_mean(axes = var_50265, keep_dims = var_30380, x = zero_mean_sq_333_cast)[name = tensor<string, []>("op_50266_cast")];
            tensor<fp16, []> var_50267_to_fp16 = const()[name = tensor<string, []>("op_50267_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_50268_cast = add(x = var_50266_cast, y = var_50267_to_fp16)[name = tensor<string, []>("op_50268_cast")];
            tensor<fp16, []> denom_333_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_333_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_333_cast = rsqrt(epsilon = denom_333_epsilon_0_to_fp16, x = var_50268_cast)[name = tensor<string, []>("denom_333_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_333_cast = mul(x = zero_mean_333_cast, y = denom_333_cast)[name = tensor<string, []>("out_333_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_50272_to_fp16 = const()[name = tensor<string, []>("op_50272_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4254719296)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_50273_cast = add(x = out_333_cast, y = var_50272_to_fp16)[name = tensor<string, []>("op_50273_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_50275_to_fp16 = const()[name = tensor<string, []>("op_50275_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4254721920)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_443_cast = mul(x = var_50273_cast, y = var_50275_to_fp16)[name = tensor<string, []>("hidden_states_443_cast")];
            tensor<int32, [2]> var_50282 = const()[name = tensor<string, []>("op_50282"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_50284 = const()[name = tensor<string, []>("op_50284"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_223_pad_type_0 = const()[name = tensor<string, []>("q_223_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_223_pad_0 = const()[name = tensor<string, []>("q_223_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4254724544)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_223_cast = conv(dilations = var_50284, groups = var_30385, pad = q_223_pad_0, pad_type = q_223_pad_type_0, strides = var_50282, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_443_cast)[name = tensor<string, []>("q_223_cast")];
            tensor<int32, [2]> var_50288 = const()[name = tensor<string, []>("op_50288"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_50290 = const()[name = tensor<string, []>("op_50290"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_445_pad_type_0 = const()[name = tensor<string, []>("k_445_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_445_pad_0 = const()[name = tensor<string, []>("k_445_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4258001408)))];
            tensor<fp16, [2, 1280, 1, 77]> k_445_cast = conv(dilations = var_50290, groups = var_30385, pad = k_445_pad_0, pad_type = k_445_pad_type_0, strides = var_50288, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_445_cast")];
            tensor<int32, [2]> var_50294 = const()[name = tensor<string, []>("op_50294"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_50296 = const()[name = tensor<string, []>("op_50296"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_223_pad_type_0 = const()[name = tensor<string, []>("v_223_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_223_pad_0 = const()[name = tensor<string, []>("v_223_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4263244352)))];
            tensor<fp16, [2, 1280, 1, 77]> v_223_cast = conv(dilations = var_50296, groups = var_30385, pad = v_223_pad_0, pad_type = v_223_pad_type_0, strides = var_50294, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_223_cast")];
            tensor<int32, [4]> var_50300_begin_0 = const()[name = tensor<string, []>("op_50300_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_50300_end_0 = const()[name = tensor<string, []>("op_50300_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_50300_end_mask_0 = const()[name = tensor<string, []>("op_50300_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50300_cast = slice_by_index(begin = var_50300_begin_0, end = var_50300_end_0, end_mask = var_50300_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50300_cast")];
            tensor<int32, [4]> var_50304_begin_0 = const()[name = tensor<string, []>("op_50304_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_50304_end_0 = const()[name = tensor<string, []>("op_50304_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_50304_end_mask_0 = const()[name = tensor<string, []>("op_50304_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50304_cast = slice_by_index(begin = var_50304_begin_0, end = var_50304_end_0, end_mask = var_50304_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50304_cast")];
            tensor<int32, [4]> var_50308_begin_0 = const()[name = tensor<string, []>("op_50308_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_50308_end_0 = const()[name = tensor<string, []>("op_50308_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_50308_end_mask_0 = const()[name = tensor<string, []>("op_50308_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50308_cast = slice_by_index(begin = var_50308_begin_0, end = var_50308_end_0, end_mask = var_50308_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50308_cast")];
            tensor<int32, [4]> var_50312_begin_0 = const()[name = tensor<string, []>("op_50312_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_50312_end_0 = const()[name = tensor<string, []>("op_50312_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_50312_end_mask_0 = const()[name = tensor<string, []>("op_50312_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50312_cast = slice_by_index(begin = var_50312_begin_0, end = var_50312_end_0, end_mask = var_50312_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50312_cast")];
            tensor<int32, [4]> var_50316_begin_0 = const()[name = tensor<string, []>("op_50316_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_50316_end_0 = const()[name = tensor<string, []>("op_50316_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_50316_end_mask_0 = const()[name = tensor<string, []>("op_50316_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50316_cast = slice_by_index(begin = var_50316_begin_0, end = var_50316_end_0, end_mask = var_50316_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50316_cast")];
            tensor<int32, [4]> var_50320_begin_0 = const()[name = tensor<string, []>("op_50320_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_50320_end_0 = const()[name = tensor<string, []>("op_50320_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_50320_end_mask_0 = const()[name = tensor<string, []>("op_50320_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50320_cast = slice_by_index(begin = var_50320_begin_0, end = var_50320_end_0, end_mask = var_50320_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50320_cast")];
            tensor<int32, [4]> var_50324_begin_0 = const()[name = tensor<string, []>("op_50324_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_50324_end_0 = const()[name = tensor<string, []>("op_50324_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_50324_end_mask_0 = const()[name = tensor<string, []>("op_50324_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50324_cast = slice_by_index(begin = var_50324_begin_0, end = var_50324_end_0, end_mask = var_50324_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50324_cast")];
            tensor<int32, [4]> var_50328_begin_0 = const()[name = tensor<string, []>("op_50328_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_50328_end_0 = const()[name = tensor<string, []>("op_50328_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_50328_end_mask_0 = const()[name = tensor<string, []>("op_50328_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50328_cast = slice_by_index(begin = var_50328_begin_0, end = var_50328_end_0, end_mask = var_50328_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50328_cast")];
            tensor<int32, [4]> var_50332_begin_0 = const()[name = tensor<string, []>("op_50332_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_50332_end_0 = const()[name = tensor<string, []>("op_50332_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_50332_end_mask_0 = const()[name = tensor<string, []>("op_50332_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50332_cast = slice_by_index(begin = var_50332_begin_0, end = var_50332_end_0, end_mask = var_50332_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50332_cast")];
            tensor<int32, [4]> var_50336_begin_0 = const()[name = tensor<string, []>("op_50336_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_50336_end_0 = const()[name = tensor<string, []>("op_50336_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_50336_end_mask_0 = const()[name = tensor<string, []>("op_50336_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50336_cast = slice_by_index(begin = var_50336_begin_0, end = var_50336_end_0, end_mask = var_50336_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50336_cast")];
            tensor<int32, [4]> var_50340_begin_0 = const()[name = tensor<string, []>("op_50340_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_50340_end_0 = const()[name = tensor<string, []>("op_50340_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_50340_end_mask_0 = const()[name = tensor<string, []>("op_50340_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50340_cast = slice_by_index(begin = var_50340_begin_0, end = var_50340_end_0, end_mask = var_50340_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50340_cast")];
            tensor<int32, [4]> var_50344_begin_0 = const()[name = tensor<string, []>("op_50344_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_50344_end_0 = const()[name = tensor<string, []>("op_50344_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_50344_end_mask_0 = const()[name = tensor<string, []>("op_50344_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50344_cast = slice_by_index(begin = var_50344_begin_0, end = var_50344_end_0, end_mask = var_50344_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50344_cast")];
            tensor<int32, [4]> var_50348_begin_0 = const()[name = tensor<string, []>("op_50348_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_50348_end_0 = const()[name = tensor<string, []>("op_50348_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_50348_end_mask_0 = const()[name = tensor<string, []>("op_50348_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50348_cast = slice_by_index(begin = var_50348_begin_0, end = var_50348_end_0, end_mask = var_50348_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50348_cast")];
            tensor<int32, [4]> var_50352_begin_0 = const()[name = tensor<string, []>("op_50352_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_50352_end_0 = const()[name = tensor<string, []>("op_50352_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_50352_end_mask_0 = const()[name = tensor<string, []>("op_50352_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50352_cast = slice_by_index(begin = var_50352_begin_0, end = var_50352_end_0, end_mask = var_50352_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50352_cast")];
            tensor<int32, [4]> var_50356_begin_0 = const()[name = tensor<string, []>("op_50356_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_50356_end_0 = const()[name = tensor<string, []>("op_50356_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_50356_end_mask_0 = const()[name = tensor<string, []>("op_50356_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50356_cast = slice_by_index(begin = var_50356_begin_0, end = var_50356_end_0, end_mask = var_50356_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50356_cast")];
            tensor<int32, [4]> var_50360_begin_0 = const()[name = tensor<string, []>("op_50360_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_50360_end_0 = const()[name = tensor<string, []>("op_50360_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_50360_end_mask_0 = const()[name = tensor<string, []>("op_50360_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50360_cast = slice_by_index(begin = var_50360_begin_0, end = var_50360_end_0, end_mask = var_50360_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50360_cast")];
            tensor<int32, [4]> var_50364_begin_0 = const()[name = tensor<string, []>("op_50364_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_50364_end_0 = const()[name = tensor<string, []>("op_50364_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_50364_end_mask_0 = const()[name = tensor<string, []>("op_50364_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50364_cast = slice_by_index(begin = var_50364_begin_0, end = var_50364_end_0, end_mask = var_50364_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50364_cast")];
            tensor<int32, [4]> var_50368_begin_0 = const()[name = tensor<string, []>("op_50368_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_50368_end_0 = const()[name = tensor<string, []>("op_50368_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_50368_end_mask_0 = const()[name = tensor<string, []>("op_50368_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50368_cast = slice_by_index(begin = var_50368_begin_0, end = var_50368_end_0, end_mask = var_50368_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50368_cast")];
            tensor<int32, [4]> var_50372_begin_0 = const()[name = tensor<string, []>("op_50372_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_50372_end_0 = const()[name = tensor<string, []>("op_50372_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_50372_end_mask_0 = const()[name = tensor<string, []>("op_50372_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50372_cast = slice_by_index(begin = var_50372_begin_0, end = var_50372_end_0, end_mask = var_50372_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50372_cast")];
            tensor<int32, [4]> var_50376_begin_0 = const()[name = tensor<string, []>("op_50376_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_50376_end_0 = const()[name = tensor<string, []>("op_50376_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_50376_end_mask_0 = const()[name = tensor<string, []>("op_50376_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50376_cast = slice_by_index(begin = var_50376_begin_0, end = var_50376_end_0, end_mask = var_50376_end_mask_0, x = q_223_cast)[name = tensor<string, []>("op_50376_cast")];
            tensor<int32, [4]> k_447_perm_0 = const()[name = tensor<string, []>("k_447_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_50383_begin_0 = const()[name = tensor<string, []>("op_50383_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_50383_end_0 = const()[name = tensor<string, []>("op_50383_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_50383_end_mask_0 = const()[name = tensor<string, []>("op_50383_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_28 = transpose(perm = k_447_perm_0, x = k_445_cast)[name = tensor<string, []>("transpose_28")];
            tensor<fp16, [2, 77, 1, 64]> var_50383_cast = slice_by_index(begin = var_50383_begin_0, end = var_50383_end_0, end_mask = var_50383_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50383_cast")];
            tensor<int32, [4]> var_50387_begin_0 = const()[name = tensor<string, []>("op_50387_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_50387_end_0 = const()[name = tensor<string, []>("op_50387_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_50387_end_mask_0 = const()[name = tensor<string, []>("op_50387_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50387_cast = slice_by_index(begin = var_50387_begin_0, end = var_50387_end_0, end_mask = var_50387_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50387_cast")];
            tensor<int32, [4]> var_50391_begin_0 = const()[name = tensor<string, []>("op_50391_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_50391_end_0 = const()[name = tensor<string, []>("op_50391_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_50391_end_mask_0 = const()[name = tensor<string, []>("op_50391_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50391_cast = slice_by_index(begin = var_50391_begin_0, end = var_50391_end_0, end_mask = var_50391_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50391_cast")];
            tensor<int32, [4]> var_50395_begin_0 = const()[name = tensor<string, []>("op_50395_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_50395_end_0 = const()[name = tensor<string, []>("op_50395_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_50395_end_mask_0 = const()[name = tensor<string, []>("op_50395_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50395_cast = slice_by_index(begin = var_50395_begin_0, end = var_50395_end_0, end_mask = var_50395_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50395_cast")];
            tensor<int32, [4]> var_50399_begin_0 = const()[name = tensor<string, []>("op_50399_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_50399_end_0 = const()[name = tensor<string, []>("op_50399_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_50399_end_mask_0 = const()[name = tensor<string, []>("op_50399_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50399_cast = slice_by_index(begin = var_50399_begin_0, end = var_50399_end_0, end_mask = var_50399_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50399_cast")];
            tensor<int32, [4]> var_50403_begin_0 = const()[name = tensor<string, []>("op_50403_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_50403_end_0 = const()[name = tensor<string, []>("op_50403_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_50403_end_mask_0 = const()[name = tensor<string, []>("op_50403_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50403_cast = slice_by_index(begin = var_50403_begin_0, end = var_50403_end_0, end_mask = var_50403_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50403_cast")];
            tensor<int32, [4]> var_50407_begin_0 = const()[name = tensor<string, []>("op_50407_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_50407_end_0 = const()[name = tensor<string, []>("op_50407_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_50407_end_mask_0 = const()[name = tensor<string, []>("op_50407_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50407_cast = slice_by_index(begin = var_50407_begin_0, end = var_50407_end_0, end_mask = var_50407_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50407_cast")];
            tensor<int32, [4]> var_50411_begin_0 = const()[name = tensor<string, []>("op_50411_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_50411_end_0 = const()[name = tensor<string, []>("op_50411_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_50411_end_mask_0 = const()[name = tensor<string, []>("op_50411_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50411_cast = slice_by_index(begin = var_50411_begin_0, end = var_50411_end_0, end_mask = var_50411_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50411_cast")];
            tensor<int32, [4]> var_50415_begin_0 = const()[name = tensor<string, []>("op_50415_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_50415_end_0 = const()[name = tensor<string, []>("op_50415_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_50415_end_mask_0 = const()[name = tensor<string, []>("op_50415_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50415_cast = slice_by_index(begin = var_50415_begin_0, end = var_50415_end_0, end_mask = var_50415_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50415_cast")];
            tensor<int32, [4]> var_50419_begin_0 = const()[name = tensor<string, []>("op_50419_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_50419_end_0 = const()[name = tensor<string, []>("op_50419_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_50419_end_mask_0 = const()[name = tensor<string, []>("op_50419_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50419_cast = slice_by_index(begin = var_50419_begin_0, end = var_50419_end_0, end_mask = var_50419_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50419_cast")];
            tensor<int32, [4]> var_50423_begin_0 = const()[name = tensor<string, []>("op_50423_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_50423_end_0 = const()[name = tensor<string, []>("op_50423_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_50423_end_mask_0 = const()[name = tensor<string, []>("op_50423_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50423_cast = slice_by_index(begin = var_50423_begin_0, end = var_50423_end_0, end_mask = var_50423_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50423_cast")];
            tensor<int32, [4]> var_50427_begin_0 = const()[name = tensor<string, []>("op_50427_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_50427_end_0 = const()[name = tensor<string, []>("op_50427_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_50427_end_mask_0 = const()[name = tensor<string, []>("op_50427_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50427_cast = slice_by_index(begin = var_50427_begin_0, end = var_50427_end_0, end_mask = var_50427_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50427_cast")];
            tensor<int32, [4]> var_50431_begin_0 = const()[name = tensor<string, []>("op_50431_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_50431_end_0 = const()[name = tensor<string, []>("op_50431_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_50431_end_mask_0 = const()[name = tensor<string, []>("op_50431_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50431_cast = slice_by_index(begin = var_50431_begin_0, end = var_50431_end_0, end_mask = var_50431_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50431_cast")];
            tensor<int32, [4]> var_50435_begin_0 = const()[name = tensor<string, []>("op_50435_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_50435_end_0 = const()[name = tensor<string, []>("op_50435_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_50435_end_mask_0 = const()[name = tensor<string, []>("op_50435_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50435_cast = slice_by_index(begin = var_50435_begin_0, end = var_50435_end_0, end_mask = var_50435_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50435_cast")];
            tensor<int32, [4]> var_50439_begin_0 = const()[name = tensor<string, []>("op_50439_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_50439_end_0 = const()[name = tensor<string, []>("op_50439_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_50439_end_mask_0 = const()[name = tensor<string, []>("op_50439_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50439_cast = slice_by_index(begin = var_50439_begin_0, end = var_50439_end_0, end_mask = var_50439_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50439_cast")];
            tensor<int32, [4]> var_50443_begin_0 = const()[name = tensor<string, []>("op_50443_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_50443_end_0 = const()[name = tensor<string, []>("op_50443_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_50443_end_mask_0 = const()[name = tensor<string, []>("op_50443_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50443_cast = slice_by_index(begin = var_50443_begin_0, end = var_50443_end_0, end_mask = var_50443_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50443_cast")];
            tensor<int32, [4]> var_50447_begin_0 = const()[name = tensor<string, []>("op_50447_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_50447_end_0 = const()[name = tensor<string, []>("op_50447_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_50447_end_mask_0 = const()[name = tensor<string, []>("op_50447_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50447_cast = slice_by_index(begin = var_50447_begin_0, end = var_50447_end_0, end_mask = var_50447_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50447_cast")];
            tensor<int32, [4]> var_50451_begin_0 = const()[name = tensor<string, []>("op_50451_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_50451_end_0 = const()[name = tensor<string, []>("op_50451_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_50451_end_mask_0 = const()[name = tensor<string, []>("op_50451_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50451_cast = slice_by_index(begin = var_50451_begin_0, end = var_50451_end_0, end_mask = var_50451_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50451_cast")];
            tensor<int32, [4]> var_50455_begin_0 = const()[name = tensor<string, []>("op_50455_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_50455_end_0 = const()[name = tensor<string, []>("op_50455_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_50455_end_mask_0 = const()[name = tensor<string, []>("op_50455_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50455_cast = slice_by_index(begin = var_50455_begin_0, end = var_50455_end_0, end_mask = var_50455_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50455_cast")];
            tensor<int32, [4]> var_50459_begin_0 = const()[name = tensor<string, []>("op_50459_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_50459_end_0 = const()[name = tensor<string, []>("op_50459_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_50459_end_mask_0 = const()[name = tensor<string, []>("op_50459_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_50459_cast = slice_by_index(begin = var_50459_begin_0, end = var_50459_end_0, end_mask = var_50459_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_50459_cast")];
            tensor<int32, [4]> var_50461_begin_0 = const()[name = tensor<string, []>("op_50461_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_50461_end_0 = const()[name = tensor<string, []>("op_50461_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_50461_end_mask_0 = const()[name = tensor<string, []>("op_50461_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50461_cast = slice_by_index(begin = var_50461_begin_0, end = var_50461_end_0, end_mask = var_50461_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50461_cast")];
            tensor<int32, [4]> var_50465_begin_0 = const()[name = tensor<string, []>("op_50465_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_50465_end_0 = const()[name = tensor<string, []>("op_50465_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_50465_end_mask_0 = const()[name = tensor<string, []>("op_50465_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50465_cast = slice_by_index(begin = var_50465_begin_0, end = var_50465_end_0, end_mask = var_50465_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50465_cast")];
            tensor<int32, [4]> var_50469_begin_0 = const()[name = tensor<string, []>("op_50469_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_50469_end_0 = const()[name = tensor<string, []>("op_50469_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_50469_end_mask_0 = const()[name = tensor<string, []>("op_50469_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50469_cast = slice_by_index(begin = var_50469_begin_0, end = var_50469_end_0, end_mask = var_50469_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50469_cast")];
            tensor<int32, [4]> var_50473_begin_0 = const()[name = tensor<string, []>("op_50473_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_50473_end_0 = const()[name = tensor<string, []>("op_50473_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_50473_end_mask_0 = const()[name = tensor<string, []>("op_50473_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50473_cast = slice_by_index(begin = var_50473_begin_0, end = var_50473_end_0, end_mask = var_50473_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50473_cast")];
            tensor<int32, [4]> var_50477_begin_0 = const()[name = tensor<string, []>("op_50477_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_50477_end_0 = const()[name = tensor<string, []>("op_50477_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_50477_end_mask_0 = const()[name = tensor<string, []>("op_50477_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50477_cast = slice_by_index(begin = var_50477_begin_0, end = var_50477_end_0, end_mask = var_50477_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50477_cast")];
            tensor<int32, [4]> var_50481_begin_0 = const()[name = tensor<string, []>("op_50481_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_50481_end_0 = const()[name = tensor<string, []>("op_50481_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_50481_end_mask_0 = const()[name = tensor<string, []>("op_50481_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50481_cast = slice_by_index(begin = var_50481_begin_0, end = var_50481_end_0, end_mask = var_50481_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50481_cast")];
            tensor<int32, [4]> var_50485_begin_0 = const()[name = tensor<string, []>("op_50485_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_50485_end_0 = const()[name = tensor<string, []>("op_50485_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_50485_end_mask_0 = const()[name = tensor<string, []>("op_50485_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50485_cast = slice_by_index(begin = var_50485_begin_0, end = var_50485_end_0, end_mask = var_50485_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50485_cast")];
            tensor<int32, [4]> var_50489_begin_0 = const()[name = tensor<string, []>("op_50489_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_50489_end_0 = const()[name = tensor<string, []>("op_50489_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_50489_end_mask_0 = const()[name = tensor<string, []>("op_50489_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50489_cast = slice_by_index(begin = var_50489_begin_0, end = var_50489_end_0, end_mask = var_50489_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50489_cast")];
            tensor<int32, [4]> var_50493_begin_0 = const()[name = tensor<string, []>("op_50493_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_50493_end_0 = const()[name = tensor<string, []>("op_50493_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_50493_end_mask_0 = const()[name = tensor<string, []>("op_50493_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50493_cast = slice_by_index(begin = var_50493_begin_0, end = var_50493_end_0, end_mask = var_50493_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50493_cast")];
            tensor<int32, [4]> var_50497_begin_0 = const()[name = tensor<string, []>("op_50497_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_50497_end_0 = const()[name = tensor<string, []>("op_50497_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_50497_end_mask_0 = const()[name = tensor<string, []>("op_50497_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50497_cast = slice_by_index(begin = var_50497_begin_0, end = var_50497_end_0, end_mask = var_50497_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50497_cast")];
            tensor<int32, [4]> var_50501_begin_0 = const()[name = tensor<string, []>("op_50501_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_50501_end_0 = const()[name = tensor<string, []>("op_50501_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_50501_end_mask_0 = const()[name = tensor<string, []>("op_50501_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50501_cast = slice_by_index(begin = var_50501_begin_0, end = var_50501_end_0, end_mask = var_50501_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50501_cast")];
            tensor<int32, [4]> var_50505_begin_0 = const()[name = tensor<string, []>("op_50505_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_50505_end_0 = const()[name = tensor<string, []>("op_50505_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_50505_end_mask_0 = const()[name = tensor<string, []>("op_50505_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50505_cast = slice_by_index(begin = var_50505_begin_0, end = var_50505_end_0, end_mask = var_50505_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50505_cast")];
            tensor<int32, [4]> var_50509_begin_0 = const()[name = tensor<string, []>("op_50509_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_50509_end_0 = const()[name = tensor<string, []>("op_50509_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_50509_end_mask_0 = const()[name = tensor<string, []>("op_50509_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50509_cast = slice_by_index(begin = var_50509_begin_0, end = var_50509_end_0, end_mask = var_50509_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50509_cast")];
            tensor<int32, [4]> var_50513_begin_0 = const()[name = tensor<string, []>("op_50513_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_50513_end_0 = const()[name = tensor<string, []>("op_50513_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_50513_end_mask_0 = const()[name = tensor<string, []>("op_50513_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50513_cast = slice_by_index(begin = var_50513_begin_0, end = var_50513_end_0, end_mask = var_50513_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50513_cast")];
            tensor<int32, [4]> var_50517_begin_0 = const()[name = tensor<string, []>("op_50517_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_50517_end_0 = const()[name = tensor<string, []>("op_50517_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_50517_end_mask_0 = const()[name = tensor<string, []>("op_50517_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50517_cast = slice_by_index(begin = var_50517_begin_0, end = var_50517_end_0, end_mask = var_50517_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50517_cast")];
            tensor<int32, [4]> var_50521_begin_0 = const()[name = tensor<string, []>("op_50521_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_50521_end_0 = const()[name = tensor<string, []>("op_50521_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_50521_end_mask_0 = const()[name = tensor<string, []>("op_50521_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50521_cast = slice_by_index(begin = var_50521_begin_0, end = var_50521_end_0, end_mask = var_50521_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50521_cast")];
            tensor<int32, [4]> var_50525_begin_0 = const()[name = tensor<string, []>("op_50525_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_50525_end_0 = const()[name = tensor<string, []>("op_50525_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_50525_end_mask_0 = const()[name = tensor<string, []>("op_50525_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50525_cast = slice_by_index(begin = var_50525_begin_0, end = var_50525_end_0, end_mask = var_50525_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50525_cast")];
            tensor<int32, [4]> var_50529_begin_0 = const()[name = tensor<string, []>("op_50529_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_50529_end_0 = const()[name = tensor<string, []>("op_50529_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_50529_end_mask_0 = const()[name = tensor<string, []>("op_50529_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50529_cast = slice_by_index(begin = var_50529_begin_0, end = var_50529_end_0, end_mask = var_50529_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50529_cast")];
            tensor<int32, [4]> var_50533_begin_0 = const()[name = tensor<string, []>("op_50533_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_50533_end_0 = const()[name = tensor<string, []>("op_50533_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_50533_end_mask_0 = const()[name = tensor<string, []>("op_50533_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50533_cast = slice_by_index(begin = var_50533_begin_0, end = var_50533_end_0, end_mask = var_50533_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50533_cast")];
            tensor<int32, [4]> var_50537_begin_0 = const()[name = tensor<string, []>("op_50537_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_50537_end_0 = const()[name = tensor<string, []>("op_50537_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_50537_end_mask_0 = const()[name = tensor<string, []>("op_50537_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_50537_cast = slice_by_index(begin = var_50537_begin_0, end = var_50537_end_0, end_mask = var_50537_end_mask_0, x = v_223_cast)[name = tensor<string, []>("op_50537_cast")];
            tensor<string, []> var_50541_equation_0 = const()[name = tensor<string, []>("op_50541_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50541_cast = einsum(equation = var_50541_equation_0, values = (var_50383_cast, var_50300_cast))[name = tensor<string, []>("op_50541_cast")];
            tensor<fp16, []> var_50542_to_fp16 = const()[name = tensor<string, []>("op_50542_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4281_cast = mul(x = var_50541_cast, y = var_50542_to_fp16)[name = tensor<string, []>("aw_4281_cast")];
            tensor<string, []> var_50545_equation_0 = const()[name = tensor<string, []>("op_50545_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50545_cast = einsum(equation = var_50545_equation_0, values = (var_50387_cast, var_50304_cast))[name = tensor<string, []>("op_50545_cast")];
            tensor<fp16, []> var_50546_to_fp16 = const()[name = tensor<string, []>("op_50546_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4283_cast = mul(x = var_50545_cast, y = var_50546_to_fp16)[name = tensor<string, []>("aw_4283_cast")];
            tensor<string, []> var_50549_equation_0 = const()[name = tensor<string, []>("op_50549_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50549_cast = einsum(equation = var_50549_equation_0, values = (var_50391_cast, var_50308_cast))[name = tensor<string, []>("op_50549_cast")];
            tensor<fp16, []> var_50550_to_fp16 = const()[name = tensor<string, []>("op_50550_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4285_cast = mul(x = var_50549_cast, y = var_50550_to_fp16)[name = tensor<string, []>("aw_4285_cast")];
            tensor<string, []> var_50553_equation_0 = const()[name = tensor<string, []>("op_50553_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50553_cast = einsum(equation = var_50553_equation_0, values = (var_50395_cast, var_50312_cast))[name = tensor<string, []>("op_50553_cast")];
            tensor<fp16, []> var_50554_to_fp16 = const()[name = tensor<string, []>("op_50554_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4287_cast = mul(x = var_50553_cast, y = var_50554_to_fp16)[name = tensor<string, []>("aw_4287_cast")];
            tensor<string, []> var_50557_equation_0 = const()[name = tensor<string, []>("op_50557_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50557_cast = einsum(equation = var_50557_equation_0, values = (var_50399_cast, var_50316_cast))[name = tensor<string, []>("op_50557_cast")];
            tensor<fp16, []> var_50558_to_fp16 = const()[name = tensor<string, []>("op_50558_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4289_cast = mul(x = var_50557_cast, y = var_50558_to_fp16)[name = tensor<string, []>("aw_4289_cast")];
            tensor<string, []> var_50561_equation_0 = const()[name = tensor<string, []>("op_50561_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50561_cast = einsum(equation = var_50561_equation_0, values = (var_50403_cast, var_50320_cast))[name = tensor<string, []>("op_50561_cast")];
            tensor<fp16, []> var_50562_to_fp16 = const()[name = tensor<string, []>("op_50562_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4291_cast = mul(x = var_50561_cast, y = var_50562_to_fp16)[name = tensor<string, []>("aw_4291_cast")];
            tensor<string, []> var_50565_equation_0 = const()[name = tensor<string, []>("op_50565_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50565_cast = einsum(equation = var_50565_equation_0, values = (var_50407_cast, var_50324_cast))[name = tensor<string, []>("op_50565_cast")];
            tensor<fp16, []> var_50566_to_fp16 = const()[name = tensor<string, []>("op_50566_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4293_cast = mul(x = var_50565_cast, y = var_50566_to_fp16)[name = tensor<string, []>("aw_4293_cast")];
            tensor<string, []> var_50569_equation_0 = const()[name = tensor<string, []>("op_50569_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50569_cast = einsum(equation = var_50569_equation_0, values = (var_50411_cast, var_50328_cast))[name = tensor<string, []>("op_50569_cast")];
            tensor<fp16, []> var_50570_to_fp16 = const()[name = tensor<string, []>("op_50570_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4295_cast = mul(x = var_50569_cast, y = var_50570_to_fp16)[name = tensor<string, []>("aw_4295_cast")];
            tensor<string, []> var_50573_equation_0 = const()[name = tensor<string, []>("op_50573_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50573_cast = einsum(equation = var_50573_equation_0, values = (var_50415_cast, var_50332_cast))[name = tensor<string, []>("op_50573_cast")];
            tensor<fp16, []> var_50574_to_fp16 = const()[name = tensor<string, []>("op_50574_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4297_cast = mul(x = var_50573_cast, y = var_50574_to_fp16)[name = tensor<string, []>("aw_4297_cast")];
            tensor<string, []> var_50577_equation_0 = const()[name = tensor<string, []>("op_50577_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50577_cast = einsum(equation = var_50577_equation_0, values = (var_50419_cast, var_50336_cast))[name = tensor<string, []>("op_50577_cast")];
            tensor<fp16, []> var_50578_to_fp16 = const()[name = tensor<string, []>("op_50578_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4299_cast = mul(x = var_50577_cast, y = var_50578_to_fp16)[name = tensor<string, []>("aw_4299_cast")];
            tensor<string, []> var_50581_equation_0 = const()[name = tensor<string, []>("op_50581_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50581_cast = einsum(equation = var_50581_equation_0, values = (var_50423_cast, var_50340_cast))[name = tensor<string, []>("op_50581_cast")];
            tensor<fp16, []> var_50582_to_fp16 = const()[name = tensor<string, []>("op_50582_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4301_cast = mul(x = var_50581_cast, y = var_50582_to_fp16)[name = tensor<string, []>("aw_4301_cast")];
            tensor<string, []> var_50585_equation_0 = const()[name = tensor<string, []>("op_50585_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50585_cast = einsum(equation = var_50585_equation_0, values = (var_50427_cast, var_50344_cast))[name = tensor<string, []>("op_50585_cast")];
            tensor<fp16, []> var_50586_to_fp16 = const()[name = tensor<string, []>("op_50586_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4303_cast = mul(x = var_50585_cast, y = var_50586_to_fp16)[name = tensor<string, []>("aw_4303_cast")];
            tensor<string, []> var_50589_equation_0 = const()[name = tensor<string, []>("op_50589_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50589_cast = einsum(equation = var_50589_equation_0, values = (var_50431_cast, var_50348_cast))[name = tensor<string, []>("op_50589_cast")];
            tensor<fp16, []> var_50590_to_fp16 = const()[name = tensor<string, []>("op_50590_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4305_cast = mul(x = var_50589_cast, y = var_50590_to_fp16)[name = tensor<string, []>("aw_4305_cast")];
            tensor<string, []> var_50593_equation_0 = const()[name = tensor<string, []>("op_50593_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50593_cast = einsum(equation = var_50593_equation_0, values = (var_50435_cast, var_50352_cast))[name = tensor<string, []>("op_50593_cast")];
            tensor<fp16, []> var_50594_to_fp16 = const()[name = tensor<string, []>("op_50594_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4307_cast = mul(x = var_50593_cast, y = var_50594_to_fp16)[name = tensor<string, []>("aw_4307_cast")];
            tensor<string, []> var_50597_equation_0 = const()[name = tensor<string, []>("op_50597_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50597_cast = einsum(equation = var_50597_equation_0, values = (var_50439_cast, var_50356_cast))[name = tensor<string, []>("op_50597_cast")];
            tensor<fp16, []> var_50598_to_fp16 = const()[name = tensor<string, []>("op_50598_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4309_cast = mul(x = var_50597_cast, y = var_50598_to_fp16)[name = tensor<string, []>("aw_4309_cast")];
            tensor<string, []> var_50601_equation_0 = const()[name = tensor<string, []>("op_50601_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50601_cast = einsum(equation = var_50601_equation_0, values = (var_50443_cast, var_50360_cast))[name = tensor<string, []>("op_50601_cast")];
            tensor<fp16, []> var_50602_to_fp16 = const()[name = tensor<string, []>("op_50602_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4311_cast = mul(x = var_50601_cast, y = var_50602_to_fp16)[name = tensor<string, []>("aw_4311_cast")];
            tensor<string, []> var_50605_equation_0 = const()[name = tensor<string, []>("op_50605_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50605_cast = einsum(equation = var_50605_equation_0, values = (var_50447_cast, var_50364_cast))[name = tensor<string, []>("op_50605_cast")];
            tensor<fp16, []> var_50606_to_fp16 = const()[name = tensor<string, []>("op_50606_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4313_cast = mul(x = var_50605_cast, y = var_50606_to_fp16)[name = tensor<string, []>("aw_4313_cast")];
            tensor<string, []> var_50609_equation_0 = const()[name = tensor<string, []>("op_50609_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50609_cast = einsum(equation = var_50609_equation_0, values = (var_50451_cast, var_50368_cast))[name = tensor<string, []>("op_50609_cast")];
            tensor<fp16, []> var_50610_to_fp16 = const()[name = tensor<string, []>("op_50610_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4315_cast = mul(x = var_50609_cast, y = var_50610_to_fp16)[name = tensor<string, []>("aw_4315_cast")];
            tensor<string, []> var_50613_equation_0 = const()[name = tensor<string, []>("op_50613_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50613_cast = einsum(equation = var_50613_equation_0, values = (var_50455_cast, var_50372_cast))[name = tensor<string, []>("op_50613_cast")];
            tensor<fp16, []> var_50614_to_fp16 = const()[name = tensor<string, []>("op_50614_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4317_cast = mul(x = var_50613_cast, y = var_50614_to_fp16)[name = tensor<string, []>("aw_4317_cast")];
            tensor<string, []> var_50617_equation_0 = const()[name = tensor<string, []>("op_50617_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_50617_cast = einsum(equation = var_50617_equation_0, values = (var_50459_cast, var_50376_cast))[name = tensor<string, []>("op_50617_cast")];
            tensor<fp16, []> var_50618_to_fp16 = const()[name = tensor<string, []>("op_50618_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4319_cast = mul(x = var_50617_cast, y = var_50618_to_fp16)[name = tensor<string, []>("aw_4319_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50620_cast = softmax(axis = var_30385, x = aw_4281_cast)[name = tensor<string, []>("op_50620_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50621_cast = softmax(axis = var_30385, x = aw_4283_cast)[name = tensor<string, []>("op_50621_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50622_cast = softmax(axis = var_30385, x = aw_4285_cast)[name = tensor<string, []>("op_50622_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50623_cast = softmax(axis = var_30385, x = aw_4287_cast)[name = tensor<string, []>("op_50623_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50624_cast = softmax(axis = var_30385, x = aw_4289_cast)[name = tensor<string, []>("op_50624_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50625_cast = softmax(axis = var_30385, x = aw_4291_cast)[name = tensor<string, []>("op_50625_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50626_cast = softmax(axis = var_30385, x = aw_4293_cast)[name = tensor<string, []>("op_50626_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50627_cast = softmax(axis = var_30385, x = aw_4295_cast)[name = tensor<string, []>("op_50627_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50628_cast = softmax(axis = var_30385, x = aw_4297_cast)[name = tensor<string, []>("op_50628_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50629_cast = softmax(axis = var_30385, x = aw_4299_cast)[name = tensor<string, []>("op_50629_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50630_cast = softmax(axis = var_30385, x = aw_4301_cast)[name = tensor<string, []>("op_50630_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50631_cast = softmax(axis = var_30385, x = aw_4303_cast)[name = tensor<string, []>("op_50631_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50632_cast = softmax(axis = var_30385, x = aw_4305_cast)[name = tensor<string, []>("op_50632_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50633_cast = softmax(axis = var_30385, x = aw_4307_cast)[name = tensor<string, []>("op_50633_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50634_cast = softmax(axis = var_30385, x = aw_4309_cast)[name = tensor<string, []>("op_50634_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50635_cast = softmax(axis = var_30385, x = aw_4311_cast)[name = tensor<string, []>("op_50635_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50636_cast = softmax(axis = var_30385, x = aw_4313_cast)[name = tensor<string, []>("op_50636_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50637_cast = softmax(axis = var_30385, x = aw_4315_cast)[name = tensor<string, []>("op_50637_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50638_cast = softmax(axis = var_30385, x = aw_4317_cast)[name = tensor<string, []>("op_50638_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_50639_cast = softmax(axis = var_30385, x = aw_4319_cast)[name = tensor<string, []>("op_50639_cast")];
            tensor<string, []> var_50641_equation_0 = const()[name = tensor<string, []>("op_50641_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50641_cast = einsum(equation = var_50641_equation_0, values = (var_50461_cast, var_50620_cast))[name = tensor<string, []>("op_50641_cast")];
            tensor<string, []> var_50643_equation_0 = const()[name = tensor<string, []>("op_50643_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50643_cast = einsum(equation = var_50643_equation_0, values = (var_50465_cast, var_50621_cast))[name = tensor<string, []>("op_50643_cast")];
            tensor<string, []> var_50645_equation_0 = const()[name = tensor<string, []>("op_50645_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50645_cast = einsum(equation = var_50645_equation_0, values = (var_50469_cast, var_50622_cast))[name = tensor<string, []>("op_50645_cast")];
            tensor<string, []> var_50647_equation_0 = const()[name = tensor<string, []>("op_50647_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50647_cast = einsum(equation = var_50647_equation_0, values = (var_50473_cast, var_50623_cast))[name = tensor<string, []>("op_50647_cast")];
            tensor<string, []> var_50649_equation_0 = const()[name = tensor<string, []>("op_50649_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50649_cast = einsum(equation = var_50649_equation_0, values = (var_50477_cast, var_50624_cast))[name = tensor<string, []>("op_50649_cast")];
            tensor<string, []> var_50651_equation_0 = const()[name = tensor<string, []>("op_50651_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50651_cast = einsum(equation = var_50651_equation_0, values = (var_50481_cast, var_50625_cast))[name = tensor<string, []>("op_50651_cast")];
            tensor<string, []> var_50653_equation_0 = const()[name = tensor<string, []>("op_50653_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50653_cast = einsum(equation = var_50653_equation_0, values = (var_50485_cast, var_50626_cast))[name = tensor<string, []>("op_50653_cast")];
            tensor<string, []> var_50655_equation_0 = const()[name = tensor<string, []>("op_50655_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50655_cast = einsum(equation = var_50655_equation_0, values = (var_50489_cast, var_50627_cast))[name = tensor<string, []>("op_50655_cast")];
            tensor<string, []> var_50657_equation_0 = const()[name = tensor<string, []>("op_50657_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50657_cast = einsum(equation = var_50657_equation_0, values = (var_50493_cast, var_50628_cast))[name = tensor<string, []>("op_50657_cast")];
            tensor<string, []> var_50659_equation_0 = const()[name = tensor<string, []>("op_50659_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50659_cast = einsum(equation = var_50659_equation_0, values = (var_50497_cast, var_50629_cast))[name = tensor<string, []>("op_50659_cast")];
            tensor<string, []> var_50661_equation_0 = const()[name = tensor<string, []>("op_50661_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50661_cast = einsum(equation = var_50661_equation_0, values = (var_50501_cast, var_50630_cast))[name = tensor<string, []>("op_50661_cast")];
            tensor<string, []> var_50663_equation_0 = const()[name = tensor<string, []>("op_50663_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50663_cast = einsum(equation = var_50663_equation_0, values = (var_50505_cast, var_50631_cast))[name = tensor<string, []>("op_50663_cast")];
            tensor<string, []> var_50665_equation_0 = const()[name = tensor<string, []>("op_50665_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50665_cast = einsum(equation = var_50665_equation_0, values = (var_50509_cast, var_50632_cast))[name = tensor<string, []>("op_50665_cast")];
            tensor<string, []> var_50667_equation_0 = const()[name = tensor<string, []>("op_50667_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50667_cast = einsum(equation = var_50667_equation_0, values = (var_50513_cast, var_50633_cast))[name = tensor<string, []>("op_50667_cast")];
            tensor<string, []> var_50669_equation_0 = const()[name = tensor<string, []>("op_50669_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50669_cast = einsum(equation = var_50669_equation_0, values = (var_50517_cast, var_50634_cast))[name = tensor<string, []>("op_50669_cast")];
            tensor<string, []> var_50671_equation_0 = const()[name = tensor<string, []>("op_50671_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50671_cast = einsum(equation = var_50671_equation_0, values = (var_50521_cast, var_50635_cast))[name = tensor<string, []>("op_50671_cast")];
            tensor<string, []> var_50673_equation_0 = const()[name = tensor<string, []>("op_50673_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50673_cast = einsum(equation = var_50673_equation_0, values = (var_50525_cast, var_50636_cast))[name = tensor<string, []>("op_50673_cast")];
            tensor<string, []> var_50675_equation_0 = const()[name = tensor<string, []>("op_50675_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50675_cast = einsum(equation = var_50675_equation_0, values = (var_50529_cast, var_50637_cast))[name = tensor<string, []>("op_50675_cast")];
            tensor<string, []> var_50677_equation_0 = const()[name = tensor<string, []>("op_50677_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50677_cast = einsum(equation = var_50677_equation_0, values = (var_50533_cast, var_50638_cast))[name = tensor<string, []>("op_50677_cast")];
            tensor<string, []> var_50679_equation_0 = const()[name = tensor<string, []>("op_50679_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_50679_cast = einsum(equation = var_50679_equation_0, values = (var_50537_cast, var_50639_cast))[name = tensor<string, []>("op_50679_cast")];
            tensor<bool, []> input_647_interleave_0 = const()[name = tensor<string, []>("input_647_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_647_cast = concat(axis = var_30385, interleave = input_647_interleave_0, values = (var_50641_cast, var_50643_cast, var_50645_cast, var_50647_cast, var_50649_cast, var_50651_cast, var_50653_cast, var_50655_cast, var_50657_cast, var_50659_cast, var_50661_cast, var_50663_cast, var_50665_cast, var_50667_cast, var_50669_cast, var_50671_cast, var_50673_cast, var_50675_cast, var_50677_cast, var_50679_cast))[name = tensor<string, []>("input_647_cast")];
            tensor<int32, [2]> var_50685 = const()[name = tensor<string, []>("op_50685"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_50687 = const()[name = tensor<string, []>("op_50687"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_50689_pad_type_0 = const()[name = tensor<string, []>("op_50689_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_50689_pad_0 = const()[name = tensor<string, []>("op_50689_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4268487296)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4271764160)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_50689_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_50687, groups = var_30385, pad = var_50689_pad_0, pad_type = var_50689_pad_type_0, strides = var_50685, weight = up_blocks_0_attentions_2_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_647_cast)[name = tensor<string, []>("op_50689_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_335_cast = add(x = var_50689_cast, y = inputs_333_cast)[name = tensor<string, []>("inputs_335_cast")];
            tensor<int32, [1]> var_50693 = const()[name = tensor<string, []>("op_50693"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_335_cast = reduce_mean(axes = var_50693, keep_dims = var_30380, x = inputs_335_cast)[name = tensor<string, []>("channels_mean_335_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_335_cast = sub(x = inputs_335_cast, y = channels_mean_335_cast)[name = tensor<string, []>("zero_mean_335_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_335_cast = mul(x = zero_mean_335_cast, y = zero_mean_335_cast)[name = tensor<string, []>("zero_mean_sq_335_cast")];
            tensor<int32, [1]> var_50697 = const()[name = tensor<string, []>("op_50697"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_50698_cast = reduce_mean(axes = var_50697, keep_dims = var_30380, x = zero_mean_sq_335_cast)[name = tensor<string, []>("op_50698_cast")];
            tensor<fp16, []> var_50699_to_fp16 = const()[name = tensor<string, []>("op_50699_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_50700_cast = add(x = var_50698_cast, y = var_50699_to_fp16)[name = tensor<string, []>("op_50700_cast")];
            tensor<fp16, []> denom_335_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_335_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_335_cast = rsqrt(epsilon = denom_335_epsilon_0_to_fp16, x = var_50700_cast)[name = tensor<string, []>("denom_335_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_335_cast = mul(x = zero_mean_335_cast, y = denom_335_cast)[name = tensor<string, []>("out_335_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_50704_to_fp16 = const()[name = tensor<string, []>("op_50704_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4271766784)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_50705_cast = add(x = out_335_cast, y = var_50704_to_fp16)[name = tensor<string, []>("op_50705_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_50707_to_fp16 = const()[name = tensor<string, []>("op_50707_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4271769408)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_649_cast = mul(x = var_50705_cast, y = var_50707_to_fp16)[name = tensor<string, []>("input_649_cast")];
            tensor<int32, [2]> var_50715 = const()[name = tensor<string, []>("op_50715"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_50717 = const()[name = tensor<string, []>("op_50717"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_50719_pad_type_0 = const()[name = tensor<string, []>("op_50719_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_50719_pad_0 = const()[name = tensor<string, []>("op_50719_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4271772032)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_2_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4297986496)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_50719_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_50717, groups = var_30385, pad = var_50719_pad_0, pad_type = var_50719_pad_type_0, strides = var_50715, weight = up_blocks_0_attentions_2_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_649_cast)[name = tensor<string, []>("op_50719_cast")];
            tensor<int32, [2]> var_50720_split_sizes_0 = const()[name = tensor<string, []>("op_50720_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_50720_axis_0 = const()[name = tensor<string, []>("op_50720_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_50720_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_50720_cast_1 = split(axis = var_50720_axis_0, split_sizes = var_50720_split_sizes_0, x = var_50719_cast)[name = tensor<string, []>("op_50720_cast")];
            tensor<string, []> var_50722_mode_0 = const()[name = tensor<string, []>("op_50722_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_50722_cast = gelu(mode = var_50722_mode_0, x = var_50720_cast_1)[name = tensor<string, []>("op_50722_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_651_cast = mul(x = var_50720_cast_0, y = var_50722_cast)[name = tensor<string, []>("input_651_cast")];
            tensor<int32, [2]> var_50726 = const()[name = tensor<string, []>("op_50726"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_50728 = const()[name = tensor<string, []>("op_50728"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_50730_pad_type_0 = const()[name = tensor<string, []>("op_50730_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_50730_pad_0 = const()[name = tensor<string, []>("op_50730_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4298007040)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4311114304)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_50730_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_50728, groups = var_30385, pad = var_50730_pad_0, pad_type = var_50730_pad_type_0, strides = var_50726, weight = up_blocks_0_attentions_2_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_651_cast)[name = tensor<string, []>("op_50730_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_337_cast = add(x = var_50730_cast, y = inputs_335_cast)[name = tensor<string, []>("inputs_337_cast")];
            tensor<int32, [1]> var_50740 = const()[name = tensor<string, []>("op_50740"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_337_cast = reduce_mean(axes = var_50740, keep_dims = var_30380, x = inputs_337_cast)[name = tensor<string, []>("channels_mean_337_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_337_cast = sub(x = inputs_337_cast, y = channels_mean_337_cast)[name = tensor<string, []>("zero_mean_337_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_337_cast = mul(x = zero_mean_337_cast, y = zero_mean_337_cast)[name = tensor<string, []>("zero_mean_sq_337_cast")];
            tensor<int32, [1]> var_50744 = const()[name = tensor<string, []>("op_50744"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_50745_cast = reduce_mean(axes = var_50744, keep_dims = var_30380, x = zero_mean_sq_337_cast)[name = tensor<string, []>("op_50745_cast")];
            tensor<fp16, []> var_50746_to_fp16 = const()[name = tensor<string, []>("op_50746_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_50747_cast = add(x = var_50745_cast, y = var_50746_to_fp16)[name = tensor<string, []>("op_50747_cast")];
            tensor<fp16, []> denom_337_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_337_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_337_cast = rsqrt(epsilon = denom_337_epsilon_0_to_fp16, x = var_50747_cast)[name = tensor<string, []>("denom_337_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_337_cast = mul(x = zero_mean_337_cast, y = denom_337_cast)[name = tensor<string, []>("out_337_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_50751_to_fp16 = const()[name = tensor<string, []>("op_50751_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4311116928)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_50752_cast = add(x = out_337_cast, y = var_50751_to_fp16)[name = tensor<string, []>("op_50752_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_50754_to_fp16 = const()[name = tensor<string, []>("op_50754_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4311119552)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_447_cast = mul(x = var_50752_cast, y = var_50754_to_fp16)[name = tensor<string, []>("hidden_states_447_cast")];
            tensor<int32, [2]> var_50761 = const()[name = tensor<string, []>("op_50761"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_50763 = const()[name = tensor<string, []>("op_50763"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_225_pad_type_0 = const()[name = tensor<string, []>("q_225_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_225_pad_0 = const()[name = tensor<string, []>("q_225_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4311122176)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_225_cast = conv(dilations = var_50763, groups = var_30385, pad = q_225_pad_0, pad_type = q_225_pad_type_0, strides = var_50761, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_q_weight_to_fp16, x = hidden_states_447_cast)[name = tensor<string, []>("q_225_cast")];
            tensor<int32, [2]> var_50767 = const()[name = tensor<string, []>("op_50767"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_50769 = const()[name = tensor<string, []>("op_50769"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_449_pad_type_0 = const()[name = tensor<string, []>("k_449_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_449_pad_0 = const()[name = tensor<string, []>("k_449_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4314399040)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_449_cast = conv(dilations = var_50769, groups = var_30385, pad = k_449_pad_0, pad_type = k_449_pad_type_0, strides = var_50767, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_k_weight_to_fp16, x = hidden_states_447_cast)[name = tensor<string, []>("k_449_cast")];
            tensor<int32, [2]> var_50773 = const()[name = tensor<string, []>("op_50773"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_50775 = const()[name = tensor<string, []>("op_50775"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_225_pad_type_0 = const()[name = tensor<string, []>("v_225_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_225_pad_0 = const()[name = tensor<string, []>("v_225_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4317675904)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_225_cast = conv(dilations = var_50775, groups = var_30385, pad = v_225_pad_0, pad_type = v_225_pad_type_0, strides = var_50773, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_v_weight_to_fp16, x = hidden_states_447_cast)[name = tensor<string, []>("v_225_cast")];
            tensor<int32, [4]> var_50779_begin_0 = const()[name = tensor<string, []>("op_50779_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_50779_end_0 = const()[name = tensor<string, []>("op_50779_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_50779_end_mask_0 = const()[name = tensor<string, []>("op_50779_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50779_cast = slice_by_index(begin = var_50779_begin_0, end = var_50779_end_0, end_mask = var_50779_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50779_cast")];
            tensor<int32, [4]> var_50783_begin_0 = const()[name = tensor<string, []>("op_50783_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_50783_end_0 = const()[name = tensor<string, []>("op_50783_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_50783_end_mask_0 = const()[name = tensor<string, []>("op_50783_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50783_cast = slice_by_index(begin = var_50783_begin_0, end = var_50783_end_0, end_mask = var_50783_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50783_cast")];
            tensor<int32, [4]> var_50787_begin_0 = const()[name = tensor<string, []>("op_50787_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_50787_end_0 = const()[name = tensor<string, []>("op_50787_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_50787_end_mask_0 = const()[name = tensor<string, []>("op_50787_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50787_cast = slice_by_index(begin = var_50787_begin_0, end = var_50787_end_0, end_mask = var_50787_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50787_cast")];
            tensor<int32, [4]> var_50791_begin_0 = const()[name = tensor<string, []>("op_50791_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_50791_end_0 = const()[name = tensor<string, []>("op_50791_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_50791_end_mask_0 = const()[name = tensor<string, []>("op_50791_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50791_cast = slice_by_index(begin = var_50791_begin_0, end = var_50791_end_0, end_mask = var_50791_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50791_cast")];
            tensor<int32, [4]> var_50795_begin_0 = const()[name = tensor<string, []>("op_50795_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_50795_end_0 = const()[name = tensor<string, []>("op_50795_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_50795_end_mask_0 = const()[name = tensor<string, []>("op_50795_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50795_cast = slice_by_index(begin = var_50795_begin_0, end = var_50795_end_0, end_mask = var_50795_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50795_cast")];
            tensor<int32, [4]> var_50799_begin_0 = const()[name = tensor<string, []>("op_50799_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_50799_end_0 = const()[name = tensor<string, []>("op_50799_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_50799_end_mask_0 = const()[name = tensor<string, []>("op_50799_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50799_cast = slice_by_index(begin = var_50799_begin_0, end = var_50799_end_0, end_mask = var_50799_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50799_cast")];
            tensor<int32, [4]> var_50803_begin_0 = const()[name = tensor<string, []>("op_50803_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_50803_end_0 = const()[name = tensor<string, []>("op_50803_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_50803_end_mask_0 = const()[name = tensor<string, []>("op_50803_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50803_cast = slice_by_index(begin = var_50803_begin_0, end = var_50803_end_0, end_mask = var_50803_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50803_cast")];
            tensor<int32, [4]> var_50807_begin_0 = const()[name = tensor<string, []>("op_50807_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_50807_end_0 = const()[name = tensor<string, []>("op_50807_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_50807_end_mask_0 = const()[name = tensor<string, []>("op_50807_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50807_cast = slice_by_index(begin = var_50807_begin_0, end = var_50807_end_0, end_mask = var_50807_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50807_cast")];
            tensor<int32, [4]> var_50811_begin_0 = const()[name = tensor<string, []>("op_50811_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_50811_end_0 = const()[name = tensor<string, []>("op_50811_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_50811_end_mask_0 = const()[name = tensor<string, []>("op_50811_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50811_cast = slice_by_index(begin = var_50811_begin_0, end = var_50811_end_0, end_mask = var_50811_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50811_cast")];
            tensor<int32, [4]> var_50815_begin_0 = const()[name = tensor<string, []>("op_50815_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_50815_end_0 = const()[name = tensor<string, []>("op_50815_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_50815_end_mask_0 = const()[name = tensor<string, []>("op_50815_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50815_cast = slice_by_index(begin = var_50815_begin_0, end = var_50815_end_0, end_mask = var_50815_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50815_cast")];
            tensor<int32, [4]> var_50819_begin_0 = const()[name = tensor<string, []>("op_50819_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_50819_end_0 = const()[name = tensor<string, []>("op_50819_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_50819_end_mask_0 = const()[name = tensor<string, []>("op_50819_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50819_cast = slice_by_index(begin = var_50819_begin_0, end = var_50819_end_0, end_mask = var_50819_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50819_cast")];
            tensor<int32, [4]> var_50823_begin_0 = const()[name = tensor<string, []>("op_50823_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_50823_end_0 = const()[name = tensor<string, []>("op_50823_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_50823_end_mask_0 = const()[name = tensor<string, []>("op_50823_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50823_cast = slice_by_index(begin = var_50823_begin_0, end = var_50823_end_0, end_mask = var_50823_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50823_cast")];
            tensor<int32, [4]> var_50827_begin_0 = const()[name = tensor<string, []>("op_50827_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_50827_end_0 = const()[name = tensor<string, []>("op_50827_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_50827_end_mask_0 = const()[name = tensor<string, []>("op_50827_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50827_cast = slice_by_index(begin = var_50827_begin_0, end = var_50827_end_0, end_mask = var_50827_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50827_cast")];
            tensor<int32, [4]> var_50831_begin_0 = const()[name = tensor<string, []>("op_50831_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_50831_end_0 = const()[name = tensor<string, []>("op_50831_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_50831_end_mask_0 = const()[name = tensor<string, []>("op_50831_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50831_cast = slice_by_index(begin = var_50831_begin_0, end = var_50831_end_0, end_mask = var_50831_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50831_cast")];
            tensor<int32, [4]> var_50835_begin_0 = const()[name = tensor<string, []>("op_50835_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_50835_end_0 = const()[name = tensor<string, []>("op_50835_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_50835_end_mask_0 = const()[name = tensor<string, []>("op_50835_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50835_cast = slice_by_index(begin = var_50835_begin_0, end = var_50835_end_0, end_mask = var_50835_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50835_cast")];
            tensor<int32, [4]> var_50839_begin_0 = const()[name = tensor<string, []>("op_50839_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_50839_end_0 = const()[name = tensor<string, []>("op_50839_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_50839_end_mask_0 = const()[name = tensor<string, []>("op_50839_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50839_cast = slice_by_index(begin = var_50839_begin_0, end = var_50839_end_0, end_mask = var_50839_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50839_cast")];
            tensor<int32, [4]> var_50843_begin_0 = const()[name = tensor<string, []>("op_50843_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_50843_end_0 = const()[name = tensor<string, []>("op_50843_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_50843_end_mask_0 = const()[name = tensor<string, []>("op_50843_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50843_cast = slice_by_index(begin = var_50843_begin_0, end = var_50843_end_0, end_mask = var_50843_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50843_cast")];
            tensor<int32, [4]> var_50847_begin_0 = const()[name = tensor<string, []>("op_50847_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_50847_end_0 = const()[name = tensor<string, []>("op_50847_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_50847_end_mask_0 = const()[name = tensor<string, []>("op_50847_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50847_cast = slice_by_index(begin = var_50847_begin_0, end = var_50847_end_0, end_mask = var_50847_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50847_cast")];
            tensor<int32, [4]> var_50851_begin_0 = const()[name = tensor<string, []>("op_50851_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_50851_end_0 = const()[name = tensor<string, []>("op_50851_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_50851_end_mask_0 = const()[name = tensor<string, []>("op_50851_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50851_cast = slice_by_index(begin = var_50851_begin_0, end = var_50851_end_0, end_mask = var_50851_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50851_cast")];
            tensor<int32, [4]> var_50855_begin_0 = const()[name = tensor<string, []>("op_50855_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_50855_end_0 = const()[name = tensor<string, []>("op_50855_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_50855_end_mask_0 = const()[name = tensor<string, []>("op_50855_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50855_cast = slice_by_index(begin = var_50855_begin_0, end = var_50855_end_0, end_mask = var_50855_end_mask_0, x = q_225_cast)[name = tensor<string, []>("op_50855_cast")];
            tensor<int32, [4]> k_451_perm_0 = const()[name = tensor<string, []>("k_451_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_50862_begin_0 = const()[name = tensor<string, []>("op_50862_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_50862_end_0 = const()[name = tensor<string, []>("op_50862_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_50862_end_mask_0 = const()[name = tensor<string, []>("op_50862_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_27 = transpose(perm = k_451_perm_0, x = k_449_cast)[name = tensor<string, []>("transpose_27")];
            tensor<fp16, [2, 1024, 1, 64]> var_50862_cast = slice_by_index(begin = var_50862_begin_0, end = var_50862_end_0, end_mask = var_50862_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50862_cast")];
            tensor<int32, [4]> var_50866_begin_0 = const()[name = tensor<string, []>("op_50866_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_50866_end_0 = const()[name = tensor<string, []>("op_50866_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_50866_end_mask_0 = const()[name = tensor<string, []>("op_50866_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50866_cast = slice_by_index(begin = var_50866_begin_0, end = var_50866_end_0, end_mask = var_50866_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50866_cast")];
            tensor<int32, [4]> var_50870_begin_0 = const()[name = tensor<string, []>("op_50870_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_50870_end_0 = const()[name = tensor<string, []>("op_50870_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_50870_end_mask_0 = const()[name = tensor<string, []>("op_50870_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50870_cast = slice_by_index(begin = var_50870_begin_0, end = var_50870_end_0, end_mask = var_50870_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50870_cast")];
            tensor<int32, [4]> var_50874_begin_0 = const()[name = tensor<string, []>("op_50874_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_50874_end_0 = const()[name = tensor<string, []>("op_50874_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_50874_end_mask_0 = const()[name = tensor<string, []>("op_50874_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50874_cast = slice_by_index(begin = var_50874_begin_0, end = var_50874_end_0, end_mask = var_50874_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50874_cast")];
            tensor<int32, [4]> var_50878_begin_0 = const()[name = tensor<string, []>("op_50878_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_50878_end_0 = const()[name = tensor<string, []>("op_50878_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_50878_end_mask_0 = const()[name = tensor<string, []>("op_50878_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50878_cast = slice_by_index(begin = var_50878_begin_0, end = var_50878_end_0, end_mask = var_50878_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50878_cast")];
            tensor<int32, [4]> var_50882_begin_0 = const()[name = tensor<string, []>("op_50882_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_50882_end_0 = const()[name = tensor<string, []>("op_50882_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_50882_end_mask_0 = const()[name = tensor<string, []>("op_50882_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50882_cast = slice_by_index(begin = var_50882_begin_0, end = var_50882_end_0, end_mask = var_50882_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50882_cast")];
            tensor<int32, [4]> var_50886_begin_0 = const()[name = tensor<string, []>("op_50886_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_50886_end_0 = const()[name = tensor<string, []>("op_50886_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_50886_end_mask_0 = const()[name = tensor<string, []>("op_50886_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50886_cast = slice_by_index(begin = var_50886_begin_0, end = var_50886_end_0, end_mask = var_50886_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50886_cast")];
            tensor<int32, [4]> var_50890_begin_0 = const()[name = tensor<string, []>("op_50890_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_50890_end_0 = const()[name = tensor<string, []>("op_50890_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_50890_end_mask_0 = const()[name = tensor<string, []>("op_50890_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50890_cast = slice_by_index(begin = var_50890_begin_0, end = var_50890_end_0, end_mask = var_50890_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50890_cast")];
            tensor<int32, [4]> var_50894_begin_0 = const()[name = tensor<string, []>("op_50894_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_50894_end_0 = const()[name = tensor<string, []>("op_50894_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_50894_end_mask_0 = const()[name = tensor<string, []>("op_50894_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50894_cast = slice_by_index(begin = var_50894_begin_0, end = var_50894_end_0, end_mask = var_50894_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50894_cast")];
            tensor<int32, [4]> var_50898_begin_0 = const()[name = tensor<string, []>("op_50898_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_50898_end_0 = const()[name = tensor<string, []>("op_50898_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_50898_end_mask_0 = const()[name = tensor<string, []>("op_50898_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50898_cast = slice_by_index(begin = var_50898_begin_0, end = var_50898_end_0, end_mask = var_50898_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50898_cast")];
            tensor<int32, [4]> var_50902_begin_0 = const()[name = tensor<string, []>("op_50902_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_50902_end_0 = const()[name = tensor<string, []>("op_50902_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_50902_end_mask_0 = const()[name = tensor<string, []>("op_50902_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50902_cast = slice_by_index(begin = var_50902_begin_0, end = var_50902_end_0, end_mask = var_50902_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50902_cast")];
            tensor<int32, [4]> var_50906_begin_0 = const()[name = tensor<string, []>("op_50906_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_50906_end_0 = const()[name = tensor<string, []>("op_50906_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_50906_end_mask_0 = const()[name = tensor<string, []>("op_50906_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50906_cast = slice_by_index(begin = var_50906_begin_0, end = var_50906_end_0, end_mask = var_50906_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50906_cast")];
            tensor<int32, [4]> var_50910_begin_0 = const()[name = tensor<string, []>("op_50910_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_50910_end_0 = const()[name = tensor<string, []>("op_50910_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_50910_end_mask_0 = const()[name = tensor<string, []>("op_50910_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50910_cast = slice_by_index(begin = var_50910_begin_0, end = var_50910_end_0, end_mask = var_50910_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50910_cast")];
            tensor<int32, [4]> var_50914_begin_0 = const()[name = tensor<string, []>("op_50914_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_50914_end_0 = const()[name = tensor<string, []>("op_50914_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_50914_end_mask_0 = const()[name = tensor<string, []>("op_50914_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50914_cast = slice_by_index(begin = var_50914_begin_0, end = var_50914_end_0, end_mask = var_50914_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50914_cast")];
            tensor<int32, [4]> var_50918_begin_0 = const()[name = tensor<string, []>("op_50918_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_50918_end_0 = const()[name = tensor<string, []>("op_50918_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_50918_end_mask_0 = const()[name = tensor<string, []>("op_50918_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50918_cast = slice_by_index(begin = var_50918_begin_0, end = var_50918_end_0, end_mask = var_50918_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50918_cast")];
            tensor<int32, [4]> var_50922_begin_0 = const()[name = tensor<string, []>("op_50922_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_50922_end_0 = const()[name = tensor<string, []>("op_50922_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_50922_end_mask_0 = const()[name = tensor<string, []>("op_50922_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50922_cast = slice_by_index(begin = var_50922_begin_0, end = var_50922_end_0, end_mask = var_50922_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50922_cast")];
            tensor<int32, [4]> var_50926_begin_0 = const()[name = tensor<string, []>("op_50926_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_50926_end_0 = const()[name = tensor<string, []>("op_50926_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_50926_end_mask_0 = const()[name = tensor<string, []>("op_50926_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50926_cast = slice_by_index(begin = var_50926_begin_0, end = var_50926_end_0, end_mask = var_50926_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50926_cast")];
            tensor<int32, [4]> var_50930_begin_0 = const()[name = tensor<string, []>("op_50930_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_50930_end_0 = const()[name = tensor<string, []>("op_50930_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_50930_end_mask_0 = const()[name = tensor<string, []>("op_50930_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50930_cast = slice_by_index(begin = var_50930_begin_0, end = var_50930_end_0, end_mask = var_50930_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50930_cast")];
            tensor<int32, [4]> var_50934_begin_0 = const()[name = tensor<string, []>("op_50934_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_50934_end_0 = const()[name = tensor<string, []>("op_50934_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_50934_end_mask_0 = const()[name = tensor<string, []>("op_50934_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50934_cast = slice_by_index(begin = var_50934_begin_0, end = var_50934_end_0, end_mask = var_50934_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50934_cast")];
            tensor<int32, [4]> var_50938_begin_0 = const()[name = tensor<string, []>("op_50938_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_50938_end_0 = const()[name = tensor<string, []>("op_50938_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_50938_end_mask_0 = const()[name = tensor<string, []>("op_50938_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_50938_cast = slice_by_index(begin = var_50938_begin_0, end = var_50938_end_0, end_mask = var_50938_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_50938_cast")];
            tensor<int32, [4]> var_50940_begin_0 = const()[name = tensor<string, []>("op_50940_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_50940_end_0 = const()[name = tensor<string, []>("op_50940_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_50940_end_mask_0 = const()[name = tensor<string, []>("op_50940_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50940_cast = slice_by_index(begin = var_50940_begin_0, end = var_50940_end_0, end_mask = var_50940_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_50940_cast")];
            tensor<int32, [4]> var_50944_begin_0 = const()[name = tensor<string, []>("op_50944_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_50944_end_0 = const()[name = tensor<string, []>("op_50944_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_50944_end_mask_0 = const()[name = tensor<string, []>("op_50944_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50944_cast = slice_by_index(begin = var_50944_begin_0, end = var_50944_end_0, end_mask = var_50944_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_50944_cast")];
            tensor<int32, [4]> var_50948_begin_0 = const()[name = tensor<string, []>("op_50948_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_50948_end_0 = const()[name = tensor<string, []>("op_50948_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_50948_end_mask_0 = const()[name = tensor<string, []>("op_50948_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50948_cast = slice_by_index(begin = var_50948_begin_0, end = var_50948_end_0, end_mask = var_50948_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_50948_cast")];
            tensor<int32, [4]> var_50952_begin_0 = const()[name = tensor<string, []>("op_50952_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_50952_end_0 = const()[name = tensor<string, []>("op_50952_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_50952_end_mask_0 = const()[name = tensor<string, []>("op_50952_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50952_cast = slice_by_index(begin = var_50952_begin_0, end = var_50952_end_0, end_mask = var_50952_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_50952_cast")];
            tensor<int32, [4]> var_50956_begin_0 = const()[name = tensor<string, []>("op_50956_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_50956_end_0 = const()[name = tensor<string, []>("op_50956_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_50956_end_mask_0 = const()[name = tensor<string, []>("op_50956_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50956_cast = slice_by_index(begin = var_50956_begin_0, end = var_50956_end_0, end_mask = var_50956_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_50956_cast")];
            tensor<int32, [4]> var_50960_begin_0 = const()[name = tensor<string, []>("op_50960_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_50960_end_0 = const()[name = tensor<string, []>("op_50960_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_50960_end_mask_0 = const()[name = tensor<string, []>("op_50960_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50960_cast = slice_by_index(begin = var_50960_begin_0, end = var_50960_end_0, end_mask = var_50960_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_50960_cast")];
            tensor<int32, [4]> var_50964_begin_0 = const()[name = tensor<string, []>("op_50964_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_50964_end_0 = const()[name = tensor<string, []>("op_50964_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_50964_end_mask_0 = const()[name = tensor<string, []>("op_50964_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50964_cast = slice_by_index(begin = var_50964_begin_0, end = var_50964_end_0, end_mask = var_50964_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_50964_cast")];
            tensor<int32, [4]> var_50968_begin_0 = const()[name = tensor<string, []>("op_50968_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_50968_end_0 = const()[name = tensor<string, []>("op_50968_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_50968_end_mask_0 = const()[name = tensor<string, []>("op_50968_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50968_cast = slice_by_index(begin = var_50968_begin_0, end = var_50968_end_0, end_mask = var_50968_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_50968_cast")];
            tensor<int32, [4]> var_50972_begin_0 = const()[name = tensor<string, []>("op_50972_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_50972_end_0 = const()[name = tensor<string, []>("op_50972_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_50972_end_mask_0 = const()[name = tensor<string, []>("op_50972_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50972_cast = slice_by_index(begin = var_50972_begin_0, end = var_50972_end_0, end_mask = var_50972_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_50972_cast")];
            tensor<int32, [4]> var_50976_begin_0 = const()[name = tensor<string, []>("op_50976_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_50976_end_0 = const()[name = tensor<string, []>("op_50976_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_50976_end_mask_0 = const()[name = tensor<string, []>("op_50976_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50976_cast = slice_by_index(begin = var_50976_begin_0, end = var_50976_end_0, end_mask = var_50976_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_50976_cast")];
            tensor<int32, [4]> var_50980_begin_0 = const()[name = tensor<string, []>("op_50980_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_50980_end_0 = const()[name = tensor<string, []>("op_50980_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_50980_end_mask_0 = const()[name = tensor<string, []>("op_50980_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50980_cast = slice_by_index(begin = var_50980_begin_0, end = var_50980_end_0, end_mask = var_50980_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_50980_cast")];
            tensor<int32, [4]> var_50984_begin_0 = const()[name = tensor<string, []>("op_50984_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_50984_end_0 = const()[name = tensor<string, []>("op_50984_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_50984_end_mask_0 = const()[name = tensor<string, []>("op_50984_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50984_cast = slice_by_index(begin = var_50984_begin_0, end = var_50984_end_0, end_mask = var_50984_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_50984_cast")];
            tensor<int32, [4]> var_50988_begin_0 = const()[name = tensor<string, []>("op_50988_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_50988_end_0 = const()[name = tensor<string, []>("op_50988_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_50988_end_mask_0 = const()[name = tensor<string, []>("op_50988_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50988_cast = slice_by_index(begin = var_50988_begin_0, end = var_50988_end_0, end_mask = var_50988_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_50988_cast")];
            tensor<int32, [4]> var_50992_begin_0 = const()[name = tensor<string, []>("op_50992_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_50992_end_0 = const()[name = tensor<string, []>("op_50992_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_50992_end_mask_0 = const()[name = tensor<string, []>("op_50992_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50992_cast = slice_by_index(begin = var_50992_begin_0, end = var_50992_end_0, end_mask = var_50992_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_50992_cast")];
            tensor<int32, [4]> var_50996_begin_0 = const()[name = tensor<string, []>("op_50996_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_50996_end_0 = const()[name = tensor<string, []>("op_50996_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_50996_end_mask_0 = const()[name = tensor<string, []>("op_50996_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_50996_cast = slice_by_index(begin = var_50996_begin_0, end = var_50996_end_0, end_mask = var_50996_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_50996_cast")];
            tensor<int32, [4]> var_51000_begin_0 = const()[name = tensor<string, []>("op_51000_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_51000_end_0 = const()[name = tensor<string, []>("op_51000_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_51000_end_mask_0 = const()[name = tensor<string, []>("op_51000_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51000_cast = slice_by_index(begin = var_51000_begin_0, end = var_51000_end_0, end_mask = var_51000_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_51000_cast")];
            tensor<int32, [4]> var_51004_begin_0 = const()[name = tensor<string, []>("op_51004_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_51004_end_0 = const()[name = tensor<string, []>("op_51004_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_51004_end_mask_0 = const()[name = tensor<string, []>("op_51004_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51004_cast = slice_by_index(begin = var_51004_begin_0, end = var_51004_end_0, end_mask = var_51004_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_51004_cast")];
            tensor<int32, [4]> var_51008_begin_0 = const()[name = tensor<string, []>("op_51008_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_51008_end_0 = const()[name = tensor<string, []>("op_51008_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_51008_end_mask_0 = const()[name = tensor<string, []>("op_51008_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51008_cast = slice_by_index(begin = var_51008_begin_0, end = var_51008_end_0, end_mask = var_51008_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_51008_cast")];
            tensor<int32, [4]> var_51012_begin_0 = const()[name = tensor<string, []>("op_51012_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_51012_end_0 = const()[name = tensor<string, []>("op_51012_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_51012_end_mask_0 = const()[name = tensor<string, []>("op_51012_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51012_cast = slice_by_index(begin = var_51012_begin_0, end = var_51012_end_0, end_mask = var_51012_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_51012_cast")];
            tensor<int32, [4]> var_51016_begin_0 = const()[name = tensor<string, []>("op_51016_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_51016_end_0 = const()[name = tensor<string, []>("op_51016_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_51016_end_mask_0 = const()[name = tensor<string, []>("op_51016_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51016_cast = slice_by_index(begin = var_51016_begin_0, end = var_51016_end_0, end_mask = var_51016_end_mask_0, x = v_225_cast)[name = tensor<string, []>("op_51016_cast")];
            tensor<string, []> var_51020_equation_0 = const()[name = tensor<string, []>("op_51020_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51020_cast = einsum(equation = var_51020_equation_0, values = (var_50862_cast, var_50779_cast))[name = tensor<string, []>("op_51020_cast")];
            tensor<fp16, []> var_51021_to_fp16 = const()[name = tensor<string, []>("op_51021_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4321_cast = mul(x = var_51020_cast, y = var_51021_to_fp16)[name = tensor<string, []>("aw_4321_cast")];
            tensor<string, []> var_51024_equation_0 = const()[name = tensor<string, []>("op_51024_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51024_cast = einsum(equation = var_51024_equation_0, values = (var_50866_cast, var_50783_cast))[name = tensor<string, []>("op_51024_cast")];
            tensor<fp16, []> var_51025_to_fp16 = const()[name = tensor<string, []>("op_51025_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4323_cast = mul(x = var_51024_cast, y = var_51025_to_fp16)[name = tensor<string, []>("aw_4323_cast")];
            tensor<string, []> var_51028_equation_0 = const()[name = tensor<string, []>("op_51028_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51028_cast = einsum(equation = var_51028_equation_0, values = (var_50870_cast, var_50787_cast))[name = tensor<string, []>("op_51028_cast")];
            tensor<fp16, []> var_51029_to_fp16 = const()[name = tensor<string, []>("op_51029_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4325_cast = mul(x = var_51028_cast, y = var_51029_to_fp16)[name = tensor<string, []>("aw_4325_cast")];
            tensor<string, []> var_51032_equation_0 = const()[name = tensor<string, []>("op_51032_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51032_cast = einsum(equation = var_51032_equation_0, values = (var_50874_cast, var_50791_cast))[name = tensor<string, []>("op_51032_cast")];
            tensor<fp16, []> var_51033_to_fp16 = const()[name = tensor<string, []>("op_51033_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4327_cast = mul(x = var_51032_cast, y = var_51033_to_fp16)[name = tensor<string, []>("aw_4327_cast")];
            tensor<string, []> var_51036_equation_0 = const()[name = tensor<string, []>("op_51036_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51036_cast = einsum(equation = var_51036_equation_0, values = (var_50878_cast, var_50795_cast))[name = tensor<string, []>("op_51036_cast")];
            tensor<fp16, []> var_51037_to_fp16 = const()[name = tensor<string, []>("op_51037_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4329_cast = mul(x = var_51036_cast, y = var_51037_to_fp16)[name = tensor<string, []>("aw_4329_cast")];
            tensor<string, []> var_51040_equation_0 = const()[name = tensor<string, []>("op_51040_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51040_cast = einsum(equation = var_51040_equation_0, values = (var_50882_cast, var_50799_cast))[name = tensor<string, []>("op_51040_cast")];
            tensor<fp16, []> var_51041_to_fp16 = const()[name = tensor<string, []>("op_51041_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4331_cast = mul(x = var_51040_cast, y = var_51041_to_fp16)[name = tensor<string, []>("aw_4331_cast")];
            tensor<string, []> var_51044_equation_0 = const()[name = tensor<string, []>("op_51044_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51044_cast = einsum(equation = var_51044_equation_0, values = (var_50886_cast, var_50803_cast))[name = tensor<string, []>("op_51044_cast")];
            tensor<fp16, []> var_51045_to_fp16 = const()[name = tensor<string, []>("op_51045_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4333_cast = mul(x = var_51044_cast, y = var_51045_to_fp16)[name = tensor<string, []>("aw_4333_cast")];
            tensor<string, []> var_51048_equation_0 = const()[name = tensor<string, []>("op_51048_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51048_cast = einsum(equation = var_51048_equation_0, values = (var_50890_cast, var_50807_cast))[name = tensor<string, []>("op_51048_cast")];
            tensor<fp16, []> var_51049_to_fp16 = const()[name = tensor<string, []>("op_51049_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4335_cast = mul(x = var_51048_cast, y = var_51049_to_fp16)[name = tensor<string, []>("aw_4335_cast")];
            tensor<string, []> var_51052_equation_0 = const()[name = tensor<string, []>("op_51052_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51052_cast = einsum(equation = var_51052_equation_0, values = (var_50894_cast, var_50811_cast))[name = tensor<string, []>("op_51052_cast")];
            tensor<fp16, []> var_51053_to_fp16 = const()[name = tensor<string, []>("op_51053_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4337_cast = mul(x = var_51052_cast, y = var_51053_to_fp16)[name = tensor<string, []>("aw_4337_cast")];
            tensor<string, []> var_51056_equation_0 = const()[name = tensor<string, []>("op_51056_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51056_cast = einsum(equation = var_51056_equation_0, values = (var_50898_cast, var_50815_cast))[name = tensor<string, []>("op_51056_cast")];
            tensor<fp16, []> var_51057_to_fp16 = const()[name = tensor<string, []>("op_51057_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4339_cast = mul(x = var_51056_cast, y = var_51057_to_fp16)[name = tensor<string, []>("aw_4339_cast")];
            tensor<string, []> var_51060_equation_0 = const()[name = tensor<string, []>("op_51060_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51060_cast = einsum(equation = var_51060_equation_0, values = (var_50902_cast, var_50819_cast))[name = tensor<string, []>("op_51060_cast")];
            tensor<fp16, []> var_51061_to_fp16 = const()[name = tensor<string, []>("op_51061_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4341_cast = mul(x = var_51060_cast, y = var_51061_to_fp16)[name = tensor<string, []>("aw_4341_cast")];
            tensor<string, []> var_51064_equation_0 = const()[name = tensor<string, []>("op_51064_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51064_cast = einsum(equation = var_51064_equation_0, values = (var_50906_cast, var_50823_cast))[name = tensor<string, []>("op_51064_cast")];
            tensor<fp16, []> var_51065_to_fp16 = const()[name = tensor<string, []>("op_51065_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4343_cast = mul(x = var_51064_cast, y = var_51065_to_fp16)[name = tensor<string, []>("aw_4343_cast")];
            tensor<string, []> var_51068_equation_0 = const()[name = tensor<string, []>("op_51068_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51068_cast = einsum(equation = var_51068_equation_0, values = (var_50910_cast, var_50827_cast))[name = tensor<string, []>("op_51068_cast")];
            tensor<fp16, []> var_51069_to_fp16 = const()[name = tensor<string, []>("op_51069_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4345_cast = mul(x = var_51068_cast, y = var_51069_to_fp16)[name = tensor<string, []>("aw_4345_cast")];
            tensor<string, []> var_51072_equation_0 = const()[name = tensor<string, []>("op_51072_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51072_cast = einsum(equation = var_51072_equation_0, values = (var_50914_cast, var_50831_cast))[name = tensor<string, []>("op_51072_cast")];
            tensor<fp16, []> var_51073_to_fp16 = const()[name = tensor<string, []>("op_51073_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4347_cast = mul(x = var_51072_cast, y = var_51073_to_fp16)[name = tensor<string, []>("aw_4347_cast")];
            tensor<string, []> var_51076_equation_0 = const()[name = tensor<string, []>("op_51076_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51076_cast = einsum(equation = var_51076_equation_0, values = (var_50918_cast, var_50835_cast))[name = tensor<string, []>("op_51076_cast")];
            tensor<fp16, []> var_51077_to_fp16 = const()[name = tensor<string, []>("op_51077_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4349_cast = mul(x = var_51076_cast, y = var_51077_to_fp16)[name = tensor<string, []>("aw_4349_cast")];
            tensor<string, []> var_51080_equation_0 = const()[name = tensor<string, []>("op_51080_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51080_cast = einsum(equation = var_51080_equation_0, values = (var_50922_cast, var_50839_cast))[name = tensor<string, []>("op_51080_cast")];
            tensor<fp16, []> var_51081_to_fp16 = const()[name = tensor<string, []>("op_51081_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4351_cast = mul(x = var_51080_cast, y = var_51081_to_fp16)[name = tensor<string, []>("aw_4351_cast")];
            tensor<string, []> var_51084_equation_0 = const()[name = tensor<string, []>("op_51084_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51084_cast = einsum(equation = var_51084_equation_0, values = (var_50926_cast, var_50843_cast))[name = tensor<string, []>("op_51084_cast")];
            tensor<fp16, []> var_51085_to_fp16 = const()[name = tensor<string, []>("op_51085_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4353_cast = mul(x = var_51084_cast, y = var_51085_to_fp16)[name = tensor<string, []>("aw_4353_cast")];
            tensor<string, []> var_51088_equation_0 = const()[name = tensor<string, []>("op_51088_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51088_cast = einsum(equation = var_51088_equation_0, values = (var_50930_cast, var_50847_cast))[name = tensor<string, []>("op_51088_cast")];
            tensor<fp16, []> var_51089_to_fp16 = const()[name = tensor<string, []>("op_51089_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4355_cast = mul(x = var_51088_cast, y = var_51089_to_fp16)[name = tensor<string, []>("aw_4355_cast")];
            tensor<string, []> var_51092_equation_0 = const()[name = tensor<string, []>("op_51092_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51092_cast = einsum(equation = var_51092_equation_0, values = (var_50934_cast, var_50851_cast))[name = tensor<string, []>("op_51092_cast")];
            tensor<fp16, []> var_51093_to_fp16 = const()[name = tensor<string, []>("op_51093_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4357_cast = mul(x = var_51092_cast, y = var_51093_to_fp16)[name = tensor<string, []>("aw_4357_cast")];
            tensor<string, []> var_51096_equation_0 = const()[name = tensor<string, []>("op_51096_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51096_cast = einsum(equation = var_51096_equation_0, values = (var_50938_cast, var_50855_cast))[name = tensor<string, []>("op_51096_cast")];
            tensor<fp16, []> var_51097_to_fp16 = const()[name = tensor<string, []>("op_51097_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4359_cast = mul(x = var_51096_cast, y = var_51097_to_fp16)[name = tensor<string, []>("aw_4359_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51099_cast = softmax(axis = var_30385, x = aw_4321_cast)[name = tensor<string, []>("op_51099_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51100_cast = softmax(axis = var_30385, x = aw_4323_cast)[name = tensor<string, []>("op_51100_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51101_cast = softmax(axis = var_30385, x = aw_4325_cast)[name = tensor<string, []>("op_51101_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51102_cast = softmax(axis = var_30385, x = aw_4327_cast)[name = tensor<string, []>("op_51102_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51103_cast = softmax(axis = var_30385, x = aw_4329_cast)[name = tensor<string, []>("op_51103_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51104_cast = softmax(axis = var_30385, x = aw_4331_cast)[name = tensor<string, []>("op_51104_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51105_cast = softmax(axis = var_30385, x = aw_4333_cast)[name = tensor<string, []>("op_51105_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51106_cast = softmax(axis = var_30385, x = aw_4335_cast)[name = tensor<string, []>("op_51106_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51107_cast = softmax(axis = var_30385, x = aw_4337_cast)[name = tensor<string, []>("op_51107_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51108_cast = softmax(axis = var_30385, x = aw_4339_cast)[name = tensor<string, []>("op_51108_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51109_cast = softmax(axis = var_30385, x = aw_4341_cast)[name = tensor<string, []>("op_51109_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51110_cast = softmax(axis = var_30385, x = aw_4343_cast)[name = tensor<string, []>("op_51110_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51111_cast = softmax(axis = var_30385, x = aw_4345_cast)[name = tensor<string, []>("op_51111_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51112_cast = softmax(axis = var_30385, x = aw_4347_cast)[name = tensor<string, []>("op_51112_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51113_cast = softmax(axis = var_30385, x = aw_4349_cast)[name = tensor<string, []>("op_51113_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51114_cast = softmax(axis = var_30385, x = aw_4351_cast)[name = tensor<string, []>("op_51114_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51115_cast = softmax(axis = var_30385, x = aw_4353_cast)[name = tensor<string, []>("op_51115_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51116_cast = softmax(axis = var_30385, x = aw_4355_cast)[name = tensor<string, []>("op_51116_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51117_cast = softmax(axis = var_30385, x = aw_4357_cast)[name = tensor<string, []>("op_51117_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51118_cast = softmax(axis = var_30385, x = aw_4359_cast)[name = tensor<string, []>("op_51118_cast")];
            tensor<string, []> var_51120_equation_0 = const()[name = tensor<string, []>("op_51120_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51120_cast = einsum(equation = var_51120_equation_0, values = (var_50940_cast, var_51099_cast))[name = tensor<string, []>("op_51120_cast")];
            tensor<string, []> var_51122_equation_0 = const()[name = tensor<string, []>("op_51122_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51122_cast = einsum(equation = var_51122_equation_0, values = (var_50944_cast, var_51100_cast))[name = tensor<string, []>("op_51122_cast")];
            tensor<string, []> var_51124_equation_0 = const()[name = tensor<string, []>("op_51124_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51124_cast = einsum(equation = var_51124_equation_0, values = (var_50948_cast, var_51101_cast))[name = tensor<string, []>("op_51124_cast")];
            tensor<string, []> var_51126_equation_0 = const()[name = tensor<string, []>("op_51126_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51126_cast = einsum(equation = var_51126_equation_0, values = (var_50952_cast, var_51102_cast))[name = tensor<string, []>("op_51126_cast")];
            tensor<string, []> var_51128_equation_0 = const()[name = tensor<string, []>("op_51128_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51128_cast = einsum(equation = var_51128_equation_0, values = (var_50956_cast, var_51103_cast))[name = tensor<string, []>("op_51128_cast")];
            tensor<string, []> var_51130_equation_0 = const()[name = tensor<string, []>("op_51130_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51130_cast = einsum(equation = var_51130_equation_0, values = (var_50960_cast, var_51104_cast))[name = tensor<string, []>("op_51130_cast")];
            tensor<string, []> var_51132_equation_0 = const()[name = tensor<string, []>("op_51132_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51132_cast = einsum(equation = var_51132_equation_0, values = (var_50964_cast, var_51105_cast))[name = tensor<string, []>("op_51132_cast")];
            tensor<string, []> var_51134_equation_0 = const()[name = tensor<string, []>("op_51134_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51134_cast = einsum(equation = var_51134_equation_0, values = (var_50968_cast, var_51106_cast))[name = tensor<string, []>("op_51134_cast")];
            tensor<string, []> var_51136_equation_0 = const()[name = tensor<string, []>("op_51136_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51136_cast = einsum(equation = var_51136_equation_0, values = (var_50972_cast, var_51107_cast))[name = tensor<string, []>("op_51136_cast")];
            tensor<string, []> var_51138_equation_0 = const()[name = tensor<string, []>("op_51138_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51138_cast = einsum(equation = var_51138_equation_0, values = (var_50976_cast, var_51108_cast))[name = tensor<string, []>("op_51138_cast")];
            tensor<string, []> var_51140_equation_0 = const()[name = tensor<string, []>("op_51140_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51140_cast = einsum(equation = var_51140_equation_0, values = (var_50980_cast, var_51109_cast))[name = tensor<string, []>("op_51140_cast")];
            tensor<string, []> var_51142_equation_0 = const()[name = tensor<string, []>("op_51142_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51142_cast = einsum(equation = var_51142_equation_0, values = (var_50984_cast, var_51110_cast))[name = tensor<string, []>("op_51142_cast")];
            tensor<string, []> var_51144_equation_0 = const()[name = tensor<string, []>("op_51144_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51144_cast = einsum(equation = var_51144_equation_0, values = (var_50988_cast, var_51111_cast))[name = tensor<string, []>("op_51144_cast")];
            tensor<string, []> var_51146_equation_0 = const()[name = tensor<string, []>("op_51146_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51146_cast = einsum(equation = var_51146_equation_0, values = (var_50992_cast, var_51112_cast))[name = tensor<string, []>("op_51146_cast")];
            tensor<string, []> var_51148_equation_0 = const()[name = tensor<string, []>("op_51148_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51148_cast = einsum(equation = var_51148_equation_0, values = (var_50996_cast, var_51113_cast))[name = tensor<string, []>("op_51148_cast")];
            tensor<string, []> var_51150_equation_0 = const()[name = tensor<string, []>("op_51150_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51150_cast = einsum(equation = var_51150_equation_0, values = (var_51000_cast, var_51114_cast))[name = tensor<string, []>("op_51150_cast")];
            tensor<string, []> var_51152_equation_0 = const()[name = tensor<string, []>("op_51152_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51152_cast = einsum(equation = var_51152_equation_0, values = (var_51004_cast, var_51115_cast))[name = tensor<string, []>("op_51152_cast")];
            tensor<string, []> var_51154_equation_0 = const()[name = tensor<string, []>("op_51154_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51154_cast = einsum(equation = var_51154_equation_0, values = (var_51008_cast, var_51116_cast))[name = tensor<string, []>("op_51154_cast")];
            tensor<string, []> var_51156_equation_0 = const()[name = tensor<string, []>("op_51156_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51156_cast = einsum(equation = var_51156_equation_0, values = (var_51012_cast, var_51117_cast))[name = tensor<string, []>("op_51156_cast")];
            tensor<string, []> var_51158_equation_0 = const()[name = tensor<string, []>("op_51158_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51158_cast = einsum(equation = var_51158_equation_0, values = (var_51016_cast, var_51118_cast))[name = tensor<string, []>("op_51158_cast")];
            tensor<bool, []> input_653_interleave_0 = const()[name = tensor<string, []>("input_653_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_653_cast = concat(axis = var_30385, interleave = input_653_interleave_0, values = (var_51120_cast, var_51122_cast, var_51124_cast, var_51126_cast, var_51128_cast, var_51130_cast, var_51132_cast, var_51134_cast, var_51136_cast, var_51138_cast, var_51140_cast, var_51142_cast, var_51144_cast, var_51146_cast, var_51148_cast, var_51150_cast, var_51152_cast, var_51154_cast, var_51156_cast, var_51158_cast))[name = tensor<string, []>("input_653_cast")];
            tensor<int32, [2]> var_51164 = const()[name = tensor<string, []>("op_51164"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_51166 = const()[name = tensor<string, []>("op_51166"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_51168_pad_type_0 = const()[name = tensor<string, []>("op_51168_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_51168_pad_0 = const()[name = tensor<string, []>("op_51168_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4320952768)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4324229632)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_51168_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_out_0_bias_to_fp16, dilations = var_51166, groups = var_30385, pad = var_51168_pad_0, pad_type = var_51168_pad_type_0, strides = var_51164, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn1_to_out_0_weight_to_fp16, x = input_653_cast)[name = tensor<string, []>("op_51168_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_339_cast = add(x = var_51168_cast, y = inputs_337_cast)[name = tensor<string, []>("inputs_339_cast")];
            tensor<int32, [1]> var_51172 = const()[name = tensor<string, []>("op_51172"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_339_cast = reduce_mean(axes = var_51172, keep_dims = var_30380, x = inputs_339_cast)[name = tensor<string, []>("channels_mean_339_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_339_cast = sub(x = inputs_339_cast, y = channels_mean_339_cast)[name = tensor<string, []>("zero_mean_339_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_339_cast = mul(x = zero_mean_339_cast, y = zero_mean_339_cast)[name = tensor<string, []>("zero_mean_sq_339_cast")];
            tensor<int32, [1]> var_51176 = const()[name = tensor<string, []>("op_51176"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_51177_cast = reduce_mean(axes = var_51176, keep_dims = var_30380, x = zero_mean_sq_339_cast)[name = tensor<string, []>("op_51177_cast")];
            tensor<fp16, []> var_51178_to_fp16 = const()[name = tensor<string, []>("op_51178_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_51179_cast = add(x = var_51177_cast, y = var_51178_to_fp16)[name = tensor<string, []>("op_51179_cast")];
            tensor<fp16, []> denom_339_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_339_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_339_cast = rsqrt(epsilon = denom_339_epsilon_0_to_fp16, x = var_51179_cast)[name = tensor<string, []>("denom_339_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_339_cast = mul(x = zero_mean_339_cast, y = denom_339_cast)[name = tensor<string, []>("out_339_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_51183_to_fp16 = const()[name = tensor<string, []>("op_51183_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4324232256)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_51184_cast = add(x = out_339_cast, y = var_51183_to_fp16)[name = tensor<string, []>("op_51184_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_51186_to_fp16 = const()[name = tensor<string, []>("op_51186_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4324234880)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_449_cast = mul(x = var_51184_cast, y = var_51186_to_fp16)[name = tensor<string, []>("hidden_states_449_cast")];
            tensor<int32, [2]> var_51193 = const()[name = tensor<string, []>("op_51193"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_51195 = const()[name = tensor<string, []>("op_51195"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_227_pad_type_0 = const()[name = tensor<string, []>("q_227_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_227_pad_0 = const()[name = tensor<string, []>("q_227_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4324237504)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_227_cast = conv(dilations = var_51195, groups = var_30385, pad = q_227_pad_0, pad_type = q_227_pad_type_0, strides = var_51193, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_q_weight_to_fp16, x = hidden_states_449_cast)[name = tensor<string, []>("q_227_cast")];
            tensor<int32, [2]> var_51199 = const()[name = tensor<string, []>("op_51199"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_51201 = const()[name = tensor<string, []>("op_51201"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_453_pad_type_0 = const()[name = tensor<string, []>("k_453_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_453_pad_0 = const()[name = tensor<string, []>("k_453_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4327514368)))];
            tensor<fp16, [2, 1280, 1, 77]> k_453_cast = conv(dilations = var_51201, groups = var_30385, pad = k_453_pad_0, pad_type = k_453_pad_type_0, strides = var_51199, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_453_cast")];
            tensor<int32, [2]> var_51205 = const()[name = tensor<string, []>("op_51205"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_51207 = const()[name = tensor<string, []>("op_51207"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_227_pad_type_0 = const()[name = tensor<string, []>("v_227_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_227_pad_0 = const()[name = tensor<string, []>("v_227_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4332757312)))];
            tensor<fp16, [2, 1280, 1, 77]> v_227_cast = conv(dilations = var_51207, groups = var_30385, pad = v_227_pad_0, pad_type = v_227_pad_type_0, strides = var_51205, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_227_cast")];
            tensor<int32, [4]> var_51211_begin_0 = const()[name = tensor<string, []>("op_51211_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_51211_end_0 = const()[name = tensor<string, []>("op_51211_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_51211_end_mask_0 = const()[name = tensor<string, []>("op_51211_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51211_cast = slice_by_index(begin = var_51211_begin_0, end = var_51211_end_0, end_mask = var_51211_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51211_cast")];
            tensor<int32, [4]> var_51215_begin_0 = const()[name = tensor<string, []>("op_51215_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_51215_end_0 = const()[name = tensor<string, []>("op_51215_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_51215_end_mask_0 = const()[name = tensor<string, []>("op_51215_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51215_cast = slice_by_index(begin = var_51215_begin_0, end = var_51215_end_0, end_mask = var_51215_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51215_cast")];
            tensor<int32, [4]> var_51219_begin_0 = const()[name = tensor<string, []>("op_51219_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_51219_end_0 = const()[name = tensor<string, []>("op_51219_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_51219_end_mask_0 = const()[name = tensor<string, []>("op_51219_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51219_cast = slice_by_index(begin = var_51219_begin_0, end = var_51219_end_0, end_mask = var_51219_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51219_cast")];
            tensor<int32, [4]> var_51223_begin_0 = const()[name = tensor<string, []>("op_51223_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_51223_end_0 = const()[name = tensor<string, []>("op_51223_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_51223_end_mask_0 = const()[name = tensor<string, []>("op_51223_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51223_cast = slice_by_index(begin = var_51223_begin_0, end = var_51223_end_0, end_mask = var_51223_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51223_cast")];
            tensor<int32, [4]> var_51227_begin_0 = const()[name = tensor<string, []>("op_51227_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_51227_end_0 = const()[name = tensor<string, []>("op_51227_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_51227_end_mask_0 = const()[name = tensor<string, []>("op_51227_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51227_cast = slice_by_index(begin = var_51227_begin_0, end = var_51227_end_0, end_mask = var_51227_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51227_cast")];
            tensor<int32, [4]> var_51231_begin_0 = const()[name = tensor<string, []>("op_51231_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_51231_end_0 = const()[name = tensor<string, []>("op_51231_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_51231_end_mask_0 = const()[name = tensor<string, []>("op_51231_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51231_cast = slice_by_index(begin = var_51231_begin_0, end = var_51231_end_0, end_mask = var_51231_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51231_cast")];
            tensor<int32, [4]> var_51235_begin_0 = const()[name = tensor<string, []>("op_51235_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_51235_end_0 = const()[name = tensor<string, []>("op_51235_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_51235_end_mask_0 = const()[name = tensor<string, []>("op_51235_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51235_cast = slice_by_index(begin = var_51235_begin_0, end = var_51235_end_0, end_mask = var_51235_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51235_cast")];
            tensor<int32, [4]> var_51239_begin_0 = const()[name = tensor<string, []>("op_51239_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_51239_end_0 = const()[name = tensor<string, []>("op_51239_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_51239_end_mask_0 = const()[name = tensor<string, []>("op_51239_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51239_cast = slice_by_index(begin = var_51239_begin_0, end = var_51239_end_0, end_mask = var_51239_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51239_cast")];
            tensor<int32, [4]> var_51243_begin_0 = const()[name = tensor<string, []>("op_51243_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_51243_end_0 = const()[name = tensor<string, []>("op_51243_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_51243_end_mask_0 = const()[name = tensor<string, []>("op_51243_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51243_cast = slice_by_index(begin = var_51243_begin_0, end = var_51243_end_0, end_mask = var_51243_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51243_cast")];
            tensor<int32, [4]> var_51247_begin_0 = const()[name = tensor<string, []>("op_51247_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_51247_end_0 = const()[name = tensor<string, []>("op_51247_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_51247_end_mask_0 = const()[name = tensor<string, []>("op_51247_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51247_cast = slice_by_index(begin = var_51247_begin_0, end = var_51247_end_0, end_mask = var_51247_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51247_cast")];
            tensor<int32, [4]> var_51251_begin_0 = const()[name = tensor<string, []>("op_51251_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_51251_end_0 = const()[name = tensor<string, []>("op_51251_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_51251_end_mask_0 = const()[name = tensor<string, []>("op_51251_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51251_cast = slice_by_index(begin = var_51251_begin_0, end = var_51251_end_0, end_mask = var_51251_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51251_cast")];
            tensor<int32, [4]> var_51255_begin_0 = const()[name = tensor<string, []>("op_51255_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_51255_end_0 = const()[name = tensor<string, []>("op_51255_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_51255_end_mask_0 = const()[name = tensor<string, []>("op_51255_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51255_cast = slice_by_index(begin = var_51255_begin_0, end = var_51255_end_0, end_mask = var_51255_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51255_cast")];
            tensor<int32, [4]> var_51259_begin_0 = const()[name = tensor<string, []>("op_51259_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_51259_end_0 = const()[name = tensor<string, []>("op_51259_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_51259_end_mask_0 = const()[name = tensor<string, []>("op_51259_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51259_cast = slice_by_index(begin = var_51259_begin_0, end = var_51259_end_0, end_mask = var_51259_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51259_cast")];
            tensor<int32, [4]> var_51263_begin_0 = const()[name = tensor<string, []>("op_51263_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_51263_end_0 = const()[name = tensor<string, []>("op_51263_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_51263_end_mask_0 = const()[name = tensor<string, []>("op_51263_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51263_cast = slice_by_index(begin = var_51263_begin_0, end = var_51263_end_0, end_mask = var_51263_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51263_cast")];
            tensor<int32, [4]> var_51267_begin_0 = const()[name = tensor<string, []>("op_51267_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_51267_end_0 = const()[name = tensor<string, []>("op_51267_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_51267_end_mask_0 = const()[name = tensor<string, []>("op_51267_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51267_cast = slice_by_index(begin = var_51267_begin_0, end = var_51267_end_0, end_mask = var_51267_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51267_cast")];
            tensor<int32, [4]> var_51271_begin_0 = const()[name = tensor<string, []>("op_51271_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_51271_end_0 = const()[name = tensor<string, []>("op_51271_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_51271_end_mask_0 = const()[name = tensor<string, []>("op_51271_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51271_cast = slice_by_index(begin = var_51271_begin_0, end = var_51271_end_0, end_mask = var_51271_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51271_cast")];
            tensor<int32, [4]> var_51275_begin_0 = const()[name = tensor<string, []>("op_51275_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_51275_end_0 = const()[name = tensor<string, []>("op_51275_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_51275_end_mask_0 = const()[name = tensor<string, []>("op_51275_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51275_cast = slice_by_index(begin = var_51275_begin_0, end = var_51275_end_0, end_mask = var_51275_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51275_cast")];
            tensor<int32, [4]> var_51279_begin_0 = const()[name = tensor<string, []>("op_51279_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_51279_end_0 = const()[name = tensor<string, []>("op_51279_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_51279_end_mask_0 = const()[name = tensor<string, []>("op_51279_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51279_cast = slice_by_index(begin = var_51279_begin_0, end = var_51279_end_0, end_mask = var_51279_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51279_cast")];
            tensor<int32, [4]> var_51283_begin_0 = const()[name = tensor<string, []>("op_51283_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_51283_end_0 = const()[name = tensor<string, []>("op_51283_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_51283_end_mask_0 = const()[name = tensor<string, []>("op_51283_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51283_cast = slice_by_index(begin = var_51283_begin_0, end = var_51283_end_0, end_mask = var_51283_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51283_cast")];
            tensor<int32, [4]> var_51287_begin_0 = const()[name = tensor<string, []>("op_51287_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_51287_end_0 = const()[name = tensor<string, []>("op_51287_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_51287_end_mask_0 = const()[name = tensor<string, []>("op_51287_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51287_cast = slice_by_index(begin = var_51287_begin_0, end = var_51287_end_0, end_mask = var_51287_end_mask_0, x = q_227_cast)[name = tensor<string, []>("op_51287_cast")];
            tensor<int32, [4]> k_455_perm_0 = const()[name = tensor<string, []>("k_455_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_51294_begin_0 = const()[name = tensor<string, []>("op_51294_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_51294_end_0 = const()[name = tensor<string, []>("op_51294_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_51294_end_mask_0 = const()[name = tensor<string, []>("op_51294_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_26 = transpose(perm = k_455_perm_0, x = k_453_cast)[name = tensor<string, []>("transpose_26")];
            tensor<fp16, [2, 77, 1, 64]> var_51294_cast = slice_by_index(begin = var_51294_begin_0, end = var_51294_end_0, end_mask = var_51294_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51294_cast")];
            tensor<int32, [4]> var_51298_begin_0 = const()[name = tensor<string, []>("op_51298_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_51298_end_0 = const()[name = tensor<string, []>("op_51298_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_51298_end_mask_0 = const()[name = tensor<string, []>("op_51298_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51298_cast = slice_by_index(begin = var_51298_begin_0, end = var_51298_end_0, end_mask = var_51298_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51298_cast")];
            tensor<int32, [4]> var_51302_begin_0 = const()[name = tensor<string, []>("op_51302_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_51302_end_0 = const()[name = tensor<string, []>("op_51302_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_51302_end_mask_0 = const()[name = tensor<string, []>("op_51302_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51302_cast = slice_by_index(begin = var_51302_begin_0, end = var_51302_end_0, end_mask = var_51302_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51302_cast")];
            tensor<int32, [4]> var_51306_begin_0 = const()[name = tensor<string, []>("op_51306_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_51306_end_0 = const()[name = tensor<string, []>("op_51306_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_51306_end_mask_0 = const()[name = tensor<string, []>("op_51306_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51306_cast = slice_by_index(begin = var_51306_begin_0, end = var_51306_end_0, end_mask = var_51306_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51306_cast")];
            tensor<int32, [4]> var_51310_begin_0 = const()[name = tensor<string, []>("op_51310_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_51310_end_0 = const()[name = tensor<string, []>("op_51310_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_51310_end_mask_0 = const()[name = tensor<string, []>("op_51310_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51310_cast = slice_by_index(begin = var_51310_begin_0, end = var_51310_end_0, end_mask = var_51310_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51310_cast")];
            tensor<int32, [4]> var_51314_begin_0 = const()[name = tensor<string, []>("op_51314_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_51314_end_0 = const()[name = tensor<string, []>("op_51314_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_51314_end_mask_0 = const()[name = tensor<string, []>("op_51314_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51314_cast = slice_by_index(begin = var_51314_begin_0, end = var_51314_end_0, end_mask = var_51314_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51314_cast")];
            tensor<int32, [4]> var_51318_begin_0 = const()[name = tensor<string, []>("op_51318_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_51318_end_0 = const()[name = tensor<string, []>("op_51318_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_51318_end_mask_0 = const()[name = tensor<string, []>("op_51318_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51318_cast = slice_by_index(begin = var_51318_begin_0, end = var_51318_end_0, end_mask = var_51318_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51318_cast")];
            tensor<int32, [4]> var_51322_begin_0 = const()[name = tensor<string, []>("op_51322_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_51322_end_0 = const()[name = tensor<string, []>("op_51322_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_51322_end_mask_0 = const()[name = tensor<string, []>("op_51322_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51322_cast = slice_by_index(begin = var_51322_begin_0, end = var_51322_end_0, end_mask = var_51322_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51322_cast")];
            tensor<int32, [4]> var_51326_begin_0 = const()[name = tensor<string, []>("op_51326_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_51326_end_0 = const()[name = tensor<string, []>("op_51326_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_51326_end_mask_0 = const()[name = tensor<string, []>("op_51326_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51326_cast = slice_by_index(begin = var_51326_begin_0, end = var_51326_end_0, end_mask = var_51326_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51326_cast")];
            tensor<int32, [4]> var_51330_begin_0 = const()[name = tensor<string, []>("op_51330_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_51330_end_0 = const()[name = tensor<string, []>("op_51330_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_51330_end_mask_0 = const()[name = tensor<string, []>("op_51330_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51330_cast = slice_by_index(begin = var_51330_begin_0, end = var_51330_end_0, end_mask = var_51330_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51330_cast")];
            tensor<int32, [4]> var_51334_begin_0 = const()[name = tensor<string, []>("op_51334_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_51334_end_0 = const()[name = tensor<string, []>("op_51334_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_51334_end_mask_0 = const()[name = tensor<string, []>("op_51334_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51334_cast = slice_by_index(begin = var_51334_begin_0, end = var_51334_end_0, end_mask = var_51334_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51334_cast")];
            tensor<int32, [4]> var_51338_begin_0 = const()[name = tensor<string, []>("op_51338_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_51338_end_0 = const()[name = tensor<string, []>("op_51338_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_51338_end_mask_0 = const()[name = tensor<string, []>("op_51338_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51338_cast = slice_by_index(begin = var_51338_begin_0, end = var_51338_end_0, end_mask = var_51338_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51338_cast")];
            tensor<int32, [4]> var_51342_begin_0 = const()[name = tensor<string, []>("op_51342_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_51342_end_0 = const()[name = tensor<string, []>("op_51342_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_51342_end_mask_0 = const()[name = tensor<string, []>("op_51342_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51342_cast = slice_by_index(begin = var_51342_begin_0, end = var_51342_end_0, end_mask = var_51342_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51342_cast")];
            tensor<int32, [4]> var_51346_begin_0 = const()[name = tensor<string, []>("op_51346_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_51346_end_0 = const()[name = tensor<string, []>("op_51346_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_51346_end_mask_0 = const()[name = tensor<string, []>("op_51346_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51346_cast = slice_by_index(begin = var_51346_begin_0, end = var_51346_end_0, end_mask = var_51346_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51346_cast")];
            tensor<int32, [4]> var_51350_begin_0 = const()[name = tensor<string, []>("op_51350_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_51350_end_0 = const()[name = tensor<string, []>("op_51350_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_51350_end_mask_0 = const()[name = tensor<string, []>("op_51350_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51350_cast = slice_by_index(begin = var_51350_begin_0, end = var_51350_end_0, end_mask = var_51350_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51350_cast")];
            tensor<int32, [4]> var_51354_begin_0 = const()[name = tensor<string, []>("op_51354_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_51354_end_0 = const()[name = tensor<string, []>("op_51354_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_51354_end_mask_0 = const()[name = tensor<string, []>("op_51354_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51354_cast = slice_by_index(begin = var_51354_begin_0, end = var_51354_end_0, end_mask = var_51354_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51354_cast")];
            tensor<int32, [4]> var_51358_begin_0 = const()[name = tensor<string, []>("op_51358_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_51358_end_0 = const()[name = tensor<string, []>("op_51358_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_51358_end_mask_0 = const()[name = tensor<string, []>("op_51358_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51358_cast = slice_by_index(begin = var_51358_begin_0, end = var_51358_end_0, end_mask = var_51358_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51358_cast")];
            tensor<int32, [4]> var_51362_begin_0 = const()[name = tensor<string, []>("op_51362_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_51362_end_0 = const()[name = tensor<string, []>("op_51362_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_51362_end_mask_0 = const()[name = tensor<string, []>("op_51362_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51362_cast = slice_by_index(begin = var_51362_begin_0, end = var_51362_end_0, end_mask = var_51362_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51362_cast")];
            tensor<int32, [4]> var_51366_begin_0 = const()[name = tensor<string, []>("op_51366_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_51366_end_0 = const()[name = tensor<string, []>("op_51366_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_51366_end_mask_0 = const()[name = tensor<string, []>("op_51366_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51366_cast = slice_by_index(begin = var_51366_begin_0, end = var_51366_end_0, end_mask = var_51366_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51366_cast")];
            tensor<int32, [4]> var_51370_begin_0 = const()[name = tensor<string, []>("op_51370_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_51370_end_0 = const()[name = tensor<string, []>("op_51370_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_51370_end_mask_0 = const()[name = tensor<string, []>("op_51370_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_51370_cast = slice_by_index(begin = var_51370_begin_0, end = var_51370_end_0, end_mask = var_51370_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_51370_cast")];
            tensor<int32, [4]> var_51372_begin_0 = const()[name = tensor<string, []>("op_51372_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_51372_end_0 = const()[name = tensor<string, []>("op_51372_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_51372_end_mask_0 = const()[name = tensor<string, []>("op_51372_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51372_cast = slice_by_index(begin = var_51372_begin_0, end = var_51372_end_0, end_mask = var_51372_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51372_cast")];
            tensor<int32, [4]> var_51376_begin_0 = const()[name = tensor<string, []>("op_51376_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_51376_end_0 = const()[name = tensor<string, []>("op_51376_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_51376_end_mask_0 = const()[name = tensor<string, []>("op_51376_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51376_cast = slice_by_index(begin = var_51376_begin_0, end = var_51376_end_0, end_mask = var_51376_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51376_cast")];
            tensor<int32, [4]> var_51380_begin_0 = const()[name = tensor<string, []>("op_51380_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_51380_end_0 = const()[name = tensor<string, []>("op_51380_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_51380_end_mask_0 = const()[name = tensor<string, []>("op_51380_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51380_cast = slice_by_index(begin = var_51380_begin_0, end = var_51380_end_0, end_mask = var_51380_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51380_cast")];
            tensor<int32, [4]> var_51384_begin_0 = const()[name = tensor<string, []>("op_51384_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_51384_end_0 = const()[name = tensor<string, []>("op_51384_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_51384_end_mask_0 = const()[name = tensor<string, []>("op_51384_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51384_cast = slice_by_index(begin = var_51384_begin_0, end = var_51384_end_0, end_mask = var_51384_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51384_cast")];
            tensor<int32, [4]> var_51388_begin_0 = const()[name = tensor<string, []>("op_51388_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_51388_end_0 = const()[name = tensor<string, []>("op_51388_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_51388_end_mask_0 = const()[name = tensor<string, []>("op_51388_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51388_cast = slice_by_index(begin = var_51388_begin_0, end = var_51388_end_0, end_mask = var_51388_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51388_cast")];
            tensor<int32, [4]> var_51392_begin_0 = const()[name = tensor<string, []>("op_51392_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_51392_end_0 = const()[name = tensor<string, []>("op_51392_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_51392_end_mask_0 = const()[name = tensor<string, []>("op_51392_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51392_cast = slice_by_index(begin = var_51392_begin_0, end = var_51392_end_0, end_mask = var_51392_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51392_cast")];
            tensor<int32, [4]> var_51396_begin_0 = const()[name = tensor<string, []>("op_51396_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_51396_end_0 = const()[name = tensor<string, []>("op_51396_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_51396_end_mask_0 = const()[name = tensor<string, []>("op_51396_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51396_cast = slice_by_index(begin = var_51396_begin_0, end = var_51396_end_0, end_mask = var_51396_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51396_cast")];
            tensor<int32, [4]> var_51400_begin_0 = const()[name = tensor<string, []>("op_51400_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_51400_end_0 = const()[name = tensor<string, []>("op_51400_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_51400_end_mask_0 = const()[name = tensor<string, []>("op_51400_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51400_cast = slice_by_index(begin = var_51400_begin_0, end = var_51400_end_0, end_mask = var_51400_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51400_cast")];
            tensor<int32, [4]> var_51404_begin_0 = const()[name = tensor<string, []>("op_51404_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_51404_end_0 = const()[name = tensor<string, []>("op_51404_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_51404_end_mask_0 = const()[name = tensor<string, []>("op_51404_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51404_cast = slice_by_index(begin = var_51404_begin_0, end = var_51404_end_0, end_mask = var_51404_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51404_cast")];
            tensor<int32, [4]> var_51408_begin_0 = const()[name = tensor<string, []>("op_51408_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_51408_end_0 = const()[name = tensor<string, []>("op_51408_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_51408_end_mask_0 = const()[name = tensor<string, []>("op_51408_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51408_cast = slice_by_index(begin = var_51408_begin_0, end = var_51408_end_0, end_mask = var_51408_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51408_cast")];
            tensor<int32, [4]> var_51412_begin_0 = const()[name = tensor<string, []>("op_51412_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_51412_end_0 = const()[name = tensor<string, []>("op_51412_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_51412_end_mask_0 = const()[name = tensor<string, []>("op_51412_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51412_cast = slice_by_index(begin = var_51412_begin_0, end = var_51412_end_0, end_mask = var_51412_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51412_cast")];
            tensor<int32, [4]> var_51416_begin_0 = const()[name = tensor<string, []>("op_51416_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_51416_end_0 = const()[name = tensor<string, []>("op_51416_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_51416_end_mask_0 = const()[name = tensor<string, []>("op_51416_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51416_cast = slice_by_index(begin = var_51416_begin_0, end = var_51416_end_0, end_mask = var_51416_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51416_cast")];
            tensor<int32, [4]> var_51420_begin_0 = const()[name = tensor<string, []>("op_51420_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_51420_end_0 = const()[name = tensor<string, []>("op_51420_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_51420_end_mask_0 = const()[name = tensor<string, []>("op_51420_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51420_cast = slice_by_index(begin = var_51420_begin_0, end = var_51420_end_0, end_mask = var_51420_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51420_cast")];
            tensor<int32, [4]> var_51424_begin_0 = const()[name = tensor<string, []>("op_51424_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_51424_end_0 = const()[name = tensor<string, []>("op_51424_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_51424_end_mask_0 = const()[name = tensor<string, []>("op_51424_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51424_cast = slice_by_index(begin = var_51424_begin_0, end = var_51424_end_0, end_mask = var_51424_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51424_cast")];
            tensor<int32, [4]> var_51428_begin_0 = const()[name = tensor<string, []>("op_51428_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_51428_end_0 = const()[name = tensor<string, []>("op_51428_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_51428_end_mask_0 = const()[name = tensor<string, []>("op_51428_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51428_cast = slice_by_index(begin = var_51428_begin_0, end = var_51428_end_0, end_mask = var_51428_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51428_cast")];
            tensor<int32, [4]> var_51432_begin_0 = const()[name = tensor<string, []>("op_51432_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_51432_end_0 = const()[name = tensor<string, []>("op_51432_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_51432_end_mask_0 = const()[name = tensor<string, []>("op_51432_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51432_cast = slice_by_index(begin = var_51432_begin_0, end = var_51432_end_0, end_mask = var_51432_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51432_cast")];
            tensor<int32, [4]> var_51436_begin_0 = const()[name = tensor<string, []>("op_51436_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_51436_end_0 = const()[name = tensor<string, []>("op_51436_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_51436_end_mask_0 = const()[name = tensor<string, []>("op_51436_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51436_cast = slice_by_index(begin = var_51436_begin_0, end = var_51436_end_0, end_mask = var_51436_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51436_cast")];
            tensor<int32, [4]> var_51440_begin_0 = const()[name = tensor<string, []>("op_51440_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_51440_end_0 = const()[name = tensor<string, []>("op_51440_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_51440_end_mask_0 = const()[name = tensor<string, []>("op_51440_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51440_cast = slice_by_index(begin = var_51440_begin_0, end = var_51440_end_0, end_mask = var_51440_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51440_cast")];
            tensor<int32, [4]> var_51444_begin_0 = const()[name = tensor<string, []>("op_51444_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_51444_end_0 = const()[name = tensor<string, []>("op_51444_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_51444_end_mask_0 = const()[name = tensor<string, []>("op_51444_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51444_cast = slice_by_index(begin = var_51444_begin_0, end = var_51444_end_0, end_mask = var_51444_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51444_cast")];
            tensor<int32, [4]> var_51448_begin_0 = const()[name = tensor<string, []>("op_51448_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_51448_end_0 = const()[name = tensor<string, []>("op_51448_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_51448_end_mask_0 = const()[name = tensor<string, []>("op_51448_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_51448_cast = slice_by_index(begin = var_51448_begin_0, end = var_51448_end_0, end_mask = var_51448_end_mask_0, x = v_227_cast)[name = tensor<string, []>("op_51448_cast")];
            tensor<string, []> var_51452_equation_0 = const()[name = tensor<string, []>("op_51452_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51452_cast = einsum(equation = var_51452_equation_0, values = (var_51294_cast, var_51211_cast))[name = tensor<string, []>("op_51452_cast")];
            tensor<fp16, []> var_51453_to_fp16 = const()[name = tensor<string, []>("op_51453_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4361_cast = mul(x = var_51452_cast, y = var_51453_to_fp16)[name = tensor<string, []>("aw_4361_cast")];
            tensor<string, []> var_51456_equation_0 = const()[name = tensor<string, []>("op_51456_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51456_cast = einsum(equation = var_51456_equation_0, values = (var_51298_cast, var_51215_cast))[name = tensor<string, []>("op_51456_cast")];
            tensor<fp16, []> var_51457_to_fp16 = const()[name = tensor<string, []>("op_51457_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4363_cast = mul(x = var_51456_cast, y = var_51457_to_fp16)[name = tensor<string, []>("aw_4363_cast")];
            tensor<string, []> var_51460_equation_0 = const()[name = tensor<string, []>("op_51460_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51460_cast = einsum(equation = var_51460_equation_0, values = (var_51302_cast, var_51219_cast))[name = tensor<string, []>("op_51460_cast")];
            tensor<fp16, []> var_51461_to_fp16 = const()[name = tensor<string, []>("op_51461_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4365_cast = mul(x = var_51460_cast, y = var_51461_to_fp16)[name = tensor<string, []>("aw_4365_cast")];
            tensor<string, []> var_51464_equation_0 = const()[name = tensor<string, []>("op_51464_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51464_cast = einsum(equation = var_51464_equation_0, values = (var_51306_cast, var_51223_cast))[name = tensor<string, []>("op_51464_cast")];
            tensor<fp16, []> var_51465_to_fp16 = const()[name = tensor<string, []>("op_51465_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4367_cast = mul(x = var_51464_cast, y = var_51465_to_fp16)[name = tensor<string, []>("aw_4367_cast")];
            tensor<string, []> var_51468_equation_0 = const()[name = tensor<string, []>("op_51468_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51468_cast = einsum(equation = var_51468_equation_0, values = (var_51310_cast, var_51227_cast))[name = tensor<string, []>("op_51468_cast")];
            tensor<fp16, []> var_51469_to_fp16 = const()[name = tensor<string, []>("op_51469_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4369_cast = mul(x = var_51468_cast, y = var_51469_to_fp16)[name = tensor<string, []>("aw_4369_cast")];
            tensor<string, []> var_51472_equation_0 = const()[name = tensor<string, []>("op_51472_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51472_cast = einsum(equation = var_51472_equation_0, values = (var_51314_cast, var_51231_cast))[name = tensor<string, []>("op_51472_cast")];
            tensor<fp16, []> var_51473_to_fp16 = const()[name = tensor<string, []>("op_51473_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4371_cast = mul(x = var_51472_cast, y = var_51473_to_fp16)[name = tensor<string, []>("aw_4371_cast")];
            tensor<string, []> var_51476_equation_0 = const()[name = tensor<string, []>("op_51476_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51476_cast = einsum(equation = var_51476_equation_0, values = (var_51318_cast, var_51235_cast))[name = tensor<string, []>("op_51476_cast")];
            tensor<fp16, []> var_51477_to_fp16 = const()[name = tensor<string, []>("op_51477_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4373_cast = mul(x = var_51476_cast, y = var_51477_to_fp16)[name = tensor<string, []>("aw_4373_cast")];
            tensor<string, []> var_51480_equation_0 = const()[name = tensor<string, []>("op_51480_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51480_cast = einsum(equation = var_51480_equation_0, values = (var_51322_cast, var_51239_cast))[name = tensor<string, []>("op_51480_cast")];
            tensor<fp16, []> var_51481_to_fp16 = const()[name = tensor<string, []>("op_51481_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4375_cast = mul(x = var_51480_cast, y = var_51481_to_fp16)[name = tensor<string, []>("aw_4375_cast")];
            tensor<string, []> var_51484_equation_0 = const()[name = tensor<string, []>("op_51484_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51484_cast = einsum(equation = var_51484_equation_0, values = (var_51326_cast, var_51243_cast))[name = tensor<string, []>("op_51484_cast")];
            tensor<fp16, []> var_51485_to_fp16 = const()[name = tensor<string, []>("op_51485_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4377_cast = mul(x = var_51484_cast, y = var_51485_to_fp16)[name = tensor<string, []>("aw_4377_cast")];
            tensor<string, []> var_51488_equation_0 = const()[name = tensor<string, []>("op_51488_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51488_cast = einsum(equation = var_51488_equation_0, values = (var_51330_cast, var_51247_cast))[name = tensor<string, []>("op_51488_cast")];
            tensor<fp16, []> var_51489_to_fp16 = const()[name = tensor<string, []>("op_51489_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4379_cast = mul(x = var_51488_cast, y = var_51489_to_fp16)[name = tensor<string, []>("aw_4379_cast")];
            tensor<string, []> var_51492_equation_0 = const()[name = tensor<string, []>("op_51492_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51492_cast = einsum(equation = var_51492_equation_0, values = (var_51334_cast, var_51251_cast))[name = tensor<string, []>("op_51492_cast")];
            tensor<fp16, []> var_51493_to_fp16 = const()[name = tensor<string, []>("op_51493_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4381_cast = mul(x = var_51492_cast, y = var_51493_to_fp16)[name = tensor<string, []>("aw_4381_cast")];
            tensor<string, []> var_51496_equation_0 = const()[name = tensor<string, []>("op_51496_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51496_cast = einsum(equation = var_51496_equation_0, values = (var_51338_cast, var_51255_cast))[name = tensor<string, []>("op_51496_cast")];
            tensor<fp16, []> var_51497_to_fp16 = const()[name = tensor<string, []>("op_51497_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4383_cast = mul(x = var_51496_cast, y = var_51497_to_fp16)[name = tensor<string, []>("aw_4383_cast")];
            tensor<string, []> var_51500_equation_0 = const()[name = tensor<string, []>("op_51500_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51500_cast = einsum(equation = var_51500_equation_0, values = (var_51342_cast, var_51259_cast))[name = tensor<string, []>("op_51500_cast")];
            tensor<fp16, []> var_51501_to_fp16 = const()[name = tensor<string, []>("op_51501_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4385_cast = mul(x = var_51500_cast, y = var_51501_to_fp16)[name = tensor<string, []>("aw_4385_cast")];
            tensor<string, []> var_51504_equation_0 = const()[name = tensor<string, []>("op_51504_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51504_cast = einsum(equation = var_51504_equation_0, values = (var_51346_cast, var_51263_cast))[name = tensor<string, []>("op_51504_cast")];
            tensor<fp16, []> var_51505_to_fp16 = const()[name = tensor<string, []>("op_51505_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4387_cast = mul(x = var_51504_cast, y = var_51505_to_fp16)[name = tensor<string, []>("aw_4387_cast")];
            tensor<string, []> var_51508_equation_0 = const()[name = tensor<string, []>("op_51508_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51508_cast = einsum(equation = var_51508_equation_0, values = (var_51350_cast, var_51267_cast))[name = tensor<string, []>("op_51508_cast")];
            tensor<fp16, []> var_51509_to_fp16 = const()[name = tensor<string, []>("op_51509_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4389_cast = mul(x = var_51508_cast, y = var_51509_to_fp16)[name = tensor<string, []>("aw_4389_cast")];
            tensor<string, []> var_51512_equation_0 = const()[name = tensor<string, []>("op_51512_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51512_cast = einsum(equation = var_51512_equation_0, values = (var_51354_cast, var_51271_cast))[name = tensor<string, []>("op_51512_cast")];
            tensor<fp16, []> var_51513_to_fp16 = const()[name = tensor<string, []>("op_51513_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4391_cast = mul(x = var_51512_cast, y = var_51513_to_fp16)[name = tensor<string, []>("aw_4391_cast")];
            tensor<string, []> var_51516_equation_0 = const()[name = tensor<string, []>("op_51516_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51516_cast = einsum(equation = var_51516_equation_0, values = (var_51358_cast, var_51275_cast))[name = tensor<string, []>("op_51516_cast")];
            tensor<fp16, []> var_51517_to_fp16 = const()[name = tensor<string, []>("op_51517_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4393_cast = mul(x = var_51516_cast, y = var_51517_to_fp16)[name = tensor<string, []>("aw_4393_cast")];
            tensor<string, []> var_51520_equation_0 = const()[name = tensor<string, []>("op_51520_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51520_cast = einsum(equation = var_51520_equation_0, values = (var_51362_cast, var_51279_cast))[name = tensor<string, []>("op_51520_cast")];
            tensor<fp16, []> var_51521_to_fp16 = const()[name = tensor<string, []>("op_51521_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4395_cast = mul(x = var_51520_cast, y = var_51521_to_fp16)[name = tensor<string, []>("aw_4395_cast")];
            tensor<string, []> var_51524_equation_0 = const()[name = tensor<string, []>("op_51524_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51524_cast = einsum(equation = var_51524_equation_0, values = (var_51366_cast, var_51283_cast))[name = tensor<string, []>("op_51524_cast")];
            tensor<fp16, []> var_51525_to_fp16 = const()[name = tensor<string, []>("op_51525_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4397_cast = mul(x = var_51524_cast, y = var_51525_to_fp16)[name = tensor<string, []>("aw_4397_cast")];
            tensor<string, []> var_51528_equation_0 = const()[name = tensor<string, []>("op_51528_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_51528_cast = einsum(equation = var_51528_equation_0, values = (var_51370_cast, var_51287_cast))[name = tensor<string, []>("op_51528_cast")];
            tensor<fp16, []> var_51529_to_fp16 = const()[name = tensor<string, []>("op_51529_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4399_cast = mul(x = var_51528_cast, y = var_51529_to_fp16)[name = tensor<string, []>("aw_4399_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51531_cast = softmax(axis = var_30385, x = aw_4361_cast)[name = tensor<string, []>("op_51531_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51532_cast = softmax(axis = var_30385, x = aw_4363_cast)[name = tensor<string, []>("op_51532_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51533_cast = softmax(axis = var_30385, x = aw_4365_cast)[name = tensor<string, []>("op_51533_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51534_cast = softmax(axis = var_30385, x = aw_4367_cast)[name = tensor<string, []>("op_51534_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51535_cast = softmax(axis = var_30385, x = aw_4369_cast)[name = tensor<string, []>("op_51535_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51536_cast = softmax(axis = var_30385, x = aw_4371_cast)[name = tensor<string, []>("op_51536_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51537_cast = softmax(axis = var_30385, x = aw_4373_cast)[name = tensor<string, []>("op_51537_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51538_cast = softmax(axis = var_30385, x = aw_4375_cast)[name = tensor<string, []>("op_51538_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51539_cast = softmax(axis = var_30385, x = aw_4377_cast)[name = tensor<string, []>("op_51539_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51540_cast = softmax(axis = var_30385, x = aw_4379_cast)[name = tensor<string, []>("op_51540_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51541_cast = softmax(axis = var_30385, x = aw_4381_cast)[name = tensor<string, []>("op_51541_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51542_cast = softmax(axis = var_30385, x = aw_4383_cast)[name = tensor<string, []>("op_51542_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51543_cast = softmax(axis = var_30385, x = aw_4385_cast)[name = tensor<string, []>("op_51543_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51544_cast = softmax(axis = var_30385, x = aw_4387_cast)[name = tensor<string, []>("op_51544_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51545_cast = softmax(axis = var_30385, x = aw_4389_cast)[name = tensor<string, []>("op_51545_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51546_cast = softmax(axis = var_30385, x = aw_4391_cast)[name = tensor<string, []>("op_51546_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51547_cast = softmax(axis = var_30385, x = aw_4393_cast)[name = tensor<string, []>("op_51547_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51548_cast = softmax(axis = var_30385, x = aw_4395_cast)[name = tensor<string, []>("op_51548_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51549_cast = softmax(axis = var_30385, x = aw_4397_cast)[name = tensor<string, []>("op_51549_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_51550_cast = softmax(axis = var_30385, x = aw_4399_cast)[name = tensor<string, []>("op_51550_cast")];
            tensor<string, []> var_51552_equation_0 = const()[name = tensor<string, []>("op_51552_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51552_cast = einsum(equation = var_51552_equation_0, values = (var_51372_cast, var_51531_cast))[name = tensor<string, []>("op_51552_cast")];
            tensor<string, []> var_51554_equation_0 = const()[name = tensor<string, []>("op_51554_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51554_cast = einsum(equation = var_51554_equation_0, values = (var_51376_cast, var_51532_cast))[name = tensor<string, []>("op_51554_cast")];
            tensor<string, []> var_51556_equation_0 = const()[name = tensor<string, []>("op_51556_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51556_cast = einsum(equation = var_51556_equation_0, values = (var_51380_cast, var_51533_cast))[name = tensor<string, []>("op_51556_cast")];
            tensor<string, []> var_51558_equation_0 = const()[name = tensor<string, []>("op_51558_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51558_cast = einsum(equation = var_51558_equation_0, values = (var_51384_cast, var_51534_cast))[name = tensor<string, []>("op_51558_cast")];
            tensor<string, []> var_51560_equation_0 = const()[name = tensor<string, []>("op_51560_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51560_cast = einsum(equation = var_51560_equation_0, values = (var_51388_cast, var_51535_cast))[name = tensor<string, []>("op_51560_cast")];
            tensor<string, []> var_51562_equation_0 = const()[name = tensor<string, []>("op_51562_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51562_cast = einsum(equation = var_51562_equation_0, values = (var_51392_cast, var_51536_cast))[name = tensor<string, []>("op_51562_cast")];
            tensor<string, []> var_51564_equation_0 = const()[name = tensor<string, []>("op_51564_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51564_cast = einsum(equation = var_51564_equation_0, values = (var_51396_cast, var_51537_cast))[name = tensor<string, []>("op_51564_cast")];
            tensor<string, []> var_51566_equation_0 = const()[name = tensor<string, []>("op_51566_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51566_cast = einsum(equation = var_51566_equation_0, values = (var_51400_cast, var_51538_cast))[name = tensor<string, []>("op_51566_cast")];
            tensor<string, []> var_51568_equation_0 = const()[name = tensor<string, []>("op_51568_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51568_cast = einsum(equation = var_51568_equation_0, values = (var_51404_cast, var_51539_cast))[name = tensor<string, []>("op_51568_cast")];
            tensor<string, []> var_51570_equation_0 = const()[name = tensor<string, []>("op_51570_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51570_cast = einsum(equation = var_51570_equation_0, values = (var_51408_cast, var_51540_cast))[name = tensor<string, []>("op_51570_cast")];
            tensor<string, []> var_51572_equation_0 = const()[name = tensor<string, []>("op_51572_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51572_cast = einsum(equation = var_51572_equation_0, values = (var_51412_cast, var_51541_cast))[name = tensor<string, []>("op_51572_cast")];
            tensor<string, []> var_51574_equation_0 = const()[name = tensor<string, []>("op_51574_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51574_cast = einsum(equation = var_51574_equation_0, values = (var_51416_cast, var_51542_cast))[name = tensor<string, []>("op_51574_cast")];
            tensor<string, []> var_51576_equation_0 = const()[name = tensor<string, []>("op_51576_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51576_cast = einsum(equation = var_51576_equation_0, values = (var_51420_cast, var_51543_cast))[name = tensor<string, []>("op_51576_cast")];
            tensor<string, []> var_51578_equation_0 = const()[name = tensor<string, []>("op_51578_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51578_cast = einsum(equation = var_51578_equation_0, values = (var_51424_cast, var_51544_cast))[name = tensor<string, []>("op_51578_cast")];
            tensor<string, []> var_51580_equation_0 = const()[name = tensor<string, []>("op_51580_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51580_cast = einsum(equation = var_51580_equation_0, values = (var_51428_cast, var_51545_cast))[name = tensor<string, []>("op_51580_cast")];
            tensor<string, []> var_51582_equation_0 = const()[name = tensor<string, []>("op_51582_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51582_cast = einsum(equation = var_51582_equation_0, values = (var_51432_cast, var_51546_cast))[name = tensor<string, []>("op_51582_cast")];
            tensor<string, []> var_51584_equation_0 = const()[name = tensor<string, []>("op_51584_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51584_cast = einsum(equation = var_51584_equation_0, values = (var_51436_cast, var_51547_cast))[name = tensor<string, []>("op_51584_cast")];
            tensor<string, []> var_51586_equation_0 = const()[name = tensor<string, []>("op_51586_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51586_cast = einsum(equation = var_51586_equation_0, values = (var_51440_cast, var_51548_cast))[name = tensor<string, []>("op_51586_cast")];
            tensor<string, []> var_51588_equation_0 = const()[name = tensor<string, []>("op_51588_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51588_cast = einsum(equation = var_51588_equation_0, values = (var_51444_cast, var_51549_cast))[name = tensor<string, []>("op_51588_cast")];
            tensor<string, []> var_51590_equation_0 = const()[name = tensor<string, []>("op_51590_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_51590_cast = einsum(equation = var_51590_equation_0, values = (var_51448_cast, var_51550_cast))[name = tensor<string, []>("op_51590_cast")];
            tensor<bool, []> input_655_interleave_0 = const()[name = tensor<string, []>("input_655_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_655_cast = concat(axis = var_30385, interleave = input_655_interleave_0, values = (var_51552_cast, var_51554_cast, var_51556_cast, var_51558_cast, var_51560_cast, var_51562_cast, var_51564_cast, var_51566_cast, var_51568_cast, var_51570_cast, var_51572_cast, var_51574_cast, var_51576_cast, var_51578_cast, var_51580_cast, var_51582_cast, var_51584_cast, var_51586_cast, var_51588_cast, var_51590_cast))[name = tensor<string, []>("input_655_cast")];
            tensor<int32, [2]> var_51596 = const()[name = tensor<string, []>("op_51596"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_51598 = const()[name = tensor<string, []>("op_51598"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_51600_pad_type_0 = const()[name = tensor<string, []>("op_51600_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_51600_pad_0 = const()[name = tensor<string, []>("op_51600_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4338000256)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4341277120)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_51600_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_out_0_bias_to_fp16, dilations = var_51598, groups = var_30385, pad = var_51600_pad_0, pad_type = var_51600_pad_type_0, strides = var_51596, weight = up_blocks_0_attentions_2_transformer_blocks_2_attn2_to_out_0_weight_to_fp16, x = input_655_cast)[name = tensor<string, []>("op_51600_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_341_cast = add(x = var_51600_cast, y = inputs_339_cast)[name = tensor<string, []>("inputs_341_cast")];
            tensor<int32, [1]> var_51604 = const()[name = tensor<string, []>("op_51604"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_341_cast = reduce_mean(axes = var_51604, keep_dims = var_30380, x = inputs_341_cast)[name = tensor<string, []>("channels_mean_341_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_341_cast = sub(x = inputs_341_cast, y = channels_mean_341_cast)[name = tensor<string, []>("zero_mean_341_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_341_cast = mul(x = zero_mean_341_cast, y = zero_mean_341_cast)[name = tensor<string, []>("zero_mean_sq_341_cast")];
            tensor<int32, [1]> var_51608 = const()[name = tensor<string, []>("op_51608"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_51609_cast = reduce_mean(axes = var_51608, keep_dims = var_30380, x = zero_mean_sq_341_cast)[name = tensor<string, []>("op_51609_cast")];
            tensor<fp16, []> var_51610_to_fp16 = const()[name = tensor<string, []>("op_51610_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_51611_cast = add(x = var_51609_cast, y = var_51610_to_fp16)[name = tensor<string, []>("op_51611_cast")];
            tensor<fp16, []> denom_341_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_341_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_341_cast = rsqrt(epsilon = denom_341_epsilon_0_to_fp16, x = var_51611_cast)[name = tensor<string, []>("denom_341_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_341_cast = mul(x = zero_mean_341_cast, y = denom_341_cast)[name = tensor<string, []>("out_341_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_51615_to_fp16 = const()[name = tensor<string, []>("op_51615_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4341279744)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_51616_cast = add(x = out_341_cast, y = var_51615_to_fp16)[name = tensor<string, []>("op_51616_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_51618_to_fp16 = const()[name = tensor<string, []>("op_51618_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4341282368)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_657_cast = mul(x = var_51616_cast, y = var_51618_to_fp16)[name = tensor<string, []>("input_657_cast")];
            tensor<int32, [2]> var_51626 = const()[name = tensor<string, []>("op_51626"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_51628 = const()[name = tensor<string, []>("op_51628"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_51630_pad_type_0 = const()[name = tensor<string, []>("op_51630_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_51630_pad_0 = const()[name = tensor<string, []>("op_51630_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4341284992)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_2_transformer_blocks_2_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4367499456)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_51630_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_2_ff_net_0_proj_bias_to_fp16, dilations = var_51628, groups = var_30385, pad = var_51630_pad_0, pad_type = var_51630_pad_type_0, strides = var_51626, weight = up_blocks_0_attentions_2_transformer_blocks_2_ff_net_0_proj_weight_to_fp16, x = input_657_cast)[name = tensor<string, []>("op_51630_cast")];
            tensor<int32, [2]> var_51631_split_sizes_0 = const()[name = tensor<string, []>("op_51631_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_51631_axis_0 = const()[name = tensor<string, []>("op_51631_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_51631_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_51631_cast_1 = split(axis = var_51631_axis_0, split_sizes = var_51631_split_sizes_0, x = var_51630_cast)[name = tensor<string, []>("op_51631_cast")];
            tensor<string, []> var_51633_mode_0 = const()[name = tensor<string, []>("op_51633_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_51633_cast = gelu(mode = var_51633_mode_0, x = var_51631_cast_1)[name = tensor<string, []>("op_51633_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_659_cast = mul(x = var_51631_cast_0, y = var_51633_cast)[name = tensor<string, []>("input_659_cast")];
            tensor<int32, [2]> var_51637 = const()[name = tensor<string, []>("op_51637"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_51639 = const()[name = tensor<string, []>("op_51639"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_51641_pad_type_0 = const()[name = tensor<string, []>("op_51641_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_51641_pad_0 = const()[name = tensor<string, []>("op_51641_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_2_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4367520000)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_2_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_2_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4380627264)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_51641_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_2_ff_net_2_bias_to_fp16, dilations = var_51639, groups = var_30385, pad = var_51641_pad_0, pad_type = var_51641_pad_type_0, strides = var_51637, weight = up_blocks_0_attentions_2_transformer_blocks_2_ff_net_2_weight_to_fp16, x = input_659_cast)[name = tensor<string, []>("op_51641_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_343_cast = add(x = var_51641_cast, y = inputs_341_cast)[name = tensor<string, []>("inputs_343_cast")];
            tensor<int32, [1]> var_51651 = const()[name = tensor<string, []>("op_51651"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_343_cast = reduce_mean(axes = var_51651, keep_dims = var_30380, x = inputs_343_cast)[name = tensor<string, []>("channels_mean_343_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_343_cast = sub(x = inputs_343_cast, y = channels_mean_343_cast)[name = tensor<string, []>("zero_mean_343_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_343_cast = mul(x = zero_mean_343_cast, y = zero_mean_343_cast)[name = tensor<string, []>("zero_mean_sq_343_cast")];
            tensor<int32, [1]> var_51655 = const()[name = tensor<string, []>("op_51655"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_51656_cast = reduce_mean(axes = var_51655, keep_dims = var_30380, x = zero_mean_sq_343_cast)[name = tensor<string, []>("op_51656_cast")];
            tensor<fp16, []> var_51657_to_fp16 = const()[name = tensor<string, []>("op_51657_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_51658_cast = add(x = var_51656_cast, y = var_51657_to_fp16)[name = tensor<string, []>("op_51658_cast")];
            tensor<fp16, []> denom_343_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_343_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_343_cast = rsqrt(epsilon = denom_343_epsilon_0_to_fp16, x = var_51658_cast)[name = tensor<string, []>("denom_343_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_343_cast = mul(x = zero_mean_343_cast, y = denom_343_cast)[name = tensor<string, []>("out_343_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_51662_to_fp16 = const()[name = tensor<string, []>("op_51662_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4380629888)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_51663_cast = add(x = out_343_cast, y = var_51662_to_fp16)[name = tensor<string, []>("op_51663_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_51665_to_fp16 = const()[name = tensor<string, []>("op_51665_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4380632512)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_453_cast = mul(x = var_51663_cast, y = var_51665_to_fp16)[name = tensor<string, []>("hidden_states_453_cast")];
            tensor<int32, [2]> var_51672 = const()[name = tensor<string, []>("op_51672"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_51674 = const()[name = tensor<string, []>("op_51674"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_229_pad_type_0 = const()[name = tensor<string, []>("q_229_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_229_pad_0 = const()[name = tensor<string, []>("q_229_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4380635136)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_229_cast = conv(dilations = var_51674, groups = var_30385, pad = q_229_pad_0, pad_type = q_229_pad_type_0, strides = var_51672, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_q_weight_to_fp16, x = hidden_states_453_cast)[name = tensor<string, []>("q_229_cast")];
            tensor<int32, [2]> var_51678 = const()[name = tensor<string, []>("op_51678"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_51680 = const()[name = tensor<string, []>("op_51680"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_457_pad_type_0 = const()[name = tensor<string, []>("k_457_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_457_pad_0 = const()[name = tensor<string, []>("k_457_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4383912000)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_457_cast = conv(dilations = var_51680, groups = var_30385, pad = k_457_pad_0, pad_type = k_457_pad_type_0, strides = var_51678, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_k_weight_to_fp16, x = hidden_states_453_cast)[name = tensor<string, []>("k_457_cast")];
            tensor<int32, [2]> var_51684 = const()[name = tensor<string, []>("op_51684"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_51686 = const()[name = tensor<string, []>("op_51686"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_229_pad_type_0 = const()[name = tensor<string, []>("v_229_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_229_pad_0 = const()[name = tensor<string, []>("v_229_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4387188864)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_229_cast = conv(dilations = var_51686, groups = var_30385, pad = v_229_pad_0, pad_type = v_229_pad_type_0, strides = var_51684, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_v_weight_to_fp16, x = hidden_states_453_cast)[name = tensor<string, []>("v_229_cast")];
            tensor<int32, [4]> var_51690_begin_0 = const()[name = tensor<string, []>("op_51690_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_51690_end_0 = const()[name = tensor<string, []>("op_51690_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_51690_end_mask_0 = const()[name = tensor<string, []>("op_51690_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51690_cast = slice_by_index(begin = var_51690_begin_0, end = var_51690_end_0, end_mask = var_51690_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51690_cast")];
            tensor<int32, [4]> var_51694_begin_0 = const()[name = tensor<string, []>("op_51694_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_51694_end_0 = const()[name = tensor<string, []>("op_51694_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_51694_end_mask_0 = const()[name = tensor<string, []>("op_51694_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51694_cast = slice_by_index(begin = var_51694_begin_0, end = var_51694_end_0, end_mask = var_51694_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51694_cast")];
            tensor<int32, [4]> var_51698_begin_0 = const()[name = tensor<string, []>("op_51698_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_51698_end_0 = const()[name = tensor<string, []>("op_51698_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_51698_end_mask_0 = const()[name = tensor<string, []>("op_51698_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51698_cast = slice_by_index(begin = var_51698_begin_0, end = var_51698_end_0, end_mask = var_51698_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51698_cast")];
            tensor<int32, [4]> var_51702_begin_0 = const()[name = tensor<string, []>("op_51702_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_51702_end_0 = const()[name = tensor<string, []>("op_51702_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_51702_end_mask_0 = const()[name = tensor<string, []>("op_51702_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51702_cast = slice_by_index(begin = var_51702_begin_0, end = var_51702_end_0, end_mask = var_51702_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51702_cast")];
            tensor<int32, [4]> var_51706_begin_0 = const()[name = tensor<string, []>("op_51706_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_51706_end_0 = const()[name = tensor<string, []>("op_51706_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_51706_end_mask_0 = const()[name = tensor<string, []>("op_51706_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51706_cast = slice_by_index(begin = var_51706_begin_0, end = var_51706_end_0, end_mask = var_51706_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51706_cast")];
            tensor<int32, [4]> var_51710_begin_0 = const()[name = tensor<string, []>("op_51710_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_51710_end_0 = const()[name = tensor<string, []>("op_51710_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_51710_end_mask_0 = const()[name = tensor<string, []>("op_51710_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51710_cast = slice_by_index(begin = var_51710_begin_0, end = var_51710_end_0, end_mask = var_51710_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51710_cast")];
            tensor<int32, [4]> var_51714_begin_0 = const()[name = tensor<string, []>("op_51714_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_51714_end_0 = const()[name = tensor<string, []>("op_51714_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_51714_end_mask_0 = const()[name = tensor<string, []>("op_51714_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51714_cast = slice_by_index(begin = var_51714_begin_0, end = var_51714_end_0, end_mask = var_51714_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51714_cast")];
            tensor<int32, [4]> var_51718_begin_0 = const()[name = tensor<string, []>("op_51718_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_51718_end_0 = const()[name = tensor<string, []>("op_51718_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_51718_end_mask_0 = const()[name = tensor<string, []>("op_51718_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51718_cast = slice_by_index(begin = var_51718_begin_0, end = var_51718_end_0, end_mask = var_51718_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51718_cast")];
            tensor<int32, [4]> var_51722_begin_0 = const()[name = tensor<string, []>("op_51722_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_51722_end_0 = const()[name = tensor<string, []>("op_51722_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_51722_end_mask_0 = const()[name = tensor<string, []>("op_51722_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51722_cast = slice_by_index(begin = var_51722_begin_0, end = var_51722_end_0, end_mask = var_51722_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51722_cast")];
            tensor<int32, [4]> var_51726_begin_0 = const()[name = tensor<string, []>("op_51726_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_51726_end_0 = const()[name = tensor<string, []>("op_51726_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_51726_end_mask_0 = const()[name = tensor<string, []>("op_51726_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51726_cast = slice_by_index(begin = var_51726_begin_0, end = var_51726_end_0, end_mask = var_51726_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51726_cast")];
            tensor<int32, [4]> var_51730_begin_0 = const()[name = tensor<string, []>("op_51730_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_51730_end_0 = const()[name = tensor<string, []>("op_51730_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_51730_end_mask_0 = const()[name = tensor<string, []>("op_51730_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51730_cast = slice_by_index(begin = var_51730_begin_0, end = var_51730_end_0, end_mask = var_51730_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51730_cast")];
            tensor<int32, [4]> var_51734_begin_0 = const()[name = tensor<string, []>("op_51734_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_51734_end_0 = const()[name = tensor<string, []>("op_51734_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_51734_end_mask_0 = const()[name = tensor<string, []>("op_51734_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51734_cast = slice_by_index(begin = var_51734_begin_0, end = var_51734_end_0, end_mask = var_51734_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51734_cast")];
            tensor<int32, [4]> var_51738_begin_0 = const()[name = tensor<string, []>("op_51738_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_51738_end_0 = const()[name = tensor<string, []>("op_51738_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_51738_end_mask_0 = const()[name = tensor<string, []>("op_51738_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51738_cast = slice_by_index(begin = var_51738_begin_0, end = var_51738_end_0, end_mask = var_51738_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51738_cast")];
            tensor<int32, [4]> var_51742_begin_0 = const()[name = tensor<string, []>("op_51742_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_51742_end_0 = const()[name = tensor<string, []>("op_51742_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_51742_end_mask_0 = const()[name = tensor<string, []>("op_51742_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51742_cast = slice_by_index(begin = var_51742_begin_0, end = var_51742_end_0, end_mask = var_51742_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51742_cast")];
            tensor<int32, [4]> var_51746_begin_0 = const()[name = tensor<string, []>("op_51746_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_51746_end_0 = const()[name = tensor<string, []>("op_51746_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_51746_end_mask_0 = const()[name = tensor<string, []>("op_51746_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51746_cast = slice_by_index(begin = var_51746_begin_0, end = var_51746_end_0, end_mask = var_51746_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51746_cast")];
            tensor<int32, [4]> var_51750_begin_0 = const()[name = tensor<string, []>("op_51750_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_51750_end_0 = const()[name = tensor<string, []>("op_51750_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_51750_end_mask_0 = const()[name = tensor<string, []>("op_51750_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51750_cast = slice_by_index(begin = var_51750_begin_0, end = var_51750_end_0, end_mask = var_51750_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51750_cast")];
            tensor<int32, [4]> var_51754_begin_0 = const()[name = tensor<string, []>("op_51754_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_51754_end_0 = const()[name = tensor<string, []>("op_51754_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_51754_end_mask_0 = const()[name = tensor<string, []>("op_51754_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51754_cast = slice_by_index(begin = var_51754_begin_0, end = var_51754_end_0, end_mask = var_51754_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51754_cast")];
            tensor<int32, [4]> var_51758_begin_0 = const()[name = tensor<string, []>("op_51758_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_51758_end_0 = const()[name = tensor<string, []>("op_51758_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_51758_end_mask_0 = const()[name = tensor<string, []>("op_51758_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51758_cast = slice_by_index(begin = var_51758_begin_0, end = var_51758_end_0, end_mask = var_51758_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51758_cast")];
            tensor<int32, [4]> var_51762_begin_0 = const()[name = tensor<string, []>("op_51762_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_51762_end_0 = const()[name = tensor<string, []>("op_51762_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_51762_end_mask_0 = const()[name = tensor<string, []>("op_51762_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51762_cast = slice_by_index(begin = var_51762_begin_0, end = var_51762_end_0, end_mask = var_51762_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51762_cast")];
            tensor<int32, [4]> var_51766_begin_0 = const()[name = tensor<string, []>("op_51766_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_51766_end_0 = const()[name = tensor<string, []>("op_51766_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_51766_end_mask_0 = const()[name = tensor<string, []>("op_51766_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51766_cast = slice_by_index(begin = var_51766_begin_0, end = var_51766_end_0, end_mask = var_51766_end_mask_0, x = q_229_cast)[name = tensor<string, []>("op_51766_cast")];
            tensor<int32, [4]> k_459_perm_0 = const()[name = tensor<string, []>("k_459_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_51773_begin_0 = const()[name = tensor<string, []>("op_51773_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_51773_end_0 = const()[name = tensor<string, []>("op_51773_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_51773_end_mask_0 = const()[name = tensor<string, []>("op_51773_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_25 = transpose(perm = k_459_perm_0, x = k_457_cast)[name = tensor<string, []>("transpose_25")];
            tensor<fp16, [2, 1024, 1, 64]> var_51773_cast = slice_by_index(begin = var_51773_begin_0, end = var_51773_end_0, end_mask = var_51773_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51773_cast")];
            tensor<int32, [4]> var_51777_begin_0 = const()[name = tensor<string, []>("op_51777_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_51777_end_0 = const()[name = tensor<string, []>("op_51777_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_51777_end_mask_0 = const()[name = tensor<string, []>("op_51777_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51777_cast = slice_by_index(begin = var_51777_begin_0, end = var_51777_end_0, end_mask = var_51777_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51777_cast")];
            tensor<int32, [4]> var_51781_begin_0 = const()[name = tensor<string, []>("op_51781_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_51781_end_0 = const()[name = tensor<string, []>("op_51781_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_51781_end_mask_0 = const()[name = tensor<string, []>("op_51781_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51781_cast = slice_by_index(begin = var_51781_begin_0, end = var_51781_end_0, end_mask = var_51781_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51781_cast")];
            tensor<int32, [4]> var_51785_begin_0 = const()[name = tensor<string, []>("op_51785_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_51785_end_0 = const()[name = tensor<string, []>("op_51785_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_51785_end_mask_0 = const()[name = tensor<string, []>("op_51785_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51785_cast = slice_by_index(begin = var_51785_begin_0, end = var_51785_end_0, end_mask = var_51785_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51785_cast")];
            tensor<int32, [4]> var_51789_begin_0 = const()[name = tensor<string, []>("op_51789_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_51789_end_0 = const()[name = tensor<string, []>("op_51789_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_51789_end_mask_0 = const()[name = tensor<string, []>("op_51789_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51789_cast = slice_by_index(begin = var_51789_begin_0, end = var_51789_end_0, end_mask = var_51789_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51789_cast")];
            tensor<int32, [4]> var_51793_begin_0 = const()[name = tensor<string, []>("op_51793_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_51793_end_0 = const()[name = tensor<string, []>("op_51793_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_51793_end_mask_0 = const()[name = tensor<string, []>("op_51793_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51793_cast = slice_by_index(begin = var_51793_begin_0, end = var_51793_end_0, end_mask = var_51793_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51793_cast")];
            tensor<int32, [4]> var_51797_begin_0 = const()[name = tensor<string, []>("op_51797_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_51797_end_0 = const()[name = tensor<string, []>("op_51797_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_51797_end_mask_0 = const()[name = tensor<string, []>("op_51797_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51797_cast = slice_by_index(begin = var_51797_begin_0, end = var_51797_end_0, end_mask = var_51797_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51797_cast")];
            tensor<int32, [4]> var_51801_begin_0 = const()[name = tensor<string, []>("op_51801_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_51801_end_0 = const()[name = tensor<string, []>("op_51801_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_51801_end_mask_0 = const()[name = tensor<string, []>("op_51801_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51801_cast = slice_by_index(begin = var_51801_begin_0, end = var_51801_end_0, end_mask = var_51801_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51801_cast")];
            tensor<int32, [4]> var_51805_begin_0 = const()[name = tensor<string, []>("op_51805_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_51805_end_0 = const()[name = tensor<string, []>("op_51805_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_51805_end_mask_0 = const()[name = tensor<string, []>("op_51805_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51805_cast = slice_by_index(begin = var_51805_begin_0, end = var_51805_end_0, end_mask = var_51805_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51805_cast")];
            tensor<int32, [4]> var_51809_begin_0 = const()[name = tensor<string, []>("op_51809_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_51809_end_0 = const()[name = tensor<string, []>("op_51809_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_51809_end_mask_0 = const()[name = tensor<string, []>("op_51809_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51809_cast = slice_by_index(begin = var_51809_begin_0, end = var_51809_end_0, end_mask = var_51809_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51809_cast")];
            tensor<int32, [4]> var_51813_begin_0 = const()[name = tensor<string, []>("op_51813_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_51813_end_0 = const()[name = tensor<string, []>("op_51813_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_51813_end_mask_0 = const()[name = tensor<string, []>("op_51813_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51813_cast = slice_by_index(begin = var_51813_begin_0, end = var_51813_end_0, end_mask = var_51813_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51813_cast")];
            tensor<int32, [4]> var_51817_begin_0 = const()[name = tensor<string, []>("op_51817_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_51817_end_0 = const()[name = tensor<string, []>("op_51817_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_51817_end_mask_0 = const()[name = tensor<string, []>("op_51817_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51817_cast = slice_by_index(begin = var_51817_begin_0, end = var_51817_end_0, end_mask = var_51817_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51817_cast")];
            tensor<int32, [4]> var_51821_begin_0 = const()[name = tensor<string, []>("op_51821_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_51821_end_0 = const()[name = tensor<string, []>("op_51821_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_51821_end_mask_0 = const()[name = tensor<string, []>("op_51821_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51821_cast = slice_by_index(begin = var_51821_begin_0, end = var_51821_end_0, end_mask = var_51821_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51821_cast")];
            tensor<int32, [4]> var_51825_begin_0 = const()[name = tensor<string, []>("op_51825_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_51825_end_0 = const()[name = tensor<string, []>("op_51825_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_51825_end_mask_0 = const()[name = tensor<string, []>("op_51825_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51825_cast = slice_by_index(begin = var_51825_begin_0, end = var_51825_end_0, end_mask = var_51825_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51825_cast")];
            tensor<int32, [4]> var_51829_begin_0 = const()[name = tensor<string, []>("op_51829_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_51829_end_0 = const()[name = tensor<string, []>("op_51829_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_51829_end_mask_0 = const()[name = tensor<string, []>("op_51829_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51829_cast = slice_by_index(begin = var_51829_begin_0, end = var_51829_end_0, end_mask = var_51829_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51829_cast")];
            tensor<int32, [4]> var_51833_begin_0 = const()[name = tensor<string, []>("op_51833_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_51833_end_0 = const()[name = tensor<string, []>("op_51833_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_51833_end_mask_0 = const()[name = tensor<string, []>("op_51833_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51833_cast = slice_by_index(begin = var_51833_begin_0, end = var_51833_end_0, end_mask = var_51833_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51833_cast")];
            tensor<int32, [4]> var_51837_begin_0 = const()[name = tensor<string, []>("op_51837_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_51837_end_0 = const()[name = tensor<string, []>("op_51837_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_51837_end_mask_0 = const()[name = tensor<string, []>("op_51837_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51837_cast = slice_by_index(begin = var_51837_begin_0, end = var_51837_end_0, end_mask = var_51837_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51837_cast")];
            tensor<int32, [4]> var_51841_begin_0 = const()[name = tensor<string, []>("op_51841_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_51841_end_0 = const()[name = tensor<string, []>("op_51841_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_51841_end_mask_0 = const()[name = tensor<string, []>("op_51841_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51841_cast = slice_by_index(begin = var_51841_begin_0, end = var_51841_end_0, end_mask = var_51841_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51841_cast")];
            tensor<int32, [4]> var_51845_begin_0 = const()[name = tensor<string, []>("op_51845_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_51845_end_0 = const()[name = tensor<string, []>("op_51845_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_51845_end_mask_0 = const()[name = tensor<string, []>("op_51845_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51845_cast = slice_by_index(begin = var_51845_begin_0, end = var_51845_end_0, end_mask = var_51845_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51845_cast")];
            tensor<int32, [4]> var_51849_begin_0 = const()[name = tensor<string, []>("op_51849_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_51849_end_0 = const()[name = tensor<string, []>("op_51849_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_51849_end_mask_0 = const()[name = tensor<string, []>("op_51849_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_51849_cast = slice_by_index(begin = var_51849_begin_0, end = var_51849_end_0, end_mask = var_51849_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_51849_cast")];
            tensor<int32, [4]> var_51851_begin_0 = const()[name = tensor<string, []>("op_51851_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_51851_end_0 = const()[name = tensor<string, []>("op_51851_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_51851_end_mask_0 = const()[name = tensor<string, []>("op_51851_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51851_cast = slice_by_index(begin = var_51851_begin_0, end = var_51851_end_0, end_mask = var_51851_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51851_cast")];
            tensor<int32, [4]> var_51855_begin_0 = const()[name = tensor<string, []>("op_51855_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_51855_end_0 = const()[name = tensor<string, []>("op_51855_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_51855_end_mask_0 = const()[name = tensor<string, []>("op_51855_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51855_cast = slice_by_index(begin = var_51855_begin_0, end = var_51855_end_0, end_mask = var_51855_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51855_cast")];
            tensor<int32, [4]> var_51859_begin_0 = const()[name = tensor<string, []>("op_51859_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_51859_end_0 = const()[name = tensor<string, []>("op_51859_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_51859_end_mask_0 = const()[name = tensor<string, []>("op_51859_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51859_cast = slice_by_index(begin = var_51859_begin_0, end = var_51859_end_0, end_mask = var_51859_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51859_cast")];
            tensor<int32, [4]> var_51863_begin_0 = const()[name = tensor<string, []>("op_51863_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_51863_end_0 = const()[name = tensor<string, []>("op_51863_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_51863_end_mask_0 = const()[name = tensor<string, []>("op_51863_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51863_cast = slice_by_index(begin = var_51863_begin_0, end = var_51863_end_0, end_mask = var_51863_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51863_cast")];
            tensor<int32, [4]> var_51867_begin_0 = const()[name = tensor<string, []>("op_51867_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_51867_end_0 = const()[name = tensor<string, []>("op_51867_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_51867_end_mask_0 = const()[name = tensor<string, []>("op_51867_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51867_cast = slice_by_index(begin = var_51867_begin_0, end = var_51867_end_0, end_mask = var_51867_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51867_cast")];
            tensor<int32, [4]> var_51871_begin_0 = const()[name = tensor<string, []>("op_51871_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_51871_end_0 = const()[name = tensor<string, []>("op_51871_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_51871_end_mask_0 = const()[name = tensor<string, []>("op_51871_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51871_cast = slice_by_index(begin = var_51871_begin_0, end = var_51871_end_0, end_mask = var_51871_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51871_cast")];
            tensor<int32, [4]> var_51875_begin_0 = const()[name = tensor<string, []>("op_51875_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_51875_end_0 = const()[name = tensor<string, []>("op_51875_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_51875_end_mask_0 = const()[name = tensor<string, []>("op_51875_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51875_cast = slice_by_index(begin = var_51875_begin_0, end = var_51875_end_0, end_mask = var_51875_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51875_cast")];
            tensor<int32, [4]> var_51879_begin_0 = const()[name = tensor<string, []>("op_51879_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_51879_end_0 = const()[name = tensor<string, []>("op_51879_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_51879_end_mask_0 = const()[name = tensor<string, []>("op_51879_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51879_cast = slice_by_index(begin = var_51879_begin_0, end = var_51879_end_0, end_mask = var_51879_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51879_cast")];
            tensor<int32, [4]> var_51883_begin_0 = const()[name = tensor<string, []>("op_51883_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_51883_end_0 = const()[name = tensor<string, []>("op_51883_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_51883_end_mask_0 = const()[name = tensor<string, []>("op_51883_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51883_cast = slice_by_index(begin = var_51883_begin_0, end = var_51883_end_0, end_mask = var_51883_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51883_cast")];
            tensor<int32, [4]> var_51887_begin_0 = const()[name = tensor<string, []>("op_51887_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_51887_end_0 = const()[name = tensor<string, []>("op_51887_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_51887_end_mask_0 = const()[name = tensor<string, []>("op_51887_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51887_cast = slice_by_index(begin = var_51887_begin_0, end = var_51887_end_0, end_mask = var_51887_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51887_cast")];
            tensor<int32, [4]> var_51891_begin_0 = const()[name = tensor<string, []>("op_51891_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_51891_end_0 = const()[name = tensor<string, []>("op_51891_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_51891_end_mask_0 = const()[name = tensor<string, []>("op_51891_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51891_cast = slice_by_index(begin = var_51891_begin_0, end = var_51891_end_0, end_mask = var_51891_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51891_cast")];
            tensor<int32, [4]> var_51895_begin_0 = const()[name = tensor<string, []>("op_51895_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_51895_end_0 = const()[name = tensor<string, []>("op_51895_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_51895_end_mask_0 = const()[name = tensor<string, []>("op_51895_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51895_cast = slice_by_index(begin = var_51895_begin_0, end = var_51895_end_0, end_mask = var_51895_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51895_cast")];
            tensor<int32, [4]> var_51899_begin_0 = const()[name = tensor<string, []>("op_51899_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_51899_end_0 = const()[name = tensor<string, []>("op_51899_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_51899_end_mask_0 = const()[name = tensor<string, []>("op_51899_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51899_cast = slice_by_index(begin = var_51899_begin_0, end = var_51899_end_0, end_mask = var_51899_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51899_cast")];
            tensor<int32, [4]> var_51903_begin_0 = const()[name = tensor<string, []>("op_51903_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_51903_end_0 = const()[name = tensor<string, []>("op_51903_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_51903_end_mask_0 = const()[name = tensor<string, []>("op_51903_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51903_cast = slice_by_index(begin = var_51903_begin_0, end = var_51903_end_0, end_mask = var_51903_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51903_cast")];
            tensor<int32, [4]> var_51907_begin_0 = const()[name = tensor<string, []>("op_51907_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_51907_end_0 = const()[name = tensor<string, []>("op_51907_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_51907_end_mask_0 = const()[name = tensor<string, []>("op_51907_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51907_cast = slice_by_index(begin = var_51907_begin_0, end = var_51907_end_0, end_mask = var_51907_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51907_cast")];
            tensor<int32, [4]> var_51911_begin_0 = const()[name = tensor<string, []>("op_51911_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_51911_end_0 = const()[name = tensor<string, []>("op_51911_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_51911_end_mask_0 = const()[name = tensor<string, []>("op_51911_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51911_cast = slice_by_index(begin = var_51911_begin_0, end = var_51911_end_0, end_mask = var_51911_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51911_cast")];
            tensor<int32, [4]> var_51915_begin_0 = const()[name = tensor<string, []>("op_51915_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_51915_end_0 = const()[name = tensor<string, []>("op_51915_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_51915_end_mask_0 = const()[name = tensor<string, []>("op_51915_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51915_cast = slice_by_index(begin = var_51915_begin_0, end = var_51915_end_0, end_mask = var_51915_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51915_cast")];
            tensor<int32, [4]> var_51919_begin_0 = const()[name = tensor<string, []>("op_51919_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_51919_end_0 = const()[name = tensor<string, []>("op_51919_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_51919_end_mask_0 = const()[name = tensor<string, []>("op_51919_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51919_cast = slice_by_index(begin = var_51919_begin_0, end = var_51919_end_0, end_mask = var_51919_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51919_cast")];
            tensor<int32, [4]> var_51923_begin_0 = const()[name = tensor<string, []>("op_51923_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_51923_end_0 = const()[name = tensor<string, []>("op_51923_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_51923_end_mask_0 = const()[name = tensor<string, []>("op_51923_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51923_cast = slice_by_index(begin = var_51923_begin_0, end = var_51923_end_0, end_mask = var_51923_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51923_cast")];
            tensor<int32, [4]> var_51927_begin_0 = const()[name = tensor<string, []>("op_51927_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_51927_end_0 = const()[name = tensor<string, []>("op_51927_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_51927_end_mask_0 = const()[name = tensor<string, []>("op_51927_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_51927_cast = slice_by_index(begin = var_51927_begin_0, end = var_51927_end_0, end_mask = var_51927_end_mask_0, x = v_229_cast)[name = tensor<string, []>("op_51927_cast")];
            tensor<string, []> var_51931_equation_0 = const()[name = tensor<string, []>("op_51931_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51931_cast = einsum(equation = var_51931_equation_0, values = (var_51773_cast, var_51690_cast))[name = tensor<string, []>("op_51931_cast")];
            tensor<fp16, []> var_51932_to_fp16 = const()[name = tensor<string, []>("op_51932_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4401_cast = mul(x = var_51931_cast, y = var_51932_to_fp16)[name = tensor<string, []>("aw_4401_cast")];
            tensor<string, []> var_51935_equation_0 = const()[name = tensor<string, []>("op_51935_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51935_cast = einsum(equation = var_51935_equation_0, values = (var_51777_cast, var_51694_cast))[name = tensor<string, []>("op_51935_cast")];
            tensor<fp16, []> var_51936_to_fp16 = const()[name = tensor<string, []>("op_51936_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4403_cast = mul(x = var_51935_cast, y = var_51936_to_fp16)[name = tensor<string, []>("aw_4403_cast")];
            tensor<string, []> var_51939_equation_0 = const()[name = tensor<string, []>("op_51939_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51939_cast = einsum(equation = var_51939_equation_0, values = (var_51781_cast, var_51698_cast))[name = tensor<string, []>("op_51939_cast")];
            tensor<fp16, []> var_51940_to_fp16 = const()[name = tensor<string, []>("op_51940_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4405_cast = mul(x = var_51939_cast, y = var_51940_to_fp16)[name = tensor<string, []>("aw_4405_cast")];
            tensor<string, []> var_51943_equation_0 = const()[name = tensor<string, []>("op_51943_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51943_cast = einsum(equation = var_51943_equation_0, values = (var_51785_cast, var_51702_cast))[name = tensor<string, []>("op_51943_cast")];
            tensor<fp16, []> var_51944_to_fp16 = const()[name = tensor<string, []>("op_51944_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4407_cast = mul(x = var_51943_cast, y = var_51944_to_fp16)[name = tensor<string, []>("aw_4407_cast")];
            tensor<string, []> var_51947_equation_0 = const()[name = tensor<string, []>("op_51947_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51947_cast = einsum(equation = var_51947_equation_0, values = (var_51789_cast, var_51706_cast))[name = tensor<string, []>("op_51947_cast")];
            tensor<fp16, []> var_51948_to_fp16 = const()[name = tensor<string, []>("op_51948_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4409_cast = mul(x = var_51947_cast, y = var_51948_to_fp16)[name = tensor<string, []>("aw_4409_cast")];
            tensor<string, []> var_51951_equation_0 = const()[name = tensor<string, []>("op_51951_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51951_cast = einsum(equation = var_51951_equation_0, values = (var_51793_cast, var_51710_cast))[name = tensor<string, []>("op_51951_cast")];
            tensor<fp16, []> var_51952_to_fp16 = const()[name = tensor<string, []>("op_51952_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4411_cast = mul(x = var_51951_cast, y = var_51952_to_fp16)[name = tensor<string, []>("aw_4411_cast")];
            tensor<string, []> var_51955_equation_0 = const()[name = tensor<string, []>("op_51955_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51955_cast = einsum(equation = var_51955_equation_0, values = (var_51797_cast, var_51714_cast))[name = tensor<string, []>("op_51955_cast")];
            tensor<fp16, []> var_51956_to_fp16 = const()[name = tensor<string, []>("op_51956_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4413_cast = mul(x = var_51955_cast, y = var_51956_to_fp16)[name = tensor<string, []>("aw_4413_cast")];
            tensor<string, []> var_51959_equation_0 = const()[name = tensor<string, []>("op_51959_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51959_cast = einsum(equation = var_51959_equation_0, values = (var_51801_cast, var_51718_cast))[name = tensor<string, []>("op_51959_cast")];
            tensor<fp16, []> var_51960_to_fp16 = const()[name = tensor<string, []>("op_51960_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4415_cast = mul(x = var_51959_cast, y = var_51960_to_fp16)[name = tensor<string, []>("aw_4415_cast")];
            tensor<string, []> var_51963_equation_0 = const()[name = tensor<string, []>("op_51963_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51963_cast = einsum(equation = var_51963_equation_0, values = (var_51805_cast, var_51722_cast))[name = tensor<string, []>("op_51963_cast")];
            tensor<fp16, []> var_51964_to_fp16 = const()[name = tensor<string, []>("op_51964_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4417_cast = mul(x = var_51963_cast, y = var_51964_to_fp16)[name = tensor<string, []>("aw_4417_cast")];
            tensor<string, []> var_51967_equation_0 = const()[name = tensor<string, []>("op_51967_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51967_cast = einsum(equation = var_51967_equation_0, values = (var_51809_cast, var_51726_cast))[name = tensor<string, []>("op_51967_cast")];
            tensor<fp16, []> var_51968_to_fp16 = const()[name = tensor<string, []>("op_51968_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4419_cast = mul(x = var_51967_cast, y = var_51968_to_fp16)[name = tensor<string, []>("aw_4419_cast")];
            tensor<string, []> var_51971_equation_0 = const()[name = tensor<string, []>("op_51971_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51971_cast = einsum(equation = var_51971_equation_0, values = (var_51813_cast, var_51730_cast))[name = tensor<string, []>("op_51971_cast")];
            tensor<fp16, []> var_51972_to_fp16 = const()[name = tensor<string, []>("op_51972_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4421_cast = mul(x = var_51971_cast, y = var_51972_to_fp16)[name = tensor<string, []>("aw_4421_cast")];
            tensor<string, []> var_51975_equation_0 = const()[name = tensor<string, []>("op_51975_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51975_cast = einsum(equation = var_51975_equation_0, values = (var_51817_cast, var_51734_cast))[name = tensor<string, []>("op_51975_cast")];
            tensor<fp16, []> var_51976_to_fp16 = const()[name = tensor<string, []>("op_51976_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4423_cast = mul(x = var_51975_cast, y = var_51976_to_fp16)[name = tensor<string, []>("aw_4423_cast")];
            tensor<string, []> var_51979_equation_0 = const()[name = tensor<string, []>("op_51979_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51979_cast = einsum(equation = var_51979_equation_0, values = (var_51821_cast, var_51738_cast))[name = tensor<string, []>("op_51979_cast")];
            tensor<fp16, []> var_51980_to_fp16 = const()[name = tensor<string, []>("op_51980_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4425_cast = mul(x = var_51979_cast, y = var_51980_to_fp16)[name = tensor<string, []>("aw_4425_cast")];
            tensor<string, []> var_51983_equation_0 = const()[name = tensor<string, []>("op_51983_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51983_cast = einsum(equation = var_51983_equation_0, values = (var_51825_cast, var_51742_cast))[name = tensor<string, []>("op_51983_cast")];
            tensor<fp16, []> var_51984_to_fp16 = const()[name = tensor<string, []>("op_51984_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4427_cast = mul(x = var_51983_cast, y = var_51984_to_fp16)[name = tensor<string, []>("aw_4427_cast")];
            tensor<string, []> var_51987_equation_0 = const()[name = tensor<string, []>("op_51987_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51987_cast = einsum(equation = var_51987_equation_0, values = (var_51829_cast, var_51746_cast))[name = tensor<string, []>("op_51987_cast")];
            tensor<fp16, []> var_51988_to_fp16 = const()[name = tensor<string, []>("op_51988_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4429_cast = mul(x = var_51987_cast, y = var_51988_to_fp16)[name = tensor<string, []>("aw_4429_cast")];
            tensor<string, []> var_51991_equation_0 = const()[name = tensor<string, []>("op_51991_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51991_cast = einsum(equation = var_51991_equation_0, values = (var_51833_cast, var_51750_cast))[name = tensor<string, []>("op_51991_cast")];
            tensor<fp16, []> var_51992_to_fp16 = const()[name = tensor<string, []>("op_51992_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4431_cast = mul(x = var_51991_cast, y = var_51992_to_fp16)[name = tensor<string, []>("aw_4431_cast")];
            tensor<string, []> var_51995_equation_0 = const()[name = tensor<string, []>("op_51995_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51995_cast = einsum(equation = var_51995_equation_0, values = (var_51837_cast, var_51754_cast))[name = tensor<string, []>("op_51995_cast")];
            tensor<fp16, []> var_51996_to_fp16 = const()[name = tensor<string, []>("op_51996_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4433_cast = mul(x = var_51995_cast, y = var_51996_to_fp16)[name = tensor<string, []>("aw_4433_cast")];
            tensor<string, []> var_51999_equation_0 = const()[name = tensor<string, []>("op_51999_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_51999_cast = einsum(equation = var_51999_equation_0, values = (var_51841_cast, var_51758_cast))[name = tensor<string, []>("op_51999_cast")];
            tensor<fp16, []> var_52000_to_fp16 = const()[name = tensor<string, []>("op_52000_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4435_cast = mul(x = var_51999_cast, y = var_52000_to_fp16)[name = tensor<string, []>("aw_4435_cast")];
            tensor<string, []> var_52003_equation_0 = const()[name = tensor<string, []>("op_52003_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52003_cast = einsum(equation = var_52003_equation_0, values = (var_51845_cast, var_51762_cast))[name = tensor<string, []>("op_52003_cast")];
            tensor<fp16, []> var_52004_to_fp16 = const()[name = tensor<string, []>("op_52004_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4437_cast = mul(x = var_52003_cast, y = var_52004_to_fp16)[name = tensor<string, []>("aw_4437_cast")];
            tensor<string, []> var_52007_equation_0 = const()[name = tensor<string, []>("op_52007_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52007_cast = einsum(equation = var_52007_equation_0, values = (var_51849_cast, var_51766_cast))[name = tensor<string, []>("op_52007_cast")];
            tensor<fp16, []> var_52008_to_fp16 = const()[name = tensor<string, []>("op_52008_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4439_cast = mul(x = var_52007_cast, y = var_52008_to_fp16)[name = tensor<string, []>("aw_4439_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52010_cast = softmax(axis = var_30385, x = aw_4401_cast)[name = tensor<string, []>("op_52010_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52011_cast = softmax(axis = var_30385, x = aw_4403_cast)[name = tensor<string, []>("op_52011_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52012_cast = softmax(axis = var_30385, x = aw_4405_cast)[name = tensor<string, []>("op_52012_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52013_cast = softmax(axis = var_30385, x = aw_4407_cast)[name = tensor<string, []>("op_52013_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52014_cast = softmax(axis = var_30385, x = aw_4409_cast)[name = tensor<string, []>("op_52014_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52015_cast = softmax(axis = var_30385, x = aw_4411_cast)[name = tensor<string, []>("op_52015_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52016_cast = softmax(axis = var_30385, x = aw_4413_cast)[name = tensor<string, []>("op_52016_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52017_cast = softmax(axis = var_30385, x = aw_4415_cast)[name = tensor<string, []>("op_52017_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52018_cast = softmax(axis = var_30385, x = aw_4417_cast)[name = tensor<string, []>("op_52018_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52019_cast = softmax(axis = var_30385, x = aw_4419_cast)[name = tensor<string, []>("op_52019_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52020_cast = softmax(axis = var_30385, x = aw_4421_cast)[name = tensor<string, []>("op_52020_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52021_cast = softmax(axis = var_30385, x = aw_4423_cast)[name = tensor<string, []>("op_52021_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52022_cast = softmax(axis = var_30385, x = aw_4425_cast)[name = tensor<string, []>("op_52022_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52023_cast = softmax(axis = var_30385, x = aw_4427_cast)[name = tensor<string, []>("op_52023_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52024_cast = softmax(axis = var_30385, x = aw_4429_cast)[name = tensor<string, []>("op_52024_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52025_cast = softmax(axis = var_30385, x = aw_4431_cast)[name = tensor<string, []>("op_52025_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52026_cast = softmax(axis = var_30385, x = aw_4433_cast)[name = tensor<string, []>("op_52026_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52027_cast = softmax(axis = var_30385, x = aw_4435_cast)[name = tensor<string, []>("op_52027_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52028_cast = softmax(axis = var_30385, x = aw_4437_cast)[name = tensor<string, []>("op_52028_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52029_cast = softmax(axis = var_30385, x = aw_4439_cast)[name = tensor<string, []>("op_52029_cast")];
            tensor<string, []> var_52031_equation_0 = const()[name = tensor<string, []>("op_52031_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52031_cast = einsum(equation = var_52031_equation_0, values = (var_51851_cast, var_52010_cast))[name = tensor<string, []>("op_52031_cast")];
            tensor<string, []> var_52033_equation_0 = const()[name = tensor<string, []>("op_52033_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52033_cast = einsum(equation = var_52033_equation_0, values = (var_51855_cast, var_52011_cast))[name = tensor<string, []>("op_52033_cast")];
            tensor<string, []> var_52035_equation_0 = const()[name = tensor<string, []>("op_52035_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52035_cast = einsum(equation = var_52035_equation_0, values = (var_51859_cast, var_52012_cast))[name = tensor<string, []>("op_52035_cast")];
            tensor<string, []> var_52037_equation_0 = const()[name = tensor<string, []>("op_52037_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52037_cast = einsum(equation = var_52037_equation_0, values = (var_51863_cast, var_52013_cast))[name = tensor<string, []>("op_52037_cast")];
            tensor<string, []> var_52039_equation_0 = const()[name = tensor<string, []>("op_52039_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52039_cast = einsum(equation = var_52039_equation_0, values = (var_51867_cast, var_52014_cast))[name = tensor<string, []>("op_52039_cast")];
            tensor<string, []> var_52041_equation_0 = const()[name = tensor<string, []>("op_52041_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52041_cast = einsum(equation = var_52041_equation_0, values = (var_51871_cast, var_52015_cast))[name = tensor<string, []>("op_52041_cast")];
            tensor<string, []> var_52043_equation_0 = const()[name = tensor<string, []>("op_52043_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52043_cast = einsum(equation = var_52043_equation_0, values = (var_51875_cast, var_52016_cast))[name = tensor<string, []>("op_52043_cast")];
            tensor<string, []> var_52045_equation_0 = const()[name = tensor<string, []>("op_52045_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52045_cast = einsum(equation = var_52045_equation_0, values = (var_51879_cast, var_52017_cast))[name = tensor<string, []>("op_52045_cast")];
            tensor<string, []> var_52047_equation_0 = const()[name = tensor<string, []>("op_52047_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52047_cast = einsum(equation = var_52047_equation_0, values = (var_51883_cast, var_52018_cast))[name = tensor<string, []>("op_52047_cast")];
            tensor<string, []> var_52049_equation_0 = const()[name = tensor<string, []>("op_52049_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52049_cast = einsum(equation = var_52049_equation_0, values = (var_51887_cast, var_52019_cast))[name = tensor<string, []>("op_52049_cast")];
            tensor<string, []> var_52051_equation_0 = const()[name = tensor<string, []>("op_52051_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52051_cast = einsum(equation = var_52051_equation_0, values = (var_51891_cast, var_52020_cast))[name = tensor<string, []>("op_52051_cast")];
            tensor<string, []> var_52053_equation_0 = const()[name = tensor<string, []>("op_52053_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52053_cast = einsum(equation = var_52053_equation_0, values = (var_51895_cast, var_52021_cast))[name = tensor<string, []>("op_52053_cast")];
            tensor<string, []> var_52055_equation_0 = const()[name = tensor<string, []>("op_52055_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52055_cast = einsum(equation = var_52055_equation_0, values = (var_51899_cast, var_52022_cast))[name = tensor<string, []>("op_52055_cast")];
            tensor<string, []> var_52057_equation_0 = const()[name = tensor<string, []>("op_52057_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52057_cast = einsum(equation = var_52057_equation_0, values = (var_51903_cast, var_52023_cast))[name = tensor<string, []>("op_52057_cast")];
            tensor<string, []> var_52059_equation_0 = const()[name = tensor<string, []>("op_52059_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52059_cast = einsum(equation = var_52059_equation_0, values = (var_51907_cast, var_52024_cast))[name = tensor<string, []>("op_52059_cast")];
            tensor<string, []> var_52061_equation_0 = const()[name = tensor<string, []>("op_52061_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52061_cast = einsum(equation = var_52061_equation_0, values = (var_51911_cast, var_52025_cast))[name = tensor<string, []>("op_52061_cast")];
            tensor<string, []> var_52063_equation_0 = const()[name = tensor<string, []>("op_52063_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52063_cast = einsum(equation = var_52063_equation_0, values = (var_51915_cast, var_52026_cast))[name = tensor<string, []>("op_52063_cast")];
            tensor<string, []> var_52065_equation_0 = const()[name = tensor<string, []>("op_52065_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52065_cast = einsum(equation = var_52065_equation_0, values = (var_51919_cast, var_52027_cast))[name = tensor<string, []>("op_52065_cast")];
            tensor<string, []> var_52067_equation_0 = const()[name = tensor<string, []>("op_52067_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52067_cast = einsum(equation = var_52067_equation_0, values = (var_51923_cast, var_52028_cast))[name = tensor<string, []>("op_52067_cast")];
            tensor<string, []> var_52069_equation_0 = const()[name = tensor<string, []>("op_52069_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52069_cast = einsum(equation = var_52069_equation_0, values = (var_51927_cast, var_52029_cast))[name = tensor<string, []>("op_52069_cast")];
            tensor<bool, []> input_661_interleave_0 = const()[name = tensor<string, []>("input_661_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_661_cast = concat(axis = var_30385, interleave = input_661_interleave_0, values = (var_52031_cast, var_52033_cast, var_52035_cast, var_52037_cast, var_52039_cast, var_52041_cast, var_52043_cast, var_52045_cast, var_52047_cast, var_52049_cast, var_52051_cast, var_52053_cast, var_52055_cast, var_52057_cast, var_52059_cast, var_52061_cast, var_52063_cast, var_52065_cast, var_52067_cast, var_52069_cast))[name = tensor<string, []>("input_661_cast")];
            tensor<int32, [2]> var_52075 = const()[name = tensor<string, []>("op_52075"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_52077 = const()[name = tensor<string, []>("op_52077"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_52079_pad_type_0 = const()[name = tensor<string, []>("op_52079_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_52079_pad_0 = const()[name = tensor<string, []>("op_52079_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4390465728)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4393742592)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_52079_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_out_0_bias_to_fp16, dilations = var_52077, groups = var_30385, pad = var_52079_pad_0, pad_type = var_52079_pad_type_0, strides = var_52075, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn1_to_out_0_weight_to_fp16, x = input_661_cast)[name = tensor<string, []>("op_52079_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_345_cast = add(x = var_52079_cast, y = inputs_343_cast)[name = tensor<string, []>("inputs_345_cast")];
            tensor<int32, [1]> var_52083 = const()[name = tensor<string, []>("op_52083"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_345_cast = reduce_mean(axes = var_52083, keep_dims = var_30380, x = inputs_345_cast)[name = tensor<string, []>("channels_mean_345_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_345_cast = sub(x = inputs_345_cast, y = channels_mean_345_cast)[name = tensor<string, []>("zero_mean_345_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_345_cast = mul(x = zero_mean_345_cast, y = zero_mean_345_cast)[name = tensor<string, []>("zero_mean_sq_345_cast")];
            tensor<int32, [1]> var_52087 = const()[name = tensor<string, []>("op_52087"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_52088_cast = reduce_mean(axes = var_52087, keep_dims = var_30380, x = zero_mean_sq_345_cast)[name = tensor<string, []>("op_52088_cast")];
            tensor<fp16, []> var_52089_to_fp16 = const()[name = tensor<string, []>("op_52089_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_52090_cast = add(x = var_52088_cast, y = var_52089_to_fp16)[name = tensor<string, []>("op_52090_cast")];
            tensor<fp16, []> denom_345_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_345_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_345_cast = rsqrt(epsilon = denom_345_epsilon_0_to_fp16, x = var_52090_cast)[name = tensor<string, []>("denom_345_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_345_cast = mul(x = zero_mean_345_cast, y = denom_345_cast)[name = tensor<string, []>("out_345_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_52094_to_fp16 = const()[name = tensor<string, []>("op_52094_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4393745216)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_52095_cast = add(x = out_345_cast, y = var_52094_to_fp16)[name = tensor<string, []>("op_52095_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_52097_to_fp16 = const()[name = tensor<string, []>("op_52097_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4393747840)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_455_cast = mul(x = var_52095_cast, y = var_52097_to_fp16)[name = tensor<string, []>("hidden_states_455_cast")];
            tensor<int32, [2]> var_52104 = const()[name = tensor<string, []>("op_52104"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_52106 = const()[name = tensor<string, []>("op_52106"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_231_pad_type_0 = const()[name = tensor<string, []>("q_231_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_231_pad_0 = const()[name = tensor<string, []>("q_231_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4393750464)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_231_cast = conv(dilations = var_52106, groups = var_30385, pad = q_231_pad_0, pad_type = q_231_pad_type_0, strides = var_52104, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_q_weight_to_fp16, x = hidden_states_455_cast)[name = tensor<string, []>("q_231_cast")];
            tensor<int32, [2]> var_52110 = const()[name = tensor<string, []>("op_52110"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_52112 = const()[name = tensor<string, []>("op_52112"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_461_pad_type_0 = const()[name = tensor<string, []>("k_461_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_461_pad_0 = const()[name = tensor<string, []>("k_461_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4397027328)))];
            tensor<fp16, [2, 1280, 1, 77]> k_461_cast = conv(dilations = var_52112, groups = var_30385, pad = k_461_pad_0, pad_type = k_461_pad_type_0, strides = var_52110, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_461_cast")];
            tensor<int32, [2]> var_52116 = const()[name = tensor<string, []>("op_52116"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_52118 = const()[name = tensor<string, []>("op_52118"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_231_pad_type_0 = const()[name = tensor<string, []>("v_231_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_231_pad_0 = const()[name = tensor<string, []>("v_231_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4402270272)))];
            tensor<fp16, [2, 1280, 1, 77]> v_231_cast = conv(dilations = var_52118, groups = var_30385, pad = v_231_pad_0, pad_type = v_231_pad_type_0, strides = var_52116, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_231_cast")];
            tensor<int32, [4]> var_52122_begin_0 = const()[name = tensor<string, []>("op_52122_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_52122_end_0 = const()[name = tensor<string, []>("op_52122_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_52122_end_mask_0 = const()[name = tensor<string, []>("op_52122_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52122_cast = slice_by_index(begin = var_52122_begin_0, end = var_52122_end_0, end_mask = var_52122_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52122_cast")];
            tensor<int32, [4]> var_52126_begin_0 = const()[name = tensor<string, []>("op_52126_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_52126_end_0 = const()[name = tensor<string, []>("op_52126_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_52126_end_mask_0 = const()[name = tensor<string, []>("op_52126_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52126_cast = slice_by_index(begin = var_52126_begin_0, end = var_52126_end_0, end_mask = var_52126_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52126_cast")];
            tensor<int32, [4]> var_52130_begin_0 = const()[name = tensor<string, []>("op_52130_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_52130_end_0 = const()[name = tensor<string, []>("op_52130_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_52130_end_mask_0 = const()[name = tensor<string, []>("op_52130_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52130_cast = slice_by_index(begin = var_52130_begin_0, end = var_52130_end_0, end_mask = var_52130_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52130_cast")];
            tensor<int32, [4]> var_52134_begin_0 = const()[name = tensor<string, []>("op_52134_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_52134_end_0 = const()[name = tensor<string, []>("op_52134_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_52134_end_mask_0 = const()[name = tensor<string, []>("op_52134_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52134_cast = slice_by_index(begin = var_52134_begin_0, end = var_52134_end_0, end_mask = var_52134_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52134_cast")];
            tensor<int32, [4]> var_52138_begin_0 = const()[name = tensor<string, []>("op_52138_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_52138_end_0 = const()[name = tensor<string, []>("op_52138_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_52138_end_mask_0 = const()[name = tensor<string, []>("op_52138_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52138_cast = slice_by_index(begin = var_52138_begin_0, end = var_52138_end_0, end_mask = var_52138_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52138_cast")];
            tensor<int32, [4]> var_52142_begin_0 = const()[name = tensor<string, []>("op_52142_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_52142_end_0 = const()[name = tensor<string, []>("op_52142_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_52142_end_mask_0 = const()[name = tensor<string, []>("op_52142_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52142_cast = slice_by_index(begin = var_52142_begin_0, end = var_52142_end_0, end_mask = var_52142_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52142_cast")];
            tensor<int32, [4]> var_52146_begin_0 = const()[name = tensor<string, []>("op_52146_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_52146_end_0 = const()[name = tensor<string, []>("op_52146_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_52146_end_mask_0 = const()[name = tensor<string, []>("op_52146_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52146_cast = slice_by_index(begin = var_52146_begin_0, end = var_52146_end_0, end_mask = var_52146_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52146_cast")];
            tensor<int32, [4]> var_52150_begin_0 = const()[name = tensor<string, []>("op_52150_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_52150_end_0 = const()[name = tensor<string, []>("op_52150_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_52150_end_mask_0 = const()[name = tensor<string, []>("op_52150_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52150_cast = slice_by_index(begin = var_52150_begin_0, end = var_52150_end_0, end_mask = var_52150_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52150_cast")];
            tensor<int32, [4]> var_52154_begin_0 = const()[name = tensor<string, []>("op_52154_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_52154_end_0 = const()[name = tensor<string, []>("op_52154_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_52154_end_mask_0 = const()[name = tensor<string, []>("op_52154_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52154_cast = slice_by_index(begin = var_52154_begin_0, end = var_52154_end_0, end_mask = var_52154_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52154_cast")];
            tensor<int32, [4]> var_52158_begin_0 = const()[name = tensor<string, []>("op_52158_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_52158_end_0 = const()[name = tensor<string, []>("op_52158_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_52158_end_mask_0 = const()[name = tensor<string, []>("op_52158_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52158_cast = slice_by_index(begin = var_52158_begin_0, end = var_52158_end_0, end_mask = var_52158_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52158_cast")];
            tensor<int32, [4]> var_52162_begin_0 = const()[name = tensor<string, []>("op_52162_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_52162_end_0 = const()[name = tensor<string, []>("op_52162_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_52162_end_mask_0 = const()[name = tensor<string, []>("op_52162_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52162_cast = slice_by_index(begin = var_52162_begin_0, end = var_52162_end_0, end_mask = var_52162_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52162_cast")];
            tensor<int32, [4]> var_52166_begin_0 = const()[name = tensor<string, []>("op_52166_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_52166_end_0 = const()[name = tensor<string, []>("op_52166_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_52166_end_mask_0 = const()[name = tensor<string, []>("op_52166_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52166_cast = slice_by_index(begin = var_52166_begin_0, end = var_52166_end_0, end_mask = var_52166_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52166_cast")];
            tensor<int32, [4]> var_52170_begin_0 = const()[name = tensor<string, []>("op_52170_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_52170_end_0 = const()[name = tensor<string, []>("op_52170_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_52170_end_mask_0 = const()[name = tensor<string, []>("op_52170_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52170_cast = slice_by_index(begin = var_52170_begin_0, end = var_52170_end_0, end_mask = var_52170_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52170_cast")];
            tensor<int32, [4]> var_52174_begin_0 = const()[name = tensor<string, []>("op_52174_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_52174_end_0 = const()[name = tensor<string, []>("op_52174_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_52174_end_mask_0 = const()[name = tensor<string, []>("op_52174_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52174_cast = slice_by_index(begin = var_52174_begin_0, end = var_52174_end_0, end_mask = var_52174_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52174_cast")];
            tensor<int32, [4]> var_52178_begin_0 = const()[name = tensor<string, []>("op_52178_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_52178_end_0 = const()[name = tensor<string, []>("op_52178_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_52178_end_mask_0 = const()[name = tensor<string, []>("op_52178_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52178_cast = slice_by_index(begin = var_52178_begin_0, end = var_52178_end_0, end_mask = var_52178_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52178_cast")];
            tensor<int32, [4]> var_52182_begin_0 = const()[name = tensor<string, []>("op_52182_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_52182_end_0 = const()[name = tensor<string, []>("op_52182_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_52182_end_mask_0 = const()[name = tensor<string, []>("op_52182_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52182_cast = slice_by_index(begin = var_52182_begin_0, end = var_52182_end_0, end_mask = var_52182_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52182_cast")];
            tensor<int32, [4]> var_52186_begin_0 = const()[name = tensor<string, []>("op_52186_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_52186_end_0 = const()[name = tensor<string, []>("op_52186_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_52186_end_mask_0 = const()[name = tensor<string, []>("op_52186_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52186_cast = slice_by_index(begin = var_52186_begin_0, end = var_52186_end_0, end_mask = var_52186_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52186_cast")];
            tensor<int32, [4]> var_52190_begin_0 = const()[name = tensor<string, []>("op_52190_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_52190_end_0 = const()[name = tensor<string, []>("op_52190_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_52190_end_mask_0 = const()[name = tensor<string, []>("op_52190_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52190_cast = slice_by_index(begin = var_52190_begin_0, end = var_52190_end_0, end_mask = var_52190_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52190_cast")];
            tensor<int32, [4]> var_52194_begin_0 = const()[name = tensor<string, []>("op_52194_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_52194_end_0 = const()[name = tensor<string, []>("op_52194_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_52194_end_mask_0 = const()[name = tensor<string, []>("op_52194_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52194_cast = slice_by_index(begin = var_52194_begin_0, end = var_52194_end_0, end_mask = var_52194_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52194_cast")];
            tensor<int32, [4]> var_52198_begin_0 = const()[name = tensor<string, []>("op_52198_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_52198_end_0 = const()[name = tensor<string, []>("op_52198_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_52198_end_mask_0 = const()[name = tensor<string, []>("op_52198_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52198_cast = slice_by_index(begin = var_52198_begin_0, end = var_52198_end_0, end_mask = var_52198_end_mask_0, x = q_231_cast)[name = tensor<string, []>("op_52198_cast")];
            tensor<int32, [4]> k_463_perm_0 = const()[name = tensor<string, []>("k_463_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_52205_begin_0 = const()[name = tensor<string, []>("op_52205_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_52205_end_0 = const()[name = tensor<string, []>("op_52205_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_52205_end_mask_0 = const()[name = tensor<string, []>("op_52205_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_24 = transpose(perm = k_463_perm_0, x = k_461_cast)[name = tensor<string, []>("transpose_24")];
            tensor<fp16, [2, 77, 1, 64]> var_52205_cast = slice_by_index(begin = var_52205_begin_0, end = var_52205_end_0, end_mask = var_52205_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52205_cast")];
            tensor<int32, [4]> var_52209_begin_0 = const()[name = tensor<string, []>("op_52209_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_52209_end_0 = const()[name = tensor<string, []>("op_52209_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_52209_end_mask_0 = const()[name = tensor<string, []>("op_52209_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52209_cast = slice_by_index(begin = var_52209_begin_0, end = var_52209_end_0, end_mask = var_52209_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52209_cast")];
            tensor<int32, [4]> var_52213_begin_0 = const()[name = tensor<string, []>("op_52213_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_52213_end_0 = const()[name = tensor<string, []>("op_52213_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_52213_end_mask_0 = const()[name = tensor<string, []>("op_52213_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52213_cast = slice_by_index(begin = var_52213_begin_0, end = var_52213_end_0, end_mask = var_52213_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52213_cast")];
            tensor<int32, [4]> var_52217_begin_0 = const()[name = tensor<string, []>("op_52217_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_52217_end_0 = const()[name = tensor<string, []>("op_52217_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_52217_end_mask_0 = const()[name = tensor<string, []>("op_52217_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52217_cast = slice_by_index(begin = var_52217_begin_0, end = var_52217_end_0, end_mask = var_52217_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52217_cast")];
            tensor<int32, [4]> var_52221_begin_0 = const()[name = tensor<string, []>("op_52221_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_52221_end_0 = const()[name = tensor<string, []>("op_52221_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_52221_end_mask_0 = const()[name = tensor<string, []>("op_52221_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52221_cast = slice_by_index(begin = var_52221_begin_0, end = var_52221_end_0, end_mask = var_52221_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52221_cast")];
            tensor<int32, [4]> var_52225_begin_0 = const()[name = tensor<string, []>("op_52225_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_52225_end_0 = const()[name = tensor<string, []>("op_52225_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_52225_end_mask_0 = const()[name = tensor<string, []>("op_52225_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52225_cast = slice_by_index(begin = var_52225_begin_0, end = var_52225_end_0, end_mask = var_52225_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52225_cast")];
            tensor<int32, [4]> var_52229_begin_0 = const()[name = tensor<string, []>("op_52229_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_52229_end_0 = const()[name = tensor<string, []>("op_52229_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_52229_end_mask_0 = const()[name = tensor<string, []>("op_52229_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52229_cast = slice_by_index(begin = var_52229_begin_0, end = var_52229_end_0, end_mask = var_52229_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52229_cast")];
            tensor<int32, [4]> var_52233_begin_0 = const()[name = tensor<string, []>("op_52233_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_52233_end_0 = const()[name = tensor<string, []>("op_52233_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_52233_end_mask_0 = const()[name = tensor<string, []>("op_52233_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52233_cast = slice_by_index(begin = var_52233_begin_0, end = var_52233_end_0, end_mask = var_52233_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52233_cast")];
            tensor<int32, [4]> var_52237_begin_0 = const()[name = tensor<string, []>("op_52237_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_52237_end_0 = const()[name = tensor<string, []>("op_52237_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_52237_end_mask_0 = const()[name = tensor<string, []>("op_52237_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52237_cast = slice_by_index(begin = var_52237_begin_0, end = var_52237_end_0, end_mask = var_52237_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52237_cast")];
            tensor<int32, [4]> var_52241_begin_0 = const()[name = tensor<string, []>("op_52241_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_52241_end_0 = const()[name = tensor<string, []>("op_52241_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_52241_end_mask_0 = const()[name = tensor<string, []>("op_52241_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52241_cast = slice_by_index(begin = var_52241_begin_0, end = var_52241_end_0, end_mask = var_52241_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52241_cast")];
            tensor<int32, [4]> var_52245_begin_0 = const()[name = tensor<string, []>("op_52245_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_52245_end_0 = const()[name = tensor<string, []>("op_52245_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_52245_end_mask_0 = const()[name = tensor<string, []>("op_52245_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52245_cast = slice_by_index(begin = var_52245_begin_0, end = var_52245_end_0, end_mask = var_52245_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52245_cast")];
            tensor<int32, [4]> var_52249_begin_0 = const()[name = tensor<string, []>("op_52249_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_52249_end_0 = const()[name = tensor<string, []>("op_52249_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_52249_end_mask_0 = const()[name = tensor<string, []>("op_52249_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52249_cast = slice_by_index(begin = var_52249_begin_0, end = var_52249_end_0, end_mask = var_52249_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52249_cast")];
            tensor<int32, [4]> var_52253_begin_0 = const()[name = tensor<string, []>("op_52253_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_52253_end_0 = const()[name = tensor<string, []>("op_52253_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_52253_end_mask_0 = const()[name = tensor<string, []>("op_52253_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52253_cast = slice_by_index(begin = var_52253_begin_0, end = var_52253_end_0, end_mask = var_52253_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52253_cast")];
            tensor<int32, [4]> var_52257_begin_0 = const()[name = tensor<string, []>("op_52257_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_52257_end_0 = const()[name = tensor<string, []>("op_52257_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_52257_end_mask_0 = const()[name = tensor<string, []>("op_52257_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52257_cast = slice_by_index(begin = var_52257_begin_0, end = var_52257_end_0, end_mask = var_52257_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52257_cast")];
            tensor<int32, [4]> var_52261_begin_0 = const()[name = tensor<string, []>("op_52261_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_52261_end_0 = const()[name = tensor<string, []>("op_52261_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_52261_end_mask_0 = const()[name = tensor<string, []>("op_52261_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52261_cast = slice_by_index(begin = var_52261_begin_0, end = var_52261_end_0, end_mask = var_52261_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52261_cast")];
            tensor<int32, [4]> var_52265_begin_0 = const()[name = tensor<string, []>("op_52265_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_52265_end_0 = const()[name = tensor<string, []>("op_52265_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_52265_end_mask_0 = const()[name = tensor<string, []>("op_52265_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52265_cast = slice_by_index(begin = var_52265_begin_0, end = var_52265_end_0, end_mask = var_52265_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52265_cast")];
            tensor<int32, [4]> var_52269_begin_0 = const()[name = tensor<string, []>("op_52269_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_52269_end_0 = const()[name = tensor<string, []>("op_52269_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_52269_end_mask_0 = const()[name = tensor<string, []>("op_52269_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52269_cast = slice_by_index(begin = var_52269_begin_0, end = var_52269_end_0, end_mask = var_52269_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52269_cast")];
            tensor<int32, [4]> var_52273_begin_0 = const()[name = tensor<string, []>("op_52273_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_52273_end_0 = const()[name = tensor<string, []>("op_52273_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_52273_end_mask_0 = const()[name = tensor<string, []>("op_52273_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52273_cast = slice_by_index(begin = var_52273_begin_0, end = var_52273_end_0, end_mask = var_52273_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52273_cast")];
            tensor<int32, [4]> var_52277_begin_0 = const()[name = tensor<string, []>("op_52277_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_52277_end_0 = const()[name = tensor<string, []>("op_52277_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_52277_end_mask_0 = const()[name = tensor<string, []>("op_52277_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52277_cast = slice_by_index(begin = var_52277_begin_0, end = var_52277_end_0, end_mask = var_52277_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52277_cast")];
            tensor<int32, [4]> var_52281_begin_0 = const()[name = tensor<string, []>("op_52281_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_52281_end_0 = const()[name = tensor<string, []>("op_52281_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_52281_end_mask_0 = const()[name = tensor<string, []>("op_52281_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_52281_cast = slice_by_index(begin = var_52281_begin_0, end = var_52281_end_0, end_mask = var_52281_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_52281_cast")];
            tensor<int32, [4]> var_52283_begin_0 = const()[name = tensor<string, []>("op_52283_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_52283_end_0 = const()[name = tensor<string, []>("op_52283_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_52283_end_mask_0 = const()[name = tensor<string, []>("op_52283_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52283_cast = slice_by_index(begin = var_52283_begin_0, end = var_52283_end_0, end_mask = var_52283_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52283_cast")];
            tensor<int32, [4]> var_52287_begin_0 = const()[name = tensor<string, []>("op_52287_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_52287_end_0 = const()[name = tensor<string, []>("op_52287_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_52287_end_mask_0 = const()[name = tensor<string, []>("op_52287_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52287_cast = slice_by_index(begin = var_52287_begin_0, end = var_52287_end_0, end_mask = var_52287_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52287_cast")];
            tensor<int32, [4]> var_52291_begin_0 = const()[name = tensor<string, []>("op_52291_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_52291_end_0 = const()[name = tensor<string, []>("op_52291_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_52291_end_mask_0 = const()[name = tensor<string, []>("op_52291_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52291_cast = slice_by_index(begin = var_52291_begin_0, end = var_52291_end_0, end_mask = var_52291_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52291_cast")];
            tensor<int32, [4]> var_52295_begin_0 = const()[name = tensor<string, []>("op_52295_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_52295_end_0 = const()[name = tensor<string, []>("op_52295_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_52295_end_mask_0 = const()[name = tensor<string, []>("op_52295_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52295_cast = slice_by_index(begin = var_52295_begin_0, end = var_52295_end_0, end_mask = var_52295_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52295_cast")];
            tensor<int32, [4]> var_52299_begin_0 = const()[name = tensor<string, []>("op_52299_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_52299_end_0 = const()[name = tensor<string, []>("op_52299_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_52299_end_mask_0 = const()[name = tensor<string, []>("op_52299_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52299_cast = slice_by_index(begin = var_52299_begin_0, end = var_52299_end_0, end_mask = var_52299_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52299_cast")];
            tensor<int32, [4]> var_52303_begin_0 = const()[name = tensor<string, []>("op_52303_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_52303_end_0 = const()[name = tensor<string, []>("op_52303_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_52303_end_mask_0 = const()[name = tensor<string, []>("op_52303_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52303_cast = slice_by_index(begin = var_52303_begin_0, end = var_52303_end_0, end_mask = var_52303_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52303_cast")];
            tensor<int32, [4]> var_52307_begin_0 = const()[name = tensor<string, []>("op_52307_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_52307_end_0 = const()[name = tensor<string, []>("op_52307_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_52307_end_mask_0 = const()[name = tensor<string, []>("op_52307_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52307_cast = slice_by_index(begin = var_52307_begin_0, end = var_52307_end_0, end_mask = var_52307_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52307_cast")];
            tensor<int32, [4]> var_52311_begin_0 = const()[name = tensor<string, []>("op_52311_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_52311_end_0 = const()[name = tensor<string, []>("op_52311_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_52311_end_mask_0 = const()[name = tensor<string, []>("op_52311_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52311_cast = slice_by_index(begin = var_52311_begin_0, end = var_52311_end_0, end_mask = var_52311_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52311_cast")];
            tensor<int32, [4]> var_52315_begin_0 = const()[name = tensor<string, []>("op_52315_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_52315_end_0 = const()[name = tensor<string, []>("op_52315_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_52315_end_mask_0 = const()[name = tensor<string, []>("op_52315_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52315_cast = slice_by_index(begin = var_52315_begin_0, end = var_52315_end_0, end_mask = var_52315_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52315_cast")];
            tensor<int32, [4]> var_52319_begin_0 = const()[name = tensor<string, []>("op_52319_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_52319_end_0 = const()[name = tensor<string, []>("op_52319_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_52319_end_mask_0 = const()[name = tensor<string, []>("op_52319_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52319_cast = slice_by_index(begin = var_52319_begin_0, end = var_52319_end_0, end_mask = var_52319_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52319_cast")];
            tensor<int32, [4]> var_52323_begin_0 = const()[name = tensor<string, []>("op_52323_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_52323_end_0 = const()[name = tensor<string, []>("op_52323_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_52323_end_mask_0 = const()[name = tensor<string, []>("op_52323_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52323_cast = slice_by_index(begin = var_52323_begin_0, end = var_52323_end_0, end_mask = var_52323_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52323_cast")];
            tensor<int32, [4]> var_52327_begin_0 = const()[name = tensor<string, []>("op_52327_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_52327_end_0 = const()[name = tensor<string, []>("op_52327_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_52327_end_mask_0 = const()[name = tensor<string, []>("op_52327_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52327_cast = slice_by_index(begin = var_52327_begin_0, end = var_52327_end_0, end_mask = var_52327_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52327_cast")];
            tensor<int32, [4]> var_52331_begin_0 = const()[name = tensor<string, []>("op_52331_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_52331_end_0 = const()[name = tensor<string, []>("op_52331_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_52331_end_mask_0 = const()[name = tensor<string, []>("op_52331_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52331_cast = slice_by_index(begin = var_52331_begin_0, end = var_52331_end_0, end_mask = var_52331_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52331_cast")];
            tensor<int32, [4]> var_52335_begin_0 = const()[name = tensor<string, []>("op_52335_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_52335_end_0 = const()[name = tensor<string, []>("op_52335_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_52335_end_mask_0 = const()[name = tensor<string, []>("op_52335_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52335_cast = slice_by_index(begin = var_52335_begin_0, end = var_52335_end_0, end_mask = var_52335_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52335_cast")];
            tensor<int32, [4]> var_52339_begin_0 = const()[name = tensor<string, []>("op_52339_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_52339_end_0 = const()[name = tensor<string, []>("op_52339_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_52339_end_mask_0 = const()[name = tensor<string, []>("op_52339_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52339_cast = slice_by_index(begin = var_52339_begin_0, end = var_52339_end_0, end_mask = var_52339_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52339_cast")];
            tensor<int32, [4]> var_52343_begin_0 = const()[name = tensor<string, []>("op_52343_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_52343_end_0 = const()[name = tensor<string, []>("op_52343_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_52343_end_mask_0 = const()[name = tensor<string, []>("op_52343_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52343_cast = slice_by_index(begin = var_52343_begin_0, end = var_52343_end_0, end_mask = var_52343_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52343_cast")];
            tensor<int32, [4]> var_52347_begin_0 = const()[name = tensor<string, []>("op_52347_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_52347_end_0 = const()[name = tensor<string, []>("op_52347_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_52347_end_mask_0 = const()[name = tensor<string, []>("op_52347_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52347_cast = slice_by_index(begin = var_52347_begin_0, end = var_52347_end_0, end_mask = var_52347_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52347_cast")];
            tensor<int32, [4]> var_52351_begin_0 = const()[name = tensor<string, []>("op_52351_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_52351_end_0 = const()[name = tensor<string, []>("op_52351_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_52351_end_mask_0 = const()[name = tensor<string, []>("op_52351_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52351_cast = slice_by_index(begin = var_52351_begin_0, end = var_52351_end_0, end_mask = var_52351_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52351_cast")];
            tensor<int32, [4]> var_52355_begin_0 = const()[name = tensor<string, []>("op_52355_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_52355_end_0 = const()[name = tensor<string, []>("op_52355_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_52355_end_mask_0 = const()[name = tensor<string, []>("op_52355_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52355_cast = slice_by_index(begin = var_52355_begin_0, end = var_52355_end_0, end_mask = var_52355_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52355_cast")];
            tensor<int32, [4]> var_52359_begin_0 = const()[name = tensor<string, []>("op_52359_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_52359_end_0 = const()[name = tensor<string, []>("op_52359_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_52359_end_mask_0 = const()[name = tensor<string, []>("op_52359_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_52359_cast = slice_by_index(begin = var_52359_begin_0, end = var_52359_end_0, end_mask = var_52359_end_mask_0, x = v_231_cast)[name = tensor<string, []>("op_52359_cast")];
            tensor<string, []> var_52363_equation_0 = const()[name = tensor<string, []>("op_52363_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52363_cast = einsum(equation = var_52363_equation_0, values = (var_52205_cast, var_52122_cast))[name = tensor<string, []>("op_52363_cast")];
            tensor<fp16, []> var_52364_to_fp16 = const()[name = tensor<string, []>("op_52364_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4441_cast = mul(x = var_52363_cast, y = var_52364_to_fp16)[name = tensor<string, []>("aw_4441_cast")];
            tensor<string, []> var_52367_equation_0 = const()[name = tensor<string, []>("op_52367_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52367_cast = einsum(equation = var_52367_equation_0, values = (var_52209_cast, var_52126_cast))[name = tensor<string, []>("op_52367_cast")];
            tensor<fp16, []> var_52368_to_fp16 = const()[name = tensor<string, []>("op_52368_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4443_cast = mul(x = var_52367_cast, y = var_52368_to_fp16)[name = tensor<string, []>("aw_4443_cast")];
            tensor<string, []> var_52371_equation_0 = const()[name = tensor<string, []>("op_52371_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52371_cast = einsum(equation = var_52371_equation_0, values = (var_52213_cast, var_52130_cast))[name = tensor<string, []>("op_52371_cast")];
            tensor<fp16, []> var_52372_to_fp16 = const()[name = tensor<string, []>("op_52372_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4445_cast = mul(x = var_52371_cast, y = var_52372_to_fp16)[name = tensor<string, []>("aw_4445_cast")];
            tensor<string, []> var_52375_equation_0 = const()[name = tensor<string, []>("op_52375_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52375_cast = einsum(equation = var_52375_equation_0, values = (var_52217_cast, var_52134_cast))[name = tensor<string, []>("op_52375_cast")];
            tensor<fp16, []> var_52376_to_fp16 = const()[name = tensor<string, []>("op_52376_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4447_cast = mul(x = var_52375_cast, y = var_52376_to_fp16)[name = tensor<string, []>("aw_4447_cast")];
            tensor<string, []> var_52379_equation_0 = const()[name = tensor<string, []>("op_52379_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52379_cast = einsum(equation = var_52379_equation_0, values = (var_52221_cast, var_52138_cast))[name = tensor<string, []>("op_52379_cast")];
            tensor<fp16, []> var_52380_to_fp16 = const()[name = tensor<string, []>("op_52380_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4449_cast = mul(x = var_52379_cast, y = var_52380_to_fp16)[name = tensor<string, []>("aw_4449_cast")];
            tensor<string, []> var_52383_equation_0 = const()[name = tensor<string, []>("op_52383_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52383_cast = einsum(equation = var_52383_equation_0, values = (var_52225_cast, var_52142_cast))[name = tensor<string, []>("op_52383_cast")];
            tensor<fp16, []> var_52384_to_fp16 = const()[name = tensor<string, []>("op_52384_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4451_cast = mul(x = var_52383_cast, y = var_52384_to_fp16)[name = tensor<string, []>("aw_4451_cast")];
            tensor<string, []> var_52387_equation_0 = const()[name = tensor<string, []>("op_52387_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52387_cast = einsum(equation = var_52387_equation_0, values = (var_52229_cast, var_52146_cast))[name = tensor<string, []>("op_52387_cast")];
            tensor<fp16, []> var_52388_to_fp16 = const()[name = tensor<string, []>("op_52388_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4453_cast = mul(x = var_52387_cast, y = var_52388_to_fp16)[name = tensor<string, []>("aw_4453_cast")];
            tensor<string, []> var_52391_equation_0 = const()[name = tensor<string, []>("op_52391_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52391_cast = einsum(equation = var_52391_equation_0, values = (var_52233_cast, var_52150_cast))[name = tensor<string, []>("op_52391_cast")];
            tensor<fp16, []> var_52392_to_fp16 = const()[name = tensor<string, []>("op_52392_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4455_cast = mul(x = var_52391_cast, y = var_52392_to_fp16)[name = tensor<string, []>("aw_4455_cast")];
            tensor<string, []> var_52395_equation_0 = const()[name = tensor<string, []>("op_52395_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52395_cast = einsum(equation = var_52395_equation_0, values = (var_52237_cast, var_52154_cast))[name = tensor<string, []>("op_52395_cast")];
            tensor<fp16, []> var_52396_to_fp16 = const()[name = tensor<string, []>("op_52396_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4457_cast = mul(x = var_52395_cast, y = var_52396_to_fp16)[name = tensor<string, []>("aw_4457_cast")];
            tensor<string, []> var_52399_equation_0 = const()[name = tensor<string, []>("op_52399_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52399_cast = einsum(equation = var_52399_equation_0, values = (var_52241_cast, var_52158_cast))[name = tensor<string, []>("op_52399_cast")];
            tensor<fp16, []> var_52400_to_fp16 = const()[name = tensor<string, []>("op_52400_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4459_cast = mul(x = var_52399_cast, y = var_52400_to_fp16)[name = tensor<string, []>("aw_4459_cast")];
            tensor<string, []> var_52403_equation_0 = const()[name = tensor<string, []>("op_52403_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52403_cast = einsum(equation = var_52403_equation_0, values = (var_52245_cast, var_52162_cast))[name = tensor<string, []>("op_52403_cast")];
            tensor<fp16, []> var_52404_to_fp16 = const()[name = tensor<string, []>("op_52404_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4461_cast = mul(x = var_52403_cast, y = var_52404_to_fp16)[name = tensor<string, []>("aw_4461_cast")];
            tensor<string, []> var_52407_equation_0 = const()[name = tensor<string, []>("op_52407_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52407_cast = einsum(equation = var_52407_equation_0, values = (var_52249_cast, var_52166_cast))[name = tensor<string, []>("op_52407_cast")];
            tensor<fp16, []> var_52408_to_fp16 = const()[name = tensor<string, []>("op_52408_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4463_cast = mul(x = var_52407_cast, y = var_52408_to_fp16)[name = tensor<string, []>("aw_4463_cast")];
            tensor<string, []> var_52411_equation_0 = const()[name = tensor<string, []>("op_52411_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52411_cast = einsum(equation = var_52411_equation_0, values = (var_52253_cast, var_52170_cast))[name = tensor<string, []>("op_52411_cast")];
            tensor<fp16, []> var_52412_to_fp16 = const()[name = tensor<string, []>("op_52412_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4465_cast = mul(x = var_52411_cast, y = var_52412_to_fp16)[name = tensor<string, []>("aw_4465_cast")];
            tensor<string, []> var_52415_equation_0 = const()[name = tensor<string, []>("op_52415_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52415_cast = einsum(equation = var_52415_equation_0, values = (var_52257_cast, var_52174_cast))[name = tensor<string, []>("op_52415_cast")];
            tensor<fp16, []> var_52416_to_fp16 = const()[name = tensor<string, []>("op_52416_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4467_cast = mul(x = var_52415_cast, y = var_52416_to_fp16)[name = tensor<string, []>("aw_4467_cast")];
            tensor<string, []> var_52419_equation_0 = const()[name = tensor<string, []>("op_52419_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52419_cast = einsum(equation = var_52419_equation_0, values = (var_52261_cast, var_52178_cast))[name = tensor<string, []>("op_52419_cast")];
            tensor<fp16, []> var_52420_to_fp16 = const()[name = tensor<string, []>("op_52420_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4469_cast = mul(x = var_52419_cast, y = var_52420_to_fp16)[name = tensor<string, []>("aw_4469_cast")];
            tensor<string, []> var_52423_equation_0 = const()[name = tensor<string, []>("op_52423_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52423_cast = einsum(equation = var_52423_equation_0, values = (var_52265_cast, var_52182_cast))[name = tensor<string, []>("op_52423_cast")];
            tensor<fp16, []> var_52424_to_fp16 = const()[name = tensor<string, []>("op_52424_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4471_cast = mul(x = var_52423_cast, y = var_52424_to_fp16)[name = tensor<string, []>("aw_4471_cast")];
            tensor<string, []> var_52427_equation_0 = const()[name = tensor<string, []>("op_52427_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52427_cast = einsum(equation = var_52427_equation_0, values = (var_52269_cast, var_52186_cast))[name = tensor<string, []>("op_52427_cast")];
            tensor<fp16, []> var_52428_to_fp16 = const()[name = tensor<string, []>("op_52428_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4473_cast = mul(x = var_52427_cast, y = var_52428_to_fp16)[name = tensor<string, []>("aw_4473_cast")];
            tensor<string, []> var_52431_equation_0 = const()[name = tensor<string, []>("op_52431_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52431_cast = einsum(equation = var_52431_equation_0, values = (var_52273_cast, var_52190_cast))[name = tensor<string, []>("op_52431_cast")];
            tensor<fp16, []> var_52432_to_fp16 = const()[name = tensor<string, []>("op_52432_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4475_cast = mul(x = var_52431_cast, y = var_52432_to_fp16)[name = tensor<string, []>("aw_4475_cast")];
            tensor<string, []> var_52435_equation_0 = const()[name = tensor<string, []>("op_52435_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52435_cast = einsum(equation = var_52435_equation_0, values = (var_52277_cast, var_52194_cast))[name = tensor<string, []>("op_52435_cast")];
            tensor<fp16, []> var_52436_to_fp16 = const()[name = tensor<string, []>("op_52436_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4477_cast = mul(x = var_52435_cast, y = var_52436_to_fp16)[name = tensor<string, []>("aw_4477_cast")];
            tensor<string, []> var_52439_equation_0 = const()[name = tensor<string, []>("op_52439_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_52439_cast = einsum(equation = var_52439_equation_0, values = (var_52281_cast, var_52198_cast))[name = tensor<string, []>("op_52439_cast")];
            tensor<fp16, []> var_52440_to_fp16 = const()[name = tensor<string, []>("op_52440_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4479_cast = mul(x = var_52439_cast, y = var_52440_to_fp16)[name = tensor<string, []>("aw_4479_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52442_cast = softmax(axis = var_30385, x = aw_4441_cast)[name = tensor<string, []>("op_52442_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52443_cast = softmax(axis = var_30385, x = aw_4443_cast)[name = tensor<string, []>("op_52443_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52444_cast = softmax(axis = var_30385, x = aw_4445_cast)[name = tensor<string, []>("op_52444_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52445_cast = softmax(axis = var_30385, x = aw_4447_cast)[name = tensor<string, []>("op_52445_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52446_cast = softmax(axis = var_30385, x = aw_4449_cast)[name = tensor<string, []>("op_52446_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52447_cast = softmax(axis = var_30385, x = aw_4451_cast)[name = tensor<string, []>("op_52447_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52448_cast = softmax(axis = var_30385, x = aw_4453_cast)[name = tensor<string, []>("op_52448_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52449_cast = softmax(axis = var_30385, x = aw_4455_cast)[name = tensor<string, []>("op_52449_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52450_cast = softmax(axis = var_30385, x = aw_4457_cast)[name = tensor<string, []>("op_52450_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52451_cast = softmax(axis = var_30385, x = aw_4459_cast)[name = tensor<string, []>("op_52451_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52452_cast = softmax(axis = var_30385, x = aw_4461_cast)[name = tensor<string, []>("op_52452_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52453_cast = softmax(axis = var_30385, x = aw_4463_cast)[name = tensor<string, []>("op_52453_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52454_cast = softmax(axis = var_30385, x = aw_4465_cast)[name = tensor<string, []>("op_52454_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52455_cast = softmax(axis = var_30385, x = aw_4467_cast)[name = tensor<string, []>("op_52455_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52456_cast = softmax(axis = var_30385, x = aw_4469_cast)[name = tensor<string, []>("op_52456_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52457_cast = softmax(axis = var_30385, x = aw_4471_cast)[name = tensor<string, []>("op_52457_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52458_cast = softmax(axis = var_30385, x = aw_4473_cast)[name = tensor<string, []>("op_52458_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52459_cast = softmax(axis = var_30385, x = aw_4475_cast)[name = tensor<string, []>("op_52459_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52460_cast = softmax(axis = var_30385, x = aw_4477_cast)[name = tensor<string, []>("op_52460_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_52461_cast = softmax(axis = var_30385, x = aw_4479_cast)[name = tensor<string, []>("op_52461_cast")];
            tensor<string, []> var_52463_equation_0 = const()[name = tensor<string, []>("op_52463_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52463_cast = einsum(equation = var_52463_equation_0, values = (var_52283_cast, var_52442_cast))[name = tensor<string, []>("op_52463_cast")];
            tensor<string, []> var_52465_equation_0 = const()[name = tensor<string, []>("op_52465_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52465_cast = einsum(equation = var_52465_equation_0, values = (var_52287_cast, var_52443_cast))[name = tensor<string, []>("op_52465_cast")];
            tensor<string, []> var_52467_equation_0 = const()[name = tensor<string, []>("op_52467_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52467_cast = einsum(equation = var_52467_equation_0, values = (var_52291_cast, var_52444_cast))[name = tensor<string, []>("op_52467_cast")];
            tensor<string, []> var_52469_equation_0 = const()[name = tensor<string, []>("op_52469_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52469_cast = einsum(equation = var_52469_equation_0, values = (var_52295_cast, var_52445_cast))[name = tensor<string, []>("op_52469_cast")];
            tensor<string, []> var_52471_equation_0 = const()[name = tensor<string, []>("op_52471_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52471_cast = einsum(equation = var_52471_equation_0, values = (var_52299_cast, var_52446_cast))[name = tensor<string, []>("op_52471_cast")];
            tensor<string, []> var_52473_equation_0 = const()[name = tensor<string, []>("op_52473_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52473_cast = einsum(equation = var_52473_equation_0, values = (var_52303_cast, var_52447_cast))[name = tensor<string, []>("op_52473_cast")];
            tensor<string, []> var_52475_equation_0 = const()[name = tensor<string, []>("op_52475_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52475_cast = einsum(equation = var_52475_equation_0, values = (var_52307_cast, var_52448_cast))[name = tensor<string, []>("op_52475_cast")];
            tensor<string, []> var_52477_equation_0 = const()[name = tensor<string, []>("op_52477_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52477_cast = einsum(equation = var_52477_equation_0, values = (var_52311_cast, var_52449_cast))[name = tensor<string, []>("op_52477_cast")];
            tensor<string, []> var_52479_equation_0 = const()[name = tensor<string, []>("op_52479_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52479_cast = einsum(equation = var_52479_equation_0, values = (var_52315_cast, var_52450_cast))[name = tensor<string, []>("op_52479_cast")];
            tensor<string, []> var_52481_equation_0 = const()[name = tensor<string, []>("op_52481_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52481_cast = einsum(equation = var_52481_equation_0, values = (var_52319_cast, var_52451_cast))[name = tensor<string, []>("op_52481_cast")];
            tensor<string, []> var_52483_equation_0 = const()[name = tensor<string, []>("op_52483_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52483_cast = einsum(equation = var_52483_equation_0, values = (var_52323_cast, var_52452_cast))[name = tensor<string, []>("op_52483_cast")];
            tensor<string, []> var_52485_equation_0 = const()[name = tensor<string, []>("op_52485_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52485_cast = einsum(equation = var_52485_equation_0, values = (var_52327_cast, var_52453_cast))[name = tensor<string, []>("op_52485_cast")];
            tensor<string, []> var_52487_equation_0 = const()[name = tensor<string, []>("op_52487_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52487_cast = einsum(equation = var_52487_equation_0, values = (var_52331_cast, var_52454_cast))[name = tensor<string, []>("op_52487_cast")];
            tensor<string, []> var_52489_equation_0 = const()[name = tensor<string, []>("op_52489_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52489_cast = einsum(equation = var_52489_equation_0, values = (var_52335_cast, var_52455_cast))[name = tensor<string, []>("op_52489_cast")];
            tensor<string, []> var_52491_equation_0 = const()[name = tensor<string, []>("op_52491_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52491_cast = einsum(equation = var_52491_equation_0, values = (var_52339_cast, var_52456_cast))[name = tensor<string, []>("op_52491_cast")];
            tensor<string, []> var_52493_equation_0 = const()[name = tensor<string, []>("op_52493_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52493_cast = einsum(equation = var_52493_equation_0, values = (var_52343_cast, var_52457_cast))[name = tensor<string, []>("op_52493_cast")];
            tensor<string, []> var_52495_equation_0 = const()[name = tensor<string, []>("op_52495_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52495_cast = einsum(equation = var_52495_equation_0, values = (var_52347_cast, var_52458_cast))[name = tensor<string, []>("op_52495_cast")];
            tensor<string, []> var_52497_equation_0 = const()[name = tensor<string, []>("op_52497_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52497_cast = einsum(equation = var_52497_equation_0, values = (var_52351_cast, var_52459_cast))[name = tensor<string, []>("op_52497_cast")];
            tensor<string, []> var_52499_equation_0 = const()[name = tensor<string, []>("op_52499_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52499_cast = einsum(equation = var_52499_equation_0, values = (var_52355_cast, var_52460_cast))[name = tensor<string, []>("op_52499_cast")];
            tensor<string, []> var_52501_equation_0 = const()[name = tensor<string, []>("op_52501_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52501_cast = einsum(equation = var_52501_equation_0, values = (var_52359_cast, var_52461_cast))[name = tensor<string, []>("op_52501_cast")];
            tensor<bool, []> input_663_interleave_0 = const()[name = tensor<string, []>("input_663_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_663_cast = concat(axis = var_30385, interleave = input_663_interleave_0, values = (var_52463_cast, var_52465_cast, var_52467_cast, var_52469_cast, var_52471_cast, var_52473_cast, var_52475_cast, var_52477_cast, var_52479_cast, var_52481_cast, var_52483_cast, var_52485_cast, var_52487_cast, var_52489_cast, var_52491_cast, var_52493_cast, var_52495_cast, var_52497_cast, var_52499_cast, var_52501_cast))[name = tensor<string, []>("input_663_cast")];
            tensor<int32, [2]> var_52507 = const()[name = tensor<string, []>("op_52507"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_52509 = const()[name = tensor<string, []>("op_52509"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_52511_pad_type_0 = const()[name = tensor<string, []>("op_52511_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_52511_pad_0 = const()[name = tensor<string, []>("op_52511_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4407513216)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4410790080)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_52511_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_out_0_bias_to_fp16, dilations = var_52509, groups = var_30385, pad = var_52511_pad_0, pad_type = var_52511_pad_type_0, strides = var_52507, weight = up_blocks_0_attentions_2_transformer_blocks_3_attn2_to_out_0_weight_to_fp16, x = input_663_cast)[name = tensor<string, []>("op_52511_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_347_cast = add(x = var_52511_cast, y = inputs_345_cast)[name = tensor<string, []>("inputs_347_cast")];
            tensor<int32, [1]> var_52515 = const()[name = tensor<string, []>("op_52515"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_347_cast = reduce_mean(axes = var_52515, keep_dims = var_30380, x = inputs_347_cast)[name = tensor<string, []>("channels_mean_347_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_347_cast = sub(x = inputs_347_cast, y = channels_mean_347_cast)[name = tensor<string, []>("zero_mean_347_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_347_cast = mul(x = zero_mean_347_cast, y = zero_mean_347_cast)[name = tensor<string, []>("zero_mean_sq_347_cast")];
            tensor<int32, [1]> var_52519 = const()[name = tensor<string, []>("op_52519"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_52520_cast = reduce_mean(axes = var_52519, keep_dims = var_30380, x = zero_mean_sq_347_cast)[name = tensor<string, []>("op_52520_cast")];
            tensor<fp16, []> var_52521_to_fp16 = const()[name = tensor<string, []>("op_52521_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_52522_cast = add(x = var_52520_cast, y = var_52521_to_fp16)[name = tensor<string, []>("op_52522_cast")];
            tensor<fp16, []> denom_347_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_347_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_347_cast = rsqrt(epsilon = denom_347_epsilon_0_to_fp16, x = var_52522_cast)[name = tensor<string, []>("denom_347_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_347_cast = mul(x = zero_mean_347_cast, y = denom_347_cast)[name = tensor<string, []>("out_347_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_52526_to_fp16 = const()[name = tensor<string, []>("op_52526_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4410792704)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_52527_cast = add(x = out_347_cast, y = var_52526_to_fp16)[name = tensor<string, []>("op_52527_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_52529_to_fp16 = const()[name = tensor<string, []>("op_52529_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4410795328)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_665_cast = mul(x = var_52527_cast, y = var_52529_to_fp16)[name = tensor<string, []>("input_665_cast")];
            tensor<int32, [2]> var_52537 = const()[name = tensor<string, []>("op_52537"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_52539 = const()[name = tensor<string, []>("op_52539"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_52541_pad_type_0 = const()[name = tensor<string, []>("op_52541_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_52541_pad_0 = const()[name = tensor<string, []>("op_52541_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4410797952)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_2_transformer_blocks_3_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4437012416)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_52541_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_3_ff_net_0_proj_bias_to_fp16, dilations = var_52539, groups = var_30385, pad = var_52541_pad_0, pad_type = var_52541_pad_type_0, strides = var_52537, weight = up_blocks_0_attentions_2_transformer_blocks_3_ff_net_0_proj_weight_to_fp16, x = input_665_cast)[name = tensor<string, []>("op_52541_cast")];
            tensor<int32, [2]> var_52542_split_sizes_0 = const()[name = tensor<string, []>("op_52542_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_52542_axis_0 = const()[name = tensor<string, []>("op_52542_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_52542_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_52542_cast_1 = split(axis = var_52542_axis_0, split_sizes = var_52542_split_sizes_0, x = var_52541_cast)[name = tensor<string, []>("op_52542_cast")];
            tensor<string, []> var_52544_mode_0 = const()[name = tensor<string, []>("op_52544_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_52544_cast = gelu(mode = var_52544_mode_0, x = var_52542_cast_1)[name = tensor<string, []>("op_52544_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_667_cast = mul(x = var_52542_cast_0, y = var_52544_cast)[name = tensor<string, []>("input_667_cast")];
            tensor<int32, [2]> var_52548 = const()[name = tensor<string, []>("op_52548"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_52550 = const()[name = tensor<string, []>("op_52550"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_52552_pad_type_0 = const()[name = tensor<string, []>("op_52552_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_52552_pad_0 = const()[name = tensor<string, []>("op_52552_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_3_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4437032960)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_3_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_3_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4450140224)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_52552_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_3_ff_net_2_bias_to_fp16, dilations = var_52550, groups = var_30385, pad = var_52552_pad_0, pad_type = var_52552_pad_type_0, strides = var_52548, weight = up_blocks_0_attentions_2_transformer_blocks_3_ff_net_2_weight_to_fp16, x = input_667_cast)[name = tensor<string, []>("op_52552_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_349_cast = add(x = var_52552_cast, y = inputs_347_cast)[name = tensor<string, []>("inputs_349_cast")];
            tensor<int32, [1]> var_52562 = const()[name = tensor<string, []>("op_52562"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_349_cast = reduce_mean(axes = var_52562, keep_dims = var_30380, x = inputs_349_cast)[name = tensor<string, []>("channels_mean_349_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_349_cast = sub(x = inputs_349_cast, y = channels_mean_349_cast)[name = tensor<string, []>("zero_mean_349_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_349_cast = mul(x = zero_mean_349_cast, y = zero_mean_349_cast)[name = tensor<string, []>("zero_mean_sq_349_cast")];
            tensor<int32, [1]> var_52566 = const()[name = tensor<string, []>("op_52566"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_52567_cast = reduce_mean(axes = var_52566, keep_dims = var_30380, x = zero_mean_sq_349_cast)[name = tensor<string, []>("op_52567_cast")];
            tensor<fp16, []> var_52568_to_fp16 = const()[name = tensor<string, []>("op_52568_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_52569_cast = add(x = var_52567_cast, y = var_52568_to_fp16)[name = tensor<string, []>("op_52569_cast")];
            tensor<fp16, []> denom_349_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_349_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_349_cast = rsqrt(epsilon = denom_349_epsilon_0_to_fp16, x = var_52569_cast)[name = tensor<string, []>("denom_349_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_349_cast = mul(x = zero_mean_349_cast, y = denom_349_cast)[name = tensor<string, []>("out_349_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_52573_to_fp16 = const()[name = tensor<string, []>("op_52573_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4450142848)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_52574_cast = add(x = out_349_cast, y = var_52573_to_fp16)[name = tensor<string, []>("op_52574_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_52576_to_fp16 = const()[name = tensor<string, []>("op_52576_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4450145472)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_459_cast = mul(x = var_52574_cast, y = var_52576_to_fp16)[name = tensor<string, []>("hidden_states_459_cast")];
            tensor<int32, [2]> var_52583 = const()[name = tensor<string, []>("op_52583"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_52585 = const()[name = tensor<string, []>("op_52585"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_233_pad_type_0 = const()[name = tensor<string, []>("q_233_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_233_pad_0 = const()[name = tensor<string, []>("q_233_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4450148096)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_233_cast = conv(dilations = var_52585, groups = var_30385, pad = q_233_pad_0, pad_type = q_233_pad_type_0, strides = var_52583, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_q_weight_to_fp16, x = hidden_states_459_cast)[name = tensor<string, []>("q_233_cast")];
            tensor<int32, [2]> var_52589 = const()[name = tensor<string, []>("op_52589"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_52591 = const()[name = tensor<string, []>("op_52591"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_465_pad_type_0 = const()[name = tensor<string, []>("k_465_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_465_pad_0 = const()[name = tensor<string, []>("k_465_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4453424960)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_465_cast = conv(dilations = var_52591, groups = var_30385, pad = k_465_pad_0, pad_type = k_465_pad_type_0, strides = var_52589, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_k_weight_to_fp16, x = hidden_states_459_cast)[name = tensor<string, []>("k_465_cast")];
            tensor<int32, [2]> var_52595 = const()[name = tensor<string, []>("op_52595"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_52597 = const()[name = tensor<string, []>("op_52597"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_233_pad_type_0 = const()[name = tensor<string, []>("v_233_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_233_pad_0 = const()[name = tensor<string, []>("v_233_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4456701824)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_233_cast = conv(dilations = var_52597, groups = var_30385, pad = v_233_pad_0, pad_type = v_233_pad_type_0, strides = var_52595, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_v_weight_to_fp16, x = hidden_states_459_cast)[name = tensor<string, []>("v_233_cast")];
            tensor<int32, [4]> var_52601_begin_0 = const()[name = tensor<string, []>("op_52601_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_52601_end_0 = const()[name = tensor<string, []>("op_52601_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_52601_end_mask_0 = const()[name = tensor<string, []>("op_52601_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52601_cast = slice_by_index(begin = var_52601_begin_0, end = var_52601_end_0, end_mask = var_52601_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52601_cast")];
            tensor<int32, [4]> var_52605_begin_0 = const()[name = tensor<string, []>("op_52605_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_52605_end_0 = const()[name = tensor<string, []>("op_52605_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_52605_end_mask_0 = const()[name = tensor<string, []>("op_52605_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52605_cast = slice_by_index(begin = var_52605_begin_0, end = var_52605_end_0, end_mask = var_52605_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52605_cast")];
            tensor<int32, [4]> var_52609_begin_0 = const()[name = tensor<string, []>("op_52609_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_52609_end_0 = const()[name = tensor<string, []>("op_52609_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_52609_end_mask_0 = const()[name = tensor<string, []>("op_52609_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52609_cast = slice_by_index(begin = var_52609_begin_0, end = var_52609_end_0, end_mask = var_52609_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52609_cast")];
            tensor<int32, [4]> var_52613_begin_0 = const()[name = tensor<string, []>("op_52613_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_52613_end_0 = const()[name = tensor<string, []>("op_52613_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_52613_end_mask_0 = const()[name = tensor<string, []>("op_52613_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52613_cast = slice_by_index(begin = var_52613_begin_0, end = var_52613_end_0, end_mask = var_52613_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52613_cast")];
            tensor<int32, [4]> var_52617_begin_0 = const()[name = tensor<string, []>("op_52617_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_52617_end_0 = const()[name = tensor<string, []>("op_52617_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_52617_end_mask_0 = const()[name = tensor<string, []>("op_52617_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52617_cast = slice_by_index(begin = var_52617_begin_0, end = var_52617_end_0, end_mask = var_52617_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52617_cast")];
            tensor<int32, [4]> var_52621_begin_0 = const()[name = tensor<string, []>("op_52621_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_52621_end_0 = const()[name = tensor<string, []>("op_52621_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_52621_end_mask_0 = const()[name = tensor<string, []>("op_52621_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52621_cast = slice_by_index(begin = var_52621_begin_0, end = var_52621_end_0, end_mask = var_52621_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52621_cast")];
            tensor<int32, [4]> var_52625_begin_0 = const()[name = tensor<string, []>("op_52625_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_52625_end_0 = const()[name = tensor<string, []>("op_52625_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_52625_end_mask_0 = const()[name = tensor<string, []>("op_52625_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52625_cast = slice_by_index(begin = var_52625_begin_0, end = var_52625_end_0, end_mask = var_52625_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52625_cast")];
            tensor<int32, [4]> var_52629_begin_0 = const()[name = tensor<string, []>("op_52629_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_52629_end_0 = const()[name = tensor<string, []>("op_52629_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_52629_end_mask_0 = const()[name = tensor<string, []>("op_52629_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52629_cast = slice_by_index(begin = var_52629_begin_0, end = var_52629_end_0, end_mask = var_52629_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52629_cast")];
            tensor<int32, [4]> var_52633_begin_0 = const()[name = tensor<string, []>("op_52633_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_52633_end_0 = const()[name = tensor<string, []>("op_52633_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_52633_end_mask_0 = const()[name = tensor<string, []>("op_52633_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52633_cast = slice_by_index(begin = var_52633_begin_0, end = var_52633_end_0, end_mask = var_52633_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52633_cast")];
            tensor<int32, [4]> var_52637_begin_0 = const()[name = tensor<string, []>("op_52637_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_52637_end_0 = const()[name = tensor<string, []>("op_52637_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_52637_end_mask_0 = const()[name = tensor<string, []>("op_52637_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52637_cast = slice_by_index(begin = var_52637_begin_0, end = var_52637_end_0, end_mask = var_52637_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52637_cast")];
            tensor<int32, [4]> var_52641_begin_0 = const()[name = tensor<string, []>("op_52641_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_52641_end_0 = const()[name = tensor<string, []>("op_52641_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_52641_end_mask_0 = const()[name = tensor<string, []>("op_52641_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52641_cast = slice_by_index(begin = var_52641_begin_0, end = var_52641_end_0, end_mask = var_52641_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52641_cast")];
            tensor<int32, [4]> var_52645_begin_0 = const()[name = tensor<string, []>("op_52645_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_52645_end_0 = const()[name = tensor<string, []>("op_52645_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_52645_end_mask_0 = const()[name = tensor<string, []>("op_52645_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52645_cast = slice_by_index(begin = var_52645_begin_0, end = var_52645_end_0, end_mask = var_52645_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52645_cast")];
            tensor<int32, [4]> var_52649_begin_0 = const()[name = tensor<string, []>("op_52649_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_52649_end_0 = const()[name = tensor<string, []>("op_52649_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_52649_end_mask_0 = const()[name = tensor<string, []>("op_52649_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52649_cast = slice_by_index(begin = var_52649_begin_0, end = var_52649_end_0, end_mask = var_52649_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52649_cast")];
            tensor<int32, [4]> var_52653_begin_0 = const()[name = tensor<string, []>("op_52653_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_52653_end_0 = const()[name = tensor<string, []>("op_52653_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_52653_end_mask_0 = const()[name = tensor<string, []>("op_52653_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52653_cast = slice_by_index(begin = var_52653_begin_0, end = var_52653_end_0, end_mask = var_52653_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52653_cast")];
            tensor<int32, [4]> var_52657_begin_0 = const()[name = tensor<string, []>("op_52657_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_52657_end_0 = const()[name = tensor<string, []>("op_52657_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_52657_end_mask_0 = const()[name = tensor<string, []>("op_52657_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52657_cast = slice_by_index(begin = var_52657_begin_0, end = var_52657_end_0, end_mask = var_52657_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52657_cast")];
            tensor<int32, [4]> var_52661_begin_0 = const()[name = tensor<string, []>("op_52661_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_52661_end_0 = const()[name = tensor<string, []>("op_52661_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_52661_end_mask_0 = const()[name = tensor<string, []>("op_52661_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52661_cast = slice_by_index(begin = var_52661_begin_0, end = var_52661_end_0, end_mask = var_52661_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52661_cast")];
            tensor<int32, [4]> var_52665_begin_0 = const()[name = tensor<string, []>("op_52665_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_52665_end_0 = const()[name = tensor<string, []>("op_52665_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_52665_end_mask_0 = const()[name = tensor<string, []>("op_52665_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52665_cast = slice_by_index(begin = var_52665_begin_0, end = var_52665_end_0, end_mask = var_52665_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52665_cast")];
            tensor<int32, [4]> var_52669_begin_0 = const()[name = tensor<string, []>("op_52669_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_52669_end_0 = const()[name = tensor<string, []>("op_52669_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_52669_end_mask_0 = const()[name = tensor<string, []>("op_52669_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52669_cast = slice_by_index(begin = var_52669_begin_0, end = var_52669_end_0, end_mask = var_52669_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52669_cast")];
            tensor<int32, [4]> var_52673_begin_0 = const()[name = tensor<string, []>("op_52673_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_52673_end_0 = const()[name = tensor<string, []>("op_52673_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_52673_end_mask_0 = const()[name = tensor<string, []>("op_52673_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52673_cast = slice_by_index(begin = var_52673_begin_0, end = var_52673_end_0, end_mask = var_52673_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52673_cast")];
            tensor<int32, [4]> var_52677_begin_0 = const()[name = tensor<string, []>("op_52677_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_52677_end_0 = const()[name = tensor<string, []>("op_52677_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_52677_end_mask_0 = const()[name = tensor<string, []>("op_52677_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52677_cast = slice_by_index(begin = var_52677_begin_0, end = var_52677_end_0, end_mask = var_52677_end_mask_0, x = q_233_cast)[name = tensor<string, []>("op_52677_cast")];
            tensor<int32, [4]> k_467_perm_0 = const()[name = tensor<string, []>("k_467_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_52684_begin_0 = const()[name = tensor<string, []>("op_52684_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_52684_end_0 = const()[name = tensor<string, []>("op_52684_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_52684_end_mask_0 = const()[name = tensor<string, []>("op_52684_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_23 = transpose(perm = k_467_perm_0, x = k_465_cast)[name = tensor<string, []>("transpose_23")];
            tensor<fp16, [2, 1024, 1, 64]> var_52684_cast = slice_by_index(begin = var_52684_begin_0, end = var_52684_end_0, end_mask = var_52684_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52684_cast")];
            tensor<int32, [4]> var_52688_begin_0 = const()[name = tensor<string, []>("op_52688_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_52688_end_0 = const()[name = tensor<string, []>("op_52688_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_52688_end_mask_0 = const()[name = tensor<string, []>("op_52688_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52688_cast = slice_by_index(begin = var_52688_begin_0, end = var_52688_end_0, end_mask = var_52688_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52688_cast")];
            tensor<int32, [4]> var_52692_begin_0 = const()[name = tensor<string, []>("op_52692_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_52692_end_0 = const()[name = tensor<string, []>("op_52692_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_52692_end_mask_0 = const()[name = tensor<string, []>("op_52692_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52692_cast = slice_by_index(begin = var_52692_begin_0, end = var_52692_end_0, end_mask = var_52692_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52692_cast")];
            tensor<int32, [4]> var_52696_begin_0 = const()[name = tensor<string, []>("op_52696_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_52696_end_0 = const()[name = tensor<string, []>("op_52696_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_52696_end_mask_0 = const()[name = tensor<string, []>("op_52696_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52696_cast = slice_by_index(begin = var_52696_begin_0, end = var_52696_end_0, end_mask = var_52696_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52696_cast")];
            tensor<int32, [4]> var_52700_begin_0 = const()[name = tensor<string, []>("op_52700_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_52700_end_0 = const()[name = tensor<string, []>("op_52700_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_52700_end_mask_0 = const()[name = tensor<string, []>("op_52700_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52700_cast = slice_by_index(begin = var_52700_begin_0, end = var_52700_end_0, end_mask = var_52700_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52700_cast")];
            tensor<int32, [4]> var_52704_begin_0 = const()[name = tensor<string, []>("op_52704_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_52704_end_0 = const()[name = tensor<string, []>("op_52704_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_52704_end_mask_0 = const()[name = tensor<string, []>("op_52704_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52704_cast = slice_by_index(begin = var_52704_begin_0, end = var_52704_end_0, end_mask = var_52704_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52704_cast")];
            tensor<int32, [4]> var_52708_begin_0 = const()[name = tensor<string, []>("op_52708_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_52708_end_0 = const()[name = tensor<string, []>("op_52708_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_52708_end_mask_0 = const()[name = tensor<string, []>("op_52708_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52708_cast = slice_by_index(begin = var_52708_begin_0, end = var_52708_end_0, end_mask = var_52708_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52708_cast")];
            tensor<int32, [4]> var_52712_begin_0 = const()[name = tensor<string, []>("op_52712_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_52712_end_0 = const()[name = tensor<string, []>("op_52712_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_52712_end_mask_0 = const()[name = tensor<string, []>("op_52712_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52712_cast = slice_by_index(begin = var_52712_begin_0, end = var_52712_end_0, end_mask = var_52712_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52712_cast")];
            tensor<int32, [4]> var_52716_begin_0 = const()[name = tensor<string, []>("op_52716_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_52716_end_0 = const()[name = tensor<string, []>("op_52716_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_52716_end_mask_0 = const()[name = tensor<string, []>("op_52716_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52716_cast = slice_by_index(begin = var_52716_begin_0, end = var_52716_end_0, end_mask = var_52716_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52716_cast")];
            tensor<int32, [4]> var_52720_begin_0 = const()[name = tensor<string, []>("op_52720_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_52720_end_0 = const()[name = tensor<string, []>("op_52720_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_52720_end_mask_0 = const()[name = tensor<string, []>("op_52720_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52720_cast = slice_by_index(begin = var_52720_begin_0, end = var_52720_end_0, end_mask = var_52720_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52720_cast")];
            tensor<int32, [4]> var_52724_begin_0 = const()[name = tensor<string, []>("op_52724_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_52724_end_0 = const()[name = tensor<string, []>("op_52724_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_52724_end_mask_0 = const()[name = tensor<string, []>("op_52724_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52724_cast = slice_by_index(begin = var_52724_begin_0, end = var_52724_end_0, end_mask = var_52724_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52724_cast")];
            tensor<int32, [4]> var_52728_begin_0 = const()[name = tensor<string, []>("op_52728_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_52728_end_0 = const()[name = tensor<string, []>("op_52728_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_52728_end_mask_0 = const()[name = tensor<string, []>("op_52728_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52728_cast = slice_by_index(begin = var_52728_begin_0, end = var_52728_end_0, end_mask = var_52728_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52728_cast")];
            tensor<int32, [4]> var_52732_begin_0 = const()[name = tensor<string, []>("op_52732_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_52732_end_0 = const()[name = tensor<string, []>("op_52732_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_52732_end_mask_0 = const()[name = tensor<string, []>("op_52732_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52732_cast = slice_by_index(begin = var_52732_begin_0, end = var_52732_end_0, end_mask = var_52732_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52732_cast")];
            tensor<int32, [4]> var_52736_begin_0 = const()[name = tensor<string, []>("op_52736_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_52736_end_0 = const()[name = tensor<string, []>("op_52736_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_52736_end_mask_0 = const()[name = tensor<string, []>("op_52736_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52736_cast = slice_by_index(begin = var_52736_begin_0, end = var_52736_end_0, end_mask = var_52736_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52736_cast")];
            tensor<int32, [4]> var_52740_begin_0 = const()[name = tensor<string, []>("op_52740_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_52740_end_0 = const()[name = tensor<string, []>("op_52740_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_52740_end_mask_0 = const()[name = tensor<string, []>("op_52740_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52740_cast = slice_by_index(begin = var_52740_begin_0, end = var_52740_end_0, end_mask = var_52740_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52740_cast")];
            tensor<int32, [4]> var_52744_begin_0 = const()[name = tensor<string, []>("op_52744_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_52744_end_0 = const()[name = tensor<string, []>("op_52744_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_52744_end_mask_0 = const()[name = tensor<string, []>("op_52744_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52744_cast = slice_by_index(begin = var_52744_begin_0, end = var_52744_end_0, end_mask = var_52744_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52744_cast")];
            tensor<int32, [4]> var_52748_begin_0 = const()[name = tensor<string, []>("op_52748_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_52748_end_0 = const()[name = tensor<string, []>("op_52748_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_52748_end_mask_0 = const()[name = tensor<string, []>("op_52748_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52748_cast = slice_by_index(begin = var_52748_begin_0, end = var_52748_end_0, end_mask = var_52748_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52748_cast")];
            tensor<int32, [4]> var_52752_begin_0 = const()[name = tensor<string, []>("op_52752_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_52752_end_0 = const()[name = tensor<string, []>("op_52752_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_52752_end_mask_0 = const()[name = tensor<string, []>("op_52752_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52752_cast = slice_by_index(begin = var_52752_begin_0, end = var_52752_end_0, end_mask = var_52752_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52752_cast")];
            tensor<int32, [4]> var_52756_begin_0 = const()[name = tensor<string, []>("op_52756_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_52756_end_0 = const()[name = tensor<string, []>("op_52756_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_52756_end_mask_0 = const()[name = tensor<string, []>("op_52756_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52756_cast = slice_by_index(begin = var_52756_begin_0, end = var_52756_end_0, end_mask = var_52756_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52756_cast")];
            tensor<int32, [4]> var_52760_begin_0 = const()[name = tensor<string, []>("op_52760_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_52760_end_0 = const()[name = tensor<string, []>("op_52760_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_52760_end_mask_0 = const()[name = tensor<string, []>("op_52760_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_52760_cast = slice_by_index(begin = var_52760_begin_0, end = var_52760_end_0, end_mask = var_52760_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_52760_cast")];
            tensor<int32, [4]> var_52762_begin_0 = const()[name = tensor<string, []>("op_52762_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_52762_end_0 = const()[name = tensor<string, []>("op_52762_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_52762_end_mask_0 = const()[name = tensor<string, []>("op_52762_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52762_cast = slice_by_index(begin = var_52762_begin_0, end = var_52762_end_0, end_mask = var_52762_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52762_cast")];
            tensor<int32, [4]> var_52766_begin_0 = const()[name = tensor<string, []>("op_52766_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_52766_end_0 = const()[name = tensor<string, []>("op_52766_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_52766_end_mask_0 = const()[name = tensor<string, []>("op_52766_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52766_cast = slice_by_index(begin = var_52766_begin_0, end = var_52766_end_0, end_mask = var_52766_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52766_cast")];
            tensor<int32, [4]> var_52770_begin_0 = const()[name = tensor<string, []>("op_52770_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_52770_end_0 = const()[name = tensor<string, []>("op_52770_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_52770_end_mask_0 = const()[name = tensor<string, []>("op_52770_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52770_cast = slice_by_index(begin = var_52770_begin_0, end = var_52770_end_0, end_mask = var_52770_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52770_cast")];
            tensor<int32, [4]> var_52774_begin_0 = const()[name = tensor<string, []>("op_52774_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_52774_end_0 = const()[name = tensor<string, []>("op_52774_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_52774_end_mask_0 = const()[name = tensor<string, []>("op_52774_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52774_cast = slice_by_index(begin = var_52774_begin_0, end = var_52774_end_0, end_mask = var_52774_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52774_cast")];
            tensor<int32, [4]> var_52778_begin_0 = const()[name = tensor<string, []>("op_52778_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_52778_end_0 = const()[name = tensor<string, []>("op_52778_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_52778_end_mask_0 = const()[name = tensor<string, []>("op_52778_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52778_cast = slice_by_index(begin = var_52778_begin_0, end = var_52778_end_0, end_mask = var_52778_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52778_cast")];
            tensor<int32, [4]> var_52782_begin_0 = const()[name = tensor<string, []>("op_52782_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_52782_end_0 = const()[name = tensor<string, []>("op_52782_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_52782_end_mask_0 = const()[name = tensor<string, []>("op_52782_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52782_cast = slice_by_index(begin = var_52782_begin_0, end = var_52782_end_0, end_mask = var_52782_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52782_cast")];
            tensor<int32, [4]> var_52786_begin_0 = const()[name = tensor<string, []>("op_52786_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_52786_end_0 = const()[name = tensor<string, []>("op_52786_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_52786_end_mask_0 = const()[name = tensor<string, []>("op_52786_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52786_cast = slice_by_index(begin = var_52786_begin_0, end = var_52786_end_0, end_mask = var_52786_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52786_cast")];
            tensor<int32, [4]> var_52790_begin_0 = const()[name = tensor<string, []>("op_52790_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_52790_end_0 = const()[name = tensor<string, []>("op_52790_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_52790_end_mask_0 = const()[name = tensor<string, []>("op_52790_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52790_cast = slice_by_index(begin = var_52790_begin_0, end = var_52790_end_0, end_mask = var_52790_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52790_cast")];
            tensor<int32, [4]> var_52794_begin_0 = const()[name = tensor<string, []>("op_52794_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_52794_end_0 = const()[name = tensor<string, []>("op_52794_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_52794_end_mask_0 = const()[name = tensor<string, []>("op_52794_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52794_cast = slice_by_index(begin = var_52794_begin_0, end = var_52794_end_0, end_mask = var_52794_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52794_cast")];
            tensor<int32, [4]> var_52798_begin_0 = const()[name = tensor<string, []>("op_52798_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_52798_end_0 = const()[name = tensor<string, []>("op_52798_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_52798_end_mask_0 = const()[name = tensor<string, []>("op_52798_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52798_cast = slice_by_index(begin = var_52798_begin_0, end = var_52798_end_0, end_mask = var_52798_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52798_cast")];
            tensor<int32, [4]> var_52802_begin_0 = const()[name = tensor<string, []>("op_52802_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_52802_end_0 = const()[name = tensor<string, []>("op_52802_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_52802_end_mask_0 = const()[name = tensor<string, []>("op_52802_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52802_cast = slice_by_index(begin = var_52802_begin_0, end = var_52802_end_0, end_mask = var_52802_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52802_cast")];
            tensor<int32, [4]> var_52806_begin_0 = const()[name = tensor<string, []>("op_52806_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_52806_end_0 = const()[name = tensor<string, []>("op_52806_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_52806_end_mask_0 = const()[name = tensor<string, []>("op_52806_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52806_cast = slice_by_index(begin = var_52806_begin_0, end = var_52806_end_0, end_mask = var_52806_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52806_cast")];
            tensor<int32, [4]> var_52810_begin_0 = const()[name = tensor<string, []>("op_52810_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_52810_end_0 = const()[name = tensor<string, []>("op_52810_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_52810_end_mask_0 = const()[name = tensor<string, []>("op_52810_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52810_cast = slice_by_index(begin = var_52810_begin_0, end = var_52810_end_0, end_mask = var_52810_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52810_cast")];
            tensor<int32, [4]> var_52814_begin_0 = const()[name = tensor<string, []>("op_52814_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_52814_end_0 = const()[name = tensor<string, []>("op_52814_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_52814_end_mask_0 = const()[name = tensor<string, []>("op_52814_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52814_cast = slice_by_index(begin = var_52814_begin_0, end = var_52814_end_0, end_mask = var_52814_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52814_cast")];
            tensor<int32, [4]> var_52818_begin_0 = const()[name = tensor<string, []>("op_52818_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_52818_end_0 = const()[name = tensor<string, []>("op_52818_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_52818_end_mask_0 = const()[name = tensor<string, []>("op_52818_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52818_cast = slice_by_index(begin = var_52818_begin_0, end = var_52818_end_0, end_mask = var_52818_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52818_cast")];
            tensor<int32, [4]> var_52822_begin_0 = const()[name = tensor<string, []>("op_52822_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_52822_end_0 = const()[name = tensor<string, []>("op_52822_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_52822_end_mask_0 = const()[name = tensor<string, []>("op_52822_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52822_cast = slice_by_index(begin = var_52822_begin_0, end = var_52822_end_0, end_mask = var_52822_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52822_cast")];
            tensor<int32, [4]> var_52826_begin_0 = const()[name = tensor<string, []>("op_52826_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_52826_end_0 = const()[name = tensor<string, []>("op_52826_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_52826_end_mask_0 = const()[name = tensor<string, []>("op_52826_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52826_cast = slice_by_index(begin = var_52826_begin_0, end = var_52826_end_0, end_mask = var_52826_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52826_cast")];
            tensor<int32, [4]> var_52830_begin_0 = const()[name = tensor<string, []>("op_52830_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_52830_end_0 = const()[name = tensor<string, []>("op_52830_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_52830_end_mask_0 = const()[name = tensor<string, []>("op_52830_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52830_cast = slice_by_index(begin = var_52830_begin_0, end = var_52830_end_0, end_mask = var_52830_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52830_cast")];
            tensor<int32, [4]> var_52834_begin_0 = const()[name = tensor<string, []>("op_52834_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_52834_end_0 = const()[name = tensor<string, []>("op_52834_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_52834_end_mask_0 = const()[name = tensor<string, []>("op_52834_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52834_cast = slice_by_index(begin = var_52834_begin_0, end = var_52834_end_0, end_mask = var_52834_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52834_cast")];
            tensor<int32, [4]> var_52838_begin_0 = const()[name = tensor<string, []>("op_52838_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_52838_end_0 = const()[name = tensor<string, []>("op_52838_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_52838_end_mask_0 = const()[name = tensor<string, []>("op_52838_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_52838_cast = slice_by_index(begin = var_52838_begin_0, end = var_52838_end_0, end_mask = var_52838_end_mask_0, x = v_233_cast)[name = tensor<string, []>("op_52838_cast")];
            tensor<string, []> var_52842_equation_0 = const()[name = tensor<string, []>("op_52842_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52842_cast = einsum(equation = var_52842_equation_0, values = (var_52684_cast, var_52601_cast))[name = tensor<string, []>("op_52842_cast")];
            tensor<fp16, []> var_52843_to_fp16 = const()[name = tensor<string, []>("op_52843_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4481_cast = mul(x = var_52842_cast, y = var_52843_to_fp16)[name = tensor<string, []>("aw_4481_cast")];
            tensor<string, []> var_52846_equation_0 = const()[name = tensor<string, []>("op_52846_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52846_cast = einsum(equation = var_52846_equation_0, values = (var_52688_cast, var_52605_cast))[name = tensor<string, []>("op_52846_cast")];
            tensor<fp16, []> var_52847_to_fp16 = const()[name = tensor<string, []>("op_52847_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4483_cast = mul(x = var_52846_cast, y = var_52847_to_fp16)[name = tensor<string, []>("aw_4483_cast")];
            tensor<string, []> var_52850_equation_0 = const()[name = tensor<string, []>("op_52850_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52850_cast = einsum(equation = var_52850_equation_0, values = (var_52692_cast, var_52609_cast))[name = tensor<string, []>("op_52850_cast")];
            tensor<fp16, []> var_52851_to_fp16 = const()[name = tensor<string, []>("op_52851_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4485_cast = mul(x = var_52850_cast, y = var_52851_to_fp16)[name = tensor<string, []>("aw_4485_cast")];
            tensor<string, []> var_52854_equation_0 = const()[name = tensor<string, []>("op_52854_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52854_cast = einsum(equation = var_52854_equation_0, values = (var_52696_cast, var_52613_cast))[name = tensor<string, []>("op_52854_cast")];
            tensor<fp16, []> var_52855_to_fp16 = const()[name = tensor<string, []>("op_52855_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4487_cast = mul(x = var_52854_cast, y = var_52855_to_fp16)[name = tensor<string, []>("aw_4487_cast")];
            tensor<string, []> var_52858_equation_0 = const()[name = tensor<string, []>("op_52858_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52858_cast = einsum(equation = var_52858_equation_0, values = (var_52700_cast, var_52617_cast))[name = tensor<string, []>("op_52858_cast")];
            tensor<fp16, []> var_52859_to_fp16 = const()[name = tensor<string, []>("op_52859_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4489_cast = mul(x = var_52858_cast, y = var_52859_to_fp16)[name = tensor<string, []>("aw_4489_cast")];
            tensor<string, []> var_52862_equation_0 = const()[name = tensor<string, []>("op_52862_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52862_cast = einsum(equation = var_52862_equation_0, values = (var_52704_cast, var_52621_cast))[name = tensor<string, []>("op_52862_cast")];
            tensor<fp16, []> var_52863_to_fp16 = const()[name = tensor<string, []>("op_52863_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4491_cast = mul(x = var_52862_cast, y = var_52863_to_fp16)[name = tensor<string, []>("aw_4491_cast")];
            tensor<string, []> var_52866_equation_0 = const()[name = tensor<string, []>("op_52866_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52866_cast = einsum(equation = var_52866_equation_0, values = (var_52708_cast, var_52625_cast))[name = tensor<string, []>("op_52866_cast")];
            tensor<fp16, []> var_52867_to_fp16 = const()[name = tensor<string, []>("op_52867_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4493_cast = mul(x = var_52866_cast, y = var_52867_to_fp16)[name = tensor<string, []>("aw_4493_cast")];
            tensor<string, []> var_52870_equation_0 = const()[name = tensor<string, []>("op_52870_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52870_cast = einsum(equation = var_52870_equation_0, values = (var_52712_cast, var_52629_cast))[name = tensor<string, []>("op_52870_cast")];
            tensor<fp16, []> var_52871_to_fp16 = const()[name = tensor<string, []>("op_52871_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4495_cast = mul(x = var_52870_cast, y = var_52871_to_fp16)[name = tensor<string, []>("aw_4495_cast")];
            tensor<string, []> var_52874_equation_0 = const()[name = tensor<string, []>("op_52874_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52874_cast = einsum(equation = var_52874_equation_0, values = (var_52716_cast, var_52633_cast))[name = tensor<string, []>("op_52874_cast")];
            tensor<fp16, []> var_52875_to_fp16 = const()[name = tensor<string, []>("op_52875_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4497_cast = mul(x = var_52874_cast, y = var_52875_to_fp16)[name = tensor<string, []>("aw_4497_cast")];
            tensor<string, []> var_52878_equation_0 = const()[name = tensor<string, []>("op_52878_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52878_cast = einsum(equation = var_52878_equation_0, values = (var_52720_cast, var_52637_cast))[name = tensor<string, []>("op_52878_cast")];
            tensor<fp16, []> var_52879_to_fp16 = const()[name = tensor<string, []>("op_52879_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4499_cast = mul(x = var_52878_cast, y = var_52879_to_fp16)[name = tensor<string, []>("aw_4499_cast")];
            tensor<string, []> var_52882_equation_0 = const()[name = tensor<string, []>("op_52882_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52882_cast = einsum(equation = var_52882_equation_0, values = (var_52724_cast, var_52641_cast))[name = tensor<string, []>("op_52882_cast")];
            tensor<fp16, []> var_52883_to_fp16 = const()[name = tensor<string, []>("op_52883_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4501_cast = mul(x = var_52882_cast, y = var_52883_to_fp16)[name = tensor<string, []>("aw_4501_cast")];
            tensor<string, []> var_52886_equation_0 = const()[name = tensor<string, []>("op_52886_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52886_cast = einsum(equation = var_52886_equation_0, values = (var_52728_cast, var_52645_cast))[name = tensor<string, []>("op_52886_cast")];
            tensor<fp16, []> var_52887_to_fp16 = const()[name = tensor<string, []>("op_52887_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4503_cast = mul(x = var_52886_cast, y = var_52887_to_fp16)[name = tensor<string, []>("aw_4503_cast")];
            tensor<string, []> var_52890_equation_0 = const()[name = tensor<string, []>("op_52890_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52890_cast = einsum(equation = var_52890_equation_0, values = (var_52732_cast, var_52649_cast))[name = tensor<string, []>("op_52890_cast")];
            tensor<fp16, []> var_52891_to_fp16 = const()[name = tensor<string, []>("op_52891_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4505_cast = mul(x = var_52890_cast, y = var_52891_to_fp16)[name = tensor<string, []>("aw_4505_cast")];
            tensor<string, []> var_52894_equation_0 = const()[name = tensor<string, []>("op_52894_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52894_cast = einsum(equation = var_52894_equation_0, values = (var_52736_cast, var_52653_cast))[name = tensor<string, []>("op_52894_cast")];
            tensor<fp16, []> var_52895_to_fp16 = const()[name = tensor<string, []>("op_52895_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4507_cast = mul(x = var_52894_cast, y = var_52895_to_fp16)[name = tensor<string, []>("aw_4507_cast")];
            tensor<string, []> var_52898_equation_0 = const()[name = tensor<string, []>("op_52898_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52898_cast = einsum(equation = var_52898_equation_0, values = (var_52740_cast, var_52657_cast))[name = tensor<string, []>("op_52898_cast")];
            tensor<fp16, []> var_52899_to_fp16 = const()[name = tensor<string, []>("op_52899_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4509_cast = mul(x = var_52898_cast, y = var_52899_to_fp16)[name = tensor<string, []>("aw_4509_cast")];
            tensor<string, []> var_52902_equation_0 = const()[name = tensor<string, []>("op_52902_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52902_cast = einsum(equation = var_52902_equation_0, values = (var_52744_cast, var_52661_cast))[name = tensor<string, []>("op_52902_cast")];
            tensor<fp16, []> var_52903_to_fp16 = const()[name = tensor<string, []>("op_52903_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4511_cast = mul(x = var_52902_cast, y = var_52903_to_fp16)[name = tensor<string, []>("aw_4511_cast")];
            tensor<string, []> var_52906_equation_0 = const()[name = tensor<string, []>("op_52906_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52906_cast = einsum(equation = var_52906_equation_0, values = (var_52748_cast, var_52665_cast))[name = tensor<string, []>("op_52906_cast")];
            tensor<fp16, []> var_52907_to_fp16 = const()[name = tensor<string, []>("op_52907_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4513_cast = mul(x = var_52906_cast, y = var_52907_to_fp16)[name = tensor<string, []>("aw_4513_cast")];
            tensor<string, []> var_52910_equation_0 = const()[name = tensor<string, []>("op_52910_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52910_cast = einsum(equation = var_52910_equation_0, values = (var_52752_cast, var_52669_cast))[name = tensor<string, []>("op_52910_cast")];
            tensor<fp16, []> var_52911_to_fp16 = const()[name = tensor<string, []>("op_52911_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4515_cast = mul(x = var_52910_cast, y = var_52911_to_fp16)[name = tensor<string, []>("aw_4515_cast")];
            tensor<string, []> var_52914_equation_0 = const()[name = tensor<string, []>("op_52914_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52914_cast = einsum(equation = var_52914_equation_0, values = (var_52756_cast, var_52673_cast))[name = tensor<string, []>("op_52914_cast")];
            tensor<fp16, []> var_52915_to_fp16 = const()[name = tensor<string, []>("op_52915_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4517_cast = mul(x = var_52914_cast, y = var_52915_to_fp16)[name = tensor<string, []>("aw_4517_cast")];
            tensor<string, []> var_52918_equation_0 = const()[name = tensor<string, []>("op_52918_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52918_cast = einsum(equation = var_52918_equation_0, values = (var_52760_cast, var_52677_cast))[name = tensor<string, []>("op_52918_cast")];
            tensor<fp16, []> var_52919_to_fp16 = const()[name = tensor<string, []>("op_52919_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4519_cast = mul(x = var_52918_cast, y = var_52919_to_fp16)[name = tensor<string, []>("aw_4519_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52921_cast = softmax(axis = var_30385, x = aw_4481_cast)[name = tensor<string, []>("op_52921_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52922_cast = softmax(axis = var_30385, x = aw_4483_cast)[name = tensor<string, []>("op_52922_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52923_cast = softmax(axis = var_30385, x = aw_4485_cast)[name = tensor<string, []>("op_52923_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52924_cast = softmax(axis = var_30385, x = aw_4487_cast)[name = tensor<string, []>("op_52924_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52925_cast = softmax(axis = var_30385, x = aw_4489_cast)[name = tensor<string, []>("op_52925_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52926_cast = softmax(axis = var_30385, x = aw_4491_cast)[name = tensor<string, []>("op_52926_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52927_cast = softmax(axis = var_30385, x = aw_4493_cast)[name = tensor<string, []>("op_52927_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52928_cast = softmax(axis = var_30385, x = aw_4495_cast)[name = tensor<string, []>("op_52928_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52929_cast = softmax(axis = var_30385, x = aw_4497_cast)[name = tensor<string, []>("op_52929_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52930_cast = softmax(axis = var_30385, x = aw_4499_cast)[name = tensor<string, []>("op_52930_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52931_cast = softmax(axis = var_30385, x = aw_4501_cast)[name = tensor<string, []>("op_52931_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52932_cast = softmax(axis = var_30385, x = aw_4503_cast)[name = tensor<string, []>("op_52932_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52933_cast = softmax(axis = var_30385, x = aw_4505_cast)[name = tensor<string, []>("op_52933_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52934_cast = softmax(axis = var_30385, x = aw_4507_cast)[name = tensor<string, []>("op_52934_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52935_cast = softmax(axis = var_30385, x = aw_4509_cast)[name = tensor<string, []>("op_52935_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52936_cast = softmax(axis = var_30385, x = aw_4511_cast)[name = tensor<string, []>("op_52936_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52937_cast = softmax(axis = var_30385, x = aw_4513_cast)[name = tensor<string, []>("op_52937_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52938_cast = softmax(axis = var_30385, x = aw_4515_cast)[name = tensor<string, []>("op_52938_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52939_cast = softmax(axis = var_30385, x = aw_4517_cast)[name = tensor<string, []>("op_52939_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_52940_cast = softmax(axis = var_30385, x = aw_4519_cast)[name = tensor<string, []>("op_52940_cast")];
            tensor<string, []> var_52942_equation_0 = const()[name = tensor<string, []>("op_52942_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52942_cast = einsum(equation = var_52942_equation_0, values = (var_52762_cast, var_52921_cast))[name = tensor<string, []>("op_52942_cast")];
            tensor<string, []> var_52944_equation_0 = const()[name = tensor<string, []>("op_52944_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52944_cast = einsum(equation = var_52944_equation_0, values = (var_52766_cast, var_52922_cast))[name = tensor<string, []>("op_52944_cast")];
            tensor<string, []> var_52946_equation_0 = const()[name = tensor<string, []>("op_52946_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52946_cast = einsum(equation = var_52946_equation_0, values = (var_52770_cast, var_52923_cast))[name = tensor<string, []>("op_52946_cast")];
            tensor<string, []> var_52948_equation_0 = const()[name = tensor<string, []>("op_52948_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52948_cast = einsum(equation = var_52948_equation_0, values = (var_52774_cast, var_52924_cast))[name = tensor<string, []>("op_52948_cast")];
            tensor<string, []> var_52950_equation_0 = const()[name = tensor<string, []>("op_52950_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52950_cast = einsum(equation = var_52950_equation_0, values = (var_52778_cast, var_52925_cast))[name = tensor<string, []>("op_52950_cast")];
            tensor<string, []> var_52952_equation_0 = const()[name = tensor<string, []>("op_52952_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52952_cast = einsum(equation = var_52952_equation_0, values = (var_52782_cast, var_52926_cast))[name = tensor<string, []>("op_52952_cast")];
            tensor<string, []> var_52954_equation_0 = const()[name = tensor<string, []>("op_52954_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52954_cast = einsum(equation = var_52954_equation_0, values = (var_52786_cast, var_52927_cast))[name = tensor<string, []>("op_52954_cast")];
            tensor<string, []> var_52956_equation_0 = const()[name = tensor<string, []>("op_52956_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52956_cast = einsum(equation = var_52956_equation_0, values = (var_52790_cast, var_52928_cast))[name = tensor<string, []>("op_52956_cast")];
            tensor<string, []> var_52958_equation_0 = const()[name = tensor<string, []>("op_52958_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52958_cast = einsum(equation = var_52958_equation_0, values = (var_52794_cast, var_52929_cast))[name = tensor<string, []>("op_52958_cast")];
            tensor<string, []> var_52960_equation_0 = const()[name = tensor<string, []>("op_52960_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52960_cast = einsum(equation = var_52960_equation_0, values = (var_52798_cast, var_52930_cast))[name = tensor<string, []>("op_52960_cast")];
            tensor<string, []> var_52962_equation_0 = const()[name = tensor<string, []>("op_52962_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52962_cast = einsum(equation = var_52962_equation_0, values = (var_52802_cast, var_52931_cast))[name = tensor<string, []>("op_52962_cast")];
            tensor<string, []> var_52964_equation_0 = const()[name = tensor<string, []>("op_52964_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52964_cast = einsum(equation = var_52964_equation_0, values = (var_52806_cast, var_52932_cast))[name = tensor<string, []>("op_52964_cast")];
            tensor<string, []> var_52966_equation_0 = const()[name = tensor<string, []>("op_52966_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52966_cast = einsum(equation = var_52966_equation_0, values = (var_52810_cast, var_52933_cast))[name = tensor<string, []>("op_52966_cast")];
            tensor<string, []> var_52968_equation_0 = const()[name = tensor<string, []>("op_52968_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52968_cast = einsum(equation = var_52968_equation_0, values = (var_52814_cast, var_52934_cast))[name = tensor<string, []>("op_52968_cast")];
            tensor<string, []> var_52970_equation_0 = const()[name = tensor<string, []>("op_52970_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52970_cast = einsum(equation = var_52970_equation_0, values = (var_52818_cast, var_52935_cast))[name = tensor<string, []>("op_52970_cast")];
            tensor<string, []> var_52972_equation_0 = const()[name = tensor<string, []>("op_52972_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52972_cast = einsum(equation = var_52972_equation_0, values = (var_52822_cast, var_52936_cast))[name = tensor<string, []>("op_52972_cast")];
            tensor<string, []> var_52974_equation_0 = const()[name = tensor<string, []>("op_52974_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52974_cast = einsum(equation = var_52974_equation_0, values = (var_52826_cast, var_52937_cast))[name = tensor<string, []>("op_52974_cast")];
            tensor<string, []> var_52976_equation_0 = const()[name = tensor<string, []>("op_52976_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52976_cast = einsum(equation = var_52976_equation_0, values = (var_52830_cast, var_52938_cast))[name = tensor<string, []>("op_52976_cast")];
            tensor<string, []> var_52978_equation_0 = const()[name = tensor<string, []>("op_52978_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52978_cast = einsum(equation = var_52978_equation_0, values = (var_52834_cast, var_52939_cast))[name = tensor<string, []>("op_52978_cast")];
            tensor<string, []> var_52980_equation_0 = const()[name = tensor<string, []>("op_52980_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_52980_cast = einsum(equation = var_52980_equation_0, values = (var_52838_cast, var_52940_cast))[name = tensor<string, []>("op_52980_cast")];
            tensor<bool, []> input_669_interleave_0 = const()[name = tensor<string, []>("input_669_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_669_cast = concat(axis = var_30385, interleave = input_669_interleave_0, values = (var_52942_cast, var_52944_cast, var_52946_cast, var_52948_cast, var_52950_cast, var_52952_cast, var_52954_cast, var_52956_cast, var_52958_cast, var_52960_cast, var_52962_cast, var_52964_cast, var_52966_cast, var_52968_cast, var_52970_cast, var_52972_cast, var_52974_cast, var_52976_cast, var_52978_cast, var_52980_cast))[name = tensor<string, []>("input_669_cast")];
            tensor<int32, [2]> var_52986 = const()[name = tensor<string, []>("op_52986"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_52988 = const()[name = tensor<string, []>("op_52988"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_52990_pad_type_0 = const()[name = tensor<string, []>("op_52990_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_52990_pad_0 = const()[name = tensor<string, []>("op_52990_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4459978688)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4463255552)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_52990_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_out_0_bias_to_fp16, dilations = var_52988, groups = var_30385, pad = var_52990_pad_0, pad_type = var_52990_pad_type_0, strides = var_52986, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn1_to_out_0_weight_to_fp16, x = input_669_cast)[name = tensor<string, []>("op_52990_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_351_cast = add(x = var_52990_cast, y = inputs_349_cast)[name = tensor<string, []>("inputs_351_cast")];
            tensor<int32, [1]> var_52994 = const()[name = tensor<string, []>("op_52994"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_351_cast = reduce_mean(axes = var_52994, keep_dims = var_30380, x = inputs_351_cast)[name = tensor<string, []>("channels_mean_351_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_351_cast = sub(x = inputs_351_cast, y = channels_mean_351_cast)[name = tensor<string, []>("zero_mean_351_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_351_cast = mul(x = zero_mean_351_cast, y = zero_mean_351_cast)[name = tensor<string, []>("zero_mean_sq_351_cast")];
            tensor<int32, [1]> var_52998 = const()[name = tensor<string, []>("op_52998"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_52999_cast = reduce_mean(axes = var_52998, keep_dims = var_30380, x = zero_mean_sq_351_cast)[name = tensor<string, []>("op_52999_cast")];
            tensor<fp16, []> var_53000_to_fp16 = const()[name = tensor<string, []>("op_53000_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_53001_cast = add(x = var_52999_cast, y = var_53000_to_fp16)[name = tensor<string, []>("op_53001_cast")];
            tensor<fp16, []> denom_351_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_351_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_351_cast = rsqrt(epsilon = denom_351_epsilon_0_to_fp16, x = var_53001_cast)[name = tensor<string, []>("denom_351_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_351_cast = mul(x = zero_mean_351_cast, y = denom_351_cast)[name = tensor<string, []>("out_351_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_53005_to_fp16 = const()[name = tensor<string, []>("op_53005_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4463258176)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_53006_cast = add(x = out_351_cast, y = var_53005_to_fp16)[name = tensor<string, []>("op_53006_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_53008_to_fp16 = const()[name = tensor<string, []>("op_53008_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4463260800)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_461_cast = mul(x = var_53006_cast, y = var_53008_to_fp16)[name = tensor<string, []>("hidden_states_461_cast")];
            tensor<int32, [2]> var_53015 = const()[name = tensor<string, []>("op_53015"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_53017 = const()[name = tensor<string, []>("op_53017"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_235_pad_type_0 = const()[name = tensor<string, []>("q_235_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_235_pad_0 = const()[name = tensor<string, []>("q_235_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4463263424)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_235_cast = conv(dilations = var_53017, groups = var_30385, pad = q_235_pad_0, pad_type = q_235_pad_type_0, strides = var_53015, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_q_weight_to_fp16, x = hidden_states_461_cast)[name = tensor<string, []>("q_235_cast")];
            tensor<int32, [2]> var_53021 = const()[name = tensor<string, []>("op_53021"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_53023 = const()[name = tensor<string, []>("op_53023"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_469_pad_type_0 = const()[name = tensor<string, []>("k_469_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_469_pad_0 = const()[name = tensor<string, []>("k_469_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4466540288)))];
            tensor<fp16, [2, 1280, 1, 77]> k_469_cast = conv(dilations = var_53023, groups = var_30385, pad = k_469_pad_0, pad_type = k_469_pad_type_0, strides = var_53021, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_469_cast")];
            tensor<int32, [2]> var_53027 = const()[name = tensor<string, []>("op_53027"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_53029 = const()[name = tensor<string, []>("op_53029"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_235_pad_type_0 = const()[name = tensor<string, []>("v_235_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_235_pad_0 = const()[name = tensor<string, []>("v_235_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4471783232)))];
            tensor<fp16, [2, 1280, 1, 77]> v_235_cast = conv(dilations = var_53029, groups = var_30385, pad = v_235_pad_0, pad_type = v_235_pad_type_0, strides = var_53027, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_235_cast")];
            tensor<int32, [4]> var_53033_begin_0 = const()[name = tensor<string, []>("op_53033_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_53033_end_0 = const()[name = tensor<string, []>("op_53033_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_53033_end_mask_0 = const()[name = tensor<string, []>("op_53033_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53033_cast = slice_by_index(begin = var_53033_begin_0, end = var_53033_end_0, end_mask = var_53033_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53033_cast")];
            tensor<int32, [4]> var_53037_begin_0 = const()[name = tensor<string, []>("op_53037_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_53037_end_0 = const()[name = tensor<string, []>("op_53037_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_53037_end_mask_0 = const()[name = tensor<string, []>("op_53037_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53037_cast = slice_by_index(begin = var_53037_begin_0, end = var_53037_end_0, end_mask = var_53037_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53037_cast")];
            tensor<int32, [4]> var_53041_begin_0 = const()[name = tensor<string, []>("op_53041_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_53041_end_0 = const()[name = tensor<string, []>("op_53041_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_53041_end_mask_0 = const()[name = tensor<string, []>("op_53041_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53041_cast = slice_by_index(begin = var_53041_begin_0, end = var_53041_end_0, end_mask = var_53041_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53041_cast")];
            tensor<int32, [4]> var_53045_begin_0 = const()[name = tensor<string, []>("op_53045_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_53045_end_0 = const()[name = tensor<string, []>("op_53045_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_53045_end_mask_0 = const()[name = tensor<string, []>("op_53045_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53045_cast = slice_by_index(begin = var_53045_begin_0, end = var_53045_end_0, end_mask = var_53045_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53045_cast")];
            tensor<int32, [4]> var_53049_begin_0 = const()[name = tensor<string, []>("op_53049_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_53049_end_0 = const()[name = tensor<string, []>("op_53049_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_53049_end_mask_0 = const()[name = tensor<string, []>("op_53049_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53049_cast = slice_by_index(begin = var_53049_begin_0, end = var_53049_end_0, end_mask = var_53049_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53049_cast")];
            tensor<int32, [4]> var_53053_begin_0 = const()[name = tensor<string, []>("op_53053_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_53053_end_0 = const()[name = tensor<string, []>("op_53053_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_53053_end_mask_0 = const()[name = tensor<string, []>("op_53053_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53053_cast = slice_by_index(begin = var_53053_begin_0, end = var_53053_end_0, end_mask = var_53053_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53053_cast")];
            tensor<int32, [4]> var_53057_begin_0 = const()[name = tensor<string, []>("op_53057_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_53057_end_0 = const()[name = tensor<string, []>("op_53057_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_53057_end_mask_0 = const()[name = tensor<string, []>("op_53057_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53057_cast = slice_by_index(begin = var_53057_begin_0, end = var_53057_end_0, end_mask = var_53057_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53057_cast")];
            tensor<int32, [4]> var_53061_begin_0 = const()[name = tensor<string, []>("op_53061_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_53061_end_0 = const()[name = tensor<string, []>("op_53061_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_53061_end_mask_0 = const()[name = tensor<string, []>("op_53061_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53061_cast = slice_by_index(begin = var_53061_begin_0, end = var_53061_end_0, end_mask = var_53061_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53061_cast")];
            tensor<int32, [4]> var_53065_begin_0 = const()[name = tensor<string, []>("op_53065_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_53065_end_0 = const()[name = tensor<string, []>("op_53065_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_53065_end_mask_0 = const()[name = tensor<string, []>("op_53065_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53065_cast = slice_by_index(begin = var_53065_begin_0, end = var_53065_end_0, end_mask = var_53065_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53065_cast")];
            tensor<int32, [4]> var_53069_begin_0 = const()[name = tensor<string, []>("op_53069_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_53069_end_0 = const()[name = tensor<string, []>("op_53069_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_53069_end_mask_0 = const()[name = tensor<string, []>("op_53069_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53069_cast = slice_by_index(begin = var_53069_begin_0, end = var_53069_end_0, end_mask = var_53069_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53069_cast")];
            tensor<int32, [4]> var_53073_begin_0 = const()[name = tensor<string, []>("op_53073_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_53073_end_0 = const()[name = tensor<string, []>("op_53073_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_53073_end_mask_0 = const()[name = tensor<string, []>("op_53073_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53073_cast = slice_by_index(begin = var_53073_begin_0, end = var_53073_end_0, end_mask = var_53073_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53073_cast")];
            tensor<int32, [4]> var_53077_begin_0 = const()[name = tensor<string, []>("op_53077_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_53077_end_0 = const()[name = tensor<string, []>("op_53077_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_53077_end_mask_0 = const()[name = tensor<string, []>("op_53077_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53077_cast = slice_by_index(begin = var_53077_begin_0, end = var_53077_end_0, end_mask = var_53077_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53077_cast")];
            tensor<int32, [4]> var_53081_begin_0 = const()[name = tensor<string, []>("op_53081_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_53081_end_0 = const()[name = tensor<string, []>("op_53081_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_53081_end_mask_0 = const()[name = tensor<string, []>("op_53081_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53081_cast = slice_by_index(begin = var_53081_begin_0, end = var_53081_end_0, end_mask = var_53081_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53081_cast")];
            tensor<int32, [4]> var_53085_begin_0 = const()[name = tensor<string, []>("op_53085_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_53085_end_0 = const()[name = tensor<string, []>("op_53085_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_53085_end_mask_0 = const()[name = tensor<string, []>("op_53085_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53085_cast = slice_by_index(begin = var_53085_begin_0, end = var_53085_end_0, end_mask = var_53085_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53085_cast")];
            tensor<int32, [4]> var_53089_begin_0 = const()[name = tensor<string, []>("op_53089_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_53089_end_0 = const()[name = tensor<string, []>("op_53089_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_53089_end_mask_0 = const()[name = tensor<string, []>("op_53089_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53089_cast = slice_by_index(begin = var_53089_begin_0, end = var_53089_end_0, end_mask = var_53089_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53089_cast")];
            tensor<int32, [4]> var_53093_begin_0 = const()[name = tensor<string, []>("op_53093_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_53093_end_0 = const()[name = tensor<string, []>("op_53093_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_53093_end_mask_0 = const()[name = tensor<string, []>("op_53093_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53093_cast = slice_by_index(begin = var_53093_begin_0, end = var_53093_end_0, end_mask = var_53093_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53093_cast")];
            tensor<int32, [4]> var_53097_begin_0 = const()[name = tensor<string, []>("op_53097_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_53097_end_0 = const()[name = tensor<string, []>("op_53097_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_53097_end_mask_0 = const()[name = tensor<string, []>("op_53097_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53097_cast = slice_by_index(begin = var_53097_begin_0, end = var_53097_end_0, end_mask = var_53097_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53097_cast")];
            tensor<int32, [4]> var_53101_begin_0 = const()[name = tensor<string, []>("op_53101_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_53101_end_0 = const()[name = tensor<string, []>("op_53101_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_53101_end_mask_0 = const()[name = tensor<string, []>("op_53101_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53101_cast = slice_by_index(begin = var_53101_begin_0, end = var_53101_end_0, end_mask = var_53101_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53101_cast")];
            tensor<int32, [4]> var_53105_begin_0 = const()[name = tensor<string, []>("op_53105_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_53105_end_0 = const()[name = tensor<string, []>("op_53105_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_53105_end_mask_0 = const()[name = tensor<string, []>("op_53105_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53105_cast = slice_by_index(begin = var_53105_begin_0, end = var_53105_end_0, end_mask = var_53105_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53105_cast")];
            tensor<int32, [4]> var_53109_begin_0 = const()[name = tensor<string, []>("op_53109_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_53109_end_0 = const()[name = tensor<string, []>("op_53109_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_53109_end_mask_0 = const()[name = tensor<string, []>("op_53109_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53109_cast = slice_by_index(begin = var_53109_begin_0, end = var_53109_end_0, end_mask = var_53109_end_mask_0, x = q_235_cast)[name = tensor<string, []>("op_53109_cast")];
            tensor<int32, [4]> k_471_perm_0 = const()[name = tensor<string, []>("k_471_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_53116_begin_0 = const()[name = tensor<string, []>("op_53116_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_53116_end_0 = const()[name = tensor<string, []>("op_53116_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_53116_end_mask_0 = const()[name = tensor<string, []>("op_53116_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_22 = transpose(perm = k_471_perm_0, x = k_469_cast)[name = tensor<string, []>("transpose_22")];
            tensor<fp16, [2, 77, 1, 64]> var_53116_cast = slice_by_index(begin = var_53116_begin_0, end = var_53116_end_0, end_mask = var_53116_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53116_cast")];
            tensor<int32, [4]> var_53120_begin_0 = const()[name = tensor<string, []>("op_53120_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_53120_end_0 = const()[name = tensor<string, []>("op_53120_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_53120_end_mask_0 = const()[name = tensor<string, []>("op_53120_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53120_cast = slice_by_index(begin = var_53120_begin_0, end = var_53120_end_0, end_mask = var_53120_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53120_cast")];
            tensor<int32, [4]> var_53124_begin_0 = const()[name = tensor<string, []>("op_53124_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_53124_end_0 = const()[name = tensor<string, []>("op_53124_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_53124_end_mask_0 = const()[name = tensor<string, []>("op_53124_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53124_cast = slice_by_index(begin = var_53124_begin_0, end = var_53124_end_0, end_mask = var_53124_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53124_cast")];
            tensor<int32, [4]> var_53128_begin_0 = const()[name = tensor<string, []>("op_53128_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_53128_end_0 = const()[name = tensor<string, []>("op_53128_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_53128_end_mask_0 = const()[name = tensor<string, []>("op_53128_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53128_cast = slice_by_index(begin = var_53128_begin_0, end = var_53128_end_0, end_mask = var_53128_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53128_cast")];
            tensor<int32, [4]> var_53132_begin_0 = const()[name = tensor<string, []>("op_53132_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_53132_end_0 = const()[name = tensor<string, []>("op_53132_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_53132_end_mask_0 = const()[name = tensor<string, []>("op_53132_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53132_cast = slice_by_index(begin = var_53132_begin_0, end = var_53132_end_0, end_mask = var_53132_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53132_cast")];
            tensor<int32, [4]> var_53136_begin_0 = const()[name = tensor<string, []>("op_53136_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_53136_end_0 = const()[name = tensor<string, []>("op_53136_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_53136_end_mask_0 = const()[name = tensor<string, []>("op_53136_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53136_cast = slice_by_index(begin = var_53136_begin_0, end = var_53136_end_0, end_mask = var_53136_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53136_cast")];
            tensor<int32, [4]> var_53140_begin_0 = const()[name = tensor<string, []>("op_53140_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_53140_end_0 = const()[name = tensor<string, []>("op_53140_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_53140_end_mask_0 = const()[name = tensor<string, []>("op_53140_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53140_cast = slice_by_index(begin = var_53140_begin_0, end = var_53140_end_0, end_mask = var_53140_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53140_cast")];
            tensor<int32, [4]> var_53144_begin_0 = const()[name = tensor<string, []>("op_53144_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_53144_end_0 = const()[name = tensor<string, []>("op_53144_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_53144_end_mask_0 = const()[name = tensor<string, []>("op_53144_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53144_cast = slice_by_index(begin = var_53144_begin_0, end = var_53144_end_0, end_mask = var_53144_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53144_cast")];
            tensor<int32, [4]> var_53148_begin_0 = const()[name = tensor<string, []>("op_53148_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_53148_end_0 = const()[name = tensor<string, []>("op_53148_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_53148_end_mask_0 = const()[name = tensor<string, []>("op_53148_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53148_cast = slice_by_index(begin = var_53148_begin_0, end = var_53148_end_0, end_mask = var_53148_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53148_cast")];
            tensor<int32, [4]> var_53152_begin_0 = const()[name = tensor<string, []>("op_53152_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_53152_end_0 = const()[name = tensor<string, []>("op_53152_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_53152_end_mask_0 = const()[name = tensor<string, []>("op_53152_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53152_cast = slice_by_index(begin = var_53152_begin_0, end = var_53152_end_0, end_mask = var_53152_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53152_cast")];
            tensor<int32, [4]> var_53156_begin_0 = const()[name = tensor<string, []>("op_53156_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_53156_end_0 = const()[name = tensor<string, []>("op_53156_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_53156_end_mask_0 = const()[name = tensor<string, []>("op_53156_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53156_cast = slice_by_index(begin = var_53156_begin_0, end = var_53156_end_0, end_mask = var_53156_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53156_cast")];
            tensor<int32, [4]> var_53160_begin_0 = const()[name = tensor<string, []>("op_53160_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_53160_end_0 = const()[name = tensor<string, []>("op_53160_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_53160_end_mask_0 = const()[name = tensor<string, []>("op_53160_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53160_cast = slice_by_index(begin = var_53160_begin_0, end = var_53160_end_0, end_mask = var_53160_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53160_cast")];
            tensor<int32, [4]> var_53164_begin_0 = const()[name = tensor<string, []>("op_53164_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_53164_end_0 = const()[name = tensor<string, []>("op_53164_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_53164_end_mask_0 = const()[name = tensor<string, []>("op_53164_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53164_cast = slice_by_index(begin = var_53164_begin_0, end = var_53164_end_0, end_mask = var_53164_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53164_cast")];
            tensor<int32, [4]> var_53168_begin_0 = const()[name = tensor<string, []>("op_53168_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_53168_end_0 = const()[name = tensor<string, []>("op_53168_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_53168_end_mask_0 = const()[name = tensor<string, []>("op_53168_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53168_cast = slice_by_index(begin = var_53168_begin_0, end = var_53168_end_0, end_mask = var_53168_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53168_cast")];
            tensor<int32, [4]> var_53172_begin_0 = const()[name = tensor<string, []>("op_53172_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_53172_end_0 = const()[name = tensor<string, []>("op_53172_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_53172_end_mask_0 = const()[name = tensor<string, []>("op_53172_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53172_cast = slice_by_index(begin = var_53172_begin_0, end = var_53172_end_0, end_mask = var_53172_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53172_cast")];
            tensor<int32, [4]> var_53176_begin_0 = const()[name = tensor<string, []>("op_53176_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_53176_end_0 = const()[name = tensor<string, []>("op_53176_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_53176_end_mask_0 = const()[name = tensor<string, []>("op_53176_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53176_cast = slice_by_index(begin = var_53176_begin_0, end = var_53176_end_0, end_mask = var_53176_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53176_cast")];
            tensor<int32, [4]> var_53180_begin_0 = const()[name = tensor<string, []>("op_53180_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_53180_end_0 = const()[name = tensor<string, []>("op_53180_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_53180_end_mask_0 = const()[name = tensor<string, []>("op_53180_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53180_cast = slice_by_index(begin = var_53180_begin_0, end = var_53180_end_0, end_mask = var_53180_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53180_cast")];
            tensor<int32, [4]> var_53184_begin_0 = const()[name = tensor<string, []>("op_53184_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_53184_end_0 = const()[name = tensor<string, []>("op_53184_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_53184_end_mask_0 = const()[name = tensor<string, []>("op_53184_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53184_cast = slice_by_index(begin = var_53184_begin_0, end = var_53184_end_0, end_mask = var_53184_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53184_cast")];
            tensor<int32, [4]> var_53188_begin_0 = const()[name = tensor<string, []>("op_53188_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_53188_end_0 = const()[name = tensor<string, []>("op_53188_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_53188_end_mask_0 = const()[name = tensor<string, []>("op_53188_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53188_cast = slice_by_index(begin = var_53188_begin_0, end = var_53188_end_0, end_mask = var_53188_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53188_cast")];
            tensor<int32, [4]> var_53192_begin_0 = const()[name = tensor<string, []>("op_53192_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_53192_end_0 = const()[name = tensor<string, []>("op_53192_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_53192_end_mask_0 = const()[name = tensor<string, []>("op_53192_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_53192_cast = slice_by_index(begin = var_53192_begin_0, end = var_53192_end_0, end_mask = var_53192_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_53192_cast")];
            tensor<int32, [4]> var_53194_begin_0 = const()[name = tensor<string, []>("op_53194_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_53194_end_0 = const()[name = tensor<string, []>("op_53194_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_53194_end_mask_0 = const()[name = tensor<string, []>("op_53194_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53194_cast = slice_by_index(begin = var_53194_begin_0, end = var_53194_end_0, end_mask = var_53194_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53194_cast")];
            tensor<int32, [4]> var_53198_begin_0 = const()[name = tensor<string, []>("op_53198_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_53198_end_0 = const()[name = tensor<string, []>("op_53198_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_53198_end_mask_0 = const()[name = tensor<string, []>("op_53198_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53198_cast = slice_by_index(begin = var_53198_begin_0, end = var_53198_end_0, end_mask = var_53198_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53198_cast")];
            tensor<int32, [4]> var_53202_begin_0 = const()[name = tensor<string, []>("op_53202_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_53202_end_0 = const()[name = tensor<string, []>("op_53202_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_53202_end_mask_0 = const()[name = tensor<string, []>("op_53202_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53202_cast = slice_by_index(begin = var_53202_begin_0, end = var_53202_end_0, end_mask = var_53202_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53202_cast")];
            tensor<int32, [4]> var_53206_begin_0 = const()[name = tensor<string, []>("op_53206_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_53206_end_0 = const()[name = tensor<string, []>("op_53206_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_53206_end_mask_0 = const()[name = tensor<string, []>("op_53206_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53206_cast = slice_by_index(begin = var_53206_begin_0, end = var_53206_end_0, end_mask = var_53206_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53206_cast")];
            tensor<int32, [4]> var_53210_begin_0 = const()[name = tensor<string, []>("op_53210_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_53210_end_0 = const()[name = tensor<string, []>("op_53210_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_53210_end_mask_0 = const()[name = tensor<string, []>("op_53210_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53210_cast = slice_by_index(begin = var_53210_begin_0, end = var_53210_end_0, end_mask = var_53210_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53210_cast")];
            tensor<int32, [4]> var_53214_begin_0 = const()[name = tensor<string, []>("op_53214_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_53214_end_0 = const()[name = tensor<string, []>("op_53214_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_53214_end_mask_0 = const()[name = tensor<string, []>("op_53214_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53214_cast = slice_by_index(begin = var_53214_begin_0, end = var_53214_end_0, end_mask = var_53214_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53214_cast")];
            tensor<int32, [4]> var_53218_begin_0 = const()[name = tensor<string, []>("op_53218_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_53218_end_0 = const()[name = tensor<string, []>("op_53218_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_53218_end_mask_0 = const()[name = tensor<string, []>("op_53218_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53218_cast = slice_by_index(begin = var_53218_begin_0, end = var_53218_end_0, end_mask = var_53218_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53218_cast")];
            tensor<int32, [4]> var_53222_begin_0 = const()[name = tensor<string, []>("op_53222_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_53222_end_0 = const()[name = tensor<string, []>("op_53222_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_53222_end_mask_0 = const()[name = tensor<string, []>("op_53222_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53222_cast = slice_by_index(begin = var_53222_begin_0, end = var_53222_end_0, end_mask = var_53222_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53222_cast")];
            tensor<int32, [4]> var_53226_begin_0 = const()[name = tensor<string, []>("op_53226_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_53226_end_0 = const()[name = tensor<string, []>("op_53226_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_53226_end_mask_0 = const()[name = tensor<string, []>("op_53226_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53226_cast = slice_by_index(begin = var_53226_begin_0, end = var_53226_end_0, end_mask = var_53226_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53226_cast")];
            tensor<int32, [4]> var_53230_begin_0 = const()[name = tensor<string, []>("op_53230_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_53230_end_0 = const()[name = tensor<string, []>("op_53230_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_53230_end_mask_0 = const()[name = tensor<string, []>("op_53230_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53230_cast = slice_by_index(begin = var_53230_begin_0, end = var_53230_end_0, end_mask = var_53230_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53230_cast")];
            tensor<int32, [4]> var_53234_begin_0 = const()[name = tensor<string, []>("op_53234_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_53234_end_0 = const()[name = tensor<string, []>("op_53234_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_53234_end_mask_0 = const()[name = tensor<string, []>("op_53234_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53234_cast = slice_by_index(begin = var_53234_begin_0, end = var_53234_end_0, end_mask = var_53234_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53234_cast")];
            tensor<int32, [4]> var_53238_begin_0 = const()[name = tensor<string, []>("op_53238_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_53238_end_0 = const()[name = tensor<string, []>("op_53238_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_53238_end_mask_0 = const()[name = tensor<string, []>("op_53238_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53238_cast = slice_by_index(begin = var_53238_begin_0, end = var_53238_end_0, end_mask = var_53238_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53238_cast")];
            tensor<int32, [4]> var_53242_begin_0 = const()[name = tensor<string, []>("op_53242_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_53242_end_0 = const()[name = tensor<string, []>("op_53242_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_53242_end_mask_0 = const()[name = tensor<string, []>("op_53242_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53242_cast = slice_by_index(begin = var_53242_begin_0, end = var_53242_end_0, end_mask = var_53242_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53242_cast")];
            tensor<int32, [4]> var_53246_begin_0 = const()[name = tensor<string, []>("op_53246_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_53246_end_0 = const()[name = tensor<string, []>("op_53246_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_53246_end_mask_0 = const()[name = tensor<string, []>("op_53246_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53246_cast = slice_by_index(begin = var_53246_begin_0, end = var_53246_end_0, end_mask = var_53246_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53246_cast")];
            tensor<int32, [4]> var_53250_begin_0 = const()[name = tensor<string, []>("op_53250_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_53250_end_0 = const()[name = tensor<string, []>("op_53250_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_53250_end_mask_0 = const()[name = tensor<string, []>("op_53250_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53250_cast = slice_by_index(begin = var_53250_begin_0, end = var_53250_end_0, end_mask = var_53250_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53250_cast")];
            tensor<int32, [4]> var_53254_begin_0 = const()[name = tensor<string, []>("op_53254_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_53254_end_0 = const()[name = tensor<string, []>("op_53254_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_53254_end_mask_0 = const()[name = tensor<string, []>("op_53254_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53254_cast = slice_by_index(begin = var_53254_begin_0, end = var_53254_end_0, end_mask = var_53254_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53254_cast")];
            tensor<int32, [4]> var_53258_begin_0 = const()[name = tensor<string, []>("op_53258_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_53258_end_0 = const()[name = tensor<string, []>("op_53258_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_53258_end_mask_0 = const()[name = tensor<string, []>("op_53258_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53258_cast = slice_by_index(begin = var_53258_begin_0, end = var_53258_end_0, end_mask = var_53258_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53258_cast")];
            tensor<int32, [4]> var_53262_begin_0 = const()[name = tensor<string, []>("op_53262_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_53262_end_0 = const()[name = tensor<string, []>("op_53262_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_53262_end_mask_0 = const()[name = tensor<string, []>("op_53262_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53262_cast = slice_by_index(begin = var_53262_begin_0, end = var_53262_end_0, end_mask = var_53262_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53262_cast")];
            tensor<int32, [4]> var_53266_begin_0 = const()[name = tensor<string, []>("op_53266_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_53266_end_0 = const()[name = tensor<string, []>("op_53266_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_53266_end_mask_0 = const()[name = tensor<string, []>("op_53266_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53266_cast = slice_by_index(begin = var_53266_begin_0, end = var_53266_end_0, end_mask = var_53266_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53266_cast")];
            tensor<int32, [4]> var_53270_begin_0 = const()[name = tensor<string, []>("op_53270_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_53270_end_0 = const()[name = tensor<string, []>("op_53270_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_53270_end_mask_0 = const()[name = tensor<string, []>("op_53270_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_53270_cast = slice_by_index(begin = var_53270_begin_0, end = var_53270_end_0, end_mask = var_53270_end_mask_0, x = v_235_cast)[name = tensor<string, []>("op_53270_cast")];
            tensor<string, []> var_53274_equation_0 = const()[name = tensor<string, []>("op_53274_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53274_cast = einsum(equation = var_53274_equation_0, values = (var_53116_cast, var_53033_cast))[name = tensor<string, []>("op_53274_cast")];
            tensor<fp16, []> var_53275_to_fp16 = const()[name = tensor<string, []>("op_53275_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4521_cast = mul(x = var_53274_cast, y = var_53275_to_fp16)[name = tensor<string, []>("aw_4521_cast")];
            tensor<string, []> var_53278_equation_0 = const()[name = tensor<string, []>("op_53278_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53278_cast = einsum(equation = var_53278_equation_0, values = (var_53120_cast, var_53037_cast))[name = tensor<string, []>("op_53278_cast")];
            tensor<fp16, []> var_53279_to_fp16 = const()[name = tensor<string, []>("op_53279_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4523_cast = mul(x = var_53278_cast, y = var_53279_to_fp16)[name = tensor<string, []>("aw_4523_cast")];
            tensor<string, []> var_53282_equation_0 = const()[name = tensor<string, []>("op_53282_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53282_cast = einsum(equation = var_53282_equation_0, values = (var_53124_cast, var_53041_cast))[name = tensor<string, []>("op_53282_cast")];
            tensor<fp16, []> var_53283_to_fp16 = const()[name = tensor<string, []>("op_53283_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4525_cast = mul(x = var_53282_cast, y = var_53283_to_fp16)[name = tensor<string, []>("aw_4525_cast")];
            tensor<string, []> var_53286_equation_0 = const()[name = tensor<string, []>("op_53286_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53286_cast = einsum(equation = var_53286_equation_0, values = (var_53128_cast, var_53045_cast))[name = tensor<string, []>("op_53286_cast")];
            tensor<fp16, []> var_53287_to_fp16 = const()[name = tensor<string, []>("op_53287_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4527_cast = mul(x = var_53286_cast, y = var_53287_to_fp16)[name = tensor<string, []>("aw_4527_cast")];
            tensor<string, []> var_53290_equation_0 = const()[name = tensor<string, []>("op_53290_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53290_cast = einsum(equation = var_53290_equation_0, values = (var_53132_cast, var_53049_cast))[name = tensor<string, []>("op_53290_cast")];
            tensor<fp16, []> var_53291_to_fp16 = const()[name = tensor<string, []>("op_53291_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4529_cast = mul(x = var_53290_cast, y = var_53291_to_fp16)[name = tensor<string, []>("aw_4529_cast")];
            tensor<string, []> var_53294_equation_0 = const()[name = tensor<string, []>("op_53294_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53294_cast = einsum(equation = var_53294_equation_0, values = (var_53136_cast, var_53053_cast))[name = tensor<string, []>("op_53294_cast")];
            tensor<fp16, []> var_53295_to_fp16 = const()[name = tensor<string, []>("op_53295_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4531_cast = mul(x = var_53294_cast, y = var_53295_to_fp16)[name = tensor<string, []>("aw_4531_cast")];
            tensor<string, []> var_53298_equation_0 = const()[name = tensor<string, []>("op_53298_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53298_cast = einsum(equation = var_53298_equation_0, values = (var_53140_cast, var_53057_cast))[name = tensor<string, []>("op_53298_cast")];
            tensor<fp16, []> var_53299_to_fp16 = const()[name = tensor<string, []>("op_53299_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4533_cast = mul(x = var_53298_cast, y = var_53299_to_fp16)[name = tensor<string, []>("aw_4533_cast")];
            tensor<string, []> var_53302_equation_0 = const()[name = tensor<string, []>("op_53302_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53302_cast = einsum(equation = var_53302_equation_0, values = (var_53144_cast, var_53061_cast))[name = tensor<string, []>("op_53302_cast")];
            tensor<fp16, []> var_53303_to_fp16 = const()[name = tensor<string, []>("op_53303_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4535_cast = mul(x = var_53302_cast, y = var_53303_to_fp16)[name = tensor<string, []>("aw_4535_cast")];
            tensor<string, []> var_53306_equation_0 = const()[name = tensor<string, []>("op_53306_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53306_cast = einsum(equation = var_53306_equation_0, values = (var_53148_cast, var_53065_cast))[name = tensor<string, []>("op_53306_cast")];
            tensor<fp16, []> var_53307_to_fp16 = const()[name = tensor<string, []>("op_53307_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4537_cast = mul(x = var_53306_cast, y = var_53307_to_fp16)[name = tensor<string, []>("aw_4537_cast")];
            tensor<string, []> var_53310_equation_0 = const()[name = tensor<string, []>("op_53310_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53310_cast = einsum(equation = var_53310_equation_0, values = (var_53152_cast, var_53069_cast))[name = tensor<string, []>("op_53310_cast")];
            tensor<fp16, []> var_53311_to_fp16 = const()[name = tensor<string, []>("op_53311_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4539_cast = mul(x = var_53310_cast, y = var_53311_to_fp16)[name = tensor<string, []>("aw_4539_cast")];
            tensor<string, []> var_53314_equation_0 = const()[name = tensor<string, []>("op_53314_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53314_cast = einsum(equation = var_53314_equation_0, values = (var_53156_cast, var_53073_cast))[name = tensor<string, []>("op_53314_cast")];
            tensor<fp16, []> var_53315_to_fp16 = const()[name = tensor<string, []>("op_53315_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4541_cast = mul(x = var_53314_cast, y = var_53315_to_fp16)[name = tensor<string, []>("aw_4541_cast")];
            tensor<string, []> var_53318_equation_0 = const()[name = tensor<string, []>("op_53318_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53318_cast = einsum(equation = var_53318_equation_0, values = (var_53160_cast, var_53077_cast))[name = tensor<string, []>("op_53318_cast")];
            tensor<fp16, []> var_53319_to_fp16 = const()[name = tensor<string, []>("op_53319_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4543_cast = mul(x = var_53318_cast, y = var_53319_to_fp16)[name = tensor<string, []>("aw_4543_cast")];
            tensor<string, []> var_53322_equation_0 = const()[name = tensor<string, []>("op_53322_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53322_cast = einsum(equation = var_53322_equation_0, values = (var_53164_cast, var_53081_cast))[name = tensor<string, []>("op_53322_cast")];
            tensor<fp16, []> var_53323_to_fp16 = const()[name = tensor<string, []>("op_53323_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4545_cast = mul(x = var_53322_cast, y = var_53323_to_fp16)[name = tensor<string, []>("aw_4545_cast")];
            tensor<string, []> var_53326_equation_0 = const()[name = tensor<string, []>("op_53326_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53326_cast = einsum(equation = var_53326_equation_0, values = (var_53168_cast, var_53085_cast))[name = tensor<string, []>("op_53326_cast")];
            tensor<fp16, []> var_53327_to_fp16 = const()[name = tensor<string, []>("op_53327_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4547_cast = mul(x = var_53326_cast, y = var_53327_to_fp16)[name = tensor<string, []>("aw_4547_cast")];
            tensor<string, []> var_53330_equation_0 = const()[name = tensor<string, []>("op_53330_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53330_cast = einsum(equation = var_53330_equation_0, values = (var_53172_cast, var_53089_cast))[name = tensor<string, []>("op_53330_cast")];
            tensor<fp16, []> var_53331_to_fp16 = const()[name = tensor<string, []>("op_53331_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4549_cast = mul(x = var_53330_cast, y = var_53331_to_fp16)[name = tensor<string, []>("aw_4549_cast")];
            tensor<string, []> var_53334_equation_0 = const()[name = tensor<string, []>("op_53334_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53334_cast = einsum(equation = var_53334_equation_0, values = (var_53176_cast, var_53093_cast))[name = tensor<string, []>("op_53334_cast")];
            tensor<fp16, []> var_53335_to_fp16 = const()[name = tensor<string, []>("op_53335_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4551_cast = mul(x = var_53334_cast, y = var_53335_to_fp16)[name = tensor<string, []>("aw_4551_cast")];
            tensor<string, []> var_53338_equation_0 = const()[name = tensor<string, []>("op_53338_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53338_cast = einsum(equation = var_53338_equation_0, values = (var_53180_cast, var_53097_cast))[name = tensor<string, []>("op_53338_cast")];
            tensor<fp16, []> var_53339_to_fp16 = const()[name = tensor<string, []>("op_53339_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4553_cast = mul(x = var_53338_cast, y = var_53339_to_fp16)[name = tensor<string, []>("aw_4553_cast")];
            tensor<string, []> var_53342_equation_0 = const()[name = tensor<string, []>("op_53342_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53342_cast = einsum(equation = var_53342_equation_0, values = (var_53184_cast, var_53101_cast))[name = tensor<string, []>("op_53342_cast")];
            tensor<fp16, []> var_53343_to_fp16 = const()[name = tensor<string, []>("op_53343_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4555_cast = mul(x = var_53342_cast, y = var_53343_to_fp16)[name = tensor<string, []>("aw_4555_cast")];
            tensor<string, []> var_53346_equation_0 = const()[name = tensor<string, []>("op_53346_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53346_cast = einsum(equation = var_53346_equation_0, values = (var_53188_cast, var_53105_cast))[name = tensor<string, []>("op_53346_cast")];
            tensor<fp16, []> var_53347_to_fp16 = const()[name = tensor<string, []>("op_53347_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4557_cast = mul(x = var_53346_cast, y = var_53347_to_fp16)[name = tensor<string, []>("aw_4557_cast")];
            tensor<string, []> var_53350_equation_0 = const()[name = tensor<string, []>("op_53350_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_53350_cast = einsum(equation = var_53350_equation_0, values = (var_53192_cast, var_53109_cast))[name = tensor<string, []>("op_53350_cast")];
            tensor<fp16, []> var_53351_to_fp16 = const()[name = tensor<string, []>("op_53351_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4559_cast = mul(x = var_53350_cast, y = var_53351_to_fp16)[name = tensor<string, []>("aw_4559_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53353_cast = softmax(axis = var_30385, x = aw_4521_cast)[name = tensor<string, []>("op_53353_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53354_cast = softmax(axis = var_30385, x = aw_4523_cast)[name = tensor<string, []>("op_53354_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53355_cast = softmax(axis = var_30385, x = aw_4525_cast)[name = tensor<string, []>("op_53355_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53356_cast = softmax(axis = var_30385, x = aw_4527_cast)[name = tensor<string, []>("op_53356_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53357_cast = softmax(axis = var_30385, x = aw_4529_cast)[name = tensor<string, []>("op_53357_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53358_cast = softmax(axis = var_30385, x = aw_4531_cast)[name = tensor<string, []>("op_53358_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53359_cast = softmax(axis = var_30385, x = aw_4533_cast)[name = tensor<string, []>("op_53359_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53360_cast = softmax(axis = var_30385, x = aw_4535_cast)[name = tensor<string, []>("op_53360_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53361_cast = softmax(axis = var_30385, x = aw_4537_cast)[name = tensor<string, []>("op_53361_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53362_cast = softmax(axis = var_30385, x = aw_4539_cast)[name = tensor<string, []>("op_53362_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53363_cast = softmax(axis = var_30385, x = aw_4541_cast)[name = tensor<string, []>("op_53363_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53364_cast = softmax(axis = var_30385, x = aw_4543_cast)[name = tensor<string, []>("op_53364_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53365_cast = softmax(axis = var_30385, x = aw_4545_cast)[name = tensor<string, []>("op_53365_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53366_cast = softmax(axis = var_30385, x = aw_4547_cast)[name = tensor<string, []>("op_53366_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53367_cast = softmax(axis = var_30385, x = aw_4549_cast)[name = tensor<string, []>("op_53367_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53368_cast = softmax(axis = var_30385, x = aw_4551_cast)[name = tensor<string, []>("op_53368_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53369_cast = softmax(axis = var_30385, x = aw_4553_cast)[name = tensor<string, []>("op_53369_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53370_cast = softmax(axis = var_30385, x = aw_4555_cast)[name = tensor<string, []>("op_53370_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53371_cast = softmax(axis = var_30385, x = aw_4557_cast)[name = tensor<string, []>("op_53371_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_53372_cast = softmax(axis = var_30385, x = aw_4559_cast)[name = tensor<string, []>("op_53372_cast")];
            tensor<string, []> var_53374_equation_0 = const()[name = tensor<string, []>("op_53374_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53374_cast = einsum(equation = var_53374_equation_0, values = (var_53194_cast, var_53353_cast))[name = tensor<string, []>("op_53374_cast")];
            tensor<string, []> var_53376_equation_0 = const()[name = tensor<string, []>("op_53376_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53376_cast = einsum(equation = var_53376_equation_0, values = (var_53198_cast, var_53354_cast))[name = tensor<string, []>("op_53376_cast")];
            tensor<string, []> var_53378_equation_0 = const()[name = tensor<string, []>("op_53378_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53378_cast = einsum(equation = var_53378_equation_0, values = (var_53202_cast, var_53355_cast))[name = tensor<string, []>("op_53378_cast")];
            tensor<string, []> var_53380_equation_0 = const()[name = tensor<string, []>("op_53380_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53380_cast = einsum(equation = var_53380_equation_0, values = (var_53206_cast, var_53356_cast))[name = tensor<string, []>("op_53380_cast")];
            tensor<string, []> var_53382_equation_0 = const()[name = tensor<string, []>("op_53382_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53382_cast = einsum(equation = var_53382_equation_0, values = (var_53210_cast, var_53357_cast))[name = tensor<string, []>("op_53382_cast")];
            tensor<string, []> var_53384_equation_0 = const()[name = tensor<string, []>("op_53384_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53384_cast = einsum(equation = var_53384_equation_0, values = (var_53214_cast, var_53358_cast))[name = tensor<string, []>("op_53384_cast")];
            tensor<string, []> var_53386_equation_0 = const()[name = tensor<string, []>("op_53386_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53386_cast = einsum(equation = var_53386_equation_0, values = (var_53218_cast, var_53359_cast))[name = tensor<string, []>("op_53386_cast")];
            tensor<string, []> var_53388_equation_0 = const()[name = tensor<string, []>("op_53388_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53388_cast = einsum(equation = var_53388_equation_0, values = (var_53222_cast, var_53360_cast))[name = tensor<string, []>("op_53388_cast")];
            tensor<string, []> var_53390_equation_0 = const()[name = tensor<string, []>("op_53390_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53390_cast = einsum(equation = var_53390_equation_0, values = (var_53226_cast, var_53361_cast))[name = tensor<string, []>("op_53390_cast")];
            tensor<string, []> var_53392_equation_0 = const()[name = tensor<string, []>("op_53392_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53392_cast = einsum(equation = var_53392_equation_0, values = (var_53230_cast, var_53362_cast))[name = tensor<string, []>("op_53392_cast")];
            tensor<string, []> var_53394_equation_0 = const()[name = tensor<string, []>("op_53394_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53394_cast = einsum(equation = var_53394_equation_0, values = (var_53234_cast, var_53363_cast))[name = tensor<string, []>("op_53394_cast")];
            tensor<string, []> var_53396_equation_0 = const()[name = tensor<string, []>("op_53396_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53396_cast = einsum(equation = var_53396_equation_0, values = (var_53238_cast, var_53364_cast))[name = tensor<string, []>("op_53396_cast")];
            tensor<string, []> var_53398_equation_0 = const()[name = tensor<string, []>("op_53398_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53398_cast = einsum(equation = var_53398_equation_0, values = (var_53242_cast, var_53365_cast))[name = tensor<string, []>("op_53398_cast")];
            tensor<string, []> var_53400_equation_0 = const()[name = tensor<string, []>("op_53400_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53400_cast = einsum(equation = var_53400_equation_0, values = (var_53246_cast, var_53366_cast))[name = tensor<string, []>("op_53400_cast")];
            tensor<string, []> var_53402_equation_0 = const()[name = tensor<string, []>("op_53402_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53402_cast = einsum(equation = var_53402_equation_0, values = (var_53250_cast, var_53367_cast))[name = tensor<string, []>("op_53402_cast")];
            tensor<string, []> var_53404_equation_0 = const()[name = tensor<string, []>("op_53404_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53404_cast = einsum(equation = var_53404_equation_0, values = (var_53254_cast, var_53368_cast))[name = tensor<string, []>("op_53404_cast")];
            tensor<string, []> var_53406_equation_0 = const()[name = tensor<string, []>("op_53406_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53406_cast = einsum(equation = var_53406_equation_0, values = (var_53258_cast, var_53369_cast))[name = tensor<string, []>("op_53406_cast")];
            tensor<string, []> var_53408_equation_0 = const()[name = tensor<string, []>("op_53408_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53408_cast = einsum(equation = var_53408_equation_0, values = (var_53262_cast, var_53370_cast))[name = tensor<string, []>("op_53408_cast")];
            tensor<string, []> var_53410_equation_0 = const()[name = tensor<string, []>("op_53410_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53410_cast = einsum(equation = var_53410_equation_0, values = (var_53266_cast, var_53371_cast))[name = tensor<string, []>("op_53410_cast")];
            tensor<string, []> var_53412_equation_0 = const()[name = tensor<string, []>("op_53412_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53412_cast = einsum(equation = var_53412_equation_0, values = (var_53270_cast, var_53372_cast))[name = tensor<string, []>("op_53412_cast")];
            tensor<bool, []> input_671_interleave_0 = const()[name = tensor<string, []>("input_671_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_671_cast = concat(axis = var_30385, interleave = input_671_interleave_0, values = (var_53374_cast, var_53376_cast, var_53378_cast, var_53380_cast, var_53382_cast, var_53384_cast, var_53386_cast, var_53388_cast, var_53390_cast, var_53392_cast, var_53394_cast, var_53396_cast, var_53398_cast, var_53400_cast, var_53402_cast, var_53404_cast, var_53406_cast, var_53408_cast, var_53410_cast, var_53412_cast))[name = tensor<string, []>("input_671_cast")];
            tensor<int32, [2]> var_53418 = const()[name = tensor<string, []>("op_53418"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_53420 = const()[name = tensor<string, []>("op_53420"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_53422_pad_type_0 = const()[name = tensor<string, []>("op_53422_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_53422_pad_0 = const()[name = tensor<string, []>("op_53422_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4477026176)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4480303040)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_53422_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_out_0_bias_to_fp16, dilations = var_53420, groups = var_30385, pad = var_53422_pad_0, pad_type = var_53422_pad_type_0, strides = var_53418, weight = up_blocks_0_attentions_2_transformer_blocks_4_attn2_to_out_0_weight_to_fp16, x = input_671_cast)[name = tensor<string, []>("op_53422_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_353_cast = add(x = var_53422_cast, y = inputs_351_cast)[name = tensor<string, []>("inputs_353_cast")];
            tensor<int32, [1]> var_53426 = const()[name = tensor<string, []>("op_53426"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_353_cast = reduce_mean(axes = var_53426, keep_dims = var_30380, x = inputs_353_cast)[name = tensor<string, []>("channels_mean_353_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_353_cast = sub(x = inputs_353_cast, y = channels_mean_353_cast)[name = tensor<string, []>("zero_mean_353_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_353_cast = mul(x = zero_mean_353_cast, y = zero_mean_353_cast)[name = tensor<string, []>("zero_mean_sq_353_cast")];
            tensor<int32, [1]> var_53430 = const()[name = tensor<string, []>("op_53430"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_53431_cast = reduce_mean(axes = var_53430, keep_dims = var_30380, x = zero_mean_sq_353_cast)[name = tensor<string, []>("op_53431_cast")];
            tensor<fp16, []> var_53432_to_fp16 = const()[name = tensor<string, []>("op_53432_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_53433_cast = add(x = var_53431_cast, y = var_53432_to_fp16)[name = tensor<string, []>("op_53433_cast")];
            tensor<fp16, []> denom_353_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_353_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_353_cast = rsqrt(epsilon = denom_353_epsilon_0_to_fp16, x = var_53433_cast)[name = tensor<string, []>("denom_353_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_353_cast = mul(x = zero_mean_353_cast, y = denom_353_cast)[name = tensor<string, []>("out_353_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_53437_to_fp16 = const()[name = tensor<string, []>("op_53437_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4480305664)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_53438_cast = add(x = out_353_cast, y = var_53437_to_fp16)[name = tensor<string, []>("op_53438_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_53440_to_fp16 = const()[name = tensor<string, []>("op_53440_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4480308288)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_673_cast = mul(x = var_53438_cast, y = var_53440_to_fp16)[name = tensor<string, []>("input_673_cast")];
            tensor<int32, [2]> var_53448 = const()[name = tensor<string, []>("op_53448"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_53450 = const()[name = tensor<string, []>("op_53450"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_53452_pad_type_0 = const()[name = tensor<string, []>("op_53452_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_53452_pad_0 = const()[name = tensor<string, []>("op_53452_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4480310912)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_2_transformer_blocks_4_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4506525376)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_53452_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_4_ff_net_0_proj_bias_to_fp16, dilations = var_53450, groups = var_30385, pad = var_53452_pad_0, pad_type = var_53452_pad_type_0, strides = var_53448, weight = up_blocks_0_attentions_2_transformer_blocks_4_ff_net_0_proj_weight_to_fp16, x = input_673_cast)[name = tensor<string, []>("op_53452_cast")];
            tensor<int32, [2]> var_53453_split_sizes_0 = const()[name = tensor<string, []>("op_53453_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_53453_axis_0 = const()[name = tensor<string, []>("op_53453_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_53453_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_53453_cast_1 = split(axis = var_53453_axis_0, split_sizes = var_53453_split_sizes_0, x = var_53452_cast)[name = tensor<string, []>("op_53453_cast")];
            tensor<string, []> var_53455_mode_0 = const()[name = tensor<string, []>("op_53455_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_53455_cast = gelu(mode = var_53455_mode_0, x = var_53453_cast_1)[name = tensor<string, []>("op_53455_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_675_cast = mul(x = var_53453_cast_0, y = var_53455_cast)[name = tensor<string, []>("input_675_cast")];
            tensor<int32, [2]> var_53459 = const()[name = tensor<string, []>("op_53459"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_53461 = const()[name = tensor<string, []>("op_53461"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_53463_pad_type_0 = const()[name = tensor<string, []>("op_53463_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_53463_pad_0 = const()[name = tensor<string, []>("op_53463_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_4_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4506545920)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_4_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_4_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4519653184)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_53463_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_4_ff_net_2_bias_to_fp16, dilations = var_53461, groups = var_30385, pad = var_53463_pad_0, pad_type = var_53463_pad_type_0, strides = var_53459, weight = up_blocks_0_attentions_2_transformer_blocks_4_ff_net_2_weight_to_fp16, x = input_675_cast)[name = tensor<string, []>("op_53463_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_355_cast = add(x = var_53463_cast, y = inputs_353_cast)[name = tensor<string, []>("inputs_355_cast")];
            tensor<int32, [1]> var_53473 = const()[name = tensor<string, []>("op_53473"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_355_cast = reduce_mean(axes = var_53473, keep_dims = var_30380, x = inputs_355_cast)[name = tensor<string, []>("channels_mean_355_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_355_cast = sub(x = inputs_355_cast, y = channels_mean_355_cast)[name = tensor<string, []>("zero_mean_355_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_355_cast = mul(x = zero_mean_355_cast, y = zero_mean_355_cast)[name = tensor<string, []>("zero_mean_sq_355_cast")];
            tensor<int32, [1]> var_53477 = const()[name = tensor<string, []>("op_53477"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_53478_cast = reduce_mean(axes = var_53477, keep_dims = var_30380, x = zero_mean_sq_355_cast)[name = tensor<string, []>("op_53478_cast")];
            tensor<fp16, []> var_53479_to_fp16 = const()[name = tensor<string, []>("op_53479_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_53480_cast = add(x = var_53478_cast, y = var_53479_to_fp16)[name = tensor<string, []>("op_53480_cast")];
            tensor<fp16, []> denom_355_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_355_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_355_cast = rsqrt(epsilon = denom_355_epsilon_0_to_fp16, x = var_53480_cast)[name = tensor<string, []>("denom_355_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_355_cast = mul(x = zero_mean_355_cast, y = denom_355_cast)[name = tensor<string, []>("out_355_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_53484_to_fp16 = const()[name = tensor<string, []>("op_53484_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4519655808)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_53485_cast = add(x = out_355_cast, y = var_53484_to_fp16)[name = tensor<string, []>("op_53485_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_53487_to_fp16 = const()[name = tensor<string, []>("op_53487_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4519658432)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_465_cast = mul(x = var_53485_cast, y = var_53487_to_fp16)[name = tensor<string, []>("hidden_states_465_cast")];
            tensor<int32, [2]> var_53494 = const()[name = tensor<string, []>("op_53494"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_53496 = const()[name = tensor<string, []>("op_53496"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_237_pad_type_0 = const()[name = tensor<string, []>("q_237_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_237_pad_0 = const()[name = tensor<string, []>("q_237_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4519661056)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_237_cast = conv(dilations = var_53496, groups = var_30385, pad = q_237_pad_0, pad_type = q_237_pad_type_0, strides = var_53494, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_q_weight_to_fp16, x = hidden_states_465_cast)[name = tensor<string, []>("q_237_cast")];
            tensor<int32, [2]> var_53500 = const()[name = tensor<string, []>("op_53500"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_53502 = const()[name = tensor<string, []>("op_53502"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_473_pad_type_0 = const()[name = tensor<string, []>("k_473_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_473_pad_0 = const()[name = tensor<string, []>("k_473_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4522937920)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_473_cast = conv(dilations = var_53502, groups = var_30385, pad = k_473_pad_0, pad_type = k_473_pad_type_0, strides = var_53500, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_k_weight_to_fp16, x = hidden_states_465_cast)[name = tensor<string, []>("k_473_cast")];
            tensor<int32, [2]> var_53506 = const()[name = tensor<string, []>("op_53506"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_53508 = const()[name = tensor<string, []>("op_53508"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_237_pad_type_0 = const()[name = tensor<string, []>("v_237_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_237_pad_0 = const()[name = tensor<string, []>("v_237_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4526214784)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_237_cast = conv(dilations = var_53508, groups = var_30385, pad = v_237_pad_0, pad_type = v_237_pad_type_0, strides = var_53506, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_v_weight_to_fp16, x = hidden_states_465_cast)[name = tensor<string, []>("v_237_cast")];
            tensor<int32, [4]> var_53512_begin_0 = const()[name = tensor<string, []>("op_53512_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_53512_end_0 = const()[name = tensor<string, []>("op_53512_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_53512_end_mask_0 = const()[name = tensor<string, []>("op_53512_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53512_cast = slice_by_index(begin = var_53512_begin_0, end = var_53512_end_0, end_mask = var_53512_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53512_cast")];
            tensor<int32, [4]> var_53516_begin_0 = const()[name = tensor<string, []>("op_53516_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_53516_end_0 = const()[name = tensor<string, []>("op_53516_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_53516_end_mask_0 = const()[name = tensor<string, []>("op_53516_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53516_cast = slice_by_index(begin = var_53516_begin_0, end = var_53516_end_0, end_mask = var_53516_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53516_cast")];
            tensor<int32, [4]> var_53520_begin_0 = const()[name = tensor<string, []>("op_53520_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_53520_end_0 = const()[name = tensor<string, []>("op_53520_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_53520_end_mask_0 = const()[name = tensor<string, []>("op_53520_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53520_cast = slice_by_index(begin = var_53520_begin_0, end = var_53520_end_0, end_mask = var_53520_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53520_cast")];
            tensor<int32, [4]> var_53524_begin_0 = const()[name = tensor<string, []>("op_53524_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_53524_end_0 = const()[name = tensor<string, []>("op_53524_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_53524_end_mask_0 = const()[name = tensor<string, []>("op_53524_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53524_cast = slice_by_index(begin = var_53524_begin_0, end = var_53524_end_0, end_mask = var_53524_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53524_cast")];
            tensor<int32, [4]> var_53528_begin_0 = const()[name = tensor<string, []>("op_53528_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_53528_end_0 = const()[name = tensor<string, []>("op_53528_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_53528_end_mask_0 = const()[name = tensor<string, []>("op_53528_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53528_cast = slice_by_index(begin = var_53528_begin_0, end = var_53528_end_0, end_mask = var_53528_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53528_cast")];
            tensor<int32, [4]> var_53532_begin_0 = const()[name = tensor<string, []>("op_53532_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_53532_end_0 = const()[name = tensor<string, []>("op_53532_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_53532_end_mask_0 = const()[name = tensor<string, []>("op_53532_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53532_cast = slice_by_index(begin = var_53532_begin_0, end = var_53532_end_0, end_mask = var_53532_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53532_cast")];
            tensor<int32, [4]> var_53536_begin_0 = const()[name = tensor<string, []>("op_53536_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_53536_end_0 = const()[name = tensor<string, []>("op_53536_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_53536_end_mask_0 = const()[name = tensor<string, []>("op_53536_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53536_cast = slice_by_index(begin = var_53536_begin_0, end = var_53536_end_0, end_mask = var_53536_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53536_cast")];
            tensor<int32, [4]> var_53540_begin_0 = const()[name = tensor<string, []>("op_53540_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_53540_end_0 = const()[name = tensor<string, []>("op_53540_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_53540_end_mask_0 = const()[name = tensor<string, []>("op_53540_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53540_cast = slice_by_index(begin = var_53540_begin_0, end = var_53540_end_0, end_mask = var_53540_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53540_cast")];
            tensor<int32, [4]> var_53544_begin_0 = const()[name = tensor<string, []>("op_53544_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_53544_end_0 = const()[name = tensor<string, []>("op_53544_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_53544_end_mask_0 = const()[name = tensor<string, []>("op_53544_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53544_cast = slice_by_index(begin = var_53544_begin_0, end = var_53544_end_0, end_mask = var_53544_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53544_cast")];
            tensor<int32, [4]> var_53548_begin_0 = const()[name = tensor<string, []>("op_53548_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_53548_end_0 = const()[name = tensor<string, []>("op_53548_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_53548_end_mask_0 = const()[name = tensor<string, []>("op_53548_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53548_cast = slice_by_index(begin = var_53548_begin_0, end = var_53548_end_0, end_mask = var_53548_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53548_cast")];
            tensor<int32, [4]> var_53552_begin_0 = const()[name = tensor<string, []>("op_53552_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_53552_end_0 = const()[name = tensor<string, []>("op_53552_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_53552_end_mask_0 = const()[name = tensor<string, []>("op_53552_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53552_cast = slice_by_index(begin = var_53552_begin_0, end = var_53552_end_0, end_mask = var_53552_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53552_cast")];
            tensor<int32, [4]> var_53556_begin_0 = const()[name = tensor<string, []>("op_53556_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_53556_end_0 = const()[name = tensor<string, []>("op_53556_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_53556_end_mask_0 = const()[name = tensor<string, []>("op_53556_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53556_cast = slice_by_index(begin = var_53556_begin_0, end = var_53556_end_0, end_mask = var_53556_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53556_cast")];
            tensor<int32, [4]> var_53560_begin_0 = const()[name = tensor<string, []>("op_53560_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_53560_end_0 = const()[name = tensor<string, []>("op_53560_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_53560_end_mask_0 = const()[name = tensor<string, []>("op_53560_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53560_cast = slice_by_index(begin = var_53560_begin_0, end = var_53560_end_0, end_mask = var_53560_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53560_cast")];
            tensor<int32, [4]> var_53564_begin_0 = const()[name = tensor<string, []>("op_53564_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_53564_end_0 = const()[name = tensor<string, []>("op_53564_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_53564_end_mask_0 = const()[name = tensor<string, []>("op_53564_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53564_cast = slice_by_index(begin = var_53564_begin_0, end = var_53564_end_0, end_mask = var_53564_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53564_cast")];
            tensor<int32, [4]> var_53568_begin_0 = const()[name = tensor<string, []>("op_53568_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_53568_end_0 = const()[name = tensor<string, []>("op_53568_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_53568_end_mask_0 = const()[name = tensor<string, []>("op_53568_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53568_cast = slice_by_index(begin = var_53568_begin_0, end = var_53568_end_0, end_mask = var_53568_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53568_cast")];
            tensor<int32, [4]> var_53572_begin_0 = const()[name = tensor<string, []>("op_53572_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_53572_end_0 = const()[name = tensor<string, []>("op_53572_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_53572_end_mask_0 = const()[name = tensor<string, []>("op_53572_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53572_cast = slice_by_index(begin = var_53572_begin_0, end = var_53572_end_0, end_mask = var_53572_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53572_cast")];
            tensor<int32, [4]> var_53576_begin_0 = const()[name = tensor<string, []>("op_53576_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_53576_end_0 = const()[name = tensor<string, []>("op_53576_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_53576_end_mask_0 = const()[name = tensor<string, []>("op_53576_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53576_cast = slice_by_index(begin = var_53576_begin_0, end = var_53576_end_0, end_mask = var_53576_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53576_cast")];
            tensor<int32, [4]> var_53580_begin_0 = const()[name = tensor<string, []>("op_53580_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_53580_end_0 = const()[name = tensor<string, []>("op_53580_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_53580_end_mask_0 = const()[name = tensor<string, []>("op_53580_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53580_cast = slice_by_index(begin = var_53580_begin_0, end = var_53580_end_0, end_mask = var_53580_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53580_cast")];
            tensor<int32, [4]> var_53584_begin_0 = const()[name = tensor<string, []>("op_53584_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_53584_end_0 = const()[name = tensor<string, []>("op_53584_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_53584_end_mask_0 = const()[name = tensor<string, []>("op_53584_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53584_cast = slice_by_index(begin = var_53584_begin_0, end = var_53584_end_0, end_mask = var_53584_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53584_cast")];
            tensor<int32, [4]> var_53588_begin_0 = const()[name = tensor<string, []>("op_53588_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_53588_end_0 = const()[name = tensor<string, []>("op_53588_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_53588_end_mask_0 = const()[name = tensor<string, []>("op_53588_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53588_cast = slice_by_index(begin = var_53588_begin_0, end = var_53588_end_0, end_mask = var_53588_end_mask_0, x = q_237_cast)[name = tensor<string, []>("op_53588_cast")];
            tensor<int32, [4]> k_475_perm_0 = const()[name = tensor<string, []>("k_475_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_53595_begin_0 = const()[name = tensor<string, []>("op_53595_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_53595_end_0 = const()[name = tensor<string, []>("op_53595_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_53595_end_mask_0 = const()[name = tensor<string, []>("op_53595_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_21 = transpose(perm = k_475_perm_0, x = k_473_cast)[name = tensor<string, []>("transpose_21")];
            tensor<fp16, [2, 1024, 1, 64]> var_53595_cast = slice_by_index(begin = var_53595_begin_0, end = var_53595_end_0, end_mask = var_53595_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53595_cast")];
            tensor<int32, [4]> var_53599_begin_0 = const()[name = tensor<string, []>("op_53599_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_53599_end_0 = const()[name = tensor<string, []>("op_53599_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_53599_end_mask_0 = const()[name = tensor<string, []>("op_53599_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53599_cast = slice_by_index(begin = var_53599_begin_0, end = var_53599_end_0, end_mask = var_53599_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53599_cast")];
            tensor<int32, [4]> var_53603_begin_0 = const()[name = tensor<string, []>("op_53603_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_53603_end_0 = const()[name = tensor<string, []>("op_53603_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_53603_end_mask_0 = const()[name = tensor<string, []>("op_53603_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53603_cast = slice_by_index(begin = var_53603_begin_0, end = var_53603_end_0, end_mask = var_53603_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53603_cast")];
            tensor<int32, [4]> var_53607_begin_0 = const()[name = tensor<string, []>("op_53607_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_53607_end_0 = const()[name = tensor<string, []>("op_53607_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_53607_end_mask_0 = const()[name = tensor<string, []>("op_53607_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53607_cast = slice_by_index(begin = var_53607_begin_0, end = var_53607_end_0, end_mask = var_53607_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53607_cast")];
            tensor<int32, [4]> var_53611_begin_0 = const()[name = tensor<string, []>("op_53611_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_53611_end_0 = const()[name = tensor<string, []>("op_53611_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_53611_end_mask_0 = const()[name = tensor<string, []>("op_53611_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53611_cast = slice_by_index(begin = var_53611_begin_0, end = var_53611_end_0, end_mask = var_53611_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53611_cast")];
            tensor<int32, [4]> var_53615_begin_0 = const()[name = tensor<string, []>("op_53615_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_53615_end_0 = const()[name = tensor<string, []>("op_53615_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_53615_end_mask_0 = const()[name = tensor<string, []>("op_53615_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53615_cast = slice_by_index(begin = var_53615_begin_0, end = var_53615_end_0, end_mask = var_53615_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53615_cast")];
            tensor<int32, [4]> var_53619_begin_0 = const()[name = tensor<string, []>("op_53619_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_53619_end_0 = const()[name = tensor<string, []>("op_53619_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_53619_end_mask_0 = const()[name = tensor<string, []>("op_53619_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53619_cast = slice_by_index(begin = var_53619_begin_0, end = var_53619_end_0, end_mask = var_53619_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53619_cast")];
            tensor<int32, [4]> var_53623_begin_0 = const()[name = tensor<string, []>("op_53623_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_53623_end_0 = const()[name = tensor<string, []>("op_53623_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_53623_end_mask_0 = const()[name = tensor<string, []>("op_53623_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53623_cast = slice_by_index(begin = var_53623_begin_0, end = var_53623_end_0, end_mask = var_53623_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53623_cast")];
            tensor<int32, [4]> var_53627_begin_0 = const()[name = tensor<string, []>("op_53627_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_53627_end_0 = const()[name = tensor<string, []>("op_53627_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_53627_end_mask_0 = const()[name = tensor<string, []>("op_53627_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53627_cast = slice_by_index(begin = var_53627_begin_0, end = var_53627_end_0, end_mask = var_53627_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53627_cast")];
            tensor<int32, [4]> var_53631_begin_0 = const()[name = tensor<string, []>("op_53631_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_53631_end_0 = const()[name = tensor<string, []>("op_53631_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_53631_end_mask_0 = const()[name = tensor<string, []>("op_53631_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53631_cast = slice_by_index(begin = var_53631_begin_0, end = var_53631_end_0, end_mask = var_53631_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53631_cast")];
            tensor<int32, [4]> var_53635_begin_0 = const()[name = tensor<string, []>("op_53635_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_53635_end_0 = const()[name = tensor<string, []>("op_53635_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_53635_end_mask_0 = const()[name = tensor<string, []>("op_53635_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53635_cast = slice_by_index(begin = var_53635_begin_0, end = var_53635_end_0, end_mask = var_53635_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53635_cast")];
            tensor<int32, [4]> var_53639_begin_0 = const()[name = tensor<string, []>("op_53639_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_53639_end_0 = const()[name = tensor<string, []>("op_53639_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_53639_end_mask_0 = const()[name = tensor<string, []>("op_53639_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53639_cast = slice_by_index(begin = var_53639_begin_0, end = var_53639_end_0, end_mask = var_53639_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53639_cast")];
            tensor<int32, [4]> var_53643_begin_0 = const()[name = tensor<string, []>("op_53643_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_53643_end_0 = const()[name = tensor<string, []>("op_53643_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_53643_end_mask_0 = const()[name = tensor<string, []>("op_53643_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53643_cast = slice_by_index(begin = var_53643_begin_0, end = var_53643_end_0, end_mask = var_53643_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53643_cast")];
            tensor<int32, [4]> var_53647_begin_0 = const()[name = tensor<string, []>("op_53647_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_53647_end_0 = const()[name = tensor<string, []>("op_53647_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_53647_end_mask_0 = const()[name = tensor<string, []>("op_53647_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53647_cast = slice_by_index(begin = var_53647_begin_0, end = var_53647_end_0, end_mask = var_53647_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53647_cast")];
            tensor<int32, [4]> var_53651_begin_0 = const()[name = tensor<string, []>("op_53651_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_53651_end_0 = const()[name = tensor<string, []>("op_53651_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_53651_end_mask_0 = const()[name = tensor<string, []>("op_53651_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53651_cast = slice_by_index(begin = var_53651_begin_0, end = var_53651_end_0, end_mask = var_53651_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53651_cast")];
            tensor<int32, [4]> var_53655_begin_0 = const()[name = tensor<string, []>("op_53655_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_53655_end_0 = const()[name = tensor<string, []>("op_53655_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_53655_end_mask_0 = const()[name = tensor<string, []>("op_53655_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53655_cast = slice_by_index(begin = var_53655_begin_0, end = var_53655_end_0, end_mask = var_53655_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53655_cast")];
            tensor<int32, [4]> var_53659_begin_0 = const()[name = tensor<string, []>("op_53659_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_53659_end_0 = const()[name = tensor<string, []>("op_53659_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_53659_end_mask_0 = const()[name = tensor<string, []>("op_53659_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53659_cast = slice_by_index(begin = var_53659_begin_0, end = var_53659_end_0, end_mask = var_53659_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53659_cast")];
            tensor<int32, [4]> var_53663_begin_0 = const()[name = tensor<string, []>("op_53663_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_53663_end_0 = const()[name = tensor<string, []>("op_53663_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_53663_end_mask_0 = const()[name = tensor<string, []>("op_53663_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53663_cast = slice_by_index(begin = var_53663_begin_0, end = var_53663_end_0, end_mask = var_53663_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53663_cast")];
            tensor<int32, [4]> var_53667_begin_0 = const()[name = tensor<string, []>("op_53667_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_53667_end_0 = const()[name = tensor<string, []>("op_53667_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_53667_end_mask_0 = const()[name = tensor<string, []>("op_53667_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53667_cast = slice_by_index(begin = var_53667_begin_0, end = var_53667_end_0, end_mask = var_53667_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53667_cast")];
            tensor<int32, [4]> var_53671_begin_0 = const()[name = tensor<string, []>("op_53671_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_53671_end_0 = const()[name = tensor<string, []>("op_53671_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_53671_end_mask_0 = const()[name = tensor<string, []>("op_53671_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_53671_cast = slice_by_index(begin = var_53671_begin_0, end = var_53671_end_0, end_mask = var_53671_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_53671_cast")];
            tensor<int32, [4]> var_53673_begin_0 = const()[name = tensor<string, []>("op_53673_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_53673_end_0 = const()[name = tensor<string, []>("op_53673_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_53673_end_mask_0 = const()[name = tensor<string, []>("op_53673_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53673_cast = slice_by_index(begin = var_53673_begin_0, end = var_53673_end_0, end_mask = var_53673_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53673_cast")];
            tensor<int32, [4]> var_53677_begin_0 = const()[name = tensor<string, []>("op_53677_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_53677_end_0 = const()[name = tensor<string, []>("op_53677_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_53677_end_mask_0 = const()[name = tensor<string, []>("op_53677_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53677_cast = slice_by_index(begin = var_53677_begin_0, end = var_53677_end_0, end_mask = var_53677_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53677_cast")];
            tensor<int32, [4]> var_53681_begin_0 = const()[name = tensor<string, []>("op_53681_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_53681_end_0 = const()[name = tensor<string, []>("op_53681_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_53681_end_mask_0 = const()[name = tensor<string, []>("op_53681_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53681_cast = slice_by_index(begin = var_53681_begin_0, end = var_53681_end_0, end_mask = var_53681_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53681_cast")];
            tensor<int32, [4]> var_53685_begin_0 = const()[name = tensor<string, []>("op_53685_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_53685_end_0 = const()[name = tensor<string, []>("op_53685_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_53685_end_mask_0 = const()[name = tensor<string, []>("op_53685_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53685_cast = slice_by_index(begin = var_53685_begin_0, end = var_53685_end_0, end_mask = var_53685_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53685_cast")];
            tensor<int32, [4]> var_53689_begin_0 = const()[name = tensor<string, []>("op_53689_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_53689_end_0 = const()[name = tensor<string, []>("op_53689_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_53689_end_mask_0 = const()[name = tensor<string, []>("op_53689_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53689_cast = slice_by_index(begin = var_53689_begin_0, end = var_53689_end_0, end_mask = var_53689_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53689_cast")];
            tensor<int32, [4]> var_53693_begin_0 = const()[name = tensor<string, []>("op_53693_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_53693_end_0 = const()[name = tensor<string, []>("op_53693_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_53693_end_mask_0 = const()[name = tensor<string, []>("op_53693_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53693_cast = slice_by_index(begin = var_53693_begin_0, end = var_53693_end_0, end_mask = var_53693_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53693_cast")];
            tensor<int32, [4]> var_53697_begin_0 = const()[name = tensor<string, []>("op_53697_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_53697_end_0 = const()[name = tensor<string, []>("op_53697_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_53697_end_mask_0 = const()[name = tensor<string, []>("op_53697_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53697_cast = slice_by_index(begin = var_53697_begin_0, end = var_53697_end_0, end_mask = var_53697_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53697_cast")];
            tensor<int32, [4]> var_53701_begin_0 = const()[name = tensor<string, []>("op_53701_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_53701_end_0 = const()[name = tensor<string, []>("op_53701_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_53701_end_mask_0 = const()[name = tensor<string, []>("op_53701_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53701_cast = slice_by_index(begin = var_53701_begin_0, end = var_53701_end_0, end_mask = var_53701_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53701_cast")];
            tensor<int32, [4]> var_53705_begin_0 = const()[name = tensor<string, []>("op_53705_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_53705_end_0 = const()[name = tensor<string, []>("op_53705_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_53705_end_mask_0 = const()[name = tensor<string, []>("op_53705_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53705_cast = slice_by_index(begin = var_53705_begin_0, end = var_53705_end_0, end_mask = var_53705_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53705_cast")];
            tensor<int32, [4]> var_53709_begin_0 = const()[name = tensor<string, []>("op_53709_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_53709_end_0 = const()[name = tensor<string, []>("op_53709_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_53709_end_mask_0 = const()[name = tensor<string, []>("op_53709_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53709_cast = slice_by_index(begin = var_53709_begin_0, end = var_53709_end_0, end_mask = var_53709_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53709_cast")];
            tensor<int32, [4]> var_53713_begin_0 = const()[name = tensor<string, []>("op_53713_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_53713_end_0 = const()[name = tensor<string, []>("op_53713_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_53713_end_mask_0 = const()[name = tensor<string, []>("op_53713_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53713_cast = slice_by_index(begin = var_53713_begin_0, end = var_53713_end_0, end_mask = var_53713_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53713_cast")];
            tensor<int32, [4]> var_53717_begin_0 = const()[name = tensor<string, []>("op_53717_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_53717_end_0 = const()[name = tensor<string, []>("op_53717_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_53717_end_mask_0 = const()[name = tensor<string, []>("op_53717_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53717_cast = slice_by_index(begin = var_53717_begin_0, end = var_53717_end_0, end_mask = var_53717_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53717_cast")];
            tensor<int32, [4]> var_53721_begin_0 = const()[name = tensor<string, []>("op_53721_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_53721_end_0 = const()[name = tensor<string, []>("op_53721_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_53721_end_mask_0 = const()[name = tensor<string, []>("op_53721_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53721_cast = slice_by_index(begin = var_53721_begin_0, end = var_53721_end_0, end_mask = var_53721_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53721_cast")];
            tensor<int32, [4]> var_53725_begin_0 = const()[name = tensor<string, []>("op_53725_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_53725_end_0 = const()[name = tensor<string, []>("op_53725_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_53725_end_mask_0 = const()[name = tensor<string, []>("op_53725_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53725_cast = slice_by_index(begin = var_53725_begin_0, end = var_53725_end_0, end_mask = var_53725_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53725_cast")];
            tensor<int32, [4]> var_53729_begin_0 = const()[name = tensor<string, []>("op_53729_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_53729_end_0 = const()[name = tensor<string, []>("op_53729_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_53729_end_mask_0 = const()[name = tensor<string, []>("op_53729_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53729_cast = slice_by_index(begin = var_53729_begin_0, end = var_53729_end_0, end_mask = var_53729_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53729_cast")];
            tensor<int32, [4]> var_53733_begin_0 = const()[name = tensor<string, []>("op_53733_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_53733_end_0 = const()[name = tensor<string, []>("op_53733_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_53733_end_mask_0 = const()[name = tensor<string, []>("op_53733_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53733_cast = slice_by_index(begin = var_53733_begin_0, end = var_53733_end_0, end_mask = var_53733_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53733_cast")];
            tensor<int32, [4]> var_53737_begin_0 = const()[name = tensor<string, []>("op_53737_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_53737_end_0 = const()[name = tensor<string, []>("op_53737_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_53737_end_mask_0 = const()[name = tensor<string, []>("op_53737_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53737_cast = slice_by_index(begin = var_53737_begin_0, end = var_53737_end_0, end_mask = var_53737_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53737_cast")];
            tensor<int32, [4]> var_53741_begin_0 = const()[name = tensor<string, []>("op_53741_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_53741_end_0 = const()[name = tensor<string, []>("op_53741_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_53741_end_mask_0 = const()[name = tensor<string, []>("op_53741_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53741_cast = slice_by_index(begin = var_53741_begin_0, end = var_53741_end_0, end_mask = var_53741_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53741_cast")];
            tensor<int32, [4]> var_53745_begin_0 = const()[name = tensor<string, []>("op_53745_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_53745_end_0 = const()[name = tensor<string, []>("op_53745_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_53745_end_mask_0 = const()[name = tensor<string, []>("op_53745_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53745_cast = slice_by_index(begin = var_53745_begin_0, end = var_53745_end_0, end_mask = var_53745_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53745_cast")];
            tensor<int32, [4]> var_53749_begin_0 = const()[name = tensor<string, []>("op_53749_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_53749_end_0 = const()[name = tensor<string, []>("op_53749_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_53749_end_mask_0 = const()[name = tensor<string, []>("op_53749_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53749_cast = slice_by_index(begin = var_53749_begin_0, end = var_53749_end_0, end_mask = var_53749_end_mask_0, x = v_237_cast)[name = tensor<string, []>("op_53749_cast")];
            tensor<string, []> var_53753_equation_0 = const()[name = tensor<string, []>("op_53753_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53753_cast = einsum(equation = var_53753_equation_0, values = (var_53595_cast, var_53512_cast))[name = tensor<string, []>("op_53753_cast")];
            tensor<fp16, []> var_53754_to_fp16 = const()[name = tensor<string, []>("op_53754_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4561_cast = mul(x = var_53753_cast, y = var_53754_to_fp16)[name = tensor<string, []>("aw_4561_cast")];
            tensor<string, []> var_53757_equation_0 = const()[name = tensor<string, []>("op_53757_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53757_cast = einsum(equation = var_53757_equation_0, values = (var_53599_cast, var_53516_cast))[name = tensor<string, []>("op_53757_cast")];
            tensor<fp16, []> var_53758_to_fp16 = const()[name = tensor<string, []>("op_53758_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4563_cast = mul(x = var_53757_cast, y = var_53758_to_fp16)[name = tensor<string, []>("aw_4563_cast")];
            tensor<string, []> var_53761_equation_0 = const()[name = tensor<string, []>("op_53761_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53761_cast = einsum(equation = var_53761_equation_0, values = (var_53603_cast, var_53520_cast))[name = tensor<string, []>("op_53761_cast")];
            tensor<fp16, []> var_53762_to_fp16 = const()[name = tensor<string, []>("op_53762_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4565_cast = mul(x = var_53761_cast, y = var_53762_to_fp16)[name = tensor<string, []>("aw_4565_cast")];
            tensor<string, []> var_53765_equation_0 = const()[name = tensor<string, []>("op_53765_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53765_cast = einsum(equation = var_53765_equation_0, values = (var_53607_cast, var_53524_cast))[name = tensor<string, []>("op_53765_cast")];
            tensor<fp16, []> var_53766_to_fp16 = const()[name = tensor<string, []>("op_53766_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4567_cast = mul(x = var_53765_cast, y = var_53766_to_fp16)[name = tensor<string, []>("aw_4567_cast")];
            tensor<string, []> var_53769_equation_0 = const()[name = tensor<string, []>("op_53769_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53769_cast = einsum(equation = var_53769_equation_0, values = (var_53611_cast, var_53528_cast))[name = tensor<string, []>("op_53769_cast")];
            tensor<fp16, []> var_53770_to_fp16 = const()[name = tensor<string, []>("op_53770_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4569_cast = mul(x = var_53769_cast, y = var_53770_to_fp16)[name = tensor<string, []>("aw_4569_cast")];
            tensor<string, []> var_53773_equation_0 = const()[name = tensor<string, []>("op_53773_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53773_cast = einsum(equation = var_53773_equation_0, values = (var_53615_cast, var_53532_cast))[name = tensor<string, []>("op_53773_cast")];
            tensor<fp16, []> var_53774_to_fp16 = const()[name = tensor<string, []>("op_53774_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4571_cast = mul(x = var_53773_cast, y = var_53774_to_fp16)[name = tensor<string, []>("aw_4571_cast")];
            tensor<string, []> var_53777_equation_0 = const()[name = tensor<string, []>("op_53777_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53777_cast = einsum(equation = var_53777_equation_0, values = (var_53619_cast, var_53536_cast))[name = tensor<string, []>("op_53777_cast")];
            tensor<fp16, []> var_53778_to_fp16 = const()[name = tensor<string, []>("op_53778_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4573_cast = mul(x = var_53777_cast, y = var_53778_to_fp16)[name = tensor<string, []>("aw_4573_cast")];
            tensor<string, []> var_53781_equation_0 = const()[name = tensor<string, []>("op_53781_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53781_cast = einsum(equation = var_53781_equation_0, values = (var_53623_cast, var_53540_cast))[name = tensor<string, []>("op_53781_cast")];
            tensor<fp16, []> var_53782_to_fp16 = const()[name = tensor<string, []>("op_53782_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4575_cast = mul(x = var_53781_cast, y = var_53782_to_fp16)[name = tensor<string, []>("aw_4575_cast")];
            tensor<string, []> var_53785_equation_0 = const()[name = tensor<string, []>("op_53785_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53785_cast = einsum(equation = var_53785_equation_0, values = (var_53627_cast, var_53544_cast))[name = tensor<string, []>("op_53785_cast")];
            tensor<fp16, []> var_53786_to_fp16 = const()[name = tensor<string, []>("op_53786_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4577_cast = mul(x = var_53785_cast, y = var_53786_to_fp16)[name = tensor<string, []>("aw_4577_cast")];
            tensor<string, []> var_53789_equation_0 = const()[name = tensor<string, []>("op_53789_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53789_cast = einsum(equation = var_53789_equation_0, values = (var_53631_cast, var_53548_cast))[name = tensor<string, []>("op_53789_cast")];
            tensor<fp16, []> var_53790_to_fp16 = const()[name = tensor<string, []>("op_53790_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4579_cast = mul(x = var_53789_cast, y = var_53790_to_fp16)[name = tensor<string, []>("aw_4579_cast")];
            tensor<string, []> var_53793_equation_0 = const()[name = tensor<string, []>("op_53793_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53793_cast = einsum(equation = var_53793_equation_0, values = (var_53635_cast, var_53552_cast))[name = tensor<string, []>("op_53793_cast")];
            tensor<fp16, []> var_53794_to_fp16 = const()[name = tensor<string, []>("op_53794_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4581_cast = mul(x = var_53793_cast, y = var_53794_to_fp16)[name = tensor<string, []>("aw_4581_cast")];
            tensor<string, []> var_53797_equation_0 = const()[name = tensor<string, []>("op_53797_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53797_cast = einsum(equation = var_53797_equation_0, values = (var_53639_cast, var_53556_cast))[name = tensor<string, []>("op_53797_cast")];
            tensor<fp16, []> var_53798_to_fp16 = const()[name = tensor<string, []>("op_53798_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4583_cast = mul(x = var_53797_cast, y = var_53798_to_fp16)[name = tensor<string, []>("aw_4583_cast")];
            tensor<string, []> var_53801_equation_0 = const()[name = tensor<string, []>("op_53801_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53801_cast = einsum(equation = var_53801_equation_0, values = (var_53643_cast, var_53560_cast))[name = tensor<string, []>("op_53801_cast")];
            tensor<fp16, []> var_53802_to_fp16 = const()[name = tensor<string, []>("op_53802_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4585_cast = mul(x = var_53801_cast, y = var_53802_to_fp16)[name = tensor<string, []>("aw_4585_cast")];
            tensor<string, []> var_53805_equation_0 = const()[name = tensor<string, []>("op_53805_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53805_cast = einsum(equation = var_53805_equation_0, values = (var_53647_cast, var_53564_cast))[name = tensor<string, []>("op_53805_cast")];
            tensor<fp16, []> var_53806_to_fp16 = const()[name = tensor<string, []>("op_53806_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4587_cast = mul(x = var_53805_cast, y = var_53806_to_fp16)[name = tensor<string, []>("aw_4587_cast")];
            tensor<string, []> var_53809_equation_0 = const()[name = tensor<string, []>("op_53809_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53809_cast = einsum(equation = var_53809_equation_0, values = (var_53651_cast, var_53568_cast))[name = tensor<string, []>("op_53809_cast")];
            tensor<fp16, []> var_53810_to_fp16 = const()[name = tensor<string, []>("op_53810_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4589_cast = mul(x = var_53809_cast, y = var_53810_to_fp16)[name = tensor<string, []>("aw_4589_cast")];
            tensor<string, []> var_53813_equation_0 = const()[name = tensor<string, []>("op_53813_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53813_cast = einsum(equation = var_53813_equation_0, values = (var_53655_cast, var_53572_cast))[name = tensor<string, []>("op_53813_cast")];
            tensor<fp16, []> var_53814_to_fp16 = const()[name = tensor<string, []>("op_53814_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4591_cast = mul(x = var_53813_cast, y = var_53814_to_fp16)[name = tensor<string, []>("aw_4591_cast")];
            tensor<string, []> var_53817_equation_0 = const()[name = tensor<string, []>("op_53817_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53817_cast = einsum(equation = var_53817_equation_0, values = (var_53659_cast, var_53576_cast))[name = tensor<string, []>("op_53817_cast")];
            tensor<fp16, []> var_53818_to_fp16 = const()[name = tensor<string, []>("op_53818_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4593_cast = mul(x = var_53817_cast, y = var_53818_to_fp16)[name = tensor<string, []>("aw_4593_cast")];
            tensor<string, []> var_53821_equation_0 = const()[name = tensor<string, []>("op_53821_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53821_cast = einsum(equation = var_53821_equation_0, values = (var_53663_cast, var_53580_cast))[name = tensor<string, []>("op_53821_cast")];
            tensor<fp16, []> var_53822_to_fp16 = const()[name = tensor<string, []>("op_53822_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4595_cast = mul(x = var_53821_cast, y = var_53822_to_fp16)[name = tensor<string, []>("aw_4595_cast")];
            tensor<string, []> var_53825_equation_0 = const()[name = tensor<string, []>("op_53825_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53825_cast = einsum(equation = var_53825_equation_0, values = (var_53667_cast, var_53584_cast))[name = tensor<string, []>("op_53825_cast")];
            tensor<fp16, []> var_53826_to_fp16 = const()[name = tensor<string, []>("op_53826_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4597_cast = mul(x = var_53825_cast, y = var_53826_to_fp16)[name = tensor<string, []>("aw_4597_cast")];
            tensor<string, []> var_53829_equation_0 = const()[name = tensor<string, []>("op_53829_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53829_cast = einsum(equation = var_53829_equation_0, values = (var_53671_cast, var_53588_cast))[name = tensor<string, []>("op_53829_cast")];
            tensor<fp16, []> var_53830_to_fp16 = const()[name = tensor<string, []>("op_53830_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4599_cast = mul(x = var_53829_cast, y = var_53830_to_fp16)[name = tensor<string, []>("aw_4599_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53832_cast = softmax(axis = var_30385, x = aw_4561_cast)[name = tensor<string, []>("op_53832_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53833_cast = softmax(axis = var_30385, x = aw_4563_cast)[name = tensor<string, []>("op_53833_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53834_cast = softmax(axis = var_30385, x = aw_4565_cast)[name = tensor<string, []>("op_53834_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53835_cast = softmax(axis = var_30385, x = aw_4567_cast)[name = tensor<string, []>("op_53835_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53836_cast = softmax(axis = var_30385, x = aw_4569_cast)[name = tensor<string, []>("op_53836_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53837_cast = softmax(axis = var_30385, x = aw_4571_cast)[name = tensor<string, []>("op_53837_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53838_cast = softmax(axis = var_30385, x = aw_4573_cast)[name = tensor<string, []>("op_53838_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53839_cast = softmax(axis = var_30385, x = aw_4575_cast)[name = tensor<string, []>("op_53839_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53840_cast = softmax(axis = var_30385, x = aw_4577_cast)[name = tensor<string, []>("op_53840_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53841_cast = softmax(axis = var_30385, x = aw_4579_cast)[name = tensor<string, []>("op_53841_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53842_cast = softmax(axis = var_30385, x = aw_4581_cast)[name = tensor<string, []>("op_53842_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53843_cast = softmax(axis = var_30385, x = aw_4583_cast)[name = tensor<string, []>("op_53843_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53844_cast = softmax(axis = var_30385, x = aw_4585_cast)[name = tensor<string, []>("op_53844_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53845_cast = softmax(axis = var_30385, x = aw_4587_cast)[name = tensor<string, []>("op_53845_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53846_cast = softmax(axis = var_30385, x = aw_4589_cast)[name = tensor<string, []>("op_53846_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53847_cast = softmax(axis = var_30385, x = aw_4591_cast)[name = tensor<string, []>("op_53847_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53848_cast = softmax(axis = var_30385, x = aw_4593_cast)[name = tensor<string, []>("op_53848_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53849_cast = softmax(axis = var_30385, x = aw_4595_cast)[name = tensor<string, []>("op_53849_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53850_cast = softmax(axis = var_30385, x = aw_4597_cast)[name = tensor<string, []>("op_53850_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_53851_cast = softmax(axis = var_30385, x = aw_4599_cast)[name = tensor<string, []>("op_53851_cast")];
            tensor<string, []> var_53853_equation_0 = const()[name = tensor<string, []>("op_53853_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53853_cast = einsum(equation = var_53853_equation_0, values = (var_53673_cast, var_53832_cast))[name = tensor<string, []>("op_53853_cast")];
            tensor<string, []> var_53855_equation_0 = const()[name = tensor<string, []>("op_53855_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53855_cast = einsum(equation = var_53855_equation_0, values = (var_53677_cast, var_53833_cast))[name = tensor<string, []>("op_53855_cast")];
            tensor<string, []> var_53857_equation_0 = const()[name = tensor<string, []>("op_53857_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53857_cast = einsum(equation = var_53857_equation_0, values = (var_53681_cast, var_53834_cast))[name = tensor<string, []>("op_53857_cast")];
            tensor<string, []> var_53859_equation_0 = const()[name = tensor<string, []>("op_53859_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53859_cast = einsum(equation = var_53859_equation_0, values = (var_53685_cast, var_53835_cast))[name = tensor<string, []>("op_53859_cast")];
            tensor<string, []> var_53861_equation_0 = const()[name = tensor<string, []>("op_53861_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53861_cast = einsum(equation = var_53861_equation_0, values = (var_53689_cast, var_53836_cast))[name = tensor<string, []>("op_53861_cast")];
            tensor<string, []> var_53863_equation_0 = const()[name = tensor<string, []>("op_53863_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53863_cast = einsum(equation = var_53863_equation_0, values = (var_53693_cast, var_53837_cast))[name = tensor<string, []>("op_53863_cast")];
            tensor<string, []> var_53865_equation_0 = const()[name = tensor<string, []>("op_53865_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53865_cast = einsum(equation = var_53865_equation_0, values = (var_53697_cast, var_53838_cast))[name = tensor<string, []>("op_53865_cast")];
            tensor<string, []> var_53867_equation_0 = const()[name = tensor<string, []>("op_53867_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53867_cast = einsum(equation = var_53867_equation_0, values = (var_53701_cast, var_53839_cast))[name = tensor<string, []>("op_53867_cast")];
            tensor<string, []> var_53869_equation_0 = const()[name = tensor<string, []>("op_53869_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53869_cast = einsum(equation = var_53869_equation_0, values = (var_53705_cast, var_53840_cast))[name = tensor<string, []>("op_53869_cast")];
            tensor<string, []> var_53871_equation_0 = const()[name = tensor<string, []>("op_53871_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53871_cast = einsum(equation = var_53871_equation_0, values = (var_53709_cast, var_53841_cast))[name = tensor<string, []>("op_53871_cast")];
            tensor<string, []> var_53873_equation_0 = const()[name = tensor<string, []>("op_53873_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53873_cast = einsum(equation = var_53873_equation_0, values = (var_53713_cast, var_53842_cast))[name = tensor<string, []>("op_53873_cast")];
            tensor<string, []> var_53875_equation_0 = const()[name = tensor<string, []>("op_53875_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53875_cast = einsum(equation = var_53875_equation_0, values = (var_53717_cast, var_53843_cast))[name = tensor<string, []>("op_53875_cast")];
            tensor<string, []> var_53877_equation_0 = const()[name = tensor<string, []>("op_53877_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53877_cast = einsum(equation = var_53877_equation_0, values = (var_53721_cast, var_53844_cast))[name = tensor<string, []>("op_53877_cast")];
            tensor<string, []> var_53879_equation_0 = const()[name = tensor<string, []>("op_53879_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53879_cast = einsum(equation = var_53879_equation_0, values = (var_53725_cast, var_53845_cast))[name = tensor<string, []>("op_53879_cast")];
            tensor<string, []> var_53881_equation_0 = const()[name = tensor<string, []>("op_53881_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53881_cast = einsum(equation = var_53881_equation_0, values = (var_53729_cast, var_53846_cast))[name = tensor<string, []>("op_53881_cast")];
            tensor<string, []> var_53883_equation_0 = const()[name = tensor<string, []>("op_53883_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53883_cast = einsum(equation = var_53883_equation_0, values = (var_53733_cast, var_53847_cast))[name = tensor<string, []>("op_53883_cast")];
            tensor<string, []> var_53885_equation_0 = const()[name = tensor<string, []>("op_53885_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53885_cast = einsum(equation = var_53885_equation_0, values = (var_53737_cast, var_53848_cast))[name = tensor<string, []>("op_53885_cast")];
            tensor<string, []> var_53887_equation_0 = const()[name = tensor<string, []>("op_53887_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53887_cast = einsum(equation = var_53887_equation_0, values = (var_53741_cast, var_53849_cast))[name = tensor<string, []>("op_53887_cast")];
            tensor<string, []> var_53889_equation_0 = const()[name = tensor<string, []>("op_53889_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53889_cast = einsum(equation = var_53889_equation_0, values = (var_53745_cast, var_53850_cast))[name = tensor<string, []>("op_53889_cast")];
            tensor<string, []> var_53891_equation_0 = const()[name = tensor<string, []>("op_53891_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_53891_cast = einsum(equation = var_53891_equation_0, values = (var_53749_cast, var_53851_cast))[name = tensor<string, []>("op_53891_cast")];
            tensor<bool, []> input_677_interleave_0 = const()[name = tensor<string, []>("input_677_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_677_cast = concat(axis = var_30385, interleave = input_677_interleave_0, values = (var_53853_cast, var_53855_cast, var_53857_cast, var_53859_cast, var_53861_cast, var_53863_cast, var_53865_cast, var_53867_cast, var_53869_cast, var_53871_cast, var_53873_cast, var_53875_cast, var_53877_cast, var_53879_cast, var_53881_cast, var_53883_cast, var_53885_cast, var_53887_cast, var_53889_cast, var_53891_cast))[name = tensor<string, []>("input_677_cast")];
            tensor<int32, [2]> var_53897 = const()[name = tensor<string, []>("op_53897"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_53899 = const()[name = tensor<string, []>("op_53899"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_53901_pad_type_0 = const()[name = tensor<string, []>("op_53901_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_53901_pad_0 = const()[name = tensor<string, []>("op_53901_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4529491648)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4532768512)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_53901_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_out_0_bias_to_fp16, dilations = var_53899, groups = var_30385, pad = var_53901_pad_0, pad_type = var_53901_pad_type_0, strides = var_53897, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn1_to_out_0_weight_to_fp16, x = input_677_cast)[name = tensor<string, []>("op_53901_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_357_cast = add(x = var_53901_cast, y = inputs_355_cast)[name = tensor<string, []>("inputs_357_cast")];
            tensor<int32, [1]> var_53905 = const()[name = tensor<string, []>("op_53905"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_357_cast = reduce_mean(axes = var_53905, keep_dims = var_30380, x = inputs_357_cast)[name = tensor<string, []>("channels_mean_357_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_357_cast = sub(x = inputs_357_cast, y = channels_mean_357_cast)[name = tensor<string, []>("zero_mean_357_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_357_cast = mul(x = zero_mean_357_cast, y = zero_mean_357_cast)[name = tensor<string, []>("zero_mean_sq_357_cast")];
            tensor<int32, [1]> var_53909 = const()[name = tensor<string, []>("op_53909"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_53910_cast = reduce_mean(axes = var_53909, keep_dims = var_30380, x = zero_mean_sq_357_cast)[name = tensor<string, []>("op_53910_cast")];
            tensor<fp16, []> var_53911_to_fp16 = const()[name = tensor<string, []>("op_53911_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_53912_cast = add(x = var_53910_cast, y = var_53911_to_fp16)[name = tensor<string, []>("op_53912_cast")];
            tensor<fp16, []> denom_357_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_357_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_357_cast = rsqrt(epsilon = denom_357_epsilon_0_to_fp16, x = var_53912_cast)[name = tensor<string, []>("denom_357_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_357_cast = mul(x = zero_mean_357_cast, y = denom_357_cast)[name = tensor<string, []>("out_357_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_53916_to_fp16 = const()[name = tensor<string, []>("op_53916_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4532771136)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_53917_cast = add(x = out_357_cast, y = var_53916_to_fp16)[name = tensor<string, []>("op_53917_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_53919_to_fp16 = const()[name = tensor<string, []>("op_53919_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4532773760)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_467_cast = mul(x = var_53917_cast, y = var_53919_to_fp16)[name = tensor<string, []>("hidden_states_467_cast")];
            tensor<int32, [2]> var_53926 = const()[name = tensor<string, []>("op_53926"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_53928 = const()[name = tensor<string, []>("op_53928"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_239_pad_type_0 = const()[name = tensor<string, []>("q_239_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_239_pad_0 = const()[name = tensor<string, []>("q_239_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4532776384)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_239_cast = conv(dilations = var_53928, groups = var_30385, pad = q_239_pad_0, pad_type = q_239_pad_type_0, strides = var_53926, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_q_weight_to_fp16, x = hidden_states_467_cast)[name = tensor<string, []>("q_239_cast")];
            tensor<int32, [2]> var_53932 = const()[name = tensor<string, []>("op_53932"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_53934 = const()[name = tensor<string, []>("op_53934"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_477_pad_type_0 = const()[name = tensor<string, []>("k_477_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_477_pad_0 = const()[name = tensor<string, []>("k_477_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4536053248)))];
            tensor<fp16, [2, 1280, 1, 77]> k_477_cast = conv(dilations = var_53934, groups = var_30385, pad = k_477_pad_0, pad_type = k_477_pad_type_0, strides = var_53932, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_477_cast")];
            tensor<int32, [2]> var_53938 = const()[name = tensor<string, []>("op_53938"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_53940 = const()[name = tensor<string, []>("op_53940"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_239_pad_type_0 = const()[name = tensor<string, []>("v_239_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_239_pad_0 = const()[name = tensor<string, []>("v_239_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4541296192)))];
            tensor<fp16, [2, 1280, 1, 77]> v_239_cast = conv(dilations = var_53940, groups = var_30385, pad = v_239_pad_0, pad_type = v_239_pad_type_0, strides = var_53938, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_239_cast")];
            tensor<int32, [4]> var_53944_begin_0 = const()[name = tensor<string, []>("op_53944_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_53944_end_0 = const()[name = tensor<string, []>("op_53944_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_53944_end_mask_0 = const()[name = tensor<string, []>("op_53944_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53944_cast = slice_by_index(begin = var_53944_begin_0, end = var_53944_end_0, end_mask = var_53944_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_53944_cast")];
            tensor<int32, [4]> var_53948_begin_0 = const()[name = tensor<string, []>("op_53948_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_53948_end_0 = const()[name = tensor<string, []>("op_53948_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_53948_end_mask_0 = const()[name = tensor<string, []>("op_53948_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53948_cast = slice_by_index(begin = var_53948_begin_0, end = var_53948_end_0, end_mask = var_53948_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_53948_cast")];
            tensor<int32, [4]> var_53952_begin_0 = const()[name = tensor<string, []>("op_53952_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_53952_end_0 = const()[name = tensor<string, []>("op_53952_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_53952_end_mask_0 = const()[name = tensor<string, []>("op_53952_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53952_cast = slice_by_index(begin = var_53952_begin_0, end = var_53952_end_0, end_mask = var_53952_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_53952_cast")];
            tensor<int32, [4]> var_53956_begin_0 = const()[name = tensor<string, []>("op_53956_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_53956_end_0 = const()[name = tensor<string, []>("op_53956_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_53956_end_mask_0 = const()[name = tensor<string, []>("op_53956_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53956_cast = slice_by_index(begin = var_53956_begin_0, end = var_53956_end_0, end_mask = var_53956_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_53956_cast")];
            tensor<int32, [4]> var_53960_begin_0 = const()[name = tensor<string, []>("op_53960_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_53960_end_0 = const()[name = tensor<string, []>("op_53960_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_53960_end_mask_0 = const()[name = tensor<string, []>("op_53960_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53960_cast = slice_by_index(begin = var_53960_begin_0, end = var_53960_end_0, end_mask = var_53960_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_53960_cast")];
            tensor<int32, [4]> var_53964_begin_0 = const()[name = tensor<string, []>("op_53964_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_53964_end_0 = const()[name = tensor<string, []>("op_53964_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_53964_end_mask_0 = const()[name = tensor<string, []>("op_53964_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53964_cast = slice_by_index(begin = var_53964_begin_0, end = var_53964_end_0, end_mask = var_53964_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_53964_cast")];
            tensor<int32, [4]> var_53968_begin_0 = const()[name = tensor<string, []>("op_53968_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_53968_end_0 = const()[name = tensor<string, []>("op_53968_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_53968_end_mask_0 = const()[name = tensor<string, []>("op_53968_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53968_cast = slice_by_index(begin = var_53968_begin_0, end = var_53968_end_0, end_mask = var_53968_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_53968_cast")];
            tensor<int32, [4]> var_53972_begin_0 = const()[name = tensor<string, []>("op_53972_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_53972_end_0 = const()[name = tensor<string, []>("op_53972_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_53972_end_mask_0 = const()[name = tensor<string, []>("op_53972_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53972_cast = slice_by_index(begin = var_53972_begin_0, end = var_53972_end_0, end_mask = var_53972_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_53972_cast")];
            tensor<int32, [4]> var_53976_begin_0 = const()[name = tensor<string, []>("op_53976_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_53976_end_0 = const()[name = tensor<string, []>("op_53976_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_53976_end_mask_0 = const()[name = tensor<string, []>("op_53976_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53976_cast = slice_by_index(begin = var_53976_begin_0, end = var_53976_end_0, end_mask = var_53976_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_53976_cast")];
            tensor<int32, [4]> var_53980_begin_0 = const()[name = tensor<string, []>("op_53980_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_53980_end_0 = const()[name = tensor<string, []>("op_53980_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_53980_end_mask_0 = const()[name = tensor<string, []>("op_53980_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53980_cast = slice_by_index(begin = var_53980_begin_0, end = var_53980_end_0, end_mask = var_53980_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_53980_cast")];
            tensor<int32, [4]> var_53984_begin_0 = const()[name = tensor<string, []>("op_53984_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_53984_end_0 = const()[name = tensor<string, []>("op_53984_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_53984_end_mask_0 = const()[name = tensor<string, []>("op_53984_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53984_cast = slice_by_index(begin = var_53984_begin_0, end = var_53984_end_0, end_mask = var_53984_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_53984_cast")];
            tensor<int32, [4]> var_53988_begin_0 = const()[name = tensor<string, []>("op_53988_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_53988_end_0 = const()[name = tensor<string, []>("op_53988_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_53988_end_mask_0 = const()[name = tensor<string, []>("op_53988_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53988_cast = slice_by_index(begin = var_53988_begin_0, end = var_53988_end_0, end_mask = var_53988_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_53988_cast")];
            tensor<int32, [4]> var_53992_begin_0 = const()[name = tensor<string, []>("op_53992_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_53992_end_0 = const()[name = tensor<string, []>("op_53992_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_53992_end_mask_0 = const()[name = tensor<string, []>("op_53992_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53992_cast = slice_by_index(begin = var_53992_begin_0, end = var_53992_end_0, end_mask = var_53992_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_53992_cast")];
            tensor<int32, [4]> var_53996_begin_0 = const()[name = tensor<string, []>("op_53996_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_53996_end_0 = const()[name = tensor<string, []>("op_53996_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_53996_end_mask_0 = const()[name = tensor<string, []>("op_53996_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_53996_cast = slice_by_index(begin = var_53996_begin_0, end = var_53996_end_0, end_mask = var_53996_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_53996_cast")];
            tensor<int32, [4]> var_54000_begin_0 = const()[name = tensor<string, []>("op_54000_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_54000_end_0 = const()[name = tensor<string, []>("op_54000_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_54000_end_mask_0 = const()[name = tensor<string, []>("op_54000_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54000_cast = slice_by_index(begin = var_54000_begin_0, end = var_54000_end_0, end_mask = var_54000_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_54000_cast")];
            tensor<int32, [4]> var_54004_begin_0 = const()[name = tensor<string, []>("op_54004_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_54004_end_0 = const()[name = tensor<string, []>("op_54004_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_54004_end_mask_0 = const()[name = tensor<string, []>("op_54004_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54004_cast = slice_by_index(begin = var_54004_begin_0, end = var_54004_end_0, end_mask = var_54004_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_54004_cast")];
            tensor<int32, [4]> var_54008_begin_0 = const()[name = tensor<string, []>("op_54008_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_54008_end_0 = const()[name = tensor<string, []>("op_54008_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_54008_end_mask_0 = const()[name = tensor<string, []>("op_54008_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54008_cast = slice_by_index(begin = var_54008_begin_0, end = var_54008_end_0, end_mask = var_54008_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_54008_cast")];
            tensor<int32, [4]> var_54012_begin_0 = const()[name = tensor<string, []>("op_54012_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_54012_end_0 = const()[name = tensor<string, []>("op_54012_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_54012_end_mask_0 = const()[name = tensor<string, []>("op_54012_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54012_cast = slice_by_index(begin = var_54012_begin_0, end = var_54012_end_0, end_mask = var_54012_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_54012_cast")];
            tensor<int32, [4]> var_54016_begin_0 = const()[name = tensor<string, []>("op_54016_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_54016_end_0 = const()[name = tensor<string, []>("op_54016_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_54016_end_mask_0 = const()[name = tensor<string, []>("op_54016_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54016_cast = slice_by_index(begin = var_54016_begin_0, end = var_54016_end_0, end_mask = var_54016_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_54016_cast")];
            tensor<int32, [4]> var_54020_begin_0 = const()[name = tensor<string, []>("op_54020_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_54020_end_0 = const()[name = tensor<string, []>("op_54020_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_54020_end_mask_0 = const()[name = tensor<string, []>("op_54020_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54020_cast = slice_by_index(begin = var_54020_begin_0, end = var_54020_end_0, end_mask = var_54020_end_mask_0, x = q_239_cast)[name = tensor<string, []>("op_54020_cast")];
            tensor<int32, [4]> k_479_perm_0 = const()[name = tensor<string, []>("k_479_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_54027_begin_0 = const()[name = tensor<string, []>("op_54027_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_54027_end_0 = const()[name = tensor<string, []>("op_54027_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_54027_end_mask_0 = const()[name = tensor<string, []>("op_54027_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_20 = transpose(perm = k_479_perm_0, x = k_477_cast)[name = tensor<string, []>("transpose_20")];
            tensor<fp16, [2, 77, 1, 64]> var_54027_cast = slice_by_index(begin = var_54027_begin_0, end = var_54027_end_0, end_mask = var_54027_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54027_cast")];
            tensor<int32, [4]> var_54031_begin_0 = const()[name = tensor<string, []>("op_54031_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_54031_end_0 = const()[name = tensor<string, []>("op_54031_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_54031_end_mask_0 = const()[name = tensor<string, []>("op_54031_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54031_cast = slice_by_index(begin = var_54031_begin_0, end = var_54031_end_0, end_mask = var_54031_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54031_cast")];
            tensor<int32, [4]> var_54035_begin_0 = const()[name = tensor<string, []>("op_54035_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_54035_end_0 = const()[name = tensor<string, []>("op_54035_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_54035_end_mask_0 = const()[name = tensor<string, []>("op_54035_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54035_cast = slice_by_index(begin = var_54035_begin_0, end = var_54035_end_0, end_mask = var_54035_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54035_cast")];
            tensor<int32, [4]> var_54039_begin_0 = const()[name = tensor<string, []>("op_54039_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_54039_end_0 = const()[name = tensor<string, []>("op_54039_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_54039_end_mask_0 = const()[name = tensor<string, []>("op_54039_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54039_cast = slice_by_index(begin = var_54039_begin_0, end = var_54039_end_0, end_mask = var_54039_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54039_cast")];
            tensor<int32, [4]> var_54043_begin_0 = const()[name = tensor<string, []>("op_54043_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_54043_end_0 = const()[name = tensor<string, []>("op_54043_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_54043_end_mask_0 = const()[name = tensor<string, []>("op_54043_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54043_cast = slice_by_index(begin = var_54043_begin_0, end = var_54043_end_0, end_mask = var_54043_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54043_cast")];
            tensor<int32, [4]> var_54047_begin_0 = const()[name = tensor<string, []>("op_54047_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_54047_end_0 = const()[name = tensor<string, []>("op_54047_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_54047_end_mask_0 = const()[name = tensor<string, []>("op_54047_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54047_cast = slice_by_index(begin = var_54047_begin_0, end = var_54047_end_0, end_mask = var_54047_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54047_cast")];
            tensor<int32, [4]> var_54051_begin_0 = const()[name = tensor<string, []>("op_54051_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_54051_end_0 = const()[name = tensor<string, []>("op_54051_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_54051_end_mask_0 = const()[name = tensor<string, []>("op_54051_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54051_cast = slice_by_index(begin = var_54051_begin_0, end = var_54051_end_0, end_mask = var_54051_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54051_cast")];
            tensor<int32, [4]> var_54055_begin_0 = const()[name = tensor<string, []>("op_54055_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_54055_end_0 = const()[name = tensor<string, []>("op_54055_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_54055_end_mask_0 = const()[name = tensor<string, []>("op_54055_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54055_cast = slice_by_index(begin = var_54055_begin_0, end = var_54055_end_0, end_mask = var_54055_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54055_cast")];
            tensor<int32, [4]> var_54059_begin_0 = const()[name = tensor<string, []>("op_54059_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_54059_end_0 = const()[name = tensor<string, []>("op_54059_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_54059_end_mask_0 = const()[name = tensor<string, []>("op_54059_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54059_cast = slice_by_index(begin = var_54059_begin_0, end = var_54059_end_0, end_mask = var_54059_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54059_cast")];
            tensor<int32, [4]> var_54063_begin_0 = const()[name = tensor<string, []>("op_54063_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_54063_end_0 = const()[name = tensor<string, []>("op_54063_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_54063_end_mask_0 = const()[name = tensor<string, []>("op_54063_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54063_cast = slice_by_index(begin = var_54063_begin_0, end = var_54063_end_0, end_mask = var_54063_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54063_cast")];
            tensor<int32, [4]> var_54067_begin_0 = const()[name = tensor<string, []>("op_54067_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_54067_end_0 = const()[name = tensor<string, []>("op_54067_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_54067_end_mask_0 = const()[name = tensor<string, []>("op_54067_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54067_cast = slice_by_index(begin = var_54067_begin_0, end = var_54067_end_0, end_mask = var_54067_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54067_cast")];
            tensor<int32, [4]> var_54071_begin_0 = const()[name = tensor<string, []>("op_54071_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_54071_end_0 = const()[name = tensor<string, []>("op_54071_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_54071_end_mask_0 = const()[name = tensor<string, []>("op_54071_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54071_cast = slice_by_index(begin = var_54071_begin_0, end = var_54071_end_0, end_mask = var_54071_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54071_cast")];
            tensor<int32, [4]> var_54075_begin_0 = const()[name = tensor<string, []>("op_54075_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_54075_end_0 = const()[name = tensor<string, []>("op_54075_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_54075_end_mask_0 = const()[name = tensor<string, []>("op_54075_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54075_cast = slice_by_index(begin = var_54075_begin_0, end = var_54075_end_0, end_mask = var_54075_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54075_cast")];
            tensor<int32, [4]> var_54079_begin_0 = const()[name = tensor<string, []>("op_54079_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_54079_end_0 = const()[name = tensor<string, []>("op_54079_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_54079_end_mask_0 = const()[name = tensor<string, []>("op_54079_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54079_cast = slice_by_index(begin = var_54079_begin_0, end = var_54079_end_0, end_mask = var_54079_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54079_cast")];
            tensor<int32, [4]> var_54083_begin_0 = const()[name = tensor<string, []>("op_54083_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_54083_end_0 = const()[name = tensor<string, []>("op_54083_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_54083_end_mask_0 = const()[name = tensor<string, []>("op_54083_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54083_cast = slice_by_index(begin = var_54083_begin_0, end = var_54083_end_0, end_mask = var_54083_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54083_cast")];
            tensor<int32, [4]> var_54087_begin_0 = const()[name = tensor<string, []>("op_54087_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_54087_end_0 = const()[name = tensor<string, []>("op_54087_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_54087_end_mask_0 = const()[name = tensor<string, []>("op_54087_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54087_cast = slice_by_index(begin = var_54087_begin_0, end = var_54087_end_0, end_mask = var_54087_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54087_cast")];
            tensor<int32, [4]> var_54091_begin_0 = const()[name = tensor<string, []>("op_54091_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_54091_end_0 = const()[name = tensor<string, []>("op_54091_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_54091_end_mask_0 = const()[name = tensor<string, []>("op_54091_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54091_cast = slice_by_index(begin = var_54091_begin_0, end = var_54091_end_0, end_mask = var_54091_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54091_cast")];
            tensor<int32, [4]> var_54095_begin_0 = const()[name = tensor<string, []>("op_54095_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_54095_end_0 = const()[name = tensor<string, []>("op_54095_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_54095_end_mask_0 = const()[name = tensor<string, []>("op_54095_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54095_cast = slice_by_index(begin = var_54095_begin_0, end = var_54095_end_0, end_mask = var_54095_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54095_cast")];
            tensor<int32, [4]> var_54099_begin_0 = const()[name = tensor<string, []>("op_54099_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_54099_end_0 = const()[name = tensor<string, []>("op_54099_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_54099_end_mask_0 = const()[name = tensor<string, []>("op_54099_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54099_cast = slice_by_index(begin = var_54099_begin_0, end = var_54099_end_0, end_mask = var_54099_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54099_cast")];
            tensor<int32, [4]> var_54103_begin_0 = const()[name = tensor<string, []>("op_54103_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_54103_end_0 = const()[name = tensor<string, []>("op_54103_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_54103_end_mask_0 = const()[name = tensor<string, []>("op_54103_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54103_cast = slice_by_index(begin = var_54103_begin_0, end = var_54103_end_0, end_mask = var_54103_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_54103_cast")];
            tensor<int32, [4]> var_54105_begin_0 = const()[name = tensor<string, []>("op_54105_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_54105_end_0 = const()[name = tensor<string, []>("op_54105_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_54105_end_mask_0 = const()[name = tensor<string, []>("op_54105_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54105_cast = slice_by_index(begin = var_54105_begin_0, end = var_54105_end_0, end_mask = var_54105_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54105_cast")];
            tensor<int32, [4]> var_54109_begin_0 = const()[name = tensor<string, []>("op_54109_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_54109_end_0 = const()[name = tensor<string, []>("op_54109_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_54109_end_mask_0 = const()[name = tensor<string, []>("op_54109_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54109_cast = slice_by_index(begin = var_54109_begin_0, end = var_54109_end_0, end_mask = var_54109_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54109_cast")];
            tensor<int32, [4]> var_54113_begin_0 = const()[name = tensor<string, []>("op_54113_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_54113_end_0 = const()[name = tensor<string, []>("op_54113_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_54113_end_mask_0 = const()[name = tensor<string, []>("op_54113_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54113_cast = slice_by_index(begin = var_54113_begin_0, end = var_54113_end_0, end_mask = var_54113_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54113_cast")];
            tensor<int32, [4]> var_54117_begin_0 = const()[name = tensor<string, []>("op_54117_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_54117_end_0 = const()[name = tensor<string, []>("op_54117_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_54117_end_mask_0 = const()[name = tensor<string, []>("op_54117_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54117_cast = slice_by_index(begin = var_54117_begin_0, end = var_54117_end_0, end_mask = var_54117_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54117_cast")];
            tensor<int32, [4]> var_54121_begin_0 = const()[name = tensor<string, []>("op_54121_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_54121_end_0 = const()[name = tensor<string, []>("op_54121_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_54121_end_mask_0 = const()[name = tensor<string, []>("op_54121_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54121_cast = slice_by_index(begin = var_54121_begin_0, end = var_54121_end_0, end_mask = var_54121_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54121_cast")];
            tensor<int32, [4]> var_54125_begin_0 = const()[name = tensor<string, []>("op_54125_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_54125_end_0 = const()[name = tensor<string, []>("op_54125_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_54125_end_mask_0 = const()[name = tensor<string, []>("op_54125_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54125_cast = slice_by_index(begin = var_54125_begin_0, end = var_54125_end_0, end_mask = var_54125_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54125_cast")];
            tensor<int32, [4]> var_54129_begin_0 = const()[name = tensor<string, []>("op_54129_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_54129_end_0 = const()[name = tensor<string, []>("op_54129_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_54129_end_mask_0 = const()[name = tensor<string, []>("op_54129_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54129_cast = slice_by_index(begin = var_54129_begin_0, end = var_54129_end_0, end_mask = var_54129_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54129_cast")];
            tensor<int32, [4]> var_54133_begin_0 = const()[name = tensor<string, []>("op_54133_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_54133_end_0 = const()[name = tensor<string, []>("op_54133_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_54133_end_mask_0 = const()[name = tensor<string, []>("op_54133_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54133_cast = slice_by_index(begin = var_54133_begin_0, end = var_54133_end_0, end_mask = var_54133_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54133_cast")];
            tensor<int32, [4]> var_54137_begin_0 = const()[name = tensor<string, []>("op_54137_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_54137_end_0 = const()[name = tensor<string, []>("op_54137_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_54137_end_mask_0 = const()[name = tensor<string, []>("op_54137_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54137_cast = slice_by_index(begin = var_54137_begin_0, end = var_54137_end_0, end_mask = var_54137_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54137_cast")];
            tensor<int32, [4]> var_54141_begin_0 = const()[name = tensor<string, []>("op_54141_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_54141_end_0 = const()[name = tensor<string, []>("op_54141_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_54141_end_mask_0 = const()[name = tensor<string, []>("op_54141_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54141_cast = slice_by_index(begin = var_54141_begin_0, end = var_54141_end_0, end_mask = var_54141_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54141_cast")];
            tensor<int32, [4]> var_54145_begin_0 = const()[name = tensor<string, []>("op_54145_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_54145_end_0 = const()[name = tensor<string, []>("op_54145_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_54145_end_mask_0 = const()[name = tensor<string, []>("op_54145_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54145_cast = slice_by_index(begin = var_54145_begin_0, end = var_54145_end_0, end_mask = var_54145_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54145_cast")];
            tensor<int32, [4]> var_54149_begin_0 = const()[name = tensor<string, []>("op_54149_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_54149_end_0 = const()[name = tensor<string, []>("op_54149_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_54149_end_mask_0 = const()[name = tensor<string, []>("op_54149_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54149_cast = slice_by_index(begin = var_54149_begin_0, end = var_54149_end_0, end_mask = var_54149_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54149_cast")];
            tensor<int32, [4]> var_54153_begin_0 = const()[name = tensor<string, []>("op_54153_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_54153_end_0 = const()[name = tensor<string, []>("op_54153_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_54153_end_mask_0 = const()[name = tensor<string, []>("op_54153_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54153_cast = slice_by_index(begin = var_54153_begin_0, end = var_54153_end_0, end_mask = var_54153_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54153_cast")];
            tensor<int32, [4]> var_54157_begin_0 = const()[name = tensor<string, []>("op_54157_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_54157_end_0 = const()[name = tensor<string, []>("op_54157_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_54157_end_mask_0 = const()[name = tensor<string, []>("op_54157_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54157_cast = slice_by_index(begin = var_54157_begin_0, end = var_54157_end_0, end_mask = var_54157_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54157_cast")];
            tensor<int32, [4]> var_54161_begin_0 = const()[name = tensor<string, []>("op_54161_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_54161_end_0 = const()[name = tensor<string, []>("op_54161_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_54161_end_mask_0 = const()[name = tensor<string, []>("op_54161_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54161_cast = slice_by_index(begin = var_54161_begin_0, end = var_54161_end_0, end_mask = var_54161_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54161_cast")];
            tensor<int32, [4]> var_54165_begin_0 = const()[name = tensor<string, []>("op_54165_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_54165_end_0 = const()[name = tensor<string, []>("op_54165_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_54165_end_mask_0 = const()[name = tensor<string, []>("op_54165_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54165_cast = slice_by_index(begin = var_54165_begin_0, end = var_54165_end_0, end_mask = var_54165_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54165_cast")];
            tensor<int32, [4]> var_54169_begin_0 = const()[name = tensor<string, []>("op_54169_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_54169_end_0 = const()[name = tensor<string, []>("op_54169_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_54169_end_mask_0 = const()[name = tensor<string, []>("op_54169_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54169_cast = slice_by_index(begin = var_54169_begin_0, end = var_54169_end_0, end_mask = var_54169_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54169_cast")];
            tensor<int32, [4]> var_54173_begin_0 = const()[name = tensor<string, []>("op_54173_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_54173_end_0 = const()[name = tensor<string, []>("op_54173_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_54173_end_mask_0 = const()[name = tensor<string, []>("op_54173_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54173_cast = slice_by_index(begin = var_54173_begin_0, end = var_54173_end_0, end_mask = var_54173_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54173_cast")];
            tensor<int32, [4]> var_54177_begin_0 = const()[name = tensor<string, []>("op_54177_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_54177_end_0 = const()[name = tensor<string, []>("op_54177_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_54177_end_mask_0 = const()[name = tensor<string, []>("op_54177_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54177_cast = slice_by_index(begin = var_54177_begin_0, end = var_54177_end_0, end_mask = var_54177_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54177_cast")];
            tensor<int32, [4]> var_54181_begin_0 = const()[name = tensor<string, []>("op_54181_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_54181_end_0 = const()[name = tensor<string, []>("op_54181_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_54181_end_mask_0 = const()[name = tensor<string, []>("op_54181_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_54181_cast = slice_by_index(begin = var_54181_begin_0, end = var_54181_end_0, end_mask = var_54181_end_mask_0, x = v_239_cast)[name = tensor<string, []>("op_54181_cast")];
            tensor<string, []> var_54185_equation_0 = const()[name = tensor<string, []>("op_54185_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54185_cast = einsum(equation = var_54185_equation_0, values = (var_54027_cast, var_53944_cast))[name = tensor<string, []>("op_54185_cast")];
            tensor<fp16, []> var_54186_to_fp16 = const()[name = tensor<string, []>("op_54186_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4601_cast = mul(x = var_54185_cast, y = var_54186_to_fp16)[name = tensor<string, []>("aw_4601_cast")];
            tensor<string, []> var_54189_equation_0 = const()[name = tensor<string, []>("op_54189_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54189_cast = einsum(equation = var_54189_equation_0, values = (var_54031_cast, var_53948_cast))[name = tensor<string, []>("op_54189_cast")];
            tensor<fp16, []> var_54190_to_fp16 = const()[name = tensor<string, []>("op_54190_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4603_cast = mul(x = var_54189_cast, y = var_54190_to_fp16)[name = tensor<string, []>("aw_4603_cast")];
            tensor<string, []> var_54193_equation_0 = const()[name = tensor<string, []>("op_54193_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54193_cast = einsum(equation = var_54193_equation_0, values = (var_54035_cast, var_53952_cast))[name = tensor<string, []>("op_54193_cast")];
            tensor<fp16, []> var_54194_to_fp16 = const()[name = tensor<string, []>("op_54194_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4605_cast = mul(x = var_54193_cast, y = var_54194_to_fp16)[name = tensor<string, []>("aw_4605_cast")];
            tensor<string, []> var_54197_equation_0 = const()[name = tensor<string, []>("op_54197_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54197_cast = einsum(equation = var_54197_equation_0, values = (var_54039_cast, var_53956_cast))[name = tensor<string, []>("op_54197_cast")];
            tensor<fp16, []> var_54198_to_fp16 = const()[name = tensor<string, []>("op_54198_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4607_cast = mul(x = var_54197_cast, y = var_54198_to_fp16)[name = tensor<string, []>("aw_4607_cast")];
            tensor<string, []> var_54201_equation_0 = const()[name = tensor<string, []>("op_54201_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54201_cast = einsum(equation = var_54201_equation_0, values = (var_54043_cast, var_53960_cast))[name = tensor<string, []>("op_54201_cast")];
            tensor<fp16, []> var_54202_to_fp16 = const()[name = tensor<string, []>("op_54202_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4609_cast = mul(x = var_54201_cast, y = var_54202_to_fp16)[name = tensor<string, []>("aw_4609_cast")];
            tensor<string, []> var_54205_equation_0 = const()[name = tensor<string, []>("op_54205_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54205_cast = einsum(equation = var_54205_equation_0, values = (var_54047_cast, var_53964_cast))[name = tensor<string, []>("op_54205_cast")];
            tensor<fp16, []> var_54206_to_fp16 = const()[name = tensor<string, []>("op_54206_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4611_cast = mul(x = var_54205_cast, y = var_54206_to_fp16)[name = tensor<string, []>("aw_4611_cast")];
            tensor<string, []> var_54209_equation_0 = const()[name = tensor<string, []>("op_54209_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54209_cast = einsum(equation = var_54209_equation_0, values = (var_54051_cast, var_53968_cast))[name = tensor<string, []>("op_54209_cast")];
            tensor<fp16, []> var_54210_to_fp16 = const()[name = tensor<string, []>("op_54210_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4613_cast = mul(x = var_54209_cast, y = var_54210_to_fp16)[name = tensor<string, []>("aw_4613_cast")];
            tensor<string, []> var_54213_equation_0 = const()[name = tensor<string, []>("op_54213_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54213_cast = einsum(equation = var_54213_equation_0, values = (var_54055_cast, var_53972_cast))[name = tensor<string, []>("op_54213_cast")];
            tensor<fp16, []> var_54214_to_fp16 = const()[name = tensor<string, []>("op_54214_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4615_cast = mul(x = var_54213_cast, y = var_54214_to_fp16)[name = tensor<string, []>("aw_4615_cast")];
            tensor<string, []> var_54217_equation_0 = const()[name = tensor<string, []>("op_54217_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54217_cast = einsum(equation = var_54217_equation_0, values = (var_54059_cast, var_53976_cast))[name = tensor<string, []>("op_54217_cast")];
            tensor<fp16, []> var_54218_to_fp16 = const()[name = tensor<string, []>("op_54218_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4617_cast = mul(x = var_54217_cast, y = var_54218_to_fp16)[name = tensor<string, []>("aw_4617_cast")];
            tensor<string, []> var_54221_equation_0 = const()[name = tensor<string, []>("op_54221_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54221_cast = einsum(equation = var_54221_equation_0, values = (var_54063_cast, var_53980_cast))[name = tensor<string, []>("op_54221_cast")];
            tensor<fp16, []> var_54222_to_fp16 = const()[name = tensor<string, []>("op_54222_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4619_cast = mul(x = var_54221_cast, y = var_54222_to_fp16)[name = tensor<string, []>("aw_4619_cast")];
            tensor<string, []> var_54225_equation_0 = const()[name = tensor<string, []>("op_54225_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54225_cast = einsum(equation = var_54225_equation_0, values = (var_54067_cast, var_53984_cast))[name = tensor<string, []>("op_54225_cast")];
            tensor<fp16, []> var_54226_to_fp16 = const()[name = tensor<string, []>("op_54226_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4621_cast = mul(x = var_54225_cast, y = var_54226_to_fp16)[name = tensor<string, []>("aw_4621_cast")];
            tensor<string, []> var_54229_equation_0 = const()[name = tensor<string, []>("op_54229_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54229_cast = einsum(equation = var_54229_equation_0, values = (var_54071_cast, var_53988_cast))[name = tensor<string, []>("op_54229_cast")];
            tensor<fp16, []> var_54230_to_fp16 = const()[name = tensor<string, []>("op_54230_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4623_cast = mul(x = var_54229_cast, y = var_54230_to_fp16)[name = tensor<string, []>("aw_4623_cast")];
            tensor<string, []> var_54233_equation_0 = const()[name = tensor<string, []>("op_54233_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54233_cast = einsum(equation = var_54233_equation_0, values = (var_54075_cast, var_53992_cast))[name = tensor<string, []>("op_54233_cast")];
            tensor<fp16, []> var_54234_to_fp16 = const()[name = tensor<string, []>("op_54234_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4625_cast = mul(x = var_54233_cast, y = var_54234_to_fp16)[name = tensor<string, []>("aw_4625_cast")];
            tensor<string, []> var_54237_equation_0 = const()[name = tensor<string, []>("op_54237_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54237_cast = einsum(equation = var_54237_equation_0, values = (var_54079_cast, var_53996_cast))[name = tensor<string, []>("op_54237_cast")];
            tensor<fp16, []> var_54238_to_fp16 = const()[name = tensor<string, []>("op_54238_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4627_cast = mul(x = var_54237_cast, y = var_54238_to_fp16)[name = tensor<string, []>("aw_4627_cast")];
            tensor<string, []> var_54241_equation_0 = const()[name = tensor<string, []>("op_54241_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54241_cast = einsum(equation = var_54241_equation_0, values = (var_54083_cast, var_54000_cast))[name = tensor<string, []>("op_54241_cast")];
            tensor<fp16, []> var_54242_to_fp16 = const()[name = tensor<string, []>("op_54242_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4629_cast = mul(x = var_54241_cast, y = var_54242_to_fp16)[name = tensor<string, []>("aw_4629_cast")];
            tensor<string, []> var_54245_equation_0 = const()[name = tensor<string, []>("op_54245_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54245_cast = einsum(equation = var_54245_equation_0, values = (var_54087_cast, var_54004_cast))[name = tensor<string, []>("op_54245_cast")];
            tensor<fp16, []> var_54246_to_fp16 = const()[name = tensor<string, []>("op_54246_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4631_cast = mul(x = var_54245_cast, y = var_54246_to_fp16)[name = tensor<string, []>("aw_4631_cast")];
            tensor<string, []> var_54249_equation_0 = const()[name = tensor<string, []>("op_54249_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54249_cast = einsum(equation = var_54249_equation_0, values = (var_54091_cast, var_54008_cast))[name = tensor<string, []>("op_54249_cast")];
            tensor<fp16, []> var_54250_to_fp16 = const()[name = tensor<string, []>("op_54250_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4633_cast = mul(x = var_54249_cast, y = var_54250_to_fp16)[name = tensor<string, []>("aw_4633_cast")];
            tensor<string, []> var_54253_equation_0 = const()[name = tensor<string, []>("op_54253_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54253_cast = einsum(equation = var_54253_equation_0, values = (var_54095_cast, var_54012_cast))[name = tensor<string, []>("op_54253_cast")];
            tensor<fp16, []> var_54254_to_fp16 = const()[name = tensor<string, []>("op_54254_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4635_cast = mul(x = var_54253_cast, y = var_54254_to_fp16)[name = tensor<string, []>("aw_4635_cast")];
            tensor<string, []> var_54257_equation_0 = const()[name = tensor<string, []>("op_54257_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54257_cast = einsum(equation = var_54257_equation_0, values = (var_54099_cast, var_54016_cast))[name = tensor<string, []>("op_54257_cast")];
            tensor<fp16, []> var_54258_to_fp16 = const()[name = tensor<string, []>("op_54258_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4637_cast = mul(x = var_54257_cast, y = var_54258_to_fp16)[name = tensor<string, []>("aw_4637_cast")];
            tensor<string, []> var_54261_equation_0 = const()[name = tensor<string, []>("op_54261_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_54261_cast = einsum(equation = var_54261_equation_0, values = (var_54103_cast, var_54020_cast))[name = tensor<string, []>("op_54261_cast")];
            tensor<fp16, []> var_54262_to_fp16 = const()[name = tensor<string, []>("op_54262_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4639_cast = mul(x = var_54261_cast, y = var_54262_to_fp16)[name = tensor<string, []>("aw_4639_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54264_cast = softmax(axis = var_30385, x = aw_4601_cast)[name = tensor<string, []>("op_54264_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54265_cast = softmax(axis = var_30385, x = aw_4603_cast)[name = tensor<string, []>("op_54265_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54266_cast = softmax(axis = var_30385, x = aw_4605_cast)[name = tensor<string, []>("op_54266_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54267_cast = softmax(axis = var_30385, x = aw_4607_cast)[name = tensor<string, []>("op_54267_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54268_cast = softmax(axis = var_30385, x = aw_4609_cast)[name = tensor<string, []>("op_54268_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54269_cast = softmax(axis = var_30385, x = aw_4611_cast)[name = tensor<string, []>("op_54269_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54270_cast = softmax(axis = var_30385, x = aw_4613_cast)[name = tensor<string, []>("op_54270_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54271_cast = softmax(axis = var_30385, x = aw_4615_cast)[name = tensor<string, []>("op_54271_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54272_cast = softmax(axis = var_30385, x = aw_4617_cast)[name = tensor<string, []>("op_54272_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54273_cast = softmax(axis = var_30385, x = aw_4619_cast)[name = tensor<string, []>("op_54273_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54274_cast = softmax(axis = var_30385, x = aw_4621_cast)[name = tensor<string, []>("op_54274_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54275_cast = softmax(axis = var_30385, x = aw_4623_cast)[name = tensor<string, []>("op_54275_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54276_cast = softmax(axis = var_30385, x = aw_4625_cast)[name = tensor<string, []>("op_54276_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54277_cast = softmax(axis = var_30385, x = aw_4627_cast)[name = tensor<string, []>("op_54277_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54278_cast = softmax(axis = var_30385, x = aw_4629_cast)[name = tensor<string, []>("op_54278_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54279_cast = softmax(axis = var_30385, x = aw_4631_cast)[name = tensor<string, []>("op_54279_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54280_cast = softmax(axis = var_30385, x = aw_4633_cast)[name = tensor<string, []>("op_54280_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54281_cast = softmax(axis = var_30385, x = aw_4635_cast)[name = tensor<string, []>("op_54281_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54282_cast = softmax(axis = var_30385, x = aw_4637_cast)[name = tensor<string, []>("op_54282_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_54283_cast = softmax(axis = var_30385, x = aw_4639_cast)[name = tensor<string, []>("op_54283_cast")];
            tensor<string, []> var_54285_equation_0 = const()[name = tensor<string, []>("op_54285_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54285_cast = einsum(equation = var_54285_equation_0, values = (var_54105_cast, var_54264_cast))[name = tensor<string, []>("op_54285_cast")];
            tensor<string, []> var_54287_equation_0 = const()[name = tensor<string, []>("op_54287_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54287_cast = einsum(equation = var_54287_equation_0, values = (var_54109_cast, var_54265_cast))[name = tensor<string, []>("op_54287_cast")];
            tensor<string, []> var_54289_equation_0 = const()[name = tensor<string, []>("op_54289_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54289_cast = einsum(equation = var_54289_equation_0, values = (var_54113_cast, var_54266_cast))[name = tensor<string, []>("op_54289_cast")];
            tensor<string, []> var_54291_equation_0 = const()[name = tensor<string, []>("op_54291_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54291_cast = einsum(equation = var_54291_equation_0, values = (var_54117_cast, var_54267_cast))[name = tensor<string, []>("op_54291_cast")];
            tensor<string, []> var_54293_equation_0 = const()[name = tensor<string, []>("op_54293_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54293_cast = einsum(equation = var_54293_equation_0, values = (var_54121_cast, var_54268_cast))[name = tensor<string, []>("op_54293_cast")];
            tensor<string, []> var_54295_equation_0 = const()[name = tensor<string, []>("op_54295_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54295_cast = einsum(equation = var_54295_equation_0, values = (var_54125_cast, var_54269_cast))[name = tensor<string, []>("op_54295_cast")];
            tensor<string, []> var_54297_equation_0 = const()[name = tensor<string, []>("op_54297_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54297_cast = einsum(equation = var_54297_equation_0, values = (var_54129_cast, var_54270_cast))[name = tensor<string, []>("op_54297_cast")];
            tensor<string, []> var_54299_equation_0 = const()[name = tensor<string, []>("op_54299_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54299_cast = einsum(equation = var_54299_equation_0, values = (var_54133_cast, var_54271_cast))[name = tensor<string, []>("op_54299_cast")];
            tensor<string, []> var_54301_equation_0 = const()[name = tensor<string, []>("op_54301_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54301_cast = einsum(equation = var_54301_equation_0, values = (var_54137_cast, var_54272_cast))[name = tensor<string, []>("op_54301_cast")];
            tensor<string, []> var_54303_equation_0 = const()[name = tensor<string, []>("op_54303_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54303_cast = einsum(equation = var_54303_equation_0, values = (var_54141_cast, var_54273_cast))[name = tensor<string, []>("op_54303_cast")];
            tensor<string, []> var_54305_equation_0 = const()[name = tensor<string, []>("op_54305_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54305_cast = einsum(equation = var_54305_equation_0, values = (var_54145_cast, var_54274_cast))[name = tensor<string, []>("op_54305_cast")];
            tensor<string, []> var_54307_equation_0 = const()[name = tensor<string, []>("op_54307_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54307_cast = einsum(equation = var_54307_equation_0, values = (var_54149_cast, var_54275_cast))[name = tensor<string, []>("op_54307_cast")];
            tensor<string, []> var_54309_equation_0 = const()[name = tensor<string, []>("op_54309_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54309_cast = einsum(equation = var_54309_equation_0, values = (var_54153_cast, var_54276_cast))[name = tensor<string, []>("op_54309_cast")];
            tensor<string, []> var_54311_equation_0 = const()[name = tensor<string, []>("op_54311_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54311_cast = einsum(equation = var_54311_equation_0, values = (var_54157_cast, var_54277_cast))[name = tensor<string, []>("op_54311_cast")];
            tensor<string, []> var_54313_equation_0 = const()[name = tensor<string, []>("op_54313_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54313_cast = einsum(equation = var_54313_equation_0, values = (var_54161_cast, var_54278_cast))[name = tensor<string, []>("op_54313_cast")];
            tensor<string, []> var_54315_equation_0 = const()[name = tensor<string, []>("op_54315_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54315_cast = einsum(equation = var_54315_equation_0, values = (var_54165_cast, var_54279_cast))[name = tensor<string, []>("op_54315_cast")];
            tensor<string, []> var_54317_equation_0 = const()[name = tensor<string, []>("op_54317_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54317_cast = einsum(equation = var_54317_equation_0, values = (var_54169_cast, var_54280_cast))[name = tensor<string, []>("op_54317_cast")];
            tensor<string, []> var_54319_equation_0 = const()[name = tensor<string, []>("op_54319_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54319_cast = einsum(equation = var_54319_equation_0, values = (var_54173_cast, var_54281_cast))[name = tensor<string, []>("op_54319_cast")];
            tensor<string, []> var_54321_equation_0 = const()[name = tensor<string, []>("op_54321_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54321_cast = einsum(equation = var_54321_equation_0, values = (var_54177_cast, var_54282_cast))[name = tensor<string, []>("op_54321_cast")];
            tensor<string, []> var_54323_equation_0 = const()[name = tensor<string, []>("op_54323_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54323_cast = einsum(equation = var_54323_equation_0, values = (var_54181_cast, var_54283_cast))[name = tensor<string, []>("op_54323_cast")];
            tensor<bool, []> input_679_interleave_0 = const()[name = tensor<string, []>("input_679_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_679_cast = concat(axis = var_30385, interleave = input_679_interleave_0, values = (var_54285_cast, var_54287_cast, var_54289_cast, var_54291_cast, var_54293_cast, var_54295_cast, var_54297_cast, var_54299_cast, var_54301_cast, var_54303_cast, var_54305_cast, var_54307_cast, var_54309_cast, var_54311_cast, var_54313_cast, var_54315_cast, var_54317_cast, var_54319_cast, var_54321_cast, var_54323_cast))[name = tensor<string, []>("input_679_cast")];
            tensor<int32, [2]> var_54329 = const()[name = tensor<string, []>("op_54329"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_54331 = const()[name = tensor<string, []>("op_54331"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_54333_pad_type_0 = const()[name = tensor<string, []>("op_54333_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_54333_pad_0 = const()[name = tensor<string, []>("op_54333_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4546539136)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4549816000)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_54333_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_out_0_bias_to_fp16, dilations = var_54331, groups = var_30385, pad = var_54333_pad_0, pad_type = var_54333_pad_type_0, strides = var_54329, weight = up_blocks_0_attentions_2_transformer_blocks_5_attn2_to_out_0_weight_to_fp16, x = input_679_cast)[name = tensor<string, []>("op_54333_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_359_cast = add(x = var_54333_cast, y = inputs_357_cast)[name = tensor<string, []>("inputs_359_cast")];
            tensor<int32, [1]> var_54337 = const()[name = tensor<string, []>("op_54337"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_359_cast = reduce_mean(axes = var_54337, keep_dims = var_30380, x = inputs_359_cast)[name = tensor<string, []>("channels_mean_359_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_359_cast = sub(x = inputs_359_cast, y = channels_mean_359_cast)[name = tensor<string, []>("zero_mean_359_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_359_cast = mul(x = zero_mean_359_cast, y = zero_mean_359_cast)[name = tensor<string, []>("zero_mean_sq_359_cast")];
            tensor<int32, [1]> var_54341 = const()[name = tensor<string, []>("op_54341"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_54342_cast = reduce_mean(axes = var_54341, keep_dims = var_30380, x = zero_mean_sq_359_cast)[name = tensor<string, []>("op_54342_cast")];
            tensor<fp16, []> var_54343_to_fp16 = const()[name = tensor<string, []>("op_54343_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_54344_cast = add(x = var_54342_cast, y = var_54343_to_fp16)[name = tensor<string, []>("op_54344_cast")];
            tensor<fp16, []> denom_359_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_359_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_359_cast = rsqrt(epsilon = denom_359_epsilon_0_to_fp16, x = var_54344_cast)[name = tensor<string, []>("denom_359_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_359_cast = mul(x = zero_mean_359_cast, y = denom_359_cast)[name = tensor<string, []>("out_359_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_54348_to_fp16 = const()[name = tensor<string, []>("op_54348_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4549818624)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_54349_cast = add(x = out_359_cast, y = var_54348_to_fp16)[name = tensor<string, []>("op_54349_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_54351_to_fp16 = const()[name = tensor<string, []>("op_54351_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4549821248)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_681_cast = mul(x = var_54349_cast, y = var_54351_to_fp16)[name = tensor<string, []>("input_681_cast")];
            tensor<int32, [2]> var_54359 = const()[name = tensor<string, []>("op_54359"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_54361 = const()[name = tensor<string, []>("op_54361"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_54363_pad_type_0 = const()[name = tensor<string, []>("op_54363_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_54363_pad_0 = const()[name = tensor<string, []>("op_54363_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4549823872)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_2_transformer_blocks_5_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4576038336)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_54363_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_5_ff_net_0_proj_bias_to_fp16, dilations = var_54361, groups = var_30385, pad = var_54363_pad_0, pad_type = var_54363_pad_type_0, strides = var_54359, weight = up_blocks_0_attentions_2_transformer_blocks_5_ff_net_0_proj_weight_to_fp16, x = input_681_cast)[name = tensor<string, []>("op_54363_cast")];
            tensor<int32, [2]> var_54364_split_sizes_0 = const()[name = tensor<string, []>("op_54364_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_54364_axis_0 = const()[name = tensor<string, []>("op_54364_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_54364_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_54364_cast_1 = split(axis = var_54364_axis_0, split_sizes = var_54364_split_sizes_0, x = var_54363_cast)[name = tensor<string, []>("op_54364_cast")];
            tensor<string, []> var_54366_mode_0 = const()[name = tensor<string, []>("op_54366_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_54366_cast = gelu(mode = var_54366_mode_0, x = var_54364_cast_1)[name = tensor<string, []>("op_54366_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_683_cast = mul(x = var_54364_cast_0, y = var_54366_cast)[name = tensor<string, []>("input_683_cast")];
            tensor<int32, [2]> var_54370 = const()[name = tensor<string, []>("op_54370"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_54372 = const()[name = tensor<string, []>("op_54372"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_54374_pad_type_0 = const()[name = tensor<string, []>("op_54374_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_54374_pad_0 = const()[name = tensor<string, []>("op_54374_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_5_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4576058880)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_5_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_5_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4589166144)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_54374_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_5_ff_net_2_bias_to_fp16, dilations = var_54372, groups = var_30385, pad = var_54374_pad_0, pad_type = var_54374_pad_type_0, strides = var_54370, weight = up_blocks_0_attentions_2_transformer_blocks_5_ff_net_2_weight_to_fp16, x = input_683_cast)[name = tensor<string, []>("op_54374_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_361_cast = add(x = var_54374_cast, y = inputs_359_cast)[name = tensor<string, []>("inputs_361_cast")];
            tensor<int32, [1]> var_54384 = const()[name = tensor<string, []>("op_54384"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_361_cast = reduce_mean(axes = var_54384, keep_dims = var_30380, x = inputs_361_cast)[name = tensor<string, []>("channels_mean_361_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_361_cast = sub(x = inputs_361_cast, y = channels_mean_361_cast)[name = tensor<string, []>("zero_mean_361_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_361_cast = mul(x = zero_mean_361_cast, y = zero_mean_361_cast)[name = tensor<string, []>("zero_mean_sq_361_cast")];
            tensor<int32, [1]> var_54388 = const()[name = tensor<string, []>("op_54388"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_54389_cast = reduce_mean(axes = var_54388, keep_dims = var_30380, x = zero_mean_sq_361_cast)[name = tensor<string, []>("op_54389_cast")];
            tensor<fp16, []> var_54390_to_fp16 = const()[name = tensor<string, []>("op_54390_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_54391_cast = add(x = var_54389_cast, y = var_54390_to_fp16)[name = tensor<string, []>("op_54391_cast")];
            tensor<fp16, []> denom_361_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_361_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_361_cast = rsqrt(epsilon = denom_361_epsilon_0_to_fp16, x = var_54391_cast)[name = tensor<string, []>("denom_361_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_361_cast = mul(x = zero_mean_361_cast, y = denom_361_cast)[name = tensor<string, []>("out_361_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_54395_to_fp16 = const()[name = tensor<string, []>("op_54395_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4589168768)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_54396_cast = add(x = out_361_cast, y = var_54395_to_fp16)[name = tensor<string, []>("op_54396_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_54398_to_fp16 = const()[name = tensor<string, []>("op_54398_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4589171392)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_471_cast = mul(x = var_54396_cast, y = var_54398_to_fp16)[name = tensor<string, []>("hidden_states_471_cast")];
            tensor<int32, [2]> var_54405 = const()[name = tensor<string, []>("op_54405"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_54407 = const()[name = tensor<string, []>("op_54407"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_241_pad_type_0 = const()[name = tensor<string, []>("q_241_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_241_pad_0 = const()[name = tensor<string, []>("q_241_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4589174016)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_241_cast = conv(dilations = var_54407, groups = var_30385, pad = q_241_pad_0, pad_type = q_241_pad_type_0, strides = var_54405, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_q_weight_to_fp16, x = hidden_states_471_cast)[name = tensor<string, []>("q_241_cast")];
            tensor<int32, [2]> var_54411 = const()[name = tensor<string, []>("op_54411"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_54413 = const()[name = tensor<string, []>("op_54413"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_481_pad_type_0 = const()[name = tensor<string, []>("k_481_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_481_pad_0 = const()[name = tensor<string, []>("k_481_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4592450880)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_481_cast = conv(dilations = var_54413, groups = var_30385, pad = k_481_pad_0, pad_type = k_481_pad_type_0, strides = var_54411, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_k_weight_to_fp16, x = hidden_states_471_cast)[name = tensor<string, []>("k_481_cast")];
            tensor<int32, [2]> var_54417 = const()[name = tensor<string, []>("op_54417"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_54419 = const()[name = tensor<string, []>("op_54419"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_241_pad_type_0 = const()[name = tensor<string, []>("v_241_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_241_pad_0 = const()[name = tensor<string, []>("v_241_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4595727744)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_241_cast = conv(dilations = var_54419, groups = var_30385, pad = v_241_pad_0, pad_type = v_241_pad_type_0, strides = var_54417, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_v_weight_to_fp16, x = hidden_states_471_cast)[name = tensor<string, []>("v_241_cast")];
            tensor<int32, [4]> var_54423_begin_0 = const()[name = tensor<string, []>("op_54423_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_54423_end_0 = const()[name = tensor<string, []>("op_54423_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_54423_end_mask_0 = const()[name = tensor<string, []>("op_54423_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54423_cast = slice_by_index(begin = var_54423_begin_0, end = var_54423_end_0, end_mask = var_54423_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54423_cast")];
            tensor<int32, [4]> var_54427_begin_0 = const()[name = tensor<string, []>("op_54427_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_54427_end_0 = const()[name = tensor<string, []>("op_54427_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_54427_end_mask_0 = const()[name = tensor<string, []>("op_54427_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54427_cast = slice_by_index(begin = var_54427_begin_0, end = var_54427_end_0, end_mask = var_54427_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54427_cast")];
            tensor<int32, [4]> var_54431_begin_0 = const()[name = tensor<string, []>("op_54431_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_54431_end_0 = const()[name = tensor<string, []>("op_54431_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_54431_end_mask_0 = const()[name = tensor<string, []>("op_54431_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54431_cast = slice_by_index(begin = var_54431_begin_0, end = var_54431_end_0, end_mask = var_54431_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54431_cast")];
            tensor<int32, [4]> var_54435_begin_0 = const()[name = tensor<string, []>("op_54435_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_54435_end_0 = const()[name = tensor<string, []>("op_54435_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_54435_end_mask_0 = const()[name = tensor<string, []>("op_54435_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54435_cast = slice_by_index(begin = var_54435_begin_0, end = var_54435_end_0, end_mask = var_54435_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54435_cast")];
            tensor<int32, [4]> var_54439_begin_0 = const()[name = tensor<string, []>("op_54439_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_54439_end_0 = const()[name = tensor<string, []>("op_54439_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_54439_end_mask_0 = const()[name = tensor<string, []>("op_54439_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54439_cast = slice_by_index(begin = var_54439_begin_0, end = var_54439_end_0, end_mask = var_54439_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54439_cast")];
            tensor<int32, [4]> var_54443_begin_0 = const()[name = tensor<string, []>("op_54443_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_54443_end_0 = const()[name = tensor<string, []>("op_54443_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_54443_end_mask_0 = const()[name = tensor<string, []>("op_54443_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54443_cast = slice_by_index(begin = var_54443_begin_0, end = var_54443_end_0, end_mask = var_54443_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54443_cast")];
            tensor<int32, [4]> var_54447_begin_0 = const()[name = tensor<string, []>("op_54447_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_54447_end_0 = const()[name = tensor<string, []>("op_54447_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_54447_end_mask_0 = const()[name = tensor<string, []>("op_54447_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54447_cast = slice_by_index(begin = var_54447_begin_0, end = var_54447_end_0, end_mask = var_54447_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54447_cast")];
            tensor<int32, [4]> var_54451_begin_0 = const()[name = tensor<string, []>("op_54451_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_54451_end_0 = const()[name = tensor<string, []>("op_54451_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_54451_end_mask_0 = const()[name = tensor<string, []>("op_54451_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54451_cast = slice_by_index(begin = var_54451_begin_0, end = var_54451_end_0, end_mask = var_54451_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54451_cast")];
            tensor<int32, [4]> var_54455_begin_0 = const()[name = tensor<string, []>("op_54455_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_54455_end_0 = const()[name = tensor<string, []>("op_54455_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_54455_end_mask_0 = const()[name = tensor<string, []>("op_54455_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54455_cast = slice_by_index(begin = var_54455_begin_0, end = var_54455_end_0, end_mask = var_54455_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54455_cast")];
            tensor<int32, [4]> var_54459_begin_0 = const()[name = tensor<string, []>("op_54459_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_54459_end_0 = const()[name = tensor<string, []>("op_54459_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_54459_end_mask_0 = const()[name = tensor<string, []>("op_54459_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54459_cast = slice_by_index(begin = var_54459_begin_0, end = var_54459_end_0, end_mask = var_54459_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54459_cast")];
            tensor<int32, [4]> var_54463_begin_0 = const()[name = tensor<string, []>("op_54463_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_54463_end_0 = const()[name = tensor<string, []>("op_54463_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_54463_end_mask_0 = const()[name = tensor<string, []>("op_54463_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54463_cast = slice_by_index(begin = var_54463_begin_0, end = var_54463_end_0, end_mask = var_54463_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54463_cast")];
            tensor<int32, [4]> var_54467_begin_0 = const()[name = tensor<string, []>("op_54467_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_54467_end_0 = const()[name = tensor<string, []>("op_54467_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_54467_end_mask_0 = const()[name = tensor<string, []>("op_54467_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54467_cast = slice_by_index(begin = var_54467_begin_0, end = var_54467_end_0, end_mask = var_54467_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54467_cast")];
            tensor<int32, [4]> var_54471_begin_0 = const()[name = tensor<string, []>("op_54471_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_54471_end_0 = const()[name = tensor<string, []>("op_54471_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_54471_end_mask_0 = const()[name = tensor<string, []>("op_54471_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54471_cast = slice_by_index(begin = var_54471_begin_0, end = var_54471_end_0, end_mask = var_54471_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54471_cast")];
            tensor<int32, [4]> var_54475_begin_0 = const()[name = tensor<string, []>("op_54475_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_54475_end_0 = const()[name = tensor<string, []>("op_54475_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_54475_end_mask_0 = const()[name = tensor<string, []>("op_54475_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54475_cast = slice_by_index(begin = var_54475_begin_0, end = var_54475_end_0, end_mask = var_54475_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54475_cast")];
            tensor<int32, [4]> var_54479_begin_0 = const()[name = tensor<string, []>("op_54479_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_54479_end_0 = const()[name = tensor<string, []>("op_54479_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_54479_end_mask_0 = const()[name = tensor<string, []>("op_54479_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54479_cast = slice_by_index(begin = var_54479_begin_0, end = var_54479_end_0, end_mask = var_54479_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54479_cast")];
            tensor<int32, [4]> var_54483_begin_0 = const()[name = tensor<string, []>("op_54483_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_54483_end_0 = const()[name = tensor<string, []>("op_54483_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_54483_end_mask_0 = const()[name = tensor<string, []>("op_54483_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54483_cast = slice_by_index(begin = var_54483_begin_0, end = var_54483_end_0, end_mask = var_54483_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54483_cast")];
            tensor<int32, [4]> var_54487_begin_0 = const()[name = tensor<string, []>("op_54487_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_54487_end_0 = const()[name = tensor<string, []>("op_54487_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_54487_end_mask_0 = const()[name = tensor<string, []>("op_54487_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54487_cast = slice_by_index(begin = var_54487_begin_0, end = var_54487_end_0, end_mask = var_54487_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54487_cast")];
            tensor<int32, [4]> var_54491_begin_0 = const()[name = tensor<string, []>("op_54491_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_54491_end_0 = const()[name = tensor<string, []>("op_54491_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_54491_end_mask_0 = const()[name = tensor<string, []>("op_54491_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54491_cast = slice_by_index(begin = var_54491_begin_0, end = var_54491_end_0, end_mask = var_54491_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54491_cast")];
            tensor<int32, [4]> var_54495_begin_0 = const()[name = tensor<string, []>("op_54495_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_54495_end_0 = const()[name = tensor<string, []>("op_54495_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_54495_end_mask_0 = const()[name = tensor<string, []>("op_54495_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54495_cast = slice_by_index(begin = var_54495_begin_0, end = var_54495_end_0, end_mask = var_54495_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54495_cast")];
            tensor<int32, [4]> var_54499_begin_0 = const()[name = tensor<string, []>("op_54499_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_54499_end_0 = const()[name = tensor<string, []>("op_54499_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_54499_end_mask_0 = const()[name = tensor<string, []>("op_54499_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54499_cast = slice_by_index(begin = var_54499_begin_0, end = var_54499_end_0, end_mask = var_54499_end_mask_0, x = q_241_cast)[name = tensor<string, []>("op_54499_cast")];
            tensor<int32, [4]> k_483_perm_0 = const()[name = tensor<string, []>("k_483_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_54506_begin_0 = const()[name = tensor<string, []>("op_54506_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_54506_end_0 = const()[name = tensor<string, []>("op_54506_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_54506_end_mask_0 = const()[name = tensor<string, []>("op_54506_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_19 = transpose(perm = k_483_perm_0, x = k_481_cast)[name = tensor<string, []>("transpose_19")];
            tensor<fp16, [2, 1024, 1, 64]> var_54506_cast = slice_by_index(begin = var_54506_begin_0, end = var_54506_end_0, end_mask = var_54506_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54506_cast")];
            tensor<int32, [4]> var_54510_begin_0 = const()[name = tensor<string, []>("op_54510_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_54510_end_0 = const()[name = tensor<string, []>("op_54510_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_54510_end_mask_0 = const()[name = tensor<string, []>("op_54510_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54510_cast = slice_by_index(begin = var_54510_begin_0, end = var_54510_end_0, end_mask = var_54510_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54510_cast")];
            tensor<int32, [4]> var_54514_begin_0 = const()[name = tensor<string, []>("op_54514_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_54514_end_0 = const()[name = tensor<string, []>("op_54514_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_54514_end_mask_0 = const()[name = tensor<string, []>("op_54514_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54514_cast = slice_by_index(begin = var_54514_begin_0, end = var_54514_end_0, end_mask = var_54514_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54514_cast")];
            tensor<int32, [4]> var_54518_begin_0 = const()[name = tensor<string, []>("op_54518_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_54518_end_0 = const()[name = tensor<string, []>("op_54518_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_54518_end_mask_0 = const()[name = tensor<string, []>("op_54518_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54518_cast = slice_by_index(begin = var_54518_begin_0, end = var_54518_end_0, end_mask = var_54518_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54518_cast")];
            tensor<int32, [4]> var_54522_begin_0 = const()[name = tensor<string, []>("op_54522_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_54522_end_0 = const()[name = tensor<string, []>("op_54522_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_54522_end_mask_0 = const()[name = tensor<string, []>("op_54522_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54522_cast = slice_by_index(begin = var_54522_begin_0, end = var_54522_end_0, end_mask = var_54522_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54522_cast")];
            tensor<int32, [4]> var_54526_begin_0 = const()[name = tensor<string, []>("op_54526_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_54526_end_0 = const()[name = tensor<string, []>("op_54526_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_54526_end_mask_0 = const()[name = tensor<string, []>("op_54526_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54526_cast = slice_by_index(begin = var_54526_begin_0, end = var_54526_end_0, end_mask = var_54526_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54526_cast")];
            tensor<int32, [4]> var_54530_begin_0 = const()[name = tensor<string, []>("op_54530_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_54530_end_0 = const()[name = tensor<string, []>("op_54530_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_54530_end_mask_0 = const()[name = tensor<string, []>("op_54530_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54530_cast = slice_by_index(begin = var_54530_begin_0, end = var_54530_end_0, end_mask = var_54530_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54530_cast")];
            tensor<int32, [4]> var_54534_begin_0 = const()[name = tensor<string, []>("op_54534_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_54534_end_0 = const()[name = tensor<string, []>("op_54534_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_54534_end_mask_0 = const()[name = tensor<string, []>("op_54534_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54534_cast = slice_by_index(begin = var_54534_begin_0, end = var_54534_end_0, end_mask = var_54534_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54534_cast")];
            tensor<int32, [4]> var_54538_begin_0 = const()[name = tensor<string, []>("op_54538_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_54538_end_0 = const()[name = tensor<string, []>("op_54538_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_54538_end_mask_0 = const()[name = tensor<string, []>("op_54538_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54538_cast = slice_by_index(begin = var_54538_begin_0, end = var_54538_end_0, end_mask = var_54538_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54538_cast")];
            tensor<int32, [4]> var_54542_begin_0 = const()[name = tensor<string, []>("op_54542_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_54542_end_0 = const()[name = tensor<string, []>("op_54542_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_54542_end_mask_0 = const()[name = tensor<string, []>("op_54542_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54542_cast = slice_by_index(begin = var_54542_begin_0, end = var_54542_end_0, end_mask = var_54542_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54542_cast")];
            tensor<int32, [4]> var_54546_begin_0 = const()[name = tensor<string, []>("op_54546_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_54546_end_0 = const()[name = tensor<string, []>("op_54546_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_54546_end_mask_0 = const()[name = tensor<string, []>("op_54546_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54546_cast = slice_by_index(begin = var_54546_begin_0, end = var_54546_end_0, end_mask = var_54546_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54546_cast")];
            tensor<int32, [4]> var_54550_begin_0 = const()[name = tensor<string, []>("op_54550_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_54550_end_0 = const()[name = tensor<string, []>("op_54550_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_54550_end_mask_0 = const()[name = tensor<string, []>("op_54550_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54550_cast = slice_by_index(begin = var_54550_begin_0, end = var_54550_end_0, end_mask = var_54550_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54550_cast")];
            tensor<int32, [4]> var_54554_begin_0 = const()[name = tensor<string, []>("op_54554_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_54554_end_0 = const()[name = tensor<string, []>("op_54554_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_54554_end_mask_0 = const()[name = tensor<string, []>("op_54554_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54554_cast = slice_by_index(begin = var_54554_begin_0, end = var_54554_end_0, end_mask = var_54554_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54554_cast")];
            tensor<int32, [4]> var_54558_begin_0 = const()[name = tensor<string, []>("op_54558_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_54558_end_0 = const()[name = tensor<string, []>("op_54558_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_54558_end_mask_0 = const()[name = tensor<string, []>("op_54558_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54558_cast = slice_by_index(begin = var_54558_begin_0, end = var_54558_end_0, end_mask = var_54558_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54558_cast")];
            tensor<int32, [4]> var_54562_begin_0 = const()[name = tensor<string, []>("op_54562_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_54562_end_0 = const()[name = tensor<string, []>("op_54562_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_54562_end_mask_0 = const()[name = tensor<string, []>("op_54562_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54562_cast = slice_by_index(begin = var_54562_begin_0, end = var_54562_end_0, end_mask = var_54562_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54562_cast")];
            tensor<int32, [4]> var_54566_begin_0 = const()[name = tensor<string, []>("op_54566_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_54566_end_0 = const()[name = tensor<string, []>("op_54566_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_54566_end_mask_0 = const()[name = tensor<string, []>("op_54566_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54566_cast = slice_by_index(begin = var_54566_begin_0, end = var_54566_end_0, end_mask = var_54566_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54566_cast")];
            tensor<int32, [4]> var_54570_begin_0 = const()[name = tensor<string, []>("op_54570_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_54570_end_0 = const()[name = tensor<string, []>("op_54570_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_54570_end_mask_0 = const()[name = tensor<string, []>("op_54570_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54570_cast = slice_by_index(begin = var_54570_begin_0, end = var_54570_end_0, end_mask = var_54570_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54570_cast")];
            tensor<int32, [4]> var_54574_begin_0 = const()[name = tensor<string, []>("op_54574_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_54574_end_0 = const()[name = tensor<string, []>("op_54574_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_54574_end_mask_0 = const()[name = tensor<string, []>("op_54574_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54574_cast = slice_by_index(begin = var_54574_begin_0, end = var_54574_end_0, end_mask = var_54574_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54574_cast")];
            tensor<int32, [4]> var_54578_begin_0 = const()[name = tensor<string, []>("op_54578_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_54578_end_0 = const()[name = tensor<string, []>("op_54578_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_54578_end_mask_0 = const()[name = tensor<string, []>("op_54578_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54578_cast = slice_by_index(begin = var_54578_begin_0, end = var_54578_end_0, end_mask = var_54578_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54578_cast")];
            tensor<int32, [4]> var_54582_begin_0 = const()[name = tensor<string, []>("op_54582_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_54582_end_0 = const()[name = tensor<string, []>("op_54582_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_54582_end_mask_0 = const()[name = tensor<string, []>("op_54582_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_54582_cast = slice_by_index(begin = var_54582_begin_0, end = var_54582_end_0, end_mask = var_54582_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_54582_cast")];
            tensor<int32, [4]> var_54584_begin_0 = const()[name = tensor<string, []>("op_54584_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_54584_end_0 = const()[name = tensor<string, []>("op_54584_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_54584_end_mask_0 = const()[name = tensor<string, []>("op_54584_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54584_cast = slice_by_index(begin = var_54584_begin_0, end = var_54584_end_0, end_mask = var_54584_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54584_cast")];
            tensor<int32, [4]> var_54588_begin_0 = const()[name = tensor<string, []>("op_54588_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_54588_end_0 = const()[name = tensor<string, []>("op_54588_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_54588_end_mask_0 = const()[name = tensor<string, []>("op_54588_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54588_cast = slice_by_index(begin = var_54588_begin_0, end = var_54588_end_0, end_mask = var_54588_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54588_cast")];
            tensor<int32, [4]> var_54592_begin_0 = const()[name = tensor<string, []>("op_54592_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_54592_end_0 = const()[name = tensor<string, []>("op_54592_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_54592_end_mask_0 = const()[name = tensor<string, []>("op_54592_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54592_cast = slice_by_index(begin = var_54592_begin_0, end = var_54592_end_0, end_mask = var_54592_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54592_cast")];
            tensor<int32, [4]> var_54596_begin_0 = const()[name = tensor<string, []>("op_54596_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_54596_end_0 = const()[name = tensor<string, []>("op_54596_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_54596_end_mask_0 = const()[name = tensor<string, []>("op_54596_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54596_cast = slice_by_index(begin = var_54596_begin_0, end = var_54596_end_0, end_mask = var_54596_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54596_cast")];
            tensor<int32, [4]> var_54600_begin_0 = const()[name = tensor<string, []>("op_54600_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_54600_end_0 = const()[name = tensor<string, []>("op_54600_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_54600_end_mask_0 = const()[name = tensor<string, []>("op_54600_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54600_cast = slice_by_index(begin = var_54600_begin_0, end = var_54600_end_0, end_mask = var_54600_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54600_cast")];
            tensor<int32, [4]> var_54604_begin_0 = const()[name = tensor<string, []>("op_54604_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_54604_end_0 = const()[name = tensor<string, []>("op_54604_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_54604_end_mask_0 = const()[name = tensor<string, []>("op_54604_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54604_cast = slice_by_index(begin = var_54604_begin_0, end = var_54604_end_0, end_mask = var_54604_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54604_cast")];
            tensor<int32, [4]> var_54608_begin_0 = const()[name = tensor<string, []>("op_54608_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_54608_end_0 = const()[name = tensor<string, []>("op_54608_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_54608_end_mask_0 = const()[name = tensor<string, []>("op_54608_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54608_cast = slice_by_index(begin = var_54608_begin_0, end = var_54608_end_0, end_mask = var_54608_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54608_cast")];
            tensor<int32, [4]> var_54612_begin_0 = const()[name = tensor<string, []>("op_54612_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_54612_end_0 = const()[name = tensor<string, []>("op_54612_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_54612_end_mask_0 = const()[name = tensor<string, []>("op_54612_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54612_cast = slice_by_index(begin = var_54612_begin_0, end = var_54612_end_0, end_mask = var_54612_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54612_cast")];
            tensor<int32, [4]> var_54616_begin_0 = const()[name = tensor<string, []>("op_54616_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_54616_end_0 = const()[name = tensor<string, []>("op_54616_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_54616_end_mask_0 = const()[name = tensor<string, []>("op_54616_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54616_cast = slice_by_index(begin = var_54616_begin_0, end = var_54616_end_0, end_mask = var_54616_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54616_cast")];
            tensor<int32, [4]> var_54620_begin_0 = const()[name = tensor<string, []>("op_54620_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_54620_end_0 = const()[name = tensor<string, []>("op_54620_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_54620_end_mask_0 = const()[name = tensor<string, []>("op_54620_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54620_cast = slice_by_index(begin = var_54620_begin_0, end = var_54620_end_0, end_mask = var_54620_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54620_cast")];
            tensor<int32, [4]> var_54624_begin_0 = const()[name = tensor<string, []>("op_54624_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_54624_end_0 = const()[name = tensor<string, []>("op_54624_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_54624_end_mask_0 = const()[name = tensor<string, []>("op_54624_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54624_cast = slice_by_index(begin = var_54624_begin_0, end = var_54624_end_0, end_mask = var_54624_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54624_cast")];
            tensor<int32, [4]> var_54628_begin_0 = const()[name = tensor<string, []>("op_54628_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_54628_end_0 = const()[name = tensor<string, []>("op_54628_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_54628_end_mask_0 = const()[name = tensor<string, []>("op_54628_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54628_cast = slice_by_index(begin = var_54628_begin_0, end = var_54628_end_0, end_mask = var_54628_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54628_cast")];
            tensor<int32, [4]> var_54632_begin_0 = const()[name = tensor<string, []>("op_54632_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_54632_end_0 = const()[name = tensor<string, []>("op_54632_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_54632_end_mask_0 = const()[name = tensor<string, []>("op_54632_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54632_cast = slice_by_index(begin = var_54632_begin_0, end = var_54632_end_0, end_mask = var_54632_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54632_cast")];
            tensor<int32, [4]> var_54636_begin_0 = const()[name = tensor<string, []>("op_54636_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_54636_end_0 = const()[name = tensor<string, []>("op_54636_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_54636_end_mask_0 = const()[name = tensor<string, []>("op_54636_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54636_cast = slice_by_index(begin = var_54636_begin_0, end = var_54636_end_0, end_mask = var_54636_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54636_cast")];
            tensor<int32, [4]> var_54640_begin_0 = const()[name = tensor<string, []>("op_54640_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_54640_end_0 = const()[name = tensor<string, []>("op_54640_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_54640_end_mask_0 = const()[name = tensor<string, []>("op_54640_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54640_cast = slice_by_index(begin = var_54640_begin_0, end = var_54640_end_0, end_mask = var_54640_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54640_cast")];
            tensor<int32, [4]> var_54644_begin_0 = const()[name = tensor<string, []>("op_54644_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_54644_end_0 = const()[name = tensor<string, []>("op_54644_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_54644_end_mask_0 = const()[name = tensor<string, []>("op_54644_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54644_cast = slice_by_index(begin = var_54644_begin_0, end = var_54644_end_0, end_mask = var_54644_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54644_cast")];
            tensor<int32, [4]> var_54648_begin_0 = const()[name = tensor<string, []>("op_54648_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_54648_end_0 = const()[name = tensor<string, []>("op_54648_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_54648_end_mask_0 = const()[name = tensor<string, []>("op_54648_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54648_cast = slice_by_index(begin = var_54648_begin_0, end = var_54648_end_0, end_mask = var_54648_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54648_cast")];
            tensor<int32, [4]> var_54652_begin_0 = const()[name = tensor<string, []>("op_54652_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_54652_end_0 = const()[name = tensor<string, []>("op_54652_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_54652_end_mask_0 = const()[name = tensor<string, []>("op_54652_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54652_cast = slice_by_index(begin = var_54652_begin_0, end = var_54652_end_0, end_mask = var_54652_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54652_cast")];
            tensor<int32, [4]> var_54656_begin_0 = const()[name = tensor<string, []>("op_54656_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_54656_end_0 = const()[name = tensor<string, []>("op_54656_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_54656_end_mask_0 = const()[name = tensor<string, []>("op_54656_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54656_cast = slice_by_index(begin = var_54656_begin_0, end = var_54656_end_0, end_mask = var_54656_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54656_cast")];
            tensor<int32, [4]> var_54660_begin_0 = const()[name = tensor<string, []>("op_54660_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_54660_end_0 = const()[name = tensor<string, []>("op_54660_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_54660_end_mask_0 = const()[name = tensor<string, []>("op_54660_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54660_cast = slice_by_index(begin = var_54660_begin_0, end = var_54660_end_0, end_mask = var_54660_end_mask_0, x = v_241_cast)[name = tensor<string, []>("op_54660_cast")];
            tensor<string, []> var_54664_equation_0 = const()[name = tensor<string, []>("op_54664_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54664_cast = einsum(equation = var_54664_equation_0, values = (var_54506_cast, var_54423_cast))[name = tensor<string, []>("op_54664_cast")];
            tensor<fp16, []> var_54665_to_fp16 = const()[name = tensor<string, []>("op_54665_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4641_cast = mul(x = var_54664_cast, y = var_54665_to_fp16)[name = tensor<string, []>("aw_4641_cast")];
            tensor<string, []> var_54668_equation_0 = const()[name = tensor<string, []>("op_54668_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54668_cast = einsum(equation = var_54668_equation_0, values = (var_54510_cast, var_54427_cast))[name = tensor<string, []>("op_54668_cast")];
            tensor<fp16, []> var_54669_to_fp16 = const()[name = tensor<string, []>("op_54669_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4643_cast = mul(x = var_54668_cast, y = var_54669_to_fp16)[name = tensor<string, []>("aw_4643_cast")];
            tensor<string, []> var_54672_equation_0 = const()[name = tensor<string, []>("op_54672_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54672_cast = einsum(equation = var_54672_equation_0, values = (var_54514_cast, var_54431_cast))[name = tensor<string, []>("op_54672_cast")];
            tensor<fp16, []> var_54673_to_fp16 = const()[name = tensor<string, []>("op_54673_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4645_cast = mul(x = var_54672_cast, y = var_54673_to_fp16)[name = tensor<string, []>("aw_4645_cast")];
            tensor<string, []> var_54676_equation_0 = const()[name = tensor<string, []>("op_54676_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54676_cast = einsum(equation = var_54676_equation_0, values = (var_54518_cast, var_54435_cast))[name = tensor<string, []>("op_54676_cast")];
            tensor<fp16, []> var_54677_to_fp16 = const()[name = tensor<string, []>("op_54677_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4647_cast = mul(x = var_54676_cast, y = var_54677_to_fp16)[name = tensor<string, []>("aw_4647_cast")];
            tensor<string, []> var_54680_equation_0 = const()[name = tensor<string, []>("op_54680_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54680_cast = einsum(equation = var_54680_equation_0, values = (var_54522_cast, var_54439_cast))[name = tensor<string, []>("op_54680_cast")];
            tensor<fp16, []> var_54681_to_fp16 = const()[name = tensor<string, []>("op_54681_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4649_cast = mul(x = var_54680_cast, y = var_54681_to_fp16)[name = tensor<string, []>("aw_4649_cast")];
            tensor<string, []> var_54684_equation_0 = const()[name = tensor<string, []>("op_54684_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54684_cast = einsum(equation = var_54684_equation_0, values = (var_54526_cast, var_54443_cast))[name = tensor<string, []>("op_54684_cast")];
            tensor<fp16, []> var_54685_to_fp16 = const()[name = tensor<string, []>("op_54685_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4651_cast = mul(x = var_54684_cast, y = var_54685_to_fp16)[name = tensor<string, []>("aw_4651_cast")];
            tensor<string, []> var_54688_equation_0 = const()[name = tensor<string, []>("op_54688_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54688_cast = einsum(equation = var_54688_equation_0, values = (var_54530_cast, var_54447_cast))[name = tensor<string, []>("op_54688_cast")];
            tensor<fp16, []> var_54689_to_fp16 = const()[name = tensor<string, []>("op_54689_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4653_cast = mul(x = var_54688_cast, y = var_54689_to_fp16)[name = tensor<string, []>("aw_4653_cast")];
            tensor<string, []> var_54692_equation_0 = const()[name = tensor<string, []>("op_54692_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54692_cast = einsum(equation = var_54692_equation_0, values = (var_54534_cast, var_54451_cast))[name = tensor<string, []>("op_54692_cast")];
            tensor<fp16, []> var_54693_to_fp16 = const()[name = tensor<string, []>("op_54693_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4655_cast = mul(x = var_54692_cast, y = var_54693_to_fp16)[name = tensor<string, []>("aw_4655_cast")];
            tensor<string, []> var_54696_equation_0 = const()[name = tensor<string, []>("op_54696_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54696_cast = einsum(equation = var_54696_equation_0, values = (var_54538_cast, var_54455_cast))[name = tensor<string, []>("op_54696_cast")];
            tensor<fp16, []> var_54697_to_fp16 = const()[name = tensor<string, []>("op_54697_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4657_cast = mul(x = var_54696_cast, y = var_54697_to_fp16)[name = tensor<string, []>("aw_4657_cast")];
            tensor<string, []> var_54700_equation_0 = const()[name = tensor<string, []>("op_54700_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54700_cast = einsum(equation = var_54700_equation_0, values = (var_54542_cast, var_54459_cast))[name = tensor<string, []>("op_54700_cast")];
            tensor<fp16, []> var_54701_to_fp16 = const()[name = tensor<string, []>("op_54701_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4659_cast = mul(x = var_54700_cast, y = var_54701_to_fp16)[name = tensor<string, []>("aw_4659_cast")];
            tensor<string, []> var_54704_equation_0 = const()[name = tensor<string, []>("op_54704_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54704_cast = einsum(equation = var_54704_equation_0, values = (var_54546_cast, var_54463_cast))[name = tensor<string, []>("op_54704_cast")];
            tensor<fp16, []> var_54705_to_fp16 = const()[name = tensor<string, []>("op_54705_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4661_cast = mul(x = var_54704_cast, y = var_54705_to_fp16)[name = tensor<string, []>("aw_4661_cast")];
            tensor<string, []> var_54708_equation_0 = const()[name = tensor<string, []>("op_54708_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54708_cast = einsum(equation = var_54708_equation_0, values = (var_54550_cast, var_54467_cast))[name = tensor<string, []>("op_54708_cast")];
            tensor<fp16, []> var_54709_to_fp16 = const()[name = tensor<string, []>("op_54709_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4663_cast = mul(x = var_54708_cast, y = var_54709_to_fp16)[name = tensor<string, []>("aw_4663_cast")];
            tensor<string, []> var_54712_equation_0 = const()[name = tensor<string, []>("op_54712_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54712_cast = einsum(equation = var_54712_equation_0, values = (var_54554_cast, var_54471_cast))[name = tensor<string, []>("op_54712_cast")];
            tensor<fp16, []> var_54713_to_fp16 = const()[name = tensor<string, []>("op_54713_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4665_cast = mul(x = var_54712_cast, y = var_54713_to_fp16)[name = tensor<string, []>("aw_4665_cast")];
            tensor<string, []> var_54716_equation_0 = const()[name = tensor<string, []>("op_54716_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54716_cast = einsum(equation = var_54716_equation_0, values = (var_54558_cast, var_54475_cast))[name = tensor<string, []>("op_54716_cast")];
            tensor<fp16, []> var_54717_to_fp16 = const()[name = tensor<string, []>("op_54717_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4667_cast = mul(x = var_54716_cast, y = var_54717_to_fp16)[name = tensor<string, []>("aw_4667_cast")];
            tensor<string, []> var_54720_equation_0 = const()[name = tensor<string, []>("op_54720_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54720_cast = einsum(equation = var_54720_equation_0, values = (var_54562_cast, var_54479_cast))[name = tensor<string, []>("op_54720_cast")];
            tensor<fp16, []> var_54721_to_fp16 = const()[name = tensor<string, []>("op_54721_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4669_cast = mul(x = var_54720_cast, y = var_54721_to_fp16)[name = tensor<string, []>("aw_4669_cast")];
            tensor<string, []> var_54724_equation_0 = const()[name = tensor<string, []>("op_54724_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54724_cast = einsum(equation = var_54724_equation_0, values = (var_54566_cast, var_54483_cast))[name = tensor<string, []>("op_54724_cast")];
            tensor<fp16, []> var_54725_to_fp16 = const()[name = tensor<string, []>("op_54725_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4671_cast = mul(x = var_54724_cast, y = var_54725_to_fp16)[name = tensor<string, []>("aw_4671_cast")];
            tensor<string, []> var_54728_equation_0 = const()[name = tensor<string, []>("op_54728_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54728_cast = einsum(equation = var_54728_equation_0, values = (var_54570_cast, var_54487_cast))[name = tensor<string, []>("op_54728_cast")];
            tensor<fp16, []> var_54729_to_fp16 = const()[name = tensor<string, []>("op_54729_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4673_cast = mul(x = var_54728_cast, y = var_54729_to_fp16)[name = tensor<string, []>("aw_4673_cast")];
            tensor<string, []> var_54732_equation_0 = const()[name = tensor<string, []>("op_54732_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54732_cast = einsum(equation = var_54732_equation_0, values = (var_54574_cast, var_54491_cast))[name = tensor<string, []>("op_54732_cast")];
            tensor<fp16, []> var_54733_to_fp16 = const()[name = tensor<string, []>("op_54733_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4675_cast = mul(x = var_54732_cast, y = var_54733_to_fp16)[name = tensor<string, []>("aw_4675_cast")];
            tensor<string, []> var_54736_equation_0 = const()[name = tensor<string, []>("op_54736_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54736_cast = einsum(equation = var_54736_equation_0, values = (var_54578_cast, var_54495_cast))[name = tensor<string, []>("op_54736_cast")];
            tensor<fp16, []> var_54737_to_fp16 = const()[name = tensor<string, []>("op_54737_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4677_cast = mul(x = var_54736_cast, y = var_54737_to_fp16)[name = tensor<string, []>("aw_4677_cast")];
            tensor<string, []> var_54740_equation_0 = const()[name = tensor<string, []>("op_54740_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54740_cast = einsum(equation = var_54740_equation_0, values = (var_54582_cast, var_54499_cast))[name = tensor<string, []>("op_54740_cast")];
            tensor<fp16, []> var_54741_to_fp16 = const()[name = tensor<string, []>("op_54741_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4679_cast = mul(x = var_54740_cast, y = var_54741_to_fp16)[name = tensor<string, []>("aw_4679_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54743_cast = softmax(axis = var_30385, x = aw_4641_cast)[name = tensor<string, []>("op_54743_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54744_cast = softmax(axis = var_30385, x = aw_4643_cast)[name = tensor<string, []>("op_54744_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54745_cast = softmax(axis = var_30385, x = aw_4645_cast)[name = tensor<string, []>("op_54745_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54746_cast = softmax(axis = var_30385, x = aw_4647_cast)[name = tensor<string, []>("op_54746_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54747_cast = softmax(axis = var_30385, x = aw_4649_cast)[name = tensor<string, []>("op_54747_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54748_cast = softmax(axis = var_30385, x = aw_4651_cast)[name = tensor<string, []>("op_54748_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54749_cast = softmax(axis = var_30385, x = aw_4653_cast)[name = tensor<string, []>("op_54749_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54750_cast = softmax(axis = var_30385, x = aw_4655_cast)[name = tensor<string, []>("op_54750_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54751_cast = softmax(axis = var_30385, x = aw_4657_cast)[name = tensor<string, []>("op_54751_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54752_cast = softmax(axis = var_30385, x = aw_4659_cast)[name = tensor<string, []>("op_54752_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54753_cast = softmax(axis = var_30385, x = aw_4661_cast)[name = tensor<string, []>("op_54753_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54754_cast = softmax(axis = var_30385, x = aw_4663_cast)[name = tensor<string, []>("op_54754_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54755_cast = softmax(axis = var_30385, x = aw_4665_cast)[name = tensor<string, []>("op_54755_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54756_cast = softmax(axis = var_30385, x = aw_4667_cast)[name = tensor<string, []>("op_54756_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54757_cast = softmax(axis = var_30385, x = aw_4669_cast)[name = tensor<string, []>("op_54757_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54758_cast = softmax(axis = var_30385, x = aw_4671_cast)[name = tensor<string, []>("op_54758_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54759_cast = softmax(axis = var_30385, x = aw_4673_cast)[name = tensor<string, []>("op_54759_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54760_cast = softmax(axis = var_30385, x = aw_4675_cast)[name = tensor<string, []>("op_54760_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54761_cast = softmax(axis = var_30385, x = aw_4677_cast)[name = tensor<string, []>("op_54761_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_54762_cast = softmax(axis = var_30385, x = aw_4679_cast)[name = tensor<string, []>("op_54762_cast")];
            tensor<string, []> var_54764_equation_0 = const()[name = tensor<string, []>("op_54764_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54764_cast = einsum(equation = var_54764_equation_0, values = (var_54584_cast, var_54743_cast))[name = tensor<string, []>("op_54764_cast")];
            tensor<string, []> var_54766_equation_0 = const()[name = tensor<string, []>("op_54766_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54766_cast = einsum(equation = var_54766_equation_0, values = (var_54588_cast, var_54744_cast))[name = tensor<string, []>("op_54766_cast")];
            tensor<string, []> var_54768_equation_0 = const()[name = tensor<string, []>("op_54768_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54768_cast = einsum(equation = var_54768_equation_0, values = (var_54592_cast, var_54745_cast))[name = tensor<string, []>("op_54768_cast")];
            tensor<string, []> var_54770_equation_0 = const()[name = tensor<string, []>("op_54770_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54770_cast = einsum(equation = var_54770_equation_0, values = (var_54596_cast, var_54746_cast))[name = tensor<string, []>("op_54770_cast")];
            tensor<string, []> var_54772_equation_0 = const()[name = tensor<string, []>("op_54772_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54772_cast = einsum(equation = var_54772_equation_0, values = (var_54600_cast, var_54747_cast))[name = tensor<string, []>("op_54772_cast")];
            tensor<string, []> var_54774_equation_0 = const()[name = tensor<string, []>("op_54774_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54774_cast = einsum(equation = var_54774_equation_0, values = (var_54604_cast, var_54748_cast))[name = tensor<string, []>("op_54774_cast")];
            tensor<string, []> var_54776_equation_0 = const()[name = tensor<string, []>("op_54776_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54776_cast = einsum(equation = var_54776_equation_0, values = (var_54608_cast, var_54749_cast))[name = tensor<string, []>("op_54776_cast")];
            tensor<string, []> var_54778_equation_0 = const()[name = tensor<string, []>("op_54778_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54778_cast = einsum(equation = var_54778_equation_0, values = (var_54612_cast, var_54750_cast))[name = tensor<string, []>("op_54778_cast")];
            tensor<string, []> var_54780_equation_0 = const()[name = tensor<string, []>("op_54780_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54780_cast = einsum(equation = var_54780_equation_0, values = (var_54616_cast, var_54751_cast))[name = tensor<string, []>("op_54780_cast")];
            tensor<string, []> var_54782_equation_0 = const()[name = tensor<string, []>("op_54782_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54782_cast = einsum(equation = var_54782_equation_0, values = (var_54620_cast, var_54752_cast))[name = tensor<string, []>("op_54782_cast")];
            tensor<string, []> var_54784_equation_0 = const()[name = tensor<string, []>("op_54784_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54784_cast = einsum(equation = var_54784_equation_0, values = (var_54624_cast, var_54753_cast))[name = tensor<string, []>("op_54784_cast")];
            tensor<string, []> var_54786_equation_0 = const()[name = tensor<string, []>("op_54786_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54786_cast = einsum(equation = var_54786_equation_0, values = (var_54628_cast, var_54754_cast))[name = tensor<string, []>("op_54786_cast")];
            tensor<string, []> var_54788_equation_0 = const()[name = tensor<string, []>("op_54788_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54788_cast = einsum(equation = var_54788_equation_0, values = (var_54632_cast, var_54755_cast))[name = tensor<string, []>("op_54788_cast")];
            tensor<string, []> var_54790_equation_0 = const()[name = tensor<string, []>("op_54790_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54790_cast = einsum(equation = var_54790_equation_0, values = (var_54636_cast, var_54756_cast))[name = tensor<string, []>("op_54790_cast")];
            tensor<string, []> var_54792_equation_0 = const()[name = tensor<string, []>("op_54792_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54792_cast = einsum(equation = var_54792_equation_0, values = (var_54640_cast, var_54757_cast))[name = tensor<string, []>("op_54792_cast")];
            tensor<string, []> var_54794_equation_0 = const()[name = tensor<string, []>("op_54794_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54794_cast = einsum(equation = var_54794_equation_0, values = (var_54644_cast, var_54758_cast))[name = tensor<string, []>("op_54794_cast")];
            tensor<string, []> var_54796_equation_0 = const()[name = tensor<string, []>("op_54796_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54796_cast = einsum(equation = var_54796_equation_0, values = (var_54648_cast, var_54759_cast))[name = tensor<string, []>("op_54796_cast")];
            tensor<string, []> var_54798_equation_0 = const()[name = tensor<string, []>("op_54798_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54798_cast = einsum(equation = var_54798_equation_0, values = (var_54652_cast, var_54760_cast))[name = tensor<string, []>("op_54798_cast")];
            tensor<string, []> var_54800_equation_0 = const()[name = tensor<string, []>("op_54800_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54800_cast = einsum(equation = var_54800_equation_0, values = (var_54656_cast, var_54761_cast))[name = tensor<string, []>("op_54800_cast")];
            tensor<string, []> var_54802_equation_0 = const()[name = tensor<string, []>("op_54802_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_54802_cast = einsum(equation = var_54802_equation_0, values = (var_54660_cast, var_54762_cast))[name = tensor<string, []>("op_54802_cast")];
            tensor<bool, []> input_685_interleave_0 = const()[name = tensor<string, []>("input_685_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_685_cast = concat(axis = var_30385, interleave = input_685_interleave_0, values = (var_54764_cast, var_54766_cast, var_54768_cast, var_54770_cast, var_54772_cast, var_54774_cast, var_54776_cast, var_54778_cast, var_54780_cast, var_54782_cast, var_54784_cast, var_54786_cast, var_54788_cast, var_54790_cast, var_54792_cast, var_54794_cast, var_54796_cast, var_54798_cast, var_54800_cast, var_54802_cast))[name = tensor<string, []>("input_685_cast")];
            tensor<int32, [2]> var_54808 = const()[name = tensor<string, []>("op_54808"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_54810 = const()[name = tensor<string, []>("op_54810"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_54812_pad_type_0 = const()[name = tensor<string, []>("op_54812_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_54812_pad_0 = const()[name = tensor<string, []>("op_54812_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4599004608)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4602281472)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_54812_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_out_0_bias_to_fp16, dilations = var_54810, groups = var_30385, pad = var_54812_pad_0, pad_type = var_54812_pad_type_0, strides = var_54808, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn1_to_out_0_weight_to_fp16, x = input_685_cast)[name = tensor<string, []>("op_54812_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_363_cast = add(x = var_54812_cast, y = inputs_361_cast)[name = tensor<string, []>("inputs_363_cast")];
            tensor<int32, [1]> var_54816 = const()[name = tensor<string, []>("op_54816"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_363_cast = reduce_mean(axes = var_54816, keep_dims = var_30380, x = inputs_363_cast)[name = tensor<string, []>("channels_mean_363_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_363_cast = sub(x = inputs_363_cast, y = channels_mean_363_cast)[name = tensor<string, []>("zero_mean_363_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_363_cast = mul(x = zero_mean_363_cast, y = zero_mean_363_cast)[name = tensor<string, []>("zero_mean_sq_363_cast")];
            tensor<int32, [1]> var_54820 = const()[name = tensor<string, []>("op_54820"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_54821_cast = reduce_mean(axes = var_54820, keep_dims = var_30380, x = zero_mean_sq_363_cast)[name = tensor<string, []>("op_54821_cast")];
            tensor<fp16, []> var_54822_to_fp16 = const()[name = tensor<string, []>("op_54822_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_54823_cast = add(x = var_54821_cast, y = var_54822_to_fp16)[name = tensor<string, []>("op_54823_cast")];
            tensor<fp16, []> denom_363_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_363_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_363_cast = rsqrt(epsilon = denom_363_epsilon_0_to_fp16, x = var_54823_cast)[name = tensor<string, []>("denom_363_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_363_cast = mul(x = zero_mean_363_cast, y = denom_363_cast)[name = tensor<string, []>("out_363_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_54827_to_fp16 = const()[name = tensor<string, []>("op_54827_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4602284096)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_54828_cast = add(x = out_363_cast, y = var_54827_to_fp16)[name = tensor<string, []>("op_54828_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_54830_to_fp16 = const()[name = tensor<string, []>("op_54830_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4602286720)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_473_cast = mul(x = var_54828_cast, y = var_54830_to_fp16)[name = tensor<string, []>("hidden_states_473_cast")];
            tensor<int32, [2]> var_54837 = const()[name = tensor<string, []>("op_54837"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_54839 = const()[name = tensor<string, []>("op_54839"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_243_pad_type_0 = const()[name = tensor<string, []>("q_243_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_243_pad_0 = const()[name = tensor<string, []>("q_243_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4602289344)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_243_cast = conv(dilations = var_54839, groups = var_30385, pad = q_243_pad_0, pad_type = q_243_pad_type_0, strides = var_54837, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_q_weight_to_fp16, x = hidden_states_473_cast)[name = tensor<string, []>("q_243_cast")];
            tensor<int32, [2]> var_54843 = const()[name = tensor<string, []>("op_54843"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_54845 = const()[name = tensor<string, []>("op_54845"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_485_pad_type_0 = const()[name = tensor<string, []>("k_485_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_485_pad_0 = const()[name = tensor<string, []>("k_485_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4605566208)))];
            tensor<fp16, [2, 1280, 1, 77]> k_485_cast = conv(dilations = var_54845, groups = var_30385, pad = k_485_pad_0, pad_type = k_485_pad_type_0, strides = var_54843, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_485_cast")];
            tensor<int32, [2]> var_54849 = const()[name = tensor<string, []>("op_54849"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_54851 = const()[name = tensor<string, []>("op_54851"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_243_pad_type_0 = const()[name = tensor<string, []>("v_243_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_243_pad_0 = const()[name = tensor<string, []>("v_243_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4610809152)))];
            tensor<fp16, [2, 1280, 1, 77]> v_243_cast = conv(dilations = var_54851, groups = var_30385, pad = v_243_pad_0, pad_type = v_243_pad_type_0, strides = var_54849, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_243_cast")];
            tensor<int32, [4]> var_54855_begin_0 = const()[name = tensor<string, []>("op_54855_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_54855_end_0 = const()[name = tensor<string, []>("op_54855_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_54855_end_mask_0 = const()[name = tensor<string, []>("op_54855_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54855_cast = slice_by_index(begin = var_54855_begin_0, end = var_54855_end_0, end_mask = var_54855_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54855_cast")];
            tensor<int32, [4]> var_54859_begin_0 = const()[name = tensor<string, []>("op_54859_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_54859_end_0 = const()[name = tensor<string, []>("op_54859_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_54859_end_mask_0 = const()[name = tensor<string, []>("op_54859_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54859_cast = slice_by_index(begin = var_54859_begin_0, end = var_54859_end_0, end_mask = var_54859_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54859_cast")];
            tensor<int32, [4]> var_54863_begin_0 = const()[name = tensor<string, []>("op_54863_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_54863_end_0 = const()[name = tensor<string, []>("op_54863_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_54863_end_mask_0 = const()[name = tensor<string, []>("op_54863_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54863_cast = slice_by_index(begin = var_54863_begin_0, end = var_54863_end_0, end_mask = var_54863_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54863_cast")];
            tensor<int32, [4]> var_54867_begin_0 = const()[name = tensor<string, []>("op_54867_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_54867_end_0 = const()[name = tensor<string, []>("op_54867_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_54867_end_mask_0 = const()[name = tensor<string, []>("op_54867_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54867_cast = slice_by_index(begin = var_54867_begin_0, end = var_54867_end_0, end_mask = var_54867_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54867_cast")];
            tensor<int32, [4]> var_54871_begin_0 = const()[name = tensor<string, []>("op_54871_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_54871_end_0 = const()[name = tensor<string, []>("op_54871_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_54871_end_mask_0 = const()[name = tensor<string, []>("op_54871_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54871_cast = slice_by_index(begin = var_54871_begin_0, end = var_54871_end_0, end_mask = var_54871_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54871_cast")];
            tensor<int32, [4]> var_54875_begin_0 = const()[name = tensor<string, []>("op_54875_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_54875_end_0 = const()[name = tensor<string, []>("op_54875_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_54875_end_mask_0 = const()[name = tensor<string, []>("op_54875_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54875_cast = slice_by_index(begin = var_54875_begin_0, end = var_54875_end_0, end_mask = var_54875_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54875_cast")];
            tensor<int32, [4]> var_54879_begin_0 = const()[name = tensor<string, []>("op_54879_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_54879_end_0 = const()[name = tensor<string, []>("op_54879_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_54879_end_mask_0 = const()[name = tensor<string, []>("op_54879_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54879_cast = slice_by_index(begin = var_54879_begin_0, end = var_54879_end_0, end_mask = var_54879_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54879_cast")];
            tensor<int32, [4]> var_54883_begin_0 = const()[name = tensor<string, []>("op_54883_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_54883_end_0 = const()[name = tensor<string, []>("op_54883_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_54883_end_mask_0 = const()[name = tensor<string, []>("op_54883_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54883_cast = slice_by_index(begin = var_54883_begin_0, end = var_54883_end_0, end_mask = var_54883_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54883_cast")];
            tensor<int32, [4]> var_54887_begin_0 = const()[name = tensor<string, []>("op_54887_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_54887_end_0 = const()[name = tensor<string, []>("op_54887_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_54887_end_mask_0 = const()[name = tensor<string, []>("op_54887_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54887_cast = slice_by_index(begin = var_54887_begin_0, end = var_54887_end_0, end_mask = var_54887_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54887_cast")];
            tensor<int32, [4]> var_54891_begin_0 = const()[name = tensor<string, []>("op_54891_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_54891_end_0 = const()[name = tensor<string, []>("op_54891_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_54891_end_mask_0 = const()[name = tensor<string, []>("op_54891_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54891_cast = slice_by_index(begin = var_54891_begin_0, end = var_54891_end_0, end_mask = var_54891_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54891_cast")];
            tensor<int32, [4]> var_54895_begin_0 = const()[name = tensor<string, []>("op_54895_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_54895_end_0 = const()[name = tensor<string, []>("op_54895_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_54895_end_mask_0 = const()[name = tensor<string, []>("op_54895_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54895_cast = slice_by_index(begin = var_54895_begin_0, end = var_54895_end_0, end_mask = var_54895_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54895_cast")];
            tensor<int32, [4]> var_54899_begin_0 = const()[name = tensor<string, []>("op_54899_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_54899_end_0 = const()[name = tensor<string, []>("op_54899_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_54899_end_mask_0 = const()[name = tensor<string, []>("op_54899_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54899_cast = slice_by_index(begin = var_54899_begin_0, end = var_54899_end_0, end_mask = var_54899_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54899_cast")];
            tensor<int32, [4]> var_54903_begin_0 = const()[name = tensor<string, []>("op_54903_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_54903_end_0 = const()[name = tensor<string, []>("op_54903_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_54903_end_mask_0 = const()[name = tensor<string, []>("op_54903_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54903_cast = slice_by_index(begin = var_54903_begin_0, end = var_54903_end_0, end_mask = var_54903_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54903_cast")];
            tensor<int32, [4]> var_54907_begin_0 = const()[name = tensor<string, []>("op_54907_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_54907_end_0 = const()[name = tensor<string, []>("op_54907_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_54907_end_mask_0 = const()[name = tensor<string, []>("op_54907_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54907_cast = slice_by_index(begin = var_54907_begin_0, end = var_54907_end_0, end_mask = var_54907_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54907_cast")];
            tensor<int32, [4]> var_54911_begin_0 = const()[name = tensor<string, []>("op_54911_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_54911_end_0 = const()[name = tensor<string, []>("op_54911_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_54911_end_mask_0 = const()[name = tensor<string, []>("op_54911_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54911_cast = slice_by_index(begin = var_54911_begin_0, end = var_54911_end_0, end_mask = var_54911_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54911_cast")];
            tensor<int32, [4]> var_54915_begin_0 = const()[name = tensor<string, []>("op_54915_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_54915_end_0 = const()[name = tensor<string, []>("op_54915_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_54915_end_mask_0 = const()[name = tensor<string, []>("op_54915_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54915_cast = slice_by_index(begin = var_54915_begin_0, end = var_54915_end_0, end_mask = var_54915_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54915_cast")];
            tensor<int32, [4]> var_54919_begin_0 = const()[name = tensor<string, []>("op_54919_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_54919_end_0 = const()[name = tensor<string, []>("op_54919_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_54919_end_mask_0 = const()[name = tensor<string, []>("op_54919_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54919_cast = slice_by_index(begin = var_54919_begin_0, end = var_54919_end_0, end_mask = var_54919_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54919_cast")];
            tensor<int32, [4]> var_54923_begin_0 = const()[name = tensor<string, []>("op_54923_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_54923_end_0 = const()[name = tensor<string, []>("op_54923_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_54923_end_mask_0 = const()[name = tensor<string, []>("op_54923_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54923_cast = slice_by_index(begin = var_54923_begin_0, end = var_54923_end_0, end_mask = var_54923_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54923_cast")];
            tensor<int32, [4]> var_54927_begin_0 = const()[name = tensor<string, []>("op_54927_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_54927_end_0 = const()[name = tensor<string, []>("op_54927_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_54927_end_mask_0 = const()[name = tensor<string, []>("op_54927_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54927_cast = slice_by_index(begin = var_54927_begin_0, end = var_54927_end_0, end_mask = var_54927_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54927_cast")];
            tensor<int32, [4]> var_54931_begin_0 = const()[name = tensor<string, []>("op_54931_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_54931_end_0 = const()[name = tensor<string, []>("op_54931_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_54931_end_mask_0 = const()[name = tensor<string, []>("op_54931_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_54931_cast = slice_by_index(begin = var_54931_begin_0, end = var_54931_end_0, end_mask = var_54931_end_mask_0, x = q_243_cast)[name = tensor<string, []>("op_54931_cast")];
            tensor<int32, [4]> k_487_perm_0 = const()[name = tensor<string, []>("k_487_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_54938_begin_0 = const()[name = tensor<string, []>("op_54938_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_54938_end_0 = const()[name = tensor<string, []>("op_54938_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_54938_end_mask_0 = const()[name = tensor<string, []>("op_54938_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_18 = transpose(perm = k_487_perm_0, x = k_485_cast)[name = tensor<string, []>("transpose_18")];
            tensor<fp16, [2, 77, 1, 64]> var_54938_cast = slice_by_index(begin = var_54938_begin_0, end = var_54938_end_0, end_mask = var_54938_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_54938_cast")];
            tensor<int32, [4]> var_54942_begin_0 = const()[name = tensor<string, []>("op_54942_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_54942_end_0 = const()[name = tensor<string, []>("op_54942_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_54942_end_mask_0 = const()[name = tensor<string, []>("op_54942_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54942_cast = slice_by_index(begin = var_54942_begin_0, end = var_54942_end_0, end_mask = var_54942_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_54942_cast")];
            tensor<int32, [4]> var_54946_begin_0 = const()[name = tensor<string, []>("op_54946_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_54946_end_0 = const()[name = tensor<string, []>("op_54946_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_54946_end_mask_0 = const()[name = tensor<string, []>("op_54946_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54946_cast = slice_by_index(begin = var_54946_begin_0, end = var_54946_end_0, end_mask = var_54946_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_54946_cast")];
            tensor<int32, [4]> var_54950_begin_0 = const()[name = tensor<string, []>("op_54950_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_54950_end_0 = const()[name = tensor<string, []>("op_54950_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_54950_end_mask_0 = const()[name = tensor<string, []>("op_54950_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54950_cast = slice_by_index(begin = var_54950_begin_0, end = var_54950_end_0, end_mask = var_54950_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_54950_cast")];
            tensor<int32, [4]> var_54954_begin_0 = const()[name = tensor<string, []>("op_54954_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_54954_end_0 = const()[name = tensor<string, []>("op_54954_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_54954_end_mask_0 = const()[name = tensor<string, []>("op_54954_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54954_cast = slice_by_index(begin = var_54954_begin_0, end = var_54954_end_0, end_mask = var_54954_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_54954_cast")];
            tensor<int32, [4]> var_54958_begin_0 = const()[name = tensor<string, []>("op_54958_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_54958_end_0 = const()[name = tensor<string, []>("op_54958_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_54958_end_mask_0 = const()[name = tensor<string, []>("op_54958_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54958_cast = slice_by_index(begin = var_54958_begin_0, end = var_54958_end_0, end_mask = var_54958_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_54958_cast")];
            tensor<int32, [4]> var_54962_begin_0 = const()[name = tensor<string, []>("op_54962_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_54962_end_0 = const()[name = tensor<string, []>("op_54962_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_54962_end_mask_0 = const()[name = tensor<string, []>("op_54962_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54962_cast = slice_by_index(begin = var_54962_begin_0, end = var_54962_end_0, end_mask = var_54962_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_54962_cast")];
            tensor<int32, [4]> var_54966_begin_0 = const()[name = tensor<string, []>("op_54966_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_54966_end_0 = const()[name = tensor<string, []>("op_54966_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_54966_end_mask_0 = const()[name = tensor<string, []>("op_54966_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54966_cast = slice_by_index(begin = var_54966_begin_0, end = var_54966_end_0, end_mask = var_54966_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_54966_cast")];
            tensor<int32, [4]> var_54970_begin_0 = const()[name = tensor<string, []>("op_54970_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_54970_end_0 = const()[name = tensor<string, []>("op_54970_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_54970_end_mask_0 = const()[name = tensor<string, []>("op_54970_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54970_cast = slice_by_index(begin = var_54970_begin_0, end = var_54970_end_0, end_mask = var_54970_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_54970_cast")];
            tensor<int32, [4]> var_54974_begin_0 = const()[name = tensor<string, []>("op_54974_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_54974_end_0 = const()[name = tensor<string, []>("op_54974_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_54974_end_mask_0 = const()[name = tensor<string, []>("op_54974_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54974_cast = slice_by_index(begin = var_54974_begin_0, end = var_54974_end_0, end_mask = var_54974_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_54974_cast")];
            tensor<int32, [4]> var_54978_begin_0 = const()[name = tensor<string, []>("op_54978_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_54978_end_0 = const()[name = tensor<string, []>("op_54978_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_54978_end_mask_0 = const()[name = tensor<string, []>("op_54978_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54978_cast = slice_by_index(begin = var_54978_begin_0, end = var_54978_end_0, end_mask = var_54978_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_54978_cast")];
            tensor<int32, [4]> var_54982_begin_0 = const()[name = tensor<string, []>("op_54982_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_54982_end_0 = const()[name = tensor<string, []>("op_54982_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_54982_end_mask_0 = const()[name = tensor<string, []>("op_54982_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54982_cast = slice_by_index(begin = var_54982_begin_0, end = var_54982_end_0, end_mask = var_54982_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_54982_cast")];
            tensor<int32, [4]> var_54986_begin_0 = const()[name = tensor<string, []>("op_54986_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_54986_end_0 = const()[name = tensor<string, []>("op_54986_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_54986_end_mask_0 = const()[name = tensor<string, []>("op_54986_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54986_cast = slice_by_index(begin = var_54986_begin_0, end = var_54986_end_0, end_mask = var_54986_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_54986_cast")];
            tensor<int32, [4]> var_54990_begin_0 = const()[name = tensor<string, []>("op_54990_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_54990_end_0 = const()[name = tensor<string, []>("op_54990_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_54990_end_mask_0 = const()[name = tensor<string, []>("op_54990_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54990_cast = slice_by_index(begin = var_54990_begin_0, end = var_54990_end_0, end_mask = var_54990_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_54990_cast")];
            tensor<int32, [4]> var_54994_begin_0 = const()[name = tensor<string, []>("op_54994_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_54994_end_0 = const()[name = tensor<string, []>("op_54994_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_54994_end_mask_0 = const()[name = tensor<string, []>("op_54994_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54994_cast = slice_by_index(begin = var_54994_begin_0, end = var_54994_end_0, end_mask = var_54994_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_54994_cast")];
            tensor<int32, [4]> var_54998_begin_0 = const()[name = tensor<string, []>("op_54998_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_54998_end_0 = const()[name = tensor<string, []>("op_54998_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_54998_end_mask_0 = const()[name = tensor<string, []>("op_54998_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_54998_cast = slice_by_index(begin = var_54998_begin_0, end = var_54998_end_0, end_mask = var_54998_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_54998_cast")];
            tensor<int32, [4]> var_55002_begin_0 = const()[name = tensor<string, []>("op_55002_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_55002_end_0 = const()[name = tensor<string, []>("op_55002_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_55002_end_mask_0 = const()[name = tensor<string, []>("op_55002_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55002_cast = slice_by_index(begin = var_55002_begin_0, end = var_55002_end_0, end_mask = var_55002_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_55002_cast")];
            tensor<int32, [4]> var_55006_begin_0 = const()[name = tensor<string, []>("op_55006_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_55006_end_0 = const()[name = tensor<string, []>("op_55006_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_55006_end_mask_0 = const()[name = tensor<string, []>("op_55006_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55006_cast = slice_by_index(begin = var_55006_begin_0, end = var_55006_end_0, end_mask = var_55006_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_55006_cast")];
            tensor<int32, [4]> var_55010_begin_0 = const()[name = tensor<string, []>("op_55010_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_55010_end_0 = const()[name = tensor<string, []>("op_55010_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_55010_end_mask_0 = const()[name = tensor<string, []>("op_55010_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55010_cast = slice_by_index(begin = var_55010_begin_0, end = var_55010_end_0, end_mask = var_55010_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_55010_cast")];
            tensor<int32, [4]> var_55014_begin_0 = const()[name = tensor<string, []>("op_55014_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_55014_end_0 = const()[name = tensor<string, []>("op_55014_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_55014_end_mask_0 = const()[name = tensor<string, []>("op_55014_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55014_cast = slice_by_index(begin = var_55014_begin_0, end = var_55014_end_0, end_mask = var_55014_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_55014_cast")];
            tensor<int32, [4]> var_55016_begin_0 = const()[name = tensor<string, []>("op_55016_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_55016_end_0 = const()[name = tensor<string, []>("op_55016_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_55016_end_mask_0 = const()[name = tensor<string, []>("op_55016_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55016_cast = slice_by_index(begin = var_55016_begin_0, end = var_55016_end_0, end_mask = var_55016_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55016_cast")];
            tensor<int32, [4]> var_55020_begin_0 = const()[name = tensor<string, []>("op_55020_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_55020_end_0 = const()[name = tensor<string, []>("op_55020_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_55020_end_mask_0 = const()[name = tensor<string, []>("op_55020_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55020_cast = slice_by_index(begin = var_55020_begin_0, end = var_55020_end_0, end_mask = var_55020_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55020_cast")];
            tensor<int32, [4]> var_55024_begin_0 = const()[name = tensor<string, []>("op_55024_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_55024_end_0 = const()[name = tensor<string, []>("op_55024_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_55024_end_mask_0 = const()[name = tensor<string, []>("op_55024_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55024_cast = slice_by_index(begin = var_55024_begin_0, end = var_55024_end_0, end_mask = var_55024_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55024_cast")];
            tensor<int32, [4]> var_55028_begin_0 = const()[name = tensor<string, []>("op_55028_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_55028_end_0 = const()[name = tensor<string, []>("op_55028_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_55028_end_mask_0 = const()[name = tensor<string, []>("op_55028_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55028_cast = slice_by_index(begin = var_55028_begin_0, end = var_55028_end_0, end_mask = var_55028_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55028_cast")];
            tensor<int32, [4]> var_55032_begin_0 = const()[name = tensor<string, []>("op_55032_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_55032_end_0 = const()[name = tensor<string, []>("op_55032_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_55032_end_mask_0 = const()[name = tensor<string, []>("op_55032_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55032_cast = slice_by_index(begin = var_55032_begin_0, end = var_55032_end_0, end_mask = var_55032_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55032_cast")];
            tensor<int32, [4]> var_55036_begin_0 = const()[name = tensor<string, []>("op_55036_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_55036_end_0 = const()[name = tensor<string, []>("op_55036_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_55036_end_mask_0 = const()[name = tensor<string, []>("op_55036_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55036_cast = slice_by_index(begin = var_55036_begin_0, end = var_55036_end_0, end_mask = var_55036_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55036_cast")];
            tensor<int32, [4]> var_55040_begin_0 = const()[name = tensor<string, []>("op_55040_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_55040_end_0 = const()[name = tensor<string, []>("op_55040_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_55040_end_mask_0 = const()[name = tensor<string, []>("op_55040_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55040_cast = slice_by_index(begin = var_55040_begin_0, end = var_55040_end_0, end_mask = var_55040_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55040_cast")];
            tensor<int32, [4]> var_55044_begin_0 = const()[name = tensor<string, []>("op_55044_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_55044_end_0 = const()[name = tensor<string, []>("op_55044_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_55044_end_mask_0 = const()[name = tensor<string, []>("op_55044_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55044_cast = slice_by_index(begin = var_55044_begin_0, end = var_55044_end_0, end_mask = var_55044_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55044_cast")];
            tensor<int32, [4]> var_55048_begin_0 = const()[name = tensor<string, []>("op_55048_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_55048_end_0 = const()[name = tensor<string, []>("op_55048_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_55048_end_mask_0 = const()[name = tensor<string, []>("op_55048_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55048_cast = slice_by_index(begin = var_55048_begin_0, end = var_55048_end_0, end_mask = var_55048_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55048_cast")];
            tensor<int32, [4]> var_55052_begin_0 = const()[name = tensor<string, []>("op_55052_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_55052_end_0 = const()[name = tensor<string, []>("op_55052_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_55052_end_mask_0 = const()[name = tensor<string, []>("op_55052_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55052_cast = slice_by_index(begin = var_55052_begin_0, end = var_55052_end_0, end_mask = var_55052_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55052_cast")];
            tensor<int32, [4]> var_55056_begin_0 = const()[name = tensor<string, []>("op_55056_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_55056_end_0 = const()[name = tensor<string, []>("op_55056_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_55056_end_mask_0 = const()[name = tensor<string, []>("op_55056_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55056_cast = slice_by_index(begin = var_55056_begin_0, end = var_55056_end_0, end_mask = var_55056_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55056_cast")];
            tensor<int32, [4]> var_55060_begin_0 = const()[name = tensor<string, []>("op_55060_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_55060_end_0 = const()[name = tensor<string, []>("op_55060_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_55060_end_mask_0 = const()[name = tensor<string, []>("op_55060_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55060_cast = slice_by_index(begin = var_55060_begin_0, end = var_55060_end_0, end_mask = var_55060_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55060_cast")];
            tensor<int32, [4]> var_55064_begin_0 = const()[name = tensor<string, []>("op_55064_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_55064_end_0 = const()[name = tensor<string, []>("op_55064_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_55064_end_mask_0 = const()[name = tensor<string, []>("op_55064_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55064_cast = slice_by_index(begin = var_55064_begin_0, end = var_55064_end_0, end_mask = var_55064_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55064_cast")];
            tensor<int32, [4]> var_55068_begin_0 = const()[name = tensor<string, []>("op_55068_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_55068_end_0 = const()[name = tensor<string, []>("op_55068_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_55068_end_mask_0 = const()[name = tensor<string, []>("op_55068_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55068_cast = slice_by_index(begin = var_55068_begin_0, end = var_55068_end_0, end_mask = var_55068_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55068_cast")];
            tensor<int32, [4]> var_55072_begin_0 = const()[name = tensor<string, []>("op_55072_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_55072_end_0 = const()[name = tensor<string, []>("op_55072_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_55072_end_mask_0 = const()[name = tensor<string, []>("op_55072_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55072_cast = slice_by_index(begin = var_55072_begin_0, end = var_55072_end_0, end_mask = var_55072_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55072_cast")];
            tensor<int32, [4]> var_55076_begin_0 = const()[name = tensor<string, []>("op_55076_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_55076_end_0 = const()[name = tensor<string, []>("op_55076_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_55076_end_mask_0 = const()[name = tensor<string, []>("op_55076_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55076_cast = slice_by_index(begin = var_55076_begin_0, end = var_55076_end_0, end_mask = var_55076_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55076_cast")];
            tensor<int32, [4]> var_55080_begin_0 = const()[name = tensor<string, []>("op_55080_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_55080_end_0 = const()[name = tensor<string, []>("op_55080_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_55080_end_mask_0 = const()[name = tensor<string, []>("op_55080_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55080_cast = slice_by_index(begin = var_55080_begin_0, end = var_55080_end_0, end_mask = var_55080_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55080_cast")];
            tensor<int32, [4]> var_55084_begin_0 = const()[name = tensor<string, []>("op_55084_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_55084_end_0 = const()[name = tensor<string, []>("op_55084_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_55084_end_mask_0 = const()[name = tensor<string, []>("op_55084_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55084_cast = slice_by_index(begin = var_55084_begin_0, end = var_55084_end_0, end_mask = var_55084_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55084_cast")];
            tensor<int32, [4]> var_55088_begin_0 = const()[name = tensor<string, []>("op_55088_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_55088_end_0 = const()[name = tensor<string, []>("op_55088_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_55088_end_mask_0 = const()[name = tensor<string, []>("op_55088_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55088_cast = slice_by_index(begin = var_55088_begin_0, end = var_55088_end_0, end_mask = var_55088_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55088_cast")];
            tensor<int32, [4]> var_55092_begin_0 = const()[name = tensor<string, []>("op_55092_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_55092_end_0 = const()[name = tensor<string, []>("op_55092_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_55092_end_mask_0 = const()[name = tensor<string, []>("op_55092_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55092_cast = slice_by_index(begin = var_55092_begin_0, end = var_55092_end_0, end_mask = var_55092_end_mask_0, x = v_243_cast)[name = tensor<string, []>("op_55092_cast")];
            tensor<string, []> var_55096_equation_0 = const()[name = tensor<string, []>("op_55096_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55096_cast = einsum(equation = var_55096_equation_0, values = (var_54938_cast, var_54855_cast))[name = tensor<string, []>("op_55096_cast")];
            tensor<fp16, []> var_55097_to_fp16 = const()[name = tensor<string, []>("op_55097_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4681_cast = mul(x = var_55096_cast, y = var_55097_to_fp16)[name = tensor<string, []>("aw_4681_cast")];
            tensor<string, []> var_55100_equation_0 = const()[name = tensor<string, []>("op_55100_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55100_cast = einsum(equation = var_55100_equation_0, values = (var_54942_cast, var_54859_cast))[name = tensor<string, []>("op_55100_cast")];
            tensor<fp16, []> var_55101_to_fp16 = const()[name = tensor<string, []>("op_55101_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4683_cast = mul(x = var_55100_cast, y = var_55101_to_fp16)[name = tensor<string, []>("aw_4683_cast")];
            tensor<string, []> var_55104_equation_0 = const()[name = tensor<string, []>("op_55104_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55104_cast = einsum(equation = var_55104_equation_0, values = (var_54946_cast, var_54863_cast))[name = tensor<string, []>("op_55104_cast")];
            tensor<fp16, []> var_55105_to_fp16 = const()[name = tensor<string, []>("op_55105_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4685_cast = mul(x = var_55104_cast, y = var_55105_to_fp16)[name = tensor<string, []>("aw_4685_cast")];
            tensor<string, []> var_55108_equation_0 = const()[name = tensor<string, []>("op_55108_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55108_cast = einsum(equation = var_55108_equation_0, values = (var_54950_cast, var_54867_cast))[name = tensor<string, []>("op_55108_cast")];
            tensor<fp16, []> var_55109_to_fp16 = const()[name = tensor<string, []>("op_55109_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4687_cast = mul(x = var_55108_cast, y = var_55109_to_fp16)[name = tensor<string, []>("aw_4687_cast")];
            tensor<string, []> var_55112_equation_0 = const()[name = tensor<string, []>("op_55112_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55112_cast = einsum(equation = var_55112_equation_0, values = (var_54954_cast, var_54871_cast))[name = tensor<string, []>("op_55112_cast")];
            tensor<fp16, []> var_55113_to_fp16 = const()[name = tensor<string, []>("op_55113_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4689_cast = mul(x = var_55112_cast, y = var_55113_to_fp16)[name = tensor<string, []>("aw_4689_cast")];
            tensor<string, []> var_55116_equation_0 = const()[name = tensor<string, []>("op_55116_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55116_cast = einsum(equation = var_55116_equation_0, values = (var_54958_cast, var_54875_cast))[name = tensor<string, []>("op_55116_cast")];
            tensor<fp16, []> var_55117_to_fp16 = const()[name = tensor<string, []>("op_55117_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4691_cast = mul(x = var_55116_cast, y = var_55117_to_fp16)[name = tensor<string, []>("aw_4691_cast")];
            tensor<string, []> var_55120_equation_0 = const()[name = tensor<string, []>("op_55120_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55120_cast = einsum(equation = var_55120_equation_0, values = (var_54962_cast, var_54879_cast))[name = tensor<string, []>("op_55120_cast")];
            tensor<fp16, []> var_55121_to_fp16 = const()[name = tensor<string, []>("op_55121_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4693_cast = mul(x = var_55120_cast, y = var_55121_to_fp16)[name = tensor<string, []>("aw_4693_cast")];
            tensor<string, []> var_55124_equation_0 = const()[name = tensor<string, []>("op_55124_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55124_cast = einsum(equation = var_55124_equation_0, values = (var_54966_cast, var_54883_cast))[name = tensor<string, []>("op_55124_cast")];
            tensor<fp16, []> var_55125_to_fp16 = const()[name = tensor<string, []>("op_55125_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4695_cast = mul(x = var_55124_cast, y = var_55125_to_fp16)[name = tensor<string, []>("aw_4695_cast")];
            tensor<string, []> var_55128_equation_0 = const()[name = tensor<string, []>("op_55128_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55128_cast = einsum(equation = var_55128_equation_0, values = (var_54970_cast, var_54887_cast))[name = tensor<string, []>("op_55128_cast")];
            tensor<fp16, []> var_55129_to_fp16 = const()[name = tensor<string, []>("op_55129_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4697_cast = mul(x = var_55128_cast, y = var_55129_to_fp16)[name = tensor<string, []>("aw_4697_cast")];
            tensor<string, []> var_55132_equation_0 = const()[name = tensor<string, []>("op_55132_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55132_cast = einsum(equation = var_55132_equation_0, values = (var_54974_cast, var_54891_cast))[name = tensor<string, []>("op_55132_cast")];
            tensor<fp16, []> var_55133_to_fp16 = const()[name = tensor<string, []>("op_55133_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4699_cast = mul(x = var_55132_cast, y = var_55133_to_fp16)[name = tensor<string, []>("aw_4699_cast")];
            tensor<string, []> var_55136_equation_0 = const()[name = tensor<string, []>("op_55136_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55136_cast = einsum(equation = var_55136_equation_0, values = (var_54978_cast, var_54895_cast))[name = tensor<string, []>("op_55136_cast")];
            tensor<fp16, []> var_55137_to_fp16 = const()[name = tensor<string, []>("op_55137_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4701_cast = mul(x = var_55136_cast, y = var_55137_to_fp16)[name = tensor<string, []>("aw_4701_cast")];
            tensor<string, []> var_55140_equation_0 = const()[name = tensor<string, []>("op_55140_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55140_cast = einsum(equation = var_55140_equation_0, values = (var_54982_cast, var_54899_cast))[name = tensor<string, []>("op_55140_cast")];
            tensor<fp16, []> var_55141_to_fp16 = const()[name = tensor<string, []>("op_55141_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4703_cast = mul(x = var_55140_cast, y = var_55141_to_fp16)[name = tensor<string, []>("aw_4703_cast")];
            tensor<string, []> var_55144_equation_0 = const()[name = tensor<string, []>("op_55144_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55144_cast = einsum(equation = var_55144_equation_0, values = (var_54986_cast, var_54903_cast))[name = tensor<string, []>("op_55144_cast")];
            tensor<fp16, []> var_55145_to_fp16 = const()[name = tensor<string, []>("op_55145_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4705_cast = mul(x = var_55144_cast, y = var_55145_to_fp16)[name = tensor<string, []>("aw_4705_cast")];
            tensor<string, []> var_55148_equation_0 = const()[name = tensor<string, []>("op_55148_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55148_cast = einsum(equation = var_55148_equation_0, values = (var_54990_cast, var_54907_cast))[name = tensor<string, []>("op_55148_cast")];
            tensor<fp16, []> var_55149_to_fp16 = const()[name = tensor<string, []>("op_55149_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4707_cast = mul(x = var_55148_cast, y = var_55149_to_fp16)[name = tensor<string, []>("aw_4707_cast")];
            tensor<string, []> var_55152_equation_0 = const()[name = tensor<string, []>("op_55152_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55152_cast = einsum(equation = var_55152_equation_0, values = (var_54994_cast, var_54911_cast))[name = tensor<string, []>("op_55152_cast")];
            tensor<fp16, []> var_55153_to_fp16 = const()[name = tensor<string, []>("op_55153_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4709_cast = mul(x = var_55152_cast, y = var_55153_to_fp16)[name = tensor<string, []>("aw_4709_cast")];
            tensor<string, []> var_55156_equation_0 = const()[name = tensor<string, []>("op_55156_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55156_cast = einsum(equation = var_55156_equation_0, values = (var_54998_cast, var_54915_cast))[name = tensor<string, []>("op_55156_cast")];
            tensor<fp16, []> var_55157_to_fp16 = const()[name = tensor<string, []>("op_55157_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4711_cast = mul(x = var_55156_cast, y = var_55157_to_fp16)[name = tensor<string, []>("aw_4711_cast")];
            tensor<string, []> var_55160_equation_0 = const()[name = tensor<string, []>("op_55160_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55160_cast = einsum(equation = var_55160_equation_0, values = (var_55002_cast, var_54919_cast))[name = tensor<string, []>("op_55160_cast")];
            tensor<fp16, []> var_55161_to_fp16 = const()[name = tensor<string, []>("op_55161_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4713_cast = mul(x = var_55160_cast, y = var_55161_to_fp16)[name = tensor<string, []>("aw_4713_cast")];
            tensor<string, []> var_55164_equation_0 = const()[name = tensor<string, []>("op_55164_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55164_cast = einsum(equation = var_55164_equation_0, values = (var_55006_cast, var_54923_cast))[name = tensor<string, []>("op_55164_cast")];
            tensor<fp16, []> var_55165_to_fp16 = const()[name = tensor<string, []>("op_55165_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4715_cast = mul(x = var_55164_cast, y = var_55165_to_fp16)[name = tensor<string, []>("aw_4715_cast")];
            tensor<string, []> var_55168_equation_0 = const()[name = tensor<string, []>("op_55168_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55168_cast = einsum(equation = var_55168_equation_0, values = (var_55010_cast, var_54927_cast))[name = tensor<string, []>("op_55168_cast")];
            tensor<fp16, []> var_55169_to_fp16 = const()[name = tensor<string, []>("op_55169_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4717_cast = mul(x = var_55168_cast, y = var_55169_to_fp16)[name = tensor<string, []>("aw_4717_cast")];
            tensor<string, []> var_55172_equation_0 = const()[name = tensor<string, []>("op_55172_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_55172_cast = einsum(equation = var_55172_equation_0, values = (var_55014_cast, var_54931_cast))[name = tensor<string, []>("op_55172_cast")];
            tensor<fp16, []> var_55173_to_fp16 = const()[name = tensor<string, []>("op_55173_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4719_cast = mul(x = var_55172_cast, y = var_55173_to_fp16)[name = tensor<string, []>("aw_4719_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55175_cast = softmax(axis = var_30385, x = aw_4681_cast)[name = tensor<string, []>("op_55175_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55176_cast = softmax(axis = var_30385, x = aw_4683_cast)[name = tensor<string, []>("op_55176_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55177_cast = softmax(axis = var_30385, x = aw_4685_cast)[name = tensor<string, []>("op_55177_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55178_cast = softmax(axis = var_30385, x = aw_4687_cast)[name = tensor<string, []>("op_55178_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55179_cast = softmax(axis = var_30385, x = aw_4689_cast)[name = tensor<string, []>("op_55179_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55180_cast = softmax(axis = var_30385, x = aw_4691_cast)[name = tensor<string, []>("op_55180_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55181_cast = softmax(axis = var_30385, x = aw_4693_cast)[name = tensor<string, []>("op_55181_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55182_cast = softmax(axis = var_30385, x = aw_4695_cast)[name = tensor<string, []>("op_55182_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55183_cast = softmax(axis = var_30385, x = aw_4697_cast)[name = tensor<string, []>("op_55183_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55184_cast = softmax(axis = var_30385, x = aw_4699_cast)[name = tensor<string, []>("op_55184_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55185_cast = softmax(axis = var_30385, x = aw_4701_cast)[name = tensor<string, []>("op_55185_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55186_cast = softmax(axis = var_30385, x = aw_4703_cast)[name = tensor<string, []>("op_55186_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55187_cast = softmax(axis = var_30385, x = aw_4705_cast)[name = tensor<string, []>("op_55187_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55188_cast = softmax(axis = var_30385, x = aw_4707_cast)[name = tensor<string, []>("op_55188_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55189_cast = softmax(axis = var_30385, x = aw_4709_cast)[name = tensor<string, []>("op_55189_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55190_cast = softmax(axis = var_30385, x = aw_4711_cast)[name = tensor<string, []>("op_55190_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55191_cast = softmax(axis = var_30385, x = aw_4713_cast)[name = tensor<string, []>("op_55191_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55192_cast = softmax(axis = var_30385, x = aw_4715_cast)[name = tensor<string, []>("op_55192_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55193_cast = softmax(axis = var_30385, x = aw_4717_cast)[name = tensor<string, []>("op_55193_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_55194_cast = softmax(axis = var_30385, x = aw_4719_cast)[name = tensor<string, []>("op_55194_cast")];
            tensor<string, []> var_55196_equation_0 = const()[name = tensor<string, []>("op_55196_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55196_cast = einsum(equation = var_55196_equation_0, values = (var_55016_cast, var_55175_cast))[name = tensor<string, []>("op_55196_cast")];
            tensor<string, []> var_55198_equation_0 = const()[name = tensor<string, []>("op_55198_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55198_cast = einsum(equation = var_55198_equation_0, values = (var_55020_cast, var_55176_cast))[name = tensor<string, []>("op_55198_cast")];
            tensor<string, []> var_55200_equation_0 = const()[name = tensor<string, []>("op_55200_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55200_cast = einsum(equation = var_55200_equation_0, values = (var_55024_cast, var_55177_cast))[name = tensor<string, []>("op_55200_cast")];
            tensor<string, []> var_55202_equation_0 = const()[name = tensor<string, []>("op_55202_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55202_cast = einsum(equation = var_55202_equation_0, values = (var_55028_cast, var_55178_cast))[name = tensor<string, []>("op_55202_cast")];
            tensor<string, []> var_55204_equation_0 = const()[name = tensor<string, []>("op_55204_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55204_cast = einsum(equation = var_55204_equation_0, values = (var_55032_cast, var_55179_cast))[name = tensor<string, []>("op_55204_cast")];
            tensor<string, []> var_55206_equation_0 = const()[name = tensor<string, []>("op_55206_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55206_cast = einsum(equation = var_55206_equation_0, values = (var_55036_cast, var_55180_cast))[name = tensor<string, []>("op_55206_cast")];
            tensor<string, []> var_55208_equation_0 = const()[name = tensor<string, []>("op_55208_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55208_cast = einsum(equation = var_55208_equation_0, values = (var_55040_cast, var_55181_cast))[name = tensor<string, []>("op_55208_cast")];
            tensor<string, []> var_55210_equation_0 = const()[name = tensor<string, []>("op_55210_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55210_cast = einsum(equation = var_55210_equation_0, values = (var_55044_cast, var_55182_cast))[name = tensor<string, []>("op_55210_cast")];
            tensor<string, []> var_55212_equation_0 = const()[name = tensor<string, []>("op_55212_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55212_cast = einsum(equation = var_55212_equation_0, values = (var_55048_cast, var_55183_cast))[name = tensor<string, []>("op_55212_cast")];
            tensor<string, []> var_55214_equation_0 = const()[name = tensor<string, []>("op_55214_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55214_cast = einsum(equation = var_55214_equation_0, values = (var_55052_cast, var_55184_cast))[name = tensor<string, []>("op_55214_cast")];
            tensor<string, []> var_55216_equation_0 = const()[name = tensor<string, []>("op_55216_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55216_cast = einsum(equation = var_55216_equation_0, values = (var_55056_cast, var_55185_cast))[name = tensor<string, []>("op_55216_cast")];
            tensor<string, []> var_55218_equation_0 = const()[name = tensor<string, []>("op_55218_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55218_cast = einsum(equation = var_55218_equation_0, values = (var_55060_cast, var_55186_cast))[name = tensor<string, []>("op_55218_cast")];
            tensor<string, []> var_55220_equation_0 = const()[name = tensor<string, []>("op_55220_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55220_cast = einsum(equation = var_55220_equation_0, values = (var_55064_cast, var_55187_cast))[name = tensor<string, []>("op_55220_cast")];
            tensor<string, []> var_55222_equation_0 = const()[name = tensor<string, []>("op_55222_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55222_cast = einsum(equation = var_55222_equation_0, values = (var_55068_cast, var_55188_cast))[name = tensor<string, []>("op_55222_cast")];
            tensor<string, []> var_55224_equation_0 = const()[name = tensor<string, []>("op_55224_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55224_cast = einsum(equation = var_55224_equation_0, values = (var_55072_cast, var_55189_cast))[name = tensor<string, []>("op_55224_cast")];
            tensor<string, []> var_55226_equation_0 = const()[name = tensor<string, []>("op_55226_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55226_cast = einsum(equation = var_55226_equation_0, values = (var_55076_cast, var_55190_cast))[name = tensor<string, []>("op_55226_cast")];
            tensor<string, []> var_55228_equation_0 = const()[name = tensor<string, []>("op_55228_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55228_cast = einsum(equation = var_55228_equation_0, values = (var_55080_cast, var_55191_cast))[name = tensor<string, []>("op_55228_cast")];
            tensor<string, []> var_55230_equation_0 = const()[name = tensor<string, []>("op_55230_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55230_cast = einsum(equation = var_55230_equation_0, values = (var_55084_cast, var_55192_cast))[name = tensor<string, []>("op_55230_cast")];
            tensor<string, []> var_55232_equation_0 = const()[name = tensor<string, []>("op_55232_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55232_cast = einsum(equation = var_55232_equation_0, values = (var_55088_cast, var_55193_cast))[name = tensor<string, []>("op_55232_cast")];
            tensor<string, []> var_55234_equation_0 = const()[name = tensor<string, []>("op_55234_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55234_cast = einsum(equation = var_55234_equation_0, values = (var_55092_cast, var_55194_cast))[name = tensor<string, []>("op_55234_cast")];
            tensor<bool, []> input_687_interleave_0 = const()[name = tensor<string, []>("input_687_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_687_cast = concat(axis = var_30385, interleave = input_687_interleave_0, values = (var_55196_cast, var_55198_cast, var_55200_cast, var_55202_cast, var_55204_cast, var_55206_cast, var_55208_cast, var_55210_cast, var_55212_cast, var_55214_cast, var_55216_cast, var_55218_cast, var_55220_cast, var_55222_cast, var_55224_cast, var_55226_cast, var_55228_cast, var_55230_cast, var_55232_cast, var_55234_cast))[name = tensor<string, []>("input_687_cast")];
            tensor<int32, [2]> var_55240 = const()[name = tensor<string, []>("op_55240"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_55242 = const()[name = tensor<string, []>("op_55242"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_55244_pad_type_0 = const()[name = tensor<string, []>("op_55244_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_55244_pad_0 = const()[name = tensor<string, []>("op_55244_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4616052096)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4619328960)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_55244_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_out_0_bias_to_fp16, dilations = var_55242, groups = var_30385, pad = var_55244_pad_0, pad_type = var_55244_pad_type_0, strides = var_55240, weight = up_blocks_0_attentions_2_transformer_blocks_6_attn2_to_out_0_weight_to_fp16, x = input_687_cast)[name = tensor<string, []>("op_55244_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_365_cast = add(x = var_55244_cast, y = inputs_363_cast)[name = tensor<string, []>("inputs_365_cast")];
            tensor<int32, [1]> var_55248 = const()[name = tensor<string, []>("op_55248"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_365_cast = reduce_mean(axes = var_55248, keep_dims = var_30380, x = inputs_365_cast)[name = tensor<string, []>("channels_mean_365_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_365_cast = sub(x = inputs_365_cast, y = channels_mean_365_cast)[name = tensor<string, []>("zero_mean_365_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_365_cast = mul(x = zero_mean_365_cast, y = zero_mean_365_cast)[name = tensor<string, []>("zero_mean_sq_365_cast")];
            tensor<int32, [1]> var_55252 = const()[name = tensor<string, []>("op_55252"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_55253_cast = reduce_mean(axes = var_55252, keep_dims = var_30380, x = zero_mean_sq_365_cast)[name = tensor<string, []>("op_55253_cast")];
            tensor<fp16, []> var_55254_to_fp16 = const()[name = tensor<string, []>("op_55254_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_55255_cast = add(x = var_55253_cast, y = var_55254_to_fp16)[name = tensor<string, []>("op_55255_cast")];
            tensor<fp16, []> denom_365_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_365_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_365_cast = rsqrt(epsilon = denom_365_epsilon_0_to_fp16, x = var_55255_cast)[name = tensor<string, []>("denom_365_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_365_cast = mul(x = zero_mean_365_cast, y = denom_365_cast)[name = tensor<string, []>("out_365_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_55259_to_fp16 = const()[name = tensor<string, []>("op_55259_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4619331584)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_55260_cast = add(x = out_365_cast, y = var_55259_to_fp16)[name = tensor<string, []>("op_55260_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_55262_to_fp16 = const()[name = tensor<string, []>("op_55262_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4619334208)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_689_cast = mul(x = var_55260_cast, y = var_55262_to_fp16)[name = tensor<string, []>("input_689_cast")];
            tensor<int32, [2]> var_55270 = const()[name = tensor<string, []>("op_55270"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_55272 = const()[name = tensor<string, []>("op_55272"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_55274_pad_type_0 = const()[name = tensor<string, []>("op_55274_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_55274_pad_0 = const()[name = tensor<string, []>("op_55274_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4619336832)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_2_transformer_blocks_6_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4645551296)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_55274_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_6_ff_net_0_proj_bias_to_fp16, dilations = var_55272, groups = var_30385, pad = var_55274_pad_0, pad_type = var_55274_pad_type_0, strides = var_55270, weight = up_blocks_0_attentions_2_transformer_blocks_6_ff_net_0_proj_weight_to_fp16, x = input_689_cast)[name = tensor<string, []>("op_55274_cast")];
            tensor<int32, [2]> var_55275_split_sizes_0 = const()[name = tensor<string, []>("op_55275_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_55275_axis_0 = const()[name = tensor<string, []>("op_55275_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_55275_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_55275_cast_1 = split(axis = var_55275_axis_0, split_sizes = var_55275_split_sizes_0, x = var_55274_cast)[name = tensor<string, []>("op_55275_cast")];
            tensor<string, []> var_55277_mode_0 = const()[name = tensor<string, []>("op_55277_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_55277_cast = gelu(mode = var_55277_mode_0, x = var_55275_cast_1)[name = tensor<string, []>("op_55277_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_691_cast = mul(x = var_55275_cast_0, y = var_55277_cast)[name = tensor<string, []>("input_691_cast")];
            tensor<int32, [2]> var_55281 = const()[name = tensor<string, []>("op_55281"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_55283 = const()[name = tensor<string, []>("op_55283"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_55285_pad_type_0 = const()[name = tensor<string, []>("op_55285_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_55285_pad_0 = const()[name = tensor<string, []>("op_55285_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_6_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4645571840)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_6_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_6_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4658679104)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_55285_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_6_ff_net_2_bias_to_fp16, dilations = var_55283, groups = var_30385, pad = var_55285_pad_0, pad_type = var_55285_pad_type_0, strides = var_55281, weight = up_blocks_0_attentions_2_transformer_blocks_6_ff_net_2_weight_to_fp16, x = input_691_cast)[name = tensor<string, []>("op_55285_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_367_cast = add(x = var_55285_cast, y = inputs_365_cast)[name = tensor<string, []>("inputs_367_cast")];
            tensor<int32, [1]> var_55295 = const()[name = tensor<string, []>("op_55295"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_367_cast = reduce_mean(axes = var_55295, keep_dims = var_30380, x = inputs_367_cast)[name = tensor<string, []>("channels_mean_367_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_367_cast = sub(x = inputs_367_cast, y = channels_mean_367_cast)[name = tensor<string, []>("zero_mean_367_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_367_cast = mul(x = zero_mean_367_cast, y = zero_mean_367_cast)[name = tensor<string, []>("zero_mean_sq_367_cast")];
            tensor<int32, [1]> var_55299 = const()[name = tensor<string, []>("op_55299"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_55300_cast = reduce_mean(axes = var_55299, keep_dims = var_30380, x = zero_mean_sq_367_cast)[name = tensor<string, []>("op_55300_cast")];
            tensor<fp16, []> var_55301_to_fp16 = const()[name = tensor<string, []>("op_55301_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_55302_cast = add(x = var_55300_cast, y = var_55301_to_fp16)[name = tensor<string, []>("op_55302_cast")];
            tensor<fp16, []> denom_367_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_367_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_367_cast = rsqrt(epsilon = denom_367_epsilon_0_to_fp16, x = var_55302_cast)[name = tensor<string, []>("denom_367_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_367_cast = mul(x = zero_mean_367_cast, y = denom_367_cast)[name = tensor<string, []>("out_367_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_55306_to_fp16 = const()[name = tensor<string, []>("op_55306_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4658681728)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_55307_cast = add(x = out_367_cast, y = var_55306_to_fp16)[name = tensor<string, []>("op_55307_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_55309_to_fp16 = const()[name = tensor<string, []>("op_55309_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4658684352)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_477_cast = mul(x = var_55307_cast, y = var_55309_to_fp16)[name = tensor<string, []>("hidden_states_477_cast")];
            tensor<int32, [2]> var_55316 = const()[name = tensor<string, []>("op_55316"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_55318 = const()[name = tensor<string, []>("op_55318"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_245_pad_type_0 = const()[name = tensor<string, []>("q_245_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_245_pad_0 = const()[name = tensor<string, []>("q_245_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4658686976)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_245_cast = conv(dilations = var_55318, groups = var_30385, pad = q_245_pad_0, pad_type = q_245_pad_type_0, strides = var_55316, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_q_weight_to_fp16, x = hidden_states_477_cast)[name = tensor<string, []>("q_245_cast")];
            tensor<int32, [2]> var_55322 = const()[name = tensor<string, []>("op_55322"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_55324 = const()[name = tensor<string, []>("op_55324"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_489_pad_type_0 = const()[name = tensor<string, []>("k_489_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_489_pad_0 = const()[name = tensor<string, []>("k_489_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4661963840)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_489_cast = conv(dilations = var_55324, groups = var_30385, pad = k_489_pad_0, pad_type = k_489_pad_type_0, strides = var_55322, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_k_weight_to_fp16, x = hidden_states_477_cast)[name = tensor<string, []>("k_489_cast")];
            tensor<int32, [2]> var_55328 = const()[name = tensor<string, []>("op_55328"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_55330 = const()[name = tensor<string, []>("op_55330"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_245_pad_type_0 = const()[name = tensor<string, []>("v_245_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_245_pad_0 = const()[name = tensor<string, []>("v_245_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4665240704)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_245_cast = conv(dilations = var_55330, groups = var_30385, pad = v_245_pad_0, pad_type = v_245_pad_type_0, strides = var_55328, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_v_weight_to_fp16, x = hidden_states_477_cast)[name = tensor<string, []>("v_245_cast")];
            tensor<int32, [4]> var_55334_begin_0 = const()[name = tensor<string, []>("op_55334_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_55334_end_0 = const()[name = tensor<string, []>("op_55334_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_55334_end_mask_0 = const()[name = tensor<string, []>("op_55334_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55334_cast = slice_by_index(begin = var_55334_begin_0, end = var_55334_end_0, end_mask = var_55334_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55334_cast")];
            tensor<int32, [4]> var_55338_begin_0 = const()[name = tensor<string, []>("op_55338_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_55338_end_0 = const()[name = tensor<string, []>("op_55338_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_55338_end_mask_0 = const()[name = tensor<string, []>("op_55338_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55338_cast = slice_by_index(begin = var_55338_begin_0, end = var_55338_end_0, end_mask = var_55338_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55338_cast")];
            tensor<int32, [4]> var_55342_begin_0 = const()[name = tensor<string, []>("op_55342_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_55342_end_0 = const()[name = tensor<string, []>("op_55342_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_55342_end_mask_0 = const()[name = tensor<string, []>("op_55342_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55342_cast = slice_by_index(begin = var_55342_begin_0, end = var_55342_end_0, end_mask = var_55342_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55342_cast")];
            tensor<int32, [4]> var_55346_begin_0 = const()[name = tensor<string, []>("op_55346_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_55346_end_0 = const()[name = tensor<string, []>("op_55346_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_55346_end_mask_0 = const()[name = tensor<string, []>("op_55346_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55346_cast = slice_by_index(begin = var_55346_begin_0, end = var_55346_end_0, end_mask = var_55346_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55346_cast")];
            tensor<int32, [4]> var_55350_begin_0 = const()[name = tensor<string, []>("op_55350_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_55350_end_0 = const()[name = tensor<string, []>("op_55350_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_55350_end_mask_0 = const()[name = tensor<string, []>("op_55350_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55350_cast = slice_by_index(begin = var_55350_begin_0, end = var_55350_end_0, end_mask = var_55350_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55350_cast")];
            tensor<int32, [4]> var_55354_begin_0 = const()[name = tensor<string, []>("op_55354_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_55354_end_0 = const()[name = tensor<string, []>("op_55354_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_55354_end_mask_0 = const()[name = tensor<string, []>("op_55354_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55354_cast = slice_by_index(begin = var_55354_begin_0, end = var_55354_end_0, end_mask = var_55354_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55354_cast")];
            tensor<int32, [4]> var_55358_begin_0 = const()[name = tensor<string, []>("op_55358_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_55358_end_0 = const()[name = tensor<string, []>("op_55358_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_55358_end_mask_0 = const()[name = tensor<string, []>("op_55358_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55358_cast = slice_by_index(begin = var_55358_begin_0, end = var_55358_end_0, end_mask = var_55358_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55358_cast")];
            tensor<int32, [4]> var_55362_begin_0 = const()[name = tensor<string, []>("op_55362_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_55362_end_0 = const()[name = tensor<string, []>("op_55362_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_55362_end_mask_0 = const()[name = tensor<string, []>("op_55362_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55362_cast = slice_by_index(begin = var_55362_begin_0, end = var_55362_end_0, end_mask = var_55362_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55362_cast")];
            tensor<int32, [4]> var_55366_begin_0 = const()[name = tensor<string, []>("op_55366_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_55366_end_0 = const()[name = tensor<string, []>("op_55366_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_55366_end_mask_0 = const()[name = tensor<string, []>("op_55366_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55366_cast = slice_by_index(begin = var_55366_begin_0, end = var_55366_end_0, end_mask = var_55366_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55366_cast")];
            tensor<int32, [4]> var_55370_begin_0 = const()[name = tensor<string, []>("op_55370_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_55370_end_0 = const()[name = tensor<string, []>("op_55370_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_55370_end_mask_0 = const()[name = tensor<string, []>("op_55370_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55370_cast = slice_by_index(begin = var_55370_begin_0, end = var_55370_end_0, end_mask = var_55370_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55370_cast")];
            tensor<int32, [4]> var_55374_begin_0 = const()[name = tensor<string, []>("op_55374_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_55374_end_0 = const()[name = tensor<string, []>("op_55374_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_55374_end_mask_0 = const()[name = tensor<string, []>("op_55374_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55374_cast = slice_by_index(begin = var_55374_begin_0, end = var_55374_end_0, end_mask = var_55374_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55374_cast")];
            tensor<int32, [4]> var_55378_begin_0 = const()[name = tensor<string, []>("op_55378_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_55378_end_0 = const()[name = tensor<string, []>("op_55378_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_55378_end_mask_0 = const()[name = tensor<string, []>("op_55378_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55378_cast = slice_by_index(begin = var_55378_begin_0, end = var_55378_end_0, end_mask = var_55378_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55378_cast")];
            tensor<int32, [4]> var_55382_begin_0 = const()[name = tensor<string, []>("op_55382_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_55382_end_0 = const()[name = tensor<string, []>("op_55382_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_55382_end_mask_0 = const()[name = tensor<string, []>("op_55382_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55382_cast = slice_by_index(begin = var_55382_begin_0, end = var_55382_end_0, end_mask = var_55382_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55382_cast")];
            tensor<int32, [4]> var_55386_begin_0 = const()[name = tensor<string, []>("op_55386_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_55386_end_0 = const()[name = tensor<string, []>("op_55386_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_55386_end_mask_0 = const()[name = tensor<string, []>("op_55386_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55386_cast = slice_by_index(begin = var_55386_begin_0, end = var_55386_end_0, end_mask = var_55386_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55386_cast")];
            tensor<int32, [4]> var_55390_begin_0 = const()[name = tensor<string, []>("op_55390_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_55390_end_0 = const()[name = tensor<string, []>("op_55390_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_55390_end_mask_0 = const()[name = tensor<string, []>("op_55390_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55390_cast = slice_by_index(begin = var_55390_begin_0, end = var_55390_end_0, end_mask = var_55390_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55390_cast")];
            tensor<int32, [4]> var_55394_begin_0 = const()[name = tensor<string, []>("op_55394_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_55394_end_0 = const()[name = tensor<string, []>("op_55394_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_55394_end_mask_0 = const()[name = tensor<string, []>("op_55394_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55394_cast = slice_by_index(begin = var_55394_begin_0, end = var_55394_end_0, end_mask = var_55394_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55394_cast")];
            tensor<int32, [4]> var_55398_begin_0 = const()[name = tensor<string, []>("op_55398_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_55398_end_0 = const()[name = tensor<string, []>("op_55398_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_55398_end_mask_0 = const()[name = tensor<string, []>("op_55398_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55398_cast = slice_by_index(begin = var_55398_begin_0, end = var_55398_end_0, end_mask = var_55398_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55398_cast")];
            tensor<int32, [4]> var_55402_begin_0 = const()[name = tensor<string, []>("op_55402_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_55402_end_0 = const()[name = tensor<string, []>("op_55402_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_55402_end_mask_0 = const()[name = tensor<string, []>("op_55402_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55402_cast = slice_by_index(begin = var_55402_begin_0, end = var_55402_end_0, end_mask = var_55402_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55402_cast")];
            tensor<int32, [4]> var_55406_begin_0 = const()[name = tensor<string, []>("op_55406_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_55406_end_0 = const()[name = tensor<string, []>("op_55406_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_55406_end_mask_0 = const()[name = tensor<string, []>("op_55406_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55406_cast = slice_by_index(begin = var_55406_begin_0, end = var_55406_end_0, end_mask = var_55406_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55406_cast")];
            tensor<int32, [4]> var_55410_begin_0 = const()[name = tensor<string, []>("op_55410_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_55410_end_0 = const()[name = tensor<string, []>("op_55410_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_55410_end_mask_0 = const()[name = tensor<string, []>("op_55410_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55410_cast = slice_by_index(begin = var_55410_begin_0, end = var_55410_end_0, end_mask = var_55410_end_mask_0, x = q_245_cast)[name = tensor<string, []>("op_55410_cast")];
            tensor<int32, [4]> k_491_perm_0 = const()[name = tensor<string, []>("k_491_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_55417_begin_0 = const()[name = tensor<string, []>("op_55417_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_55417_end_0 = const()[name = tensor<string, []>("op_55417_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_55417_end_mask_0 = const()[name = tensor<string, []>("op_55417_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_17 = transpose(perm = k_491_perm_0, x = k_489_cast)[name = tensor<string, []>("transpose_17")];
            tensor<fp16, [2, 1024, 1, 64]> var_55417_cast = slice_by_index(begin = var_55417_begin_0, end = var_55417_end_0, end_mask = var_55417_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55417_cast")];
            tensor<int32, [4]> var_55421_begin_0 = const()[name = tensor<string, []>("op_55421_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_55421_end_0 = const()[name = tensor<string, []>("op_55421_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_55421_end_mask_0 = const()[name = tensor<string, []>("op_55421_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55421_cast = slice_by_index(begin = var_55421_begin_0, end = var_55421_end_0, end_mask = var_55421_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55421_cast")];
            tensor<int32, [4]> var_55425_begin_0 = const()[name = tensor<string, []>("op_55425_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_55425_end_0 = const()[name = tensor<string, []>("op_55425_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_55425_end_mask_0 = const()[name = tensor<string, []>("op_55425_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55425_cast = slice_by_index(begin = var_55425_begin_0, end = var_55425_end_0, end_mask = var_55425_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55425_cast")];
            tensor<int32, [4]> var_55429_begin_0 = const()[name = tensor<string, []>("op_55429_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_55429_end_0 = const()[name = tensor<string, []>("op_55429_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_55429_end_mask_0 = const()[name = tensor<string, []>("op_55429_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55429_cast = slice_by_index(begin = var_55429_begin_0, end = var_55429_end_0, end_mask = var_55429_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55429_cast")];
            tensor<int32, [4]> var_55433_begin_0 = const()[name = tensor<string, []>("op_55433_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_55433_end_0 = const()[name = tensor<string, []>("op_55433_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_55433_end_mask_0 = const()[name = tensor<string, []>("op_55433_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55433_cast = slice_by_index(begin = var_55433_begin_0, end = var_55433_end_0, end_mask = var_55433_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55433_cast")];
            tensor<int32, [4]> var_55437_begin_0 = const()[name = tensor<string, []>("op_55437_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_55437_end_0 = const()[name = tensor<string, []>("op_55437_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_55437_end_mask_0 = const()[name = tensor<string, []>("op_55437_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55437_cast = slice_by_index(begin = var_55437_begin_0, end = var_55437_end_0, end_mask = var_55437_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55437_cast")];
            tensor<int32, [4]> var_55441_begin_0 = const()[name = tensor<string, []>("op_55441_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_55441_end_0 = const()[name = tensor<string, []>("op_55441_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_55441_end_mask_0 = const()[name = tensor<string, []>("op_55441_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55441_cast = slice_by_index(begin = var_55441_begin_0, end = var_55441_end_0, end_mask = var_55441_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55441_cast")];
            tensor<int32, [4]> var_55445_begin_0 = const()[name = tensor<string, []>("op_55445_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_55445_end_0 = const()[name = tensor<string, []>("op_55445_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_55445_end_mask_0 = const()[name = tensor<string, []>("op_55445_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55445_cast = slice_by_index(begin = var_55445_begin_0, end = var_55445_end_0, end_mask = var_55445_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55445_cast")];
            tensor<int32, [4]> var_55449_begin_0 = const()[name = tensor<string, []>("op_55449_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_55449_end_0 = const()[name = tensor<string, []>("op_55449_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_55449_end_mask_0 = const()[name = tensor<string, []>("op_55449_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55449_cast = slice_by_index(begin = var_55449_begin_0, end = var_55449_end_0, end_mask = var_55449_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55449_cast")];
            tensor<int32, [4]> var_55453_begin_0 = const()[name = tensor<string, []>("op_55453_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_55453_end_0 = const()[name = tensor<string, []>("op_55453_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_55453_end_mask_0 = const()[name = tensor<string, []>("op_55453_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55453_cast = slice_by_index(begin = var_55453_begin_0, end = var_55453_end_0, end_mask = var_55453_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55453_cast")];
            tensor<int32, [4]> var_55457_begin_0 = const()[name = tensor<string, []>("op_55457_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_55457_end_0 = const()[name = tensor<string, []>("op_55457_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_55457_end_mask_0 = const()[name = tensor<string, []>("op_55457_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55457_cast = slice_by_index(begin = var_55457_begin_0, end = var_55457_end_0, end_mask = var_55457_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55457_cast")];
            tensor<int32, [4]> var_55461_begin_0 = const()[name = tensor<string, []>("op_55461_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_55461_end_0 = const()[name = tensor<string, []>("op_55461_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_55461_end_mask_0 = const()[name = tensor<string, []>("op_55461_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55461_cast = slice_by_index(begin = var_55461_begin_0, end = var_55461_end_0, end_mask = var_55461_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55461_cast")];
            tensor<int32, [4]> var_55465_begin_0 = const()[name = tensor<string, []>("op_55465_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_55465_end_0 = const()[name = tensor<string, []>("op_55465_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_55465_end_mask_0 = const()[name = tensor<string, []>("op_55465_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55465_cast = slice_by_index(begin = var_55465_begin_0, end = var_55465_end_0, end_mask = var_55465_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55465_cast")];
            tensor<int32, [4]> var_55469_begin_0 = const()[name = tensor<string, []>("op_55469_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_55469_end_0 = const()[name = tensor<string, []>("op_55469_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_55469_end_mask_0 = const()[name = tensor<string, []>("op_55469_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55469_cast = slice_by_index(begin = var_55469_begin_0, end = var_55469_end_0, end_mask = var_55469_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55469_cast")];
            tensor<int32, [4]> var_55473_begin_0 = const()[name = tensor<string, []>("op_55473_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_55473_end_0 = const()[name = tensor<string, []>("op_55473_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_55473_end_mask_0 = const()[name = tensor<string, []>("op_55473_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55473_cast = slice_by_index(begin = var_55473_begin_0, end = var_55473_end_0, end_mask = var_55473_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55473_cast")];
            tensor<int32, [4]> var_55477_begin_0 = const()[name = tensor<string, []>("op_55477_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_55477_end_0 = const()[name = tensor<string, []>("op_55477_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_55477_end_mask_0 = const()[name = tensor<string, []>("op_55477_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55477_cast = slice_by_index(begin = var_55477_begin_0, end = var_55477_end_0, end_mask = var_55477_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55477_cast")];
            tensor<int32, [4]> var_55481_begin_0 = const()[name = tensor<string, []>("op_55481_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_55481_end_0 = const()[name = tensor<string, []>("op_55481_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_55481_end_mask_0 = const()[name = tensor<string, []>("op_55481_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55481_cast = slice_by_index(begin = var_55481_begin_0, end = var_55481_end_0, end_mask = var_55481_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55481_cast")];
            tensor<int32, [4]> var_55485_begin_0 = const()[name = tensor<string, []>("op_55485_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_55485_end_0 = const()[name = tensor<string, []>("op_55485_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_55485_end_mask_0 = const()[name = tensor<string, []>("op_55485_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55485_cast = slice_by_index(begin = var_55485_begin_0, end = var_55485_end_0, end_mask = var_55485_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55485_cast")];
            tensor<int32, [4]> var_55489_begin_0 = const()[name = tensor<string, []>("op_55489_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_55489_end_0 = const()[name = tensor<string, []>("op_55489_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_55489_end_mask_0 = const()[name = tensor<string, []>("op_55489_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55489_cast = slice_by_index(begin = var_55489_begin_0, end = var_55489_end_0, end_mask = var_55489_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55489_cast")];
            tensor<int32, [4]> var_55493_begin_0 = const()[name = tensor<string, []>("op_55493_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_55493_end_0 = const()[name = tensor<string, []>("op_55493_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_55493_end_mask_0 = const()[name = tensor<string, []>("op_55493_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_55493_cast = slice_by_index(begin = var_55493_begin_0, end = var_55493_end_0, end_mask = var_55493_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_55493_cast")];
            tensor<int32, [4]> var_55495_begin_0 = const()[name = tensor<string, []>("op_55495_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_55495_end_0 = const()[name = tensor<string, []>("op_55495_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_55495_end_mask_0 = const()[name = tensor<string, []>("op_55495_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55495_cast = slice_by_index(begin = var_55495_begin_0, end = var_55495_end_0, end_mask = var_55495_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55495_cast")];
            tensor<int32, [4]> var_55499_begin_0 = const()[name = tensor<string, []>("op_55499_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_55499_end_0 = const()[name = tensor<string, []>("op_55499_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_55499_end_mask_0 = const()[name = tensor<string, []>("op_55499_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55499_cast = slice_by_index(begin = var_55499_begin_0, end = var_55499_end_0, end_mask = var_55499_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55499_cast")];
            tensor<int32, [4]> var_55503_begin_0 = const()[name = tensor<string, []>("op_55503_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_55503_end_0 = const()[name = tensor<string, []>("op_55503_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_55503_end_mask_0 = const()[name = tensor<string, []>("op_55503_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55503_cast = slice_by_index(begin = var_55503_begin_0, end = var_55503_end_0, end_mask = var_55503_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55503_cast")];
            tensor<int32, [4]> var_55507_begin_0 = const()[name = tensor<string, []>("op_55507_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_55507_end_0 = const()[name = tensor<string, []>("op_55507_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_55507_end_mask_0 = const()[name = tensor<string, []>("op_55507_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55507_cast = slice_by_index(begin = var_55507_begin_0, end = var_55507_end_0, end_mask = var_55507_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55507_cast")];
            tensor<int32, [4]> var_55511_begin_0 = const()[name = tensor<string, []>("op_55511_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_55511_end_0 = const()[name = tensor<string, []>("op_55511_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_55511_end_mask_0 = const()[name = tensor<string, []>("op_55511_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55511_cast = slice_by_index(begin = var_55511_begin_0, end = var_55511_end_0, end_mask = var_55511_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55511_cast")];
            tensor<int32, [4]> var_55515_begin_0 = const()[name = tensor<string, []>("op_55515_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_55515_end_0 = const()[name = tensor<string, []>("op_55515_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_55515_end_mask_0 = const()[name = tensor<string, []>("op_55515_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55515_cast = slice_by_index(begin = var_55515_begin_0, end = var_55515_end_0, end_mask = var_55515_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55515_cast")];
            tensor<int32, [4]> var_55519_begin_0 = const()[name = tensor<string, []>("op_55519_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_55519_end_0 = const()[name = tensor<string, []>("op_55519_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_55519_end_mask_0 = const()[name = tensor<string, []>("op_55519_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55519_cast = slice_by_index(begin = var_55519_begin_0, end = var_55519_end_0, end_mask = var_55519_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55519_cast")];
            tensor<int32, [4]> var_55523_begin_0 = const()[name = tensor<string, []>("op_55523_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_55523_end_0 = const()[name = tensor<string, []>("op_55523_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_55523_end_mask_0 = const()[name = tensor<string, []>("op_55523_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55523_cast = slice_by_index(begin = var_55523_begin_0, end = var_55523_end_0, end_mask = var_55523_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55523_cast")];
            tensor<int32, [4]> var_55527_begin_0 = const()[name = tensor<string, []>("op_55527_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_55527_end_0 = const()[name = tensor<string, []>("op_55527_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_55527_end_mask_0 = const()[name = tensor<string, []>("op_55527_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55527_cast = slice_by_index(begin = var_55527_begin_0, end = var_55527_end_0, end_mask = var_55527_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55527_cast")];
            tensor<int32, [4]> var_55531_begin_0 = const()[name = tensor<string, []>("op_55531_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_55531_end_0 = const()[name = tensor<string, []>("op_55531_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_55531_end_mask_0 = const()[name = tensor<string, []>("op_55531_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55531_cast = slice_by_index(begin = var_55531_begin_0, end = var_55531_end_0, end_mask = var_55531_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55531_cast")];
            tensor<int32, [4]> var_55535_begin_0 = const()[name = tensor<string, []>("op_55535_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_55535_end_0 = const()[name = tensor<string, []>("op_55535_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_55535_end_mask_0 = const()[name = tensor<string, []>("op_55535_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55535_cast = slice_by_index(begin = var_55535_begin_0, end = var_55535_end_0, end_mask = var_55535_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55535_cast")];
            tensor<int32, [4]> var_55539_begin_0 = const()[name = tensor<string, []>("op_55539_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_55539_end_0 = const()[name = tensor<string, []>("op_55539_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_55539_end_mask_0 = const()[name = tensor<string, []>("op_55539_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55539_cast = slice_by_index(begin = var_55539_begin_0, end = var_55539_end_0, end_mask = var_55539_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55539_cast")];
            tensor<int32, [4]> var_55543_begin_0 = const()[name = tensor<string, []>("op_55543_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_55543_end_0 = const()[name = tensor<string, []>("op_55543_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_55543_end_mask_0 = const()[name = tensor<string, []>("op_55543_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55543_cast = slice_by_index(begin = var_55543_begin_0, end = var_55543_end_0, end_mask = var_55543_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55543_cast")];
            tensor<int32, [4]> var_55547_begin_0 = const()[name = tensor<string, []>("op_55547_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_55547_end_0 = const()[name = tensor<string, []>("op_55547_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_55547_end_mask_0 = const()[name = tensor<string, []>("op_55547_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55547_cast = slice_by_index(begin = var_55547_begin_0, end = var_55547_end_0, end_mask = var_55547_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55547_cast")];
            tensor<int32, [4]> var_55551_begin_0 = const()[name = tensor<string, []>("op_55551_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_55551_end_0 = const()[name = tensor<string, []>("op_55551_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_55551_end_mask_0 = const()[name = tensor<string, []>("op_55551_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55551_cast = slice_by_index(begin = var_55551_begin_0, end = var_55551_end_0, end_mask = var_55551_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55551_cast")];
            tensor<int32, [4]> var_55555_begin_0 = const()[name = tensor<string, []>("op_55555_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_55555_end_0 = const()[name = tensor<string, []>("op_55555_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_55555_end_mask_0 = const()[name = tensor<string, []>("op_55555_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55555_cast = slice_by_index(begin = var_55555_begin_0, end = var_55555_end_0, end_mask = var_55555_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55555_cast")];
            tensor<int32, [4]> var_55559_begin_0 = const()[name = tensor<string, []>("op_55559_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_55559_end_0 = const()[name = tensor<string, []>("op_55559_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_55559_end_mask_0 = const()[name = tensor<string, []>("op_55559_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55559_cast = slice_by_index(begin = var_55559_begin_0, end = var_55559_end_0, end_mask = var_55559_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55559_cast")];
            tensor<int32, [4]> var_55563_begin_0 = const()[name = tensor<string, []>("op_55563_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_55563_end_0 = const()[name = tensor<string, []>("op_55563_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_55563_end_mask_0 = const()[name = tensor<string, []>("op_55563_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55563_cast = slice_by_index(begin = var_55563_begin_0, end = var_55563_end_0, end_mask = var_55563_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55563_cast")];
            tensor<int32, [4]> var_55567_begin_0 = const()[name = tensor<string, []>("op_55567_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_55567_end_0 = const()[name = tensor<string, []>("op_55567_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_55567_end_mask_0 = const()[name = tensor<string, []>("op_55567_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55567_cast = slice_by_index(begin = var_55567_begin_0, end = var_55567_end_0, end_mask = var_55567_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55567_cast")];
            tensor<int32, [4]> var_55571_begin_0 = const()[name = tensor<string, []>("op_55571_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_55571_end_0 = const()[name = tensor<string, []>("op_55571_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_55571_end_mask_0 = const()[name = tensor<string, []>("op_55571_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55571_cast = slice_by_index(begin = var_55571_begin_0, end = var_55571_end_0, end_mask = var_55571_end_mask_0, x = v_245_cast)[name = tensor<string, []>("op_55571_cast")];
            tensor<string, []> var_55575_equation_0 = const()[name = tensor<string, []>("op_55575_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55575_cast = einsum(equation = var_55575_equation_0, values = (var_55417_cast, var_55334_cast))[name = tensor<string, []>("op_55575_cast")];
            tensor<fp16, []> var_55576_to_fp16 = const()[name = tensor<string, []>("op_55576_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4721_cast = mul(x = var_55575_cast, y = var_55576_to_fp16)[name = tensor<string, []>("aw_4721_cast")];
            tensor<string, []> var_55579_equation_0 = const()[name = tensor<string, []>("op_55579_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55579_cast = einsum(equation = var_55579_equation_0, values = (var_55421_cast, var_55338_cast))[name = tensor<string, []>("op_55579_cast")];
            tensor<fp16, []> var_55580_to_fp16 = const()[name = tensor<string, []>("op_55580_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4723_cast = mul(x = var_55579_cast, y = var_55580_to_fp16)[name = tensor<string, []>("aw_4723_cast")];
            tensor<string, []> var_55583_equation_0 = const()[name = tensor<string, []>("op_55583_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55583_cast = einsum(equation = var_55583_equation_0, values = (var_55425_cast, var_55342_cast))[name = tensor<string, []>("op_55583_cast")];
            tensor<fp16, []> var_55584_to_fp16 = const()[name = tensor<string, []>("op_55584_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4725_cast = mul(x = var_55583_cast, y = var_55584_to_fp16)[name = tensor<string, []>("aw_4725_cast")];
            tensor<string, []> var_55587_equation_0 = const()[name = tensor<string, []>("op_55587_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55587_cast = einsum(equation = var_55587_equation_0, values = (var_55429_cast, var_55346_cast))[name = tensor<string, []>("op_55587_cast")];
            tensor<fp16, []> var_55588_to_fp16 = const()[name = tensor<string, []>("op_55588_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4727_cast = mul(x = var_55587_cast, y = var_55588_to_fp16)[name = tensor<string, []>("aw_4727_cast")];
            tensor<string, []> var_55591_equation_0 = const()[name = tensor<string, []>("op_55591_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55591_cast = einsum(equation = var_55591_equation_0, values = (var_55433_cast, var_55350_cast))[name = tensor<string, []>("op_55591_cast")];
            tensor<fp16, []> var_55592_to_fp16 = const()[name = tensor<string, []>("op_55592_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4729_cast = mul(x = var_55591_cast, y = var_55592_to_fp16)[name = tensor<string, []>("aw_4729_cast")];
            tensor<string, []> var_55595_equation_0 = const()[name = tensor<string, []>("op_55595_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55595_cast = einsum(equation = var_55595_equation_0, values = (var_55437_cast, var_55354_cast))[name = tensor<string, []>("op_55595_cast")];
            tensor<fp16, []> var_55596_to_fp16 = const()[name = tensor<string, []>("op_55596_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4731_cast = mul(x = var_55595_cast, y = var_55596_to_fp16)[name = tensor<string, []>("aw_4731_cast")];
            tensor<string, []> var_55599_equation_0 = const()[name = tensor<string, []>("op_55599_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55599_cast = einsum(equation = var_55599_equation_0, values = (var_55441_cast, var_55358_cast))[name = tensor<string, []>("op_55599_cast")];
            tensor<fp16, []> var_55600_to_fp16 = const()[name = tensor<string, []>("op_55600_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4733_cast = mul(x = var_55599_cast, y = var_55600_to_fp16)[name = tensor<string, []>("aw_4733_cast")];
            tensor<string, []> var_55603_equation_0 = const()[name = tensor<string, []>("op_55603_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55603_cast = einsum(equation = var_55603_equation_0, values = (var_55445_cast, var_55362_cast))[name = tensor<string, []>("op_55603_cast")];
            tensor<fp16, []> var_55604_to_fp16 = const()[name = tensor<string, []>("op_55604_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4735_cast = mul(x = var_55603_cast, y = var_55604_to_fp16)[name = tensor<string, []>("aw_4735_cast")];
            tensor<string, []> var_55607_equation_0 = const()[name = tensor<string, []>("op_55607_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55607_cast = einsum(equation = var_55607_equation_0, values = (var_55449_cast, var_55366_cast))[name = tensor<string, []>("op_55607_cast")];
            tensor<fp16, []> var_55608_to_fp16 = const()[name = tensor<string, []>("op_55608_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4737_cast = mul(x = var_55607_cast, y = var_55608_to_fp16)[name = tensor<string, []>("aw_4737_cast")];
            tensor<string, []> var_55611_equation_0 = const()[name = tensor<string, []>("op_55611_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55611_cast = einsum(equation = var_55611_equation_0, values = (var_55453_cast, var_55370_cast))[name = tensor<string, []>("op_55611_cast")];
            tensor<fp16, []> var_55612_to_fp16 = const()[name = tensor<string, []>("op_55612_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4739_cast = mul(x = var_55611_cast, y = var_55612_to_fp16)[name = tensor<string, []>("aw_4739_cast")];
            tensor<string, []> var_55615_equation_0 = const()[name = tensor<string, []>("op_55615_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55615_cast = einsum(equation = var_55615_equation_0, values = (var_55457_cast, var_55374_cast))[name = tensor<string, []>("op_55615_cast")];
            tensor<fp16, []> var_55616_to_fp16 = const()[name = tensor<string, []>("op_55616_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4741_cast = mul(x = var_55615_cast, y = var_55616_to_fp16)[name = tensor<string, []>("aw_4741_cast")];
            tensor<string, []> var_55619_equation_0 = const()[name = tensor<string, []>("op_55619_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55619_cast = einsum(equation = var_55619_equation_0, values = (var_55461_cast, var_55378_cast))[name = tensor<string, []>("op_55619_cast")];
            tensor<fp16, []> var_55620_to_fp16 = const()[name = tensor<string, []>("op_55620_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4743_cast = mul(x = var_55619_cast, y = var_55620_to_fp16)[name = tensor<string, []>("aw_4743_cast")];
            tensor<string, []> var_55623_equation_0 = const()[name = tensor<string, []>("op_55623_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55623_cast = einsum(equation = var_55623_equation_0, values = (var_55465_cast, var_55382_cast))[name = tensor<string, []>("op_55623_cast")];
            tensor<fp16, []> var_55624_to_fp16 = const()[name = tensor<string, []>("op_55624_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4745_cast = mul(x = var_55623_cast, y = var_55624_to_fp16)[name = tensor<string, []>("aw_4745_cast")];
            tensor<string, []> var_55627_equation_0 = const()[name = tensor<string, []>("op_55627_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55627_cast = einsum(equation = var_55627_equation_0, values = (var_55469_cast, var_55386_cast))[name = tensor<string, []>("op_55627_cast")];
            tensor<fp16, []> var_55628_to_fp16 = const()[name = tensor<string, []>("op_55628_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4747_cast = mul(x = var_55627_cast, y = var_55628_to_fp16)[name = tensor<string, []>("aw_4747_cast")];
            tensor<string, []> var_55631_equation_0 = const()[name = tensor<string, []>("op_55631_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55631_cast = einsum(equation = var_55631_equation_0, values = (var_55473_cast, var_55390_cast))[name = tensor<string, []>("op_55631_cast")];
            tensor<fp16, []> var_55632_to_fp16 = const()[name = tensor<string, []>("op_55632_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4749_cast = mul(x = var_55631_cast, y = var_55632_to_fp16)[name = tensor<string, []>("aw_4749_cast")];
            tensor<string, []> var_55635_equation_0 = const()[name = tensor<string, []>("op_55635_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55635_cast = einsum(equation = var_55635_equation_0, values = (var_55477_cast, var_55394_cast))[name = tensor<string, []>("op_55635_cast")];
            tensor<fp16, []> var_55636_to_fp16 = const()[name = tensor<string, []>("op_55636_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4751_cast = mul(x = var_55635_cast, y = var_55636_to_fp16)[name = tensor<string, []>("aw_4751_cast")];
            tensor<string, []> var_55639_equation_0 = const()[name = tensor<string, []>("op_55639_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55639_cast = einsum(equation = var_55639_equation_0, values = (var_55481_cast, var_55398_cast))[name = tensor<string, []>("op_55639_cast")];
            tensor<fp16, []> var_55640_to_fp16 = const()[name = tensor<string, []>("op_55640_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4753_cast = mul(x = var_55639_cast, y = var_55640_to_fp16)[name = tensor<string, []>("aw_4753_cast")];
            tensor<string, []> var_55643_equation_0 = const()[name = tensor<string, []>("op_55643_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55643_cast = einsum(equation = var_55643_equation_0, values = (var_55485_cast, var_55402_cast))[name = tensor<string, []>("op_55643_cast")];
            tensor<fp16, []> var_55644_to_fp16 = const()[name = tensor<string, []>("op_55644_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4755_cast = mul(x = var_55643_cast, y = var_55644_to_fp16)[name = tensor<string, []>("aw_4755_cast")];
            tensor<string, []> var_55647_equation_0 = const()[name = tensor<string, []>("op_55647_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55647_cast = einsum(equation = var_55647_equation_0, values = (var_55489_cast, var_55406_cast))[name = tensor<string, []>("op_55647_cast")];
            tensor<fp16, []> var_55648_to_fp16 = const()[name = tensor<string, []>("op_55648_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4757_cast = mul(x = var_55647_cast, y = var_55648_to_fp16)[name = tensor<string, []>("aw_4757_cast")];
            tensor<string, []> var_55651_equation_0 = const()[name = tensor<string, []>("op_55651_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55651_cast = einsum(equation = var_55651_equation_0, values = (var_55493_cast, var_55410_cast))[name = tensor<string, []>("op_55651_cast")];
            tensor<fp16, []> var_55652_to_fp16 = const()[name = tensor<string, []>("op_55652_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4759_cast = mul(x = var_55651_cast, y = var_55652_to_fp16)[name = tensor<string, []>("aw_4759_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55654_cast = softmax(axis = var_30385, x = aw_4721_cast)[name = tensor<string, []>("op_55654_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55655_cast = softmax(axis = var_30385, x = aw_4723_cast)[name = tensor<string, []>("op_55655_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55656_cast = softmax(axis = var_30385, x = aw_4725_cast)[name = tensor<string, []>("op_55656_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55657_cast = softmax(axis = var_30385, x = aw_4727_cast)[name = tensor<string, []>("op_55657_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55658_cast = softmax(axis = var_30385, x = aw_4729_cast)[name = tensor<string, []>("op_55658_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55659_cast = softmax(axis = var_30385, x = aw_4731_cast)[name = tensor<string, []>("op_55659_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55660_cast = softmax(axis = var_30385, x = aw_4733_cast)[name = tensor<string, []>("op_55660_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55661_cast = softmax(axis = var_30385, x = aw_4735_cast)[name = tensor<string, []>("op_55661_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55662_cast = softmax(axis = var_30385, x = aw_4737_cast)[name = tensor<string, []>("op_55662_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55663_cast = softmax(axis = var_30385, x = aw_4739_cast)[name = tensor<string, []>("op_55663_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55664_cast = softmax(axis = var_30385, x = aw_4741_cast)[name = tensor<string, []>("op_55664_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55665_cast = softmax(axis = var_30385, x = aw_4743_cast)[name = tensor<string, []>("op_55665_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55666_cast = softmax(axis = var_30385, x = aw_4745_cast)[name = tensor<string, []>("op_55666_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55667_cast = softmax(axis = var_30385, x = aw_4747_cast)[name = tensor<string, []>("op_55667_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55668_cast = softmax(axis = var_30385, x = aw_4749_cast)[name = tensor<string, []>("op_55668_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55669_cast = softmax(axis = var_30385, x = aw_4751_cast)[name = tensor<string, []>("op_55669_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55670_cast = softmax(axis = var_30385, x = aw_4753_cast)[name = tensor<string, []>("op_55670_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55671_cast = softmax(axis = var_30385, x = aw_4755_cast)[name = tensor<string, []>("op_55671_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55672_cast = softmax(axis = var_30385, x = aw_4757_cast)[name = tensor<string, []>("op_55672_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_55673_cast = softmax(axis = var_30385, x = aw_4759_cast)[name = tensor<string, []>("op_55673_cast")];
            tensor<string, []> var_55675_equation_0 = const()[name = tensor<string, []>("op_55675_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55675_cast = einsum(equation = var_55675_equation_0, values = (var_55495_cast, var_55654_cast))[name = tensor<string, []>("op_55675_cast")];
            tensor<string, []> var_55677_equation_0 = const()[name = tensor<string, []>("op_55677_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55677_cast = einsum(equation = var_55677_equation_0, values = (var_55499_cast, var_55655_cast))[name = tensor<string, []>("op_55677_cast")];
            tensor<string, []> var_55679_equation_0 = const()[name = tensor<string, []>("op_55679_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55679_cast = einsum(equation = var_55679_equation_0, values = (var_55503_cast, var_55656_cast))[name = tensor<string, []>("op_55679_cast")];
            tensor<string, []> var_55681_equation_0 = const()[name = tensor<string, []>("op_55681_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55681_cast = einsum(equation = var_55681_equation_0, values = (var_55507_cast, var_55657_cast))[name = tensor<string, []>("op_55681_cast")];
            tensor<string, []> var_55683_equation_0 = const()[name = tensor<string, []>("op_55683_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55683_cast = einsum(equation = var_55683_equation_0, values = (var_55511_cast, var_55658_cast))[name = tensor<string, []>("op_55683_cast")];
            tensor<string, []> var_55685_equation_0 = const()[name = tensor<string, []>("op_55685_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55685_cast = einsum(equation = var_55685_equation_0, values = (var_55515_cast, var_55659_cast))[name = tensor<string, []>("op_55685_cast")];
            tensor<string, []> var_55687_equation_0 = const()[name = tensor<string, []>("op_55687_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55687_cast = einsum(equation = var_55687_equation_0, values = (var_55519_cast, var_55660_cast))[name = tensor<string, []>("op_55687_cast")];
            tensor<string, []> var_55689_equation_0 = const()[name = tensor<string, []>("op_55689_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55689_cast = einsum(equation = var_55689_equation_0, values = (var_55523_cast, var_55661_cast))[name = tensor<string, []>("op_55689_cast")];
            tensor<string, []> var_55691_equation_0 = const()[name = tensor<string, []>("op_55691_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55691_cast = einsum(equation = var_55691_equation_0, values = (var_55527_cast, var_55662_cast))[name = tensor<string, []>("op_55691_cast")];
            tensor<string, []> var_55693_equation_0 = const()[name = tensor<string, []>("op_55693_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55693_cast = einsum(equation = var_55693_equation_0, values = (var_55531_cast, var_55663_cast))[name = tensor<string, []>("op_55693_cast")];
            tensor<string, []> var_55695_equation_0 = const()[name = tensor<string, []>("op_55695_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55695_cast = einsum(equation = var_55695_equation_0, values = (var_55535_cast, var_55664_cast))[name = tensor<string, []>("op_55695_cast")];
            tensor<string, []> var_55697_equation_0 = const()[name = tensor<string, []>("op_55697_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55697_cast = einsum(equation = var_55697_equation_0, values = (var_55539_cast, var_55665_cast))[name = tensor<string, []>("op_55697_cast")];
            tensor<string, []> var_55699_equation_0 = const()[name = tensor<string, []>("op_55699_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55699_cast = einsum(equation = var_55699_equation_0, values = (var_55543_cast, var_55666_cast))[name = tensor<string, []>("op_55699_cast")];
            tensor<string, []> var_55701_equation_0 = const()[name = tensor<string, []>("op_55701_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55701_cast = einsum(equation = var_55701_equation_0, values = (var_55547_cast, var_55667_cast))[name = tensor<string, []>("op_55701_cast")];
            tensor<string, []> var_55703_equation_0 = const()[name = tensor<string, []>("op_55703_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55703_cast = einsum(equation = var_55703_equation_0, values = (var_55551_cast, var_55668_cast))[name = tensor<string, []>("op_55703_cast")];
            tensor<string, []> var_55705_equation_0 = const()[name = tensor<string, []>("op_55705_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55705_cast = einsum(equation = var_55705_equation_0, values = (var_55555_cast, var_55669_cast))[name = tensor<string, []>("op_55705_cast")];
            tensor<string, []> var_55707_equation_0 = const()[name = tensor<string, []>("op_55707_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55707_cast = einsum(equation = var_55707_equation_0, values = (var_55559_cast, var_55670_cast))[name = tensor<string, []>("op_55707_cast")];
            tensor<string, []> var_55709_equation_0 = const()[name = tensor<string, []>("op_55709_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55709_cast = einsum(equation = var_55709_equation_0, values = (var_55563_cast, var_55671_cast))[name = tensor<string, []>("op_55709_cast")];
            tensor<string, []> var_55711_equation_0 = const()[name = tensor<string, []>("op_55711_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55711_cast = einsum(equation = var_55711_equation_0, values = (var_55567_cast, var_55672_cast))[name = tensor<string, []>("op_55711_cast")];
            tensor<string, []> var_55713_equation_0 = const()[name = tensor<string, []>("op_55713_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_55713_cast = einsum(equation = var_55713_equation_0, values = (var_55571_cast, var_55673_cast))[name = tensor<string, []>("op_55713_cast")];
            tensor<bool, []> input_693_interleave_0 = const()[name = tensor<string, []>("input_693_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_693_cast = concat(axis = var_30385, interleave = input_693_interleave_0, values = (var_55675_cast, var_55677_cast, var_55679_cast, var_55681_cast, var_55683_cast, var_55685_cast, var_55687_cast, var_55689_cast, var_55691_cast, var_55693_cast, var_55695_cast, var_55697_cast, var_55699_cast, var_55701_cast, var_55703_cast, var_55705_cast, var_55707_cast, var_55709_cast, var_55711_cast, var_55713_cast))[name = tensor<string, []>("input_693_cast")];
            tensor<int32, [2]> var_55719 = const()[name = tensor<string, []>("op_55719"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_55721 = const()[name = tensor<string, []>("op_55721"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_55723_pad_type_0 = const()[name = tensor<string, []>("op_55723_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_55723_pad_0 = const()[name = tensor<string, []>("op_55723_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4668517568)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4671794432)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_55723_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_out_0_bias_to_fp16, dilations = var_55721, groups = var_30385, pad = var_55723_pad_0, pad_type = var_55723_pad_type_0, strides = var_55719, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn1_to_out_0_weight_to_fp16, x = input_693_cast)[name = tensor<string, []>("op_55723_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_369_cast = add(x = var_55723_cast, y = inputs_367_cast)[name = tensor<string, []>("inputs_369_cast")];
            tensor<int32, [1]> var_55727 = const()[name = tensor<string, []>("op_55727"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_369_cast = reduce_mean(axes = var_55727, keep_dims = var_30380, x = inputs_369_cast)[name = tensor<string, []>("channels_mean_369_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_369_cast = sub(x = inputs_369_cast, y = channels_mean_369_cast)[name = tensor<string, []>("zero_mean_369_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_369_cast = mul(x = zero_mean_369_cast, y = zero_mean_369_cast)[name = tensor<string, []>("zero_mean_sq_369_cast")];
            tensor<int32, [1]> var_55731 = const()[name = tensor<string, []>("op_55731"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_55732_cast = reduce_mean(axes = var_55731, keep_dims = var_30380, x = zero_mean_sq_369_cast)[name = tensor<string, []>("op_55732_cast")];
            tensor<fp16, []> var_55733_to_fp16 = const()[name = tensor<string, []>("op_55733_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_55734_cast = add(x = var_55732_cast, y = var_55733_to_fp16)[name = tensor<string, []>("op_55734_cast")];
            tensor<fp16, []> denom_369_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_369_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_369_cast = rsqrt(epsilon = denom_369_epsilon_0_to_fp16, x = var_55734_cast)[name = tensor<string, []>("denom_369_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_369_cast = mul(x = zero_mean_369_cast, y = denom_369_cast)[name = tensor<string, []>("out_369_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_55738_to_fp16 = const()[name = tensor<string, []>("op_55738_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4671797056)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_55739_cast = add(x = out_369_cast, y = var_55738_to_fp16)[name = tensor<string, []>("op_55739_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_55741_to_fp16 = const()[name = tensor<string, []>("op_55741_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4671799680)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_479_cast = mul(x = var_55739_cast, y = var_55741_to_fp16)[name = tensor<string, []>("hidden_states_479_cast")];
            tensor<int32, [2]> var_55748 = const()[name = tensor<string, []>("op_55748"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_55750 = const()[name = tensor<string, []>("op_55750"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_247_pad_type_0 = const()[name = tensor<string, []>("q_247_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_247_pad_0 = const()[name = tensor<string, []>("q_247_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4671802304)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_247_cast = conv(dilations = var_55750, groups = var_30385, pad = q_247_pad_0, pad_type = q_247_pad_type_0, strides = var_55748, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_q_weight_to_fp16, x = hidden_states_479_cast)[name = tensor<string, []>("q_247_cast")];
            tensor<int32, [2]> var_55754 = const()[name = tensor<string, []>("op_55754"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_55756 = const()[name = tensor<string, []>("op_55756"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_493_pad_type_0 = const()[name = tensor<string, []>("k_493_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_493_pad_0 = const()[name = tensor<string, []>("k_493_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4675079168)))];
            tensor<fp16, [2, 1280, 1, 77]> k_493_cast = conv(dilations = var_55756, groups = var_30385, pad = k_493_pad_0, pad_type = k_493_pad_type_0, strides = var_55754, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_493_cast")];
            tensor<int32, [2]> var_55760 = const()[name = tensor<string, []>("op_55760"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_55762 = const()[name = tensor<string, []>("op_55762"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_247_pad_type_0 = const()[name = tensor<string, []>("v_247_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_247_pad_0 = const()[name = tensor<string, []>("v_247_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4680322112)))];
            tensor<fp16, [2, 1280, 1, 77]> v_247_cast = conv(dilations = var_55762, groups = var_30385, pad = v_247_pad_0, pad_type = v_247_pad_type_0, strides = var_55760, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_247_cast")];
            tensor<int32, [4]> var_55766_begin_0 = const()[name = tensor<string, []>("op_55766_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_55766_end_0 = const()[name = tensor<string, []>("op_55766_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_55766_end_mask_0 = const()[name = tensor<string, []>("op_55766_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55766_cast = slice_by_index(begin = var_55766_begin_0, end = var_55766_end_0, end_mask = var_55766_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55766_cast")];
            tensor<int32, [4]> var_55770_begin_0 = const()[name = tensor<string, []>("op_55770_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_55770_end_0 = const()[name = tensor<string, []>("op_55770_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_55770_end_mask_0 = const()[name = tensor<string, []>("op_55770_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55770_cast = slice_by_index(begin = var_55770_begin_0, end = var_55770_end_0, end_mask = var_55770_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55770_cast")];
            tensor<int32, [4]> var_55774_begin_0 = const()[name = tensor<string, []>("op_55774_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_55774_end_0 = const()[name = tensor<string, []>("op_55774_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_55774_end_mask_0 = const()[name = tensor<string, []>("op_55774_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55774_cast = slice_by_index(begin = var_55774_begin_0, end = var_55774_end_0, end_mask = var_55774_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55774_cast")];
            tensor<int32, [4]> var_55778_begin_0 = const()[name = tensor<string, []>("op_55778_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_55778_end_0 = const()[name = tensor<string, []>("op_55778_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_55778_end_mask_0 = const()[name = tensor<string, []>("op_55778_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55778_cast = slice_by_index(begin = var_55778_begin_0, end = var_55778_end_0, end_mask = var_55778_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55778_cast")];
            tensor<int32, [4]> var_55782_begin_0 = const()[name = tensor<string, []>("op_55782_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_55782_end_0 = const()[name = tensor<string, []>("op_55782_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_55782_end_mask_0 = const()[name = tensor<string, []>("op_55782_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55782_cast = slice_by_index(begin = var_55782_begin_0, end = var_55782_end_0, end_mask = var_55782_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55782_cast")];
            tensor<int32, [4]> var_55786_begin_0 = const()[name = tensor<string, []>("op_55786_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_55786_end_0 = const()[name = tensor<string, []>("op_55786_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_55786_end_mask_0 = const()[name = tensor<string, []>("op_55786_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55786_cast = slice_by_index(begin = var_55786_begin_0, end = var_55786_end_0, end_mask = var_55786_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55786_cast")];
            tensor<int32, [4]> var_55790_begin_0 = const()[name = tensor<string, []>("op_55790_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_55790_end_0 = const()[name = tensor<string, []>("op_55790_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_55790_end_mask_0 = const()[name = tensor<string, []>("op_55790_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55790_cast = slice_by_index(begin = var_55790_begin_0, end = var_55790_end_0, end_mask = var_55790_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55790_cast")];
            tensor<int32, [4]> var_55794_begin_0 = const()[name = tensor<string, []>("op_55794_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_55794_end_0 = const()[name = tensor<string, []>("op_55794_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_55794_end_mask_0 = const()[name = tensor<string, []>("op_55794_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55794_cast = slice_by_index(begin = var_55794_begin_0, end = var_55794_end_0, end_mask = var_55794_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55794_cast")];
            tensor<int32, [4]> var_55798_begin_0 = const()[name = tensor<string, []>("op_55798_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_55798_end_0 = const()[name = tensor<string, []>("op_55798_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_55798_end_mask_0 = const()[name = tensor<string, []>("op_55798_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55798_cast = slice_by_index(begin = var_55798_begin_0, end = var_55798_end_0, end_mask = var_55798_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55798_cast")];
            tensor<int32, [4]> var_55802_begin_0 = const()[name = tensor<string, []>("op_55802_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_55802_end_0 = const()[name = tensor<string, []>("op_55802_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_55802_end_mask_0 = const()[name = tensor<string, []>("op_55802_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55802_cast = slice_by_index(begin = var_55802_begin_0, end = var_55802_end_0, end_mask = var_55802_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55802_cast")];
            tensor<int32, [4]> var_55806_begin_0 = const()[name = tensor<string, []>("op_55806_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_55806_end_0 = const()[name = tensor<string, []>("op_55806_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_55806_end_mask_0 = const()[name = tensor<string, []>("op_55806_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55806_cast = slice_by_index(begin = var_55806_begin_0, end = var_55806_end_0, end_mask = var_55806_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55806_cast")];
            tensor<int32, [4]> var_55810_begin_0 = const()[name = tensor<string, []>("op_55810_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_55810_end_0 = const()[name = tensor<string, []>("op_55810_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_55810_end_mask_0 = const()[name = tensor<string, []>("op_55810_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55810_cast = slice_by_index(begin = var_55810_begin_0, end = var_55810_end_0, end_mask = var_55810_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55810_cast")];
            tensor<int32, [4]> var_55814_begin_0 = const()[name = tensor<string, []>("op_55814_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_55814_end_0 = const()[name = tensor<string, []>("op_55814_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_55814_end_mask_0 = const()[name = tensor<string, []>("op_55814_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55814_cast = slice_by_index(begin = var_55814_begin_0, end = var_55814_end_0, end_mask = var_55814_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55814_cast")];
            tensor<int32, [4]> var_55818_begin_0 = const()[name = tensor<string, []>("op_55818_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_55818_end_0 = const()[name = tensor<string, []>("op_55818_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_55818_end_mask_0 = const()[name = tensor<string, []>("op_55818_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55818_cast = slice_by_index(begin = var_55818_begin_0, end = var_55818_end_0, end_mask = var_55818_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55818_cast")];
            tensor<int32, [4]> var_55822_begin_0 = const()[name = tensor<string, []>("op_55822_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_55822_end_0 = const()[name = tensor<string, []>("op_55822_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_55822_end_mask_0 = const()[name = tensor<string, []>("op_55822_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55822_cast = slice_by_index(begin = var_55822_begin_0, end = var_55822_end_0, end_mask = var_55822_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55822_cast")];
            tensor<int32, [4]> var_55826_begin_0 = const()[name = tensor<string, []>("op_55826_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_55826_end_0 = const()[name = tensor<string, []>("op_55826_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_55826_end_mask_0 = const()[name = tensor<string, []>("op_55826_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55826_cast = slice_by_index(begin = var_55826_begin_0, end = var_55826_end_0, end_mask = var_55826_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55826_cast")];
            tensor<int32, [4]> var_55830_begin_0 = const()[name = tensor<string, []>("op_55830_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_55830_end_0 = const()[name = tensor<string, []>("op_55830_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_55830_end_mask_0 = const()[name = tensor<string, []>("op_55830_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55830_cast = slice_by_index(begin = var_55830_begin_0, end = var_55830_end_0, end_mask = var_55830_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55830_cast")];
            tensor<int32, [4]> var_55834_begin_0 = const()[name = tensor<string, []>("op_55834_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_55834_end_0 = const()[name = tensor<string, []>("op_55834_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_55834_end_mask_0 = const()[name = tensor<string, []>("op_55834_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55834_cast = slice_by_index(begin = var_55834_begin_0, end = var_55834_end_0, end_mask = var_55834_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55834_cast")];
            tensor<int32, [4]> var_55838_begin_0 = const()[name = tensor<string, []>("op_55838_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_55838_end_0 = const()[name = tensor<string, []>("op_55838_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_55838_end_mask_0 = const()[name = tensor<string, []>("op_55838_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55838_cast = slice_by_index(begin = var_55838_begin_0, end = var_55838_end_0, end_mask = var_55838_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55838_cast")];
            tensor<int32, [4]> var_55842_begin_0 = const()[name = tensor<string, []>("op_55842_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_55842_end_0 = const()[name = tensor<string, []>("op_55842_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_55842_end_mask_0 = const()[name = tensor<string, []>("op_55842_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_55842_cast = slice_by_index(begin = var_55842_begin_0, end = var_55842_end_0, end_mask = var_55842_end_mask_0, x = q_247_cast)[name = tensor<string, []>("op_55842_cast")];
            tensor<int32, [4]> k_495_perm_0 = const()[name = tensor<string, []>("k_495_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_55849_begin_0 = const()[name = tensor<string, []>("op_55849_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_55849_end_0 = const()[name = tensor<string, []>("op_55849_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_55849_end_mask_0 = const()[name = tensor<string, []>("op_55849_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_16 = transpose(perm = k_495_perm_0, x = k_493_cast)[name = tensor<string, []>("transpose_16")];
            tensor<fp16, [2, 77, 1, 64]> var_55849_cast = slice_by_index(begin = var_55849_begin_0, end = var_55849_end_0, end_mask = var_55849_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55849_cast")];
            tensor<int32, [4]> var_55853_begin_0 = const()[name = tensor<string, []>("op_55853_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_55853_end_0 = const()[name = tensor<string, []>("op_55853_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_55853_end_mask_0 = const()[name = tensor<string, []>("op_55853_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55853_cast = slice_by_index(begin = var_55853_begin_0, end = var_55853_end_0, end_mask = var_55853_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55853_cast")];
            tensor<int32, [4]> var_55857_begin_0 = const()[name = tensor<string, []>("op_55857_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_55857_end_0 = const()[name = tensor<string, []>("op_55857_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_55857_end_mask_0 = const()[name = tensor<string, []>("op_55857_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55857_cast = slice_by_index(begin = var_55857_begin_0, end = var_55857_end_0, end_mask = var_55857_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55857_cast")];
            tensor<int32, [4]> var_55861_begin_0 = const()[name = tensor<string, []>("op_55861_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_55861_end_0 = const()[name = tensor<string, []>("op_55861_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_55861_end_mask_0 = const()[name = tensor<string, []>("op_55861_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55861_cast = slice_by_index(begin = var_55861_begin_0, end = var_55861_end_0, end_mask = var_55861_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55861_cast")];
            tensor<int32, [4]> var_55865_begin_0 = const()[name = tensor<string, []>("op_55865_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_55865_end_0 = const()[name = tensor<string, []>("op_55865_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_55865_end_mask_0 = const()[name = tensor<string, []>("op_55865_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55865_cast = slice_by_index(begin = var_55865_begin_0, end = var_55865_end_0, end_mask = var_55865_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55865_cast")];
            tensor<int32, [4]> var_55869_begin_0 = const()[name = tensor<string, []>("op_55869_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_55869_end_0 = const()[name = tensor<string, []>("op_55869_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_55869_end_mask_0 = const()[name = tensor<string, []>("op_55869_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55869_cast = slice_by_index(begin = var_55869_begin_0, end = var_55869_end_0, end_mask = var_55869_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55869_cast")];
            tensor<int32, [4]> var_55873_begin_0 = const()[name = tensor<string, []>("op_55873_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_55873_end_0 = const()[name = tensor<string, []>("op_55873_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_55873_end_mask_0 = const()[name = tensor<string, []>("op_55873_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55873_cast = slice_by_index(begin = var_55873_begin_0, end = var_55873_end_0, end_mask = var_55873_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55873_cast")];
            tensor<int32, [4]> var_55877_begin_0 = const()[name = tensor<string, []>("op_55877_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_55877_end_0 = const()[name = tensor<string, []>("op_55877_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_55877_end_mask_0 = const()[name = tensor<string, []>("op_55877_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55877_cast = slice_by_index(begin = var_55877_begin_0, end = var_55877_end_0, end_mask = var_55877_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55877_cast")];
            tensor<int32, [4]> var_55881_begin_0 = const()[name = tensor<string, []>("op_55881_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_55881_end_0 = const()[name = tensor<string, []>("op_55881_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_55881_end_mask_0 = const()[name = tensor<string, []>("op_55881_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55881_cast = slice_by_index(begin = var_55881_begin_0, end = var_55881_end_0, end_mask = var_55881_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55881_cast")];
            tensor<int32, [4]> var_55885_begin_0 = const()[name = tensor<string, []>("op_55885_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_55885_end_0 = const()[name = tensor<string, []>("op_55885_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_55885_end_mask_0 = const()[name = tensor<string, []>("op_55885_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55885_cast = slice_by_index(begin = var_55885_begin_0, end = var_55885_end_0, end_mask = var_55885_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55885_cast")];
            tensor<int32, [4]> var_55889_begin_0 = const()[name = tensor<string, []>("op_55889_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_55889_end_0 = const()[name = tensor<string, []>("op_55889_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_55889_end_mask_0 = const()[name = tensor<string, []>("op_55889_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55889_cast = slice_by_index(begin = var_55889_begin_0, end = var_55889_end_0, end_mask = var_55889_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55889_cast")];
            tensor<int32, [4]> var_55893_begin_0 = const()[name = tensor<string, []>("op_55893_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_55893_end_0 = const()[name = tensor<string, []>("op_55893_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_55893_end_mask_0 = const()[name = tensor<string, []>("op_55893_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55893_cast = slice_by_index(begin = var_55893_begin_0, end = var_55893_end_0, end_mask = var_55893_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55893_cast")];
            tensor<int32, [4]> var_55897_begin_0 = const()[name = tensor<string, []>("op_55897_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_55897_end_0 = const()[name = tensor<string, []>("op_55897_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_55897_end_mask_0 = const()[name = tensor<string, []>("op_55897_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55897_cast = slice_by_index(begin = var_55897_begin_0, end = var_55897_end_0, end_mask = var_55897_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55897_cast")];
            tensor<int32, [4]> var_55901_begin_0 = const()[name = tensor<string, []>("op_55901_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_55901_end_0 = const()[name = tensor<string, []>("op_55901_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_55901_end_mask_0 = const()[name = tensor<string, []>("op_55901_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55901_cast = slice_by_index(begin = var_55901_begin_0, end = var_55901_end_0, end_mask = var_55901_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55901_cast")];
            tensor<int32, [4]> var_55905_begin_0 = const()[name = tensor<string, []>("op_55905_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_55905_end_0 = const()[name = tensor<string, []>("op_55905_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_55905_end_mask_0 = const()[name = tensor<string, []>("op_55905_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55905_cast = slice_by_index(begin = var_55905_begin_0, end = var_55905_end_0, end_mask = var_55905_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55905_cast")];
            tensor<int32, [4]> var_55909_begin_0 = const()[name = tensor<string, []>("op_55909_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_55909_end_0 = const()[name = tensor<string, []>("op_55909_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_55909_end_mask_0 = const()[name = tensor<string, []>("op_55909_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55909_cast = slice_by_index(begin = var_55909_begin_0, end = var_55909_end_0, end_mask = var_55909_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55909_cast")];
            tensor<int32, [4]> var_55913_begin_0 = const()[name = tensor<string, []>("op_55913_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_55913_end_0 = const()[name = tensor<string, []>("op_55913_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_55913_end_mask_0 = const()[name = tensor<string, []>("op_55913_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55913_cast = slice_by_index(begin = var_55913_begin_0, end = var_55913_end_0, end_mask = var_55913_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55913_cast")];
            tensor<int32, [4]> var_55917_begin_0 = const()[name = tensor<string, []>("op_55917_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_55917_end_0 = const()[name = tensor<string, []>("op_55917_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_55917_end_mask_0 = const()[name = tensor<string, []>("op_55917_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55917_cast = slice_by_index(begin = var_55917_begin_0, end = var_55917_end_0, end_mask = var_55917_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55917_cast")];
            tensor<int32, [4]> var_55921_begin_0 = const()[name = tensor<string, []>("op_55921_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_55921_end_0 = const()[name = tensor<string, []>("op_55921_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_55921_end_mask_0 = const()[name = tensor<string, []>("op_55921_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55921_cast = slice_by_index(begin = var_55921_begin_0, end = var_55921_end_0, end_mask = var_55921_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55921_cast")];
            tensor<int32, [4]> var_55925_begin_0 = const()[name = tensor<string, []>("op_55925_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_55925_end_0 = const()[name = tensor<string, []>("op_55925_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_55925_end_mask_0 = const()[name = tensor<string, []>("op_55925_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_55925_cast = slice_by_index(begin = var_55925_begin_0, end = var_55925_end_0, end_mask = var_55925_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_55925_cast")];
            tensor<int32, [4]> var_55927_begin_0 = const()[name = tensor<string, []>("op_55927_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_55927_end_0 = const()[name = tensor<string, []>("op_55927_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_55927_end_mask_0 = const()[name = tensor<string, []>("op_55927_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55927_cast = slice_by_index(begin = var_55927_begin_0, end = var_55927_end_0, end_mask = var_55927_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_55927_cast")];
            tensor<int32, [4]> var_55931_begin_0 = const()[name = tensor<string, []>("op_55931_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_55931_end_0 = const()[name = tensor<string, []>("op_55931_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_55931_end_mask_0 = const()[name = tensor<string, []>("op_55931_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55931_cast = slice_by_index(begin = var_55931_begin_0, end = var_55931_end_0, end_mask = var_55931_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_55931_cast")];
            tensor<int32, [4]> var_55935_begin_0 = const()[name = tensor<string, []>("op_55935_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_55935_end_0 = const()[name = tensor<string, []>("op_55935_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_55935_end_mask_0 = const()[name = tensor<string, []>("op_55935_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55935_cast = slice_by_index(begin = var_55935_begin_0, end = var_55935_end_0, end_mask = var_55935_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_55935_cast")];
            tensor<int32, [4]> var_55939_begin_0 = const()[name = tensor<string, []>("op_55939_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_55939_end_0 = const()[name = tensor<string, []>("op_55939_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_55939_end_mask_0 = const()[name = tensor<string, []>("op_55939_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55939_cast = slice_by_index(begin = var_55939_begin_0, end = var_55939_end_0, end_mask = var_55939_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_55939_cast")];
            tensor<int32, [4]> var_55943_begin_0 = const()[name = tensor<string, []>("op_55943_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_55943_end_0 = const()[name = tensor<string, []>("op_55943_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_55943_end_mask_0 = const()[name = tensor<string, []>("op_55943_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55943_cast = slice_by_index(begin = var_55943_begin_0, end = var_55943_end_0, end_mask = var_55943_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_55943_cast")];
            tensor<int32, [4]> var_55947_begin_0 = const()[name = tensor<string, []>("op_55947_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_55947_end_0 = const()[name = tensor<string, []>("op_55947_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_55947_end_mask_0 = const()[name = tensor<string, []>("op_55947_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55947_cast = slice_by_index(begin = var_55947_begin_0, end = var_55947_end_0, end_mask = var_55947_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_55947_cast")];
            tensor<int32, [4]> var_55951_begin_0 = const()[name = tensor<string, []>("op_55951_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_55951_end_0 = const()[name = tensor<string, []>("op_55951_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_55951_end_mask_0 = const()[name = tensor<string, []>("op_55951_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55951_cast = slice_by_index(begin = var_55951_begin_0, end = var_55951_end_0, end_mask = var_55951_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_55951_cast")];
            tensor<int32, [4]> var_55955_begin_0 = const()[name = tensor<string, []>("op_55955_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_55955_end_0 = const()[name = tensor<string, []>("op_55955_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_55955_end_mask_0 = const()[name = tensor<string, []>("op_55955_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55955_cast = slice_by_index(begin = var_55955_begin_0, end = var_55955_end_0, end_mask = var_55955_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_55955_cast")];
            tensor<int32, [4]> var_55959_begin_0 = const()[name = tensor<string, []>("op_55959_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_55959_end_0 = const()[name = tensor<string, []>("op_55959_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_55959_end_mask_0 = const()[name = tensor<string, []>("op_55959_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55959_cast = slice_by_index(begin = var_55959_begin_0, end = var_55959_end_0, end_mask = var_55959_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_55959_cast")];
            tensor<int32, [4]> var_55963_begin_0 = const()[name = tensor<string, []>("op_55963_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_55963_end_0 = const()[name = tensor<string, []>("op_55963_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_55963_end_mask_0 = const()[name = tensor<string, []>("op_55963_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55963_cast = slice_by_index(begin = var_55963_begin_0, end = var_55963_end_0, end_mask = var_55963_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_55963_cast")];
            tensor<int32, [4]> var_55967_begin_0 = const()[name = tensor<string, []>("op_55967_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_55967_end_0 = const()[name = tensor<string, []>("op_55967_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_55967_end_mask_0 = const()[name = tensor<string, []>("op_55967_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55967_cast = slice_by_index(begin = var_55967_begin_0, end = var_55967_end_0, end_mask = var_55967_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_55967_cast")];
            tensor<int32, [4]> var_55971_begin_0 = const()[name = tensor<string, []>("op_55971_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_55971_end_0 = const()[name = tensor<string, []>("op_55971_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_55971_end_mask_0 = const()[name = tensor<string, []>("op_55971_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55971_cast = slice_by_index(begin = var_55971_begin_0, end = var_55971_end_0, end_mask = var_55971_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_55971_cast")];
            tensor<int32, [4]> var_55975_begin_0 = const()[name = tensor<string, []>("op_55975_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_55975_end_0 = const()[name = tensor<string, []>("op_55975_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_55975_end_mask_0 = const()[name = tensor<string, []>("op_55975_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55975_cast = slice_by_index(begin = var_55975_begin_0, end = var_55975_end_0, end_mask = var_55975_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_55975_cast")];
            tensor<int32, [4]> var_55979_begin_0 = const()[name = tensor<string, []>("op_55979_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_55979_end_0 = const()[name = tensor<string, []>("op_55979_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_55979_end_mask_0 = const()[name = tensor<string, []>("op_55979_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55979_cast = slice_by_index(begin = var_55979_begin_0, end = var_55979_end_0, end_mask = var_55979_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_55979_cast")];
            tensor<int32, [4]> var_55983_begin_0 = const()[name = tensor<string, []>("op_55983_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_55983_end_0 = const()[name = tensor<string, []>("op_55983_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_55983_end_mask_0 = const()[name = tensor<string, []>("op_55983_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55983_cast = slice_by_index(begin = var_55983_begin_0, end = var_55983_end_0, end_mask = var_55983_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_55983_cast")];
            tensor<int32, [4]> var_55987_begin_0 = const()[name = tensor<string, []>("op_55987_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_55987_end_0 = const()[name = tensor<string, []>("op_55987_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_55987_end_mask_0 = const()[name = tensor<string, []>("op_55987_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55987_cast = slice_by_index(begin = var_55987_begin_0, end = var_55987_end_0, end_mask = var_55987_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_55987_cast")];
            tensor<int32, [4]> var_55991_begin_0 = const()[name = tensor<string, []>("op_55991_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_55991_end_0 = const()[name = tensor<string, []>("op_55991_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_55991_end_mask_0 = const()[name = tensor<string, []>("op_55991_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55991_cast = slice_by_index(begin = var_55991_begin_0, end = var_55991_end_0, end_mask = var_55991_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_55991_cast")];
            tensor<int32, [4]> var_55995_begin_0 = const()[name = tensor<string, []>("op_55995_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_55995_end_0 = const()[name = tensor<string, []>("op_55995_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_55995_end_mask_0 = const()[name = tensor<string, []>("op_55995_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55995_cast = slice_by_index(begin = var_55995_begin_0, end = var_55995_end_0, end_mask = var_55995_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_55995_cast")];
            tensor<int32, [4]> var_55999_begin_0 = const()[name = tensor<string, []>("op_55999_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_55999_end_0 = const()[name = tensor<string, []>("op_55999_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_55999_end_mask_0 = const()[name = tensor<string, []>("op_55999_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_55999_cast = slice_by_index(begin = var_55999_begin_0, end = var_55999_end_0, end_mask = var_55999_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_55999_cast")];
            tensor<int32, [4]> var_56003_begin_0 = const()[name = tensor<string, []>("op_56003_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_56003_end_0 = const()[name = tensor<string, []>("op_56003_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_56003_end_mask_0 = const()[name = tensor<string, []>("op_56003_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56003_cast = slice_by_index(begin = var_56003_begin_0, end = var_56003_end_0, end_mask = var_56003_end_mask_0, x = v_247_cast)[name = tensor<string, []>("op_56003_cast")];
            tensor<string, []> var_56007_equation_0 = const()[name = tensor<string, []>("op_56007_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56007_cast = einsum(equation = var_56007_equation_0, values = (var_55849_cast, var_55766_cast))[name = tensor<string, []>("op_56007_cast")];
            tensor<fp16, []> var_56008_to_fp16 = const()[name = tensor<string, []>("op_56008_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4761_cast = mul(x = var_56007_cast, y = var_56008_to_fp16)[name = tensor<string, []>("aw_4761_cast")];
            tensor<string, []> var_56011_equation_0 = const()[name = tensor<string, []>("op_56011_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56011_cast = einsum(equation = var_56011_equation_0, values = (var_55853_cast, var_55770_cast))[name = tensor<string, []>("op_56011_cast")];
            tensor<fp16, []> var_56012_to_fp16 = const()[name = tensor<string, []>("op_56012_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4763_cast = mul(x = var_56011_cast, y = var_56012_to_fp16)[name = tensor<string, []>("aw_4763_cast")];
            tensor<string, []> var_56015_equation_0 = const()[name = tensor<string, []>("op_56015_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56015_cast = einsum(equation = var_56015_equation_0, values = (var_55857_cast, var_55774_cast))[name = tensor<string, []>("op_56015_cast")];
            tensor<fp16, []> var_56016_to_fp16 = const()[name = tensor<string, []>("op_56016_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4765_cast = mul(x = var_56015_cast, y = var_56016_to_fp16)[name = tensor<string, []>("aw_4765_cast")];
            tensor<string, []> var_56019_equation_0 = const()[name = tensor<string, []>("op_56019_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56019_cast = einsum(equation = var_56019_equation_0, values = (var_55861_cast, var_55778_cast))[name = tensor<string, []>("op_56019_cast")];
            tensor<fp16, []> var_56020_to_fp16 = const()[name = tensor<string, []>("op_56020_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4767_cast = mul(x = var_56019_cast, y = var_56020_to_fp16)[name = tensor<string, []>("aw_4767_cast")];
            tensor<string, []> var_56023_equation_0 = const()[name = tensor<string, []>("op_56023_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56023_cast = einsum(equation = var_56023_equation_0, values = (var_55865_cast, var_55782_cast))[name = tensor<string, []>("op_56023_cast")];
            tensor<fp16, []> var_56024_to_fp16 = const()[name = tensor<string, []>("op_56024_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4769_cast = mul(x = var_56023_cast, y = var_56024_to_fp16)[name = tensor<string, []>("aw_4769_cast")];
            tensor<string, []> var_56027_equation_0 = const()[name = tensor<string, []>("op_56027_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56027_cast = einsum(equation = var_56027_equation_0, values = (var_55869_cast, var_55786_cast))[name = tensor<string, []>("op_56027_cast")];
            tensor<fp16, []> var_56028_to_fp16 = const()[name = tensor<string, []>("op_56028_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4771_cast = mul(x = var_56027_cast, y = var_56028_to_fp16)[name = tensor<string, []>("aw_4771_cast")];
            tensor<string, []> var_56031_equation_0 = const()[name = tensor<string, []>("op_56031_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56031_cast = einsum(equation = var_56031_equation_0, values = (var_55873_cast, var_55790_cast))[name = tensor<string, []>("op_56031_cast")];
            tensor<fp16, []> var_56032_to_fp16 = const()[name = tensor<string, []>("op_56032_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4773_cast = mul(x = var_56031_cast, y = var_56032_to_fp16)[name = tensor<string, []>("aw_4773_cast")];
            tensor<string, []> var_56035_equation_0 = const()[name = tensor<string, []>("op_56035_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56035_cast = einsum(equation = var_56035_equation_0, values = (var_55877_cast, var_55794_cast))[name = tensor<string, []>("op_56035_cast")];
            tensor<fp16, []> var_56036_to_fp16 = const()[name = tensor<string, []>("op_56036_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4775_cast = mul(x = var_56035_cast, y = var_56036_to_fp16)[name = tensor<string, []>("aw_4775_cast")];
            tensor<string, []> var_56039_equation_0 = const()[name = tensor<string, []>("op_56039_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56039_cast = einsum(equation = var_56039_equation_0, values = (var_55881_cast, var_55798_cast))[name = tensor<string, []>("op_56039_cast")];
            tensor<fp16, []> var_56040_to_fp16 = const()[name = tensor<string, []>("op_56040_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4777_cast = mul(x = var_56039_cast, y = var_56040_to_fp16)[name = tensor<string, []>("aw_4777_cast")];
            tensor<string, []> var_56043_equation_0 = const()[name = tensor<string, []>("op_56043_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56043_cast = einsum(equation = var_56043_equation_0, values = (var_55885_cast, var_55802_cast))[name = tensor<string, []>("op_56043_cast")];
            tensor<fp16, []> var_56044_to_fp16 = const()[name = tensor<string, []>("op_56044_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4779_cast = mul(x = var_56043_cast, y = var_56044_to_fp16)[name = tensor<string, []>("aw_4779_cast")];
            tensor<string, []> var_56047_equation_0 = const()[name = tensor<string, []>("op_56047_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56047_cast = einsum(equation = var_56047_equation_0, values = (var_55889_cast, var_55806_cast))[name = tensor<string, []>("op_56047_cast")];
            tensor<fp16, []> var_56048_to_fp16 = const()[name = tensor<string, []>("op_56048_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4781_cast = mul(x = var_56047_cast, y = var_56048_to_fp16)[name = tensor<string, []>("aw_4781_cast")];
            tensor<string, []> var_56051_equation_0 = const()[name = tensor<string, []>("op_56051_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56051_cast = einsum(equation = var_56051_equation_0, values = (var_55893_cast, var_55810_cast))[name = tensor<string, []>("op_56051_cast")];
            tensor<fp16, []> var_56052_to_fp16 = const()[name = tensor<string, []>("op_56052_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4783_cast = mul(x = var_56051_cast, y = var_56052_to_fp16)[name = tensor<string, []>("aw_4783_cast")];
            tensor<string, []> var_56055_equation_0 = const()[name = tensor<string, []>("op_56055_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56055_cast = einsum(equation = var_56055_equation_0, values = (var_55897_cast, var_55814_cast))[name = tensor<string, []>("op_56055_cast")];
            tensor<fp16, []> var_56056_to_fp16 = const()[name = tensor<string, []>("op_56056_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4785_cast = mul(x = var_56055_cast, y = var_56056_to_fp16)[name = tensor<string, []>("aw_4785_cast")];
            tensor<string, []> var_56059_equation_0 = const()[name = tensor<string, []>("op_56059_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56059_cast = einsum(equation = var_56059_equation_0, values = (var_55901_cast, var_55818_cast))[name = tensor<string, []>("op_56059_cast")];
            tensor<fp16, []> var_56060_to_fp16 = const()[name = tensor<string, []>("op_56060_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4787_cast = mul(x = var_56059_cast, y = var_56060_to_fp16)[name = tensor<string, []>("aw_4787_cast")];
            tensor<string, []> var_56063_equation_0 = const()[name = tensor<string, []>("op_56063_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56063_cast = einsum(equation = var_56063_equation_0, values = (var_55905_cast, var_55822_cast))[name = tensor<string, []>("op_56063_cast")];
            tensor<fp16, []> var_56064_to_fp16 = const()[name = tensor<string, []>("op_56064_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4789_cast = mul(x = var_56063_cast, y = var_56064_to_fp16)[name = tensor<string, []>("aw_4789_cast")];
            tensor<string, []> var_56067_equation_0 = const()[name = tensor<string, []>("op_56067_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56067_cast = einsum(equation = var_56067_equation_0, values = (var_55909_cast, var_55826_cast))[name = tensor<string, []>("op_56067_cast")];
            tensor<fp16, []> var_56068_to_fp16 = const()[name = tensor<string, []>("op_56068_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4791_cast = mul(x = var_56067_cast, y = var_56068_to_fp16)[name = tensor<string, []>("aw_4791_cast")];
            tensor<string, []> var_56071_equation_0 = const()[name = tensor<string, []>("op_56071_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56071_cast = einsum(equation = var_56071_equation_0, values = (var_55913_cast, var_55830_cast))[name = tensor<string, []>("op_56071_cast")];
            tensor<fp16, []> var_56072_to_fp16 = const()[name = tensor<string, []>("op_56072_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4793_cast = mul(x = var_56071_cast, y = var_56072_to_fp16)[name = tensor<string, []>("aw_4793_cast")];
            tensor<string, []> var_56075_equation_0 = const()[name = tensor<string, []>("op_56075_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56075_cast = einsum(equation = var_56075_equation_0, values = (var_55917_cast, var_55834_cast))[name = tensor<string, []>("op_56075_cast")];
            tensor<fp16, []> var_56076_to_fp16 = const()[name = tensor<string, []>("op_56076_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4795_cast = mul(x = var_56075_cast, y = var_56076_to_fp16)[name = tensor<string, []>("aw_4795_cast")];
            tensor<string, []> var_56079_equation_0 = const()[name = tensor<string, []>("op_56079_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56079_cast = einsum(equation = var_56079_equation_0, values = (var_55921_cast, var_55838_cast))[name = tensor<string, []>("op_56079_cast")];
            tensor<fp16, []> var_56080_to_fp16 = const()[name = tensor<string, []>("op_56080_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4797_cast = mul(x = var_56079_cast, y = var_56080_to_fp16)[name = tensor<string, []>("aw_4797_cast")];
            tensor<string, []> var_56083_equation_0 = const()[name = tensor<string, []>("op_56083_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56083_cast = einsum(equation = var_56083_equation_0, values = (var_55925_cast, var_55842_cast))[name = tensor<string, []>("op_56083_cast")];
            tensor<fp16, []> var_56084_to_fp16 = const()[name = tensor<string, []>("op_56084_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4799_cast = mul(x = var_56083_cast, y = var_56084_to_fp16)[name = tensor<string, []>("aw_4799_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56086_cast = softmax(axis = var_30385, x = aw_4761_cast)[name = tensor<string, []>("op_56086_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56087_cast = softmax(axis = var_30385, x = aw_4763_cast)[name = tensor<string, []>("op_56087_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56088_cast = softmax(axis = var_30385, x = aw_4765_cast)[name = tensor<string, []>("op_56088_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56089_cast = softmax(axis = var_30385, x = aw_4767_cast)[name = tensor<string, []>("op_56089_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56090_cast = softmax(axis = var_30385, x = aw_4769_cast)[name = tensor<string, []>("op_56090_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56091_cast = softmax(axis = var_30385, x = aw_4771_cast)[name = tensor<string, []>("op_56091_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56092_cast = softmax(axis = var_30385, x = aw_4773_cast)[name = tensor<string, []>("op_56092_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56093_cast = softmax(axis = var_30385, x = aw_4775_cast)[name = tensor<string, []>("op_56093_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56094_cast = softmax(axis = var_30385, x = aw_4777_cast)[name = tensor<string, []>("op_56094_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56095_cast = softmax(axis = var_30385, x = aw_4779_cast)[name = tensor<string, []>("op_56095_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56096_cast = softmax(axis = var_30385, x = aw_4781_cast)[name = tensor<string, []>("op_56096_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56097_cast = softmax(axis = var_30385, x = aw_4783_cast)[name = tensor<string, []>("op_56097_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56098_cast = softmax(axis = var_30385, x = aw_4785_cast)[name = tensor<string, []>("op_56098_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56099_cast = softmax(axis = var_30385, x = aw_4787_cast)[name = tensor<string, []>("op_56099_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56100_cast = softmax(axis = var_30385, x = aw_4789_cast)[name = tensor<string, []>("op_56100_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56101_cast = softmax(axis = var_30385, x = aw_4791_cast)[name = tensor<string, []>("op_56101_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56102_cast = softmax(axis = var_30385, x = aw_4793_cast)[name = tensor<string, []>("op_56102_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56103_cast = softmax(axis = var_30385, x = aw_4795_cast)[name = tensor<string, []>("op_56103_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56104_cast = softmax(axis = var_30385, x = aw_4797_cast)[name = tensor<string, []>("op_56104_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56105_cast = softmax(axis = var_30385, x = aw_4799_cast)[name = tensor<string, []>("op_56105_cast")];
            tensor<string, []> var_56107_equation_0 = const()[name = tensor<string, []>("op_56107_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56107_cast = einsum(equation = var_56107_equation_0, values = (var_55927_cast, var_56086_cast))[name = tensor<string, []>("op_56107_cast")];
            tensor<string, []> var_56109_equation_0 = const()[name = tensor<string, []>("op_56109_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56109_cast = einsum(equation = var_56109_equation_0, values = (var_55931_cast, var_56087_cast))[name = tensor<string, []>("op_56109_cast")];
            tensor<string, []> var_56111_equation_0 = const()[name = tensor<string, []>("op_56111_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56111_cast = einsum(equation = var_56111_equation_0, values = (var_55935_cast, var_56088_cast))[name = tensor<string, []>("op_56111_cast")];
            tensor<string, []> var_56113_equation_0 = const()[name = tensor<string, []>("op_56113_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56113_cast = einsum(equation = var_56113_equation_0, values = (var_55939_cast, var_56089_cast))[name = tensor<string, []>("op_56113_cast")];
            tensor<string, []> var_56115_equation_0 = const()[name = tensor<string, []>("op_56115_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56115_cast = einsum(equation = var_56115_equation_0, values = (var_55943_cast, var_56090_cast))[name = tensor<string, []>("op_56115_cast")];
            tensor<string, []> var_56117_equation_0 = const()[name = tensor<string, []>("op_56117_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56117_cast = einsum(equation = var_56117_equation_0, values = (var_55947_cast, var_56091_cast))[name = tensor<string, []>("op_56117_cast")];
            tensor<string, []> var_56119_equation_0 = const()[name = tensor<string, []>("op_56119_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56119_cast = einsum(equation = var_56119_equation_0, values = (var_55951_cast, var_56092_cast))[name = tensor<string, []>("op_56119_cast")];
            tensor<string, []> var_56121_equation_0 = const()[name = tensor<string, []>("op_56121_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56121_cast = einsum(equation = var_56121_equation_0, values = (var_55955_cast, var_56093_cast))[name = tensor<string, []>("op_56121_cast")];
            tensor<string, []> var_56123_equation_0 = const()[name = tensor<string, []>("op_56123_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56123_cast = einsum(equation = var_56123_equation_0, values = (var_55959_cast, var_56094_cast))[name = tensor<string, []>("op_56123_cast")];
            tensor<string, []> var_56125_equation_0 = const()[name = tensor<string, []>("op_56125_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56125_cast = einsum(equation = var_56125_equation_0, values = (var_55963_cast, var_56095_cast))[name = tensor<string, []>("op_56125_cast")];
            tensor<string, []> var_56127_equation_0 = const()[name = tensor<string, []>("op_56127_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56127_cast = einsum(equation = var_56127_equation_0, values = (var_55967_cast, var_56096_cast))[name = tensor<string, []>("op_56127_cast")];
            tensor<string, []> var_56129_equation_0 = const()[name = tensor<string, []>("op_56129_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56129_cast = einsum(equation = var_56129_equation_0, values = (var_55971_cast, var_56097_cast))[name = tensor<string, []>("op_56129_cast")];
            tensor<string, []> var_56131_equation_0 = const()[name = tensor<string, []>("op_56131_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56131_cast = einsum(equation = var_56131_equation_0, values = (var_55975_cast, var_56098_cast))[name = tensor<string, []>("op_56131_cast")];
            tensor<string, []> var_56133_equation_0 = const()[name = tensor<string, []>("op_56133_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56133_cast = einsum(equation = var_56133_equation_0, values = (var_55979_cast, var_56099_cast))[name = tensor<string, []>("op_56133_cast")];
            tensor<string, []> var_56135_equation_0 = const()[name = tensor<string, []>("op_56135_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56135_cast = einsum(equation = var_56135_equation_0, values = (var_55983_cast, var_56100_cast))[name = tensor<string, []>("op_56135_cast")];
            tensor<string, []> var_56137_equation_0 = const()[name = tensor<string, []>("op_56137_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56137_cast = einsum(equation = var_56137_equation_0, values = (var_55987_cast, var_56101_cast))[name = tensor<string, []>("op_56137_cast")];
            tensor<string, []> var_56139_equation_0 = const()[name = tensor<string, []>("op_56139_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56139_cast = einsum(equation = var_56139_equation_0, values = (var_55991_cast, var_56102_cast))[name = tensor<string, []>("op_56139_cast")];
            tensor<string, []> var_56141_equation_0 = const()[name = tensor<string, []>("op_56141_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56141_cast = einsum(equation = var_56141_equation_0, values = (var_55995_cast, var_56103_cast))[name = tensor<string, []>("op_56141_cast")];
            tensor<string, []> var_56143_equation_0 = const()[name = tensor<string, []>("op_56143_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56143_cast = einsum(equation = var_56143_equation_0, values = (var_55999_cast, var_56104_cast))[name = tensor<string, []>("op_56143_cast")];
            tensor<string, []> var_56145_equation_0 = const()[name = tensor<string, []>("op_56145_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56145_cast = einsum(equation = var_56145_equation_0, values = (var_56003_cast, var_56105_cast))[name = tensor<string, []>("op_56145_cast")];
            tensor<bool, []> input_695_interleave_0 = const()[name = tensor<string, []>("input_695_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_695_cast = concat(axis = var_30385, interleave = input_695_interleave_0, values = (var_56107_cast, var_56109_cast, var_56111_cast, var_56113_cast, var_56115_cast, var_56117_cast, var_56119_cast, var_56121_cast, var_56123_cast, var_56125_cast, var_56127_cast, var_56129_cast, var_56131_cast, var_56133_cast, var_56135_cast, var_56137_cast, var_56139_cast, var_56141_cast, var_56143_cast, var_56145_cast))[name = tensor<string, []>("input_695_cast")];
            tensor<int32, [2]> var_56151 = const()[name = tensor<string, []>("op_56151"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_56153 = const()[name = tensor<string, []>("op_56153"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_56155_pad_type_0 = const()[name = tensor<string, []>("op_56155_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_56155_pad_0 = const()[name = tensor<string, []>("op_56155_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4685565056)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4688841920)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_56155_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_out_0_bias_to_fp16, dilations = var_56153, groups = var_30385, pad = var_56155_pad_0, pad_type = var_56155_pad_type_0, strides = var_56151, weight = up_blocks_0_attentions_2_transformer_blocks_7_attn2_to_out_0_weight_to_fp16, x = input_695_cast)[name = tensor<string, []>("op_56155_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_371_cast = add(x = var_56155_cast, y = inputs_369_cast)[name = tensor<string, []>("inputs_371_cast")];
            tensor<int32, [1]> var_56159 = const()[name = tensor<string, []>("op_56159"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_371_cast = reduce_mean(axes = var_56159, keep_dims = var_30380, x = inputs_371_cast)[name = tensor<string, []>("channels_mean_371_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_371_cast = sub(x = inputs_371_cast, y = channels_mean_371_cast)[name = tensor<string, []>("zero_mean_371_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_371_cast = mul(x = zero_mean_371_cast, y = zero_mean_371_cast)[name = tensor<string, []>("zero_mean_sq_371_cast")];
            tensor<int32, [1]> var_56163 = const()[name = tensor<string, []>("op_56163"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_56164_cast = reduce_mean(axes = var_56163, keep_dims = var_30380, x = zero_mean_sq_371_cast)[name = tensor<string, []>("op_56164_cast")];
            tensor<fp16, []> var_56165_to_fp16 = const()[name = tensor<string, []>("op_56165_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_56166_cast = add(x = var_56164_cast, y = var_56165_to_fp16)[name = tensor<string, []>("op_56166_cast")];
            tensor<fp16, []> denom_371_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_371_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_371_cast = rsqrt(epsilon = denom_371_epsilon_0_to_fp16, x = var_56166_cast)[name = tensor<string, []>("denom_371_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_371_cast = mul(x = zero_mean_371_cast, y = denom_371_cast)[name = tensor<string, []>("out_371_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_56170_to_fp16 = const()[name = tensor<string, []>("op_56170_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4688844544)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_56171_cast = add(x = out_371_cast, y = var_56170_to_fp16)[name = tensor<string, []>("op_56171_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_56173_to_fp16 = const()[name = tensor<string, []>("op_56173_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4688847168)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_697_cast = mul(x = var_56171_cast, y = var_56173_to_fp16)[name = tensor<string, []>("input_697_cast")];
            tensor<int32, [2]> var_56181 = const()[name = tensor<string, []>("op_56181"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_56183 = const()[name = tensor<string, []>("op_56183"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_56185_pad_type_0 = const()[name = tensor<string, []>("op_56185_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_56185_pad_0 = const()[name = tensor<string, []>("op_56185_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4688849792)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_2_transformer_blocks_7_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4715064256)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_56185_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_7_ff_net_0_proj_bias_to_fp16, dilations = var_56183, groups = var_30385, pad = var_56185_pad_0, pad_type = var_56185_pad_type_0, strides = var_56181, weight = up_blocks_0_attentions_2_transformer_blocks_7_ff_net_0_proj_weight_to_fp16, x = input_697_cast)[name = tensor<string, []>("op_56185_cast")];
            tensor<int32, [2]> var_56186_split_sizes_0 = const()[name = tensor<string, []>("op_56186_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_56186_axis_0 = const()[name = tensor<string, []>("op_56186_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_56186_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_56186_cast_1 = split(axis = var_56186_axis_0, split_sizes = var_56186_split_sizes_0, x = var_56185_cast)[name = tensor<string, []>("op_56186_cast")];
            tensor<string, []> var_56188_mode_0 = const()[name = tensor<string, []>("op_56188_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_56188_cast = gelu(mode = var_56188_mode_0, x = var_56186_cast_1)[name = tensor<string, []>("op_56188_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_699_cast = mul(x = var_56186_cast_0, y = var_56188_cast)[name = tensor<string, []>("input_699_cast")];
            tensor<int32, [2]> var_56192 = const()[name = tensor<string, []>("op_56192"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_56194 = const()[name = tensor<string, []>("op_56194"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_56196_pad_type_0 = const()[name = tensor<string, []>("op_56196_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_56196_pad_0 = const()[name = tensor<string, []>("op_56196_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_7_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4715084800)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_7_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_7_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4728192064)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_56196_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_7_ff_net_2_bias_to_fp16, dilations = var_56194, groups = var_30385, pad = var_56196_pad_0, pad_type = var_56196_pad_type_0, strides = var_56192, weight = up_blocks_0_attentions_2_transformer_blocks_7_ff_net_2_weight_to_fp16, x = input_699_cast)[name = tensor<string, []>("op_56196_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_373_cast = add(x = var_56196_cast, y = inputs_371_cast)[name = tensor<string, []>("inputs_373_cast")];
            tensor<int32, [1]> var_56206 = const()[name = tensor<string, []>("op_56206"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_373_cast = reduce_mean(axes = var_56206, keep_dims = var_30380, x = inputs_373_cast)[name = tensor<string, []>("channels_mean_373_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_373_cast = sub(x = inputs_373_cast, y = channels_mean_373_cast)[name = tensor<string, []>("zero_mean_373_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_373_cast = mul(x = zero_mean_373_cast, y = zero_mean_373_cast)[name = tensor<string, []>("zero_mean_sq_373_cast")];
            tensor<int32, [1]> var_56210 = const()[name = tensor<string, []>("op_56210"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_56211_cast = reduce_mean(axes = var_56210, keep_dims = var_30380, x = zero_mean_sq_373_cast)[name = tensor<string, []>("op_56211_cast")];
            tensor<fp16, []> var_56212_to_fp16 = const()[name = tensor<string, []>("op_56212_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_56213_cast = add(x = var_56211_cast, y = var_56212_to_fp16)[name = tensor<string, []>("op_56213_cast")];
            tensor<fp16, []> denom_373_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_373_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_373_cast = rsqrt(epsilon = denom_373_epsilon_0_to_fp16, x = var_56213_cast)[name = tensor<string, []>("denom_373_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_373_cast = mul(x = zero_mean_373_cast, y = denom_373_cast)[name = tensor<string, []>("out_373_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_56217_to_fp16 = const()[name = tensor<string, []>("op_56217_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4728194688)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_56218_cast = add(x = out_373_cast, y = var_56217_to_fp16)[name = tensor<string, []>("op_56218_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_56220_to_fp16 = const()[name = tensor<string, []>("op_56220_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4728197312)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_483_cast = mul(x = var_56218_cast, y = var_56220_to_fp16)[name = tensor<string, []>("hidden_states_483_cast")];
            tensor<int32, [2]> var_56227 = const()[name = tensor<string, []>("op_56227"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_56229 = const()[name = tensor<string, []>("op_56229"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_249_pad_type_0 = const()[name = tensor<string, []>("q_249_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_249_pad_0 = const()[name = tensor<string, []>("q_249_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4728199936)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_249_cast = conv(dilations = var_56229, groups = var_30385, pad = q_249_pad_0, pad_type = q_249_pad_type_0, strides = var_56227, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_q_weight_to_fp16, x = hidden_states_483_cast)[name = tensor<string, []>("q_249_cast")];
            tensor<int32, [2]> var_56233 = const()[name = tensor<string, []>("op_56233"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_56235 = const()[name = tensor<string, []>("op_56235"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_497_pad_type_0 = const()[name = tensor<string, []>("k_497_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_497_pad_0 = const()[name = tensor<string, []>("k_497_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4731476800)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_497_cast = conv(dilations = var_56235, groups = var_30385, pad = k_497_pad_0, pad_type = k_497_pad_type_0, strides = var_56233, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_k_weight_to_fp16, x = hidden_states_483_cast)[name = tensor<string, []>("k_497_cast")];
            tensor<int32, [2]> var_56239 = const()[name = tensor<string, []>("op_56239"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_56241 = const()[name = tensor<string, []>("op_56241"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_249_pad_type_0 = const()[name = tensor<string, []>("v_249_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_249_pad_0 = const()[name = tensor<string, []>("v_249_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4734753664)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_249_cast = conv(dilations = var_56241, groups = var_30385, pad = v_249_pad_0, pad_type = v_249_pad_type_0, strides = var_56239, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_v_weight_to_fp16, x = hidden_states_483_cast)[name = tensor<string, []>("v_249_cast")];
            tensor<int32, [4]> var_56245_begin_0 = const()[name = tensor<string, []>("op_56245_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_56245_end_0 = const()[name = tensor<string, []>("op_56245_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_56245_end_mask_0 = const()[name = tensor<string, []>("op_56245_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56245_cast = slice_by_index(begin = var_56245_begin_0, end = var_56245_end_0, end_mask = var_56245_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56245_cast")];
            tensor<int32, [4]> var_56249_begin_0 = const()[name = tensor<string, []>("op_56249_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_56249_end_0 = const()[name = tensor<string, []>("op_56249_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_56249_end_mask_0 = const()[name = tensor<string, []>("op_56249_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56249_cast = slice_by_index(begin = var_56249_begin_0, end = var_56249_end_0, end_mask = var_56249_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56249_cast")];
            tensor<int32, [4]> var_56253_begin_0 = const()[name = tensor<string, []>("op_56253_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_56253_end_0 = const()[name = tensor<string, []>("op_56253_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_56253_end_mask_0 = const()[name = tensor<string, []>("op_56253_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56253_cast = slice_by_index(begin = var_56253_begin_0, end = var_56253_end_0, end_mask = var_56253_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56253_cast")];
            tensor<int32, [4]> var_56257_begin_0 = const()[name = tensor<string, []>("op_56257_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_56257_end_0 = const()[name = tensor<string, []>("op_56257_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_56257_end_mask_0 = const()[name = tensor<string, []>("op_56257_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56257_cast = slice_by_index(begin = var_56257_begin_0, end = var_56257_end_0, end_mask = var_56257_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56257_cast")];
            tensor<int32, [4]> var_56261_begin_0 = const()[name = tensor<string, []>("op_56261_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_56261_end_0 = const()[name = tensor<string, []>("op_56261_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_56261_end_mask_0 = const()[name = tensor<string, []>("op_56261_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56261_cast = slice_by_index(begin = var_56261_begin_0, end = var_56261_end_0, end_mask = var_56261_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56261_cast")];
            tensor<int32, [4]> var_56265_begin_0 = const()[name = tensor<string, []>("op_56265_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_56265_end_0 = const()[name = tensor<string, []>("op_56265_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_56265_end_mask_0 = const()[name = tensor<string, []>("op_56265_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56265_cast = slice_by_index(begin = var_56265_begin_0, end = var_56265_end_0, end_mask = var_56265_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56265_cast")];
            tensor<int32, [4]> var_56269_begin_0 = const()[name = tensor<string, []>("op_56269_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_56269_end_0 = const()[name = tensor<string, []>("op_56269_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_56269_end_mask_0 = const()[name = tensor<string, []>("op_56269_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56269_cast = slice_by_index(begin = var_56269_begin_0, end = var_56269_end_0, end_mask = var_56269_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56269_cast")];
            tensor<int32, [4]> var_56273_begin_0 = const()[name = tensor<string, []>("op_56273_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_56273_end_0 = const()[name = tensor<string, []>("op_56273_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_56273_end_mask_0 = const()[name = tensor<string, []>("op_56273_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56273_cast = slice_by_index(begin = var_56273_begin_0, end = var_56273_end_0, end_mask = var_56273_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56273_cast")];
            tensor<int32, [4]> var_56277_begin_0 = const()[name = tensor<string, []>("op_56277_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_56277_end_0 = const()[name = tensor<string, []>("op_56277_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_56277_end_mask_0 = const()[name = tensor<string, []>("op_56277_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56277_cast = slice_by_index(begin = var_56277_begin_0, end = var_56277_end_0, end_mask = var_56277_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56277_cast")];
            tensor<int32, [4]> var_56281_begin_0 = const()[name = tensor<string, []>("op_56281_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_56281_end_0 = const()[name = tensor<string, []>("op_56281_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_56281_end_mask_0 = const()[name = tensor<string, []>("op_56281_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56281_cast = slice_by_index(begin = var_56281_begin_0, end = var_56281_end_0, end_mask = var_56281_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56281_cast")];
            tensor<int32, [4]> var_56285_begin_0 = const()[name = tensor<string, []>("op_56285_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_56285_end_0 = const()[name = tensor<string, []>("op_56285_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_56285_end_mask_0 = const()[name = tensor<string, []>("op_56285_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56285_cast = slice_by_index(begin = var_56285_begin_0, end = var_56285_end_0, end_mask = var_56285_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56285_cast")];
            tensor<int32, [4]> var_56289_begin_0 = const()[name = tensor<string, []>("op_56289_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_56289_end_0 = const()[name = tensor<string, []>("op_56289_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_56289_end_mask_0 = const()[name = tensor<string, []>("op_56289_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56289_cast = slice_by_index(begin = var_56289_begin_0, end = var_56289_end_0, end_mask = var_56289_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56289_cast")];
            tensor<int32, [4]> var_56293_begin_0 = const()[name = tensor<string, []>("op_56293_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_56293_end_0 = const()[name = tensor<string, []>("op_56293_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_56293_end_mask_0 = const()[name = tensor<string, []>("op_56293_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56293_cast = slice_by_index(begin = var_56293_begin_0, end = var_56293_end_0, end_mask = var_56293_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56293_cast")];
            tensor<int32, [4]> var_56297_begin_0 = const()[name = tensor<string, []>("op_56297_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_56297_end_0 = const()[name = tensor<string, []>("op_56297_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_56297_end_mask_0 = const()[name = tensor<string, []>("op_56297_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56297_cast = slice_by_index(begin = var_56297_begin_0, end = var_56297_end_0, end_mask = var_56297_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56297_cast")];
            tensor<int32, [4]> var_56301_begin_0 = const()[name = tensor<string, []>("op_56301_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_56301_end_0 = const()[name = tensor<string, []>("op_56301_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_56301_end_mask_0 = const()[name = tensor<string, []>("op_56301_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56301_cast = slice_by_index(begin = var_56301_begin_0, end = var_56301_end_0, end_mask = var_56301_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56301_cast")];
            tensor<int32, [4]> var_56305_begin_0 = const()[name = tensor<string, []>("op_56305_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_56305_end_0 = const()[name = tensor<string, []>("op_56305_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_56305_end_mask_0 = const()[name = tensor<string, []>("op_56305_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56305_cast = slice_by_index(begin = var_56305_begin_0, end = var_56305_end_0, end_mask = var_56305_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56305_cast")];
            tensor<int32, [4]> var_56309_begin_0 = const()[name = tensor<string, []>("op_56309_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_56309_end_0 = const()[name = tensor<string, []>("op_56309_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_56309_end_mask_0 = const()[name = tensor<string, []>("op_56309_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56309_cast = slice_by_index(begin = var_56309_begin_0, end = var_56309_end_0, end_mask = var_56309_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56309_cast")];
            tensor<int32, [4]> var_56313_begin_0 = const()[name = tensor<string, []>("op_56313_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_56313_end_0 = const()[name = tensor<string, []>("op_56313_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_56313_end_mask_0 = const()[name = tensor<string, []>("op_56313_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56313_cast = slice_by_index(begin = var_56313_begin_0, end = var_56313_end_0, end_mask = var_56313_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56313_cast")];
            tensor<int32, [4]> var_56317_begin_0 = const()[name = tensor<string, []>("op_56317_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_56317_end_0 = const()[name = tensor<string, []>("op_56317_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_56317_end_mask_0 = const()[name = tensor<string, []>("op_56317_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56317_cast = slice_by_index(begin = var_56317_begin_0, end = var_56317_end_0, end_mask = var_56317_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56317_cast")];
            tensor<int32, [4]> var_56321_begin_0 = const()[name = tensor<string, []>("op_56321_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_56321_end_0 = const()[name = tensor<string, []>("op_56321_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_56321_end_mask_0 = const()[name = tensor<string, []>("op_56321_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56321_cast = slice_by_index(begin = var_56321_begin_0, end = var_56321_end_0, end_mask = var_56321_end_mask_0, x = q_249_cast)[name = tensor<string, []>("op_56321_cast")];
            tensor<int32, [4]> k_499_perm_0 = const()[name = tensor<string, []>("k_499_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_56328_begin_0 = const()[name = tensor<string, []>("op_56328_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_56328_end_0 = const()[name = tensor<string, []>("op_56328_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_56328_end_mask_0 = const()[name = tensor<string, []>("op_56328_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_15 = transpose(perm = k_499_perm_0, x = k_497_cast)[name = tensor<string, []>("transpose_15")];
            tensor<fp16, [2, 1024, 1, 64]> var_56328_cast = slice_by_index(begin = var_56328_begin_0, end = var_56328_end_0, end_mask = var_56328_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56328_cast")];
            tensor<int32, [4]> var_56332_begin_0 = const()[name = tensor<string, []>("op_56332_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_56332_end_0 = const()[name = tensor<string, []>("op_56332_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_56332_end_mask_0 = const()[name = tensor<string, []>("op_56332_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56332_cast = slice_by_index(begin = var_56332_begin_0, end = var_56332_end_0, end_mask = var_56332_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56332_cast")];
            tensor<int32, [4]> var_56336_begin_0 = const()[name = tensor<string, []>("op_56336_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_56336_end_0 = const()[name = tensor<string, []>("op_56336_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_56336_end_mask_0 = const()[name = tensor<string, []>("op_56336_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56336_cast = slice_by_index(begin = var_56336_begin_0, end = var_56336_end_0, end_mask = var_56336_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56336_cast")];
            tensor<int32, [4]> var_56340_begin_0 = const()[name = tensor<string, []>("op_56340_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_56340_end_0 = const()[name = tensor<string, []>("op_56340_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_56340_end_mask_0 = const()[name = tensor<string, []>("op_56340_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56340_cast = slice_by_index(begin = var_56340_begin_0, end = var_56340_end_0, end_mask = var_56340_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56340_cast")];
            tensor<int32, [4]> var_56344_begin_0 = const()[name = tensor<string, []>("op_56344_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_56344_end_0 = const()[name = tensor<string, []>("op_56344_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_56344_end_mask_0 = const()[name = tensor<string, []>("op_56344_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56344_cast = slice_by_index(begin = var_56344_begin_0, end = var_56344_end_0, end_mask = var_56344_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56344_cast")];
            tensor<int32, [4]> var_56348_begin_0 = const()[name = tensor<string, []>("op_56348_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_56348_end_0 = const()[name = tensor<string, []>("op_56348_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_56348_end_mask_0 = const()[name = tensor<string, []>("op_56348_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56348_cast = slice_by_index(begin = var_56348_begin_0, end = var_56348_end_0, end_mask = var_56348_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56348_cast")];
            tensor<int32, [4]> var_56352_begin_0 = const()[name = tensor<string, []>("op_56352_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_56352_end_0 = const()[name = tensor<string, []>("op_56352_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_56352_end_mask_0 = const()[name = tensor<string, []>("op_56352_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56352_cast = slice_by_index(begin = var_56352_begin_0, end = var_56352_end_0, end_mask = var_56352_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56352_cast")];
            tensor<int32, [4]> var_56356_begin_0 = const()[name = tensor<string, []>("op_56356_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_56356_end_0 = const()[name = tensor<string, []>("op_56356_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_56356_end_mask_0 = const()[name = tensor<string, []>("op_56356_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56356_cast = slice_by_index(begin = var_56356_begin_0, end = var_56356_end_0, end_mask = var_56356_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56356_cast")];
            tensor<int32, [4]> var_56360_begin_0 = const()[name = tensor<string, []>("op_56360_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_56360_end_0 = const()[name = tensor<string, []>("op_56360_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_56360_end_mask_0 = const()[name = tensor<string, []>("op_56360_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56360_cast = slice_by_index(begin = var_56360_begin_0, end = var_56360_end_0, end_mask = var_56360_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56360_cast")];
            tensor<int32, [4]> var_56364_begin_0 = const()[name = tensor<string, []>("op_56364_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_56364_end_0 = const()[name = tensor<string, []>("op_56364_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_56364_end_mask_0 = const()[name = tensor<string, []>("op_56364_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56364_cast = slice_by_index(begin = var_56364_begin_0, end = var_56364_end_0, end_mask = var_56364_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56364_cast")];
            tensor<int32, [4]> var_56368_begin_0 = const()[name = tensor<string, []>("op_56368_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_56368_end_0 = const()[name = tensor<string, []>("op_56368_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_56368_end_mask_0 = const()[name = tensor<string, []>("op_56368_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56368_cast = slice_by_index(begin = var_56368_begin_0, end = var_56368_end_0, end_mask = var_56368_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56368_cast")];
            tensor<int32, [4]> var_56372_begin_0 = const()[name = tensor<string, []>("op_56372_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_56372_end_0 = const()[name = tensor<string, []>("op_56372_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_56372_end_mask_0 = const()[name = tensor<string, []>("op_56372_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56372_cast = slice_by_index(begin = var_56372_begin_0, end = var_56372_end_0, end_mask = var_56372_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56372_cast")];
            tensor<int32, [4]> var_56376_begin_0 = const()[name = tensor<string, []>("op_56376_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_56376_end_0 = const()[name = tensor<string, []>("op_56376_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_56376_end_mask_0 = const()[name = tensor<string, []>("op_56376_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56376_cast = slice_by_index(begin = var_56376_begin_0, end = var_56376_end_0, end_mask = var_56376_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56376_cast")];
            tensor<int32, [4]> var_56380_begin_0 = const()[name = tensor<string, []>("op_56380_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_56380_end_0 = const()[name = tensor<string, []>("op_56380_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_56380_end_mask_0 = const()[name = tensor<string, []>("op_56380_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56380_cast = slice_by_index(begin = var_56380_begin_0, end = var_56380_end_0, end_mask = var_56380_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56380_cast")];
            tensor<int32, [4]> var_56384_begin_0 = const()[name = tensor<string, []>("op_56384_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_56384_end_0 = const()[name = tensor<string, []>("op_56384_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_56384_end_mask_0 = const()[name = tensor<string, []>("op_56384_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56384_cast = slice_by_index(begin = var_56384_begin_0, end = var_56384_end_0, end_mask = var_56384_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56384_cast")];
            tensor<int32, [4]> var_56388_begin_0 = const()[name = tensor<string, []>("op_56388_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_56388_end_0 = const()[name = tensor<string, []>("op_56388_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_56388_end_mask_0 = const()[name = tensor<string, []>("op_56388_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56388_cast = slice_by_index(begin = var_56388_begin_0, end = var_56388_end_0, end_mask = var_56388_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56388_cast")];
            tensor<int32, [4]> var_56392_begin_0 = const()[name = tensor<string, []>("op_56392_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_56392_end_0 = const()[name = tensor<string, []>("op_56392_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_56392_end_mask_0 = const()[name = tensor<string, []>("op_56392_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56392_cast = slice_by_index(begin = var_56392_begin_0, end = var_56392_end_0, end_mask = var_56392_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56392_cast")];
            tensor<int32, [4]> var_56396_begin_0 = const()[name = tensor<string, []>("op_56396_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_56396_end_0 = const()[name = tensor<string, []>("op_56396_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_56396_end_mask_0 = const()[name = tensor<string, []>("op_56396_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56396_cast = slice_by_index(begin = var_56396_begin_0, end = var_56396_end_0, end_mask = var_56396_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56396_cast")];
            tensor<int32, [4]> var_56400_begin_0 = const()[name = tensor<string, []>("op_56400_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_56400_end_0 = const()[name = tensor<string, []>("op_56400_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_56400_end_mask_0 = const()[name = tensor<string, []>("op_56400_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56400_cast = slice_by_index(begin = var_56400_begin_0, end = var_56400_end_0, end_mask = var_56400_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56400_cast")];
            tensor<int32, [4]> var_56404_begin_0 = const()[name = tensor<string, []>("op_56404_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_56404_end_0 = const()[name = tensor<string, []>("op_56404_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_56404_end_mask_0 = const()[name = tensor<string, []>("op_56404_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_56404_cast = slice_by_index(begin = var_56404_begin_0, end = var_56404_end_0, end_mask = var_56404_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_56404_cast")];
            tensor<int32, [4]> var_56406_begin_0 = const()[name = tensor<string, []>("op_56406_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_56406_end_0 = const()[name = tensor<string, []>("op_56406_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_56406_end_mask_0 = const()[name = tensor<string, []>("op_56406_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56406_cast = slice_by_index(begin = var_56406_begin_0, end = var_56406_end_0, end_mask = var_56406_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56406_cast")];
            tensor<int32, [4]> var_56410_begin_0 = const()[name = tensor<string, []>("op_56410_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_56410_end_0 = const()[name = tensor<string, []>("op_56410_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_56410_end_mask_0 = const()[name = tensor<string, []>("op_56410_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56410_cast = slice_by_index(begin = var_56410_begin_0, end = var_56410_end_0, end_mask = var_56410_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56410_cast")];
            tensor<int32, [4]> var_56414_begin_0 = const()[name = tensor<string, []>("op_56414_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_56414_end_0 = const()[name = tensor<string, []>("op_56414_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_56414_end_mask_0 = const()[name = tensor<string, []>("op_56414_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56414_cast = slice_by_index(begin = var_56414_begin_0, end = var_56414_end_0, end_mask = var_56414_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56414_cast")];
            tensor<int32, [4]> var_56418_begin_0 = const()[name = tensor<string, []>("op_56418_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_56418_end_0 = const()[name = tensor<string, []>("op_56418_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_56418_end_mask_0 = const()[name = tensor<string, []>("op_56418_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56418_cast = slice_by_index(begin = var_56418_begin_0, end = var_56418_end_0, end_mask = var_56418_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56418_cast")];
            tensor<int32, [4]> var_56422_begin_0 = const()[name = tensor<string, []>("op_56422_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_56422_end_0 = const()[name = tensor<string, []>("op_56422_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_56422_end_mask_0 = const()[name = tensor<string, []>("op_56422_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56422_cast = slice_by_index(begin = var_56422_begin_0, end = var_56422_end_0, end_mask = var_56422_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56422_cast")];
            tensor<int32, [4]> var_56426_begin_0 = const()[name = tensor<string, []>("op_56426_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_56426_end_0 = const()[name = tensor<string, []>("op_56426_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_56426_end_mask_0 = const()[name = tensor<string, []>("op_56426_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56426_cast = slice_by_index(begin = var_56426_begin_0, end = var_56426_end_0, end_mask = var_56426_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56426_cast")];
            tensor<int32, [4]> var_56430_begin_0 = const()[name = tensor<string, []>("op_56430_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_56430_end_0 = const()[name = tensor<string, []>("op_56430_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_56430_end_mask_0 = const()[name = tensor<string, []>("op_56430_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56430_cast = slice_by_index(begin = var_56430_begin_0, end = var_56430_end_0, end_mask = var_56430_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56430_cast")];
            tensor<int32, [4]> var_56434_begin_0 = const()[name = tensor<string, []>("op_56434_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_56434_end_0 = const()[name = tensor<string, []>("op_56434_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_56434_end_mask_0 = const()[name = tensor<string, []>("op_56434_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56434_cast = slice_by_index(begin = var_56434_begin_0, end = var_56434_end_0, end_mask = var_56434_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56434_cast")];
            tensor<int32, [4]> var_56438_begin_0 = const()[name = tensor<string, []>("op_56438_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_56438_end_0 = const()[name = tensor<string, []>("op_56438_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_56438_end_mask_0 = const()[name = tensor<string, []>("op_56438_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56438_cast = slice_by_index(begin = var_56438_begin_0, end = var_56438_end_0, end_mask = var_56438_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56438_cast")];
            tensor<int32, [4]> var_56442_begin_0 = const()[name = tensor<string, []>("op_56442_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_56442_end_0 = const()[name = tensor<string, []>("op_56442_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_56442_end_mask_0 = const()[name = tensor<string, []>("op_56442_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56442_cast = slice_by_index(begin = var_56442_begin_0, end = var_56442_end_0, end_mask = var_56442_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56442_cast")];
            tensor<int32, [4]> var_56446_begin_0 = const()[name = tensor<string, []>("op_56446_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_56446_end_0 = const()[name = tensor<string, []>("op_56446_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_56446_end_mask_0 = const()[name = tensor<string, []>("op_56446_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56446_cast = slice_by_index(begin = var_56446_begin_0, end = var_56446_end_0, end_mask = var_56446_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56446_cast")];
            tensor<int32, [4]> var_56450_begin_0 = const()[name = tensor<string, []>("op_56450_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_56450_end_0 = const()[name = tensor<string, []>("op_56450_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_56450_end_mask_0 = const()[name = tensor<string, []>("op_56450_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56450_cast = slice_by_index(begin = var_56450_begin_0, end = var_56450_end_0, end_mask = var_56450_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56450_cast")];
            tensor<int32, [4]> var_56454_begin_0 = const()[name = tensor<string, []>("op_56454_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_56454_end_0 = const()[name = tensor<string, []>("op_56454_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_56454_end_mask_0 = const()[name = tensor<string, []>("op_56454_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56454_cast = slice_by_index(begin = var_56454_begin_0, end = var_56454_end_0, end_mask = var_56454_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56454_cast")];
            tensor<int32, [4]> var_56458_begin_0 = const()[name = tensor<string, []>("op_56458_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_56458_end_0 = const()[name = tensor<string, []>("op_56458_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_56458_end_mask_0 = const()[name = tensor<string, []>("op_56458_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56458_cast = slice_by_index(begin = var_56458_begin_0, end = var_56458_end_0, end_mask = var_56458_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56458_cast")];
            tensor<int32, [4]> var_56462_begin_0 = const()[name = tensor<string, []>("op_56462_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_56462_end_0 = const()[name = tensor<string, []>("op_56462_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_56462_end_mask_0 = const()[name = tensor<string, []>("op_56462_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56462_cast = slice_by_index(begin = var_56462_begin_0, end = var_56462_end_0, end_mask = var_56462_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56462_cast")];
            tensor<int32, [4]> var_56466_begin_0 = const()[name = tensor<string, []>("op_56466_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_56466_end_0 = const()[name = tensor<string, []>("op_56466_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_56466_end_mask_0 = const()[name = tensor<string, []>("op_56466_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56466_cast = slice_by_index(begin = var_56466_begin_0, end = var_56466_end_0, end_mask = var_56466_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56466_cast")];
            tensor<int32, [4]> var_56470_begin_0 = const()[name = tensor<string, []>("op_56470_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_56470_end_0 = const()[name = tensor<string, []>("op_56470_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_56470_end_mask_0 = const()[name = tensor<string, []>("op_56470_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56470_cast = slice_by_index(begin = var_56470_begin_0, end = var_56470_end_0, end_mask = var_56470_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56470_cast")];
            tensor<int32, [4]> var_56474_begin_0 = const()[name = tensor<string, []>("op_56474_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_56474_end_0 = const()[name = tensor<string, []>("op_56474_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_56474_end_mask_0 = const()[name = tensor<string, []>("op_56474_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56474_cast = slice_by_index(begin = var_56474_begin_0, end = var_56474_end_0, end_mask = var_56474_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56474_cast")];
            tensor<int32, [4]> var_56478_begin_0 = const()[name = tensor<string, []>("op_56478_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_56478_end_0 = const()[name = tensor<string, []>("op_56478_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_56478_end_mask_0 = const()[name = tensor<string, []>("op_56478_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56478_cast = slice_by_index(begin = var_56478_begin_0, end = var_56478_end_0, end_mask = var_56478_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56478_cast")];
            tensor<int32, [4]> var_56482_begin_0 = const()[name = tensor<string, []>("op_56482_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_56482_end_0 = const()[name = tensor<string, []>("op_56482_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_56482_end_mask_0 = const()[name = tensor<string, []>("op_56482_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56482_cast = slice_by_index(begin = var_56482_begin_0, end = var_56482_end_0, end_mask = var_56482_end_mask_0, x = v_249_cast)[name = tensor<string, []>("op_56482_cast")];
            tensor<string, []> var_56486_equation_0 = const()[name = tensor<string, []>("op_56486_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56486_cast = einsum(equation = var_56486_equation_0, values = (var_56328_cast, var_56245_cast))[name = tensor<string, []>("op_56486_cast")];
            tensor<fp16, []> var_56487_to_fp16 = const()[name = tensor<string, []>("op_56487_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4801_cast = mul(x = var_56486_cast, y = var_56487_to_fp16)[name = tensor<string, []>("aw_4801_cast")];
            tensor<string, []> var_56490_equation_0 = const()[name = tensor<string, []>("op_56490_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56490_cast = einsum(equation = var_56490_equation_0, values = (var_56332_cast, var_56249_cast))[name = tensor<string, []>("op_56490_cast")];
            tensor<fp16, []> var_56491_to_fp16 = const()[name = tensor<string, []>("op_56491_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4803_cast = mul(x = var_56490_cast, y = var_56491_to_fp16)[name = tensor<string, []>("aw_4803_cast")];
            tensor<string, []> var_56494_equation_0 = const()[name = tensor<string, []>("op_56494_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56494_cast = einsum(equation = var_56494_equation_0, values = (var_56336_cast, var_56253_cast))[name = tensor<string, []>("op_56494_cast")];
            tensor<fp16, []> var_56495_to_fp16 = const()[name = tensor<string, []>("op_56495_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4805_cast = mul(x = var_56494_cast, y = var_56495_to_fp16)[name = tensor<string, []>("aw_4805_cast")];
            tensor<string, []> var_56498_equation_0 = const()[name = tensor<string, []>("op_56498_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56498_cast = einsum(equation = var_56498_equation_0, values = (var_56340_cast, var_56257_cast))[name = tensor<string, []>("op_56498_cast")];
            tensor<fp16, []> var_56499_to_fp16 = const()[name = tensor<string, []>("op_56499_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4807_cast = mul(x = var_56498_cast, y = var_56499_to_fp16)[name = tensor<string, []>("aw_4807_cast")];
            tensor<string, []> var_56502_equation_0 = const()[name = tensor<string, []>("op_56502_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56502_cast = einsum(equation = var_56502_equation_0, values = (var_56344_cast, var_56261_cast))[name = tensor<string, []>("op_56502_cast")];
            tensor<fp16, []> var_56503_to_fp16 = const()[name = tensor<string, []>("op_56503_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4809_cast = mul(x = var_56502_cast, y = var_56503_to_fp16)[name = tensor<string, []>("aw_4809_cast")];
            tensor<string, []> var_56506_equation_0 = const()[name = tensor<string, []>("op_56506_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56506_cast = einsum(equation = var_56506_equation_0, values = (var_56348_cast, var_56265_cast))[name = tensor<string, []>("op_56506_cast")];
            tensor<fp16, []> var_56507_to_fp16 = const()[name = tensor<string, []>("op_56507_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4811_cast = mul(x = var_56506_cast, y = var_56507_to_fp16)[name = tensor<string, []>("aw_4811_cast")];
            tensor<string, []> var_56510_equation_0 = const()[name = tensor<string, []>("op_56510_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56510_cast = einsum(equation = var_56510_equation_0, values = (var_56352_cast, var_56269_cast))[name = tensor<string, []>("op_56510_cast")];
            tensor<fp16, []> var_56511_to_fp16 = const()[name = tensor<string, []>("op_56511_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4813_cast = mul(x = var_56510_cast, y = var_56511_to_fp16)[name = tensor<string, []>("aw_4813_cast")];
            tensor<string, []> var_56514_equation_0 = const()[name = tensor<string, []>("op_56514_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56514_cast = einsum(equation = var_56514_equation_0, values = (var_56356_cast, var_56273_cast))[name = tensor<string, []>("op_56514_cast")];
            tensor<fp16, []> var_56515_to_fp16 = const()[name = tensor<string, []>("op_56515_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4815_cast = mul(x = var_56514_cast, y = var_56515_to_fp16)[name = tensor<string, []>("aw_4815_cast")];
            tensor<string, []> var_56518_equation_0 = const()[name = tensor<string, []>("op_56518_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56518_cast = einsum(equation = var_56518_equation_0, values = (var_56360_cast, var_56277_cast))[name = tensor<string, []>("op_56518_cast")];
            tensor<fp16, []> var_56519_to_fp16 = const()[name = tensor<string, []>("op_56519_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4817_cast = mul(x = var_56518_cast, y = var_56519_to_fp16)[name = tensor<string, []>("aw_4817_cast")];
            tensor<string, []> var_56522_equation_0 = const()[name = tensor<string, []>("op_56522_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56522_cast = einsum(equation = var_56522_equation_0, values = (var_56364_cast, var_56281_cast))[name = tensor<string, []>("op_56522_cast")];
            tensor<fp16, []> var_56523_to_fp16 = const()[name = tensor<string, []>("op_56523_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4819_cast = mul(x = var_56522_cast, y = var_56523_to_fp16)[name = tensor<string, []>("aw_4819_cast")];
            tensor<string, []> var_56526_equation_0 = const()[name = tensor<string, []>("op_56526_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56526_cast = einsum(equation = var_56526_equation_0, values = (var_56368_cast, var_56285_cast))[name = tensor<string, []>("op_56526_cast")];
            tensor<fp16, []> var_56527_to_fp16 = const()[name = tensor<string, []>("op_56527_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4821_cast = mul(x = var_56526_cast, y = var_56527_to_fp16)[name = tensor<string, []>("aw_4821_cast")];
            tensor<string, []> var_56530_equation_0 = const()[name = tensor<string, []>("op_56530_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56530_cast = einsum(equation = var_56530_equation_0, values = (var_56372_cast, var_56289_cast))[name = tensor<string, []>("op_56530_cast")];
            tensor<fp16, []> var_56531_to_fp16 = const()[name = tensor<string, []>("op_56531_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4823_cast = mul(x = var_56530_cast, y = var_56531_to_fp16)[name = tensor<string, []>("aw_4823_cast")];
            tensor<string, []> var_56534_equation_0 = const()[name = tensor<string, []>("op_56534_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56534_cast = einsum(equation = var_56534_equation_0, values = (var_56376_cast, var_56293_cast))[name = tensor<string, []>("op_56534_cast")];
            tensor<fp16, []> var_56535_to_fp16 = const()[name = tensor<string, []>("op_56535_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4825_cast = mul(x = var_56534_cast, y = var_56535_to_fp16)[name = tensor<string, []>("aw_4825_cast")];
            tensor<string, []> var_56538_equation_0 = const()[name = tensor<string, []>("op_56538_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56538_cast = einsum(equation = var_56538_equation_0, values = (var_56380_cast, var_56297_cast))[name = tensor<string, []>("op_56538_cast")];
            tensor<fp16, []> var_56539_to_fp16 = const()[name = tensor<string, []>("op_56539_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4827_cast = mul(x = var_56538_cast, y = var_56539_to_fp16)[name = tensor<string, []>("aw_4827_cast")];
            tensor<string, []> var_56542_equation_0 = const()[name = tensor<string, []>("op_56542_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56542_cast = einsum(equation = var_56542_equation_0, values = (var_56384_cast, var_56301_cast))[name = tensor<string, []>("op_56542_cast")];
            tensor<fp16, []> var_56543_to_fp16 = const()[name = tensor<string, []>("op_56543_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4829_cast = mul(x = var_56542_cast, y = var_56543_to_fp16)[name = tensor<string, []>("aw_4829_cast")];
            tensor<string, []> var_56546_equation_0 = const()[name = tensor<string, []>("op_56546_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56546_cast = einsum(equation = var_56546_equation_0, values = (var_56388_cast, var_56305_cast))[name = tensor<string, []>("op_56546_cast")];
            tensor<fp16, []> var_56547_to_fp16 = const()[name = tensor<string, []>("op_56547_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4831_cast = mul(x = var_56546_cast, y = var_56547_to_fp16)[name = tensor<string, []>("aw_4831_cast")];
            tensor<string, []> var_56550_equation_0 = const()[name = tensor<string, []>("op_56550_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56550_cast = einsum(equation = var_56550_equation_0, values = (var_56392_cast, var_56309_cast))[name = tensor<string, []>("op_56550_cast")];
            tensor<fp16, []> var_56551_to_fp16 = const()[name = tensor<string, []>("op_56551_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4833_cast = mul(x = var_56550_cast, y = var_56551_to_fp16)[name = tensor<string, []>("aw_4833_cast")];
            tensor<string, []> var_56554_equation_0 = const()[name = tensor<string, []>("op_56554_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56554_cast = einsum(equation = var_56554_equation_0, values = (var_56396_cast, var_56313_cast))[name = tensor<string, []>("op_56554_cast")];
            tensor<fp16, []> var_56555_to_fp16 = const()[name = tensor<string, []>("op_56555_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4835_cast = mul(x = var_56554_cast, y = var_56555_to_fp16)[name = tensor<string, []>("aw_4835_cast")];
            tensor<string, []> var_56558_equation_0 = const()[name = tensor<string, []>("op_56558_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56558_cast = einsum(equation = var_56558_equation_0, values = (var_56400_cast, var_56317_cast))[name = tensor<string, []>("op_56558_cast")];
            tensor<fp16, []> var_56559_to_fp16 = const()[name = tensor<string, []>("op_56559_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4837_cast = mul(x = var_56558_cast, y = var_56559_to_fp16)[name = tensor<string, []>("aw_4837_cast")];
            tensor<string, []> var_56562_equation_0 = const()[name = tensor<string, []>("op_56562_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56562_cast = einsum(equation = var_56562_equation_0, values = (var_56404_cast, var_56321_cast))[name = tensor<string, []>("op_56562_cast")];
            tensor<fp16, []> var_56563_to_fp16 = const()[name = tensor<string, []>("op_56563_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4839_cast = mul(x = var_56562_cast, y = var_56563_to_fp16)[name = tensor<string, []>("aw_4839_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56565_cast = softmax(axis = var_30385, x = aw_4801_cast)[name = tensor<string, []>("op_56565_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56566_cast = softmax(axis = var_30385, x = aw_4803_cast)[name = tensor<string, []>("op_56566_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56567_cast = softmax(axis = var_30385, x = aw_4805_cast)[name = tensor<string, []>("op_56567_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56568_cast = softmax(axis = var_30385, x = aw_4807_cast)[name = tensor<string, []>("op_56568_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56569_cast = softmax(axis = var_30385, x = aw_4809_cast)[name = tensor<string, []>("op_56569_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56570_cast = softmax(axis = var_30385, x = aw_4811_cast)[name = tensor<string, []>("op_56570_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56571_cast = softmax(axis = var_30385, x = aw_4813_cast)[name = tensor<string, []>("op_56571_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56572_cast = softmax(axis = var_30385, x = aw_4815_cast)[name = tensor<string, []>("op_56572_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56573_cast = softmax(axis = var_30385, x = aw_4817_cast)[name = tensor<string, []>("op_56573_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56574_cast = softmax(axis = var_30385, x = aw_4819_cast)[name = tensor<string, []>("op_56574_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56575_cast = softmax(axis = var_30385, x = aw_4821_cast)[name = tensor<string, []>("op_56575_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56576_cast = softmax(axis = var_30385, x = aw_4823_cast)[name = tensor<string, []>("op_56576_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56577_cast = softmax(axis = var_30385, x = aw_4825_cast)[name = tensor<string, []>("op_56577_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56578_cast = softmax(axis = var_30385, x = aw_4827_cast)[name = tensor<string, []>("op_56578_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56579_cast = softmax(axis = var_30385, x = aw_4829_cast)[name = tensor<string, []>("op_56579_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56580_cast = softmax(axis = var_30385, x = aw_4831_cast)[name = tensor<string, []>("op_56580_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56581_cast = softmax(axis = var_30385, x = aw_4833_cast)[name = tensor<string, []>("op_56581_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56582_cast = softmax(axis = var_30385, x = aw_4835_cast)[name = tensor<string, []>("op_56582_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56583_cast = softmax(axis = var_30385, x = aw_4837_cast)[name = tensor<string, []>("op_56583_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_56584_cast = softmax(axis = var_30385, x = aw_4839_cast)[name = tensor<string, []>("op_56584_cast")];
            tensor<string, []> var_56586_equation_0 = const()[name = tensor<string, []>("op_56586_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56586_cast = einsum(equation = var_56586_equation_0, values = (var_56406_cast, var_56565_cast))[name = tensor<string, []>("op_56586_cast")];
            tensor<string, []> var_56588_equation_0 = const()[name = tensor<string, []>("op_56588_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56588_cast = einsum(equation = var_56588_equation_0, values = (var_56410_cast, var_56566_cast))[name = tensor<string, []>("op_56588_cast")];
            tensor<string, []> var_56590_equation_0 = const()[name = tensor<string, []>("op_56590_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56590_cast = einsum(equation = var_56590_equation_0, values = (var_56414_cast, var_56567_cast))[name = tensor<string, []>("op_56590_cast")];
            tensor<string, []> var_56592_equation_0 = const()[name = tensor<string, []>("op_56592_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56592_cast = einsum(equation = var_56592_equation_0, values = (var_56418_cast, var_56568_cast))[name = tensor<string, []>("op_56592_cast")];
            tensor<string, []> var_56594_equation_0 = const()[name = tensor<string, []>("op_56594_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56594_cast = einsum(equation = var_56594_equation_0, values = (var_56422_cast, var_56569_cast))[name = tensor<string, []>("op_56594_cast")];
            tensor<string, []> var_56596_equation_0 = const()[name = tensor<string, []>("op_56596_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56596_cast = einsum(equation = var_56596_equation_0, values = (var_56426_cast, var_56570_cast))[name = tensor<string, []>("op_56596_cast")];
            tensor<string, []> var_56598_equation_0 = const()[name = tensor<string, []>("op_56598_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56598_cast = einsum(equation = var_56598_equation_0, values = (var_56430_cast, var_56571_cast))[name = tensor<string, []>("op_56598_cast")];
            tensor<string, []> var_56600_equation_0 = const()[name = tensor<string, []>("op_56600_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56600_cast = einsum(equation = var_56600_equation_0, values = (var_56434_cast, var_56572_cast))[name = tensor<string, []>("op_56600_cast")];
            tensor<string, []> var_56602_equation_0 = const()[name = tensor<string, []>("op_56602_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56602_cast = einsum(equation = var_56602_equation_0, values = (var_56438_cast, var_56573_cast))[name = tensor<string, []>("op_56602_cast")];
            tensor<string, []> var_56604_equation_0 = const()[name = tensor<string, []>("op_56604_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56604_cast = einsum(equation = var_56604_equation_0, values = (var_56442_cast, var_56574_cast))[name = tensor<string, []>("op_56604_cast")];
            tensor<string, []> var_56606_equation_0 = const()[name = tensor<string, []>("op_56606_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56606_cast = einsum(equation = var_56606_equation_0, values = (var_56446_cast, var_56575_cast))[name = tensor<string, []>("op_56606_cast")];
            tensor<string, []> var_56608_equation_0 = const()[name = tensor<string, []>("op_56608_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56608_cast = einsum(equation = var_56608_equation_0, values = (var_56450_cast, var_56576_cast))[name = tensor<string, []>("op_56608_cast")];
            tensor<string, []> var_56610_equation_0 = const()[name = tensor<string, []>("op_56610_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56610_cast = einsum(equation = var_56610_equation_0, values = (var_56454_cast, var_56577_cast))[name = tensor<string, []>("op_56610_cast")];
            tensor<string, []> var_56612_equation_0 = const()[name = tensor<string, []>("op_56612_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56612_cast = einsum(equation = var_56612_equation_0, values = (var_56458_cast, var_56578_cast))[name = tensor<string, []>("op_56612_cast")];
            tensor<string, []> var_56614_equation_0 = const()[name = tensor<string, []>("op_56614_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56614_cast = einsum(equation = var_56614_equation_0, values = (var_56462_cast, var_56579_cast))[name = tensor<string, []>("op_56614_cast")];
            tensor<string, []> var_56616_equation_0 = const()[name = tensor<string, []>("op_56616_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56616_cast = einsum(equation = var_56616_equation_0, values = (var_56466_cast, var_56580_cast))[name = tensor<string, []>("op_56616_cast")];
            tensor<string, []> var_56618_equation_0 = const()[name = tensor<string, []>("op_56618_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56618_cast = einsum(equation = var_56618_equation_0, values = (var_56470_cast, var_56581_cast))[name = tensor<string, []>("op_56618_cast")];
            tensor<string, []> var_56620_equation_0 = const()[name = tensor<string, []>("op_56620_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56620_cast = einsum(equation = var_56620_equation_0, values = (var_56474_cast, var_56582_cast))[name = tensor<string, []>("op_56620_cast")];
            tensor<string, []> var_56622_equation_0 = const()[name = tensor<string, []>("op_56622_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56622_cast = einsum(equation = var_56622_equation_0, values = (var_56478_cast, var_56583_cast))[name = tensor<string, []>("op_56622_cast")];
            tensor<string, []> var_56624_equation_0 = const()[name = tensor<string, []>("op_56624_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_56624_cast = einsum(equation = var_56624_equation_0, values = (var_56482_cast, var_56584_cast))[name = tensor<string, []>("op_56624_cast")];
            tensor<bool, []> input_701_interleave_0 = const()[name = tensor<string, []>("input_701_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_701_cast = concat(axis = var_30385, interleave = input_701_interleave_0, values = (var_56586_cast, var_56588_cast, var_56590_cast, var_56592_cast, var_56594_cast, var_56596_cast, var_56598_cast, var_56600_cast, var_56602_cast, var_56604_cast, var_56606_cast, var_56608_cast, var_56610_cast, var_56612_cast, var_56614_cast, var_56616_cast, var_56618_cast, var_56620_cast, var_56622_cast, var_56624_cast))[name = tensor<string, []>("input_701_cast")];
            tensor<int32, [2]> var_56630 = const()[name = tensor<string, []>("op_56630"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_56632 = const()[name = tensor<string, []>("op_56632"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_56634_pad_type_0 = const()[name = tensor<string, []>("op_56634_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_56634_pad_0 = const()[name = tensor<string, []>("op_56634_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4738030528)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4741307392)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_56634_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_out_0_bias_to_fp16, dilations = var_56632, groups = var_30385, pad = var_56634_pad_0, pad_type = var_56634_pad_type_0, strides = var_56630, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn1_to_out_0_weight_to_fp16, x = input_701_cast)[name = tensor<string, []>("op_56634_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_375_cast = add(x = var_56634_cast, y = inputs_373_cast)[name = tensor<string, []>("inputs_375_cast")];
            tensor<int32, [1]> var_56638 = const()[name = tensor<string, []>("op_56638"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_375_cast = reduce_mean(axes = var_56638, keep_dims = var_30380, x = inputs_375_cast)[name = tensor<string, []>("channels_mean_375_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_375_cast = sub(x = inputs_375_cast, y = channels_mean_375_cast)[name = tensor<string, []>("zero_mean_375_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_375_cast = mul(x = zero_mean_375_cast, y = zero_mean_375_cast)[name = tensor<string, []>("zero_mean_sq_375_cast")];
            tensor<int32, [1]> var_56642 = const()[name = tensor<string, []>("op_56642"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_56643_cast = reduce_mean(axes = var_56642, keep_dims = var_30380, x = zero_mean_sq_375_cast)[name = tensor<string, []>("op_56643_cast")];
            tensor<fp16, []> var_56644_to_fp16 = const()[name = tensor<string, []>("op_56644_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_56645_cast = add(x = var_56643_cast, y = var_56644_to_fp16)[name = tensor<string, []>("op_56645_cast")];
            tensor<fp16, []> denom_375_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_375_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_375_cast = rsqrt(epsilon = denom_375_epsilon_0_to_fp16, x = var_56645_cast)[name = tensor<string, []>("denom_375_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_375_cast = mul(x = zero_mean_375_cast, y = denom_375_cast)[name = tensor<string, []>("out_375_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_56649_to_fp16 = const()[name = tensor<string, []>("op_56649_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4741310016)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_56650_cast = add(x = out_375_cast, y = var_56649_to_fp16)[name = tensor<string, []>("op_56650_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_56652_to_fp16 = const()[name = tensor<string, []>("op_56652_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4741312640)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_485_cast = mul(x = var_56650_cast, y = var_56652_to_fp16)[name = tensor<string, []>("hidden_states_485_cast")];
            tensor<int32, [2]> var_56659 = const()[name = tensor<string, []>("op_56659"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_56661 = const()[name = tensor<string, []>("op_56661"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_251_pad_type_0 = const()[name = tensor<string, []>("q_251_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_251_pad_0 = const()[name = tensor<string, []>("q_251_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4741315264)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_251_cast = conv(dilations = var_56661, groups = var_30385, pad = q_251_pad_0, pad_type = q_251_pad_type_0, strides = var_56659, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_q_weight_to_fp16, x = hidden_states_485_cast)[name = tensor<string, []>("q_251_cast")];
            tensor<int32, [2]> var_56665 = const()[name = tensor<string, []>("op_56665"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_56667 = const()[name = tensor<string, []>("op_56667"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_501_pad_type_0 = const()[name = tensor<string, []>("k_501_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_501_pad_0 = const()[name = tensor<string, []>("k_501_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4744592128)))];
            tensor<fp16, [2, 1280, 1, 77]> k_501_cast = conv(dilations = var_56667, groups = var_30385, pad = k_501_pad_0, pad_type = k_501_pad_type_0, strides = var_56665, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_501_cast")];
            tensor<int32, [2]> var_56671 = const()[name = tensor<string, []>("op_56671"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_56673 = const()[name = tensor<string, []>("op_56673"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_251_pad_type_0 = const()[name = tensor<string, []>("v_251_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_251_pad_0 = const()[name = tensor<string, []>("v_251_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4749835072)))];
            tensor<fp16, [2, 1280, 1, 77]> v_251_cast = conv(dilations = var_56673, groups = var_30385, pad = v_251_pad_0, pad_type = v_251_pad_type_0, strides = var_56671, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_251_cast")];
            tensor<int32, [4]> var_56677_begin_0 = const()[name = tensor<string, []>("op_56677_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_56677_end_0 = const()[name = tensor<string, []>("op_56677_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_56677_end_mask_0 = const()[name = tensor<string, []>("op_56677_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56677_cast = slice_by_index(begin = var_56677_begin_0, end = var_56677_end_0, end_mask = var_56677_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56677_cast")];
            tensor<int32, [4]> var_56681_begin_0 = const()[name = tensor<string, []>("op_56681_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_56681_end_0 = const()[name = tensor<string, []>("op_56681_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_56681_end_mask_0 = const()[name = tensor<string, []>("op_56681_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56681_cast = slice_by_index(begin = var_56681_begin_0, end = var_56681_end_0, end_mask = var_56681_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56681_cast")];
            tensor<int32, [4]> var_56685_begin_0 = const()[name = tensor<string, []>("op_56685_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_56685_end_0 = const()[name = tensor<string, []>("op_56685_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_56685_end_mask_0 = const()[name = tensor<string, []>("op_56685_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56685_cast = slice_by_index(begin = var_56685_begin_0, end = var_56685_end_0, end_mask = var_56685_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56685_cast")];
            tensor<int32, [4]> var_56689_begin_0 = const()[name = tensor<string, []>("op_56689_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_56689_end_0 = const()[name = tensor<string, []>("op_56689_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_56689_end_mask_0 = const()[name = tensor<string, []>("op_56689_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56689_cast = slice_by_index(begin = var_56689_begin_0, end = var_56689_end_0, end_mask = var_56689_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56689_cast")];
            tensor<int32, [4]> var_56693_begin_0 = const()[name = tensor<string, []>("op_56693_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_56693_end_0 = const()[name = tensor<string, []>("op_56693_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_56693_end_mask_0 = const()[name = tensor<string, []>("op_56693_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56693_cast = slice_by_index(begin = var_56693_begin_0, end = var_56693_end_0, end_mask = var_56693_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56693_cast")];
            tensor<int32, [4]> var_56697_begin_0 = const()[name = tensor<string, []>("op_56697_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_56697_end_0 = const()[name = tensor<string, []>("op_56697_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_56697_end_mask_0 = const()[name = tensor<string, []>("op_56697_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56697_cast = slice_by_index(begin = var_56697_begin_0, end = var_56697_end_0, end_mask = var_56697_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56697_cast")];
            tensor<int32, [4]> var_56701_begin_0 = const()[name = tensor<string, []>("op_56701_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_56701_end_0 = const()[name = tensor<string, []>("op_56701_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_56701_end_mask_0 = const()[name = tensor<string, []>("op_56701_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56701_cast = slice_by_index(begin = var_56701_begin_0, end = var_56701_end_0, end_mask = var_56701_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56701_cast")];
            tensor<int32, [4]> var_56705_begin_0 = const()[name = tensor<string, []>("op_56705_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_56705_end_0 = const()[name = tensor<string, []>("op_56705_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_56705_end_mask_0 = const()[name = tensor<string, []>("op_56705_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56705_cast = slice_by_index(begin = var_56705_begin_0, end = var_56705_end_0, end_mask = var_56705_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56705_cast")];
            tensor<int32, [4]> var_56709_begin_0 = const()[name = tensor<string, []>("op_56709_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_56709_end_0 = const()[name = tensor<string, []>("op_56709_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_56709_end_mask_0 = const()[name = tensor<string, []>("op_56709_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56709_cast = slice_by_index(begin = var_56709_begin_0, end = var_56709_end_0, end_mask = var_56709_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56709_cast")];
            tensor<int32, [4]> var_56713_begin_0 = const()[name = tensor<string, []>("op_56713_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_56713_end_0 = const()[name = tensor<string, []>("op_56713_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_56713_end_mask_0 = const()[name = tensor<string, []>("op_56713_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56713_cast = slice_by_index(begin = var_56713_begin_0, end = var_56713_end_0, end_mask = var_56713_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56713_cast")];
            tensor<int32, [4]> var_56717_begin_0 = const()[name = tensor<string, []>("op_56717_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_56717_end_0 = const()[name = tensor<string, []>("op_56717_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_56717_end_mask_0 = const()[name = tensor<string, []>("op_56717_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56717_cast = slice_by_index(begin = var_56717_begin_0, end = var_56717_end_0, end_mask = var_56717_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56717_cast")];
            tensor<int32, [4]> var_56721_begin_0 = const()[name = tensor<string, []>("op_56721_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_56721_end_0 = const()[name = tensor<string, []>("op_56721_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_56721_end_mask_0 = const()[name = tensor<string, []>("op_56721_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56721_cast = slice_by_index(begin = var_56721_begin_0, end = var_56721_end_0, end_mask = var_56721_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56721_cast")];
            tensor<int32, [4]> var_56725_begin_0 = const()[name = tensor<string, []>("op_56725_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_56725_end_0 = const()[name = tensor<string, []>("op_56725_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_56725_end_mask_0 = const()[name = tensor<string, []>("op_56725_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56725_cast = slice_by_index(begin = var_56725_begin_0, end = var_56725_end_0, end_mask = var_56725_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56725_cast")];
            tensor<int32, [4]> var_56729_begin_0 = const()[name = tensor<string, []>("op_56729_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_56729_end_0 = const()[name = tensor<string, []>("op_56729_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_56729_end_mask_0 = const()[name = tensor<string, []>("op_56729_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56729_cast = slice_by_index(begin = var_56729_begin_0, end = var_56729_end_0, end_mask = var_56729_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56729_cast")];
            tensor<int32, [4]> var_56733_begin_0 = const()[name = tensor<string, []>("op_56733_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_56733_end_0 = const()[name = tensor<string, []>("op_56733_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_56733_end_mask_0 = const()[name = tensor<string, []>("op_56733_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56733_cast = slice_by_index(begin = var_56733_begin_0, end = var_56733_end_0, end_mask = var_56733_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56733_cast")];
            tensor<int32, [4]> var_56737_begin_0 = const()[name = tensor<string, []>("op_56737_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_56737_end_0 = const()[name = tensor<string, []>("op_56737_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_56737_end_mask_0 = const()[name = tensor<string, []>("op_56737_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56737_cast = slice_by_index(begin = var_56737_begin_0, end = var_56737_end_0, end_mask = var_56737_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56737_cast")];
            tensor<int32, [4]> var_56741_begin_0 = const()[name = tensor<string, []>("op_56741_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_56741_end_0 = const()[name = tensor<string, []>("op_56741_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_56741_end_mask_0 = const()[name = tensor<string, []>("op_56741_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56741_cast = slice_by_index(begin = var_56741_begin_0, end = var_56741_end_0, end_mask = var_56741_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56741_cast")];
            tensor<int32, [4]> var_56745_begin_0 = const()[name = tensor<string, []>("op_56745_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_56745_end_0 = const()[name = tensor<string, []>("op_56745_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_56745_end_mask_0 = const()[name = tensor<string, []>("op_56745_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56745_cast = slice_by_index(begin = var_56745_begin_0, end = var_56745_end_0, end_mask = var_56745_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56745_cast")];
            tensor<int32, [4]> var_56749_begin_0 = const()[name = tensor<string, []>("op_56749_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_56749_end_0 = const()[name = tensor<string, []>("op_56749_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_56749_end_mask_0 = const()[name = tensor<string, []>("op_56749_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56749_cast = slice_by_index(begin = var_56749_begin_0, end = var_56749_end_0, end_mask = var_56749_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56749_cast")];
            tensor<int32, [4]> var_56753_begin_0 = const()[name = tensor<string, []>("op_56753_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_56753_end_0 = const()[name = tensor<string, []>("op_56753_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_56753_end_mask_0 = const()[name = tensor<string, []>("op_56753_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_56753_cast = slice_by_index(begin = var_56753_begin_0, end = var_56753_end_0, end_mask = var_56753_end_mask_0, x = q_251_cast)[name = tensor<string, []>("op_56753_cast")];
            tensor<int32, [4]> k_503_perm_0 = const()[name = tensor<string, []>("k_503_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_56760_begin_0 = const()[name = tensor<string, []>("op_56760_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_56760_end_0 = const()[name = tensor<string, []>("op_56760_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_56760_end_mask_0 = const()[name = tensor<string, []>("op_56760_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_14 = transpose(perm = k_503_perm_0, x = k_501_cast)[name = tensor<string, []>("transpose_14")];
            tensor<fp16, [2, 77, 1, 64]> var_56760_cast = slice_by_index(begin = var_56760_begin_0, end = var_56760_end_0, end_mask = var_56760_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56760_cast")];
            tensor<int32, [4]> var_56764_begin_0 = const()[name = tensor<string, []>("op_56764_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_56764_end_0 = const()[name = tensor<string, []>("op_56764_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_56764_end_mask_0 = const()[name = tensor<string, []>("op_56764_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56764_cast = slice_by_index(begin = var_56764_begin_0, end = var_56764_end_0, end_mask = var_56764_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56764_cast")];
            tensor<int32, [4]> var_56768_begin_0 = const()[name = tensor<string, []>("op_56768_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_56768_end_0 = const()[name = tensor<string, []>("op_56768_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_56768_end_mask_0 = const()[name = tensor<string, []>("op_56768_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56768_cast = slice_by_index(begin = var_56768_begin_0, end = var_56768_end_0, end_mask = var_56768_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56768_cast")];
            tensor<int32, [4]> var_56772_begin_0 = const()[name = tensor<string, []>("op_56772_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_56772_end_0 = const()[name = tensor<string, []>("op_56772_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_56772_end_mask_0 = const()[name = tensor<string, []>("op_56772_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56772_cast = slice_by_index(begin = var_56772_begin_0, end = var_56772_end_0, end_mask = var_56772_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56772_cast")];
            tensor<int32, [4]> var_56776_begin_0 = const()[name = tensor<string, []>("op_56776_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_56776_end_0 = const()[name = tensor<string, []>("op_56776_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_56776_end_mask_0 = const()[name = tensor<string, []>("op_56776_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56776_cast = slice_by_index(begin = var_56776_begin_0, end = var_56776_end_0, end_mask = var_56776_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56776_cast")];
            tensor<int32, [4]> var_56780_begin_0 = const()[name = tensor<string, []>("op_56780_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_56780_end_0 = const()[name = tensor<string, []>("op_56780_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_56780_end_mask_0 = const()[name = tensor<string, []>("op_56780_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56780_cast = slice_by_index(begin = var_56780_begin_0, end = var_56780_end_0, end_mask = var_56780_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56780_cast")];
            tensor<int32, [4]> var_56784_begin_0 = const()[name = tensor<string, []>("op_56784_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_56784_end_0 = const()[name = tensor<string, []>("op_56784_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_56784_end_mask_0 = const()[name = tensor<string, []>("op_56784_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56784_cast = slice_by_index(begin = var_56784_begin_0, end = var_56784_end_0, end_mask = var_56784_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56784_cast")];
            tensor<int32, [4]> var_56788_begin_0 = const()[name = tensor<string, []>("op_56788_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_56788_end_0 = const()[name = tensor<string, []>("op_56788_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_56788_end_mask_0 = const()[name = tensor<string, []>("op_56788_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56788_cast = slice_by_index(begin = var_56788_begin_0, end = var_56788_end_0, end_mask = var_56788_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56788_cast")];
            tensor<int32, [4]> var_56792_begin_0 = const()[name = tensor<string, []>("op_56792_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_56792_end_0 = const()[name = tensor<string, []>("op_56792_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_56792_end_mask_0 = const()[name = tensor<string, []>("op_56792_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56792_cast = slice_by_index(begin = var_56792_begin_0, end = var_56792_end_0, end_mask = var_56792_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56792_cast")];
            tensor<int32, [4]> var_56796_begin_0 = const()[name = tensor<string, []>("op_56796_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_56796_end_0 = const()[name = tensor<string, []>("op_56796_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_56796_end_mask_0 = const()[name = tensor<string, []>("op_56796_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56796_cast = slice_by_index(begin = var_56796_begin_0, end = var_56796_end_0, end_mask = var_56796_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56796_cast")];
            tensor<int32, [4]> var_56800_begin_0 = const()[name = tensor<string, []>("op_56800_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_56800_end_0 = const()[name = tensor<string, []>("op_56800_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_56800_end_mask_0 = const()[name = tensor<string, []>("op_56800_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56800_cast = slice_by_index(begin = var_56800_begin_0, end = var_56800_end_0, end_mask = var_56800_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56800_cast")];
            tensor<int32, [4]> var_56804_begin_0 = const()[name = tensor<string, []>("op_56804_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_56804_end_0 = const()[name = tensor<string, []>("op_56804_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_56804_end_mask_0 = const()[name = tensor<string, []>("op_56804_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56804_cast = slice_by_index(begin = var_56804_begin_0, end = var_56804_end_0, end_mask = var_56804_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56804_cast")];
            tensor<int32, [4]> var_56808_begin_0 = const()[name = tensor<string, []>("op_56808_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_56808_end_0 = const()[name = tensor<string, []>("op_56808_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_56808_end_mask_0 = const()[name = tensor<string, []>("op_56808_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56808_cast = slice_by_index(begin = var_56808_begin_0, end = var_56808_end_0, end_mask = var_56808_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56808_cast")];
            tensor<int32, [4]> var_56812_begin_0 = const()[name = tensor<string, []>("op_56812_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_56812_end_0 = const()[name = tensor<string, []>("op_56812_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_56812_end_mask_0 = const()[name = tensor<string, []>("op_56812_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56812_cast = slice_by_index(begin = var_56812_begin_0, end = var_56812_end_0, end_mask = var_56812_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56812_cast")];
            tensor<int32, [4]> var_56816_begin_0 = const()[name = tensor<string, []>("op_56816_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_56816_end_0 = const()[name = tensor<string, []>("op_56816_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_56816_end_mask_0 = const()[name = tensor<string, []>("op_56816_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56816_cast = slice_by_index(begin = var_56816_begin_0, end = var_56816_end_0, end_mask = var_56816_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56816_cast")];
            tensor<int32, [4]> var_56820_begin_0 = const()[name = tensor<string, []>("op_56820_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_56820_end_0 = const()[name = tensor<string, []>("op_56820_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_56820_end_mask_0 = const()[name = tensor<string, []>("op_56820_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56820_cast = slice_by_index(begin = var_56820_begin_0, end = var_56820_end_0, end_mask = var_56820_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56820_cast")];
            tensor<int32, [4]> var_56824_begin_0 = const()[name = tensor<string, []>("op_56824_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_56824_end_0 = const()[name = tensor<string, []>("op_56824_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_56824_end_mask_0 = const()[name = tensor<string, []>("op_56824_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56824_cast = slice_by_index(begin = var_56824_begin_0, end = var_56824_end_0, end_mask = var_56824_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56824_cast")];
            tensor<int32, [4]> var_56828_begin_0 = const()[name = tensor<string, []>("op_56828_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_56828_end_0 = const()[name = tensor<string, []>("op_56828_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_56828_end_mask_0 = const()[name = tensor<string, []>("op_56828_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56828_cast = slice_by_index(begin = var_56828_begin_0, end = var_56828_end_0, end_mask = var_56828_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56828_cast")];
            tensor<int32, [4]> var_56832_begin_0 = const()[name = tensor<string, []>("op_56832_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_56832_end_0 = const()[name = tensor<string, []>("op_56832_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_56832_end_mask_0 = const()[name = tensor<string, []>("op_56832_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56832_cast = slice_by_index(begin = var_56832_begin_0, end = var_56832_end_0, end_mask = var_56832_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56832_cast")];
            tensor<int32, [4]> var_56836_begin_0 = const()[name = tensor<string, []>("op_56836_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_56836_end_0 = const()[name = tensor<string, []>("op_56836_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_56836_end_mask_0 = const()[name = tensor<string, []>("op_56836_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_56836_cast = slice_by_index(begin = var_56836_begin_0, end = var_56836_end_0, end_mask = var_56836_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_56836_cast")];
            tensor<int32, [4]> var_56838_begin_0 = const()[name = tensor<string, []>("op_56838_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_56838_end_0 = const()[name = tensor<string, []>("op_56838_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_56838_end_mask_0 = const()[name = tensor<string, []>("op_56838_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56838_cast = slice_by_index(begin = var_56838_begin_0, end = var_56838_end_0, end_mask = var_56838_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56838_cast")];
            tensor<int32, [4]> var_56842_begin_0 = const()[name = tensor<string, []>("op_56842_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_56842_end_0 = const()[name = tensor<string, []>("op_56842_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_56842_end_mask_0 = const()[name = tensor<string, []>("op_56842_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56842_cast = slice_by_index(begin = var_56842_begin_0, end = var_56842_end_0, end_mask = var_56842_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56842_cast")];
            tensor<int32, [4]> var_56846_begin_0 = const()[name = tensor<string, []>("op_56846_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_56846_end_0 = const()[name = tensor<string, []>("op_56846_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_56846_end_mask_0 = const()[name = tensor<string, []>("op_56846_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56846_cast = slice_by_index(begin = var_56846_begin_0, end = var_56846_end_0, end_mask = var_56846_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56846_cast")];
            tensor<int32, [4]> var_56850_begin_0 = const()[name = tensor<string, []>("op_56850_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_56850_end_0 = const()[name = tensor<string, []>("op_56850_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_56850_end_mask_0 = const()[name = tensor<string, []>("op_56850_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56850_cast = slice_by_index(begin = var_56850_begin_0, end = var_56850_end_0, end_mask = var_56850_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56850_cast")];
            tensor<int32, [4]> var_56854_begin_0 = const()[name = tensor<string, []>("op_56854_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_56854_end_0 = const()[name = tensor<string, []>("op_56854_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_56854_end_mask_0 = const()[name = tensor<string, []>("op_56854_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56854_cast = slice_by_index(begin = var_56854_begin_0, end = var_56854_end_0, end_mask = var_56854_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56854_cast")];
            tensor<int32, [4]> var_56858_begin_0 = const()[name = tensor<string, []>("op_56858_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_56858_end_0 = const()[name = tensor<string, []>("op_56858_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_56858_end_mask_0 = const()[name = tensor<string, []>("op_56858_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56858_cast = slice_by_index(begin = var_56858_begin_0, end = var_56858_end_0, end_mask = var_56858_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56858_cast")];
            tensor<int32, [4]> var_56862_begin_0 = const()[name = tensor<string, []>("op_56862_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_56862_end_0 = const()[name = tensor<string, []>("op_56862_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_56862_end_mask_0 = const()[name = tensor<string, []>("op_56862_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56862_cast = slice_by_index(begin = var_56862_begin_0, end = var_56862_end_0, end_mask = var_56862_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56862_cast")];
            tensor<int32, [4]> var_56866_begin_0 = const()[name = tensor<string, []>("op_56866_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_56866_end_0 = const()[name = tensor<string, []>("op_56866_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_56866_end_mask_0 = const()[name = tensor<string, []>("op_56866_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56866_cast = slice_by_index(begin = var_56866_begin_0, end = var_56866_end_0, end_mask = var_56866_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56866_cast")];
            tensor<int32, [4]> var_56870_begin_0 = const()[name = tensor<string, []>("op_56870_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_56870_end_0 = const()[name = tensor<string, []>("op_56870_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_56870_end_mask_0 = const()[name = tensor<string, []>("op_56870_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56870_cast = slice_by_index(begin = var_56870_begin_0, end = var_56870_end_0, end_mask = var_56870_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56870_cast")];
            tensor<int32, [4]> var_56874_begin_0 = const()[name = tensor<string, []>("op_56874_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_56874_end_0 = const()[name = tensor<string, []>("op_56874_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_56874_end_mask_0 = const()[name = tensor<string, []>("op_56874_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56874_cast = slice_by_index(begin = var_56874_begin_0, end = var_56874_end_0, end_mask = var_56874_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56874_cast")];
            tensor<int32, [4]> var_56878_begin_0 = const()[name = tensor<string, []>("op_56878_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_56878_end_0 = const()[name = tensor<string, []>("op_56878_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_56878_end_mask_0 = const()[name = tensor<string, []>("op_56878_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56878_cast = slice_by_index(begin = var_56878_begin_0, end = var_56878_end_0, end_mask = var_56878_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56878_cast")];
            tensor<int32, [4]> var_56882_begin_0 = const()[name = tensor<string, []>("op_56882_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_56882_end_0 = const()[name = tensor<string, []>("op_56882_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_56882_end_mask_0 = const()[name = tensor<string, []>("op_56882_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56882_cast = slice_by_index(begin = var_56882_begin_0, end = var_56882_end_0, end_mask = var_56882_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56882_cast")];
            tensor<int32, [4]> var_56886_begin_0 = const()[name = tensor<string, []>("op_56886_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_56886_end_0 = const()[name = tensor<string, []>("op_56886_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_56886_end_mask_0 = const()[name = tensor<string, []>("op_56886_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56886_cast = slice_by_index(begin = var_56886_begin_0, end = var_56886_end_0, end_mask = var_56886_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56886_cast")];
            tensor<int32, [4]> var_56890_begin_0 = const()[name = tensor<string, []>("op_56890_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_56890_end_0 = const()[name = tensor<string, []>("op_56890_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_56890_end_mask_0 = const()[name = tensor<string, []>("op_56890_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56890_cast = slice_by_index(begin = var_56890_begin_0, end = var_56890_end_0, end_mask = var_56890_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56890_cast")];
            tensor<int32, [4]> var_56894_begin_0 = const()[name = tensor<string, []>("op_56894_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_56894_end_0 = const()[name = tensor<string, []>("op_56894_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_56894_end_mask_0 = const()[name = tensor<string, []>("op_56894_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56894_cast = slice_by_index(begin = var_56894_begin_0, end = var_56894_end_0, end_mask = var_56894_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56894_cast")];
            tensor<int32, [4]> var_56898_begin_0 = const()[name = tensor<string, []>("op_56898_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_56898_end_0 = const()[name = tensor<string, []>("op_56898_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_56898_end_mask_0 = const()[name = tensor<string, []>("op_56898_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56898_cast = slice_by_index(begin = var_56898_begin_0, end = var_56898_end_0, end_mask = var_56898_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56898_cast")];
            tensor<int32, [4]> var_56902_begin_0 = const()[name = tensor<string, []>("op_56902_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_56902_end_0 = const()[name = tensor<string, []>("op_56902_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_56902_end_mask_0 = const()[name = tensor<string, []>("op_56902_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56902_cast = slice_by_index(begin = var_56902_begin_0, end = var_56902_end_0, end_mask = var_56902_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56902_cast")];
            tensor<int32, [4]> var_56906_begin_0 = const()[name = tensor<string, []>("op_56906_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_56906_end_0 = const()[name = tensor<string, []>("op_56906_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_56906_end_mask_0 = const()[name = tensor<string, []>("op_56906_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56906_cast = slice_by_index(begin = var_56906_begin_0, end = var_56906_end_0, end_mask = var_56906_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56906_cast")];
            tensor<int32, [4]> var_56910_begin_0 = const()[name = tensor<string, []>("op_56910_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_56910_end_0 = const()[name = tensor<string, []>("op_56910_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_56910_end_mask_0 = const()[name = tensor<string, []>("op_56910_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56910_cast = slice_by_index(begin = var_56910_begin_0, end = var_56910_end_0, end_mask = var_56910_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56910_cast")];
            tensor<int32, [4]> var_56914_begin_0 = const()[name = tensor<string, []>("op_56914_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_56914_end_0 = const()[name = tensor<string, []>("op_56914_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_56914_end_mask_0 = const()[name = tensor<string, []>("op_56914_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_56914_cast = slice_by_index(begin = var_56914_begin_0, end = var_56914_end_0, end_mask = var_56914_end_mask_0, x = v_251_cast)[name = tensor<string, []>("op_56914_cast")];
            tensor<string, []> var_56918_equation_0 = const()[name = tensor<string, []>("op_56918_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56918_cast = einsum(equation = var_56918_equation_0, values = (var_56760_cast, var_56677_cast))[name = tensor<string, []>("op_56918_cast")];
            tensor<fp16, []> var_56919_to_fp16 = const()[name = tensor<string, []>("op_56919_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4841_cast = mul(x = var_56918_cast, y = var_56919_to_fp16)[name = tensor<string, []>("aw_4841_cast")];
            tensor<string, []> var_56922_equation_0 = const()[name = tensor<string, []>("op_56922_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56922_cast = einsum(equation = var_56922_equation_0, values = (var_56764_cast, var_56681_cast))[name = tensor<string, []>("op_56922_cast")];
            tensor<fp16, []> var_56923_to_fp16 = const()[name = tensor<string, []>("op_56923_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4843_cast = mul(x = var_56922_cast, y = var_56923_to_fp16)[name = tensor<string, []>("aw_4843_cast")];
            tensor<string, []> var_56926_equation_0 = const()[name = tensor<string, []>("op_56926_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56926_cast = einsum(equation = var_56926_equation_0, values = (var_56768_cast, var_56685_cast))[name = tensor<string, []>("op_56926_cast")];
            tensor<fp16, []> var_56927_to_fp16 = const()[name = tensor<string, []>("op_56927_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4845_cast = mul(x = var_56926_cast, y = var_56927_to_fp16)[name = tensor<string, []>("aw_4845_cast")];
            tensor<string, []> var_56930_equation_0 = const()[name = tensor<string, []>("op_56930_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56930_cast = einsum(equation = var_56930_equation_0, values = (var_56772_cast, var_56689_cast))[name = tensor<string, []>("op_56930_cast")];
            tensor<fp16, []> var_56931_to_fp16 = const()[name = tensor<string, []>("op_56931_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4847_cast = mul(x = var_56930_cast, y = var_56931_to_fp16)[name = tensor<string, []>("aw_4847_cast")];
            tensor<string, []> var_56934_equation_0 = const()[name = tensor<string, []>("op_56934_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56934_cast = einsum(equation = var_56934_equation_0, values = (var_56776_cast, var_56693_cast))[name = tensor<string, []>("op_56934_cast")];
            tensor<fp16, []> var_56935_to_fp16 = const()[name = tensor<string, []>("op_56935_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4849_cast = mul(x = var_56934_cast, y = var_56935_to_fp16)[name = tensor<string, []>("aw_4849_cast")];
            tensor<string, []> var_56938_equation_0 = const()[name = tensor<string, []>("op_56938_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56938_cast = einsum(equation = var_56938_equation_0, values = (var_56780_cast, var_56697_cast))[name = tensor<string, []>("op_56938_cast")];
            tensor<fp16, []> var_56939_to_fp16 = const()[name = tensor<string, []>("op_56939_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4851_cast = mul(x = var_56938_cast, y = var_56939_to_fp16)[name = tensor<string, []>("aw_4851_cast")];
            tensor<string, []> var_56942_equation_0 = const()[name = tensor<string, []>("op_56942_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56942_cast = einsum(equation = var_56942_equation_0, values = (var_56784_cast, var_56701_cast))[name = tensor<string, []>("op_56942_cast")];
            tensor<fp16, []> var_56943_to_fp16 = const()[name = tensor<string, []>("op_56943_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4853_cast = mul(x = var_56942_cast, y = var_56943_to_fp16)[name = tensor<string, []>("aw_4853_cast")];
            tensor<string, []> var_56946_equation_0 = const()[name = tensor<string, []>("op_56946_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56946_cast = einsum(equation = var_56946_equation_0, values = (var_56788_cast, var_56705_cast))[name = tensor<string, []>("op_56946_cast")];
            tensor<fp16, []> var_56947_to_fp16 = const()[name = tensor<string, []>("op_56947_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4855_cast = mul(x = var_56946_cast, y = var_56947_to_fp16)[name = tensor<string, []>("aw_4855_cast")];
            tensor<string, []> var_56950_equation_0 = const()[name = tensor<string, []>("op_56950_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56950_cast = einsum(equation = var_56950_equation_0, values = (var_56792_cast, var_56709_cast))[name = tensor<string, []>("op_56950_cast")];
            tensor<fp16, []> var_56951_to_fp16 = const()[name = tensor<string, []>("op_56951_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4857_cast = mul(x = var_56950_cast, y = var_56951_to_fp16)[name = tensor<string, []>("aw_4857_cast")];
            tensor<string, []> var_56954_equation_0 = const()[name = tensor<string, []>("op_56954_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56954_cast = einsum(equation = var_56954_equation_0, values = (var_56796_cast, var_56713_cast))[name = tensor<string, []>("op_56954_cast")];
            tensor<fp16, []> var_56955_to_fp16 = const()[name = tensor<string, []>("op_56955_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4859_cast = mul(x = var_56954_cast, y = var_56955_to_fp16)[name = tensor<string, []>("aw_4859_cast")];
            tensor<string, []> var_56958_equation_0 = const()[name = tensor<string, []>("op_56958_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56958_cast = einsum(equation = var_56958_equation_0, values = (var_56800_cast, var_56717_cast))[name = tensor<string, []>("op_56958_cast")];
            tensor<fp16, []> var_56959_to_fp16 = const()[name = tensor<string, []>("op_56959_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4861_cast = mul(x = var_56958_cast, y = var_56959_to_fp16)[name = tensor<string, []>("aw_4861_cast")];
            tensor<string, []> var_56962_equation_0 = const()[name = tensor<string, []>("op_56962_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56962_cast = einsum(equation = var_56962_equation_0, values = (var_56804_cast, var_56721_cast))[name = tensor<string, []>("op_56962_cast")];
            tensor<fp16, []> var_56963_to_fp16 = const()[name = tensor<string, []>("op_56963_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4863_cast = mul(x = var_56962_cast, y = var_56963_to_fp16)[name = tensor<string, []>("aw_4863_cast")];
            tensor<string, []> var_56966_equation_0 = const()[name = tensor<string, []>("op_56966_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56966_cast = einsum(equation = var_56966_equation_0, values = (var_56808_cast, var_56725_cast))[name = tensor<string, []>("op_56966_cast")];
            tensor<fp16, []> var_56967_to_fp16 = const()[name = tensor<string, []>("op_56967_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4865_cast = mul(x = var_56966_cast, y = var_56967_to_fp16)[name = tensor<string, []>("aw_4865_cast")];
            tensor<string, []> var_56970_equation_0 = const()[name = tensor<string, []>("op_56970_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56970_cast = einsum(equation = var_56970_equation_0, values = (var_56812_cast, var_56729_cast))[name = tensor<string, []>("op_56970_cast")];
            tensor<fp16, []> var_56971_to_fp16 = const()[name = tensor<string, []>("op_56971_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4867_cast = mul(x = var_56970_cast, y = var_56971_to_fp16)[name = tensor<string, []>("aw_4867_cast")];
            tensor<string, []> var_56974_equation_0 = const()[name = tensor<string, []>("op_56974_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56974_cast = einsum(equation = var_56974_equation_0, values = (var_56816_cast, var_56733_cast))[name = tensor<string, []>("op_56974_cast")];
            tensor<fp16, []> var_56975_to_fp16 = const()[name = tensor<string, []>("op_56975_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4869_cast = mul(x = var_56974_cast, y = var_56975_to_fp16)[name = tensor<string, []>("aw_4869_cast")];
            tensor<string, []> var_56978_equation_0 = const()[name = tensor<string, []>("op_56978_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56978_cast = einsum(equation = var_56978_equation_0, values = (var_56820_cast, var_56737_cast))[name = tensor<string, []>("op_56978_cast")];
            tensor<fp16, []> var_56979_to_fp16 = const()[name = tensor<string, []>("op_56979_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4871_cast = mul(x = var_56978_cast, y = var_56979_to_fp16)[name = tensor<string, []>("aw_4871_cast")];
            tensor<string, []> var_56982_equation_0 = const()[name = tensor<string, []>("op_56982_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56982_cast = einsum(equation = var_56982_equation_0, values = (var_56824_cast, var_56741_cast))[name = tensor<string, []>("op_56982_cast")];
            tensor<fp16, []> var_56983_to_fp16 = const()[name = tensor<string, []>("op_56983_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4873_cast = mul(x = var_56982_cast, y = var_56983_to_fp16)[name = tensor<string, []>("aw_4873_cast")];
            tensor<string, []> var_56986_equation_0 = const()[name = tensor<string, []>("op_56986_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56986_cast = einsum(equation = var_56986_equation_0, values = (var_56828_cast, var_56745_cast))[name = tensor<string, []>("op_56986_cast")];
            tensor<fp16, []> var_56987_to_fp16 = const()[name = tensor<string, []>("op_56987_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4875_cast = mul(x = var_56986_cast, y = var_56987_to_fp16)[name = tensor<string, []>("aw_4875_cast")];
            tensor<string, []> var_56990_equation_0 = const()[name = tensor<string, []>("op_56990_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56990_cast = einsum(equation = var_56990_equation_0, values = (var_56832_cast, var_56749_cast))[name = tensor<string, []>("op_56990_cast")];
            tensor<fp16, []> var_56991_to_fp16 = const()[name = tensor<string, []>("op_56991_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4877_cast = mul(x = var_56990_cast, y = var_56991_to_fp16)[name = tensor<string, []>("aw_4877_cast")];
            tensor<string, []> var_56994_equation_0 = const()[name = tensor<string, []>("op_56994_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_56994_cast = einsum(equation = var_56994_equation_0, values = (var_56836_cast, var_56753_cast))[name = tensor<string, []>("op_56994_cast")];
            tensor<fp16, []> var_56995_to_fp16 = const()[name = tensor<string, []>("op_56995_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4879_cast = mul(x = var_56994_cast, y = var_56995_to_fp16)[name = tensor<string, []>("aw_4879_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56997_cast = softmax(axis = var_30385, x = aw_4841_cast)[name = tensor<string, []>("op_56997_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56998_cast = softmax(axis = var_30385, x = aw_4843_cast)[name = tensor<string, []>("op_56998_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_56999_cast = softmax(axis = var_30385, x = aw_4845_cast)[name = tensor<string, []>("op_56999_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57000_cast = softmax(axis = var_30385, x = aw_4847_cast)[name = tensor<string, []>("op_57000_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57001_cast = softmax(axis = var_30385, x = aw_4849_cast)[name = tensor<string, []>("op_57001_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57002_cast = softmax(axis = var_30385, x = aw_4851_cast)[name = tensor<string, []>("op_57002_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57003_cast = softmax(axis = var_30385, x = aw_4853_cast)[name = tensor<string, []>("op_57003_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57004_cast = softmax(axis = var_30385, x = aw_4855_cast)[name = tensor<string, []>("op_57004_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57005_cast = softmax(axis = var_30385, x = aw_4857_cast)[name = tensor<string, []>("op_57005_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57006_cast = softmax(axis = var_30385, x = aw_4859_cast)[name = tensor<string, []>("op_57006_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57007_cast = softmax(axis = var_30385, x = aw_4861_cast)[name = tensor<string, []>("op_57007_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57008_cast = softmax(axis = var_30385, x = aw_4863_cast)[name = tensor<string, []>("op_57008_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57009_cast = softmax(axis = var_30385, x = aw_4865_cast)[name = tensor<string, []>("op_57009_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57010_cast = softmax(axis = var_30385, x = aw_4867_cast)[name = tensor<string, []>("op_57010_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57011_cast = softmax(axis = var_30385, x = aw_4869_cast)[name = tensor<string, []>("op_57011_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57012_cast = softmax(axis = var_30385, x = aw_4871_cast)[name = tensor<string, []>("op_57012_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57013_cast = softmax(axis = var_30385, x = aw_4873_cast)[name = tensor<string, []>("op_57013_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57014_cast = softmax(axis = var_30385, x = aw_4875_cast)[name = tensor<string, []>("op_57014_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57015_cast = softmax(axis = var_30385, x = aw_4877_cast)[name = tensor<string, []>("op_57015_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57016_cast = softmax(axis = var_30385, x = aw_4879_cast)[name = tensor<string, []>("op_57016_cast")];
            tensor<string, []> var_57018_equation_0 = const()[name = tensor<string, []>("op_57018_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57018_cast = einsum(equation = var_57018_equation_0, values = (var_56838_cast, var_56997_cast))[name = tensor<string, []>("op_57018_cast")];
            tensor<string, []> var_57020_equation_0 = const()[name = tensor<string, []>("op_57020_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57020_cast = einsum(equation = var_57020_equation_0, values = (var_56842_cast, var_56998_cast))[name = tensor<string, []>("op_57020_cast")];
            tensor<string, []> var_57022_equation_0 = const()[name = tensor<string, []>("op_57022_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57022_cast = einsum(equation = var_57022_equation_0, values = (var_56846_cast, var_56999_cast))[name = tensor<string, []>("op_57022_cast")];
            tensor<string, []> var_57024_equation_0 = const()[name = tensor<string, []>("op_57024_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57024_cast = einsum(equation = var_57024_equation_0, values = (var_56850_cast, var_57000_cast))[name = tensor<string, []>("op_57024_cast")];
            tensor<string, []> var_57026_equation_0 = const()[name = tensor<string, []>("op_57026_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57026_cast = einsum(equation = var_57026_equation_0, values = (var_56854_cast, var_57001_cast))[name = tensor<string, []>("op_57026_cast")];
            tensor<string, []> var_57028_equation_0 = const()[name = tensor<string, []>("op_57028_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57028_cast = einsum(equation = var_57028_equation_0, values = (var_56858_cast, var_57002_cast))[name = tensor<string, []>("op_57028_cast")];
            tensor<string, []> var_57030_equation_0 = const()[name = tensor<string, []>("op_57030_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57030_cast = einsum(equation = var_57030_equation_0, values = (var_56862_cast, var_57003_cast))[name = tensor<string, []>("op_57030_cast")];
            tensor<string, []> var_57032_equation_0 = const()[name = tensor<string, []>("op_57032_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57032_cast = einsum(equation = var_57032_equation_0, values = (var_56866_cast, var_57004_cast))[name = tensor<string, []>("op_57032_cast")];
            tensor<string, []> var_57034_equation_0 = const()[name = tensor<string, []>("op_57034_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57034_cast = einsum(equation = var_57034_equation_0, values = (var_56870_cast, var_57005_cast))[name = tensor<string, []>("op_57034_cast")];
            tensor<string, []> var_57036_equation_0 = const()[name = tensor<string, []>("op_57036_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57036_cast = einsum(equation = var_57036_equation_0, values = (var_56874_cast, var_57006_cast))[name = tensor<string, []>("op_57036_cast")];
            tensor<string, []> var_57038_equation_0 = const()[name = tensor<string, []>("op_57038_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57038_cast = einsum(equation = var_57038_equation_0, values = (var_56878_cast, var_57007_cast))[name = tensor<string, []>("op_57038_cast")];
            tensor<string, []> var_57040_equation_0 = const()[name = tensor<string, []>("op_57040_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57040_cast = einsum(equation = var_57040_equation_0, values = (var_56882_cast, var_57008_cast))[name = tensor<string, []>("op_57040_cast")];
            tensor<string, []> var_57042_equation_0 = const()[name = tensor<string, []>("op_57042_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57042_cast = einsum(equation = var_57042_equation_0, values = (var_56886_cast, var_57009_cast))[name = tensor<string, []>("op_57042_cast")];
            tensor<string, []> var_57044_equation_0 = const()[name = tensor<string, []>("op_57044_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57044_cast = einsum(equation = var_57044_equation_0, values = (var_56890_cast, var_57010_cast))[name = tensor<string, []>("op_57044_cast")];
            tensor<string, []> var_57046_equation_0 = const()[name = tensor<string, []>("op_57046_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57046_cast = einsum(equation = var_57046_equation_0, values = (var_56894_cast, var_57011_cast))[name = tensor<string, []>("op_57046_cast")];
            tensor<string, []> var_57048_equation_0 = const()[name = tensor<string, []>("op_57048_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57048_cast = einsum(equation = var_57048_equation_0, values = (var_56898_cast, var_57012_cast))[name = tensor<string, []>("op_57048_cast")];
            tensor<string, []> var_57050_equation_0 = const()[name = tensor<string, []>("op_57050_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57050_cast = einsum(equation = var_57050_equation_0, values = (var_56902_cast, var_57013_cast))[name = tensor<string, []>("op_57050_cast")];
            tensor<string, []> var_57052_equation_0 = const()[name = tensor<string, []>("op_57052_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57052_cast = einsum(equation = var_57052_equation_0, values = (var_56906_cast, var_57014_cast))[name = tensor<string, []>("op_57052_cast")];
            tensor<string, []> var_57054_equation_0 = const()[name = tensor<string, []>("op_57054_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57054_cast = einsum(equation = var_57054_equation_0, values = (var_56910_cast, var_57015_cast))[name = tensor<string, []>("op_57054_cast")];
            tensor<string, []> var_57056_equation_0 = const()[name = tensor<string, []>("op_57056_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57056_cast = einsum(equation = var_57056_equation_0, values = (var_56914_cast, var_57016_cast))[name = tensor<string, []>("op_57056_cast")];
            tensor<bool, []> input_703_interleave_0 = const()[name = tensor<string, []>("input_703_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_703_cast = concat(axis = var_30385, interleave = input_703_interleave_0, values = (var_57018_cast, var_57020_cast, var_57022_cast, var_57024_cast, var_57026_cast, var_57028_cast, var_57030_cast, var_57032_cast, var_57034_cast, var_57036_cast, var_57038_cast, var_57040_cast, var_57042_cast, var_57044_cast, var_57046_cast, var_57048_cast, var_57050_cast, var_57052_cast, var_57054_cast, var_57056_cast))[name = tensor<string, []>("input_703_cast")];
            tensor<int32, [2]> var_57062 = const()[name = tensor<string, []>("op_57062"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_57064 = const()[name = tensor<string, []>("op_57064"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_57066_pad_type_0 = const()[name = tensor<string, []>("op_57066_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_57066_pad_0 = const()[name = tensor<string, []>("op_57066_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4755078016)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4758354880)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_57066_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_out_0_bias_to_fp16, dilations = var_57064, groups = var_30385, pad = var_57066_pad_0, pad_type = var_57066_pad_type_0, strides = var_57062, weight = up_blocks_0_attentions_2_transformer_blocks_8_attn2_to_out_0_weight_to_fp16, x = input_703_cast)[name = tensor<string, []>("op_57066_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_377_cast = add(x = var_57066_cast, y = inputs_375_cast)[name = tensor<string, []>("inputs_377_cast")];
            tensor<int32, [1]> var_57070 = const()[name = tensor<string, []>("op_57070"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_377_cast = reduce_mean(axes = var_57070, keep_dims = var_30380, x = inputs_377_cast)[name = tensor<string, []>("channels_mean_377_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_377_cast = sub(x = inputs_377_cast, y = channels_mean_377_cast)[name = tensor<string, []>("zero_mean_377_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_377_cast = mul(x = zero_mean_377_cast, y = zero_mean_377_cast)[name = tensor<string, []>("zero_mean_sq_377_cast")];
            tensor<int32, [1]> var_57074 = const()[name = tensor<string, []>("op_57074"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_57075_cast = reduce_mean(axes = var_57074, keep_dims = var_30380, x = zero_mean_sq_377_cast)[name = tensor<string, []>("op_57075_cast")];
            tensor<fp16, []> var_57076_to_fp16 = const()[name = tensor<string, []>("op_57076_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_57077_cast = add(x = var_57075_cast, y = var_57076_to_fp16)[name = tensor<string, []>("op_57077_cast")];
            tensor<fp16, []> denom_377_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_377_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_377_cast = rsqrt(epsilon = denom_377_epsilon_0_to_fp16, x = var_57077_cast)[name = tensor<string, []>("denom_377_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_377_cast = mul(x = zero_mean_377_cast, y = denom_377_cast)[name = tensor<string, []>("out_377_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_57081_to_fp16 = const()[name = tensor<string, []>("op_57081_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4758357504)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_57082_cast = add(x = out_377_cast, y = var_57081_to_fp16)[name = tensor<string, []>("op_57082_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_57084_to_fp16 = const()[name = tensor<string, []>("op_57084_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4758360128)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_705_cast = mul(x = var_57082_cast, y = var_57084_to_fp16)[name = tensor<string, []>("input_705_cast")];
            tensor<int32, [2]> var_57092 = const()[name = tensor<string, []>("op_57092"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_57094 = const()[name = tensor<string, []>("op_57094"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_57096_pad_type_0 = const()[name = tensor<string, []>("op_57096_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_57096_pad_0 = const()[name = tensor<string, []>("op_57096_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4758362752)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_2_transformer_blocks_8_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4784577216)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_57096_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_8_ff_net_0_proj_bias_to_fp16, dilations = var_57094, groups = var_30385, pad = var_57096_pad_0, pad_type = var_57096_pad_type_0, strides = var_57092, weight = up_blocks_0_attentions_2_transformer_blocks_8_ff_net_0_proj_weight_to_fp16, x = input_705_cast)[name = tensor<string, []>("op_57096_cast")];
            tensor<int32, [2]> var_57097_split_sizes_0 = const()[name = tensor<string, []>("op_57097_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_57097_axis_0 = const()[name = tensor<string, []>("op_57097_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_57097_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_57097_cast_1 = split(axis = var_57097_axis_0, split_sizes = var_57097_split_sizes_0, x = var_57096_cast)[name = tensor<string, []>("op_57097_cast")];
            tensor<string, []> var_57099_mode_0 = const()[name = tensor<string, []>("op_57099_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_57099_cast = gelu(mode = var_57099_mode_0, x = var_57097_cast_1)[name = tensor<string, []>("op_57099_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_707_cast = mul(x = var_57097_cast_0, y = var_57099_cast)[name = tensor<string, []>("input_707_cast")];
            tensor<int32, [2]> var_57103 = const()[name = tensor<string, []>("op_57103"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_57105 = const()[name = tensor<string, []>("op_57105"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_57107_pad_type_0 = const()[name = tensor<string, []>("op_57107_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_57107_pad_0 = const()[name = tensor<string, []>("op_57107_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_8_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4784597760)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_8_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_8_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4797705024)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_57107_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_8_ff_net_2_bias_to_fp16, dilations = var_57105, groups = var_30385, pad = var_57107_pad_0, pad_type = var_57107_pad_type_0, strides = var_57103, weight = up_blocks_0_attentions_2_transformer_blocks_8_ff_net_2_weight_to_fp16, x = input_707_cast)[name = tensor<string, []>("op_57107_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_379_cast = add(x = var_57107_cast, y = inputs_377_cast)[name = tensor<string, []>("inputs_379_cast")];
            tensor<int32, [1]> var_57117 = const()[name = tensor<string, []>("op_57117"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_379_cast = reduce_mean(axes = var_57117, keep_dims = var_30380, x = inputs_379_cast)[name = tensor<string, []>("channels_mean_379_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_379_cast = sub(x = inputs_379_cast, y = channels_mean_379_cast)[name = tensor<string, []>("zero_mean_379_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_379_cast = mul(x = zero_mean_379_cast, y = zero_mean_379_cast)[name = tensor<string, []>("zero_mean_sq_379_cast")];
            tensor<int32, [1]> var_57121 = const()[name = tensor<string, []>("op_57121"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_57122_cast = reduce_mean(axes = var_57121, keep_dims = var_30380, x = zero_mean_sq_379_cast)[name = tensor<string, []>("op_57122_cast")];
            tensor<fp16, []> var_57123_to_fp16 = const()[name = tensor<string, []>("op_57123_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_57124_cast = add(x = var_57122_cast, y = var_57123_to_fp16)[name = tensor<string, []>("op_57124_cast")];
            tensor<fp16, []> denom_379_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_379_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_379_cast = rsqrt(epsilon = denom_379_epsilon_0_to_fp16, x = var_57124_cast)[name = tensor<string, []>("denom_379_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_379_cast = mul(x = zero_mean_379_cast, y = denom_379_cast)[name = tensor<string, []>("out_379_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_57128_to_fp16 = const()[name = tensor<string, []>("op_57128_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4797707648)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_57129_cast = add(x = out_379_cast, y = var_57128_to_fp16)[name = tensor<string, []>("op_57129_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_57131_to_fp16 = const()[name = tensor<string, []>("op_57131_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4797710272)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_489_cast = mul(x = var_57129_cast, y = var_57131_to_fp16)[name = tensor<string, []>("hidden_states_489_cast")];
            tensor<int32, [2]> var_57138 = const()[name = tensor<string, []>("op_57138"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_57140 = const()[name = tensor<string, []>("op_57140"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_253_pad_type_0 = const()[name = tensor<string, []>("q_253_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_253_pad_0 = const()[name = tensor<string, []>("q_253_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4797712896)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_253_cast = conv(dilations = var_57140, groups = var_30385, pad = q_253_pad_0, pad_type = q_253_pad_type_0, strides = var_57138, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_q_weight_to_fp16, x = hidden_states_489_cast)[name = tensor<string, []>("q_253_cast")];
            tensor<int32, [2]> var_57144 = const()[name = tensor<string, []>("op_57144"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_57146 = const()[name = tensor<string, []>("op_57146"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_505_pad_type_0 = const()[name = tensor<string, []>("k_505_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_505_pad_0 = const()[name = tensor<string, []>("k_505_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4800989760)))];
            tensor<fp16, [2, 1280, 1, 1024]> k_505_cast = conv(dilations = var_57146, groups = var_30385, pad = k_505_pad_0, pad_type = k_505_pad_type_0, strides = var_57144, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_k_weight_to_fp16, x = hidden_states_489_cast)[name = tensor<string, []>("k_505_cast")];
            tensor<int32, [2]> var_57150 = const()[name = tensor<string, []>("op_57150"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_57152 = const()[name = tensor<string, []>("op_57152"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_253_pad_type_0 = const()[name = tensor<string, []>("v_253_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_253_pad_0 = const()[name = tensor<string, []>("v_253_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4804266624)))];
            tensor<fp16, [2, 1280, 1, 1024]> v_253_cast = conv(dilations = var_57152, groups = var_30385, pad = v_253_pad_0, pad_type = v_253_pad_type_0, strides = var_57150, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_v_weight_to_fp16, x = hidden_states_489_cast)[name = tensor<string, []>("v_253_cast")];
            tensor<int32, [4]> var_57156_begin_0 = const()[name = tensor<string, []>("op_57156_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_57156_end_0 = const()[name = tensor<string, []>("op_57156_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_57156_end_mask_0 = const()[name = tensor<string, []>("op_57156_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57156_cast = slice_by_index(begin = var_57156_begin_0, end = var_57156_end_0, end_mask = var_57156_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57156_cast")];
            tensor<int32, [4]> var_57160_begin_0 = const()[name = tensor<string, []>("op_57160_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_57160_end_0 = const()[name = tensor<string, []>("op_57160_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_57160_end_mask_0 = const()[name = tensor<string, []>("op_57160_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57160_cast = slice_by_index(begin = var_57160_begin_0, end = var_57160_end_0, end_mask = var_57160_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57160_cast")];
            tensor<int32, [4]> var_57164_begin_0 = const()[name = tensor<string, []>("op_57164_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_57164_end_0 = const()[name = tensor<string, []>("op_57164_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_57164_end_mask_0 = const()[name = tensor<string, []>("op_57164_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57164_cast = slice_by_index(begin = var_57164_begin_0, end = var_57164_end_0, end_mask = var_57164_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57164_cast")];
            tensor<int32, [4]> var_57168_begin_0 = const()[name = tensor<string, []>("op_57168_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_57168_end_0 = const()[name = tensor<string, []>("op_57168_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_57168_end_mask_0 = const()[name = tensor<string, []>("op_57168_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57168_cast = slice_by_index(begin = var_57168_begin_0, end = var_57168_end_0, end_mask = var_57168_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57168_cast")];
            tensor<int32, [4]> var_57172_begin_0 = const()[name = tensor<string, []>("op_57172_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_57172_end_0 = const()[name = tensor<string, []>("op_57172_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_57172_end_mask_0 = const()[name = tensor<string, []>("op_57172_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57172_cast = slice_by_index(begin = var_57172_begin_0, end = var_57172_end_0, end_mask = var_57172_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57172_cast")];
            tensor<int32, [4]> var_57176_begin_0 = const()[name = tensor<string, []>("op_57176_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_57176_end_0 = const()[name = tensor<string, []>("op_57176_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_57176_end_mask_0 = const()[name = tensor<string, []>("op_57176_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57176_cast = slice_by_index(begin = var_57176_begin_0, end = var_57176_end_0, end_mask = var_57176_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57176_cast")];
            tensor<int32, [4]> var_57180_begin_0 = const()[name = tensor<string, []>("op_57180_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_57180_end_0 = const()[name = tensor<string, []>("op_57180_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_57180_end_mask_0 = const()[name = tensor<string, []>("op_57180_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57180_cast = slice_by_index(begin = var_57180_begin_0, end = var_57180_end_0, end_mask = var_57180_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57180_cast")];
            tensor<int32, [4]> var_57184_begin_0 = const()[name = tensor<string, []>("op_57184_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_57184_end_0 = const()[name = tensor<string, []>("op_57184_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_57184_end_mask_0 = const()[name = tensor<string, []>("op_57184_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57184_cast = slice_by_index(begin = var_57184_begin_0, end = var_57184_end_0, end_mask = var_57184_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57184_cast")];
            tensor<int32, [4]> var_57188_begin_0 = const()[name = tensor<string, []>("op_57188_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_57188_end_0 = const()[name = tensor<string, []>("op_57188_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_57188_end_mask_0 = const()[name = tensor<string, []>("op_57188_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57188_cast = slice_by_index(begin = var_57188_begin_0, end = var_57188_end_0, end_mask = var_57188_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57188_cast")];
            tensor<int32, [4]> var_57192_begin_0 = const()[name = tensor<string, []>("op_57192_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_57192_end_0 = const()[name = tensor<string, []>("op_57192_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_57192_end_mask_0 = const()[name = tensor<string, []>("op_57192_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57192_cast = slice_by_index(begin = var_57192_begin_0, end = var_57192_end_0, end_mask = var_57192_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57192_cast")];
            tensor<int32, [4]> var_57196_begin_0 = const()[name = tensor<string, []>("op_57196_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_57196_end_0 = const()[name = tensor<string, []>("op_57196_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_57196_end_mask_0 = const()[name = tensor<string, []>("op_57196_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57196_cast = slice_by_index(begin = var_57196_begin_0, end = var_57196_end_0, end_mask = var_57196_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57196_cast")];
            tensor<int32, [4]> var_57200_begin_0 = const()[name = tensor<string, []>("op_57200_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_57200_end_0 = const()[name = tensor<string, []>("op_57200_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_57200_end_mask_0 = const()[name = tensor<string, []>("op_57200_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57200_cast = slice_by_index(begin = var_57200_begin_0, end = var_57200_end_0, end_mask = var_57200_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57200_cast")];
            tensor<int32, [4]> var_57204_begin_0 = const()[name = tensor<string, []>("op_57204_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_57204_end_0 = const()[name = tensor<string, []>("op_57204_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_57204_end_mask_0 = const()[name = tensor<string, []>("op_57204_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57204_cast = slice_by_index(begin = var_57204_begin_0, end = var_57204_end_0, end_mask = var_57204_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57204_cast")];
            tensor<int32, [4]> var_57208_begin_0 = const()[name = tensor<string, []>("op_57208_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_57208_end_0 = const()[name = tensor<string, []>("op_57208_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_57208_end_mask_0 = const()[name = tensor<string, []>("op_57208_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57208_cast = slice_by_index(begin = var_57208_begin_0, end = var_57208_end_0, end_mask = var_57208_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57208_cast")];
            tensor<int32, [4]> var_57212_begin_0 = const()[name = tensor<string, []>("op_57212_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_57212_end_0 = const()[name = tensor<string, []>("op_57212_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_57212_end_mask_0 = const()[name = tensor<string, []>("op_57212_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57212_cast = slice_by_index(begin = var_57212_begin_0, end = var_57212_end_0, end_mask = var_57212_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57212_cast")];
            tensor<int32, [4]> var_57216_begin_0 = const()[name = tensor<string, []>("op_57216_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_57216_end_0 = const()[name = tensor<string, []>("op_57216_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_57216_end_mask_0 = const()[name = tensor<string, []>("op_57216_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57216_cast = slice_by_index(begin = var_57216_begin_0, end = var_57216_end_0, end_mask = var_57216_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57216_cast")];
            tensor<int32, [4]> var_57220_begin_0 = const()[name = tensor<string, []>("op_57220_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_57220_end_0 = const()[name = tensor<string, []>("op_57220_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_57220_end_mask_0 = const()[name = tensor<string, []>("op_57220_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57220_cast = slice_by_index(begin = var_57220_begin_0, end = var_57220_end_0, end_mask = var_57220_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57220_cast")];
            tensor<int32, [4]> var_57224_begin_0 = const()[name = tensor<string, []>("op_57224_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_57224_end_0 = const()[name = tensor<string, []>("op_57224_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_57224_end_mask_0 = const()[name = tensor<string, []>("op_57224_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57224_cast = slice_by_index(begin = var_57224_begin_0, end = var_57224_end_0, end_mask = var_57224_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57224_cast")];
            tensor<int32, [4]> var_57228_begin_0 = const()[name = tensor<string, []>("op_57228_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_57228_end_0 = const()[name = tensor<string, []>("op_57228_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_57228_end_mask_0 = const()[name = tensor<string, []>("op_57228_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57228_cast = slice_by_index(begin = var_57228_begin_0, end = var_57228_end_0, end_mask = var_57228_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57228_cast")];
            tensor<int32, [4]> var_57232_begin_0 = const()[name = tensor<string, []>("op_57232_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_57232_end_0 = const()[name = tensor<string, []>("op_57232_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_57232_end_mask_0 = const()[name = tensor<string, []>("op_57232_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57232_cast = slice_by_index(begin = var_57232_begin_0, end = var_57232_end_0, end_mask = var_57232_end_mask_0, x = q_253_cast)[name = tensor<string, []>("op_57232_cast")];
            tensor<int32, [4]> k_507_perm_0 = const()[name = tensor<string, []>("k_507_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_57239_begin_0 = const()[name = tensor<string, []>("op_57239_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_57239_end_0 = const()[name = tensor<string, []>("op_57239_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_57239_end_mask_0 = const()[name = tensor<string, []>("op_57239_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 1280]> transpose_13 = transpose(perm = k_507_perm_0, x = k_505_cast)[name = tensor<string, []>("transpose_13")];
            tensor<fp16, [2, 1024, 1, 64]> var_57239_cast = slice_by_index(begin = var_57239_begin_0, end = var_57239_end_0, end_mask = var_57239_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57239_cast")];
            tensor<int32, [4]> var_57243_begin_0 = const()[name = tensor<string, []>("op_57243_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_57243_end_0 = const()[name = tensor<string, []>("op_57243_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_57243_end_mask_0 = const()[name = tensor<string, []>("op_57243_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57243_cast = slice_by_index(begin = var_57243_begin_0, end = var_57243_end_0, end_mask = var_57243_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57243_cast")];
            tensor<int32, [4]> var_57247_begin_0 = const()[name = tensor<string, []>("op_57247_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_57247_end_0 = const()[name = tensor<string, []>("op_57247_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_57247_end_mask_0 = const()[name = tensor<string, []>("op_57247_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57247_cast = slice_by_index(begin = var_57247_begin_0, end = var_57247_end_0, end_mask = var_57247_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57247_cast")];
            tensor<int32, [4]> var_57251_begin_0 = const()[name = tensor<string, []>("op_57251_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_57251_end_0 = const()[name = tensor<string, []>("op_57251_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_57251_end_mask_0 = const()[name = tensor<string, []>("op_57251_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57251_cast = slice_by_index(begin = var_57251_begin_0, end = var_57251_end_0, end_mask = var_57251_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57251_cast")];
            tensor<int32, [4]> var_57255_begin_0 = const()[name = tensor<string, []>("op_57255_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_57255_end_0 = const()[name = tensor<string, []>("op_57255_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_57255_end_mask_0 = const()[name = tensor<string, []>("op_57255_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57255_cast = slice_by_index(begin = var_57255_begin_0, end = var_57255_end_0, end_mask = var_57255_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57255_cast")];
            tensor<int32, [4]> var_57259_begin_0 = const()[name = tensor<string, []>("op_57259_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_57259_end_0 = const()[name = tensor<string, []>("op_57259_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_57259_end_mask_0 = const()[name = tensor<string, []>("op_57259_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57259_cast = slice_by_index(begin = var_57259_begin_0, end = var_57259_end_0, end_mask = var_57259_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57259_cast")];
            tensor<int32, [4]> var_57263_begin_0 = const()[name = tensor<string, []>("op_57263_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_57263_end_0 = const()[name = tensor<string, []>("op_57263_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_57263_end_mask_0 = const()[name = tensor<string, []>("op_57263_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57263_cast = slice_by_index(begin = var_57263_begin_0, end = var_57263_end_0, end_mask = var_57263_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57263_cast")];
            tensor<int32, [4]> var_57267_begin_0 = const()[name = tensor<string, []>("op_57267_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_57267_end_0 = const()[name = tensor<string, []>("op_57267_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_57267_end_mask_0 = const()[name = tensor<string, []>("op_57267_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57267_cast = slice_by_index(begin = var_57267_begin_0, end = var_57267_end_0, end_mask = var_57267_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57267_cast")];
            tensor<int32, [4]> var_57271_begin_0 = const()[name = tensor<string, []>("op_57271_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_57271_end_0 = const()[name = tensor<string, []>("op_57271_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_57271_end_mask_0 = const()[name = tensor<string, []>("op_57271_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57271_cast = slice_by_index(begin = var_57271_begin_0, end = var_57271_end_0, end_mask = var_57271_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57271_cast")];
            tensor<int32, [4]> var_57275_begin_0 = const()[name = tensor<string, []>("op_57275_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_57275_end_0 = const()[name = tensor<string, []>("op_57275_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_57275_end_mask_0 = const()[name = tensor<string, []>("op_57275_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57275_cast = slice_by_index(begin = var_57275_begin_0, end = var_57275_end_0, end_mask = var_57275_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57275_cast")];
            tensor<int32, [4]> var_57279_begin_0 = const()[name = tensor<string, []>("op_57279_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_57279_end_0 = const()[name = tensor<string, []>("op_57279_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 704])];
            tensor<bool, [4]> var_57279_end_mask_0 = const()[name = tensor<string, []>("op_57279_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57279_cast = slice_by_index(begin = var_57279_begin_0, end = var_57279_end_0, end_mask = var_57279_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57279_cast")];
            tensor<int32, [4]> var_57283_begin_0 = const()[name = tensor<string, []>("op_57283_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_57283_end_0 = const()[name = tensor<string, []>("op_57283_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 768])];
            tensor<bool, [4]> var_57283_end_mask_0 = const()[name = tensor<string, []>("op_57283_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57283_cast = slice_by_index(begin = var_57283_begin_0, end = var_57283_end_0, end_mask = var_57283_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57283_cast")];
            tensor<int32, [4]> var_57287_begin_0 = const()[name = tensor<string, []>("op_57287_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_57287_end_0 = const()[name = tensor<string, []>("op_57287_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 832])];
            tensor<bool, [4]> var_57287_end_mask_0 = const()[name = tensor<string, []>("op_57287_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57287_cast = slice_by_index(begin = var_57287_begin_0, end = var_57287_end_0, end_mask = var_57287_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57287_cast")];
            tensor<int32, [4]> var_57291_begin_0 = const()[name = tensor<string, []>("op_57291_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_57291_end_0 = const()[name = tensor<string, []>("op_57291_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 896])];
            tensor<bool, [4]> var_57291_end_mask_0 = const()[name = tensor<string, []>("op_57291_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57291_cast = slice_by_index(begin = var_57291_begin_0, end = var_57291_end_0, end_mask = var_57291_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57291_cast")];
            tensor<int32, [4]> var_57295_begin_0 = const()[name = tensor<string, []>("op_57295_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_57295_end_0 = const()[name = tensor<string, []>("op_57295_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 960])];
            tensor<bool, [4]> var_57295_end_mask_0 = const()[name = tensor<string, []>("op_57295_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57295_cast = slice_by_index(begin = var_57295_begin_0, end = var_57295_end_0, end_mask = var_57295_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57295_cast")];
            tensor<int32, [4]> var_57299_begin_0 = const()[name = tensor<string, []>("op_57299_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_57299_end_0 = const()[name = tensor<string, []>("op_57299_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_57299_end_mask_0 = const()[name = tensor<string, []>("op_57299_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57299_cast = slice_by_index(begin = var_57299_begin_0, end = var_57299_end_0, end_mask = var_57299_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57299_cast")];
            tensor<int32, [4]> var_57303_begin_0 = const()[name = tensor<string, []>("op_57303_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_57303_end_0 = const()[name = tensor<string, []>("op_57303_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1088])];
            tensor<bool, [4]> var_57303_end_mask_0 = const()[name = tensor<string, []>("op_57303_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57303_cast = slice_by_index(begin = var_57303_begin_0, end = var_57303_end_0, end_mask = var_57303_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57303_cast")];
            tensor<int32, [4]> var_57307_begin_0 = const()[name = tensor<string, []>("op_57307_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_57307_end_0 = const()[name = tensor<string, []>("op_57307_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1152])];
            tensor<bool, [4]> var_57307_end_mask_0 = const()[name = tensor<string, []>("op_57307_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57307_cast = slice_by_index(begin = var_57307_begin_0, end = var_57307_end_0, end_mask = var_57307_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57307_cast")];
            tensor<int32, [4]> var_57311_begin_0 = const()[name = tensor<string, []>("op_57311_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_57311_end_0 = const()[name = tensor<string, []>("op_57311_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1216])];
            tensor<bool, [4]> var_57311_end_mask_0 = const()[name = tensor<string, []>("op_57311_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57311_cast = slice_by_index(begin = var_57311_begin_0, end = var_57311_end_0, end_mask = var_57311_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57311_cast")];
            tensor<int32, [4]> var_57315_begin_0 = const()[name = tensor<string, []>("op_57315_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_57315_end_0 = const()[name = tensor<string, []>("op_57315_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1280])];
            tensor<bool, [4]> var_57315_end_mask_0 = const()[name = tensor<string, []>("op_57315_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_57315_cast = slice_by_index(begin = var_57315_begin_0, end = var_57315_end_0, end_mask = var_57315_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_57315_cast")];
            tensor<int32, [4]> var_57317_begin_0 = const()[name = tensor<string, []>("op_57317_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_57317_end_0 = const()[name = tensor<string, []>("op_57317_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_57317_end_mask_0 = const()[name = tensor<string, []>("op_57317_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57317_cast = slice_by_index(begin = var_57317_begin_0, end = var_57317_end_0, end_mask = var_57317_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57317_cast")];
            tensor<int32, [4]> var_57321_begin_0 = const()[name = tensor<string, []>("op_57321_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_57321_end_0 = const()[name = tensor<string, []>("op_57321_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_57321_end_mask_0 = const()[name = tensor<string, []>("op_57321_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57321_cast = slice_by_index(begin = var_57321_begin_0, end = var_57321_end_0, end_mask = var_57321_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57321_cast")];
            tensor<int32, [4]> var_57325_begin_0 = const()[name = tensor<string, []>("op_57325_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_57325_end_0 = const()[name = tensor<string, []>("op_57325_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_57325_end_mask_0 = const()[name = tensor<string, []>("op_57325_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57325_cast = slice_by_index(begin = var_57325_begin_0, end = var_57325_end_0, end_mask = var_57325_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57325_cast")];
            tensor<int32, [4]> var_57329_begin_0 = const()[name = tensor<string, []>("op_57329_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_57329_end_0 = const()[name = tensor<string, []>("op_57329_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_57329_end_mask_0 = const()[name = tensor<string, []>("op_57329_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57329_cast = slice_by_index(begin = var_57329_begin_0, end = var_57329_end_0, end_mask = var_57329_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57329_cast")];
            tensor<int32, [4]> var_57333_begin_0 = const()[name = tensor<string, []>("op_57333_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_57333_end_0 = const()[name = tensor<string, []>("op_57333_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_57333_end_mask_0 = const()[name = tensor<string, []>("op_57333_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57333_cast = slice_by_index(begin = var_57333_begin_0, end = var_57333_end_0, end_mask = var_57333_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57333_cast")];
            tensor<int32, [4]> var_57337_begin_0 = const()[name = tensor<string, []>("op_57337_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_57337_end_0 = const()[name = tensor<string, []>("op_57337_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_57337_end_mask_0 = const()[name = tensor<string, []>("op_57337_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57337_cast = slice_by_index(begin = var_57337_begin_0, end = var_57337_end_0, end_mask = var_57337_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57337_cast")];
            tensor<int32, [4]> var_57341_begin_0 = const()[name = tensor<string, []>("op_57341_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_57341_end_0 = const()[name = tensor<string, []>("op_57341_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_57341_end_mask_0 = const()[name = tensor<string, []>("op_57341_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57341_cast = slice_by_index(begin = var_57341_begin_0, end = var_57341_end_0, end_mask = var_57341_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57341_cast")];
            tensor<int32, [4]> var_57345_begin_0 = const()[name = tensor<string, []>("op_57345_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_57345_end_0 = const()[name = tensor<string, []>("op_57345_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_57345_end_mask_0 = const()[name = tensor<string, []>("op_57345_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57345_cast = slice_by_index(begin = var_57345_begin_0, end = var_57345_end_0, end_mask = var_57345_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57345_cast")];
            tensor<int32, [4]> var_57349_begin_0 = const()[name = tensor<string, []>("op_57349_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_57349_end_0 = const()[name = tensor<string, []>("op_57349_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_57349_end_mask_0 = const()[name = tensor<string, []>("op_57349_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57349_cast = slice_by_index(begin = var_57349_begin_0, end = var_57349_end_0, end_mask = var_57349_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57349_cast")];
            tensor<int32, [4]> var_57353_begin_0 = const()[name = tensor<string, []>("op_57353_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_57353_end_0 = const()[name = tensor<string, []>("op_57353_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_57353_end_mask_0 = const()[name = tensor<string, []>("op_57353_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57353_cast = slice_by_index(begin = var_57353_begin_0, end = var_57353_end_0, end_mask = var_57353_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57353_cast")];
            tensor<int32, [4]> var_57357_begin_0 = const()[name = tensor<string, []>("op_57357_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_57357_end_0 = const()[name = tensor<string, []>("op_57357_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_57357_end_mask_0 = const()[name = tensor<string, []>("op_57357_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57357_cast = slice_by_index(begin = var_57357_begin_0, end = var_57357_end_0, end_mask = var_57357_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57357_cast")];
            tensor<int32, [4]> var_57361_begin_0 = const()[name = tensor<string, []>("op_57361_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_57361_end_0 = const()[name = tensor<string, []>("op_57361_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_57361_end_mask_0 = const()[name = tensor<string, []>("op_57361_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57361_cast = slice_by_index(begin = var_57361_begin_0, end = var_57361_end_0, end_mask = var_57361_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57361_cast")];
            tensor<int32, [4]> var_57365_begin_0 = const()[name = tensor<string, []>("op_57365_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_57365_end_0 = const()[name = tensor<string, []>("op_57365_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_57365_end_mask_0 = const()[name = tensor<string, []>("op_57365_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57365_cast = slice_by_index(begin = var_57365_begin_0, end = var_57365_end_0, end_mask = var_57365_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57365_cast")];
            tensor<int32, [4]> var_57369_begin_0 = const()[name = tensor<string, []>("op_57369_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_57369_end_0 = const()[name = tensor<string, []>("op_57369_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_57369_end_mask_0 = const()[name = tensor<string, []>("op_57369_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57369_cast = slice_by_index(begin = var_57369_begin_0, end = var_57369_end_0, end_mask = var_57369_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57369_cast")];
            tensor<int32, [4]> var_57373_begin_0 = const()[name = tensor<string, []>("op_57373_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_57373_end_0 = const()[name = tensor<string, []>("op_57373_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_57373_end_mask_0 = const()[name = tensor<string, []>("op_57373_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57373_cast = slice_by_index(begin = var_57373_begin_0, end = var_57373_end_0, end_mask = var_57373_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57373_cast")];
            tensor<int32, [4]> var_57377_begin_0 = const()[name = tensor<string, []>("op_57377_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_57377_end_0 = const()[name = tensor<string, []>("op_57377_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_57377_end_mask_0 = const()[name = tensor<string, []>("op_57377_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57377_cast = slice_by_index(begin = var_57377_begin_0, end = var_57377_end_0, end_mask = var_57377_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57377_cast")];
            tensor<int32, [4]> var_57381_begin_0 = const()[name = tensor<string, []>("op_57381_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_57381_end_0 = const()[name = tensor<string, []>("op_57381_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_57381_end_mask_0 = const()[name = tensor<string, []>("op_57381_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57381_cast = slice_by_index(begin = var_57381_begin_0, end = var_57381_end_0, end_mask = var_57381_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57381_cast")];
            tensor<int32, [4]> var_57385_begin_0 = const()[name = tensor<string, []>("op_57385_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_57385_end_0 = const()[name = tensor<string, []>("op_57385_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_57385_end_mask_0 = const()[name = tensor<string, []>("op_57385_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57385_cast = slice_by_index(begin = var_57385_begin_0, end = var_57385_end_0, end_mask = var_57385_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57385_cast")];
            tensor<int32, [4]> var_57389_begin_0 = const()[name = tensor<string, []>("op_57389_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_57389_end_0 = const()[name = tensor<string, []>("op_57389_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_57389_end_mask_0 = const()[name = tensor<string, []>("op_57389_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57389_cast = slice_by_index(begin = var_57389_begin_0, end = var_57389_end_0, end_mask = var_57389_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57389_cast")];
            tensor<int32, [4]> var_57393_begin_0 = const()[name = tensor<string, []>("op_57393_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_57393_end_0 = const()[name = tensor<string, []>("op_57393_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_57393_end_mask_0 = const()[name = tensor<string, []>("op_57393_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57393_cast = slice_by_index(begin = var_57393_begin_0, end = var_57393_end_0, end_mask = var_57393_end_mask_0, x = v_253_cast)[name = tensor<string, []>("op_57393_cast")];
            tensor<string, []> var_57397_equation_0 = const()[name = tensor<string, []>("op_57397_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57397_cast = einsum(equation = var_57397_equation_0, values = (var_57239_cast, var_57156_cast))[name = tensor<string, []>("op_57397_cast")];
            tensor<fp16, []> var_57398_to_fp16 = const()[name = tensor<string, []>("op_57398_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4881_cast = mul(x = var_57397_cast, y = var_57398_to_fp16)[name = tensor<string, []>("aw_4881_cast")];
            tensor<string, []> var_57401_equation_0 = const()[name = tensor<string, []>("op_57401_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57401_cast = einsum(equation = var_57401_equation_0, values = (var_57243_cast, var_57160_cast))[name = tensor<string, []>("op_57401_cast")];
            tensor<fp16, []> var_57402_to_fp16 = const()[name = tensor<string, []>("op_57402_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4883_cast = mul(x = var_57401_cast, y = var_57402_to_fp16)[name = tensor<string, []>("aw_4883_cast")];
            tensor<string, []> var_57405_equation_0 = const()[name = tensor<string, []>("op_57405_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57405_cast = einsum(equation = var_57405_equation_0, values = (var_57247_cast, var_57164_cast))[name = tensor<string, []>("op_57405_cast")];
            tensor<fp16, []> var_57406_to_fp16 = const()[name = tensor<string, []>("op_57406_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4885_cast = mul(x = var_57405_cast, y = var_57406_to_fp16)[name = tensor<string, []>("aw_4885_cast")];
            tensor<string, []> var_57409_equation_0 = const()[name = tensor<string, []>("op_57409_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57409_cast = einsum(equation = var_57409_equation_0, values = (var_57251_cast, var_57168_cast))[name = tensor<string, []>("op_57409_cast")];
            tensor<fp16, []> var_57410_to_fp16 = const()[name = tensor<string, []>("op_57410_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4887_cast = mul(x = var_57409_cast, y = var_57410_to_fp16)[name = tensor<string, []>("aw_4887_cast")];
            tensor<string, []> var_57413_equation_0 = const()[name = tensor<string, []>("op_57413_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57413_cast = einsum(equation = var_57413_equation_0, values = (var_57255_cast, var_57172_cast))[name = tensor<string, []>("op_57413_cast")];
            tensor<fp16, []> var_57414_to_fp16 = const()[name = tensor<string, []>("op_57414_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4889_cast = mul(x = var_57413_cast, y = var_57414_to_fp16)[name = tensor<string, []>("aw_4889_cast")];
            tensor<string, []> var_57417_equation_0 = const()[name = tensor<string, []>("op_57417_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57417_cast = einsum(equation = var_57417_equation_0, values = (var_57259_cast, var_57176_cast))[name = tensor<string, []>("op_57417_cast")];
            tensor<fp16, []> var_57418_to_fp16 = const()[name = tensor<string, []>("op_57418_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4891_cast = mul(x = var_57417_cast, y = var_57418_to_fp16)[name = tensor<string, []>("aw_4891_cast")];
            tensor<string, []> var_57421_equation_0 = const()[name = tensor<string, []>("op_57421_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57421_cast = einsum(equation = var_57421_equation_0, values = (var_57263_cast, var_57180_cast))[name = tensor<string, []>("op_57421_cast")];
            tensor<fp16, []> var_57422_to_fp16 = const()[name = tensor<string, []>("op_57422_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4893_cast = mul(x = var_57421_cast, y = var_57422_to_fp16)[name = tensor<string, []>("aw_4893_cast")];
            tensor<string, []> var_57425_equation_0 = const()[name = tensor<string, []>("op_57425_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57425_cast = einsum(equation = var_57425_equation_0, values = (var_57267_cast, var_57184_cast))[name = tensor<string, []>("op_57425_cast")];
            tensor<fp16, []> var_57426_to_fp16 = const()[name = tensor<string, []>("op_57426_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4895_cast = mul(x = var_57425_cast, y = var_57426_to_fp16)[name = tensor<string, []>("aw_4895_cast")];
            tensor<string, []> var_57429_equation_0 = const()[name = tensor<string, []>("op_57429_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57429_cast = einsum(equation = var_57429_equation_0, values = (var_57271_cast, var_57188_cast))[name = tensor<string, []>("op_57429_cast")];
            tensor<fp16, []> var_57430_to_fp16 = const()[name = tensor<string, []>("op_57430_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4897_cast = mul(x = var_57429_cast, y = var_57430_to_fp16)[name = tensor<string, []>("aw_4897_cast")];
            tensor<string, []> var_57433_equation_0 = const()[name = tensor<string, []>("op_57433_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57433_cast = einsum(equation = var_57433_equation_0, values = (var_57275_cast, var_57192_cast))[name = tensor<string, []>("op_57433_cast")];
            tensor<fp16, []> var_57434_to_fp16 = const()[name = tensor<string, []>("op_57434_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4899_cast = mul(x = var_57433_cast, y = var_57434_to_fp16)[name = tensor<string, []>("aw_4899_cast")];
            tensor<string, []> var_57437_equation_0 = const()[name = tensor<string, []>("op_57437_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57437_cast = einsum(equation = var_57437_equation_0, values = (var_57279_cast, var_57196_cast))[name = tensor<string, []>("op_57437_cast")];
            tensor<fp16, []> var_57438_to_fp16 = const()[name = tensor<string, []>("op_57438_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4901_cast = mul(x = var_57437_cast, y = var_57438_to_fp16)[name = tensor<string, []>("aw_4901_cast")];
            tensor<string, []> var_57441_equation_0 = const()[name = tensor<string, []>("op_57441_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57441_cast = einsum(equation = var_57441_equation_0, values = (var_57283_cast, var_57200_cast))[name = tensor<string, []>("op_57441_cast")];
            tensor<fp16, []> var_57442_to_fp16 = const()[name = tensor<string, []>("op_57442_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4903_cast = mul(x = var_57441_cast, y = var_57442_to_fp16)[name = tensor<string, []>("aw_4903_cast")];
            tensor<string, []> var_57445_equation_0 = const()[name = tensor<string, []>("op_57445_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57445_cast = einsum(equation = var_57445_equation_0, values = (var_57287_cast, var_57204_cast))[name = tensor<string, []>("op_57445_cast")];
            tensor<fp16, []> var_57446_to_fp16 = const()[name = tensor<string, []>("op_57446_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4905_cast = mul(x = var_57445_cast, y = var_57446_to_fp16)[name = tensor<string, []>("aw_4905_cast")];
            tensor<string, []> var_57449_equation_0 = const()[name = tensor<string, []>("op_57449_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57449_cast = einsum(equation = var_57449_equation_0, values = (var_57291_cast, var_57208_cast))[name = tensor<string, []>("op_57449_cast")];
            tensor<fp16, []> var_57450_to_fp16 = const()[name = tensor<string, []>("op_57450_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4907_cast = mul(x = var_57449_cast, y = var_57450_to_fp16)[name = tensor<string, []>("aw_4907_cast")];
            tensor<string, []> var_57453_equation_0 = const()[name = tensor<string, []>("op_57453_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57453_cast = einsum(equation = var_57453_equation_0, values = (var_57295_cast, var_57212_cast))[name = tensor<string, []>("op_57453_cast")];
            tensor<fp16, []> var_57454_to_fp16 = const()[name = tensor<string, []>("op_57454_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4909_cast = mul(x = var_57453_cast, y = var_57454_to_fp16)[name = tensor<string, []>("aw_4909_cast")];
            tensor<string, []> var_57457_equation_0 = const()[name = tensor<string, []>("op_57457_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57457_cast = einsum(equation = var_57457_equation_0, values = (var_57299_cast, var_57216_cast))[name = tensor<string, []>("op_57457_cast")];
            tensor<fp16, []> var_57458_to_fp16 = const()[name = tensor<string, []>("op_57458_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4911_cast = mul(x = var_57457_cast, y = var_57458_to_fp16)[name = tensor<string, []>("aw_4911_cast")];
            tensor<string, []> var_57461_equation_0 = const()[name = tensor<string, []>("op_57461_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57461_cast = einsum(equation = var_57461_equation_0, values = (var_57303_cast, var_57220_cast))[name = tensor<string, []>("op_57461_cast")];
            tensor<fp16, []> var_57462_to_fp16 = const()[name = tensor<string, []>("op_57462_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4913_cast = mul(x = var_57461_cast, y = var_57462_to_fp16)[name = tensor<string, []>("aw_4913_cast")];
            tensor<string, []> var_57465_equation_0 = const()[name = tensor<string, []>("op_57465_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57465_cast = einsum(equation = var_57465_equation_0, values = (var_57307_cast, var_57224_cast))[name = tensor<string, []>("op_57465_cast")];
            tensor<fp16, []> var_57466_to_fp16 = const()[name = tensor<string, []>("op_57466_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4915_cast = mul(x = var_57465_cast, y = var_57466_to_fp16)[name = tensor<string, []>("aw_4915_cast")];
            tensor<string, []> var_57469_equation_0 = const()[name = tensor<string, []>("op_57469_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57469_cast = einsum(equation = var_57469_equation_0, values = (var_57311_cast, var_57228_cast))[name = tensor<string, []>("op_57469_cast")];
            tensor<fp16, []> var_57470_to_fp16 = const()[name = tensor<string, []>("op_57470_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4917_cast = mul(x = var_57469_cast, y = var_57470_to_fp16)[name = tensor<string, []>("aw_4917_cast")];
            tensor<string, []> var_57473_equation_0 = const()[name = tensor<string, []>("op_57473_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57473_cast = einsum(equation = var_57473_equation_0, values = (var_57315_cast, var_57232_cast))[name = tensor<string, []>("op_57473_cast")];
            tensor<fp16, []> var_57474_to_fp16 = const()[name = tensor<string, []>("op_57474_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_4919_cast = mul(x = var_57473_cast, y = var_57474_to_fp16)[name = tensor<string, []>("aw_4919_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57476_cast = softmax(axis = var_30385, x = aw_4881_cast)[name = tensor<string, []>("op_57476_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57477_cast = softmax(axis = var_30385, x = aw_4883_cast)[name = tensor<string, []>("op_57477_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57478_cast = softmax(axis = var_30385, x = aw_4885_cast)[name = tensor<string, []>("op_57478_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57479_cast = softmax(axis = var_30385, x = aw_4887_cast)[name = tensor<string, []>("op_57479_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57480_cast = softmax(axis = var_30385, x = aw_4889_cast)[name = tensor<string, []>("op_57480_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57481_cast = softmax(axis = var_30385, x = aw_4891_cast)[name = tensor<string, []>("op_57481_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57482_cast = softmax(axis = var_30385, x = aw_4893_cast)[name = tensor<string, []>("op_57482_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57483_cast = softmax(axis = var_30385, x = aw_4895_cast)[name = tensor<string, []>("op_57483_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57484_cast = softmax(axis = var_30385, x = aw_4897_cast)[name = tensor<string, []>("op_57484_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57485_cast = softmax(axis = var_30385, x = aw_4899_cast)[name = tensor<string, []>("op_57485_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57486_cast = softmax(axis = var_30385, x = aw_4901_cast)[name = tensor<string, []>("op_57486_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57487_cast = softmax(axis = var_30385, x = aw_4903_cast)[name = tensor<string, []>("op_57487_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57488_cast = softmax(axis = var_30385, x = aw_4905_cast)[name = tensor<string, []>("op_57488_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57489_cast = softmax(axis = var_30385, x = aw_4907_cast)[name = tensor<string, []>("op_57489_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57490_cast = softmax(axis = var_30385, x = aw_4909_cast)[name = tensor<string, []>("op_57490_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57491_cast = softmax(axis = var_30385, x = aw_4911_cast)[name = tensor<string, []>("op_57491_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57492_cast = softmax(axis = var_30385, x = aw_4913_cast)[name = tensor<string, []>("op_57492_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57493_cast = softmax(axis = var_30385, x = aw_4915_cast)[name = tensor<string, []>("op_57493_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57494_cast = softmax(axis = var_30385, x = aw_4917_cast)[name = tensor<string, []>("op_57494_cast")];
            tensor<fp16, [2, 1024, 1, 1024]> var_57495_cast = softmax(axis = var_30385, x = aw_4919_cast)[name = tensor<string, []>("op_57495_cast")];
            tensor<string, []> var_57497_equation_0 = const()[name = tensor<string, []>("op_57497_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57497_cast = einsum(equation = var_57497_equation_0, values = (var_57317_cast, var_57476_cast))[name = tensor<string, []>("op_57497_cast")];
            tensor<string, []> var_57499_equation_0 = const()[name = tensor<string, []>("op_57499_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57499_cast = einsum(equation = var_57499_equation_0, values = (var_57321_cast, var_57477_cast))[name = tensor<string, []>("op_57499_cast")];
            tensor<string, []> var_57501_equation_0 = const()[name = tensor<string, []>("op_57501_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57501_cast = einsum(equation = var_57501_equation_0, values = (var_57325_cast, var_57478_cast))[name = tensor<string, []>("op_57501_cast")];
            tensor<string, []> var_57503_equation_0 = const()[name = tensor<string, []>("op_57503_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57503_cast = einsum(equation = var_57503_equation_0, values = (var_57329_cast, var_57479_cast))[name = tensor<string, []>("op_57503_cast")];
            tensor<string, []> var_57505_equation_0 = const()[name = tensor<string, []>("op_57505_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57505_cast = einsum(equation = var_57505_equation_0, values = (var_57333_cast, var_57480_cast))[name = tensor<string, []>("op_57505_cast")];
            tensor<string, []> var_57507_equation_0 = const()[name = tensor<string, []>("op_57507_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57507_cast = einsum(equation = var_57507_equation_0, values = (var_57337_cast, var_57481_cast))[name = tensor<string, []>("op_57507_cast")];
            tensor<string, []> var_57509_equation_0 = const()[name = tensor<string, []>("op_57509_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57509_cast = einsum(equation = var_57509_equation_0, values = (var_57341_cast, var_57482_cast))[name = tensor<string, []>("op_57509_cast")];
            tensor<string, []> var_57511_equation_0 = const()[name = tensor<string, []>("op_57511_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57511_cast = einsum(equation = var_57511_equation_0, values = (var_57345_cast, var_57483_cast))[name = tensor<string, []>("op_57511_cast")];
            tensor<string, []> var_57513_equation_0 = const()[name = tensor<string, []>("op_57513_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57513_cast = einsum(equation = var_57513_equation_0, values = (var_57349_cast, var_57484_cast))[name = tensor<string, []>("op_57513_cast")];
            tensor<string, []> var_57515_equation_0 = const()[name = tensor<string, []>("op_57515_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57515_cast = einsum(equation = var_57515_equation_0, values = (var_57353_cast, var_57485_cast))[name = tensor<string, []>("op_57515_cast")];
            tensor<string, []> var_57517_equation_0 = const()[name = tensor<string, []>("op_57517_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57517_cast = einsum(equation = var_57517_equation_0, values = (var_57357_cast, var_57486_cast))[name = tensor<string, []>("op_57517_cast")];
            tensor<string, []> var_57519_equation_0 = const()[name = tensor<string, []>("op_57519_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57519_cast = einsum(equation = var_57519_equation_0, values = (var_57361_cast, var_57487_cast))[name = tensor<string, []>("op_57519_cast")];
            tensor<string, []> var_57521_equation_0 = const()[name = tensor<string, []>("op_57521_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57521_cast = einsum(equation = var_57521_equation_0, values = (var_57365_cast, var_57488_cast))[name = tensor<string, []>("op_57521_cast")];
            tensor<string, []> var_57523_equation_0 = const()[name = tensor<string, []>("op_57523_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57523_cast = einsum(equation = var_57523_equation_0, values = (var_57369_cast, var_57489_cast))[name = tensor<string, []>("op_57523_cast")];
            tensor<string, []> var_57525_equation_0 = const()[name = tensor<string, []>("op_57525_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57525_cast = einsum(equation = var_57525_equation_0, values = (var_57373_cast, var_57490_cast))[name = tensor<string, []>("op_57525_cast")];
            tensor<string, []> var_57527_equation_0 = const()[name = tensor<string, []>("op_57527_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57527_cast = einsum(equation = var_57527_equation_0, values = (var_57377_cast, var_57491_cast))[name = tensor<string, []>("op_57527_cast")];
            tensor<string, []> var_57529_equation_0 = const()[name = tensor<string, []>("op_57529_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57529_cast = einsum(equation = var_57529_equation_0, values = (var_57381_cast, var_57492_cast))[name = tensor<string, []>("op_57529_cast")];
            tensor<string, []> var_57531_equation_0 = const()[name = tensor<string, []>("op_57531_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57531_cast = einsum(equation = var_57531_equation_0, values = (var_57385_cast, var_57493_cast))[name = tensor<string, []>("op_57531_cast")];
            tensor<string, []> var_57533_equation_0 = const()[name = tensor<string, []>("op_57533_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57533_cast = einsum(equation = var_57533_equation_0, values = (var_57389_cast, var_57494_cast))[name = tensor<string, []>("op_57533_cast")];
            tensor<string, []> var_57535_equation_0 = const()[name = tensor<string, []>("op_57535_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57535_cast = einsum(equation = var_57535_equation_0, values = (var_57393_cast, var_57495_cast))[name = tensor<string, []>("op_57535_cast")];
            tensor<bool, []> input_709_interleave_0 = const()[name = tensor<string, []>("input_709_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_709_cast = concat(axis = var_30385, interleave = input_709_interleave_0, values = (var_57497_cast, var_57499_cast, var_57501_cast, var_57503_cast, var_57505_cast, var_57507_cast, var_57509_cast, var_57511_cast, var_57513_cast, var_57515_cast, var_57517_cast, var_57519_cast, var_57521_cast, var_57523_cast, var_57525_cast, var_57527_cast, var_57529_cast, var_57531_cast, var_57533_cast, var_57535_cast))[name = tensor<string, []>("input_709_cast")];
            tensor<int32, [2]> var_57541 = const()[name = tensor<string, []>("op_57541"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_57543 = const()[name = tensor<string, []>("op_57543"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_57545_pad_type_0 = const()[name = tensor<string, []>("op_57545_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_57545_pad_0 = const()[name = tensor<string, []>("op_57545_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4807543488)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4810820352)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_57545_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_out_0_bias_to_fp16, dilations = var_57543, groups = var_30385, pad = var_57545_pad_0, pad_type = var_57545_pad_type_0, strides = var_57541, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn1_to_out_0_weight_to_fp16, x = input_709_cast)[name = tensor<string, []>("op_57545_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_381_cast = add(x = var_57545_cast, y = inputs_379_cast)[name = tensor<string, []>("inputs_381_cast")];
            tensor<int32, [1]> var_57549 = const()[name = tensor<string, []>("op_57549"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_381_cast = reduce_mean(axes = var_57549, keep_dims = var_30380, x = inputs_381_cast)[name = tensor<string, []>("channels_mean_381_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_381_cast = sub(x = inputs_381_cast, y = channels_mean_381_cast)[name = tensor<string, []>("zero_mean_381_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_381_cast = mul(x = zero_mean_381_cast, y = zero_mean_381_cast)[name = tensor<string, []>("zero_mean_sq_381_cast")];
            tensor<int32, [1]> var_57553 = const()[name = tensor<string, []>("op_57553"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_57554_cast = reduce_mean(axes = var_57553, keep_dims = var_30380, x = zero_mean_sq_381_cast)[name = tensor<string, []>("op_57554_cast")];
            tensor<fp16, []> var_57555_to_fp16 = const()[name = tensor<string, []>("op_57555_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_57556_cast = add(x = var_57554_cast, y = var_57555_to_fp16)[name = tensor<string, []>("op_57556_cast")];
            tensor<fp16, []> denom_381_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_381_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_381_cast = rsqrt(epsilon = denom_381_epsilon_0_to_fp16, x = var_57556_cast)[name = tensor<string, []>("denom_381_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_381_cast = mul(x = zero_mean_381_cast, y = denom_381_cast)[name = tensor<string, []>("out_381_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_57560_to_fp16 = const()[name = tensor<string, []>("op_57560_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4810822976)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_57561_cast = add(x = out_381_cast, y = var_57560_to_fp16)[name = tensor<string, []>("op_57561_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_57563_to_fp16 = const()[name = tensor<string, []>("op_57563_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4810825600)))];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_491_cast = mul(x = var_57561_cast, y = var_57563_to_fp16)[name = tensor<string, []>("hidden_states_491_cast")];
            tensor<int32, [2]> var_57570 = const()[name = tensor<string, []>("op_57570"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_57572 = const()[name = tensor<string, []>("op_57572"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_255_pad_type_0 = const()[name = tensor<string, []>("q_255_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_255_pad_0 = const()[name = tensor<string, []>("q_255_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4810828224)))];
            tensor<fp16, [2, 1280, 1, 1024]> q_255_cast = conv(dilations = var_57572, groups = var_30385, pad = q_255_pad_0, pad_type = q_255_pad_type_0, strides = var_57570, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_q_weight_to_fp16, x = hidden_states_491_cast)[name = tensor<string, []>("q_255_cast")];
            tensor<int32, [2]> var_57576 = const()[name = tensor<string, []>("op_57576"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_57578 = const()[name = tensor<string, []>("op_57578"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_509_pad_type_0 = const()[name = tensor<string, []>("k_509_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_509_pad_0 = const()[name = tensor<string, []>("k_509_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4814105088)))];
            tensor<fp16, [2, 1280, 1, 77]> k_509_cast = conv(dilations = var_57578, groups = var_30385, pad = k_509_pad_0, pad_type = k_509_pad_type_0, strides = var_57576, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_509_cast")];
            tensor<int32, [2]> var_57582 = const()[name = tensor<string, []>("op_57582"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_57584 = const()[name = tensor<string, []>("op_57584"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_255_pad_type_0 = const()[name = tensor<string, []>("v_255_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_255_pad_0 = const()[name = tensor<string, []>("v_255_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2048, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4819348032)))];
            tensor<fp16, [2, 1280, 1, 77]> v_255_cast = conv(dilations = var_57584, groups = var_30385, pad = v_255_pad_0, pad_type = v_255_pad_type_0, strides = var_57582, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_255_cast")];
            tensor<int32, [4]> var_57588_begin_0 = const()[name = tensor<string, []>("op_57588_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_57588_end_0 = const()[name = tensor<string, []>("op_57588_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_57588_end_mask_0 = const()[name = tensor<string, []>("op_57588_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57588_cast = slice_by_index(begin = var_57588_begin_0, end = var_57588_end_0, end_mask = var_57588_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57588_cast")];
            tensor<int32, [4]> var_57592_begin_0 = const()[name = tensor<string, []>("op_57592_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_57592_end_0 = const()[name = tensor<string, []>("op_57592_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_57592_end_mask_0 = const()[name = tensor<string, []>("op_57592_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57592_cast = slice_by_index(begin = var_57592_begin_0, end = var_57592_end_0, end_mask = var_57592_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57592_cast")];
            tensor<int32, [4]> var_57596_begin_0 = const()[name = tensor<string, []>("op_57596_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_57596_end_0 = const()[name = tensor<string, []>("op_57596_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_57596_end_mask_0 = const()[name = tensor<string, []>("op_57596_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57596_cast = slice_by_index(begin = var_57596_begin_0, end = var_57596_end_0, end_mask = var_57596_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57596_cast")];
            tensor<int32, [4]> var_57600_begin_0 = const()[name = tensor<string, []>("op_57600_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_57600_end_0 = const()[name = tensor<string, []>("op_57600_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_57600_end_mask_0 = const()[name = tensor<string, []>("op_57600_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57600_cast = slice_by_index(begin = var_57600_begin_0, end = var_57600_end_0, end_mask = var_57600_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57600_cast")];
            tensor<int32, [4]> var_57604_begin_0 = const()[name = tensor<string, []>("op_57604_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_57604_end_0 = const()[name = tensor<string, []>("op_57604_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_57604_end_mask_0 = const()[name = tensor<string, []>("op_57604_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57604_cast = slice_by_index(begin = var_57604_begin_0, end = var_57604_end_0, end_mask = var_57604_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57604_cast")];
            tensor<int32, [4]> var_57608_begin_0 = const()[name = tensor<string, []>("op_57608_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_57608_end_0 = const()[name = tensor<string, []>("op_57608_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_57608_end_mask_0 = const()[name = tensor<string, []>("op_57608_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57608_cast = slice_by_index(begin = var_57608_begin_0, end = var_57608_end_0, end_mask = var_57608_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57608_cast")];
            tensor<int32, [4]> var_57612_begin_0 = const()[name = tensor<string, []>("op_57612_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_57612_end_0 = const()[name = tensor<string, []>("op_57612_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_57612_end_mask_0 = const()[name = tensor<string, []>("op_57612_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57612_cast = slice_by_index(begin = var_57612_begin_0, end = var_57612_end_0, end_mask = var_57612_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57612_cast")];
            tensor<int32, [4]> var_57616_begin_0 = const()[name = tensor<string, []>("op_57616_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_57616_end_0 = const()[name = tensor<string, []>("op_57616_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_57616_end_mask_0 = const()[name = tensor<string, []>("op_57616_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57616_cast = slice_by_index(begin = var_57616_begin_0, end = var_57616_end_0, end_mask = var_57616_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57616_cast")];
            tensor<int32, [4]> var_57620_begin_0 = const()[name = tensor<string, []>("op_57620_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_57620_end_0 = const()[name = tensor<string, []>("op_57620_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_57620_end_mask_0 = const()[name = tensor<string, []>("op_57620_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57620_cast = slice_by_index(begin = var_57620_begin_0, end = var_57620_end_0, end_mask = var_57620_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57620_cast")];
            tensor<int32, [4]> var_57624_begin_0 = const()[name = tensor<string, []>("op_57624_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_57624_end_0 = const()[name = tensor<string, []>("op_57624_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_57624_end_mask_0 = const()[name = tensor<string, []>("op_57624_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57624_cast = slice_by_index(begin = var_57624_begin_0, end = var_57624_end_0, end_mask = var_57624_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57624_cast")];
            tensor<int32, [4]> var_57628_begin_0 = const()[name = tensor<string, []>("op_57628_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_57628_end_0 = const()[name = tensor<string, []>("op_57628_end_0"), val = tensor<int32, [4]>([2, 704, 1, 1024])];
            tensor<bool, [4]> var_57628_end_mask_0 = const()[name = tensor<string, []>("op_57628_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57628_cast = slice_by_index(begin = var_57628_begin_0, end = var_57628_end_0, end_mask = var_57628_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57628_cast")];
            tensor<int32, [4]> var_57632_begin_0 = const()[name = tensor<string, []>("op_57632_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_57632_end_0 = const()[name = tensor<string, []>("op_57632_end_0"), val = tensor<int32, [4]>([2, 768, 1, 1024])];
            tensor<bool, [4]> var_57632_end_mask_0 = const()[name = tensor<string, []>("op_57632_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57632_cast = slice_by_index(begin = var_57632_begin_0, end = var_57632_end_0, end_mask = var_57632_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57632_cast")];
            tensor<int32, [4]> var_57636_begin_0 = const()[name = tensor<string, []>("op_57636_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_57636_end_0 = const()[name = tensor<string, []>("op_57636_end_0"), val = tensor<int32, [4]>([2, 832, 1, 1024])];
            tensor<bool, [4]> var_57636_end_mask_0 = const()[name = tensor<string, []>("op_57636_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57636_cast = slice_by_index(begin = var_57636_begin_0, end = var_57636_end_0, end_mask = var_57636_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57636_cast")];
            tensor<int32, [4]> var_57640_begin_0 = const()[name = tensor<string, []>("op_57640_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_57640_end_0 = const()[name = tensor<string, []>("op_57640_end_0"), val = tensor<int32, [4]>([2, 896, 1, 1024])];
            tensor<bool, [4]> var_57640_end_mask_0 = const()[name = tensor<string, []>("op_57640_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57640_cast = slice_by_index(begin = var_57640_begin_0, end = var_57640_end_0, end_mask = var_57640_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57640_cast")];
            tensor<int32, [4]> var_57644_begin_0 = const()[name = tensor<string, []>("op_57644_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_57644_end_0 = const()[name = tensor<string, []>("op_57644_end_0"), val = tensor<int32, [4]>([2, 960, 1, 1024])];
            tensor<bool, [4]> var_57644_end_mask_0 = const()[name = tensor<string, []>("op_57644_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57644_cast = slice_by_index(begin = var_57644_begin_0, end = var_57644_end_0, end_mask = var_57644_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57644_cast")];
            tensor<int32, [4]> var_57648_begin_0 = const()[name = tensor<string, []>("op_57648_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_57648_end_0 = const()[name = tensor<string, []>("op_57648_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 1024])];
            tensor<bool, [4]> var_57648_end_mask_0 = const()[name = tensor<string, []>("op_57648_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57648_cast = slice_by_index(begin = var_57648_begin_0, end = var_57648_end_0, end_mask = var_57648_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57648_cast")];
            tensor<int32, [4]> var_57652_begin_0 = const()[name = tensor<string, []>("op_57652_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_57652_end_0 = const()[name = tensor<string, []>("op_57652_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 1024])];
            tensor<bool, [4]> var_57652_end_mask_0 = const()[name = tensor<string, []>("op_57652_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57652_cast = slice_by_index(begin = var_57652_begin_0, end = var_57652_end_0, end_mask = var_57652_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57652_cast")];
            tensor<int32, [4]> var_57656_begin_0 = const()[name = tensor<string, []>("op_57656_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_57656_end_0 = const()[name = tensor<string, []>("op_57656_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 1024])];
            tensor<bool, [4]> var_57656_end_mask_0 = const()[name = tensor<string, []>("op_57656_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57656_cast = slice_by_index(begin = var_57656_begin_0, end = var_57656_end_0, end_mask = var_57656_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57656_cast")];
            tensor<int32, [4]> var_57660_begin_0 = const()[name = tensor<string, []>("op_57660_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_57660_end_0 = const()[name = tensor<string, []>("op_57660_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 1024])];
            tensor<bool, [4]> var_57660_end_mask_0 = const()[name = tensor<string, []>("op_57660_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57660_cast = slice_by_index(begin = var_57660_begin_0, end = var_57660_end_0, end_mask = var_57660_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57660_cast")];
            tensor<int32, [4]> var_57664_begin_0 = const()[name = tensor<string, []>("op_57664_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_57664_end_0 = const()[name = tensor<string, []>("op_57664_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1024])];
            tensor<bool, [4]> var_57664_end_mask_0 = const()[name = tensor<string, []>("op_57664_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_57664_cast = slice_by_index(begin = var_57664_begin_0, end = var_57664_end_0, end_mask = var_57664_end_mask_0, x = q_255_cast)[name = tensor<string, []>("op_57664_cast")];
            tensor<int32, [4]> k_511_perm_0 = const()[name = tensor<string, []>("k_511_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_57671_begin_0 = const()[name = tensor<string, []>("op_57671_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_57671_end_0 = const()[name = tensor<string, []>("op_57671_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_57671_end_mask_0 = const()[name = tensor<string, []>("op_57671_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_12 = transpose(perm = k_511_perm_0, x = k_509_cast)[name = tensor<string, []>("transpose_12")];
            tensor<fp16, [2, 77, 1, 64]> var_57671_cast = slice_by_index(begin = var_57671_begin_0, end = var_57671_end_0, end_mask = var_57671_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57671_cast")];
            tensor<int32, [4]> var_57675_begin_0 = const()[name = tensor<string, []>("op_57675_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_57675_end_0 = const()[name = tensor<string, []>("op_57675_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_57675_end_mask_0 = const()[name = tensor<string, []>("op_57675_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57675_cast = slice_by_index(begin = var_57675_begin_0, end = var_57675_end_0, end_mask = var_57675_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57675_cast")];
            tensor<int32, [4]> var_57679_begin_0 = const()[name = tensor<string, []>("op_57679_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_57679_end_0 = const()[name = tensor<string, []>("op_57679_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_57679_end_mask_0 = const()[name = tensor<string, []>("op_57679_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57679_cast = slice_by_index(begin = var_57679_begin_0, end = var_57679_end_0, end_mask = var_57679_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57679_cast")];
            tensor<int32, [4]> var_57683_begin_0 = const()[name = tensor<string, []>("op_57683_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_57683_end_0 = const()[name = tensor<string, []>("op_57683_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_57683_end_mask_0 = const()[name = tensor<string, []>("op_57683_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57683_cast = slice_by_index(begin = var_57683_begin_0, end = var_57683_end_0, end_mask = var_57683_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57683_cast")];
            tensor<int32, [4]> var_57687_begin_0 = const()[name = tensor<string, []>("op_57687_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_57687_end_0 = const()[name = tensor<string, []>("op_57687_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_57687_end_mask_0 = const()[name = tensor<string, []>("op_57687_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57687_cast = slice_by_index(begin = var_57687_begin_0, end = var_57687_end_0, end_mask = var_57687_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57687_cast")];
            tensor<int32, [4]> var_57691_begin_0 = const()[name = tensor<string, []>("op_57691_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_57691_end_0 = const()[name = tensor<string, []>("op_57691_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_57691_end_mask_0 = const()[name = tensor<string, []>("op_57691_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57691_cast = slice_by_index(begin = var_57691_begin_0, end = var_57691_end_0, end_mask = var_57691_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57691_cast")];
            tensor<int32, [4]> var_57695_begin_0 = const()[name = tensor<string, []>("op_57695_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_57695_end_0 = const()[name = tensor<string, []>("op_57695_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_57695_end_mask_0 = const()[name = tensor<string, []>("op_57695_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57695_cast = slice_by_index(begin = var_57695_begin_0, end = var_57695_end_0, end_mask = var_57695_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57695_cast")];
            tensor<int32, [4]> var_57699_begin_0 = const()[name = tensor<string, []>("op_57699_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_57699_end_0 = const()[name = tensor<string, []>("op_57699_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_57699_end_mask_0 = const()[name = tensor<string, []>("op_57699_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57699_cast = slice_by_index(begin = var_57699_begin_0, end = var_57699_end_0, end_mask = var_57699_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57699_cast")];
            tensor<int32, [4]> var_57703_begin_0 = const()[name = tensor<string, []>("op_57703_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_57703_end_0 = const()[name = tensor<string, []>("op_57703_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_57703_end_mask_0 = const()[name = tensor<string, []>("op_57703_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57703_cast = slice_by_index(begin = var_57703_begin_0, end = var_57703_end_0, end_mask = var_57703_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57703_cast")];
            tensor<int32, [4]> var_57707_begin_0 = const()[name = tensor<string, []>("op_57707_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_57707_end_0 = const()[name = tensor<string, []>("op_57707_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_57707_end_mask_0 = const()[name = tensor<string, []>("op_57707_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57707_cast = slice_by_index(begin = var_57707_begin_0, end = var_57707_end_0, end_mask = var_57707_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57707_cast")];
            tensor<int32, [4]> var_57711_begin_0 = const()[name = tensor<string, []>("op_57711_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_57711_end_0 = const()[name = tensor<string, []>("op_57711_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_57711_end_mask_0 = const()[name = tensor<string, []>("op_57711_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57711_cast = slice_by_index(begin = var_57711_begin_0, end = var_57711_end_0, end_mask = var_57711_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57711_cast")];
            tensor<int32, [4]> var_57715_begin_0 = const()[name = tensor<string, []>("op_57715_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_57715_end_0 = const()[name = tensor<string, []>("op_57715_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_57715_end_mask_0 = const()[name = tensor<string, []>("op_57715_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57715_cast = slice_by_index(begin = var_57715_begin_0, end = var_57715_end_0, end_mask = var_57715_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57715_cast")];
            tensor<int32, [4]> var_57719_begin_0 = const()[name = tensor<string, []>("op_57719_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_57719_end_0 = const()[name = tensor<string, []>("op_57719_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_57719_end_mask_0 = const()[name = tensor<string, []>("op_57719_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57719_cast = slice_by_index(begin = var_57719_begin_0, end = var_57719_end_0, end_mask = var_57719_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57719_cast")];
            tensor<int32, [4]> var_57723_begin_0 = const()[name = tensor<string, []>("op_57723_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_57723_end_0 = const()[name = tensor<string, []>("op_57723_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_57723_end_mask_0 = const()[name = tensor<string, []>("op_57723_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57723_cast = slice_by_index(begin = var_57723_begin_0, end = var_57723_end_0, end_mask = var_57723_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57723_cast")];
            tensor<int32, [4]> var_57727_begin_0 = const()[name = tensor<string, []>("op_57727_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_57727_end_0 = const()[name = tensor<string, []>("op_57727_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_57727_end_mask_0 = const()[name = tensor<string, []>("op_57727_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57727_cast = slice_by_index(begin = var_57727_begin_0, end = var_57727_end_0, end_mask = var_57727_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57727_cast")];
            tensor<int32, [4]> var_57731_begin_0 = const()[name = tensor<string, []>("op_57731_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_57731_end_0 = const()[name = tensor<string, []>("op_57731_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_57731_end_mask_0 = const()[name = tensor<string, []>("op_57731_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57731_cast = slice_by_index(begin = var_57731_begin_0, end = var_57731_end_0, end_mask = var_57731_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57731_cast")];
            tensor<int32, [4]> var_57735_begin_0 = const()[name = tensor<string, []>("op_57735_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_57735_end_0 = const()[name = tensor<string, []>("op_57735_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_57735_end_mask_0 = const()[name = tensor<string, []>("op_57735_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57735_cast = slice_by_index(begin = var_57735_begin_0, end = var_57735_end_0, end_mask = var_57735_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57735_cast")];
            tensor<int32, [4]> var_57739_begin_0 = const()[name = tensor<string, []>("op_57739_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_57739_end_0 = const()[name = tensor<string, []>("op_57739_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_57739_end_mask_0 = const()[name = tensor<string, []>("op_57739_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57739_cast = slice_by_index(begin = var_57739_begin_0, end = var_57739_end_0, end_mask = var_57739_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57739_cast")];
            tensor<int32, [4]> var_57743_begin_0 = const()[name = tensor<string, []>("op_57743_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_57743_end_0 = const()[name = tensor<string, []>("op_57743_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_57743_end_mask_0 = const()[name = tensor<string, []>("op_57743_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57743_cast = slice_by_index(begin = var_57743_begin_0, end = var_57743_end_0, end_mask = var_57743_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57743_cast")];
            tensor<int32, [4]> var_57747_begin_0 = const()[name = tensor<string, []>("op_57747_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_57747_end_0 = const()[name = tensor<string, []>("op_57747_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_57747_end_mask_0 = const()[name = tensor<string, []>("op_57747_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_57747_cast = slice_by_index(begin = var_57747_begin_0, end = var_57747_end_0, end_mask = var_57747_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_57747_cast")];
            tensor<int32, [4]> var_57749_begin_0 = const()[name = tensor<string, []>("op_57749_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_57749_end_0 = const()[name = tensor<string, []>("op_57749_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_57749_end_mask_0 = const()[name = tensor<string, []>("op_57749_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57749_cast = slice_by_index(begin = var_57749_begin_0, end = var_57749_end_0, end_mask = var_57749_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57749_cast")];
            tensor<int32, [4]> var_57753_begin_0 = const()[name = tensor<string, []>("op_57753_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_57753_end_0 = const()[name = tensor<string, []>("op_57753_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_57753_end_mask_0 = const()[name = tensor<string, []>("op_57753_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57753_cast = slice_by_index(begin = var_57753_begin_0, end = var_57753_end_0, end_mask = var_57753_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57753_cast")];
            tensor<int32, [4]> var_57757_begin_0 = const()[name = tensor<string, []>("op_57757_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_57757_end_0 = const()[name = tensor<string, []>("op_57757_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_57757_end_mask_0 = const()[name = tensor<string, []>("op_57757_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57757_cast = slice_by_index(begin = var_57757_begin_0, end = var_57757_end_0, end_mask = var_57757_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57757_cast")];
            tensor<int32, [4]> var_57761_begin_0 = const()[name = tensor<string, []>("op_57761_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_57761_end_0 = const()[name = tensor<string, []>("op_57761_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_57761_end_mask_0 = const()[name = tensor<string, []>("op_57761_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57761_cast = slice_by_index(begin = var_57761_begin_0, end = var_57761_end_0, end_mask = var_57761_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57761_cast")];
            tensor<int32, [4]> var_57765_begin_0 = const()[name = tensor<string, []>("op_57765_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_57765_end_0 = const()[name = tensor<string, []>("op_57765_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_57765_end_mask_0 = const()[name = tensor<string, []>("op_57765_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57765_cast = slice_by_index(begin = var_57765_begin_0, end = var_57765_end_0, end_mask = var_57765_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57765_cast")];
            tensor<int32, [4]> var_57769_begin_0 = const()[name = tensor<string, []>("op_57769_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_57769_end_0 = const()[name = tensor<string, []>("op_57769_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_57769_end_mask_0 = const()[name = tensor<string, []>("op_57769_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57769_cast = slice_by_index(begin = var_57769_begin_0, end = var_57769_end_0, end_mask = var_57769_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57769_cast")];
            tensor<int32, [4]> var_57773_begin_0 = const()[name = tensor<string, []>("op_57773_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_57773_end_0 = const()[name = tensor<string, []>("op_57773_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_57773_end_mask_0 = const()[name = tensor<string, []>("op_57773_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57773_cast = slice_by_index(begin = var_57773_begin_0, end = var_57773_end_0, end_mask = var_57773_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57773_cast")];
            tensor<int32, [4]> var_57777_begin_0 = const()[name = tensor<string, []>("op_57777_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_57777_end_0 = const()[name = tensor<string, []>("op_57777_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_57777_end_mask_0 = const()[name = tensor<string, []>("op_57777_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57777_cast = slice_by_index(begin = var_57777_begin_0, end = var_57777_end_0, end_mask = var_57777_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57777_cast")];
            tensor<int32, [4]> var_57781_begin_0 = const()[name = tensor<string, []>("op_57781_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_57781_end_0 = const()[name = tensor<string, []>("op_57781_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_57781_end_mask_0 = const()[name = tensor<string, []>("op_57781_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57781_cast = slice_by_index(begin = var_57781_begin_0, end = var_57781_end_0, end_mask = var_57781_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57781_cast")];
            tensor<int32, [4]> var_57785_begin_0 = const()[name = tensor<string, []>("op_57785_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_57785_end_0 = const()[name = tensor<string, []>("op_57785_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_57785_end_mask_0 = const()[name = tensor<string, []>("op_57785_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57785_cast = slice_by_index(begin = var_57785_begin_0, end = var_57785_end_0, end_mask = var_57785_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57785_cast")];
            tensor<int32, [4]> var_57789_begin_0 = const()[name = tensor<string, []>("op_57789_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_57789_end_0 = const()[name = tensor<string, []>("op_57789_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_57789_end_mask_0 = const()[name = tensor<string, []>("op_57789_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57789_cast = slice_by_index(begin = var_57789_begin_0, end = var_57789_end_0, end_mask = var_57789_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57789_cast")];
            tensor<int32, [4]> var_57793_begin_0 = const()[name = tensor<string, []>("op_57793_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_57793_end_0 = const()[name = tensor<string, []>("op_57793_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_57793_end_mask_0 = const()[name = tensor<string, []>("op_57793_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57793_cast = slice_by_index(begin = var_57793_begin_0, end = var_57793_end_0, end_mask = var_57793_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57793_cast")];
            tensor<int32, [4]> var_57797_begin_0 = const()[name = tensor<string, []>("op_57797_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_57797_end_0 = const()[name = tensor<string, []>("op_57797_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_57797_end_mask_0 = const()[name = tensor<string, []>("op_57797_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57797_cast = slice_by_index(begin = var_57797_begin_0, end = var_57797_end_0, end_mask = var_57797_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57797_cast")];
            tensor<int32, [4]> var_57801_begin_0 = const()[name = tensor<string, []>("op_57801_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_57801_end_0 = const()[name = tensor<string, []>("op_57801_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_57801_end_mask_0 = const()[name = tensor<string, []>("op_57801_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57801_cast = slice_by_index(begin = var_57801_begin_0, end = var_57801_end_0, end_mask = var_57801_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57801_cast")];
            tensor<int32, [4]> var_57805_begin_0 = const()[name = tensor<string, []>("op_57805_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_57805_end_0 = const()[name = tensor<string, []>("op_57805_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_57805_end_mask_0 = const()[name = tensor<string, []>("op_57805_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57805_cast = slice_by_index(begin = var_57805_begin_0, end = var_57805_end_0, end_mask = var_57805_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57805_cast")];
            tensor<int32, [4]> var_57809_begin_0 = const()[name = tensor<string, []>("op_57809_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_57809_end_0 = const()[name = tensor<string, []>("op_57809_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_57809_end_mask_0 = const()[name = tensor<string, []>("op_57809_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57809_cast = slice_by_index(begin = var_57809_begin_0, end = var_57809_end_0, end_mask = var_57809_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57809_cast")];
            tensor<int32, [4]> var_57813_begin_0 = const()[name = tensor<string, []>("op_57813_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_57813_end_0 = const()[name = tensor<string, []>("op_57813_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_57813_end_mask_0 = const()[name = tensor<string, []>("op_57813_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57813_cast = slice_by_index(begin = var_57813_begin_0, end = var_57813_end_0, end_mask = var_57813_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57813_cast")];
            tensor<int32, [4]> var_57817_begin_0 = const()[name = tensor<string, []>("op_57817_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_57817_end_0 = const()[name = tensor<string, []>("op_57817_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_57817_end_mask_0 = const()[name = tensor<string, []>("op_57817_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57817_cast = slice_by_index(begin = var_57817_begin_0, end = var_57817_end_0, end_mask = var_57817_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57817_cast")];
            tensor<int32, [4]> var_57821_begin_0 = const()[name = tensor<string, []>("op_57821_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_57821_end_0 = const()[name = tensor<string, []>("op_57821_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_57821_end_mask_0 = const()[name = tensor<string, []>("op_57821_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57821_cast = slice_by_index(begin = var_57821_begin_0, end = var_57821_end_0, end_mask = var_57821_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57821_cast")];
            tensor<int32, [4]> var_57825_begin_0 = const()[name = tensor<string, []>("op_57825_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_57825_end_0 = const()[name = tensor<string, []>("op_57825_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_57825_end_mask_0 = const()[name = tensor<string, []>("op_57825_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_57825_cast = slice_by_index(begin = var_57825_begin_0, end = var_57825_end_0, end_mask = var_57825_end_mask_0, x = v_255_cast)[name = tensor<string, []>("op_57825_cast")];
            tensor<string, []> var_57829_equation_0 = const()[name = tensor<string, []>("op_57829_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57829_cast = einsum(equation = var_57829_equation_0, values = (var_57671_cast, var_57588_cast))[name = tensor<string, []>("op_57829_cast")];
            tensor<fp16, []> var_57830_to_fp16 = const()[name = tensor<string, []>("op_57830_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4921_cast = mul(x = var_57829_cast, y = var_57830_to_fp16)[name = tensor<string, []>("aw_4921_cast")];
            tensor<string, []> var_57833_equation_0 = const()[name = tensor<string, []>("op_57833_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57833_cast = einsum(equation = var_57833_equation_0, values = (var_57675_cast, var_57592_cast))[name = tensor<string, []>("op_57833_cast")];
            tensor<fp16, []> var_57834_to_fp16 = const()[name = tensor<string, []>("op_57834_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4923_cast = mul(x = var_57833_cast, y = var_57834_to_fp16)[name = tensor<string, []>("aw_4923_cast")];
            tensor<string, []> var_57837_equation_0 = const()[name = tensor<string, []>("op_57837_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57837_cast = einsum(equation = var_57837_equation_0, values = (var_57679_cast, var_57596_cast))[name = tensor<string, []>("op_57837_cast")];
            tensor<fp16, []> var_57838_to_fp16 = const()[name = tensor<string, []>("op_57838_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4925_cast = mul(x = var_57837_cast, y = var_57838_to_fp16)[name = tensor<string, []>("aw_4925_cast")];
            tensor<string, []> var_57841_equation_0 = const()[name = tensor<string, []>("op_57841_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57841_cast = einsum(equation = var_57841_equation_0, values = (var_57683_cast, var_57600_cast))[name = tensor<string, []>("op_57841_cast")];
            tensor<fp16, []> var_57842_to_fp16 = const()[name = tensor<string, []>("op_57842_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4927_cast = mul(x = var_57841_cast, y = var_57842_to_fp16)[name = tensor<string, []>("aw_4927_cast")];
            tensor<string, []> var_57845_equation_0 = const()[name = tensor<string, []>("op_57845_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57845_cast = einsum(equation = var_57845_equation_0, values = (var_57687_cast, var_57604_cast))[name = tensor<string, []>("op_57845_cast")];
            tensor<fp16, []> var_57846_to_fp16 = const()[name = tensor<string, []>("op_57846_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4929_cast = mul(x = var_57845_cast, y = var_57846_to_fp16)[name = tensor<string, []>("aw_4929_cast")];
            tensor<string, []> var_57849_equation_0 = const()[name = tensor<string, []>("op_57849_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57849_cast = einsum(equation = var_57849_equation_0, values = (var_57691_cast, var_57608_cast))[name = tensor<string, []>("op_57849_cast")];
            tensor<fp16, []> var_57850_to_fp16 = const()[name = tensor<string, []>("op_57850_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4931_cast = mul(x = var_57849_cast, y = var_57850_to_fp16)[name = tensor<string, []>("aw_4931_cast")];
            tensor<string, []> var_57853_equation_0 = const()[name = tensor<string, []>("op_57853_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57853_cast = einsum(equation = var_57853_equation_0, values = (var_57695_cast, var_57612_cast))[name = tensor<string, []>("op_57853_cast")];
            tensor<fp16, []> var_57854_to_fp16 = const()[name = tensor<string, []>("op_57854_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4933_cast = mul(x = var_57853_cast, y = var_57854_to_fp16)[name = tensor<string, []>("aw_4933_cast")];
            tensor<string, []> var_57857_equation_0 = const()[name = tensor<string, []>("op_57857_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57857_cast = einsum(equation = var_57857_equation_0, values = (var_57699_cast, var_57616_cast))[name = tensor<string, []>("op_57857_cast")];
            tensor<fp16, []> var_57858_to_fp16 = const()[name = tensor<string, []>("op_57858_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4935_cast = mul(x = var_57857_cast, y = var_57858_to_fp16)[name = tensor<string, []>("aw_4935_cast")];
            tensor<string, []> var_57861_equation_0 = const()[name = tensor<string, []>("op_57861_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57861_cast = einsum(equation = var_57861_equation_0, values = (var_57703_cast, var_57620_cast))[name = tensor<string, []>("op_57861_cast")];
            tensor<fp16, []> var_57862_to_fp16 = const()[name = tensor<string, []>("op_57862_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4937_cast = mul(x = var_57861_cast, y = var_57862_to_fp16)[name = tensor<string, []>("aw_4937_cast")];
            tensor<string, []> var_57865_equation_0 = const()[name = tensor<string, []>("op_57865_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57865_cast = einsum(equation = var_57865_equation_0, values = (var_57707_cast, var_57624_cast))[name = tensor<string, []>("op_57865_cast")];
            tensor<fp16, []> var_57866_to_fp16 = const()[name = tensor<string, []>("op_57866_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4939_cast = mul(x = var_57865_cast, y = var_57866_to_fp16)[name = tensor<string, []>("aw_4939_cast")];
            tensor<string, []> var_57869_equation_0 = const()[name = tensor<string, []>("op_57869_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57869_cast = einsum(equation = var_57869_equation_0, values = (var_57711_cast, var_57628_cast))[name = tensor<string, []>("op_57869_cast")];
            tensor<fp16, []> var_57870_to_fp16 = const()[name = tensor<string, []>("op_57870_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4941_cast = mul(x = var_57869_cast, y = var_57870_to_fp16)[name = tensor<string, []>("aw_4941_cast")];
            tensor<string, []> var_57873_equation_0 = const()[name = tensor<string, []>("op_57873_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57873_cast = einsum(equation = var_57873_equation_0, values = (var_57715_cast, var_57632_cast))[name = tensor<string, []>("op_57873_cast")];
            tensor<fp16, []> var_57874_to_fp16 = const()[name = tensor<string, []>("op_57874_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4943_cast = mul(x = var_57873_cast, y = var_57874_to_fp16)[name = tensor<string, []>("aw_4943_cast")];
            tensor<string, []> var_57877_equation_0 = const()[name = tensor<string, []>("op_57877_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57877_cast = einsum(equation = var_57877_equation_0, values = (var_57719_cast, var_57636_cast))[name = tensor<string, []>("op_57877_cast")];
            tensor<fp16, []> var_57878_to_fp16 = const()[name = tensor<string, []>("op_57878_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4945_cast = mul(x = var_57877_cast, y = var_57878_to_fp16)[name = tensor<string, []>("aw_4945_cast")];
            tensor<string, []> var_57881_equation_0 = const()[name = tensor<string, []>("op_57881_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57881_cast = einsum(equation = var_57881_equation_0, values = (var_57723_cast, var_57640_cast))[name = tensor<string, []>("op_57881_cast")];
            tensor<fp16, []> var_57882_to_fp16 = const()[name = tensor<string, []>("op_57882_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4947_cast = mul(x = var_57881_cast, y = var_57882_to_fp16)[name = tensor<string, []>("aw_4947_cast")];
            tensor<string, []> var_57885_equation_0 = const()[name = tensor<string, []>("op_57885_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57885_cast = einsum(equation = var_57885_equation_0, values = (var_57727_cast, var_57644_cast))[name = tensor<string, []>("op_57885_cast")];
            tensor<fp16, []> var_57886_to_fp16 = const()[name = tensor<string, []>("op_57886_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4949_cast = mul(x = var_57885_cast, y = var_57886_to_fp16)[name = tensor<string, []>("aw_4949_cast")];
            tensor<string, []> var_57889_equation_0 = const()[name = tensor<string, []>("op_57889_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57889_cast = einsum(equation = var_57889_equation_0, values = (var_57731_cast, var_57648_cast))[name = tensor<string, []>("op_57889_cast")];
            tensor<fp16, []> var_57890_to_fp16 = const()[name = tensor<string, []>("op_57890_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4951_cast = mul(x = var_57889_cast, y = var_57890_to_fp16)[name = tensor<string, []>("aw_4951_cast")];
            tensor<string, []> var_57893_equation_0 = const()[name = tensor<string, []>("op_57893_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57893_cast = einsum(equation = var_57893_equation_0, values = (var_57735_cast, var_57652_cast))[name = tensor<string, []>("op_57893_cast")];
            tensor<fp16, []> var_57894_to_fp16 = const()[name = tensor<string, []>("op_57894_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4953_cast = mul(x = var_57893_cast, y = var_57894_to_fp16)[name = tensor<string, []>("aw_4953_cast")];
            tensor<string, []> var_57897_equation_0 = const()[name = tensor<string, []>("op_57897_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57897_cast = einsum(equation = var_57897_equation_0, values = (var_57739_cast, var_57656_cast))[name = tensor<string, []>("op_57897_cast")];
            tensor<fp16, []> var_57898_to_fp16 = const()[name = tensor<string, []>("op_57898_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4955_cast = mul(x = var_57897_cast, y = var_57898_to_fp16)[name = tensor<string, []>("aw_4955_cast")];
            tensor<string, []> var_57901_equation_0 = const()[name = tensor<string, []>("op_57901_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57901_cast = einsum(equation = var_57901_equation_0, values = (var_57743_cast, var_57660_cast))[name = tensor<string, []>("op_57901_cast")];
            tensor<fp16, []> var_57902_to_fp16 = const()[name = tensor<string, []>("op_57902_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4957_cast = mul(x = var_57901_cast, y = var_57902_to_fp16)[name = tensor<string, []>("aw_4957_cast")];
            tensor<string, []> var_57905_equation_0 = const()[name = tensor<string, []>("op_57905_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_57905_cast = einsum(equation = var_57905_equation_0, values = (var_57747_cast, var_57664_cast))[name = tensor<string, []>("op_57905_cast")];
            tensor<fp16, []> var_57906_to_fp16 = const()[name = tensor<string, []>("op_57906_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_4959_cast = mul(x = var_57905_cast, y = var_57906_to_fp16)[name = tensor<string, []>("aw_4959_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57908_cast = softmax(axis = var_30385, x = aw_4921_cast)[name = tensor<string, []>("op_57908_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57909_cast = softmax(axis = var_30385, x = aw_4923_cast)[name = tensor<string, []>("op_57909_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57910_cast = softmax(axis = var_30385, x = aw_4925_cast)[name = tensor<string, []>("op_57910_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57911_cast = softmax(axis = var_30385, x = aw_4927_cast)[name = tensor<string, []>("op_57911_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57912_cast = softmax(axis = var_30385, x = aw_4929_cast)[name = tensor<string, []>("op_57912_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57913_cast = softmax(axis = var_30385, x = aw_4931_cast)[name = tensor<string, []>("op_57913_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57914_cast = softmax(axis = var_30385, x = aw_4933_cast)[name = tensor<string, []>("op_57914_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57915_cast = softmax(axis = var_30385, x = aw_4935_cast)[name = tensor<string, []>("op_57915_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57916_cast = softmax(axis = var_30385, x = aw_4937_cast)[name = tensor<string, []>("op_57916_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57917_cast = softmax(axis = var_30385, x = aw_4939_cast)[name = tensor<string, []>("op_57917_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57918_cast = softmax(axis = var_30385, x = aw_4941_cast)[name = tensor<string, []>("op_57918_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57919_cast = softmax(axis = var_30385, x = aw_4943_cast)[name = tensor<string, []>("op_57919_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57920_cast = softmax(axis = var_30385, x = aw_4945_cast)[name = tensor<string, []>("op_57920_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57921_cast = softmax(axis = var_30385, x = aw_4947_cast)[name = tensor<string, []>("op_57921_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57922_cast = softmax(axis = var_30385, x = aw_4949_cast)[name = tensor<string, []>("op_57922_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57923_cast = softmax(axis = var_30385, x = aw_4951_cast)[name = tensor<string, []>("op_57923_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57924_cast = softmax(axis = var_30385, x = aw_4953_cast)[name = tensor<string, []>("op_57924_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57925_cast = softmax(axis = var_30385, x = aw_4955_cast)[name = tensor<string, []>("op_57925_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57926_cast = softmax(axis = var_30385, x = aw_4957_cast)[name = tensor<string, []>("op_57926_cast")];
            tensor<fp16, [2, 77, 1, 1024]> var_57927_cast = softmax(axis = var_30385, x = aw_4959_cast)[name = tensor<string, []>("op_57927_cast")];
            tensor<string, []> var_57929_equation_0 = const()[name = tensor<string, []>("op_57929_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57929_cast = einsum(equation = var_57929_equation_0, values = (var_57749_cast, var_57908_cast))[name = tensor<string, []>("op_57929_cast")];
            tensor<string, []> var_57931_equation_0 = const()[name = tensor<string, []>("op_57931_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57931_cast = einsum(equation = var_57931_equation_0, values = (var_57753_cast, var_57909_cast))[name = tensor<string, []>("op_57931_cast")];
            tensor<string, []> var_57933_equation_0 = const()[name = tensor<string, []>("op_57933_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57933_cast = einsum(equation = var_57933_equation_0, values = (var_57757_cast, var_57910_cast))[name = tensor<string, []>("op_57933_cast")];
            tensor<string, []> var_57935_equation_0 = const()[name = tensor<string, []>("op_57935_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57935_cast = einsum(equation = var_57935_equation_0, values = (var_57761_cast, var_57911_cast))[name = tensor<string, []>("op_57935_cast")];
            tensor<string, []> var_57937_equation_0 = const()[name = tensor<string, []>("op_57937_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57937_cast = einsum(equation = var_57937_equation_0, values = (var_57765_cast, var_57912_cast))[name = tensor<string, []>("op_57937_cast")];
            tensor<string, []> var_57939_equation_0 = const()[name = tensor<string, []>("op_57939_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57939_cast = einsum(equation = var_57939_equation_0, values = (var_57769_cast, var_57913_cast))[name = tensor<string, []>("op_57939_cast")];
            tensor<string, []> var_57941_equation_0 = const()[name = tensor<string, []>("op_57941_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57941_cast = einsum(equation = var_57941_equation_0, values = (var_57773_cast, var_57914_cast))[name = tensor<string, []>("op_57941_cast")];
            tensor<string, []> var_57943_equation_0 = const()[name = tensor<string, []>("op_57943_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57943_cast = einsum(equation = var_57943_equation_0, values = (var_57777_cast, var_57915_cast))[name = tensor<string, []>("op_57943_cast")];
            tensor<string, []> var_57945_equation_0 = const()[name = tensor<string, []>("op_57945_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57945_cast = einsum(equation = var_57945_equation_0, values = (var_57781_cast, var_57916_cast))[name = tensor<string, []>("op_57945_cast")];
            tensor<string, []> var_57947_equation_0 = const()[name = tensor<string, []>("op_57947_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57947_cast = einsum(equation = var_57947_equation_0, values = (var_57785_cast, var_57917_cast))[name = tensor<string, []>("op_57947_cast")];
            tensor<string, []> var_57949_equation_0 = const()[name = tensor<string, []>("op_57949_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57949_cast = einsum(equation = var_57949_equation_0, values = (var_57789_cast, var_57918_cast))[name = tensor<string, []>("op_57949_cast")];
            tensor<string, []> var_57951_equation_0 = const()[name = tensor<string, []>("op_57951_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57951_cast = einsum(equation = var_57951_equation_0, values = (var_57793_cast, var_57919_cast))[name = tensor<string, []>("op_57951_cast")];
            tensor<string, []> var_57953_equation_0 = const()[name = tensor<string, []>("op_57953_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57953_cast = einsum(equation = var_57953_equation_0, values = (var_57797_cast, var_57920_cast))[name = tensor<string, []>("op_57953_cast")];
            tensor<string, []> var_57955_equation_0 = const()[name = tensor<string, []>("op_57955_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57955_cast = einsum(equation = var_57955_equation_0, values = (var_57801_cast, var_57921_cast))[name = tensor<string, []>("op_57955_cast")];
            tensor<string, []> var_57957_equation_0 = const()[name = tensor<string, []>("op_57957_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57957_cast = einsum(equation = var_57957_equation_0, values = (var_57805_cast, var_57922_cast))[name = tensor<string, []>("op_57957_cast")];
            tensor<string, []> var_57959_equation_0 = const()[name = tensor<string, []>("op_57959_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57959_cast = einsum(equation = var_57959_equation_0, values = (var_57809_cast, var_57923_cast))[name = tensor<string, []>("op_57959_cast")];
            tensor<string, []> var_57961_equation_0 = const()[name = tensor<string, []>("op_57961_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57961_cast = einsum(equation = var_57961_equation_0, values = (var_57813_cast, var_57924_cast))[name = tensor<string, []>("op_57961_cast")];
            tensor<string, []> var_57963_equation_0 = const()[name = tensor<string, []>("op_57963_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57963_cast = einsum(equation = var_57963_equation_0, values = (var_57817_cast, var_57925_cast))[name = tensor<string, []>("op_57963_cast")];
            tensor<string, []> var_57965_equation_0 = const()[name = tensor<string, []>("op_57965_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57965_cast = einsum(equation = var_57965_equation_0, values = (var_57821_cast, var_57926_cast))[name = tensor<string, []>("op_57965_cast")];
            tensor<string, []> var_57967_equation_0 = const()[name = tensor<string, []>("op_57967_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_57967_cast = einsum(equation = var_57967_equation_0, values = (var_57825_cast, var_57927_cast))[name = tensor<string, []>("op_57967_cast")];
            tensor<bool, []> input_711_interleave_0 = const()[name = tensor<string, []>("input_711_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 1024]> input_711_cast = concat(axis = var_30385, interleave = input_711_interleave_0, values = (var_57929_cast, var_57931_cast, var_57933_cast, var_57935_cast, var_57937_cast, var_57939_cast, var_57941_cast, var_57943_cast, var_57945_cast, var_57947_cast, var_57949_cast, var_57951_cast, var_57953_cast, var_57955_cast, var_57957_cast, var_57959_cast, var_57961_cast, var_57963_cast, var_57965_cast, var_57967_cast))[name = tensor<string, []>("input_711_cast")];
            tensor<int32, [2]> var_57973 = const()[name = tensor<string, []>("op_57973"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_57975 = const()[name = tensor<string, []>("op_57975"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_57977_pad_type_0 = const()[name = tensor<string, []>("op_57977_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_57977_pad_0 = const()[name = tensor<string, []>("op_57977_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4824590976)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4827867840)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_57977_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_out_0_bias_to_fp16, dilations = var_57975, groups = var_30385, pad = var_57977_pad_0, pad_type = var_57977_pad_type_0, strides = var_57973, weight = up_blocks_0_attentions_2_transformer_blocks_9_attn2_to_out_0_weight_to_fp16, x = input_711_cast)[name = tensor<string, []>("op_57977_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> inputs_383_cast = add(x = var_57977_cast, y = inputs_381_cast)[name = tensor<string, []>("inputs_383_cast")];
            tensor<int32, [1]> var_57981 = const()[name = tensor<string, []>("op_57981"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_383_cast = reduce_mean(axes = var_57981, keep_dims = var_30380, x = inputs_383_cast)[name = tensor<string, []>("channels_mean_383_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_383_cast = sub(x = inputs_383_cast, y = channels_mean_383_cast)[name = tensor<string, []>("zero_mean_383_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> zero_mean_sq_383_cast = mul(x = zero_mean_383_cast, y = zero_mean_383_cast)[name = tensor<string, []>("zero_mean_sq_383_cast")];
            tensor<int32, [1]> var_57985 = const()[name = tensor<string, []>("op_57985"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_57986_cast = reduce_mean(axes = var_57985, keep_dims = var_30380, x = zero_mean_sq_383_cast)[name = tensor<string, []>("op_57986_cast")];
            tensor<fp16, []> var_57987_to_fp16 = const()[name = tensor<string, []>("op_57987_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_57988_cast = add(x = var_57986_cast, y = var_57987_to_fp16)[name = tensor<string, []>("op_57988_cast")];
            tensor<fp16, []> denom_383_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_383_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_383_cast = rsqrt(epsilon = denom_383_epsilon_0_to_fp16, x = var_57988_cast)[name = tensor<string, []>("denom_383_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> out_383_cast = mul(x = zero_mean_383_cast, y = denom_383_cast)[name = tensor<string, []>("out_383_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_57992_to_fp16 = const()[name = tensor<string, []>("op_57992_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4827870464)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_57993_cast = add(x = out_383_cast, y = var_57992_to_fp16)[name = tensor<string, []>("op_57993_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_57995_to_fp16 = const()[name = tensor<string, []>("op_57995_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4827873088)))];
            tensor<fp16, [2, 1280, 1, 1024]> input_713_cast = mul(x = var_57993_cast, y = var_57995_to_fp16)[name = tensor<string, []>("input_713_cast")];
            tensor<int32, [2]> var_58003 = const()[name = tensor<string, []>("op_58003"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58005 = const()[name = tensor<string, []>("op_58005"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_58007_pad_type_0 = const()[name = tensor<string, []>("op_58007_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_58007_pad_0 = const()[name = tensor<string, []>("op_58007_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4827875712)))];
            tensor<fp16, [10240]> up_blocks_0_attentions_2_transformer_blocks_9_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4854090176)))];
            tensor<fp16, [2, 10240, 1, 1024]> var_58007_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_9_ff_net_0_proj_bias_to_fp16, dilations = var_58005, groups = var_30385, pad = var_58007_pad_0, pad_type = var_58007_pad_type_0, strides = var_58003, weight = up_blocks_0_attentions_2_transformer_blocks_9_ff_net_0_proj_weight_to_fp16, x = input_713_cast)[name = tensor<string, []>("op_58007_cast")];
            tensor<int32, [2]> var_58008_split_sizes_0 = const()[name = tensor<string, []>("op_58008_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_58008_axis_0 = const()[name = tensor<string, []>("op_58008_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 1024]> var_58008_cast_0, tensor<fp16, [2, 5120, 1, 1024]> var_58008_cast_1 = split(axis = var_58008_axis_0, split_sizes = var_58008_split_sizes_0, x = var_58007_cast)[name = tensor<string, []>("op_58008_cast")];
            tensor<string, []> var_58010_mode_0 = const()[name = tensor<string, []>("op_58010_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 1024]> var_58010_cast = gelu(mode = var_58010_mode_0, x = var_58008_cast_1)[name = tensor<string, []>("op_58010_cast")];
            tensor<fp16, [2, 5120, 1, 1024]> input_715_cast = mul(x = var_58008_cast_0, y = var_58010_cast)[name = tensor<string, []>("input_715_cast")];
            tensor<int32, [2]> var_58014 = const()[name = tensor<string, []>("op_58014"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58016 = const()[name = tensor<string, []>("op_58016"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_58018_pad_type_0 = const()[name = tensor<string, []>("op_58018_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_58018_pad_0 = const()[name = tensor<string, []>("op_58018_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_0_attentions_2_transformer_blocks_9_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4854110720)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_transformer_blocks_9_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_transformer_blocks_9_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4867217984)))];
            tensor<fp16, [2, 1280, 1, 1024]> var_58018_cast = conv(bias = up_blocks_0_attentions_2_transformer_blocks_9_ff_net_2_bias_to_fp16, dilations = var_58016, groups = var_30385, pad = var_58018_pad_0, pad_type = var_58018_pad_type_0, strides = var_58014, weight = up_blocks_0_attentions_2_transformer_blocks_9_ff_net_2_weight_to_fp16, x = input_715_cast)[name = tensor<string, []>("op_58018_cast")];
            tensor<fp16, [2, 1280, 1, 1024]> hidden_states_495_cast = add(x = var_58018_cast, y = inputs_383_cast)[name = tensor<string, []>("hidden_states_495_cast")];
            tensor<int32, [4]> var_58020 = const()[name = tensor<string, []>("op_58020"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> input_717_cast = reshape(shape = var_58020, x = hidden_states_495_cast)[name = tensor<string, []>("input_717_cast")];
            tensor<int32, [2]> var_58024 = const()[name = tensor<string, []>("op_58024"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58026 = const()[name = tensor<string, []>("op_58026"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_497_pad_type_0 = const()[name = tensor<string, []>("hidden_states_497_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_497_pad_0 = const()[name = tensor<string, []>("hidden_states_497_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_attentions_2_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_proj_out_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4867220608)))];
            tensor<fp16, [1280]> up_blocks_0_attentions_2_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_attentions_2_proj_out_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4870497472)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_497_cast = conv(bias = up_blocks_0_attentions_2_proj_out_bias_to_fp16, dilations = var_58026, groups = var_30385, pad = hidden_states_497_pad_0, pad_type = hidden_states_497_pad_type_0, strides = var_58024, weight = up_blocks_0_attentions_2_proj_out_weight_to_fp16, x = input_717_cast)[name = tensor<string, []>("hidden_states_497_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_719_cast = add(x = hidden_states_497_cast, y = hidden_states_431_cast)[name = tensor<string, []>("input_719_cast")];
            tensor<fp32, []> input_721_scale_factor_height_0 = const()[name = tensor<string, []>("input_721_scale_factor_height_0"), val = tensor<fp32, []>(0x1p+1)];
            tensor<fp32, []> input_721_scale_factor_width_0 = const()[name = tensor<string, []>("input_721_scale_factor_width_0"), val = tensor<fp32, []>(0x1p+1)];
            tensor<fp16, [2, 1280, 64, 64]> input_721_cast = upsample_nearest_neighbor(scale_factor_height = input_721_scale_factor_height_0, scale_factor_width = input_721_scale_factor_width_0, x = input_719_cast)[name = tensor<string, []>("input_721_cast")];
            tensor<int32, [2]> var_58035 = const()[name = tensor<string, []>("op_58035"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58037 = const()[name = tensor<string, []>("op_58037"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_499_pad_type_0 = const()[name = tensor<string, []>("hidden_states_499_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_499_pad_0 = const()[name = tensor<string, []>("hidden_states_499_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> up_blocks_0_upsamplers_0_conv_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_upsamplers_0_conv_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4870500096)))];
            tensor<fp16, [1280]> up_blocks_0_upsamplers_0_conv_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_upsamplers_0_conv_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4899991360)))];
            tensor<fp16, [2, 1280, 64, 64]> hidden_states_499_cast = conv(bias = up_blocks_0_upsamplers_0_conv_bias_to_fp16, dilations = var_58037, groups = var_30385, pad = hidden_states_499_pad_0, pad_type = hidden_states_499_pad_type_0, strides = var_58035, weight = up_blocks_0_upsamplers_0_conv_weight_to_fp16, x = input_721_cast)[name = tensor<string, []>("hidden_states_499_cast")];
            tensor<bool, []> var_58059 = const()[name = tensor<string, []>("op_58059"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_58064 = const()[name = tensor<string, []>("op_58064"), val = tensor<int32, []>(1)];
            tensor<bool, []> input_723_interleave_0 = const()[name = tensor<string, []>("input_723_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1920, 64, 64]> input_723_cast = concat(axis = var_58064, interleave = input_723_interleave_0, values = (hidden_states_499_cast, res_hidden_states_7_cast))[name = tensor<string, []>("input_723_cast")];
            tensor<int32, [5]> reshape_120_shape_0 = const()[name = tensor<string, []>("reshape_120_shape_0"), val = tensor<int32, [5]>([2, 32, 60, 64, 64])];
            tensor<fp16, [2, 32, 60, 64, 64]> reshape_120_cast = reshape(shape = reshape_120_shape_0, x = input_723_cast)[name = tensor<string, []>("reshape_120_cast")];
            tensor<int32, [3]> reduce_mean_90_axes_0 = const()[name = tensor<string, []>("reduce_mean_90_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_90_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_90_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_90_cast = reduce_mean(axes = reduce_mean_90_axes_0, keep_dims = reduce_mean_90_keep_dims_0, x = reshape_120_cast)[name = tensor<string, []>("reduce_mean_90_cast")];
            tensor<fp16, [2, 32, 60, 64, 64]> sub_60_cast = sub(x = reshape_120_cast, y = reduce_mean_90_cast)[name = tensor<string, []>("sub_60_cast")];
            tensor<fp16, [2, 32, 60, 64, 64]> square_30_cast = square(x = sub_60_cast)[name = tensor<string, []>("square_30_cast")];
            tensor<int32, [3]> reduce_mean_92_axes_0 = const()[name = tensor<string, []>("reduce_mean_92_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_92_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_92_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_92_cast = reduce_mean(axes = reduce_mean_92_axes_0, keep_dims = reduce_mean_92_keep_dims_0, x = square_30_cast)[name = tensor<string, []>("reduce_mean_92_cast")];
            tensor<fp16, []> add_60_y_0_to_fp16 = const()[name = tensor<string, []>("add_60_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_60_cast = add(x = reduce_mean_92_cast, y = add_60_y_0_to_fp16)[name = tensor<string, []>("add_60_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_30_cast = sqrt(x = add_60_cast)[name = tensor<string, []>("sqrt_30_cast")];
            tensor<fp16, [2, 32, 60, 64, 64]> real_div_30_cast = real_div(x = sub_60_cast, y = sqrt_30_cast)[name = tensor<string, []>("real_div_30_cast")];
            tensor<int32, [4]> reshape_121_shape_0 = const()[name = tensor<string, []>("reshape_121_shape_0"), val = tensor<int32, [4]>([2, 1920, 64, 64])];
            tensor<fp16, [2, 1920, 64, 64]> reshape_121_cast = reshape(shape = reshape_121_shape_0, x = real_div_30_cast)[name = tensor<string, []>("reshape_121_cast")];
            tensor<fp16, [1920]> add_61_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_61_gamma_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4899993984)))];
            tensor<fp16, [1920]> add_61_beta_0_to_fp16 = const()[name = tensor<string, []>("add_61_beta_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4899997888)))];
            tensor<fp16, []> add_61_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_61_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1920, 64, 64]> add_61_cast = batch_norm(beta = add_61_beta_0_to_fp16, epsilon = add_61_epsilon_0_to_fp16, gamma = add_61_gamma_0_to_fp16, mean = add_55_mean_0_to_fp16, variance = add_55_variance_0_to_fp16, x = reshape_121_cast)[name = tensor<string, []>("add_61_cast")];
            tensor<fp16, [2, 1920, 64, 64]> input_727_cast = silu(x = add_61_cast)[name = tensor<string, []>("input_727_cast")];
            tensor<int32, [2]> var_58093 = const()[name = tensor<string, []>("op_58093"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58095 = const()[name = tensor<string, []>("op_58095"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_501_pad_type_0 = const()[name = tensor<string, []>("hidden_states_501_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_501_pad_0 = const()[name = tensor<string, []>("hidden_states_501_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 1920, 3, 3]> up_blocks_1_resnets_0_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [640, 1920, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4900001792)))];
            tensor<fp16, [640]> up_blocks_1_resnets_0_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4922120256)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_501_cast = conv(bias = up_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = var_58095, groups = var_58064, pad = hidden_states_501_pad_0, pad_type = hidden_states_501_pad_type_0, strides = var_58093, weight = up_blocks_1_resnets_0_conv1_weight_to_fp16, x = input_727_cast)[name = tensor<string, []>("hidden_states_501_cast")];
            tensor<int32, [2]> var_58101 = const()[name = tensor<string, []>("op_58101"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58103 = const()[name = tensor<string, []>("op_58103"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_23_pad_type_0 = const()[name = tensor<string, []>("temb_23_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_23_pad_0 = const()[name = tensor<string, []>("temb_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1280, 1, 1]> up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [640, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4922121600)))];
            tensor<fp16, [640]> up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4923760064)))];
            tensor<fp16, [2, 640, 1, 1]> temb_23_cast = conv(bias = up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_58103, groups = var_58064, pad = temb_23_pad_0, pad_type = temb_23_pad_type_0, strides = var_58101, weight = up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16, x = input_23_cast)[name = tensor<string, []>("temb_23_cast")];
            tensor<fp16, [2, 640, 64, 64]> input_731_cast = add(x = hidden_states_501_cast, y = temb_23_cast)[name = tensor<string, []>("input_731_cast")];
            tensor<int32, [5]> reshape_124_shape_0 = const()[name = tensor<string, []>("reshape_124_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 64, 64])];
            tensor<fp16, [2, 32, 20, 64, 64]> reshape_124_cast = reshape(shape = reshape_124_shape_0, x = input_731_cast)[name = tensor<string, []>("reshape_124_cast")];
            tensor<int32, [3]> reduce_mean_93_axes_0 = const()[name = tensor<string, []>("reduce_mean_93_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_93_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_93_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_93_cast = reduce_mean(axes = reduce_mean_93_axes_0, keep_dims = reduce_mean_93_keep_dims_0, x = reshape_124_cast)[name = tensor<string, []>("reduce_mean_93_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> sub_62_cast = sub(x = reshape_124_cast, y = reduce_mean_93_cast)[name = tensor<string, []>("sub_62_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> square_31_cast = square(x = sub_62_cast)[name = tensor<string, []>("square_31_cast")];
            tensor<int32, [3]> reduce_mean_95_axes_0 = const()[name = tensor<string, []>("reduce_mean_95_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_95_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_95_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_95_cast = reduce_mean(axes = reduce_mean_95_axes_0, keep_dims = reduce_mean_95_keep_dims_0, x = square_31_cast)[name = tensor<string, []>("reduce_mean_95_cast")];
            tensor<fp16, []> add_62_y_0_to_fp16 = const()[name = tensor<string, []>("add_62_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_62_cast = add(x = reduce_mean_95_cast, y = add_62_y_0_to_fp16)[name = tensor<string, []>("add_62_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_31_cast = sqrt(x = add_62_cast)[name = tensor<string, []>("sqrt_31_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_31_cast = real_div(x = sub_62_cast, y = sqrt_31_cast)[name = tensor<string, []>("real_div_31_cast")];
            tensor<int32, [4]> reshape_125_shape_0 = const()[name = tensor<string, []>("reshape_125_shape_0"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> reshape_125_cast = reshape(shape = reshape_125_shape_0, x = real_div_31_cast)[name = tensor<string, []>("reshape_125_cast")];
            tensor<fp16, [640]> add_63_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_63_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4923761408)))];
            tensor<fp16, [640]> add_63_beta_0_to_fp16 = const()[name = tensor<string, []>("add_63_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4923762752)))];
            tensor<fp16, []> add_63_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_63_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 64, 64]> add_63_cast = batch_norm(beta = add_63_beta_0_to_fp16, epsilon = add_63_epsilon_0_to_fp16, gamma = add_63_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_125_cast)[name = tensor<string, []>("add_63_cast")];
            tensor<fp16, [2, 640, 64, 64]> input_735_cast = silu(x = add_63_cast)[name = tensor<string, []>("input_735_cast")];
            tensor<int32, [2]> var_58113 = const()[name = tensor<string, []>("op_58113"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58115 = const()[name = tensor<string, []>("op_58115"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_503_pad_type_0 = const()[name = tensor<string, []>("hidden_states_503_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_503_pad_0 = const()[name = tensor<string, []>("hidden_states_503_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 640, 3, 3]> up_blocks_1_resnets_0_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [640, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4923764096)))];
            tensor<fp16, [640]> up_blocks_1_resnets_0_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4931136960)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_503_cast = conv(bias = up_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = var_58115, groups = var_58064, pad = hidden_states_503_pad_0, pad_type = hidden_states_503_pad_type_0, strides = var_58113, weight = up_blocks_1_resnets_0_conv2_weight_to_fp16, x = input_735_cast)[name = tensor<string, []>("hidden_states_503_cast")];
            tensor<int32, [2]> var_58120 = const()[name = tensor<string, []>("op_58120"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58122 = const()[name = tensor<string, []>("op_58122"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_11_pad_type_0 = const()[name = tensor<string, []>("x_11_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_11_pad_0 = const()[name = tensor<string, []>("x_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1920, 1, 1]> up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [640, 1920, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4931138304)))];
            tensor<fp16, [640]> up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4933595968)))];
            tensor<fp16, [2, 640, 64, 64]> x_11_cast = conv(bias = up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_58122, groups = var_58064, pad = x_11_pad_0, pad_type = x_11_pad_type_0, strides = var_58120, weight = up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16, x = input_723_cast)[name = tensor<string, []>("x_11_cast")];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_505_cast = add(x = x_11_cast, y = hidden_states_503_cast)[name = tensor<string, []>("hidden_states_505_cast")];
            tensor<int32, [5]> reshape_128_shape_0 = const()[name = tensor<string, []>("reshape_128_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 64, 64])];
            tensor<fp16, [2, 32, 20, 64, 64]> reshape_128_cast = reshape(shape = reshape_128_shape_0, x = hidden_states_505_cast)[name = tensor<string, []>("reshape_128_cast")];
            tensor<int32, [3]> reduce_mean_96_axes_0 = const()[name = tensor<string, []>("reduce_mean_96_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_96_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_96_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_96_cast = reduce_mean(axes = reduce_mean_96_axes_0, keep_dims = reduce_mean_96_keep_dims_0, x = reshape_128_cast)[name = tensor<string, []>("reduce_mean_96_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> sub_64_cast = sub(x = reshape_128_cast, y = reduce_mean_96_cast)[name = tensor<string, []>("sub_64_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> square_32_cast = square(x = sub_64_cast)[name = tensor<string, []>("square_32_cast")];
            tensor<int32, [3]> reduce_mean_98_axes_0 = const()[name = tensor<string, []>("reduce_mean_98_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_98_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_98_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_98_cast = reduce_mean(axes = reduce_mean_98_axes_0, keep_dims = reduce_mean_98_keep_dims_0, x = square_32_cast)[name = tensor<string, []>("reduce_mean_98_cast")];
            tensor<fp16, []> add_64_y_0_to_fp16 = const()[name = tensor<string, []>("add_64_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_64_cast = add(x = reduce_mean_98_cast, y = add_64_y_0_to_fp16)[name = tensor<string, []>("add_64_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_32_cast = sqrt(x = add_64_cast)[name = tensor<string, []>("sqrt_32_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_32_cast = real_div(x = sub_64_cast, y = sqrt_32_cast)[name = tensor<string, []>("real_div_32_cast")];
            tensor<int32, [4]> reshape_129_shape_0 = const()[name = tensor<string, []>("reshape_129_shape_0"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> reshape_129_cast = reshape(shape = reshape_129_shape_0, x = real_div_32_cast)[name = tensor<string, []>("reshape_129_cast")];
            tensor<fp16, [640]> add_65_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_65_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4933597312)))];
            tensor<fp16, [640]> add_65_beta_0_to_fp16 = const()[name = tensor<string, []>("add_65_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4933598656)))];
            tensor<fp16, []> add_65_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_65_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 64, 64]> add_65_cast = batch_norm(beta = add_65_beta_0_to_fp16, epsilon = add_65_epsilon_0_to_fp16, gamma = add_65_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_129_cast)[name = tensor<string, []>("add_65_cast")];
            tensor<int32, [2]> var_58144 = const()[name = tensor<string, []>("op_58144"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58146 = const()[name = tensor<string, []>("op_58146"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_507_pad_type_0 = const()[name = tensor<string, []>("hidden_states_507_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_507_pad_0 = const()[name = tensor<string, []>("hidden_states_507_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_proj_in_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4933600000)))];
            tensor<fp16, [640]> up_blocks_1_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_proj_in_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4934419264)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_507_cast = conv(bias = up_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = var_58146, groups = var_58064, pad = hidden_states_507_pad_0, pad_type = hidden_states_507_pad_type_0, strides = var_58144, weight = up_blocks_1_attentions_0_proj_in_weight_to_fp16, x = add_65_cast)[name = tensor<string, []>("hidden_states_507_cast")];
            tensor<int32, [4]> var_58151 = const()[name = tensor<string, []>("op_58151"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<fp16, [2, 640, 1, 4096]> inputs_385_cast = reshape(shape = var_58151, x = hidden_states_507_cast)[name = tensor<string, []>("inputs_385_cast")];
            tensor<int32, [1]> var_58161 = const()[name = tensor<string, []>("op_58161"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_385_cast = reduce_mean(axes = var_58161, keep_dims = var_58059, x = inputs_385_cast)[name = tensor<string, []>("channels_mean_385_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_385_cast = sub(x = inputs_385_cast, y = channels_mean_385_cast)[name = tensor<string, []>("zero_mean_385_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_385_cast = mul(x = zero_mean_385_cast, y = zero_mean_385_cast)[name = tensor<string, []>("zero_mean_sq_385_cast")];
            tensor<int32, [1]> var_58165 = const()[name = tensor<string, []>("op_58165"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_58166_cast = reduce_mean(axes = var_58165, keep_dims = var_58059, x = zero_mean_sq_385_cast)[name = tensor<string, []>("op_58166_cast")];
            tensor<fp16, []> var_58167_to_fp16 = const()[name = tensor<string, []>("op_58167_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_58168_cast = add(x = var_58166_cast, y = var_58167_to_fp16)[name = tensor<string, []>("op_58168_cast")];
            tensor<fp16, []> denom_385_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_385_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_385_cast = rsqrt(epsilon = denom_385_epsilon_0_to_fp16, x = var_58168_cast)[name = tensor<string, []>("denom_385_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_385_cast = mul(x = zero_mean_385_cast, y = denom_385_cast)[name = tensor<string, []>("out_385_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_58172_to_fp16 = const()[name = tensor<string, []>("op_58172_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4934420608)))];
            tensor<fp16, [2, 640, 1, 4096]> var_58173_cast = add(x = out_385_cast, y = var_58172_to_fp16)[name = tensor<string, []>("op_58173_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_58175_to_fp16 = const()[name = tensor<string, []>("op_58175_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4934421952)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_509_cast = mul(x = var_58173_cast, y = var_58175_to_fp16)[name = tensor<string, []>("hidden_states_509_cast")];
            tensor<int32, [2]> var_58182 = const()[name = tensor<string, []>("op_58182"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58184 = const()[name = tensor<string, []>("op_58184"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_257_pad_type_0 = const()[name = tensor<string, []>("q_257_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_257_pad_0 = const()[name = tensor<string, []>("q_257_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4934423296)))];
            tensor<fp16, [2, 640, 1, 4096]> q_257_cast = conv(dilations = var_58184, groups = var_58064, pad = q_257_pad_0, pad_type = q_257_pad_type_0, strides = var_58182, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_509_cast)[name = tensor<string, []>("q_257_cast")];
            tensor<int32, [2]> var_58188 = const()[name = tensor<string, []>("op_58188"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58190 = const()[name = tensor<string, []>("op_58190"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_513_pad_type_0 = const()[name = tensor<string, []>("k_513_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_513_pad_0 = const()[name = tensor<string, []>("k_513_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4935242560)))];
            tensor<fp16, [2, 640, 1, 4096]> k_513_cast = conv(dilations = var_58190, groups = var_58064, pad = k_513_pad_0, pad_type = k_513_pad_type_0, strides = var_58188, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_509_cast)[name = tensor<string, []>("k_513_cast")];
            tensor<int32, [2]> var_58194 = const()[name = tensor<string, []>("op_58194"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58196 = const()[name = tensor<string, []>("op_58196"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_257_pad_type_0 = const()[name = tensor<string, []>("v_257_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_257_pad_0 = const()[name = tensor<string, []>("v_257_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4936061824)))];
            tensor<fp16, [2, 640, 1, 4096]> v_257_cast = conv(dilations = var_58196, groups = var_58064, pad = v_257_pad_0, pad_type = v_257_pad_type_0, strides = var_58194, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_509_cast)[name = tensor<string, []>("v_257_cast")];
            tensor<int32, [4]> var_58200_begin_0 = const()[name = tensor<string, []>("op_58200_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58200_end_0 = const()[name = tensor<string, []>("op_58200_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_58200_end_mask_0 = const()[name = tensor<string, []>("op_58200_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58200_cast = slice_by_index(begin = var_58200_begin_0, end = var_58200_end_0, end_mask = var_58200_end_mask_0, x = q_257_cast)[name = tensor<string, []>("op_58200_cast")];
            tensor<int32, [4]> var_58204_begin_0 = const()[name = tensor<string, []>("op_58204_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_58204_end_0 = const()[name = tensor<string, []>("op_58204_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_58204_end_mask_0 = const()[name = tensor<string, []>("op_58204_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58204_cast = slice_by_index(begin = var_58204_begin_0, end = var_58204_end_0, end_mask = var_58204_end_mask_0, x = q_257_cast)[name = tensor<string, []>("op_58204_cast")];
            tensor<int32, [4]> var_58208_begin_0 = const()[name = tensor<string, []>("op_58208_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_58208_end_0 = const()[name = tensor<string, []>("op_58208_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_58208_end_mask_0 = const()[name = tensor<string, []>("op_58208_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58208_cast = slice_by_index(begin = var_58208_begin_0, end = var_58208_end_0, end_mask = var_58208_end_mask_0, x = q_257_cast)[name = tensor<string, []>("op_58208_cast")];
            tensor<int32, [4]> var_58212_begin_0 = const()[name = tensor<string, []>("op_58212_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_58212_end_0 = const()[name = tensor<string, []>("op_58212_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_58212_end_mask_0 = const()[name = tensor<string, []>("op_58212_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58212_cast = slice_by_index(begin = var_58212_begin_0, end = var_58212_end_0, end_mask = var_58212_end_mask_0, x = q_257_cast)[name = tensor<string, []>("op_58212_cast")];
            tensor<int32, [4]> var_58216_begin_0 = const()[name = tensor<string, []>("op_58216_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_58216_end_0 = const()[name = tensor<string, []>("op_58216_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_58216_end_mask_0 = const()[name = tensor<string, []>("op_58216_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58216_cast = slice_by_index(begin = var_58216_begin_0, end = var_58216_end_0, end_mask = var_58216_end_mask_0, x = q_257_cast)[name = tensor<string, []>("op_58216_cast")];
            tensor<int32, [4]> var_58220_begin_0 = const()[name = tensor<string, []>("op_58220_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_58220_end_0 = const()[name = tensor<string, []>("op_58220_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_58220_end_mask_0 = const()[name = tensor<string, []>("op_58220_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58220_cast = slice_by_index(begin = var_58220_begin_0, end = var_58220_end_0, end_mask = var_58220_end_mask_0, x = q_257_cast)[name = tensor<string, []>("op_58220_cast")];
            tensor<int32, [4]> var_58224_begin_0 = const()[name = tensor<string, []>("op_58224_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_58224_end_0 = const()[name = tensor<string, []>("op_58224_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_58224_end_mask_0 = const()[name = tensor<string, []>("op_58224_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58224_cast = slice_by_index(begin = var_58224_begin_0, end = var_58224_end_0, end_mask = var_58224_end_mask_0, x = q_257_cast)[name = tensor<string, []>("op_58224_cast")];
            tensor<int32, [4]> var_58228_begin_0 = const()[name = tensor<string, []>("op_58228_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_58228_end_0 = const()[name = tensor<string, []>("op_58228_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_58228_end_mask_0 = const()[name = tensor<string, []>("op_58228_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58228_cast = slice_by_index(begin = var_58228_begin_0, end = var_58228_end_0, end_mask = var_58228_end_mask_0, x = q_257_cast)[name = tensor<string, []>("op_58228_cast")];
            tensor<int32, [4]> var_58232_begin_0 = const()[name = tensor<string, []>("op_58232_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_58232_end_0 = const()[name = tensor<string, []>("op_58232_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_58232_end_mask_0 = const()[name = tensor<string, []>("op_58232_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58232_cast = slice_by_index(begin = var_58232_begin_0, end = var_58232_end_0, end_mask = var_58232_end_mask_0, x = q_257_cast)[name = tensor<string, []>("op_58232_cast")];
            tensor<int32, [4]> var_58236_begin_0 = const()[name = tensor<string, []>("op_58236_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_58236_end_0 = const()[name = tensor<string, []>("op_58236_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_58236_end_mask_0 = const()[name = tensor<string, []>("op_58236_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58236_cast = slice_by_index(begin = var_58236_begin_0, end = var_58236_end_0, end_mask = var_58236_end_mask_0, x = q_257_cast)[name = tensor<string, []>("op_58236_cast")];
            tensor<int32, [4]> k_515_perm_0 = const()[name = tensor<string, []>("k_515_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_58243_begin_0 = const()[name = tensor<string, []>("op_58243_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58243_end_0 = const()[name = tensor<string, []>("op_58243_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_58243_end_mask_0 = const()[name = tensor<string, []>("op_58243_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 640]> transpose_11 = transpose(perm = k_515_perm_0, x = k_513_cast)[name = tensor<string, []>("transpose_11")];
            tensor<fp16, [2, 4096, 1, 64]> var_58243_cast = slice_by_index(begin = var_58243_begin_0, end = var_58243_end_0, end_mask = var_58243_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_58243_cast")];
            tensor<int32, [4]> var_58247_begin_0 = const()[name = tensor<string, []>("op_58247_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_58247_end_0 = const()[name = tensor<string, []>("op_58247_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_58247_end_mask_0 = const()[name = tensor<string, []>("op_58247_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58247_cast = slice_by_index(begin = var_58247_begin_0, end = var_58247_end_0, end_mask = var_58247_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_58247_cast")];
            tensor<int32, [4]> var_58251_begin_0 = const()[name = tensor<string, []>("op_58251_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_58251_end_0 = const()[name = tensor<string, []>("op_58251_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_58251_end_mask_0 = const()[name = tensor<string, []>("op_58251_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58251_cast = slice_by_index(begin = var_58251_begin_0, end = var_58251_end_0, end_mask = var_58251_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_58251_cast")];
            tensor<int32, [4]> var_58255_begin_0 = const()[name = tensor<string, []>("op_58255_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_58255_end_0 = const()[name = tensor<string, []>("op_58255_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_58255_end_mask_0 = const()[name = tensor<string, []>("op_58255_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58255_cast = slice_by_index(begin = var_58255_begin_0, end = var_58255_end_0, end_mask = var_58255_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_58255_cast")];
            tensor<int32, [4]> var_58259_begin_0 = const()[name = tensor<string, []>("op_58259_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_58259_end_0 = const()[name = tensor<string, []>("op_58259_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_58259_end_mask_0 = const()[name = tensor<string, []>("op_58259_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58259_cast = slice_by_index(begin = var_58259_begin_0, end = var_58259_end_0, end_mask = var_58259_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_58259_cast")];
            tensor<int32, [4]> var_58263_begin_0 = const()[name = tensor<string, []>("op_58263_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_58263_end_0 = const()[name = tensor<string, []>("op_58263_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 384])];
            tensor<bool, [4]> var_58263_end_mask_0 = const()[name = tensor<string, []>("op_58263_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58263_cast = slice_by_index(begin = var_58263_begin_0, end = var_58263_end_0, end_mask = var_58263_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_58263_cast")];
            tensor<int32, [4]> var_58267_begin_0 = const()[name = tensor<string, []>("op_58267_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_58267_end_0 = const()[name = tensor<string, []>("op_58267_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 448])];
            tensor<bool, [4]> var_58267_end_mask_0 = const()[name = tensor<string, []>("op_58267_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58267_cast = slice_by_index(begin = var_58267_begin_0, end = var_58267_end_0, end_mask = var_58267_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_58267_cast")];
            tensor<int32, [4]> var_58271_begin_0 = const()[name = tensor<string, []>("op_58271_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_58271_end_0 = const()[name = tensor<string, []>("op_58271_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 512])];
            tensor<bool, [4]> var_58271_end_mask_0 = const()[name = tensor<string, []>("op_58271_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58271_cast = slice_by_index(begin = var_58271_begin_0, end = var_58271_end_0, end_mask = var_58271_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_58271_cast")];
            tensor<int32, [4]> var_58275_begin_0 = const()[name = tensor<string, []>("op_58275_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_58275_end_0 = const()[name = tensor<string, []>("op_58275_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 576])];
            tensor<bool, [4]> var_58275_end_mask_0 = const()[name = tensor<string, []>("op_58275_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58275_cast = slice_by_index(begin = var_58275_begin_0, end = var_58275_end_0, end_mask = var_58275_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_58275_cast")];
            tensor<int32, [4]> var_58279_begin_0 = const()[name = tensor<string, []>("op_58279_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_58279_end_0 = const()[name = tensor<string, []>("op_58279_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 640])];
            tensor<bool, [4]> var_58279_end_mask_0 = const()[name = tensor<string, []>("op_58279_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58279_cast = slice_by_index(begin = var_58279_begin_0, end = var_58279_end_0, end_mask = var_58279_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_58279_cast")];
            tensor<int32, [4]> var_58281_begin_0 = const()[name = tensor<string, []>("op_58281_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58281_end_0 = const()[name = tensor<string, []>("op_58281_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_58281_end_mask_0 = const()[name = tensor<string, []>("op_58281_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58281_cast = slice_by_index(begin = var_58281_begin_0, end = var_58281_end_0, end_mask = var_58281_end_mask_0, x = v_257_cast)[name = tensor<string, []>("op_58281_cast")];
            tensor<int32, [4]> var_58285_begin_0 = const()[name = tensor<string, []>("op_58285_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_58285_end_0 = const()[name = tensor<string, []>("op_58285_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_58285_end_mask_0 = const()[name = tensor<string, []>("op_58285_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58285_cast = slice_by_index(begin = var_58285_begin_0, end = var_58285_end_0, end_mask = var_58285_end_mask_0, x = v_257_cast)[name = tensor<string, []>("op_58285_cast")];
            tensor<int32, [4]> var_58289_begin_0 = const()[name = tensor<string, []>("op_58289_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_58289_end_0 = const()[name = tensor<string, []>("op_58289_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_58289_end_mask_0 = const()[name = tensor<string, []>("op_58289_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58289_cast = slice_by_index(begin = var_58289_begin_0, end = var_58289_end_0, end_mask = var_58289_end_mask_0, x = v_257_cast)[name = tensor<string, []>("op_58289_cast")];
            tensor<int32, [4]> var_58293_begin_0 = const()[name = tensor<string, []>("op_58293_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_58293_end_0 = const()[name = tensor<string, []>("op_58293_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_58293_end_mask_0 = const()[name = tensor<string, []>("op_58293_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58293_cast = slice_by_index(begin = var_58293_begin_0, end = var_58293_end_0, end_mask = var_58293_end_mask_0, x = v_257_cast)[name = tensor<string, []>("op_58293_cast")];
            tensor<int32, [4]> var_58297_begin_0 = const()[name = tensor<string, []>("op_58297_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_58297_end_0 = const()[name = tensor<string, []>("op_58297_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_58297_end_mask_0 = const()[name = tensor<string, []>("op_58297_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58297_cast = slice_by_index(begin = var_58297_begin_0, end = var_58297_end_0, end_mask = var_58297_end_mask_0, x = v_257_cast)[name = tensor<string, []>("op_58297_cast")];
            tensor<int32, [4]> var_58301_begin_0 = const()[name = tensor<string, []>("op_58301_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_58301_end_0 = const()[name = tensor<string, []>("op_58301_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_58301_end_mask_0 = const()[name = tensor<string, []>("op_58301_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58301_cast = slice_by_index(begin = var_58301_begin_0, end = var_58301_end_0, end_mask = var_58301_end_mask_0, x = v_257_cast)[name = tensor<string, []>("op_58301_cast")];
            tensor<int32, [4]> var_58305_begin_0 = const()[name = tensor<string, []>("op_58305_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_58305_end_0 = const()[name = tensor<string, []>("op_58305_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_58305_end_mask_0 = const()[name = tensor<string, []>("op_58305_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58305_cast = slice_by_index(begin = var_58305_begin_0, end = var_58305_end_0, end_mask = var_58305_end_mask_0, x = v_257_cast)[name = tensor<string, []>("op_58305_cast")];
            tensor<int32, [4]> var_58309_begin_0 = const()[name = tensor<string, []>("op_58309_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_58309_end_0 = const()[name = tensor<string, []>("op_58309_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_58309_end_mask_0 = const()[name = tensor<string, []>("op_58309_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58309_cast = slice_by_index(begin = var_58309_begin_0, end = var_58309_end_0, end_mask = var_58309_end_mask_0, x = v_257_cast)[name = tensor<string, []>("op_58309_cast")];
            tensor<int32, [4]> var_58313_begin_0 = const()[name = tensor<string, []>("op_58313_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_58313_end_0 = const()[name = tensor<string, []>("op_58313_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_58313_end_mask_0 = const()[name = tensor<string, []>("op_58313_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58313_cast = slice_by_index(begin = var_58313_begin_0, end = var_58313_end_0, end_mask = var_58313_end_mask_0, x = v_257_cast)[name = tensor<string, []>("op_58313_cast")];
            tensor<int32, [4]> var_58317_begin_0 = const()[name = tensor<string, []>("op_58317_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_58317_end_0 = const()[name = tensor<string, []>("op_58317_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_58317_end_mask_0 = const()[name = tensor<string, []>("op_58317_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58317_cast = slice_by_index(begin = var_58317_begin_0, end = var_58317_end_0, end_mask = var_58317_end_mask_0, x = v_257_cast)[name = tensor<string, []>("op_58317_cast")];
            tensor<string, []> var_58321_equation_0 = const()[name = tensor<string, []>("op_58321_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58321_cast = einsum(equation = var_58321_equation_0, values = (var_58243_cast, var_58200_cast))[name = tensor<string, []>("op_58321_cast")];
            tensor<fp16, []> var_58322_to_fp16 = const()[name = tensor<string, []>("op_58322_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4961_cast = mul(x = var_58321_cast, y = var_58322_to_fp16)[name = tensor<string, []>("aw_4961_cast")];
            tensor<string, []> var_58325_equation_0 = const()[name = tensor<string, []>("op_58325_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58325_cast = einsum(equation = var_58325_equation_0, values = (var_58247_cast, var_58204_cast))[name = tensor<string, []>("op_58325_cast")];
            tensor<fp16, []> var_58326_to_fp16 = const()[name = tensor<string, []>("op_58326_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4963_cast = mul(x = var_58325_cast, y = var_58326_to_fp16)[name = tensor<string, []>("aw_4963_cast")];
            tensor<string, []> var_58329_equation_0 = const()[name = tensor<string, []>("op_58329_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58329_cast = einsum(equation = var_58329_equation_0, values = (var_58251_cast, var_58208_cast))[name = tensor<string, []>("op_58329_cast")];
            tensor<fp16, []> var_58330_to_fp16 = const()[name = tensor<string, []>("op_58330_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4965_cast = mul(x = var_58329_cast, y = var_58330_to_fp16)[name = tensor<string, []>("aw_4965_cast")];
            tensor<string, []> var_58333_equation_0 = const()[name = tensor<string, []>("op_58333_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58333_cast = einsum(equation = var_58333_equation_0, values = (var_58255_cast, var_58212_cast))[name = tensor<string, []>("op_58333_cast")];
            tensor<fp16, []> var_58334_to_fp16 = const()[name = tensor<string, []>("op_58334_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4967_cast = mul(x = var_58333_cast, y = var_58334_to_fp16)[name = tensor<string, []>("aw_4967_cast")];
            tensor<string, []> var_58337_equation_0 = const()[name = tensor<string, []>("op_58337_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58337_cast = einsum(equation = var_58337_equation_0, values = (var_58259_cast, var_58216_cast))[name = tensor<string, []>("op_58337_cast")];
            tensor<fp16, []> var_58338_to_fp16 = const()[name = tensor<string, []>("op_58338_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4969_cast = mul(x = var_58337_cast, y = var_58338_to_fp16)[name = tensor<string, []>("aw_4969_cast")];
            tensor<string, []> var_58341_equation_0 = const()[name = tensor<string, []>("op_58341_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58341_cast = einsum(equation = var_58341_equation_0, values = (var_58263_cast, var_58220_cast))[name = tensor<string, []>("op_58341_cast")];
            tensor<fp16, []> var_58342_to_fp16 = const()[name = tensor<string, []>("op_58342_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4971_cast = mul(x = var_58341_cast, y = var_58342_to_fp16)[name = tensor<string, []>("aw_4971_cast")];
            tensor<string, []> var_58345_equation_0 = const()[name = tensor<string, []>("op_58345_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58345_cast = einsum(equation = var_58345_equation_0, values = (var_58267_cast, var_58224_cast))[name = tensor<string, []>("op_58345_cast")];
            tensor<fp16, []> var_58346_to_fp16 = const()[name = tensor<string, []>("op_58346_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4973_cast = mul(x = var_58345_cast, y = var_58346_to_fp16)[name = tensor<string, []>("aw_4973_cast")];
            tensor<string, []> var_58349_equation_0 = const()[name = tensor<string, []>("op_58349_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58349_cast = einsum(equation = var_58349_equation_0, values = (var_58271_cast, var_58228_cast))[name = tensor<string, []>("op_58349_cast")];
            tensor<fp16, []> var_58350_to_fp16 = const()[name = tensor<string, []>("op_58350_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4975_cast = mul(x = var_58349_cast, y = var_58350_to_fp16)[name = tensor<string, []>("aw_4975_cast")];
            tensor<string, []> var_58353_equation_0 = const()[name = tensor<string, []>("op_58353_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58353_cast = einsum(equation = var_58353_equation_0, values = (var_58275_cast, var_58232_cast))[name = tensor<string, []>("op_58353_cast")];
            tensor<fp16, []> var_58354_to_fp16 = const()[name = tensor<string, []>("op_58354_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4977_cast = mul(x = var_58353_cast, y = var_58354_to_fp16)[name = tensor<string, []>("aw_4977_cast")];
            tensor<string, []> var_58357_equation_0 = const()[name = tensor<string, []>("op_58357_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58357_cast = einsum(equation = var_58357_equation_0, values = (var_58279_cast, var_58236_cast))[name = tensor<string, []>("op_58357_cast")];
            tensor<fp16, []> var_58358_to_fp16 = const()[name = tensor<string, []>("op_58358_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_4979_cast = mul(x = var_58357_cast, y = var_58358_to_fp16)[name = tensor<string, []>("aw_4979_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58360_cast = softmax(axis = var_58064, x = aw_4961_cast)[name = tensor<string, []>("op_58360_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58361_cast = softmax(axis = var_58064, x = aw_4963_cast)[name = tensor<string, []>("op_58361_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58362_cast = softmax(axis = var_58064, x = aw_4965_cast)[name = tensor<string, []>("op_58362_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58363_cast = softmax(axis = var_58064, x = aw_4967_cast)[name = tensor<string, []>("op_58363_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58364_cast = softmax(axis = var_58064, x = aw_4969_cast)[name = tensor<string, []>("op_58364_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58365_cast = softmax(axis = var_58064, x = aw_4971_cast)[name = tensor<string, []>("op_58365_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58366_cast = softmax(axis = var_58064, x = aw_4973_cast)[name = tensor<string, []>("op_58366_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58367_cast = softmax(axis = var_58064, x = aw_4975_cast)[name = tensor<string, []>("op_58367_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58368_cast = softmax(axis = var_58064, x = aw_4977_cast)[name = tensor<string, []>("op_58368_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58369_cast = softmax(axis = var_58064, x = aw_4979_cast)[name = tensor<string, []>("op_58369_cast")];
            tensor<string, []> var_58371_equation_0 = const()[name = tensor<string, []>("op_58371_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58371_cast = einsum(equation = var_58371_equation_0, values = (var_58281_cast, var_58360_cast))[name = tensor<string, []>("op_58371_cast")];
            tensor<string, []> var_58373_equation_0 = const()[name = tensor<string, []>("op_58373_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58373_cast = einsum(equation = var_58373_equation_0, values = (var_58285_cast, var_58361_cast))[name = tensor<string, []>("op_58373_cast")];
            tensor<string, []> var_58375_equation_0 = const()[name = tensor<string, []>("op_58375_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58375_cast = einsum(equation = var_58375_equation_0, values = (var_58289_cast, var_58362_cast))[name = tensor<string, []>("op_58375_cast")];
            tensor<string, []> var_58377_equation_0 = const()[name = tensor<string, []>("op_58377_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58377_cast = einsum(equation = var_58377_equation_0, values = (var_58293_cast, var_58363_cast))[name = tensor<string, []>("op_58377_cast")];
            tensor<string, []> var_58379_equation_0 = const()[name = tensor<string, []>("op_58379_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58379_cast = einsum(equation = var_58379_equation_0, values = (var_58297_cast, var_58364_cast))[name = tensor<string, []>("op_58379_cast")];
            tensor<string, []> var_58381_equation_0 = const()[name = tensor<string, []>("op_58381_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58381_cast = einsum(equation = var_58381_equation_0, values = (var_58301_cast, var_58365_cast))[name = tensor<string, []>("op_58381_cast")];
            tensor<string, []> var_58383_equation_0 = const()[name = tensor<string, []>("op_58383_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58383_cast = einsum(equation = var_58383_equation_0, values = (var_58305_cast, var_58366_cast))[name = tensor<string, []>("op_58383_cast")];
            tensor<string, []> var_58385_equation_0 = const()[name = tensor<string, []>("op_58385_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58385_cast = einsum(equation = var_58385_equation_0, values = (var_58309_cast, var_58367_cast))[name = tensor<string, []>("op_58385_cast")];
            tensor<string, []> var_58387_equation_0 = const()[name = tensor<string, []>("op_58387_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58387_cast = einsum(equation = var_58387_equation_0, values = (var_58313_cast, var_58368_cast))[name = tensor<string, []>("op_58387_cast")];
            tensor<string, []> var_58389_equation_0 = const()[name = tensor<string, []>("op_58389_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58389_cast = einsum(equation = var_58389_equation_0, values = (var_58317_cast, var_58369_cast))[name = tensor<string, []>("op_58389_cast")];
            tensor<bool, []> input_739_interleave_0 = const()[name = tensor<string, []>("input_739_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_739_cast = concat(axis = var_58064, interleave = input_739_interleave_0, values = (var_58371_cast, var_58373_cast, var_58375_cast, var_58377_cast, var_58379_cast, var_58381_cast, var_58383_cast, var_58385_cast, var_58387_cast, var_58389_cast))[name = tensor<string, []>("input_739_cast")];
            tensor<int32, [2]> var_58395 = const()[name = tensor<string, []>("op_58395"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58397 = const()[name = tensor<string, []>("op_58397"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_58399_pad_type_0 = const()[name = tensor<string, []>("op_58399_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_58399_pad_0 = const()[name = tensor<string, []>("op_58399_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4936881088)))];
            tensor<fp16, [640]> up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4937700352)))];
            tensor<fp16, [2, 640, 1, 4096]> var_58399_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_58397, groups = var_58064, pad = var_58399_pad_0, pad_type = var_58399_pad_type_0, strides = var_58395, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_739_cast)[name = tensor<string, []>("op_58399_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_387_cast = add(x = var_58399_cast, y = inputs_385_cast)[name = tensor<string, []>("inputs_387_cast")];
            tensor<int32, [1]> var_58403 = const()[name = tensor<string, []>("op_58403"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_387_cast = reduce_mean(axes = var_58403, keep_dims = var_58059, x = inputs_387_cast)[name = tensor<string, []>("channels_mean_387_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_387_cast = sub(x = inputs_387_cast, y = channels_mean_387_cast)[name = tensor<string, []>("zero_mean_387_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_387_cast = mul(x = zero_mean_387_cast, y = zero_mean_387_cast)[name = tensor<string, []>("zero_mean_sq_387_cast")];
            tensor<int32, [1]> var_58407 = const()[name = tensor<string, []>("op_58407"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_58408_cast = reduce_mean(axes = var_58407, keep_dims = var_58059, x = zero_mean_sq_387_cast)[name = tensor<string, []>("op_58408_cast")];
            tensor<fp16, []> var_58409_to_fp16 = const()[name = tensor<string, []>("op_58409_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_58410_cast = add(x = var_58408_cast, y = var_58409_to_fp16)[name = tensor<string, []>("op_58410_cast")];
            tensor<fp16, []> denom_387_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_387_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_387_cast = rsqrt(epsilon = denom_387_epsilon_0_to_fp16, x = var_58410_cast)[name = tensor<string, []>("denom_387_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_387_cast = mul(x = zero_mean_387_cast, y = denom_387_cast)[name = tensor<string, []>("out_387_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_58414_to_fp16 = const()[name = tensor<string, []>("op_58414_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4937701696)))];
            tensor<fp16, [2, 640, 1, 4096]> var_58415_cast = add(x = out_387_cast, y = var_58414_to_fp16)[name = tensor<string, []>("op_58415_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_58417_to_fp16 = const()[name = tensor<string, []>("op_58417_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4937703040)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_511_cast = mul(x = var_58415_cast, y = var_58417_to_fp16)[name = tensor<string, []>("hidden_states_511_cast")];
            tensor<int32, [2]> var_58424 = const()[name = tensor<string, []>("op_58424"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58426 = const()[name = tensor<string, []>("op_58426"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_259_pad_type_0 = const()[name = tensor<string, []>("q_259_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_259_pad_0 = const()[name = tensor<string, []>("q_259_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4937704384)))];
            tensor<fp16, [2, 640, 1, 4096]> q_259_cast = conv(dilations = var_58426, groups = var_58064, pad = q_259_pad_0, pad_type = q_259_pad_type_0, strides = var_58424, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_511_cast)[name = tensor<string, []>("q_259_cast")];
            tensor<int32, [2]> var_58430 = const()[name = tensor<string, []>("op_58430"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58432 = const()[name = tensor<string, []>("op_58432"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_517_pad_type_0 = const()[name = tensor<string, []>("k_517_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_517_pad_0 = const()[name = tensor<string, []>("k_517_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4938523648)))];
            tensor<fp16, [2, 640, 1, 77]> k_517_cast = conv(dilations = var_58432, groups = var_58064, pad = k_517_pad_0, pad_type = k_517_pad_type_0, strides = var_58430, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_517_cast")];
            tensor<int32, [2]> var_58436 = const()[name = tensor<string, []>("op_58436"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58438 = const()[name = tensor<string, []>("op_58438"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_259_pad_type_0 = const()[name = tensor<string, []>("v_259_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_259_pad_0 = const()[name = tensor<string, []>("v_259_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4941145152)))];
            tensor<fp16, [2, 640, 1, 77]> v_259_cast = conv(dilations = var_58438, groups = var_58064, pad = v_259_pad_0, pad_type = v_259_pad_type_0, strides = var_58436, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_259_cast")];
            tensor<int32, [4]> var_58442_begin_0 = const()[name = tensor<string, []>("op_58442_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58442_end_0 = const()[name = tensor<string, []>("op_58442_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_58442_end_mask_0 = const()[name = tensor<string, []>("op_58442_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58442_cast = slice_by_index(begin = var_58442_begin_0, end = var_58442_end_0, end_mask = var_58442_end_mask_0, x = q_259_cast)[name = tensor<string, []>("op_58442_cast")];
            tensor<int32, [4]> var_58446_begin_0 = const()[name = tensor<string, []>("op_58446_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_58446_end_0 = const()[name = tensor<string, []>("op_58446_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_58446_end_mask_0 = const()[name = tensor<string, []>("op_58446_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58446_cast = slice_by_index(begin = var_58446_begin_0, end = var_58446_end_0, end_mask = var_58446_end_mask_0, x = q_259_cast)[name = tensor<string, []>("op_58446_cast")];
            tensor<int32, [4]> var_58450_begin_0 = const()[name = tensor<string, []>("op_58450_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_58450_end_0 = const()[name = tensor<string, []>("op_58450_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_58450_end_mask_0 = const()[name = tensor<string, []>("op_58450_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58450_cast = slice_by_index(begin = var_58450_begin_0, end = var_58450_end_0, end_mask = var_58450_end_mask_0, x = q_259_cast)[name = tensor<string, []>("op_58450_cast")];
            tensor<int32, [4]> var_58454_begin_0 = const()[name = tensor<string, []>("op_58454_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_58454_end_0 = const()[name = tensor<string, []>("op_58454_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_58454_end_mask_0 = const()[name = tensor<string, []>("op_58454_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58454_cast = slice_by_index(begin = var_58454_begin_0, end = var_58454_end_0, end_mask = var_58454_end_mask_0, x = q_259_cast)[name = tensor<string, []>("op_58454_cast")];
            tensor<int32, [4]> var_58458_begin_0 = const()[name = tensor<string, []>("op_58458_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_58458_end_0 = const()[name = tensor<string, []>("op_58458_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_58458_end_mask_0 = const()[name = tensor<string, []>("op_58458_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58458_cast = slice_by_index(begin = var_58458_begin_0, end = var_58458_end_0, end_mask = var_58458_end_mask_0, x = q_259_cast)[name = tensor<string, []>("op_58458_cast")];
            tensor<int32, [4]> var_58462_begin_0 = const()[name = tensor<string, []>("op_58462_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_58462_end_0 = const()[name = tensor<string, []>("op_58462_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_58462_end_mask_0 = const()[name = tensor<string, []>("op_58462_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58462_cast = slice_by_index(begin = var_58462_begin_0, end = var_58462_end_0, end_mask = var_58462_end_mask_0, x = q_259_cast)[name = tensor<string, []>("op_58462_cast")];
            tensor<int32, [4]> var_58466_begin_0 = const()[name = tensor<string, []>("op_58466_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_58466_end_0 = const()[name = tensor<string, []>("op_58466_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_58466_end_mask_0 = const()[name = tensor<string, []>("op_58466_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58466_cast = slice_by_index(begin = var_58466_begin_0, end = var_58466_end_0, end_mask = var_58466_end_mask_0, x = q_259_cast)[name = tensor<string, []>("op_58466_cast")];
            tensor<int32, [4]> var_58470_begin_0 = const()[name = tensor<string, []>("op_58470_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_58470_end_0 = const()[name = tensor<string, []>("op_58470_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_58470_end_mask_0 = const()[name = tensor<string, []>("op_58470_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58470_cast = slice_by_index(begin = var_58470_begin_0, end = var_58470_end_0, end_mask = var_58470_end_mask_0, x = q_259_cast)[name = tensor<string, []>("op_58470_cast")];
            tensor<int32, [4]> var_58474_begin_0 = const()[name = tensor<string, []>("op_58474_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_58474_end_0 = const()[name = tensor<string, []>("op_58474_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_58474_end_mask_0 = const()[name = tensor<string, []>("op_58474_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58474_cast = slice_by_index(begin = var_58474_begin_0, end = var_58474_end_0, end_mask = var_58474_end_mask_0, x = q_259_cast)[name = tensor<string, []>("op_58474_cast")];
            tensor<int32, [4]> var_58478_begin_0 = const()[name = tensor<string, []>("op_58478_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_58478_end_0 = const()[name = tensor<string, []>("op_58478_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_58478_end_mask_0 = const()[name = tensor<string, []>("op_58478_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58478_cast = slice_by_index(begin = var_58478_begin_0, end = var_58478_end_0, end_mask = var_58478_end_mask_0, x = q_259_cast)[name = tensor<string, []>("op_58478_cast")];
            tensor<int32, [4]> k_519_perm_0 = const()[name = tensor<string, []>("k_519_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_58485_begin_0 = const()[name = tensor<string, []>("op_58485_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58485_end_0 = const()[name = tensor<string, []>("op_58485_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_58485_end_mask_0 = const()[name = tensor<string, []>("op_58485_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> transpose_10 = transpose(perm = k_519_perm_0, x = k_517_cast)[name = tensor<string, []>("transpose_10")];
            tensor<fp16, [2, 77, 1, 64]> var_58485_cast = slice_by_index(begin = var_58485_begin_0, end = var_58485_end_0, end_mask = var_58485_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_58485_cast")];
            tensor<int32, [4]> var_58489_begin_0 = const()[name = tensor<string, []>("op_58489_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_58489_end_0 = const()[name = tensor<string, []>("op_58489_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_58489_end_mask_0 = const()[name = tensor<string, []>("op_58489_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58489_cast = slice_by_index(begin = var_58489_begin_0, end = var_58489_end_0, end_mask = var_58489_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_58489_cast")];
            tensor<int32, [4]> var_58493_begin_0 = const()[name = tensor<string, []>("op_58493_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_58493_end_0 = const()[name = tensor<string, []>("op_58493_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_58493_end_mask_0 = const()[name = tensor<string, []>("op_58493_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58493_cast = slice_by_index(begin = var_58493_begin_0, end = var_58493_end_0, end_mask = var_58493_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_58493_cast")];
            tensor<int32, [4]> var_58497_begin_0 = const()[name = tensor<string, []>("op_58497_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_58497_end_0 = const()[name = tensor<string, []>("op_58497_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_58497_end_mask_0 = const()[name = tensor<string, []>("op_58497_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58497_cast = slice_by_index(begin = var_58497_begin_0, end = var_58497_end_0, end_mask = var_58497_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_58497_cast")];
            tensor<int32, [4]> var_58501_begin_0 = const()[name = tensor<string, []>("op_58501_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_58501_end_0 = const()[name = tensor<string, []>("op_58501_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_58501_end_mask_0 = const()[name = tensor<string, []>("op_58501_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58501_cast = slice_by_index(begin = var_58501_begin_0, end = var_58501_end_0, end_mask = var_58501_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_58501_cast")];
            tensor<int32, [4]> var_58505_begin_0 = const()[name = tensor<string, []>("op_58505_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_58505_end_0 = const()[name = tensor<string, []>("op_58505_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_58505_end_mask_0 = const()[name = tensor<string, []>("op_58505_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58505_cast = slice_by_index(begin = var_58505_begin_0, end = var_58505_end_0, end_mask = var_58505_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_58505_cast")];
            tensor<int32, [4]> var_58509_begin_0 = const()[name = tensor<string, []>("op_58509_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_58509_end_0 = const()[name = tensor<string, []>("op_58509_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_58509_end_mask_0 = const()[name = tensor<string, []>("op_58509_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58509_cast = slice_by_index(begin = var_58509_begin_0, end = var_58509_end_0, end_mask = var_58509_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_58509_cast")];
            tensor<int32, [4]> var_58513_begin_0 = const()[name = tensor<string, []>("op_58513_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_58513_end_0 = const()[name = tensor<string, []>("op_58513_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_58513_end_mask_0 = const()[name = tensor<string, []>("op_58513_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58513_cast = slice_by_index(begin = var_58513_begin_0, end = var_58513_end_0, end_mask = var_58513_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_58513_cast")];
            tensor<int32, [4]> var_58517_begin_0 = const()[name = tensor<string, []>("op_58517_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_58517_end_0 = const()[name = tensor<string, []>("op_58517_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_58517_end_mask_0 = const()[name = tensor<string, []>("op_58517_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58517_cast = slice_by_index(begin = var_58517_begin_0, end = var_58517_end_0, end_mask = var_58517_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_58517_cast")];
            tensor<int32, [4]> var_58521_begin_0 = const()[name = tensor<string, []>("op_58521_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_58521_end_0 = const()[name = tensor<string, []>("op_58521_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_58521_end_mask_0 = const()[name = tensor<string, []>("op_58521_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_58521_cast = slice_by_index(begin = var_58521_begin_0, end = var_58521_end_0, end_mask = var_58521_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_58521_cast")];
            tensor<int32, [4]> var_58523_begin_0 = const()[name = tensor<string, []>("op_58523_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58523_end_0 = const()[name = tensor<string, []>("op_58523_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_58523_end_mask_0 = const()[name = tensor<string, []>("op_58523_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58523_cast = slice_by_index(begin = var_58523_begin_0, end = var_58523_end_0, end_mask = var_58523_end_mask_0, x = v_259_cast)[name = tensor<string, []>("op_58523_cast")];
            tensor<int32, [4]> var_58527_begin_0 = const()[name = tensor<string, []>("op_58527_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_58527_end_0 = const()[name = tensor<string, []>("op_58527_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_58527_end_mask_0 = const()[name = tensor<string, []>("op_58527_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58527_cast = slice_by_index(begin = var_58527_begin_0, end = var_58527_end_0, end_mask = var_58527_end_mask_0, x = v_259_cast)[name = tensor<string, []>("op_58527_cast")];
            tensor<int32, [4]> var_58531_begin_0 = const()[name = tensor<string, []>("op_58531_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_58531_end_0 = const()[name = tensor<string, []>("op_58531_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_58531_end_mask_0 = const()[name = tensor<string, []>("op_58531_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58531_cast = slice_by_index(begin = var_58531_begin_0, end = var_58531_end_0, end_mask = var_58531_end_mask_0, x = v_259_cast)[name = tensor<string, []>("op_58531_cast")];
            tensor<int32, [4]> var_58535_begin_0 = const()[name = tensor<string, []>("op_58535_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_58535_end_0 = const()[name = tensor<string, []>("op_58535_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_58535_end_mask_0 = const()[name = tensor<string, []>("op_58535_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58535_cast = slice_by_index(begin = var_58535_begin_0, end = var_58535_end_0, end_mask = var_58535_end_mask_0, x = v_259_cast)[name = tensor<string, []>("op_58535_cast")];
            tensor<int32, [4]> var_58539_begin_0 = const()[name = tensor<string, []>("op_58539_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_58539_end_0 = const()[name = tensor<string, []>("op_58539_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_58539_end_mask_0 = const()[name = tensor<string, []>("op_58539_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58539_cast = slice_by_index(begin = var_58539_begin_0, end = var_58539_end_0, end_mask = var_58539_end_mask_0, x = v_259_cast)[name = tensor<string, []>("op_58539_cast")];
            tensor<int32, [4]> var_58543_begin_0 = const()[name = tensor<string, []>("op_58543_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_58543_end_0 = const()[name = tensor<string, []>("op_58543_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_58543_end_mask_0 = const()[name = tensor<string, []>("op_58543_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58543_cast = slice_by_index(begin = var_58543_begin_0, end = var_58543_end_0, end_mask = var_58543_end_mask_0, x = v_259_cast)[name = tensor<string, []>("op_58543_cast")];
            tensor<int32, [4]> var_58547_begin_0 = const()[name = tensor<string, []>("op_58547_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_58547_end_0 = const()[name = tensor<string, []>("op_58547_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_58547_end_mask_0 = const()[name = tensor<string, []>("op_58547_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58547_cast = slice_by_index(begin = var_58547_begin_0, end = var_58547_end_0, end_mask = var_58547_end_mask_0, x = v_259_cast)[name = tensor<string, []>("op_58547_cast")];
            tensor<int32, [4]> var_58551_begin_0 = const()[name = tensor<string, []>("op_58551_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_58551_end_0 = const()[name = tensor<string, []>("op_58551_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_58551_end_mask_0 = const()[name = tensor<string, []>("op_58551_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58551_cast = slice_by_index(begin = var_58551_begin_0, end = var_58551_end_0, end_mask = var_58551_end_mask_0, x = v_259_cast)[name = tensor<string, []>("op_58551_cast")];
            tensor<int32, [4]> var_58555_begin_0 = const()[name = tensor<string, []>("op_58555_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_58555_end_0 = const()[name = tensor<string, []>("op_58555_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_58555_end_mask_0 = const()[name = tensor<string, []>("op_58555_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58555_cast = slice_by_index(begin = var_58555_begin_0, end = var_58555_end_0, end_mask = var_58555_end_mask_0, x = v_259_cast)[name = tensor<string, []>("op_58555_cast")];
            tensor<int32, [4]> var_58559_begin_0 = const()[name = tensor<string, []>("op_58559_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_58559_end_0 = const()[name = tensor<string, []>("op_58559_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_58559_end_mask_0 = const()[name = tensor<string, []>("op_58559_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_58559_cast = slice_by_index(begin = var_58559_begin_0, end = var_58559_end_0, end_mask = var_58559_end_mask_0, x = v_259_cast)[name = tensor<string, []>("op_58559_cast")];
            tensor<string, []> var_58563_equation_0 = const()[name = tensor<string, []>("op_58563_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58563_cast = einsum(equation = var_58563_equation_0, values = (var_58485_cast, var_58442_cast))[name = tensor<string, []>("op_58563_cast")];
            tensor<fp16, []> var_58564_to_fp16 = const()[name = tensor<string, []>("op_58564_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4981_cast = mul(x = var_58563_cast, y = var_58564_to_fp16)[name = tensor<string, []>("aw_4981_cast")];
            tensor<string, []> var_58567_equation_0 = const()[name = tensor<string, []>("op_58567_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58567_cast = einsum(equation = var_58567_equation_0, values = (var_58489_cast, var_58446_cast))[name = tensor<string, []>("op_58567_cast")];
            tensor<fp16, []> var_58568_to_fp16 = const()[name = tensor<string, []>("op_58568_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4983_cast = mul(x = var_58567_cast, y = var_58568_to_fp16)[name = tensor<string, []>("aw_4983_cast")];
            tensor<string, []> var_58571_equation_0 = const()[name = tensor<string, []>("op_58571_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58571_cast = einsum(equation = var_58571_equation_0, values = (var_58493_cast, var_58450_cast))[name = tensor<string, []>("op_58571_cast")];
            tensor<fp16, []> var_58572_to_fp16 = const()[name = tensor<string, []>("op_58572_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4985_cast = mul(x = var_58571_cast, y = var_58572_to_fp16)[name = tensor<string, []>("aw_4985_cast")];
            tensor<string, []> var_58575_equation_0 = const()[name = tensor<string, []>("op_58575_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58575_cast = einsum(equation = var_58575_equation_0, values = (var_58497_cast, var_58454_cast))[name = tensor<string, []>("op_58575_cast")];
            tensor<fp16, []> var_58576_to_fp16 = const()[name = tensor<string, []>("op_58576_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4987_cast = mul(x = var_58575_cast, y = var_58576_to_fp16)[name = tensor<string, []>("aw_4987_cast")];
            tensor<string, []> var_58579_equation_0 = const()[name = tensor<string, []>("op_58579_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58579_cast = einsum(equation = var_58579_equation_0, values = (var_58501_cast, var_58458_cast))[name = tensor<string, []>("op_58579_cast")];
            tensor<fp16, []> var_58580_to_fp16 = const()[name = tensor<string, []>("op_58580_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4989_cast = mul(x = var_58579_cast, y = var_58580_to_fp16)[name = tensor<string, []>("aw_4989_cast")];
            tensor<string, []> var_58583_equation_0 = const()[name = tensor<string, []>("op_58583_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58583_cast = einsum(equation = var_58583_equation_0, values = (var_58505_cast, var_58462_cast))[name = tensor<string, []>("op_58583_cast")];
            tensor<fp16, []> var_58584_to_fp16 = const()[name = tensor<string, []>("op_58584_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4991_cast = mul(x = var_58583_cast, y = var_58584_to_fp16)[name = tensor<string, []>("aw_4991_cast")];
            tensor<string, []> var_58587_equation_0 = const()[name = tensor<string, []>("op_58587_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58587_cast = einsum(equation = var_58587_equation_0, values = (var_58509_cast, var_58466_cast))[name = tensor<string, []>("op_58587_cast")];
            tensor<fp16, []> var_58588_to_fp16 = const()[name = tensor<string, []>("op_58588_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4993_cast = mul(x = var_58587_cast, y = var_58588_to_fp16)[name = tensor<string, []>("aw_4993_cast")];
            tensor<string, []> var_58591_equation_0 = const()[name = tensor<string, []>("op_58591_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58591_cast = einsum(equation = var_58591_equation_0, values = (var_58513_cast, var_58470_cast))[name = tensor<string, []>("op_58591_cast")];
            tensor<fp16, []> var_58592_to_fp16 = const()[name = tensor<string, []>("op_58592_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4995_cast = mul(x = var_58591_cast, y = var_58592_to_fp16)[name = tensor<string, []>("aw_4995_cast")];
            tensor<string, []> var_58595_equation_0 = const()[name = tensor<string, []>("op_58595_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58595_cast = einsum(equation = var_58595_equation_0, values = (var_58517_cast, var_58474_cast))[name = tensor<string, []>("op_58595_cast")];
            tensor<fp16, []> var_58596_to_fp16 = const()[name = tensor<string, []>("op_58596_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4997_cast = mul(x = var_58595_cast, y = var_58596_to_fp16)[name = tensor<string, []>("aw_4997_cast")];
            tensor<string, []> var_58599_equation_0 = const()[name = tensor<string, []>("op_58599_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_58599_cast = einsum(equation = var_58599_equation_0, values = (var_58521_cast, var_58478_cast))[name = tensor<string, []>("op_58599_cast")];
            tensor<fp16, []> var_58600_to_fp16 = const()[name = tensor<string, []>("op_58600_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_4999_cast = mul(x = var_58599_cast, y = var_58600_to_fp16)[name = tensor<string, []>("aw_4999_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_58602_cast = softmax(axis = var_58064, x = aw_4981_cast)[name = tensor<string, []>("op_58602_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_58603_cast = softmax(axis = var_58064, x = aw_4983_cast)[name = tensor<string, []>("op_58603_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_58604_cast = softmax(axis = var_58064, x = aw_4985_cast)[name = tensor<string, []>("op_58604_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_58605_cast = softmax(axis = var_58064, x = aw_4987_cast)[name = tensor<string, []>("op_58605_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_58606_cast = softmax(axis = var_58064, x = aw_4989_cast)[name = tensor<string, []>("op_58606_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_58607_cast = softmax(axis = var_58064, x = aw_4991_cast)[name = tensor<string, []>("op_58607_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_58608_cast = softmax(axis = var_58064, x = aw_4993_cast)[name = tensor<string, []>("op_58608_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_58609_cast = softmax(axis = var_58064, x = aw_4995_cast)[name = tensor<string, []>("op_58609_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_58610_cast = softmax(axis = var_58064, x = aw_4997_cast)[name = tensor<string, []>("op_58610_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_58611_cast = softmax(axis = var_58064, x = aw_4999_cast)[name = tensor<string, []>("op_58611_cast")];
            tensor<string, []> var_58613_equation_0 = const()[name = tensor<string, []>("op_58613_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58613_cast = einsum(equation = var_58613_equation_0, values = (var_58523_cast, var_58602_cast))[name = tensor<string, []>("op_58613_cast")];
            tensor<string, []> var_58615_equation_0 = const()[name = tensor<string, []>("op_58615_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58615_cast = einsum(equation = var_58615_equation_0, values = (var_58527_cast, var_58603_cast))[name = tensor<string, []>("op_58615_cast")];
            tensor<string, []> var_58617_equation_0 = const()[name = tensor<string, []>("op_58617_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58617_cast = einsum(equation = var_58617_equation_0, values = (var_58531_cast, var_58604_cast))[name = tensor<string, []>("op_58617_cast")];
            tensor<string, []> var_58619_equation_0 = const()[name = tensor<string, []>("op_58619_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58619_cast = einsum(equation = var_58619_equation_0, values = (var_58535_cast, var_58605_cast))[name = tensor<string, []>("op_58619_cast")];
            tensor<string, []> var_58621_equation_0 = const()[name = tensor<string, []>("op_58621_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58621_cast = einsum(equation = var_58621_equation_0, values = (var_58539_cast, var_58606_cast))[name = tensor<string, []>("op_58621_cast")];
            tensor<string, []> var_58623_equation_0 = const()[name = tensor<string, []>("op_58623_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58623_cast = einsum(equation = var_58623_equation_0, values = (var_58543_cast, var_58607_cast))[name = tensor<string, []>("op_58623_cast")];
            tensor<string, []> var_58625_equation_0 = const()[name = tensor<string, []>("op_58625_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58625_cast = einsum(equation = var_58625_equation_0, values = (var_58547_cast, var_58608_cast))[name = tensor<string, []>("op_58625_cast")];
            tensor<string, []> var_58627_equation_0 = const()[name = tensor<string, []>("op_58627_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58627_cast = einsum(equation = var_58627_equation_0, values = (var_58551_cast, var_58609_cast))[name = tensor<string, []>("op_58627_cast")];
            tensor<string, []> var_58629_equation_0 = const()[name = tensor<string, []>("op_58629_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58629_cast = einsum(equation = var_58629_equation_0, values = (var_58555_cast, var_58610_cast))[name = tensor<string, []>("op_58629_cast")];
            tensor<string, []> var_58631_equation_0 = const()[name = tensor<string, []>("op_58631_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58631_cast = einsum(equation = var_58631_equation_0, values = (var_58559_cast, var_58611_cast))[name = tensor<string, []>("op_58631_cast")];
            tensor<bool, []> input_741_interleave_0 = const()[name = tensor<string, []>("input_741_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_741_cast = concat(axis = var_58064, interleave = input_741_interleave_0, values = (var_58613_cast, var_58615_cast, var_58617_cast, var_58619_cast, var_58621_cast, var_58623_cast, var_58625_cast, var_58627_cast, var_58629_cast, var_58631_cast))[name = tensor<string, []>("input_741_cast")];
            tensor<int32, [2]> var_58637 = const()[name = tensor<string, []>("op_58637"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58639 = const()[name = tensor<string, []>("op_58639"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_58641_pad_type_0 = const()[name = tensor<string, []>("op_58641_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_58641_pad_0 = const()[name = tensor<string, []>("op_58641_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4943766656)))];
            tensor<fp16, [640]> up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4944585920)))];
            tensor<fp16, [2, 640, 1, 4096]> var_58641_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_58639, groups = var_58064, pad = var_58641_pad_0, pad_type = var_58641_pad_type_0, strides = var_58637, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_741_cast)[name = tensor<string, []>("op_58641_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_389_cast = add(x = var_58641_cast, y = inputs_387_cast)[name = tensor<string, []>("inputs_389_cast")];
            tensor<int32, [1]> var_58645 = const()[name = tensor<string, []>("op_58645"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_389_cast = reduce_mean(axes = var_58645, keep_dims = var_58059, x = inputs_389_cast)[name = tensor<string, []>("channels_mean_389_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_389_cast = sub(x = inputs_389_cast, y = channels_mean_389_cast)[name = tensor<string, []>("zero_mean_389_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_389_cast = mul(x = zero_mean_389_cast, y = zero_mean_389_cast)[name = tensor<string, []>("zero_mean_sq_389_cast")];
            tensor<int32, [1]> var_58649 = const()[name = tensor<string, []>("op_58649"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_58650_cast = reduce_mean(axes = var_58649, keep_dims = var_58059, x = zero_mean_sq_389_cast)[name = tensor<string, []>("op_58650_cast")];
            tensor<fp16, []> var_58651_to_fp16 = const()[name = tensor<string, []>("op_58651_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_58652_cast = add(x = var_58650_cast, y = var_58651_to_fp16)[name = tensor<string, []>("op_58652_cast")];
            tensor<fp16, []> denom_389_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_389_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_389_cast = rsqrt(epsilon = denom_389_epsilon_0_to_fp16, x = var_58652_cast)[name = tensor<string, []>("denom_389_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_389_cast = mul(x = zero_mean_389_cast, y = denom_389_cast)[name = tensor<string, []>("out_389_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_58656_to_fp16 = const()[name = tensor<string, []>("op_58656_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4944587264)))];
            tensor<fp16, [2, 640, 1, 4096]> var_58657_cast = add(x = out_389_cast, y = var_58656_to_fp16)[name = tensor<string, []>("op_58657_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_58659_to_fp16 = const()[name = tensor<string, []>("op_58659_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4944588608)))];
            tensor<fp16, [2, 640, 1, 4096]> input_743_cast = mul(x = var_58657_cast, y = var_58659_to_fp16)[name = tensor<string, []>("input_743_cast")];
            tensor<int32, [2]> var_58667 = const()[name = tensor<string, []>("op_58667"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58669 = const()[name = tensor<string, []>("op_58669"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_58671_pad_type_0 = const()[name = tensor<string, []>("op_58671_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_58671_pad_0 = const()[name = tensor<string, []>("op_58671_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [5120, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [5120, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4944589952)))];
            tensor<fp16, [5120]> up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4951143616)))];
            tensor<fp16, [2, 5120, 1, 4096]> var_58671_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_58669, groups = var_58064, pad = var_58671_pad_0, pad_type = var_58671_pad_type_0, strides = var_58667, weight = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_743_cast)[name = tensor<string, []>("op_58671_cast")];
            tensor<int32, [2]> var_58672_split_sizes_0 = const()[name = tensor<string, []>("op_58672_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_58672_axis_0 = const()[name = tensor<string, []>("op_58672_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 4096]> var_58672_cast_0, tensor<fp16, [2, 2560, 1, 4096]> var_58672_cast_1 = split(axis = var_58672_axis_0, split_sizes = var_58672_split_sizes_0, x = var_58671_cast)[name = tensor<string, []>("op_58672_cast")];
            tensor<string, []> var_58674_mode_0 = const()[name = tensor<string, []>("op_58674_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 4096]> var_58674_cast = gelu(mode = var_58674_mode_0, x = var_58672_cast_1)[name = tensor<string, []>("op_58674_cast")];
            tensor<fp16, [2, 2560, 1, 4096]> input_745_cast = mul(x = var_58672_cast_0, y = var_58674_cast)[name = tensor<string, []>("input_745_cast")];
            tensor<int32, [2]> var_58678 = const()[name = tensor<string, []>("op_58678"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58680 = const()[name = tensor<string, []>("op_58680"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_58682_pad_type_0 = const()[name = tensor<string, []>("op_58682_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_58682_pad_0 = const()[name = tensor<string, []>("op_58682_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2560, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [640, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4951153920)))];
            tensor<fp16, [640]> up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4954430784)))];
            tensor<fp16, [2, 640, 1, 4096]> var_58682_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_58680, groups = var_58064, pad = var_58682_pad_0, pad_type = var_58682_pad_type_0, strides = var_58678, weight = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_745_cast)[name = tensor<string, []>("op_58682_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_391_cast = add(x = var_58682_cast, y = inputs_389_cast)[name = tensor<string, []>("inputs_391_cast")];
            tensor<int32, [1]> var_58692 = const()[name = tensor<string, []>("op_58692"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_391_cast = reduce_mean(axes = var_58692, keep_dims = var_58059, x = inputs_391_cast)[name = tensor<string, []>("channels_mean_391_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_391_cast = sub(x = inputs_391_cast, y = channels_mean_391_cast)[name = tensor<string, []>("zero_mean_391_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_391_cast = mul(x = zero_mean_391_cast, y = zero_mean_391_cast)[name = tensor<string, []>("zero_mean_sq_391_cast")];
            tensor<int32, [1]> var_58696 = const()[name = tensor<string, []>("op_58696"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_58697_cast = reduce_mean(axes = var_58696, keep_dims = var_58059, x = zero_mean_sq_391_cast)[name = tensor<string, []>("op_58697_cast")];
            tensor<fp16, []> var_58698_to_fp16 = const()[name = tensor<string, []>("op_58698_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_58699_cast = add(x = var_58697_cast, y = var_58698_to_fp16)[name = tensor<string, []>("op_58699_cast")];
            tensor<fp16, []> denom_391_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_391_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_391_cast = rsqrt(epsilon = denom_391_epsilon_0_to_fp16, x = var_58699_cast)[name = tensor<string, []>("denom_391_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_391_cast = mul(x = zero_mean_391_cast, y = denom_391_cast)[name = tensor<string, []>("out_391_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_58703_to_fp16 = const()[name = tensor<string, []>("op_58703_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4954432128)))];
            tensor<fp16, [2, 640, 1, 4096]> var_58704_cast = add(x = out_391_cast, y = var_58703_to_fp16)[name = tensor<string, []>("op_58704_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_58706_to_fp16 = const()[name = tensor<string, []>("op_58706_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4954433472)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_515_cast = mul(x = var_58704_cast, y = var_58706_to_fp16)[name = tensor<string, []>("hidden_states_515_cast")];
            tensor<int32, [2]> var_58713 = const()[name = tensor<string, []>("op_58713"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58715 = const()[name = tensor<string, []>("op_58715"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_261_pad_type_0 = const()[name = tensor<string, []>("q_261_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_261_pad_0 = const()[name = tensor<string, []>("q_261_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4954434816)))];
            tensor<fp16, [2, 640, 1, 4096]> q_261_cast = conv(dilations = var_58715, groups = var_58064, pad = q_261_pad_0, pad_type = q_261_pad_type_0, strides = var_58713, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_515_cast)[name = tensor<string, []>("q_261_cast")];
            tensor<int32, [2]> var_58719 = const()[name = tensor<string, []>("op_58719"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58721 = const()[name = tensor<string, []>("op_58721"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_521_pad_type_0 = const()[name = tensor<string, []>("k_521_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_521_pad_0 = const()[name = tensor<string, []>("k_521_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4955254080)))];
            tensor<fp16, [2, 640, 1, 4096]> k_521_cast = conv(dilations = var_58721, groups = var_58064, pad = k_521_pad_0, pad_type = k_521_pad_type_0, strides = var_58719, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_515_cast)[name = tensor<string, []>("k_521_cast")];
            tensor<int32, [2]> var_58725 = const()[name = tensor<string, []>("op_58725"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58727 = const()[name = tensor<string, []>("op_58727"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_261_pad_type_0 = const()[name = tensor<string, []>("v_261_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_261_pad_0 = const()[name = tensor<string, []>("v_261_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4956073344)))];
            tensor<fp16, [2, 640, 1, 4096]> v_261_cast = conv(dilations = var_58727, groups = var_58064, pad = v_261_pad_0, pad_type = v_261_pad_type_0, strides = var_58725, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_515_cast)[name = tensor<string, []>("v_261_cast")];
            tensor<int32, [4]> var_58731_begin_0 = const()[name = tensor<string, []>("op_58731_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58731_end_0 = const()[name = tensor<string, []>("op_58731_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_58731_end_mask_0 = const()[name = tensor<string, []>("op_58731_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58731_cast = slice_by_index(begin = var_58731_begin_0, end = var_58731_end_0, end_mask = var_58731_end_mask_0, x = q_261_cast)[name = tensor<string, []>("op_58731_cast")];
            tensor<int32, [4]> var_58735_begin_0 = const()[name = tensor<string, []>("op_58735_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_58735_end_0 = const()[name = tensor<string, []>("op_58735_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_58735_end_mask_0 = const()[name = tensor<string, []>("op_58735_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58735_cast = slice_by_index(begin = var_58735_begin_0, end = var_58735_end_0, end_mask = var_58735_end_mask_0, x = q_261_cast)[name = tensor<string, []>("op_58735_cast")];
            tensor<int32, [4]> var_58739_begin_0 = const()[name = tensor<string, []>("op_58739_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_58739_end_0 = const()[name = tensor<string, []>("op_58739_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_58739_end_mask_0 = const()[name = tensor<string, []>("op_58739_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58739_cast = slice_by_index(begin = var_58739_begin_0, end = var_58739_end_0, end_mask = var_58739_end_mask_0, x = q_261_cast)[name = tensor<string, []>("op_58739_cast")];
            tensor<int32, [4]> var_58743_begin_0 = const()[name = tensor<string, []>("op_58743_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_58743_end_0 = const()[name = tensor<string, []>("op_58743_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_58743_end_mask_0 = const()[name = tensor<string, []>("op_58743_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58743_cast = slice_by_index(begin = var_58743_begin_0, end = var_58743_end_0, end_mask = var_58743_end_mask_0, x = q_261_cast)[name = tensor<string, []>("op_58743_cast")];
            tensor<int32, [4]> var_58747_begin_0 = const()[name = tensor<string, []>("op_58747_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_58747_end_0 = const()[name = tensor<string, []>("op_58747_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_58747_end_mask_0 = const()[name = tensor<string, []>("op_58747_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58747_cast = slice_by_index(begin = var_58747_begin_0, end = var_58747_end_0, end_mask = var_58747_end_mask_0, x = q_261_cast)[name = tensor<string, []>("op_58747_cast")];
            tensor<int32, [4]> var_58751_begin_0 = const()[name = tensor<string, []>("op_58751_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_58751_end_0 = const()[name = tensor<string, []>("op_58751_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_58751_end_mask_0 = const()[name = tensor<string, []>("op_58751_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58751_cast = slice_by_index(begin = var_58751_begin_0, end = var_58751_end_0, end_mask = var_58751_end_mask_0, x = q_261_cast)[name = tensor<string, []>("op_58751_cast")];
            tensor<int32, [4]> var_58755_begin_0 = const()[name = tensor<string, []>("op_58755_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_58755_end_0 = const()[name = tensor<string, []>("op_58755_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_58755_end_mask_0 = const()[name = tensor<string, []>("op_58755_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58755_cast = slice_by_index(begin = var_58755_begin_0, end = var_58755_end_0, end_mask = var_58755_end_mask_0, x = q_261_cast)[name = tensor<string, []>("op_58755_cast")];
            tensor<int32, [4]> var_58759_begin_0 = const()[name = tensor<string, []>("op_58759_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_58759_end_0 = const()[name = tensor<string, []>("op_58759_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_58759_end_mask_0 = const()[name = tensor<string, []>("op_58759_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58759_cast = slice_by_index(begin = var_58759_begin_0, end = var_58759_end_0, end_mask = var_58759_end_mask_0, x = q_261_cast)[name = tensor<string, []>("op_58759_cast")];
            tensor<int32, [4]> var_58763_begin_0 = const()[name = tensor<string, []>("op_58763_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_58763_end_0 = const()[name = tensor<string, []>("op_58763_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_58763_end_mask_0 = const()[name = tensor<string, []>("op_58763_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58763_cast = slice_by_index(begin = var_58763_begin_0, end = var_58763_end_0, end_mask = var_58763_end_mask_0, x = q_261_cast)[name = tensor<string, []>("op_58763_cast")];
            tensor<int32, [4]> var_58767_begin_0 = const()[name = tensor<string, []>("op_58767_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_58767_end_0 = const()[name = tensor<string, []>("op_58767_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_58767_end_mask_0 = const()[name = tensor<string, []>("op_58767_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58767_cast = slice_by_index(begin = var_58767_begin_0, end = var_58767_end_0, end_mask = var_58767_end_mask_0, x = q_261_cast)[name = tensor<string, []>("op_58767_cast")];
            tensor<int32, [4]> k_523_perm_0 = const()[name = tensor<string, []>("k_523_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_58774_begin_0 = const()[name = tensor<string, []>("op_58774_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58774_end_0 = const()[name = tensor<string, []>("op_58774_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_58774_end_mask_0 = const()[name = tensor<string, []>("op_58774_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 640]> transpose_9 = transpose(perm = k_523_perm_0, x = k_521_cast)[name = tensor<string, []>("transpose_9")];
            tensor<fp16, [2, 4096, 1, 64]> var_58774_cast = slice_by_index(begin = var_58774_begin_0, end = var_58774_end_0, end_mask = var_58774_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_58774_cast")];
            tensor<int32, [4]> var_58778_begin_0 = const()[name = tensor<string, []>("op_58778_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_58778_end_0 = const()[name = tensor<string, []>("op_58778_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_58778_end_mask_0 = const()[name = tensor<string, []>("op_58778_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58778_cast = slice_by_index(begin = var_58778_begin_0, end = var_58778_end_0, end_mask = var_58778_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_58778_cast")];
            tensor<int32, [4]> var_58782_begin_0 = const()[name = tensor<string, []>("op_58782_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_58782_end_0 = const()[name = tensor<string, []>("op_58782_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_58782_end_mask_0 = const()[name = tensor<string, []>("op_58782_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58782_cast = slice_by_index(begin = var_58782_begin_0, end = var_58782_end_0, end_mask = var_58782_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_58782_cast")];
            tensor<int32, [4]> var_58786_begin_0 = const()[name = tensor<string, []>("op_58786_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_58786_end_0 = const()[name = tensor<string, []>("op_58786_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_58786_end_mask_0 = const()[name = tensor<string, []>("op_58786_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58786_cast = slice_by_index(begin = var_58786_begin_0, end = var_58786_end_0, end_mask = var_58786_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_58786_cast")];
            tensor<int32, [4]> var_58790_begin_0 = const()[name = tensor<string, []>("op_58790_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_58790_end_0 = const()[name = tensor<string, []>("op_58790_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_58790_end_mask_0 = const()[name = tensor<string, []>("op_58790_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58790_cast = slice_by_index(begin = var_58790_begin_0, end = var_58790_end_0, end_mask = var_58790_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_58790_cast")];
            tensor<int32, [4]> var_58794_begin_0 = const()[name = tensor<string, []>("op_58794_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_58794_end_0 = const()[name = tensor<string, []>("op_58794_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 384])];
            tensor<bool, [4]> var_58794_end_mask_0 = const()[name = tensor<string, []>("op_58794_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58794_cast = slice_by_index(begin = var_58794_begin_0, end = var_58794_end_0, end_mask = var_58794_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_58794_cast")];
            tensor<int32, [4]> var_58798_begin_0 = const()[name = tensor<string, []>("op_58798_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_58798_end_0 = const()[name = tensor<string, []>("op_58798_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 448])];
            tensor<bool, [4]> var_58798_end_mask_0 = const()[name = tensor<string, []>("op_58798_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58798_cast = slice_by_index(begin = var_58798_begin_0, end = var_58798_end_0, end_mask = var_58798_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_58798_cast")];
            tensor<int32, [4]> var_58802_begin_0 = const()[name = tensor<string, []>("op_58802_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_58802_end_0 = const()[name = tensor<string, []>("op_58802_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 512])];
            tensor<bool, [4]> var_58802_end_mask_0 = const()[name = tensor<string, []>("op_58802_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58802_cast = slice_by_index(begin = var_58802_begin_0, end = var_58802_end_0, end_mask = var_58802_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_58802_cast")];
            tensor<int32, [4]> var_58806_begin_0 = const()[name = tensor<string, []>("op_58806_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_58806_end_0 = const()[name = tensor<string, []>("op_58806_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 576])];
            tensor<bool, [4]> var_58806_end_mask_0 = const()[name = tensor<string, []>("op_58806_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58806_cast = slice_by_index(begin = var_58806_begin_0, end = var_58806_end_0, end_mask = var_58806_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_58806_cast")];
            tensor<int32, [4]> var_58810_begin_0 = const()[name = tensor<string, []>("op_58810_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_58810_end_0 = const()[name = tensor<string, []>("op_58810_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 640])];
            tensor<bool, [4]> var_58810_end_mask_0 = const()[name = tensor<string, []>("op_58810_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_58810_cast = slice_by_index(begin = var_58810_begin_0, end = var_58810_end_0, end_mask = var_58810_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_58810_cast")];
            tensor<int32, [4]> var_58812_begin_0 = const()[name = tensor<string, []>("op_58812_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58812_end_0 = const()[name = tensor<string, []>("op_58812_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_58812_end_mask_0 = const()[name = tensor<string, []>("op_58812_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58812_cast = slice_by_index(begin = var_58812_begin_0, end = var_58812_end_0, end_mask = var_58812_end_mask_0, x = v_261_cast)[name = tensor<string, []>("op_58812_cast")];
            tensor<int32, [4]> var_58816_begin_0 = const()[name = tensor<string, []>("op_58816_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_58816_end_0 = const()[name = tensor<string, []>("op_58816_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_58816_end_mask_0 = const()[name = tensor<string, []>("op_58816_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58816_cast = slice_by_index(begin = var_58816_begin_0, end = var_58816_end_0, end_mask = var_58816_end_mask_0, x = v_261_cast)[name = tensor<string, []>("op_58816_cast")];
            tensor<int32, [4]> var_58820_begin_0 = const()[name = tensor<string, []>("op_58820_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_58820_end_0 = const()[name = tensor<string, []>("op_58820_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_58820_end_mask_0 = const()[name = tensor<string, []>("op_58820_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58820_cast = slice_by_index(begin = var_58820_begin_0, end = var_58820_end_0, end_mask = var_58820_end_mask_0, x = v_261_cast)[name = tensor<string, []>("op_58820_cast")];
            tensor<int32, [4]> var_58824_begin_0 = const()[name = tensor<string, []>("op_58824_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_58824_end_0 = const()[name = tensor<string, []>("op_58824_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_58824_end_mask_0 = const()[name = tensor<string, []>("op_58824_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58824_cast = slice_by_index(begin = var_58824_begin_0, end = var_58824_end_0, end_mask = var_58824_end_mask_0, x = v_261_cast)[name = tensor<string, []>("op_58824_cast")];
            tensor<int32, [4]> var_58828_begin_0 = const()[name = tensor<string, []>("op_58828_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_58828_end_0 = const()[name = tensor<string, []>("op_58828_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_58828_end_mask_0 = const()[name = tensor<string, []>("op_58828_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58828_cast = slice_by_index(begin = var_58828_begin_0, end = var_58828_end_0, end_mask = var_58828_end_mask_0, x = v_261_cast)[name = tensor<string, []>("op_58828_cast")];
            tensor<int32, [4]> var_58832_begin_0 = const()[name = tensor<string, []>("op_58832_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_58832_end_0 = const()[name = tensor<string, []>("op_58832_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_58832_end_mask_0 = const()[name = tensor<string, []>("op_58832_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58832_cast = slice_by_index(begin = var_58832_begin_0, end = var_58832_end_0, end_mask = var_58832_end_mask_0, x = v_261_cast)[name = tensor<string, []>("op_58832_cast")];
            tensor<int32, [4]> var_58836_begin_0 = const()[name = tensor<string, []>("op_58836_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_58836_end_0 = const()[name = tensor<string, []>("op_58836_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_58836_end_mask_0 = const()[name = tensor<string, []>("op_58836_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58836_cast = slice_by_index(begin = var_58836_begin_0, end = var_58836_end_0, end_mask = var_58836_end_mask_0, x = v_261_cast)[name = tensor<string, []>("op_58836_cast")];
            tensor<int32, [4]> var_58840_begin_0 = const()[name = tensor<string, []>("op_58840_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_58840_end_0 = const()[name = tensor<string, []>("op_58840_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_58840_end_mask_0 = const()[name = tensor<string, []>("op_58840_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58840_cast = slice_by_index(begin = var_58840_begin_0, end = var_58840_end_0, end_mask = var_58840_end_mask_0, x = v_261_cast)[name = tensor<string, []>("op_58840_cast")];
            tensor<int32, [4]> var_58844_begin_0 = const()[name = tensor<string, []>("op_58844_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_58844_end_0 = const()[name = tensor<string, []>("op_58844_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_58844_end_mask_0 = const()[name = tensor<string, []>("op_58844_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58844_cast = slice_by_index(begin = var_58844_begin_0, end = var_58844_end_0, end_mask = var_58844_end_mask_0, x = v_261_cast)[name = tensor<string, []>("op_58844_cast")];
            tensor<int32, [4]> var_58848_begin_0 = const()[name = tensor<string, []>("op_58848_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_58848_end_0 = const()[name = tensor<string, []>("op_58848_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_58848_end_mask_0 = const()[name = tensor<string, []>("op_58848_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58848_cast = slice_by_index(begin = var_58848_begin_0, end = var_58848_end_0, end_mask = var_58848_end_mask_0, x = v_261_cast)[name = tensor<string, []>("op_58848_cast")];
            tensor<string, []> var_58852_equation_0 = const()[name = tensor<string, []>("op_58852_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58852_cast = einsum(equation = var_58852_equation_0, values = (var_58774_cast, var_58731_cast))[name = tensor<string, []>("op_58852_cast")];
            tensor<fp16, []> var_58853_to_fp16 = const()[name = tensor<string, []>("op_58853_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5001_cast = mul(x = var_58852_cast, y = var_58853_to_fp16)[name = tensor<string, []>("aw_5001_cast")];
            tensor<string, []> var_58856_equation_0 = const()[name = tensor<string, []>("op_58856_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58856_cast = einsum(equation = var_58856_equation_0, values = (var_58778_cast, var_58735_cast))[name = tensor<string, []>("op_58856_cast")];
            tensor<fp16, []> var_58857_to_fp16 = const()[name = tensor<string, []>("op_58857_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5003_cast = mul(x = var_58856_cast, y = var_58857_to_fp16)[name = tensor<string, []>("aw_5003_cast")];
            tensor<string, []> var_58860_equation_0 = const()[name = tensor<string, []>("op_58860_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58860_cast = einsum(equation = var_58860_equation_0, values = (var_58782_cast, var_58739_cast))[name = tensor<string, []>("op_58860_cast")];
            tensor<fp16, []> var_58861_to_fp16 = const()[name = tensor<string, []>("op_58861_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5005_cast = mul(x = var_58860_cast, y = var_58861_to_fp16)[name = tensor<string, []>("aw_5005_cast")];
            tensor<string, []> var_58864_equation_0 = const()[name = tensor<string, []>("op_58864_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58864_cast = einsum(equation = var_58864_equation_0, values = (var_58786_cast, var_58743_cast))[name = tensor<string, []>("op_58864_cast")];
            tensor<fp16, []> var_58865_to_fp16 = const()[name = tensor<string, []>("op_58865_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5007_cast = mul(x = var_58864_cast, y = var_58865_to_fp16)[name = tensor<string, []>("aw_5007_cast")];
            tensor<string, []> var_58868_equation_0 = const()[name = tensor<string, []>("op_58868_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58868_cast = einsum(equation = var_58868_equation_0, values = (var_58790_cast, var_58747_cast))[name = tensor<string, []>("op_58868_cast")];
            tensor<fp16, []> var_58869_to_fp16 = const()[name = tensor<string, []>("op_58869_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5009_cast = mul(x = var_58868_cast, y = var_58869_to_fp16)[name = tensor<string, []>("aw_5009_cast")];
            tensor<string, []> var_58872_equation_0 = const()[name = tensor<string, []>("op_58872_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58872_cast = einsum(equation = var_58872_equation_0, values = (var_58794_cast, var_58751_cast))[name = tensor<string, []>("op_58872_cast")];
            tensor<fp16, []> var_58873_to_fp16 = const()[name = tensor<string, []>("op_58873_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5011_cast = mul(x = var_58872_cast, y = var_58873_to_fp16)[name = tensor<string, []>("aw_5011_cast")];
            tensor<string, []> var_58876_equation_0 = const()[name = tensor<string, []>("op_58876_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58876_cast = einsum(equation = var_58876_equation_0, values = (var_58798_cast, var_58755_cast))[name = tensor<string, []>("op_58876_cast")];
            tensor<fp16, []> var_58877_to_fp16 = const()[name = tensor<string, []>("op_58877_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5013_cast = mul(x = var_58876_cast, y = var_58877_to_fp16)[name = tensor<string, []>("aw_5013_cast")];
            tensor<string, []> var_58880_equation_0 = const()[name = tensor<string, []>("op_58880_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58880_cast = einsum(equation = var_58880_equation_0, values = (var_58802_cast, var_58759_cast))[name = tensor<string, []>("op_58880_cast")];
            tensor<fp16, []> var_58881_to_fp16 = const()[name = tensor<string, []>("op_58881_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5015_cast = mul(x = var_58880_cast, y = var_58881_to_fp16)[name = tensor<string, []>("aw_5015_cast")];
            tensor<string, []> var_58884_equation_0 = const()[name = tensor<string, []>("op_58884_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58884_cast = einsum(equation = var_58884_equation_0, values = (var_58806_cast, var_58763_cast))[name = tensor<string, []>("op_58884_cast")];
            tensor<fp16, []> var_58885_to_fp16 = const()[name = tensor<string, []>("op_58885_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5017_cast = mul(x = var_58884_cast, y = var_58885_to_fp16)[name = tensor<string, []>("aw_5017_cast")];
            tensor<string, []> var_58888_equation_0 = const()[name = tensor<string, []>("op_58888_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58888_cast = einsum(equation = var_58888_equation_0, values = (var_58810_cast, var_58767_cast))[name = tensor<string, []>("op_58888_cast")];
            tensor<fp16, []> var_58889_to_fp16 = const()[name = tensor<string, []>("op_58889_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5019_cast = mul(x = var_58888_cast, y = var_58889_to_fp16)[name = tensor<string, []>("aw_5019_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58891_cast = softmax(axis = var_58064, x = aw_5001_cast)[name = tensor<string, []>("op_58891_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58892_cast = softmax(axis = var_58064, x = aw_5003_cast)[name = tensor<string, []>("op_58892_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58893_cast = softmax(axis = var_58064, x = aw_5005_cast)[name = tensor<string, []>("op_58893_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58894_cast = softmax(axis = var_58064, x = aw_5007_cast)[name = tensor<string, []>("op_58894_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58895_cast = softmax(axis = var_58064, x = aw_5009_cast)[name = tensor<string, []>("op_58895_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58896_cast = softmax(axis = var_58064, x = aw_5011_cast)[name = tensor<string, []>("op_58896_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58897_cast = softmax(axis = var_58064, x = aw_5013_cast)[name = tensor<string, []>("op_58897_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58898_cast = softmax(axis = var_58064, x = aw_5015_cast)[name = tensor<string, []>("op_58898_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58899_cast = softmax(axis = var_58064, x = aw_5017_cast)[name = tensor<string, []>("op_58899_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_58900_cast = softmax(axis = var_58064, x = aw_5019_cast)[name = tensor<string, []>("op_58900_cast")];
            tensor<string, []> var_58902_equation_0 = const()[name = tensor<string, []>("op_58902_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58902_cast = einsum(equation = var_58902_equation_0, values = (var_58812_cast, var_58891_cast))[name = tensor<string, []>("op_58902_cast")];
            tensor<string, []> var_58904_equation_0 = const()[name = tensor<string, []>("op_58904_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58904_cast = einsum(equation = var_58904_equation_0, values = (var_58816_cast, var_58892_cast))[name = tensor<string, []>("op_58904_cast")];
            tensor<string, []> var_58906_equation_0 = const()[name = tensor<string, []>("op_58906_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58906_cast = einsum(equation = var_58906_equation_0, values = (var_58820_cast, var_58893_cast))[name = tensor<string, []>("op_58906_cast")];
            tensor<string, []> var_58908_equation_0 = const()[name = tensor<string, []>("op_58908_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58908_cast = einsum(equation = var_58908_equation_0, values = (var_58824_cast, var_58894_cast))[name = tensor<string, []>("op_58908_cast")];
            tensor<string, []> var_58910_equation_0 = const()[name = tensor<string, []>("op_58910_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58910_cast = einsum(equation = var_58910_equation_0, values = (var_58828_cast, var_58895_cast))[name = tensor<string, []>("op_58910_cast")];
            tensor<string, []> var_58912_equation_0 = const()[name = tensor<string, []>("op_58912_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58912_cast = einsum(equation = var_58912_equation_0, values = (var_58832_cast, var_58896_cast))[name = tensor<string, []>("op_58912_cast")];
            tensor<string, []> var_58914_equation_0 = const()[name = tensor<string, []>("op_58914_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58914_cast = einsum(equation = var_58914_equation_0, values = (var_58836_cast, var_58897_cast))[name = tensor<string, []>("op_58914_cast")];
            tensor<string, []> var_58916_equation_0 = const()[name = tensor<string, []>("op_58916_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58916_cast = einsum(equation = var_58916_equation_0, values = (var_58840_cast, var_58898_cast))[name = tensor<string, []>("op_58916_cast")];
            tensor<string, []> var_58918_equation_0 = const()[name = tensor<string, []>("op_58918_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58918_cast = einsum(equation = var_58918_equation_0, values = (var_58844_cast, var_58899_cast))[name = tensor<string, []>("op_58918_cast")];
            tensor<string, []> var_58920_equation_0 = const()[name = tensor<string, []>("op_58920_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_58920_cast = einsum(equation = var_58920_equation_0, values = (var_58848_cast, var_58900_cast))[name = tensor<string, []>("op_58920_cast")];
            tensor<bool, []> input_747_interleave_0 = const()[name = tensor<string, []>("input_747_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_747_cast = concat(axis = var_58064, interleave = input_747_interleave_0, values = (var_58902_cast, var_58904_cast, var_58906_cast, var_58908_cast, var_58910_cast, var_58912_cast, var_58914_cast, var_58916_cast, var_58918_cast, var_58920_cast))[name = tensor<string, []>("input_747_cast")];
            tensor<int32, [2]> var_58926 = const()[name = tensor<string, []>("op_58926"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58928 = const()[name = tensor<string, []>("op_58928"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_58930_pad_type_0 = const()[name = tensor<string, []>("op_58930_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_58930_pad_0 = const()[name = tensor<string, []>("op_58930_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4956892608)))];
            tensor<fp16, [640]> up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4957711872)))];
            tensor<fp16, [2, 640, 1, 4096]> var_58930_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_58928, groups = var_58064, pad = var_58930_pad_0, pad_type = var_58930_pad_type_0, strides = var_58926, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_747_cast)[name = tensor<string, []>("op_58930_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_393_cast = add(x = var_58930_cast, y = inputs_391_cast)[name = tensor<string, []>("inputs_393_cast")];
            tensor<int32, [1]> var_58934 = const()[name = tensor<string, []>("op_58934"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_393_cast = reduce_mean(axes = var_58934, keep_dims = var_58059, x = inputs_393_cast)[name = tensor<string, []>("channels_mean_393_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_393_cast = sub(x = inputs_393_cast, y = channels_mean_393_cast)[name = tensor<string, []>("zero_mean_393_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_393_cast = mul(x = zero_mean_393_cast, y = zero_mean_393_cast)[name = tensor<string, []>("zero_mean_sq_393_cast")];
            tensor<int32, [1]> var_58938 = const()[name = tensor<string, []>("op_58938"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_58939_cast = reduce_mean(axes = var_58938, keep_dims = var_58059, x = zero_mean_sq_393_cast)[name = tensor<string, []>("op_58939_cast")];
            tensor<fp16, []> var_58940_to_fp16 = const()[name = tensor<string, []>("op_58940_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_58941_cast = add(x = var_58939_cast, y = var_58940_to_fp16)[name = tensor<string, []>("op_58941_cast")];
            tensor<fp16, []> denom_393_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_393_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_393_cast = rsqrt(epsilon = denom_393_epsilon_0_to_fp16, x = var_58941_cast)[name = tensor<string, []>("denom_393_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_393_cast = mul(x = zero_mean_393_cast, y = denom_393_cast)[name = tensor<string, []>("out_393_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_58945_to_fp16 = const()[name = tensor<string, []>("op_58945_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4957713216)))];
            tensor<fp16, [2, 640, 1, 4096]> var_58946_cast = add(x = out_393_cast, y = var_58945_to_fp16)[name = tensor<string, []>("op_58946_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_58948_to_fp16 = const()[name = tensor<string, []>("op_58948_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4957714560)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_517_cast = mul(x = var_58946_cast, y = var_58948_to_fp16)[name = tensor<string, []>("hidden_states_517_cast")];
            tensor<int32, [2]> var_58955 = const()[name = tensor<string, []>("op_58955"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58957 = const()[name = tensor<string, []>("op_58957"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_263_pad_type_0 = const()[name = tensor<string, []>("q_263_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_263_pad_0 = const()[name = tensor<string, []>("q_263_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4957715904)))];
            tensor<fp16, [2, 640, 1, 4096]> q_263_cast = conv(dilations = var_58957, groups = var_58064, pad = q_263_pad_0, pad_type = q_263_pad_type_0, strides = var_58955, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_517_cast)[name = tensor<string, []>("q_263_cast")];
            tensor<int32, [2]> var_58961 = const()[name = tensor<string, []>("op_58961"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58963 = const()[name = tensor<string, []>("op_58963"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_525_pad_type_0 = const()[name = tensor<string, []>("k_525_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_525_pad_0 = const()[name = tensor<string, []>("k_525_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4958535168)))];
            tensor<fp16, [2, 640, 1, 77]> k_525_cast = conv(dilations = var_58963, groups = var_58064, pad = k_525_pad_0, pad_type = k_525_pad_type_0, strides = var_58961, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_525_cast")];
            tensor<int32, [2]> var_58967 = const()[name = tensor<string, []>("op_58967"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58969 = const()[name = tensor<string, []>("op_58969"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_263_pad_type_0 = const()[name = tensor<string, []>("v_263_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_263_pad_0 = const()[name = tensor<string, []>("v_263_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4961156672)))];
            tensor<fp16, [2, 640, 1, 77]> v_263_cast = conv(dilations = var_58969, groups = var_58064, pad = v_263_pad_0, pad_type = v_263_pad_type_0, strides = var_58967, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_263_cast")];
            tensor<int32, [4]> var_58973_begin_0 = const()[name = tensor<string, []>("op_58973_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_58973_end_0 = const()[name = tensor<string, []>("op_58973_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_58973_end_mask_0 = const()[name = tensor<string, []>("op_58973_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58973_cast = slice_by_index(begin = var_58973_begin_0, end = var_58973_end_0, end_mask = var_58973_end_mask_0, x = q_263_cast)[name = tensor<string, []>("op_58973_cast")];
            tensor<int32, [4]> var_58977_begin_0 = const()[name = tensor<string, []>("op_58977_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_58977_end_0 = const()[name = tensor<string, []>("op_58977_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_58977_end_mask_0 = const()[name = tensor<string, []>("op_58977_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58977_cast = slice_by_index(begin = var_58977_begin_0, end = var_58977_end_0, end_mask = var_58977_end_mask_0, x = q_263_cast)[name = tensor<string, []>("op_58977_cast")];
            tensor<int32, [4]> var_58981_begin_0 = const()[name = tensor<string, []>("op_58981_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_58981_end_0 = const()[name = tensor<string, []>("op_58981_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_58981_end_mask_0 = const()[name = tensor<string, []>("op_58981_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58981_cast = slice_by_index(begin = var_58981_begin_0, end = var_58981_end_0, end_mask = var_58981_end_mask_0, x = q_263_cast)[name = tensor<string, []>("op_58981_cast")];
            tensor<int32, [4]> var_58985_begin_0 = const()[name = tensor<string, []>("op_58985_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_58985_end_0 = const()[name = tensor<string, []>("op_58985_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_58985_end_mask_0 = const()[name = tensor<string, []>("op_58985_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58985_cast = slice_by_index(begin = var_58985_begin_0, end = var_58985_end_0, end_mask = var_58985_end_mask_0, x = q_263_cast)[name = tensor<string, []>("op_58985_cast")];
            tensor<int32, [4]> var_58989_begin_0 = const()[name = tensor<string, []>("op_58989_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_58989_end_0 = const()[name = tensor<string, []>("op_58989_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_58989_end_mask_0 = const()[name = tensor<string, []>("op_58989_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58989_cast = slice_by_index(begin = var_58989_begin_0, end = var_58989_end_0, end_mask = var_58989_end_mask_0, x = q_263_cast)[name = tensor<string, []>("op_58989_cast")];
            tensor<int32, [4]> var_58993_begin_0 = const()[name = tensor<string, []>("op_58993_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_58993_end_0 = const()[name = tensor<string, []>("op_58993_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_58993_end_mask_0 = const()[name = tensor<string, []>("op_58993_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58993_cast = slice_by_index(begin = var_58993_begin_0, end = var_58993_end_0, end_mask = var_58993_end_mask_0, x = q_263_cast)[name = tensor<string, []>("op_58993_cast")];
            tensor<int32, [4]> var_58997_begin_0 = const()[name = tensor<string, []>("op_58997_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_58997_end_0 = const()[name = tensor<string, []>("op_58997_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_58997_end_mask_0 = const()[name = tensor<string, []>("op_58997_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_58997_cast = slice_by_index(begin = var_58997_begin_0, end = var_58997_end_0, end_mask = var_58997_end_mask_0, x = q_263_cast)[name = tensor<string, []>("op_58997_cast")];
            tensor<int32, [4]> var_59001_begin_0 = const()[name = tensor<string, []>("op_59001_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_59001_end_0 = const()[name = tensor<string, []>("op_59001_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_59001_end_mask_0 = const()[name = tensor<string, []>("op_59001_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59001_cast = slice_by_index(begin = var_59001_begin_0, end = var_59001_end_0, end_mask = var_59001_end_mask_0, x = q_263_cast)[name = tensor<string, []>("op_59001_cast")];
            tensor<int32, [4]> var_59005_begin_0 = const()[name = tensor<string, []>("op_59005_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_59005_end_0 = const()[name = tensor<string, []>("op_59005_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_59005_end_mask_0 = const()[name = tensor<string, []>("op_59005_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59005_cast = slice_by_index(begin = var_59005_begin_0, end = var_59005_end_0, end_mask = var_59005_end_mask_0, x = q_263_cast)[name = tensor<string, []>("op_59005_cast")];
            tensor<int32, [4]> var_59009_begin_0 = const()[name = tensor<string, []>("op_59009_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_59009_end_0 = const()[name = tensor<string, []>("op_59009_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_59009_end_mask_0 = const()[name = tensor<string, []>("op_59009_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59009_cast = slice_by_index(begin = var_59009_begin_0, end = var_59009_end_0, end_mask = var_59009_end_mask_0, x = q_263_cast)[name = tensor<string, []>("op_59009_cast")];
            tensor<int32, [4]> k_527_perm_0 = const()[name = tensor<string, []>("k_527_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_59016_begin_0 = const()[name = tensor<string, []>("op_59016_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59016_end_0 = const()[name = tensor<string, []>("op_59016_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_59016_end_mask_0 = const()[name = tensor<string, []>("op_59016_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> transpose_8 = transpose(perm = k_527_perm_0, x = k_525_cast)[name = tensor<string, []>("transpose_8")];
            tensor<fp16, [2, 77, 1, 64]> var_59016_cast = slice_by_index(begin = var_59016_begin_0, end = var_59016_end_0, end_mask = var_59016_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_59016_cast")];
            tensor<int32, [4]> var_59020_begin_0 = const()[name = tensor<string, []>("op_59020_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_59020_end_0 = const()[name = tensor<string, []>("op_59020_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_59020_end_mask_0 = const()[name = tensor<string, []>("op_59020_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59020_cast = slice_by_index(begin = var_59020_begin_0, end = var_59020_end_0, end_mask = var_59020_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_59020_cast")];
            tensor<int32, [4]> var_59024_begin_0 = const()[name = tensor<string, []>("op_59024_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_59024_end_0 = const()[name = tensor<string, []>("op_59024_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_59024_end_mask_0 = const()[name = tensor<string, []>("op_59024_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59024_cast = slice_by_index(begin = var_59024_begin_0, end = var_59024_end_0, end_mask = var_59024_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_59024_cast")];
            tensor<int32, [4]> var_59028_begin_0 = const()[name = tensor<string, []>("op_59028_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_59028_end_0 = const()[name = tensor<string, []>("op_59028_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_59028_end_mask_0 = const()[name = tensor<string, []>("op_59028_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59028_cast = slice_by_index(begin = var_59028_begin_0, end = var_59028_end_0, end_mask = var_59028_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_59028_cast")];
            tensor<int32, [4]> var_59032_begin_0 = const()[name = tensor<string, []>("op_59032_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_59032_end_0 = const()[name = tensor<string, []>("op_59032_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_59032_end_mask_0 = const()[name = tensor<string, []>("op_59032_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59032_cast = slice_by_index(begin = var_59032_begin_0, end = var_59032_end_0, end_mask = var_59032_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_59032_cast")];
            tensor<int32, [4]> var_59036_begin_0 = const()[name = tensor<string, []>("op_59036_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_59036_end_0 = const()[name = tensor<string, []>("op_59036_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_59036_end_mask_0 = const()[name = tensor<string, []>("op_59036_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59036_cast = slice_by_index(begin = var_59036_begin_0, end = var_59036_end_0, end_mask = var_59036_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_59036_cast")];
            tensor<int32, [4]> var_59040_begin_0 = const()[name = tensor<string, []>("op_59040_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_59040_end_0 = const()[name = tensor<string, []>("op_59040_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_59040_end_mask_0 = const()[name = tensor<string, []>("op_59040_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59040_cast = slice_by_index(begin = var_59040_begin_0, end = var_59040_end_0, end_mask = var_59040_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_59040_cast")];
            tensor<int32, [4]> var_59044_begin_0 = const()[name = tensor<string, []>("op_59044_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_59044_end_0 = const()[name = tensor<string, []>("op_59044_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_59044_end_mask_0 = const()[name = tensor<string, []>("op_59044_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59044_cast = slice_by_index(begin = var_59044_begin_0, end = var_59044_end_0, end_mask = var_59044_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_59044_cast")];
            tensor<int32, [4]> var_59048_begin_0 = const()[name = tensor<string, []>("op_59048_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_59048_end_0 = const()[name = tensor<string, []>("op_59048_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_59048_end_mask_0 = const()[name = tensor<string, []>("op_59048_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59048_cast = slice_by_index(begin = var_59048_begin_0, end = var_59048_end_0, end_mask = var_59048_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_59048_cast")];
            tensor<int32, [4]> var_59052_begin_0 = const()[name = tensor<string, []>("op_59052_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_59052_end_0 = const()[name = tensor<string, []>("op_59052_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_59052_end_mask_0 = const()[name = tensor<string, []>("op_59052_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59052_cast = slice_by_index(begin = var_59052_begin_0, end = var_59052_end_0, end_mask = var_59052_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_59052_cast")];
            tensor<int32, [4]> var_59054_begin_0 = const()[name = tensor<string, []>("op_59054_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59054_end_0 = const()[name = tensor<string, []>("op_59054_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_59054_end_mask_0 = const()[name = tensor<string, []>("op_59054_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59054_cast = slice_by_index(begin = var_59054_begin_0, end = var_59054_end_0, end_mask = var_59054_end_mask_0, x = v_263_cast)[name = tensor<string, []>("op_59054_cast")];
            tensor<int32, [4]> var_59058_begin_0 = const()[name = tensor<string, []>("op_59058_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_59058_end_0 = const()[name = tensor<string, []>("op_59058_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_59058_end_mask_0 = const()[name = tensor<string, []>("op_59058_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59058_cast = slice_by_index(begin = var_59058_begin_0, end = var_59058_end_0, end_mask = var_59058_end_mask_0, x = v_263_cast)[name = tensor<string, []>("op_59058_cast")];
            tensor<int32, [4]> var_59062_begin_0 = const()[name = tensor<string, []>("op_59062_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_59062_end_0 = const()[name = tensor<string, []>("op_59062_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_59062_end_mask_0 = const()[name = tensor<string, []>("op_59062_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59062_cast = slice_by_index(begin = var_59062_begin_0, end = var_59062_end_0, end_mask = var_59062_end_mask_0, x = v_263_cast)[name = tensor<string, []>("op_59062_cast")];
            tensor<int32, [4]> var_59066_begin_0 = const()[name = tensor<string, []>("op_59066_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_59066_end_0 = const()[name = tensor<string, []>("op_59066_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_59066_end_mask_0 = const()[name = tensor<string, []>("op_59066_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59066_cast = slice_by_index(begin = var_59066_begin_0, end = var_59066_end_0, end_mask = var_59066_end_mask_0, x = v_263_cast)[name = tensor<string, []>("op_59066_cast")];
            tensor<int32, [4]> var_59070_begin_0 = const()[name = tensor<string, []>("op_59070_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_59070_end_0 = const()[name = tensor<string, []>("op_59070_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_59070_end_mask_0 = const()[name = tensor<string, []>("op_59070_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59070_cast = slice_by_index(begin = var_59070_begin_0, end = var_59070_end_0, end_mask = var_59070_end_mask_0, x = v_263_cast)[name = tensor<string, []>("op_59070_cast")];
            tensor<int32, [4]> var_59074_begin_0 = const()[name = tensor<string, []>("op_59074_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_59074_end_0 = const()[name = tensor<string, []>("op_59074_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_59074_end_mask_0 = const()[name = tensor<string, []>("op_59074_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59074_cast = slice_by_index(begin = var_59074_begin_0, end = var_59074_end_0, end_mask = var_59074_end_mask_0, x = v_263_cast)[name = tensor<string, []>("op_59074_cast")];
            tensor<int32, [4]> var_59078_begin_0 = const()[name = tensor<string, []>("op_59078_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_59078_end_0 = const()[name = tensor<string, []>("op_59078_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_59078_end_mask_0 = const()[name = tensor<string, []>("op_59078_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59078_cast = slice_by_index(begin = var_59078_begin_0, end = var_59078_end_0, end_mask = var_59078_end_mask_0, x = v_263_cast)[name = tensor<string, []>("op_59078_cast")];
            tensor<int32, [4]> var_59082_begin_0 = const()[name = tensor<string, []>("op_59082_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_59082_end_0 = const()[name = tensor<string, []>("op_59082_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_59082_end_mask_0 = const()[name = tensor<string, []>("op_59082_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59082_cast = slice_by_index(begin = var_59082_begin_0, end = var_59082_end_0, end_mask = var_59082_end_mask_0, x = v_263_cast)[name = tensor<string, []>("op_59082_cast")];
            tensor<int32, [4]> var_59086_begin_0 = const()[name = tensor<string, []>("op_59086_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_59086_end_0 = const()[name = tensor<string, []>("op_59086_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_59086_end_mask_0 = const()[name = tensor<string, []>("op_59086_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59086_cast = slice_by_index(begin = var_59086_begin_0, end = var_59086_end_0, end_mask = var_59086_end_mask_0, x = v_263_cast)[name = tensor<string, []>("op_59086_cast")];
            tensor<int32, [4]> var_59090_begin_0 = const()[name = tensor<string, []>("op_59090_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_59090_end_0 = const()[name = tensor<string, []>("op_59090_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_59090_end_mask_0 = const()[name = tensor<string, []>("op_59090_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59090_cast = slice_by_index(begin = var_59090_begin_0, end = var_59090_end_0, end_mask = var_59090_end_mask_0, x = v_263_cast)[name = tensor<string, []>("op_59090_cast")];
            tensor<string, []> var_59094_equation_0 = const()[name = tensor<string, []>("op_59094_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59094_cast = einsum(equation = var_59094_equation_0, values = (var_59016_cast, var_58973_cast))[name = tensor<string, []>("op_59094_cast")];
            tensor<fp16, []> var_59095_to_fp16 = const()[name = tensor<string, []>("op_59095_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5021_cast = mul(x = var_59094_cast, y = var_59095_to_fp16)[name = tensor<string, []>("aw_5021_cast")];
            tensor<string, []> var_59098_equation_0 = const()[name = tensor<string, []>("op_59098_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59098_cast = einsum(equation = var_59098_equation_0, values = (var_59020_cast, var_58977_cast))[name = tensor<string, []>("op_59098_cast")];
            tensor<fp16, []> var_59099_to_fp16 = const()[name = tensor<string, []>("op_59099_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5023_cast = mul(x = var_59098_cast, y = var_59099_to_fp16)[name = tensor<string, []>("aw_5023_cast")];
            tensor<string, []> var_59102_equation_0 = const()[name = tensor<string, []>("op_59102_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59102_cast = einsum(equation = var_59102_equation_0, values = (var_59024_cast, var_58981_cast))[name = tensor<string, []>("op_59102_cast")];
            tensor<fp16, []> var_59103_to_fp16 = const()[name = tensor<string, []>("op_59103_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5025_cast = mul(x = var_59102_cast, y = var_59103_to_fp16)[name = tensor<string, []>("aw_5025_cast")];
            tensor<string, []> var_59106_equation_0 = const()[name = tensor<string, []>("op_59106_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59106_cast = einsum(equation = var_59106_equation_0, values = (var_59028_cast, var_58985_cast))[name = tensor<string, []>("op_59106_cast")];
            tensor<fp16, []> var_59107_to_fp16 = const()[name = tensor<string, []>("op_59107_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5027_cast = mul(x = var_59106_cast, y = var_59107_to_fp16)[name = tensor<string, []>("aw_5027_cast")];
            tensor<string, []> var_59110_equation_0 = const()[name = tensor<string, []>("op_59110_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59110_cast = einsum(equation = var_59110_equation_0, values = (var_59032_cast, var_58989_cast))[name = tensor<string, []>("op_59110_cast")];
            tensor<fp16, []> var_59111_to_fp16 = const()[name = tensor<string, []>("op_59111_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5029_cast = mul(x = var_59110_cast, y = var_59111_to_fp16)[name = tensor<string, []>("aw_5029_cast")];
            tensor<string, []> var_59114_equation_0 = const()[name = tensor<string, []>("op_59114_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59114_cast = einsum(equation = var_59114_equation_0, values = (var_59036_cast, var_58993_cast))[name = tensor<string, []>("op_59114_cast")];
            tensor<fp16, []> var_59115_to_fp16 = const()[name = tensor<string, []>("op_59115_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5031_cast = mul(x = var_59114_cast, y = var_59115_to_fp16)[name = tensor<string, []>("aw_5031_cast")];
            tensor<string, []> var_59118_equation_0 = const()[name = tensor<string, []>("op_59118_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59118_cast = einsum(equation = var_59118_equation_0, values = (var_59040_cast, var_58997_cast))[name = tensor<string, []>("op_59118_cast")];
            tensor<fp16, []> var_59119_to_fp16 = const()[name = tensor<string, []>("op_59119_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5033_cast = mul(x = var_59118_cast, y = var_59119_to_fp16)[name = tensor<string, []>("aw_5033_cast")];
            tensor<string, []> var_59122_equation_0 = const()[name = tensor<string, []>("op_59122_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59122_cast = einsum(equation = var_59122_equation_0, values = (var_59044_cast, var_59001_cast))[name = tensor<string, []>("op_59122_cast")];
            tensor<fp16, []> var_59123_to_fp16 = const()[name = tensor<string, []>("op_59123_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5035_cast = mul(x = var_59122_cast, y = var_59123_to_fp16)[name = tensor<string, []>("aw_5035_cast")];
            tensor<string, []> var_59126_equation_0 = const()[name = tensor<string, []>("op_59126_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59126_cast = einsum(equation = var_59126_equation_0, values = (var_59048_cast, var_59005_cast))[name = tensor<string, []>("op_59126_cast")];
            tensor<fp16, []> var_59127_to_fp16 = const()[name = tensor<string, []>("op_59127_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5037_cast = mul(x = var_59126_cast, y = var_59127_to_fp16)[name = tensor<string, []>("aw_5037_cast")];
            tensor<string, []> var_59130_equation_0 = const()[name = tensor<string, []>("op_59130_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59130_cast = einsum(equation = var_59130_equation_0, values = (var_59052_cast, var_59009_cast))[name = tensor<string, []>("op_59130_cast")];
            tensor<fp16, []> var_59131_to_fp16 = const()[name = tensor<string, []>("op_59131_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5039_cast = mul(x = var_59130_cast, y = var_59131_to_fp16)[name = tensor<string, []>("aw_5039_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59133_cast = softmax(axis = var_58064, x = aw_5021_cast)[name = tensor<string, []>("op_59133_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59134_cast = softmax(axis = var_58064, x = aw_5023_cast)[name = tensor<string, []>("op_59134_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59135_cast = softmax(axis = var_58064, x = aw_5025_cast)[name = tensor<string, []>("op_59135_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59136_cast = softmax(axis = var_58064, x = aw_5027_cast)[name = tensor<string, []>("op_59136_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59137_cast = softmax(axis = var_58064, x = aw_5029_cast)[name = tensor<string, []>("op_59137_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59138_cast = softmax(axis = var_58064, x = aw_5031_cast)[name = tensor<string, []>("op_59138_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59139_cast = softmax(axis = var_58064, x = aw_5033_cast)[name = tensor<string, []>("op_59139_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59140_cast = softmax(axis = var_58064, x = aw_5035_cast)[name = tensor<string, []>("op_59140_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59141_cast = softmax(axis = var_58064, x = aw_5037_cast)[name = tensor<string, []>("op_59141_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59142_cast = softmax(axis = var_58064, x = aw_5039_cast)[name = tensor<string, []>("op_59142_cast")];
            tensor<string, []> var_59144_equation_0 = const()[name = tensor<string, []>("op_59144_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59144_cast = einsum(equation = var_59144_equation_0, values = (var_59054_cast, var_59133_cast))[name = tensor<string, []>("op_59144_cast")];
            tensor<string, []> var_59146_equation_0 = const()[name = tensor<string, []>("op_59146_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59146_cast = einsum(equation = var_59146_equation_0, values = (var_59058_cast, var_59134_cast))[name = tensor<string, []>("op_59146_cast")];
            tensor<string, []> var_59148_equation_0 = const()[name = tensor<string, []>("op_59148_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59148_cast = einsum(equation = var_59148_equation_0, values = (var_59062_cast, var_59135_cast))[name = tensor<string, []>("op_59148_cast")];
            tensor<string, []> var_59150_equation_0 = const()[name = tensor<string, []>("op_59150_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59150_cast = einsum(equation = var_59150_equation_0, values = (var_59066_cast, var_59136_cast))[name = tensor<string, []>("op_59150_cast")];
            tensor<string, []> var_59152_equation_0 = const()[name = tensor<string, []>("op_59152_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59152_cast = einsum(equation = var_59152_equation_0, values = (var_59070_cast, var_59137_cast))[name = tensor<string, []>("op_59152_cast")];
            tensor<string, []> var_59154_equation_0 = const()[name = tensor<string, []>("op_59154_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59154_cast = einsum(equation = var_59154_equation_0, values = (var_59074_cast, var_59138_cast))[name = tensor<string, []>("op_59154_cast")];
            tensor<string, []> var_59156_equation_0 = const()[name = tensor<string, []>("op_59156_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59156_cast = einsum(equation = var_59156_equation_0, values = (var_59078_cast, var_59139_cast))[name = tensor<string, []>("op_59156_cast")];
            tensor<string, []> var_59158_equation_0 = const()[name = tensor<string, []>("op_59158_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59158_cast = einsum(equation = var_59158_equation_0, values = (var_59082_cast, var_59140_cast))[name = tensor<string, []>("op_59158_cast")];
            tensor<string, []> var_59160_equation_0 = const()[name = tensor<string, []>("op_59160_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59160_cast = einsum(equation = var_59160_equation_0, values = (var_59086_cast, var_59141_cast))[name = tensor<string, []>("op_59160_cast")];
            tensor<string, []> var_59162_equation_0 = const()[name = tensor<string, []>("op_59162_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59162_cast = einsum(equation = var_59162_equation_0, values = (var_59090_cast, var_59142_cast))[name = tensor<string, []>("op_59162_cast")];
            tensor<bool, []> input_749_interleave_0 = const()[name = tensor<string, []>("input_749_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_749_cast = concat(axis = var_58064, interleave = input_749_interleave_0, values = (var_59144_cast, var_59146_cast, var_59148_cast, var_59150_cast, var_59152_cast, var_59154_cast, var_59156_cast, var_59158_cast, var_59160_cast, var_59162_cast))[name = tensor<string, []>("input_749_cast")];
            tensor<int32, [2]> var_59168 = const()[name = tensor<string, []>("op_59168"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59170 = const()[name = tensor<string, []>("op_59170"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_59172_pad_type_0 = const()[name = tensor<string, []>("op_59172_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_59172_pad_0 = const()[name = tensor<string, []>("op_59172_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4963778176)))];
            tensor<fp16, [640]> up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4964597440)))];
            tensor<fp16, [2, 640, 1, 4096]> var_59172_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_59170, groups = var_58064, pad = var_59172_pad_0, pad_type = var_59172_pad_type_0, strides = var_59168, weight = up_blocks_1_attentions_0_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_749_cast)[name = tensor<string, []>("op_59172_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_395_cast = add(x = var_59172_cast, y = inputs_393_cast)[name = tensor<string, []>("inputs_395_cast")];
            tensor<int32, [1]> var_59176 = const()[name = tensor<string, []>("op_59176"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_395_cast = reduce_mean(axes = var_59176, keep_dims = var_58059, x = inputs_395_cast)[name = tensor<string, []>("channels_mean_395_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_395_cast = sub(x = inputs_395_cast, y = channels_mean_395_cast)[name = tensor<string, []>("zero_mean_395_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_395_cast = mul(x = zero_mean_395_cast, y = zero_mean_395_cast)[name = tensor<string, []>("zero_mean_sq_395_cast")];
            tensor<int32, [1]> var_59180 = const()[name = tensor<string, []>("op_59180"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_59181_cast = reduce_mean(axes = var_59180, keep_dims = var_58059, x = zero_mean_sq_395_cast)[name = tensor<string, []>("op_59181_cast")];
            tensor<fp16, []> var_59182_to_fp16 = const()[name = tensor<string, []>("op_59182_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_59183_cast = add(x = var_59181_cast, y = var_59182_to_fp16)[name = tensor<string, []>("op_59183_cast")];
            tensor<fp16, []> denom_395_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_395_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_395_cast = rsqrt(epsilon = denom_395_epsilon_0_to_fp16, x = var_59183_cast)[name = tensor<string, []>("denom_395_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_395_cast = mul(x = zero_mean_395_cast, y = denom_395_cast)[name = tensor<string, []>("out_395_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_59187_to_fp16 = const()[name = tensor<string, []>("op_59187_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4964598784)))];
            tensor<fp16, [2, 640, 1, 4096]> var_59188_cast = add(x = out_395_cast, y = var_59187_to_fp16)[name = tensor<string, []>("op_59188_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_59190_to_fp16 = const()[name = tensor<string, []>("op_59190_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4964600128)))];
            tensor<fp16, [2, 640, 1, 4096]> input_751_cast = mul(x = var_59188_cast, y = var_59190_to_fp16)[name = tensor<string, []>("input_751_cast")];
            tensor<int32, [2]> var_59198 = const()[name = tensor<string, []>("op_59198"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59200 = const()[name = tensor<string, []>("op_59200"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_59202_pad_type_0 = const()[name = tensor<string, []>("op_59202_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_59202_pad_0 = const()[name = tensor<string, []>("op_59202_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [5120, 640, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [5120, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4964601472)))];
            tensor<fp16, [5120]> up_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4971155136)))];
            tensor<fp16, [2, 5120, 1, 4096]> var_59202_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_59200, groups = var_58064, pad = var_59202_pad_0, pad_type = var_59202_pad_type_0, strides = var_59198, weight = up_blocks_1_attentions_0_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_751_cast)[name = tensor<string, []>("op_59202_cast")];
            tensor<int32, [2]> var_59203_split_sizes_0 = const()[name = tensor<string, []>("op_59203_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_59203_axis_0 = const()[name = tensor<string, []>("op_59203_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 4096]> var_59203_cast_0, tensor<fp16, [2, 2560, 1, 4096]> var_59203_cast_1 = split(axis = var_59203_axis_0, split_sizes = var_59203_split_sizes_0, x = var_59202_cast)[name = tensor<string, []>("op_59203_cast")];
            tensor<string, []> var_59205_mode_0 = const()[name = tensor<string, []>("op_59205_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 4096]> var_59205_cast = gelu(mode = var_59205_mode_0, x = var_59203_cast_1)[name = tensor<string, []>("op_59205_cast")];
            tensor<fp16, [2, 2560, 1, 4096]> input_753_cast = mul(x = var_59203_cast_0, y = var_59205_cast)[name = tensor<string, []>("input_753_cast")];
            tensor<int32, [2]> var_59209 = const()[name = tensor<string, []>("op_59209"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59211 = const()[name = tensor<string, []>("op_59211"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_59213_pad_type_0 = const()[name = tensor<string, []>("op_59213_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_59213_pad_0 = const()[name = tensor<string, []>("op_59213_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2560, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor<fp16, [640, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4971165440)))];
            tensor<fp16, [640]> up_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4974442304)))];
            tensor<fp16, [2, 640, 1, 4096]> var_59213_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_59211, groups = var_58064, pad = var_59213_pad_0, pad_type = var_59213_pad_type_0, strides = var_59209, weight = up_blocks_1_attentions_0_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_753_cast)[name = tensor<string, []>("op_59213_cast")];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_521_cast = add(x = var_59213_cast, y = inputs_395_cast)[name = tensor<string, []>("hidden_states_521_cast")];
            tensor<int32, [4]> var_59215 = const()[name = tensor<string, []>("op_59215"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> input_755_cast = reshape(shape = var_59215, x = hidden_states_521_cast)[name = tensor<string, []>("input_755_cast")];
            tensor<int32, [2]> var_59219 = const()[name = tensor<string, []>("op_59219"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59221 = const()[name = tensor<string, []>("op_59221"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_523_pad_type_0 = const()[name = tensor<string, []>("hidden_states_523_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_523_pad_0 = const()[name = tensor<string, []>("hidden_states_523_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_proj_out_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4974443648)))];
            tensor<fp16, [640]> up_blocks_1_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_proj_out_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4975262912)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_523_cast = conv(bias = up_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = var_59221, groups = var_58064, pad = hidden_states_523_pad_0, pad_type = hidden_states_523_pad_type_0, strides = var_59219, weight = up_blocks_1_attentions_0_proj_out_weight_to_fp16, x = input_755_cast)[name = tensor<string, []>("hidden_states_523_cast")];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_525_cast = add(x = hidden_states_523_cast, y = hidden_states_505_cast)[name = tensor<string, []>("hidden_states_525_cast")];
            tensor<bool, []> input_757_interleave_0 = const()[name = tensor<string, []>("input_757_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 64, 64]> input_757_cast = concat(axis = var_58064, interleave = input_757_interleave_0, values = (hidden_states_525_cast, res_hidden_states_9_cast))[name = tensor<string, []>("input_757_cast")];
            tensor<int32, [5]> reshape_132_shape_0 = const()[name = tensor<string, []>("reshape_132_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 64, 64])];
            tensor<fp16, [2, 32, 40, 64, 64]> reshape_132_cast = reshape(shape = reshape_132_shape_0, x = input_757_cast)[name = tensor<string, []>("reshape_132_cast")];
            tensor<int32, [3]> reduce_mean_99_axes_0 = const()[name = tensor<string, []>("reduce_mean_99_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_99_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_99_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_99_cast = reduce_mean(axes = reduce_mean_99_axes_0, keep_dims = reduce_mean_99_keep_dims_0, x = reshape_132_cast)[name = tensor<string, []>("reduce_mean_99_cast")];
            tensor<fp16, [2, 32, 40, 64, 64]> sub_66_cast = sub(x = reshape_132_cast, y = reduce_mean_99_cast)[name = tensor<string, []>("sub_66_cast")];
            tensor<fp16, [2, 32, 40, 64, 64]> square_33_cast = square(x = sub_66_cast)[name = tensor<string, []>("square_33_cast")];
            tensor<int32, [3]> reduce_mean_101_axes_0 = const()[name = tensor<string, []>("reduce_mean_101_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_101_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_101_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_101_cast = reduce_mean(axes = reduce_mean_101_axes_0, keep_dims = reduce_mean_101_keep_dims_0, x = square_33_cast)[name = tensor<string, []>("reduce_mean_101_cast")];
            tensor<fp16, []> add_66_y_0_to_fp16 = const()[name = tensor<string, []>("add_66_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_66_cast = add(x = reduce_mean_101_cast, y = add_66_y_0_to_fp16)[name = tensor<string, []>("add_66_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_33_cast = sqrt(x = add_66_cast)[name = tensor<string, []>("sqrt_33_cast")];
            tensor<fp16, [2, 32, 40, 64, 64]> real_div_33_cast = real_div(x = sub_66_cast, y = sqrt_33_cast)[name = tensor<string, []>("real_div_33_cast")];
            tensor<int32, [4]> reshape_133_shape_0 = const()[name = tensor<string, []>("reshape_133_shape_0"), val = tensor<int32, [4]>([2, 1280, 64, 64])];
            tensor<fp16, [2, 1280, 64, 64]> reshape_133_cast = reshape(shape = reshape_133_shape_0, x = real_div_33_cast)[name = tensor<string, []>("reshape_133_cast")];
            tensor<fp16, [1280]> add_67_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_67_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4975264256)))];
            tensor<fp16, [1280]> add_67_beta_0_to_fp16 = const()[name = tensor<string, []>("add_67_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4975266880)))];
            tensor<fp16, []> add_67_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_67_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 64, 64]> add_67_cast = batch_norm(beta = add_67_beta_0_to_fp16, epsilon = add_67_epsilon_0_to_fp16, gamma = add_67_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_133_cast)[name = tensor<string, []>("add_67_cast")];
            tensor<fp16, [2, 1280, 64, 64]> input_761_cast = silu(x = add_67_cast)[name = tensor<string, []>("input_761_cast")];
            tensor<int32, [2]> var_59239 = const()[name = tensor<string, []>("op_59239"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59241 = const()[name = tensor<string, []>("op_59241"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_527_pad_type_0 = const()[name = tensor<string, []>("hidden_states_527_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_527_pad_0 = const()[name = tensor<string, []>("hidden_states_527_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 1280, 3, 3]> up_blocks_1_resnets_1_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_1_conv1_weight_to_fp16"), val = tensor<fp16, [640, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4975269504)))];
            tensor<fp16, [640]> up_blocks_1_resnets_1_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_1_conv1_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4990015168)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_527_cast = conv(bias = up_blocks_1_resnets_1_conv1_bias_to_fp16, dilations = var_59241, groups = var_58064, pad = hidden_states_527_pad_0, pad_type = hidden_states_527_pad_type_0, strides = var_59239, weight = up_blocks_1_resnets_1_conv1_weight_to_fp16, x = input_761_cast)[name = tensor<string, []>("hidden_states_527_cast")];
            tensor<int32, [2]> var_59247 = const()[name = tensor<string, []>("op_59247"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59249 = const()[name = tensor<string, []>("op_59249"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_25_pad_type_0 = const()[name = tensor<string, []>("temb_25_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_25_pad_0 = const()[name = tensor<string, []>("temb_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1280, 1, 1]> up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [640, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4990016512)))];
            tensor<fp16, [640]> up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4991654976)))];
            tensor<fp16, [2, 640, 1, 1]> temb_25_cast = conv(bias = up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_59249, groups = var_58064, pad = temb_25_pad_0, pad_type = temb_25_pad_type_0, strides = var_59247, weight = up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16, x = input_23_cast)[name = tensor<string, []>("temb_25_cast")];
            tensor<fp16, [2, 640, 64, 64]> input_765_cast = add(x = hidden_states_527_cast, y = temb_25_cast)[name = tensor<string, []>("input_765_cast")];
            tensor<int32, [5]> reshape_136_shape_0 = const()[name = tensor<string, []>("reshape_136_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 64, 64])];
            tensor<fp16, [2, 32, 20, 64, 64]> reshape_136_cast = reshape(shape = reshape_136_shape_0, x = input_765_cast)[name = tensor<string, []>("reshape_136_cast")];
            tensor<int32, [3]> reduce_mean_102_axes_0 = const()[name = tensor<string, []>("reduce_mean_102_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_102_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_102_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_102_cast = reduce_mean(axes = reduce_mean_102_axes_0, keep_dims = reduce_mean_102_keep_dims_0, x = reshape_136_cast)[name = tensor<string, []>("reduce_mean_102_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> sub_68_cast = sub(x = reshape_136_cast, y = reduce_mean_102_cast)[name = tensor<string, []>("sub_68_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> square_34_cast = square(x = sub_68_cast)[name = tensor<string, []>("square_34_cast")];
            tensor<int32, [3]> reduce_mean_104_axes_0 = const()[name = tensor<string, []>("reduce_mean_104_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_104_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_104_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_104_cast = reduce_mean(axes = reduce_mean_104_axes_0, keep_dims = reduce_mean_104_keep_dims_0, x = square_34_cast)[name = tensor<string, []>("reduce_mean_104_cast")];
            tensor<fp16, []> add_68_y_0_to_fp16 = const()[name = tensor<string, []>("add_68_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_68_cast = add(x = reduce_mean_104_cast, y = add_68_y_0_to_fp16)[name = tensor<string, []>("add_68_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_34_cast = sqrt(x = add_68_cast)[name = tensor<string, []>("sqrt_34_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_34_cast = real_div(x = sub_68_cast, y = sqrt_34_cast)[name = tensor<string, []>("real_div_34_cast")];
            tensor<int32, [4]> reshape_137_shape_0 = const()[name = tensor<string, []>("reshape_137_shape_0"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> reshape_137_cast = reshape(shape = reshape_137_shape_0, x = real_div_34_cast)[name = tensor<string, []>("reshape_137_cast")];
            tensor<fp16, [640]> add_69_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_69_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4991656320)))];
            tensor<fp16, [640]> add_69_beta_0_to_fp16 = const()[name = tensor<string, []>("add_69_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4991657664)))];
            tensor<fp16, []> add_69_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_69_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 64, 64]> add_69_cast = batch_norm(beta = add_69_beta_0_to_fp16, epsilon = add_69_epsilon_0_to_fp16, gamma = add_69_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_137_cast)[name = tensor<string, []>("add_69_cast")];
            tensor<fp16, [2, 640, 64, 64]> input_769_cast = silu(x = add_69_cast)[name = tensor<string, []>("input_769_cast")];
            tensor<int32, [2]> var_59259 = const()[name = tensor<string, []>("op_59259"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59261 = const()[name = tensor<string, []>("op_59261"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_529_pad_type_0 = const()[name = tensor<string, []>("hidden_states_529_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_529_pad_0 = const()[name = tensor<string, []>("hidden_states_529_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 640, 3, 3]> up_blocks_1_resnets_1_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_1_conv2_weight_to_fp16"), val = tensor<fp16, [640, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4991659008)))];
            tensor<fp16, [640]> up_blocks_1_resnets_1_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_1_conv2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4999031872)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_529_cast = conv(bias = up_blocks_1_resnets_1_conv2_bias_to_fp16, dilations = var_59261, groups = var_58064, pad = hidden_states_529_pad_0, pad_type = hidden_states_529_pad_type_0, strides = var_59259, weight = up_blocks_1_resnets_1_conv2_weight_to_fp16, x = input_769_cast)[name = tensor<string, []>("hidden_states_529_cast")];
            tensor<int32, [2]> var_59266 = const()[name = tensor<string, []>("op_59266"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59268 = const()[name = tensor<string, []>("op_59268"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_13_pad_type_0 = const()[name = tensor<string, []>("x_13_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_13_pad_0 = const()[name = tensor<string, []>("x_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1280, 1, 1]> up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [640, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4999033216)))];
            tensor<fp16, [640]> up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5000671680)))];
            tensor<fp16, [2, 640, 64, 64]> x_13_cast = conv(bias = up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_59268, groups = var_58064, pad = x_13_pad_0, pad_type = x_13_pad_type_0, strides = var_59266, weight = up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16, x = input_757_cast)[name = tensor<string, []>("x_13_cast")];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_531_cast = add(x = x_13_cast, y = hidden_states_529_cast)[name = tensor<string, []>("hidden_states_531_cast")];
            tensor<int32, [5]> reshape_140_shape_0 = const()[name = tensor<string, []>("reshape_140_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 64, 64])];
            tensor<fp16, [2, 32, 20, 64, 64]> reshape_140_cast = reshape(shape = reshape_140_shape_0, x = hidden_states_531_cast)[name = tensor<string, []>("reshape_140_cast")];
            tensor<int32, [3]> reduce_mean_105_axes_0 = const()[name = tensor<string, []>("reduce_mean_105_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_105_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_105_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_105_cast = reduce_mean(axes = reduce_mean_105_axes_0, keep_dims = reduce_mean_105_keep_dims_0, x = reshape_140_cast)[name = tensor<string, []>("reduce_mean_105_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> sub_70_cast = sub(x = reshape_140_cast, y = reduce_mean_105_cast)[name = tensor<string, []>("sub_70_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> square_35_cast = square(x = sub_70_cast)[name = tensor<string, []>("square_35_cast")];
            tensor<int32, [3]> reduce_mean_107_axes_0 = const()[name = tensor<string, []>("reduce_mean_107_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_107_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_107_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_107_cast = reduce_mean(axes = reduce_mean_107_axes_0, keep_dims = reduce_mean_107_keep_dims_0, x = square_35_cast)[name = tensor<string, []>("reduce_mean_107_cast")];
            tensor<fp16, []> add_70_y_0_to_fp16 = const()[name = tensor<string, []>("add_70_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_70_cast = add(x = reduce_mean_107_cast, y = add_70_y_0_to_fp16)[name = tensor<string, []>("add_70_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_35_cast = sqrt(x = add_70_cast)[name = tensor<string, []>("sqrt_35_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_35_cast = real_div(x = sub_70_cast, y = sqrt_35_cast)[name = tensor<string, []>("real_div_35_cast")];
            tensor<int32, [4]> reshape_141_shape_0 = const()[name = tensor<string, []>("reshape_141_shape_0"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> reshape_141_cast = reshape(shape = reshape_141_shape_0, x = real_div_35_cast)[name = tensor<string, []>("reshape_141_cast")];
            tensor<fp16, [640]> add_71_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_71_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5000673024)))];
            tensor<fp16, [640]> add_71_beta_0_to_fp16 = const()[name = tensor<string, []>("add_71_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5000674368)))];
            tensor<fp16, []> add_71_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_71_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 64, 64]> add_71_cast = batch_norm(beta = add_71_beta_0_to_fp16, epsilon = add_71_epsilon_0_to_fp16, gamma = add_71_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_141_cast)[name = tensor<string, []>("add_71_cast")];
            tensor<int32, [2]> var_59290 = const()[name = tensor<string, []>("op_59290"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59292 = const()[name = tensor<string, []>("op_59292"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_533_pad_type_0 = const()[name = tensor<string, []>("hidden_states_533_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_533_pad_0 = const()[name = tensor<string, []>("hidden_states_533_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_proj_in_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5000675712)))];
            tensor<fp16, [640]> up_blocks_1_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_proj_in_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5001494976)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_533_cast = conv(bias = up_blocks_1_attentions_1_proj_in_bias_to_fp16, dilations = var_59292, groups = var_58064, pad = hidden_states_533_pad_0, pad_type = hidden_states_533_pad_type_0, strides = var_59290, weight = up_blocks_1_attentions_1_proj_in_weight_to_fp16, x = add_71_cast)[name = tensor<string, []>("hidden_states_533_cast")];
            tensor<int32, [4]> var_59297 = const()[name = tensor<string, []>("op_59297"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<fp16, [2, 640, 1, 4096]> inputs_397_cast = reshape(shape = var_59297, x = hidden_states_533_cast)[name = tensor<string, []>("inputs_397_cast")];
            tensor<int32, [1]> var_59307 = const()[name = tensor<string, []>("op_59307"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_397_cast = reduce_mean(axes = var_59307, keep_dims = var_58059, x = inputs_397_cast)[name = tensor<string, []>("channels_mean_397_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_397_cast = sub(x = inputs_397_cast, y = channels_mean_397_cast)[name = tensor<string, []>("zero_mean_397_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_397_cast = mul(x = zero_mean_397_cast, y = zero_mean_397_cast)[name = tensor<string, []>("zero_mean_sq_397_cast")];
            tensor<int32, [1]> var_59311 = const()[name = tensor<string, []>("op_59311"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_59312_cast = reduce_mean(axes = var_59311, keep_dims = var_58059, x = zero_mean_sq_397_cast)[name = tensor<string, []>("op_59312_cast")];
            tensor<fp16, []> var_59313_to_fp16 = const()[name = tensor<string, []>("op_59313_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_59314_cast = add(x = var_59312_cast, y = var_59313_to_fp16)[name = tensor<string, []>("op_59314_cast")];
            tensor<fp16, []> denom_397_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_397_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_397_cast = rsqrt(epsilon = denom_397_epsilon_0_to_fp16, x = var_59314_cast)[name = tensor<string, []>("denom_397_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_397_cast = mul(x = zero_mean_397_cast, y = denom_397_cast)[name = tensor<string, []>("out_397_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_59318_to_fp16 = const()[name = tensor<string, []>("op_59318_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5001496320)))];
            tensor<fp16, [2, 640, 1, 4096]> var_59319_cast = add(x = out_397_cast, y = var_59318_to_fp16)[name = tensor<string, []>("op_59319_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_59321_to_fp16 = const()[name = tensor<string, []>("op_59321_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5001497664)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_535_cast = mul(x = var_59319_cast, y = var_59321_to_fp16)[name = tensor<string, []>("hidden_states_535_cast")];
            tensor<int32, [2]> var_59328 = const()[name = tensor<string, []>("op_59328"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59330 = const()[name = tensor<string, []>("op_59330"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_265_pad_type_0 = const()[name = tensor<string, []>("q_265_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_265_pad_0 = const()[name = tensor<string, []>("q_265_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5001499008)))];
            tensor<fp16, [2, 640, 1, 4096]> q_265_cast = conv(dilations = var_59330, groups = var_58064, pad = q_265_pad_0, pad_type = q_265_pad_type_0, strides = var_59328, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_535_cast)[name = tensor<string, []>("q_265_cast")];
            tensor<int32, [2]> var_59334 = const()[name = tensor<string, []>("op_59334"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59336 = const()[name = tensor<string, []>("op_59336"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_529_pad_type_0 = const()[name = tensor<string, []>("k_529_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_529_pad_0 = const()[name = tensor<string, []>("k_529_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5002318272)))];
            tensor<fp16, [2, 640, 1, 4096]> k_529_cast = conv(dilations = var_59336, groups = var_58064, pad = k_529_pad_0, pad_type = k_529_pad_type_0, strides = var_59334, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_535_cast)[name = tensor<string, []>("k_529_cast")];
            tensor<int32, [2]> var_59340 = const()[name = tensor<string, []>("op_59340"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59342 = const()[name = tensor<string, []>("op_59342"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_265_pad_type_0 = const()[name = tensor<string, []>("v_265_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_265_pad_0 = const()[name = tensor<string, []>("v_265_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5003137536)))];
            tensor<fp16, [2, 640, 1, 4096]> v_265_cast = conv(dilations = var_59342, groups = var_58064, pad = v_265_pad_0, pad_type = v_265_pad_type_0, strides = var_59340, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_535_cast)[name = tensor<string, []>("v_265_cast")];
            tensor<int32, [4]> var_59346_begin_0 = const()[name = tensor<string, []>("op_59346_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59346_end_0 = const()[name = tensor<string, []>("op_59346_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_59346_end_mask_0 = const()[name = tensor<string, []>("op_59346_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59346_cast = slice_by_index(begin = var_59346_begin_0, end = var_59346_end_0, end_mask = var_59346_end_mask_0, x = q_265_cast)[name = tensor<string, []>("op_59346_cast")];
            tensor<int32, [4]> var_59350_begin_0 = const()[name = tensor<string, []>("op_59350_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_59350_end_0 = const()[name = tensor<string, []>("op_59350_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_59350_end_mask_0 = const()[name = tensor<string, []>("op_59350_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59350_cast = slice_by_index(begin = var_59350_begin_0, end = var_59350_end_0, end_mask = var_59350_end_mask_0, x = q_265_cast)[name = tensor<string, []>("op_59350_cast")];
            tensor<int32, [4]> var_59354_begin_0 = const()[name = tensor<string, []>("op_59354_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_59354_end_0 = const()[name = tensor<string, []>("op_59354_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_59354_end_mask_0 = const()[name = tensor<string, []>("op_59354_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59354_cast = slice_by_index(begin = var_59354_begin_0, end = var_59354_end_0, end_mask = var_59354_end_mask_0, x = q_265_cast)[name = tensor<string, []>("op_59354_cast")];
            tensor<int32, [4]> var_59358_begin_0 = const()[name = tensor<string, []>("op_59358_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_59358_end_0 = const()[name = tensor<string, []>("op_59358_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_59358_end_mask_0 = const()[name = tensor<string, []>("op_59358_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59358_cast = slice_by_index(begin = var_59358_begin_0, end = var_59358_end_0, end_mask = var_59358_end_mask_0, x = q_265_cast)[name = tensor<string, []>("op_59358_cast")];
            tensor<int32, [4]> var_59362_begin_0 = const()[name = tensor<string, []>("op_59362_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_59362_end_0 = const()[name = tensor<string, []>("op_59362_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_59362_end_mask_0 = const()[name = tensor<string, []>("op_59362_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59362_cast = slice_by_index(begin = var_59362_begin_0, end = var_59362_end_0, end_mask = var_59362_end_mask_0, x = q_265_cast)[name = tensor<string, []>("op_59362_cast")];
            tensor<int32, [4]> var_59366_begin_0 = const()[name = tensor<string, []>("op_59366_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_59366_end_0 = const()[name = tensor<string, []>("op_59366_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_59366_end_mask_0 = const()[name = tensor<string, []>("op_59366_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59366_cast = slice_by_index(begin = var_59366_begin_0, end = var_59366_end_0, end_mask = var_59366_end_mask_0, x = q_265_cast)[name = tensor<string, []>("op_59366_cast")];
            tensor<int32, [4]> var_59370_begin_0 = const()[name = tensor<string, []>("op_59370_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_59370_end_0 = const()[name = tensor<string, []>("op_59370_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_59370_end_mask_0 = const()[name = tensor<string, []>("op_59370_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59370_cast = slice_by_index(begin = var_59370_begin_0, end = var_59370_end_0, end_mask = var_59370_end_mask_0, x = q_265_cast)[name = tensor<string, []>("op_59370_cast")];
            tensor<int32, [4]> var_59374_begin_0 = const()[name = tensor<string, []>("op_59374_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_59374_end_0 = const()[name = tensor<string, []>("op_59374_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_59374_end_mask_0 = const()[name = tensor<string, []>("op_59374_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59374_cast = slice_by_index(begin = var_59374_begin_0, end = var_59374_end_0, end_mask = var_59374_end_mask_0, x = q_265_cast)[name = tensor<string, []>("op_59374_cast")];
            tensor<int32, [4]> var_59378_begin_0 = const()[name = tensor<string, []>("op_59378_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_59378_end_0 = const()[name = tensor<string, []>("op_59378_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_59378_end_mask_0 = const()[name = tensor<string, []>("op_59378_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59378_cast = slice_by_index(begin = var_59378_begin_0, end = var_59378_end_0, end_mask = var_59378_end_mask_0, x = q_265_cast)[name = tensor<string, []>("op_59378_cast")];
            tensor<int32, [4]> var_59382_begin_0 = const()[name = tensor<string, []>("op_59382_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_59382_end_0 = const()[name = tensor<string, []>("op_59382_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_59382_end_mask_0 = const()[name = tensor<string, []>("op_59382_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59382_cast = slice_by_index(begin = var_59382_begin_0, end = var_59382_end_0, end_mask = var_59382_end_mask_0, x = q_265_cast)[name = tensor<string, []>("op_59382_cast")];
            tensor<int32, [4]> k_531_perm_0 = const()[name = tensor<string, []>("k_531_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_59389_begin_0 = const()[name = tensor<string, []>("op_59389_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59389_end_0 = const()[name = tensor<string, []>("op_59389_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_59389_end_mask_0 = const()[name = tensor<string, []>("op_59389_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 640]> transpose_7 = transpose(perm = k_531_perm_0, x = k_529_cast)[name = tensor<string, []>("transpose_7")];
            tensor<fp16, [2, 4096, 1, 64]> var_59389_cast = slice_by_index(begin = var_59389_begin_0, end = var_59389_end_0, end_mask = var_59389_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_59389_cast")];
            tensor<int32, [4]> var_59393_begin_0 = const()[name = tensor<string, []>("op_59393_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_59393_end_0 = const()[name = tensor<string, []>("op_59393_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_59393_end_mask_0 = const()[name = tensor<string, []>("op_59393_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59393_cast = slice_by_index(begin = var_59393_begin_0, end = var_59393_end_0, end_mask = var_59393_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_59393_cast")];
            tensor<int32, [4]> var_59397_begin_0 = const()[name = tensor<string, []>("op_59397_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_59397_end_0 = const()[name = tensor<string, []>("op_59397_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_59397_end_mask_0 = const()[name = tensor<string, []>("op_59397_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59397_cast = slice_by_index(begin = var_59397_begin_0, end = var_59397_end_0, end_mask = var_59397_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_59397_cast")];
            tensor<int32, [4]> var_59401_begin_0 = const()[name = tensor<string, []>("op_59401_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_59401_end_0 = const()[name = tensor<string, []>("op_59401_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_59401_end_mask_0 = const()[name = tensor<string, []>("op_59401_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59401_cast = slice_by_index(begin = var_59401_begin_0, end = var_59401_end_0, end_mask = var_59401_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_59401_cast")];
            tensor<int32, [4]> var_59405_begin_0 = const()[name = tensor<string, []>("op_59405_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_59405_end_0 = const()[name = tensor<string, []>("op_59405_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_59405_end_mask_0 = const()[name = tensor<string, []>("op_59405_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59405_cast = slice_by_index(begin = var_59405_begin_0, end = var_59405_end_0, end_mask = var_59405_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_59405_cast")];
            tensor<int32, [4]> var_59409_begin_0 = const()[name = tensor<string, []>("op_59409_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_59409_end_0 = const()[name = tensor<string, []>("op_59409_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 384])];
            tensor<bool, [4]> var_59409_end_mask_0 = const()[name = tensor<string, []>("op_59409_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59409_cast = slice_by_index(begin = var_59409_begin_0, end = var_59409_end_0, end_mask = var_59409_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_59409_cast")];
            tensor<int32, [4]> var_59413_begin_0 = const()[name = tensor<string, []>("op_59413_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_59413_end_0 = const()[name = tensor<string, []>("op_59413_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 448])];
            tensor<bool, [4]> var_59413_end_mask_0 = const()[name = tensor<string, []>("op_59413_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59413_cast = slice_by_index(begin = var_59413_begin_0, end = var_59413_end_0, end_mask = var_59413_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_59413_cast")];
            tensor<int32, [4]> var_59417_begin_0 = const()[name = tensor<string, []>("op_59417_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_59417_end_0 = const()[name = tensor<string, []>("op_59417_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 512])];
            tensor<bool, [4]> var_59417_end_mask_0 = const()[name = tensor<string, []>("op_59417_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59417_cast = slice_by_index(begin = var_59417_begin_0, end = var_59417_end_0, end_mask = var_59417_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_59417_cast")];
            tensor<int32, [4]> var_59421_begin_0 = const()[name = tensor<string, []>("op_59421_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_59421_end_0 = const()[name = tensor<string, []>("op_59421_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 576])];
            tensor<bool, [4]> var_59421_end_mask_0 = const()[name = tensor<string, []>("op_59421_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59421_cast = slice_by_index(begin = var_59421_begin_0, end = var_59421_end_0, end_mask = var_59421_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_59421_cast")];
            tensor<int32, [4]> var_59425_begin_0 = const()[name = tensor<string, []>("op_59425_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_59425_end_0 = const()[name = tensor<string, []>("op_59425_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 640])];
            tensor<bool, [4]> var_59425_end_mask_0 = const()[name = tensor<string, []>("op_59425_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59425_cast = slice_by_index(begin = var_59425_begin_0, end = var_59425_end_0, end_mask = var_59425_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_59425_cast")];
            tensor<int32, [4]> var_59427_begin_0 = const()[name = tensor<string, []>("op_59427_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59427_end_0 = const()[name = tensor<string, []>("op_59427_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_59427_end_mask_0 = const()[name = tensor<string, []>("op_59427_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59427_cast = slice_by_index(begin = var_59427_begin_0, end = var_59427_end_0, end_mask = var_59427_end_mask_0, x = v_265_cast)[name = tensor<string, []>("op_59427_cast")];
            tensor<int32, [4]> var_59431_begin_0 = const()[name = tensor<string, []>("op_59431_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_59431_end_0 = const()[name = tensor<string, []>("op_59431_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_59431_end_mask_0 = const()[name = tensor<string, []>("op_59431_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59431_cast = slice_by_index(begin = var_59431_begin_0, end = var_59431_end_0, end_mask = var_59431_end_mask_0, x = v_265_cast)[name = tensor<string, []>("op_59431_cast")];
            tensor<int32, [4]> var_59435_begin_0 = const()[name = tensor<string, []>("op_59435_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_59435_end_0 = const()[name = tensor<string, []>("op_59435_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_59435_end_mask_0 = const()[name = tensor<string, []>("op_59435_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59435_cast = slice_by_index(begin = var_59435_begin_0, end = var_59435_end_0, end_mask = var_59435_end_mask_0, x = v_265_cast)[name = tensor<string, []>("op_59435_cast")];
            tensor<int32, [4]> var_59439_begin_0 = const()[name = tensor<string, []>("op_59439_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_59439_end_0 = const()[name = tensor<string, []>("op_59439_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_59439_end_mask_0 = const()[name = tensor<string, []>("op_59439_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59439_cast = slice_by_index(begin = var_59439_begin_0, end = var_59439_end_0, end_mask = var_59439_end_mask_0, x = v_265_cast)[name = tensor<string, []>("op_59439_cast")];
            tensor<int32, [4]> var_59443_begin_0 = const()[name = tensor<string, []>("op_59443_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_59443_end_0 = const()[name = tensor<string, []>("op_59443_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_59443_end_mask_0 = const()[name = tensor<string, []>("op_59443_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59443_cast = slice_by_index(begin = var_59443_begin_0, end = var_59443_end_0, end_mask = var_59443_end_mask_0, x = v_265_cast)[name = tensor<string, []>("op_59443_cast")];
            tensor<int32, [4]> var_59447_begin_0 = const()[name = tensor<string, []>("op_59447_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_59447_end_0 = const()[name = tensor<string, []>("op_59447_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_59447_end_mask_0 = const()[name = tensor<string, []>("op_59447_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59447_cast = slice_by_index(begin = var_59447_begin_0, end = var_59447_end_0, end_mask = var_59447_end_mask_0, x = v_265_cast)[name = tensor<string, []>("op_59447_cast")];
            tensor<int32, [4]> var_59451_begin_0 = const()[name = tensor<string, []>("op_59451_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_59451_end_0 = const()[name = tensor<string, []>("op_59451_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_59451_end_mask_0 = const()[name = tensor<string, []>("op_59451_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59451_cast = slice_by_index(begin = var_59451_begin_0, end = var_59451_end_0, end_mask = var_59451_end_mask_0, x = v_265_cast)[name = tensor<string, []>("op_59451_cast")];
            tensor<int32, [4]> var_59455_begin_0 = const()[name = tensor<string, []>("op_59455_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_59455_end_0 = const()[name = tensor<string, []>("op_59455_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_59455_end_mask_0 = const()[name = tensor<string, []>("op_59455_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59455_cast = slice_by_index(begin = var_59455_begin_0, end = var_59455_end_0, end_mask = var_59455_end_mask_0, x = v_265_cast)[name = tensor<string, []>("op_59455_cast")];
            tensor<int32, [4]> var_59459_begin_0 = const()[name = tensor<string, []>("op_59459_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_59459_end_0 = const()[name = tensor<string, []>("op_59459_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_59459_end_mask_0 = const()[name = tensor<string, []>("op_59459_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59459_cast = slice_by_index(begin = var_59459_begin_0, end = var_59459_end_0, end_mask = var_59459_end_mask_0, x = v_265_cast)[name = tensor<string, []>("op_59459_cast")];
            tensor<int32, [4]> var_59463_begin_0 = const()[name = tensor<string, []>("op_59463_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_59463_end_0 = const()[name = tensor<string, []>("op_59463_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_59463_end_mask_0 = const()[name = tensor<string, []>("op_59463_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59463_cast = slice_by_index(begin = var_59463_begin_0, end = var_59463_end_0, end_mask = var_59463_end_mask_0, x = v_265_cast)[name = tensor<string, []>("op_59463_cast")];
            tensor<string, []> var_59467_equation_0 = const()[name = tensor<string, []>("op_59467_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59467_cast = einsum(equation = var_59467_equation_0, values = (var_59389_cast, var_59346_cast))[name = tensor<string, []>("op_59467_cast")];
            tensor<fp16, []> var_59468_to_fp16 = const()[name = tensor<string, []>("op_59468_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5041_cast = mul(x = var_59467_cast, y = var_59468_to_fp16)[name = tensor<string, []>("aw_5041_cast")];
            tensor<string, []> var_59471_equation_0 = const()[name = tensor<string, []>("op_59471_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59471_cast = einsum(equation = var_59471_equation_0, values = (var_59393_cast, var_59350_cast))[name = tensor<string, []>("op_59471_cast")];
            tensor<fp16, []> var_59472_to_fp16 = const()[name = tensor<string, []>("op_59472_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5043_cast = mul(x = var_59471_cast, y = var_59472_to_fp16)[name = tensor<string, []>("aw_5043_cast")];
            tensor<string, []> var_59475_equation_0 = const()[name = tensor<string, []>("op_59475_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59475_cast = einsum(equation = var_59475_equation_0, values = (var_59397_cast, var_59354_cast))[name = tensor<string, []>("op_59475_cast")];
            tensor<fp16, []> var_59476_to_fp16 = const()[name = tensor<string, []>("op_59476_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5045_cast = mul(x = var_59475_cast, y = var_59476_to_fp16)[name = tensor<string, []>("aw_5045_cast")];
            tensor<string, []> var_59479_equation_0 = const()[name = tensor<string, []>("op_59479_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59479_cast = einsum(equation = var_59479_equation_0, values = (var_59401_cast, var_59358_cast))[name = tensor<string, []>("op_59479_cast")];
            tensor<fp16, []> var_59480_to_fp16 = const()[name = tensor<string, []>("op_59480_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5047_cast = mul(x = var_59479_cast, y = var_59480_to_fp16)[name = tensor<string, []>("aw_5047_cast")];
            tensor<string, []> var_59483_equation_0 = const()[name = tensor<string, []>("op_59483_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59483_cast = einsum(equation = var_59483_equation_0, values = (var_59405_cast, var_59362_cast))[name = tensor<string, []>("op_59483_cast")];
            tensor<fp16, []> var_59484_to_fp16 = const()[name = tensor<string, []>("op_59484_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5049_cast = mul(x = var_59483_cast, y = var_59484_to_fp16)[name = tensor<string, []>("aw_5049_cast")];
            tensor<string, []> var_59487_equation_0 = const()[name = tensor<string, []>("op_59487_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59487_cast = einsum(equation = var_59487_equation_0, values = (var_59409_cast, var_59366_cast))[name = tensor<string, []>("op_59487_cast")];
            tensor<fp16, []> var_59488_to_fp16 = const()[name = tensor<string, []>("op_59488_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5051_cast = mul(x = var_59487_cast, y = var_59488_to_fp16)[name = tensor<string, []>("aw_5051_cast")];
            tensor<string, []> var_59491_equation_0 = const()[name = tensor<string, []>("op_59491_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59491_cast = einsum(equation = var_59491_equation_0, values = (var_59413_cast, var_59370_cast))[name = tensor<string, []>("op_59491_cast")];
            tensor<fp16, []> var_59492_to_fp16 = const()[name = tensor<string, []>("op_59492_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5053_cast = mul(x = var_59491_cast, y = var_59492_to_fp16)[name = tensor<string, []>("aw_5053_cast")];
            tensor<string, []> var_59495_equation_0 = const()[name = tensor<string, []>("op_59495_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59495_cast = einsum(equation = var_59495_equation_0, values = (var_59417_cast, var_59374_cast))[name = tensor<string, []>("op_59495_cast")];
            tensor<fp16, []> var_59496_to_fp16 = const()[name = tensor<string, []>("op_59496_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5055_cast = mul(x = var_59495_cast, y = var_59496_to_fp16)[name = tensor<string, []>("aw_5055_cast")];
            tensor<string, []> var_59499_equation_0 = const()[name = tensor<string, []>("op_59499_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59499_cast = einsum(equation = var_59499_equation_0, values = (var_59421_cast, var_59378_cast))[name = tensor<string, []>("op_59499_cast")];
            tensor<fp16, []> var_59500_to_fp16 = const()[name = tensor<string, []>("op_59500_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5057_cast = mul(x = var_59499_cast, y = var_59500_to_fp16)[name = tensor<string, []>("aw_5057_cast")];
            tensor<string, []> var_59503_equation_0 = const()[name = tensor<string, []>("op_59503_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59503_cast = einsum(equation = var_59503_equation_0, values = (var_59425_cast, var_59382_cast))[name = tensor<string, []>("op_59503_cast")];
            tensor<fp16, []> var_59504_to_fp16 = const()[name = tensor<string, []>("op_59504_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5059_cast = mul(x = var_59503_cast, y = var_59504_to_fp16)[name = tensor<string, []>("aw_5059_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59506_cast = softmax(axis = var_58064, x = aw_5041_cast)[name = tensor<string, []>("op_59506_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59507_cast = softmax(axis = var_58064, x = aw_5043_cast)[name = tensor<string, []>("op_59507_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59508_cast = softmax(axis = var_58064, x = aw_5045_cast)[name = tensor<string, []>("op_59508_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59509_cast = softmax(axis = var_58064, x = aw_5047_cast)[name = tensor<string, []>("op_59509_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59510_cast = softmax(axis = var_58064, x = aw_5049_cast)[name = tensor<string, []>("op_59510_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59511_cast = softmax(axis = var_58064, x = aw_5051_cast)[name = tensor<string, []>("op_59511_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59512_cast = softmax(axis = var_58064, x = aw_5053_cast)[name = tensor<string, []>("op_59512_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59513_cast = softmax(axis = var_58064, x = aw_5055_cast)[name = tensor<string, []>("op_59513_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59514_cast = softmax(axis = var_58064, x = aw_5057_cast)[name = tensor<string, []>("op_59514_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59515_cast = softmax(axis = var_58064, x = aw_5059_cast)[name = tensor<string, []>("op_59515_cast")];
            tensor<string, []> var_59517_equation_0 = const()[name = tensor<string, []>("op_59517_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59517_cast = einsum(equation = var_59517_equation_0, values = (var_59427_cast, var_59506_cast))[name = tensor<string, []>("op_59517_cast")];
            tensor<string, []> var_59519_equation_0 = const()[name = tensor<string, []>("op_59519_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59519_cast = einsum(equation = var_59519_equation_0, values = (var_59431_cast, var_59507_cast))[name = tensor<string, []>("op_59519_cast")];
            tensor<string, []> var_59521_equation_0 = const()[name = tensor<string, []>("op_59521_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59521_cast = einsum(equation = var_59521_equation_0, values = (var_59435_cast, var_59508_cast))[name = tensor<string, []>("op_59521_cast")];
            tensor<string, []> var_59523_equation_0 = const()[name = tensor<string, []>("op_59523_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59523_cast = einsum(equation = var_59523_equation_0, values = (var_59439_cast, var_59509_cast))[name = tensor<string, []>("op_59523_cast")];
            tensor<string, []> var_59525_equation_0 = const()[name = tensor<string, []>("op_59525_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59525_cast = einsum(equation = var_59525_equation_0, values = (var_59443_cast, var_59510_cast))[name = tensor<string, []>("op_59525_cast")];
            tensor<string, []> var_59527_equation_0 = const()[name = tensor<string, []>("op_59527_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59527_cast = einsum(equation = var_59527_equation_0, values = (var_59447_cast, var_59511_cast))[name = tensor<string, []>("op_59527_cast")];
            tensor<string, []> var_59529_equation_0 = const()[name = tensor<string, []>("op_59529_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59529_cast = einsum(equation = var_59529_equation_0, values = (var_59451_cast, var_59512_cast))[name = tensor<string, []>("op_59529_cast")];
            tensor<string, []> var_59531_equation_0 = const()[name = tensor<string, []>("op_59531_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59531_cast = einsum(equation = var_59531_equation_0, values = (var_59455_cast, var_59513_cast))[name = tensor<string, []>("op_59531_cast")];
            tensor<string, []> var_59533_equation_0 = const()[name = tensor<string, []>("op_59533_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59533_cast = einsum(equation = var_59533_equation_0, values = (var_59459_cast, var_59514_cast))[name = tensor<string, []>("op_59533_cast")];
            tensor<string, []> var_59535_equation_0 = const()[name = tensor<string, []>("op_59535_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59535_cast = einsum(equation = var_59535_equation_0, values = (var_59463_cast, var_59515_cast))[name = tensor<string, []>("op_59535_cast")];
            tensor<bool, []> input_773_interleave_0 = const()[name = tensor<string, []>("input_773_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_773_cast = concat(axis = var_58064, interleave = input_773_interleave_0, values = (var_59517_cast, var_59519_cast, var_59521_cast, var_59523_cast, var_59525_cast, var_59527_cast, var_59529_cast, var_59531_cast, var_59533_cast, var_59535_cast))[name = tensor<string, []>("input_773_cast")];
            tensor<int32, [2]> var_59541 = const()[name = tensor<string, []>("op_59541"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59543 = const()[name = tensor<string, []>("op_59543"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_59545_pad_type_0 = const()[name = tensor<string, []>("op_59545_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_59545_pad_0 = const()[name = tensor<string, []>("op_59545_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5003956800)))];
            tensor<fp16, [640]> up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5004776064)))];
            tensor<fp16, [2, 640, 1, 4096]> var_59545_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_59543, groups = var_58064, pad = var_59545_pad_0, pad_type = var_59545_pad_type_0, strides = var_59541, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_773_cast)[name = tensor<string, []>("op_59545_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_399_cast = add(x = var_59545_cast, y = inputs_397_cast)[name = tensor<string, []>("inputs_399_cast")];
            tensor<int32, [1]> var_59549 = const()[name = tensor<string, []>("op_59549"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_399_cast = reduce_mean(axes = var_59549, keep_dims = var_58059, x = inputs_399_cast)[name = tensor<string, []>("channels_mean_399_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_399_cast = sub(x = inputs_399_cast, y = channels_mean_399_cast)[name = tensor<string, []>("zero_mean_399_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_399_cast = mul(x = zero_mean_399_cast, y = zero_mean_399_cast)[name = tensor<string, []>("zero_mean_sq_399_cast")];
            tensor<int32, [1]> var_59553 = const()[name = tensor<string, []>("op_59553"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_59554_cast = reduce_mean(axes = var_59553, keep_dims = var_58059, x = zero_mean_sq_399_cast)[name = tensor<string, []>("op_59554_cast")];
            tensor<fp16, []> var_59555_to_fp16 = const()[name = tensor<string, []>("op_59555_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_59556_cast = add(x = var_59554_cast, y = var_59555_to_fp16)[name = tensor<string, []>("op_59556_cast")];
            tensor<fp16, []> denom_399_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_399_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_399_cast = rsqrt(epsilon = denom_399_epsilon_0_to_fp16, x = var_59556_cast)[name = tensor<string, []>("denom_399_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_399_cast = mul(x = zero_mean_399_cast, y = denom_399_cast)[name = tensor<string, []>("out_399_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_59560_to_fp16 = const()[name = tensor<string, []>("op_59560_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5004777408)))];
            tensor<fp16, [2, 640, 1, 4096]> var_59561_cast = add(x = out_399_cast, y = var_59560_to_fp16)[name = tensor<string, []>("op_59561_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_59563_to_fp16 = const()[name = tensor<string, []>("op_59563_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5004778752)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_537_cast = mul(x = var_59561_cast, y = var_59563_to_fp16)[name = tensor<string, []>("hidden_states_537_cast")];
            tensor<int32, [2]> var_59570 = const()[name = tensor<string, []>("op_59570"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59572 = const()[name = tensor<string, []>("op_59572"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_267_pad_type_0 = const()[name = tensor<string, []>("q_267_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_267_pad_0 = const()[name = tensor<string, []>("q_267_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5004780096)))];
            tensor<fp16, [2, 640, 1, 4096]> q_267_cast = conv(dilations = var_59572, groups = var_58064, pad = q_267_pad_0, pad_type = q_267_pad_type_0, strides = var_59570, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_537_cast)[name = tensor<string, []>("q_267_cast")];
            tensor<int32, [2]> var_59576 = const()[name = tensor<string, []>("op_59576"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59578 = const()[name = tensor<string, []>("op_59578"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_533_pad_type_0 = const()[name = tensor<string, []>("k_533_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_533_pad_0 = const()[name = tensor<string, []>("k_533_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5005599360)))];
            tensor<fp16, [2, 640, 1, 77]> k_533_cast = conv(dilations = var_59578, groups = var_58064, pad = k_533_pad_0, pad_type = k_533_pad_type_0, strides = var_59576, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_533_cast")];
            tensor<int32, [2]> var_59582 = const()[name = tensor<string, []>("op_59582"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59584 = const()[name = tensor<string, []>("op_59584"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_267_pad_type_0 = const()[name = tensor<string, []>("v_267_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_267_pad_0 = const()[name = tensor<string, []>("v_267_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5008220864)))];
            tensor<fp16, [2, 640, 1, 77]> v_267_cast = conv(dilations = var_59584, groups = var_58064, pad = v_267_pad_0, pad_type = v_267_pad_type_0, strides = var_59582, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_267_cast")];
            tensor<int32, [4]> var_59588_begin_0 = const()[name = tensor<string, []>("op_59588_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59588_end_0 = const()[name = tensor<string, []>("op_59588_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_59588_end_mask_0 = const()[name = tensor<string, []>("op_59588_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59588_cast = slice_by_index(begin = var_59588_begin_0, end = var_59588_end_0, end_mask = var_59588_end_mask_0, x = q_267_cast)[name = tensor<string, []>("op_59588_cast")];
            tensor<int32, [4]> var_59592_begin_0 = const()[name = tensor<string, []>("op_59592_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_59592_end_0 = const()[name = tensor<string, []>("op_59592_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_59592_end_mask_0 = const()[name = tensor<string, []>("op_59592_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59592_cast = slice_by_index(begin = var_59592_begin_0, end = var_59592_end_0, end_mask = var_59592_end_mask_0, x = q_267_cast)[name = tensor<string, []>("op_59592_cast")];
            tensor<int32, [4]> var_59596_begin_0 = const()[name = tensor<string, []>("op_59596_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_59596_end_0 = const()[name = tensor<string, []>("op_59596_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_59596_end_mask_0 = const()[name = tensor<string, []>("op_59596_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59596_cast = slice_by_index(begin = var_59596_begin_0, end = var_59596_end_0, end_mask = var_59596_end_mask_0, x = q_267_cast)[name = tensor<string, []>("op_59596_cast")];
            tensor<int32, [4]> var_59600_begin_0 = const()[name = tensor<string, []>("op_59600_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_59600_end_0 = const()[name = tensor<string, []>("op_59600_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_59600_end_mask_0 = const()[name = tensor<string, []>("op_59600_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59600_cast = slice_by_index(begin = var_59600_begin_0, end = var_59600_end_0, end_mask = var_59600_end_mask_0, x = q_267_cast)[name = tensor<string, []>("op_59600_cast")];
            tensor<int32, [4]> var_59604_begin_0 = const()[name = tensor<string, []>("op_59604_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_59604_end_0 = const()[name = tensor<string, []>("op_59604_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_59604_end_mask_0 = const()[name = tensor<string, []>("op_59604_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59604_cast = slice_by_index(begin = var_59604_begin_0, end = var_59604_end_0, end_mask = var_59604_end_mask_0, x = q_267_cast)[name = tensor<string, []>("op_59604_cast")];
            tensor<int32, [4]> var_59608_begin_0 = const()[name = tensor<string, []>("op_59608_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_59608_end_0 = const()[name = tensor<string, []>("op_59608_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_59608_end_mask_0 = const()[name = tensor<string, []>("op_59608_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59608_cast = slice_by_index(begin = var_59608_begin_0, end = var_59608_end_0, end_mask = var_59608_end_mask_0, x = q_267_cast)[name = tensor<string, []>("op_59608_cast")];
            tensor<int32, [4]> var_59612_begin_0 = const()[name = tensor<string, []>("op_59612_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_59612_end_0 = const()[name = tensor<string, []>("op_59612_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_59612_end_mask_0 = const()[name = tensor<string, []>("op_59612_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59612_cast = slice_by_index(begin = var_59612_begin_0, end = var_59612_end_0, end_mask = var_59612_end_mask_0, x = q_267_cast)[name = tensor<string, []>("op_59612_cast")];
            tensor<int32, [4]> var_59616_begin_0 = const()[name = tensor<string, []>("op_59616_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_59616_end_0 = const()[name = tensor<string, []>("op_59616_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_59616_end_mask_0 = const()[name = tensor<string, []>("op_59616_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59616_cast = slice_by_index(begin = var_59616_begin_0, end = var_59616_end_0, end_mask = var_59616_end_mask_0, x = q_267_cast)[name = tensor<string, []>("op_59616_cast")];
            tensor<int32, [4]> var_59620_begin_0 = const()[name = tensor<string, []>("op_59620_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_59620_end_0 = const()[name = tensor<string, []>("op_59620_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_59620_end_mask_0 = const()[name = tensor<string, []>("op_59620_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59620_cast = slice_by_index(begin = var_59620_begin_0, end = var_59620_end_0, end_mask = var_59620_end_mask_0, x = q_267_cast)[name = tensor<string, []>("op_59620_cast")];
            tensor<int32, [4]> var_59624_begin_0 = const()[name = tensor<string, []>("op_59624_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_59624_end_0 = const()[name = tensor<string, []>("op_59624_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_59624_end_mask_0 = const()[name = tensor<string, []>("op_59624_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59624_cast = slice_by_index(begin = var_59624_begin_0, end = var_59624_end_0, end_mask = var_59624_end_mask_0, x = q_267_cast)[name = tensor<string, []>("op_59624_cast")];
            tensor<int32, [4]> k_535_perm_0 = const()[name = tensor<string, []>("k_535_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_59631_begin_0 = const()[name = tensor<string, []>("op_59631_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59631_end_0 = const()[name = tensor<string, []>("op_59631_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_59631_end_mask_0 = const()[name = tensor<string, []>("op_59631_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> transpose_6 = transpose(perm = k_535_perm_0, x = k_533_cast)[name = tensor<string, []>("transpose_6")];
            tensor<fp16, [2, 77, 1, 64]> var_59631_cast = slice_by_index(begin = var_59631_begin_0, end = var_59631_end_0, end_mask = var_59631_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_59631_cast")];
            tensor<int32, [4]> var_59635_begin_0 = const()[name = tensor<string, []>("op_59635_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_59635_end_0 = const()[name = tensor<string, []>("op_59635_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_59635_end_mask_0 = const()[name = tensor<string, []>("op_59635_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59635_cast = slice_by_index(begin = var_59635_begin_0, end = var_59635_end_0, end_mask = var_59635_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_59635_cast")];
            tensor<int32, [4]> var_59639_begin_0 = const()[name = tensor<string, []>("op_59639_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_59639_end_0 = const()[name = tensor<string, []>("op_59639_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_59639_end_mask_0 = const()[name = tensor<string, []>("op_59639_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59639_cast = slice_by_index(begin = var_59639_begin_0, end = var_59639_end_0, end_mask = var_59639_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_59639_cast")];
            tensor<int32, [4]> var_59643_begin_0 = const()[name = tensor<string, []>("op_59643_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_59643_end_0 = const()[name = tensor<string, []>("op_59643_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_59643_end_mask_0 = const()[name = tensor<string, []>("op_59643_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59643_cast = slice_by_index(begin = var_59643_begin_0, end = var_59643_end_0, end_mask = var_59643_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_59643_cast")];
            tensor<int32, [4]> var_59647_begin_0 = const()[name = tensor<string, []>("op_59647_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_59647_end_0 = const()[name = tensor<string, []>("op_59647_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_59647_end_mask_0 = const()[name = tensor<string, []>("op_59647_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59647_cast = slice_by_index(begin = var_59647_begin_0, end = var_59647_end_0, end_mask = var_59647_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_59647_cast")];
            tensor<int32, [4]> var_59651_begin_0 = const()[name = tensor<string, []>("op_59651_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_59651_end_0 = const()[name = tensor<string, []>("op_59651_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_59651_end_mask_0 = const()[name = tensor<string, []>("op_59651_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59651_cast = slice_by_index(begin = var_59651_begin_0, end = var_59651_end_0, end_mask = var_59651_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_59651_cast")];
            tensor<int32, [4]> var_59655_begin_0 = const()[name = tensor<string, []>("op_59655_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_59655_end_0 = const()[name = tensor<string, []>("op_59655_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_59655_end_mask_0 = const()[name = tensor<string, []>("op_59655_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59655_cast = slice_by_index(begin = var_59655_begin_0, end = var_59655_end_0, end_mask = var_59655_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_59655_cast")];
            tensor<int32, [4]> var_59659_begin_0 = const()[name = tensor<string, []>("op_59659_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_59659_end_0 = const()[name = tensor<string, []>("op_59659_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_59659_end_mask_0 = const()[name = tensor<string, []>("op_59659_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59659_cast = slice_by_index(begin = var_59659_begin_0, end = var_59659_end_0, end_mask = var_59659_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_59659_cast")];
            tensor<int32, [4]> var_59663_begin_0 = const()[name = tensor<string, []>("op_59663_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_59663_end_0 = const()[name = tensor<string, []>("op_59663_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_59663_end_mask_0 = const()[name = tensor<string, []>("op_59663_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59663_cast = slice_by_index(begin = var_59663_begin_0, end = var_59663_end_0, end_mask = var_59663_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_59663_cast")];
            tensor<int32, [4]> var_59667_begin_0 = const()[name = tensor<string, []>("op_59667_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_59667_end_0 = const()[name = tensor<string, []>("op_59667_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_59667_end_mask_0 = const()[name = tensor<string, []>("op_59667_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_59667_cast = slice_by_index(begin = var_59667_begin_0, end = var_59667_end_0, end_mask = var_59667_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_59667_cast")];
            tensor<int32, [4]> var_59669_begin_0 = const()[name = tensor<string, []>("op_59669_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59669_end_0 = const()[name = tensor<string, []>("op_59669_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_59669_end_mask_0 = const()[name = tensor<string, []>("op_59669_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59669_cast = slice_by_index(begin = var_59669_begin_0, end = var_59669_end_0, end_mask = var_59669_end_mask_0, x = v_267_cast)[name = tensor<string, []>("op_59669_cast")];
            tensor<int32, [4]> var_59673_begin_0 = const()[name = tensor<string, []>("op_59673_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_59673_end_0 = const()[name = tensor<string, []>("op_59673_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_59673_end_mask_0 = const()[name = tensor<string, []>("op_59673_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59673_cast = slice_by_index(begin = var_59673_begin_0, end = var_59673_end_0, end_mask = var_59673_end_mask_0, x = v_267_cast)[name = tensor<string, []>("op_59673_cast")];
            tensor<int32, [4]> var_59677_begin_0 = const()[name = tensor<string, []>("op_59677_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_59677_end_0 = const()[name = tensor<string, []>("op_59677_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_59677_end_mask_0 = const()[name = tensor<string, []>("op_59677_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59677_cast = slice_by_index(begin = var_59677_begin_0, end = var_59677_end_0, end_mask = var_59677_end_mask_0, x = v_267_cast)[name = tensor<string, []>("op_59677_cast")];
            tensor<int32, [4]> var_59681_begin_0 = const()[name = tensor<string, []>("op_59681_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_59681_end_0 = const()[name = tensor<string, []>("op_59681_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_59681_end_mask_0 = const()[name = tensor<string, []>("op_59681_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59681_cast = slice_by_index(begin = var_59681_begin_0, end = var_59681_end_0, end_mask = var_59681_end_mask_0, x = v_267_cast)[name = tensor<string, []>("op_59681_cast")];
            tensor<int32, [4]> var_59685_begin_0 = const()[name = tensor<string, []>("op_59685_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_59685_end_0 = const()[name = tensor<string, []>("op_59685_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_59685_end_mask_0 = const()[name = tensor<string, []>("op_59685_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59685_cast = slice_by_index(begin = var_59685_begin_0, end = var_59685_end_0, end_mask = var_59685_end_mask_0, x = v_267_cast)[name = tensor<string, []>("op_59685_cast")];
            tensor<int32, [4]> var_59689_begin_0 = const()[name = tensor<string, []>("op_59689_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_59689_end_0 = const()[name = tensor<string, []>("op_59689_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_59689_end_mask_0 = const()[name = tensor<string, []>("op_59689_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59689_cast = slice_by_index(begin = var_59689_begin_0, end = var_59689_end_0, end_mask = var_59689_end_mask_0, x = v_267_cast)[name = tensor<string, []>("op_59689_cast")];
            tensor<int32, [4]> var_59693_begin_0 = const()[name = tensor<string, []>("op_59693_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_59693_end_0 = const()[name = tensor<string, []>("op_59693_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_59693_end_mask_0 = const()[name = tensor<string, []>("op_59693_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59693_cast = slice_by_index(begin = var_59693_begin_0, end = var_59693_end_0, end_mask = var_59693_end_mask_0, x = v_267_cast)[name = tensor<string, []>("op_59693_cast")];
            tensor<int32, [4]> var_59697_begin_0 = const()[name = tensor<string, []>("op_59697_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_59697_end_0 = const()[name = tensor<string, []>("op_59697_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_59697_end_mask_0 = const()[name = tensor<string, []>("op_59697_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59697_cast = slice_by_index(begin = var_59697_begin_0, end = var_59697_end_0, end_mask = var_59697_end_mask_0, x = v_267_cast)[name = tensor<string, []>("op_59697_cast")];
            tensor<int32, [4]> var_59701_begin_0 = const()[name = tensor<string, []>("op_59701_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_59701_end_0 = const()[name = tensor<string, []>("op_59701_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_59701_end_mask_0 = const()[name = tensor<string, []>("op_59701_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59701_cast = slice_by_index(begin = var_59701_begin_0, end = var_59701_end_0, end_mask = var_59701_end_mask_0, x = v_267_cast)[name = tensor<string, []>("op_59701_cast")];
            tensor<int32, [4]> var_59705_begin_0 = const()[name = tensor<string, []>("op_59705_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_59705_end_0 = const()[name = tensor<string, []>("op_59705_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_59705_end_mask_0 = const()[name = tensor<string, []>("op_59705_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_59705_cast = slice_by_index(begin = var_59705_begin_0, end = var_59705_end_0, end_mask = var_59705_end_mask_0, x = v_267_cast)[name = tensor<string, []>("op_59705_cast")];
            tensor<string, []> var_59709_equation_0 = const()[name = tensor<string, []>("op_59709_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59709_cast = einsum(equation = var_59709_equation_0, values = (var_59631_cast, var_59588_cast))[name = tensor<string, []>("op_59709_cast")];
            tensor<fp16, []> var_59710_to_fp16 = const()[name = tensor<string, []>("op_59710_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5061_cast = mul(x = var_59709_cast, y = var_59710_to_fp16)[name = tensor<string, []>("aw_5061_cast")];
            tensor<string, []> var_59713_equation_0 = const()[name = tensor<string, []>("op_59713_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59713_cast = einsum(equation = var_59713_equation_0, values = (var_59635_cast, var_59592_cast))[name = tensor<string, []>("op_59713_cast")];
            tensor<fp16, []> var_59714_to_fp16 = const()[name = tensor<string, []>("op_59714_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5063_cast = mul(x = var_59713_cast, y = var_59714_to_fp16)[name = tensor<string, []>("aw_5063_cast")];
            tensor<string, []> var_59717_equation_0 = const()[name = tensor<string, []>("op_59717_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59717_cast = einsum(equation = var_59717_equation_0, values = (var_59639_cast, var_59596_cast))[name = tensor<string, []>("op_59717_cast")];
            tensor<fp16, []> var_59718_to_fp16 = const()[name = tensor<string, []>("op_59718_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5065_cast = mul(x = var_59717_cast, y = var_59718_to_fp16)[name = tensor<string, []>("aw_5065_cast")];
            tensor<string, []> var_59721_equation_0 = const()[name = tensor<string, []>("op_59721_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59721_cast = einsum(equation = var_59721_equation_0, values = (var_59643_cast, var_59600_cast))[name = tensor<string, []>("op_59721_cast")];
            tensor<fp16, []> var_59722_to_fp16 = const()[name = tensor<string, []>("op_59722_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5067_cast = mul(x = var_59721_cast, y = var_59722_to_fp16)[name = tensor<string, []>("aw_5067_cast")];
            tensor<string, []> var_59725_equation_0 = const()[name = tensor<string, []>("op_59725_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59725_cast = einsum(equation = var_59725_equation_0, values = (var_59647_cast, var_59604_cast))[name = tensor<string, []>("op_59725_cast")];
            tensor<fp16, []> var_59726_to_fp16 = const()[name = tensor<string, []>("op_59726_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5069_cast = mul(x = var_59725_cast, y = var_59726_to_fp16)[name = tensor<string, []>("aw_5069_cast")];
            tensor<string, []> var_59729_equation_0 = const()[name = tensor<string, []>("op_59729_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59729_cast = einsum(equation = var_59729_equation_0, values = (var_59651_cast, var_59608_cast))[name = tensor<string, []>("op_59729_cast")];
            tensor<fp16, []> var_59730_to_fp16 = const()[name = tensor<string, []>("op_59730_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5071_cast = mul(x = var_59729_cast, y = var_59730_to_fp16)[name = tensor<string, []>("aw_5071_cast")];
            tensor<string, []> var_59733_equation_0 = const()[name = tensor<string, []>("op_59733_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59733_cast = einsum(equation = var_59733_equation_0, values = (var_59655_cast, var_59612_cast))[name = tensor<string, []>("op_59733_cast")];
            tensor<fp16, []> var_59734_to_fp16 = const()[name = tensor<string, []>("op_59734_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5073_cast = mul(x = var_59733_cast, y = var_59734_to_fp16)[name = tensor<string, []>("aw_5073_cast")];
            tensor<string, []> var_59737_equation_0 = const()[name = tensor<string, []>("op_59737_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59737_cast = einsum(equation = var_59737_equation_0, values = (var_59659_cast, var_59616_cast))[name = tensor<string, []>("op_59737_cast")];
            tensor<fp16, []> var_59738_to_fp16 = const()[name = tensor<string, []>("op_59738_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5075_cast = mul(x = var_59737_cast, y = var_59738_to_fp16)[name = tensor<string, []>("aw_5075_cast")];
            tensor<string, []> var_59741_equation_0 = const()[name = tensor<string, []>("op_59741_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59741_cast = einsum(equation = var_59741_equation_0, values = (var_59663_cast, var_59620_cast))[name = tensor<string, []>("op_59741_cast")];
            tensor<fp16, []> var_59742_to_fp16 = const()[name = tensor<string, []>("op_59742_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5077_cast = mul(x = var_59741_cast, y = var_59742_to_fp16)[name = tensor<string, []>("aw_5077_cast")];
            tensor<string, []> var_59745_equation_0 = const()[name = tensor<string, []>("op_59745_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_59745_cast = einsum(equation = var_59745_equation_0, values = (var_59667_cast, var_59624_cast))[name = tensor<string, []>("op_59745_cast")];
            tensor<fp16, []> var_59746_to_fp16 = const()[name = tensor<string, []>("op_59746_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5079_cast = mul(x = var_59745_cast, y = var_59746_to_fp16)[name = tensor<string, []>("aw_5079_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59748_cast = softmax(axis = var_58064, x = aw_5061_cast)[name = tensor<string, []>("op_59748_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59749_cast = softmax(axis = var_58064, x = aw_5063_cast)[name = tensor<string, []>("op_59749_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59750_cast = softmax(axis = var_58064, x = aw_5065_cast)[name = tensor<string, []>("op_59750_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59751_cast = softmax(axis = var_58064, x = aw_5067_cast)[name = tensor<string, []>("op_59751_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59752_cast = softmax(axis = var_58064, x = aw_5069_cast)[name = tensor<string, []>("op_59752_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59753_cast = softmax(axis = var_58064, x = aw_5071_cast)[name = tensor<string, []>("op_59753_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59754_cast = softmax(axis = var_58064, x = aw_5073_cast)[name = tensor<string, []>("op_59754_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59755_cast = softmax(axis = var_58064, x = aw_5075_cast)[name = tensor<string, []>("op_59755_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59756_cast = softmax(axis = var_58064, x = aw_5077_cast)[name = tensor<string, []>("op_59756_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_59757_cast = softmax(axis = var_58064, x = aw_5079_cast)[name = tensor<string, []>("op_59757_cast")];
            tensor<string, []> var_59759_equation_0 = const()[name = tensor<string, []>("op_59759_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59759_cast = einsum(equation = var_59759_equation_0, values = (var_59669_cast, var_59748_cast))[name = tensor<string, []>("op_59759_cast")];
            tensor<string, []> var_59761_equation_0 = const()[name = tensor<string, []>("op_59761_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59761_cast = einsum(equation = var_59761_equation_0, values = (var_59673_cast, var_59749_cast))[name = tensor<string, []>("op_59761_cast")];
            tensor<string, []> var_59763_equation_0 = const()[name = tensor<string, []>("op_59763_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59763_cast = einsum(equation = var_59763_equation_0, values = (var_59677_cast, var_59750_cast))[name = tensor<string, []>("op_59763_cast")];
            tensor<string, []> var_59765_equation_0 = const()[name = tensor<string, []>("op_59765_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59765_cast = einsum(equation = var_59765_equation_0, values = (var_59681_cast, var_59751_cast))[name = tensor<string, []>("op_59765_cast")];
            tensor<string, []> var_59767_equation_0 = const()[name = tensor<string, []>("op_59767_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59767_cast = einsum(equation = var_59767_equation_0, values = (var_59685_cast, var_59752_cast))[name = tensor<string, []>("op_59767_cast")];
            tensor<string, []> var_59769_equation_0 = const()[name = tensor<string, []>("op_59769_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59769_cast = einsum(equation = var_59769_equation_0, values = (var_59689_cast, var_59753_cast))[name = tensor<string, []>("op_59769_cast")];
            tensor<string, []> var_59771_equation_0 = const()[name = tensor<string, []>("op_59771_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59771_cast = einsum(equation = var_59771_equation_0, values = (var_59693_cast, var_59754_cast))[name = tensor<string, []>("op_59771_cast")];
            tensor<string, []> var_59773_equation_0 = const()[name = tensor<string, []>("op_59773_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59773_cast = einsum(equation = var_59773_equation_0, values = (var_59697_cast, var_59755_cast))[name = tensor<string, []>("op_59773_cast")];
            tensor<string, []> var_59775_equation_0 = const()[name = tensor<string, []>("op_59775_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59775_cast = einsum(equation = var_59775_equation_0, values = (var_59701_cast, var_59756_cast))[name = tensor<string, []>("op_59775_cast")];
            tensor<string, []> var_59777_equation_0 = const()[name = tensor<string, []>("op_59777_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_59777_cast = einsum(equation = var_59777_equation_0, values = (var_59705_cast, var_59757_cast))[name = tensor<string, []>("op_59777_cast")];
            tensor<bool, []> input_775_interleave_0 = const()[name = tensor<string, []>("input_775_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_775_cast = concat(axis = var_58064, interleave = input_775_interleave_0, values = (var_59759_cast, var_59761_cast, var_59763_cast, var_59765_cast, var_59767_cast, var_59769_cast, var_59771_cast, var_59773_cast, var_59775_cast, var_59777_cast))[name = tensor<string, []>("input_775_cast")];
            tensor<int32, [2]> var_59783 = const()[name = tensor<string, []>("op_59783"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59785 = const()[name = tensor<string, []>("op_59785"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_59787_pad_type_0 = const()[name = tensor<string, []>("op_59787_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_59787_pad_0 = const()[name = tensor<string, []>("op_59787_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5010842368)))];
            tensor<fp16, [640]> up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5011661632)))];
            tensor<fp16, [2, 640, 1, 4096]> var_59787_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_59785, groups = var_58064, pad = var_59787_pad_0, pad_type = var_59787_pad_type_0, strides = var_59783, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_775_cast)[name = tensor<string, []>("op_59787_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_401_cast = add(x = var_59787_cast, y = inputs_399_cast)[name = tensor<string, []>("inputs_401_cast")];
            tensor<int32, [1]> var_59791 = const()[name = tensor<string, []>("op_59791"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_401_cast = reduce_mean(axes = var_59791, keep_dims = var_58059, x = inputs_401_cast)[name = tensor<string, []>("channels_mean_401_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_401_cast = sub(x = inputs_401_cast, y = channels_mean_401_cast)[name = tensor<string, []>("zero_mean_401_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_401_cast = mul(x = zero_mean_401_cast, y = zero_mean_401_cast)[name = tensor<string, []>("zero_mean_sq_401_cast")];
            tensor<int32, [1]> var_59795 = const()[name = tensor<string, []>("op_59795"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_59796_cast = reduce_mean(axes = var_59795, keep_dims = var_58059, x = zero_mean_sq_401_cast)[name = tensor<string, []>("op_59796_cast")];
            tensor<fp16, []> var_59797_to_fp16 = const()[name = tensor<string, []>("op_59797_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_59798_cast = add(x = var_59796_cast, y = var_59797_to_fp16)[name = tensor<string, []>("op_59798_cast")];
            tensor<fp16, []> denom_401_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_401_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_401_cast = rsqrt(epsilon = denom_401_epsilon_0_to_fp16, x = var_59798_cast)[name = tensor<string, []>("denom_401_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_401_cast = mul(x = zero_mean_401_cast, y = denom_401_cast)[name = tensor<string, []>("out_401_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_59802_to_fp16 = const()[name = tensor<string, []>("op_59802_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5011662976)))];
            tensor<fp16, [2, 640, 1, 4096]> var_59803_cast = add(x = out_401_cast, y = var_59802_to_fp16)[name = tensor<string, []>("op_59803_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_59805_to_fp16 = const()[name = tensor<string, []>("op_59805_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5011664320)))];
            tensor<fp16, [2, 640, 1, 4096]> input_777_cast = mul(x = var_59803_cast, y = var_59805_to_fp16)[name = tensor<string, []>("input_777_cast")];
            tensor<int32, [2]> var_59813 = const()[name = tensor<string, []>("op_59813"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59815 = const()[name = tensor<string, []>("op_59815"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_59817_pad_type_0 = const()[name = tensor<string, []>("op_59817_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_59817_pad_0 = const()[name = tensor<string, []>("op_59817_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [5120, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [5120, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5011665664)))];
            tensor<fp16, [5120]> up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5018219328)))];
            tensor<fp16, [2, 5120, 1, 4096]> var_59817_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_59815, groups = var_58064, pad = var_59817_pad_0, pad_type = var_59817_pad_type_0, strides = var_59813, weight = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_777_cast)[name = tensor<string, []>("op_59817_cast")];
            tensor<int32, [2]> var_59818_split_sizes_0 = const()[name = tensor<string, []>("op_59818_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_59818_axis_0 = const()[name = tensor<string, []>("op_59818_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 4096]> var_59818_cast_0, tensor<fp16, [2, 2560, 1, 4096]> var_59818_cast_1 = split(axis = var_59818_axis_0, split_sizes = var_59818_split_sizes_0, x = var_59817_cast)[name = tensor<string, []>("op_59818_cast")];
            tensor<string, []> var_59820_mode_0 = const()[name = tensor<string, []>("op_59820_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 4096]> var_59820_cast = gelu(mode = var_59820_mode_0, x = var_59818_cast_1)[name = tensor<string, []>("op_59820_cast")];
            tensor<fp16, [2, 2560, 1, 4096]> input_779_cast = mul(x = var_59818_cast_0, y = var_59820_cast)[name = tensor<string, []>("input_779_cast")];
            tensor<int32, [2]> var_59824 = const()[name = tensor<string, []>("op_59824"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59826 = const()[name = tensor<string, []>("op_59826"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_59828_pad_type_0 = const()[name = tensor<string, []>("op_59828_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_59828_pad_0 = const()[name = tensor<string, []>("op_59828_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2560, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [640, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5018229632)))];
            tensor<fp16, [640]> up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5021506496)))];
            tensor<fp16, [2, 640, 1, 4096]> var_59828_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_59826, groups = var_58064, pad = var_59828_pad_0, pad_type = var_59828_pad_type_0, strides = var_59824, weight = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_779_cast)[name = tensor<string, []>("op_59828_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_403_cast = add(x = var_59828_cast, y = inputs_401_cast)[name = tensor<string, []>("inputs_403_cast")];
            tensor<int32, [1]> var_59838 = const()[name = tensor<string, []>("op_59838"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_403_cast = reduce_mean(axes = var_59838, keep_dims = var_58059, x = inputs_403_cast)[name = tensor<string, []>("channels_mean_403_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_403_cast = sub(x = inputs_403_cast, y = channels_mean_403_cast)[name = tensor<string, []>("zero_mean_403_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_403_cast = mul(x = zero_mean_403_cast, y = zero_mean_403_cast)[name = tensor<string, []>("zero_mean_sq_403_cast")];
            tensor<int32, [1]> var_59842 = const()[name = tensor<string, []>("op_59842"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_59843_cast = reduce_mean(axes = var_59842, keep_dims = var_58059, x = zero_mean_sq_403_cast)[name = tensor<string, []>("op_59843_cast")];
            tensor<fp16, []> var_59844_to_fp16 = const()[name = tensor<string, []>("op_59844_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_59845_cast = add(x = var_59843_cast, y = var_59844_to_fp16)[name = tensor<string, []>("op_59845_cast")];
            tensor<fp16, []> denom_403_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_403_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_403_cast = rsqrt(epsilon = denom_403_epsilon_0_to_fp16, x = var_59845_cast)[name = tensor<string, []>("denom_403_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_403_cast = mul(x = zero_mean_403_cast, y = denom_403_cast)[name = tensor<string, []>("out_403_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_59849_to_fp16 = const()[name = tensor<string, []>("op_59849_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5021507840)))];
            tensor<fp16, [2, 640, 1, 4096]> var_59850_cast = add(x = out_403_cast, y = var_59849_to_fp16)[name = tensor<string, []>("op_59850_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_59852_to_fp16 = const()[name = tensor<string, []>("op_59852_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5021509184)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_541_cast = mul(x = var_59850_cast, y = var_59852_to_fp16)[name = tensor<string, []>("hidden_states_541_cast")];
            tensor<int32, [2]> var_59859 = const()[name = tensor<string, []>("op_59859"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59861 = const()[name = tensor<string, []>("op_59861"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_269_pad_type_0 = const()[name = tensor<string, []>("q_269_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_269_pad_0 = const()[name = tensor<string, []>("q_269_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5021510528)))];
            tensor<fp16, [2, 640, 1, 4096]> q_269_cast = conv(dilations = var_59861, groups = var_58064, pad = q_269_pad_0, pad_type = q_269_pad_type_0, strides = var_59859, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_541_cast)[name = tensor<string, []>("q_269_cast")];
            tensor<int32, [2]> var_59865 = const()[name = tensor<string, []>("op_59865"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59867 = const()[name = tensor<string, []>("op_59867"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_537_pad_type_0 = const()[name = tensor<string, []>("k_537_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_537_pad_0 = const()[name = tensor<string, []>("k_537_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5022329792)))];
            tensor<fp16, [2, 640, 1, 4096]> k_537_cast = conv(dilations = var_59867, groups = var_58064, pad = k_537_pad_0, pad_type = k_537_pad_type_0, strides = var_59865, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_541_cast)[name = tensor<string, []>("k_537_cast")];
            tensor<int32, [2]> var_59871 = const()[name = tensor<string, []>("op_59871"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_59873 = const()[name = tensor<string, []>("op_59873"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_269_pad_type_0 = const()[name = tensor<string, []>("v_269_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_269_pad_0 = const()[name = tensor<string, []>("v_269_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5023149056)))];
            tensor<fp16, [2, 640, 1, 4096]> v_269_cast = conv(dilations = var_59873, groups = var_58064, pad = v_269_pad_0, pad_type = v_269_pad_type_0, strides = var_59871, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_541_cast)[name = tensor<string, []>("v_269_cast")];
            tensor<int32, [4]> var_59877_begin_0 = const()[name = tensor<string, []>("op_59877_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59877_end_0 = const()[name = tensor<string, []>("op_59877_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_59877_end_mask_0 = const()[name = tensor<string, []>("op_59877_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59877_cast = slice_by_index(begin = var_59877_begin_0, end = var_59877_end_0, end_mask = var_59877_end_mask_0, x = q_269_cast)[name = tensor<string, []>("op_59877_cast")];
            tensor<int32, [4]> var_59881_begin_0 = const()[name = tensor<string, []>("op_59881_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_59881_end_0 = const()[name = tensor<string, []>("op_59881_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_59881_end_mask_0 = const()[name = tensor<string, []>("op_59881_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59881_cast = slice_by_index(begin = var_59881_begin_0, end = var_59881_end_0, end_mask = var_59881_end_mask_0, x = q_269_cast)[name = tensor<string, []>("op_59881_cast")];
            tensor<int32, [4]> var_59885_begin_0 = const()[name = tensor<string, []>("op_59885_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_59885_end_0 = const()[name = tensor<string, []>("op_59885_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_59885_end_mask_0 = const()[name = tensor<string, []>("op_59885_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59885_cast = slice_by_index(begin = var_59885_begin_0, end = var_59885_end_0, end_mask = var_59885_end_mask_0, x = q_269_cast)[name = tensor<string, []>("op_59885_cast")];
            tensor<int32, [4]> var_59889_begin_0 = const()[name = tensor<string, []>("op_59889_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_59889_end_0 = const()[name = tensor<string, []>("op_59889_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_59889_end_mask_0 = const()[name = tensor<string, []>("op_59889_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59889_cast = slice_by_index(begin = var_59889_begin_0, end = var_59889_end_0, end_mask = var_59889_end_mask_0, x = q_269_cast)[name = tensor<string, []>("op_59889_cast")];
            tensor<int32, [4]> var_59893_begin_0 = const()[name = tensor<string, []>("op_59893_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_59893_end_0 = const()[name = tensor<string, []>("op_59893_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_59893_end_mask_0 = const()[name = tensor<string, []>("op_59893_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59893_cast = slice_by_index(begin = var_59893_begin_0, end = var_59893_end_0, end_mask = var_59893_end_mask_0, x = q_269_cast)[name = tensor<string, []>("op_59893_cast")];
            tensor<int32, [4]> var_59897_begin_0 = const()[name = tensor<string, []>("op_59897_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_59897_end_0 = const()[name = tensor<string, []>("op_59897_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_59897_end_mask_0 = const()[name = tensor<string, []>("op_59897_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59897_cast = slice_by_index(begin = var_59897_begin_0, end = var_59897_end_0, end_mask = var_59897_end_mask_0, x = q_269_cast)[name = tensor<string, []>("op_59897_cast")];
            tensor<int32, [4]> var_59901_begin_0 = const()[name = tensor<string, []>("op_59901_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_59901_end_0 = const()[name = tensor<string, []>("op_59901_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_59901_end_mask_0 = const()[name = tensor<string, []>("op_59901_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59901_cast = slice_by_index(begin = var_59901_begin_0, end = var_59901_end_0, end_mask = var_59901_end_mask_0, x = q_269_cast)[name = tensor<string, []>("op_59901_cast")];
            tensor<int32, [4]> var_59905_begin_0 = const()[name = tensor<string, []>("op_59905_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_59905_end_0 = const()[name = tensor<string, []>("op_59905_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_59905_end_mask_0 = const()[name = tensor<string, []>("op_59905_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59905_cast = slice_by_index(begin = var_59905_begin_0, end = var_59905_end_0, end_mask = var_59905_end_mask_0, x = q_269_cast)[name = tensor<string, []>("op_59905_cast")];
            tensor<int32, [4]> var_59909_begin_0 = const()[name = tensor<string, []>("op_59909_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_59909_end_0 = const()[name = tensor<string, []>("op_59909_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_59909_end_mask_0 = const()[name = tensor<string, []>("op_59909_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59909_cast = slice_by_index(begin = var_59909_begin_0, end = var_59909_end_0, end_mask = var_59909_end_mask_0, x = q_269_cast)[name = tensor<string, []>("op_59909_cast")];
            tensor<int32, [4]> var_59913_begin_0 = const()[name = tensor<string, []>("op_59913_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_59913_end_0 = const()[name = tensor<string, []>("op_59913_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_59913_end_mask_0 = const()[name = tensor<string, []>("op_59913_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59913_cast = slice_by_index(begin = var_59913_begin_0, end = var_59913_end_0, end_mask = var_59913_end_mask_0, x = q_269_cast)[name = tensor<string, []>("op_59913_cast")];
            tensor<int32, [4]> k_539_perm_0 = const()[name = tensor<string, []>("k_539_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_59920_begin_0 = const()[name = tensor<string, []>("op_59920_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59920_end_0 = const()[name = tensor<string, []>("op_59920_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_59920_end_mask_0 = const()[name = tensor<string, []>("op_59920_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 640]> transpose_5 = transpose(perm = k_539_perm_0, x = k_537_cast)[name = tensor<string, []>("transpose_5")];
            tensor<fp16, [2, 4096, 1, 64]> var_59920_cast = slice_by_index(begin = var_59920_begin_0, end = var_59920_end_0, end_mask = var_59920_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_59920_cast")];
            tensor<int32, [4]> var_59924_begin_0 = const()[name = tensor<string, []>("op_59924_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_59924_end_0 = const()[name = tensor<string, []>("op_59924_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_59924_end_mask_0 = const()[name = tensor<string, []>("op_59924_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59924_cast = slice_by_index(begin = var_59924_begin_0, end = var_59924_end_0, end_mask = var_59924_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_59924_cast")];
            tensor<int32, [4]> var_59928_begin_0 = const()[name = tensor<string, []>("op_59928_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_59928_end_0 = const()[name = tensor<string, []>("op_59928_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_59928_end_mask_0 = const()[name = tensor<string, []>("op_59928_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59928_cast = slice_by_index(begin = var_59928_begin_0, end = var_59928_end_0, end_mask = var_59928_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_59928_cast")];
            tensor<int32, [4]> var_59932_begin_0 = const()[name = tensor<string, []>("op_59932_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_59932_end_0 = const()[name = tensor<string, []>("op_59932_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_59932_end_mask_0 = const()[name = tensor<string, []>("op_59932_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59932_cast = slice_by_index(begin = var_59932_begin_0, end = var_59932_end_0, end_mask = var_59932_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_59932_cast")];
            tensor<int32, [4]> var_59936_begin_0 = const()[name = tensor<string, []>("op_59936_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_59936_end_0 = const()[name = tensor<string, []>("op_59936_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_59936_end_mask_0 = const()[name = tensor<string, []>("op_59936_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59936_cast = slice_by_index(begin = var_59936_begin_0, end = var_59936_end_0, end_mask = var_59936_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_59936_cast")];
            tensor<int32, [4]> var_59940_begin_0 = const()[name = tensor<string, []>("op_59940_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_59940_end_0 = const()[name = tensor<string, []>("op_59940_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 384])];
            tensor<bool, [4]> var_59940_end_mask_0 = const()[name = tensor<string, []>("op_59940_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59940_cast = slice_by_index(begin = var_59940_begin_0, end = var_59940_end_0, end_mask = var_59940_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_59940_cast")];
            tensor<int32, [4]> var_59944_begin_0 = const()[name = tensor<string, []>("op_59944_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_59944_end_0 = const()[name = tensor<string, []>("op_59944_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 448])];
            tensor<bool, [4]> var_59944_end_mask_0 = const()[name = tensor<string, []>("op_59944_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59944_cast = slice_by_index(begin = var_59944_begin_0, end = var_59944_end_0, end_mask = var_59944_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_59944_cast")];
            tensor<int32, [4]> var_59948_begin_0 = const()[name = tensor<string, []>("op_59948_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_59948_end_0 = const()[name = tensor<string, []>("op_59948_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 512])];
            tensor<bool, [4]> var_59948_end_mask_0 = const()[name = tensor<string, []>("op_59948_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59948_cast = slice_by_index(begin = var_59948_begin_0, end = var_59948_end_0, end_mask = var_59948_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_59948_cast")];
            tensor<int32, [4]> var_59952_begin_0 = const()[name = tensor<string, []>("op_59952_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_59952_end_0 = const()[name = tensor<string, []>("op_59952_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 576])];
            tensor<bool, [4]> var_59952_end_mask_0 = const()[name = tensor<string, []>("op_59952_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59952_cast = slice_by_index(begin = var_59952_begin_0, end = var_59952_end_0, end_mask = var_59952_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_59952_cast")];
            tensor<int32, [4]> var_59956_begin_0 = const()[name = tensor<string, []>("op_59956_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_59956_end_0 = const()[name = tensor<string, []>("op_59956_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 640])];
            tensor<bool, [4]> var_59956_end_mask_0 = const()[name = tensor<string, []>("op_59956_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_59956_cast = slice_by_index(begin = var_59956_begin_0, end = var_59956_end_0, end_mask = var_59956_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_59956_cast")];
            tensor<int32, [4]> var_59958_begin_0 = const()[name = tensor<string, []>("op_59958_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_59958_end_0 = const()[name = tensor<string, []>("op_59958_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_59958_end_mask_0 = const()[name = tensor<string, []>("op_59958_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59958_cast = slice_by_index(begin = var_59958_begin_0, end = var_59958_end_0, end_mask = var_59958_end_mask_0, x = v_269_cast)[name = tensor<string, []>("op_59958_cast")];
            tensor<int32, [4]> var_59962_begin_0 = const()[name = tensor<string, []>("op_59962_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_59962_end_0 = const()[name = tensor<string, []>("op_59962_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_59962_end_mask_0 = const()[name = tensor<string, []>("op_59962_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59962_cast = slice_by_index(begin = var_59962_begin_0, end = var_59962_end_0, end_mask = var_59962_end_mask_0, x = v_269_cast)[name = tensor<string, []>("op_59962_cast")];
            tensor<int32, [4]> var_59966_begin_0 = const()[name = tensor<string, []>("op_59966_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_59966_end_0 = const()[name = tensor<string, []>("op_59966_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_59966_end_mask_0 = const()[name = tensor<string, []>("op_59966_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59966_cast = slice_by_index(begin = var_59966_begin_0, end = var_59966_end_0, end_mask = var_59966_end_mask_0, x = v_269_cast)[name = tensor<string, []>("op_59966_cast")];
            tensor<int32, [4]> var_59970_begin_0 = const()[name = tensor<string, []>("op_59970_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_59970_end_0 = const()[name = tensor<string, []>("op_59970_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_59970_end_mask_0 = const()[name = tensor<string, []>("op_59970_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59970_cast = slice_by_index(begin = var_59970_begin_0, end = var_59970_end_0, end_mask = var_59970_end_mask_0, x = v_269_cast)[name = tensor<string, []>("op_59970_cast")];
            tensor<int32, [4]> var_59974_begin_0 = const()[name = tensor<string, []>("op_59974_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_59974_end_0 = const()[name = tensor<string, []>("op_59974_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_59974_end_mask_0 = const()[name = tensor<string, []>("op_59974_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59974_cast = slice_by_index(begin = var_59974_begin_0, end = var_59974_end_0, end_mask = var_59974_end_mask_0, x = v_269_cast)[name = tensor<string, []>("op_59974_cast")];
            tensor<int32, [4]> var_59978_begin_0 = const()[name = tensor<string, []>("op_59978_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_59978_end_0 = const()[name = tensor<string, []>("op_59978_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_59978_end_mask_0 = const()[name = tensor<string, []>("op_59978_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59978_cast = slice_by_index(begin = var_59978_begin_0, end = var_59978_end_0, end_mask = var_59978_end_mask_0, x = v_269_cast)[name = tensor<string, []>("op_59978_cast")];
            tensor<int32, [4]> var_59982_begin_0 = const()[name = tensor<string, []>("op_59982_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_59982_end_0 = const()[name = tensor<string, []>("op_59982_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_59982_end_mask_0 = const()[name = tensor<string, []>("op_59982_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59982_cast = slice_by_index(begin = var_59982_begin_0, end = var_59982_end_0, end_mask = var_59982_end_mask_0, x = v_269_cast)[name = tensor<string, []>("op_59982_cast")];
            tensor<int32, [4]> var_59986_begin_0 = const()[name = tensor<string, []>("op_59986_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_59986_end_0 = const()[name = tensor<string, []>("op_59986_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_59986_end_mask_0 = const()[name = tensor<string, []>("op_59986_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59986_cast = slice_by_index(begin = var_59986_begin_0, end = var_59986_end_0, end_mask = var_59986_end_mask_0, x = v_269_cast)[name = tensor<string, []>("op_59986_cast")];
            tensor<int32, [4]> var_59990_begin_0 = const()[name = tensor<string, []>("op_59990_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_59990_end_0 = const()[name = tensor<string, []>("op_59990_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_59990_end_mask_0 = const()[name = tensor<string, []>("op_59990_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59990_cast = slice_by_index(begin = var_59990_begin_0, end = var_59990_end_0, end_mask = var_59990_end_mask_0, x = v_269_cast)[name = tensor<string, []>("op_59990_cast")];
            tensor<int32, [4]> var_59994_begin_0 = const()[name = tensor<string, []>("op_59994_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_59994_end_0 = const()[name = tensor<string, []>("op_59994_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_59994_end_mask_0 = const()[name = tensor<string, []>("op_59994_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_59994_cast = slice_by_index(begin = var_59994_begin_0, end = var_59994_end_0, end_mask = var_59994_end_mask_0, x = v_269_cast)[name = tensor<string, []>("op_59994_cast")];
            tensor<string, []> var_59998_equation_0 = const()[name = tensor<string, []>("op_59998_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_59998_cast = einsum(equation = var_59998_equation_0, values = (var_59920_cast, var_59877_cast))[name = tensor<string, []>("op_59998_cast")];
            tensor<fp16, []> var_59999_to_fp16 = const()[name = tensor<string, []>("op_59999_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5081_cast = mul(x = var_59998_cast, y = var_59999_to_fp16)[name = tensor<string, []>("aw_5081_cast")];
            tensor<string, []> var_60002_equation_0 = const()[name = tensor<string, []>("op_60002_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60002_cast = einsum(equation = var_60002_equation_0, values = (var_59924_cast, var_59881_cast))[name = tensor<string, []>("op_60002_cast")];
            tensor<fp16, []> var_60003_to_fp16 = const()[name = tensor<string, []>("op_60003_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5083_cast = mul(x = var_60002_cast, y = var_60003_to_fp16)[name = tensor<string, []>("aw_5083_cast")];
            tensor<string, []> var_60006_equation_0 = const()[name = tensor<string, []>("op_60006_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60006_cast = einsum(equation = var_60006_equation_0, values = (var_59928_cast, var_59885_cast))[name = tensor<string, []>("op_60006_cast")];
            tensor<fp16, []> var_60007_to_fp16 = const()[name = tensor<string, []>("op_60007_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5085_cast = mul(x = var_60006_cast, y = var_60007_to_fp16)[name = tensor<string, []>("aw_5085_cast")];
            tensor<string, []> var_60010_equation_0 = const()[name = tensor<string, []>("op_60010_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60010_cast = einsum(equation = var_60010_equation_0, values = (var_59932_cast, var_59889_cast))[name = tensor<string, []>("op_60010_cast")];
            tensor<fp16, []> var_60011_to_fp16 = const()[name = tensor<string, []>("op_60011_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5087_cast = mul(x = var_60010_cast, y = var_60011_to_fp16)[name = tensor<string, []>("aw_5087_cast")];
            tensor<string, []> var_60014_equation_0 = const()[name = tensor<string, []>("op_60014_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60014_cast = einsum(equation = var_60014_equation_0, values = (var_59936_cast, var_59893_cast))[name = tensor<string, []>("op_60014_cast")];
            tensor<fp16, []> var_60015_to_fp16 = const()[name = tensor<string, []>("op_60015_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5089_cast = mul(x = var_60014_cast, y = var_60015_to_fp16)[name = tensor<string, []>("aw_5089_cast")];
            tensor<string, []> var_60018_equation_0 = const()[name = tensor<string, []>("op_60018_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60018_cast = einsum(equation = var_60018_equation_0, values = (var_59940_cast, var_59897_cast))[name = tensor<string, []>("op_60018_cast")];
            tensor<fp16, []> var_60019_to_fp16 = const()[name = tensor<string, []>("op_60019_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5091_cast = mul(x = var_60018_cast, y = var_60019_to_fp16)[name = tensor<string, []>("aw_5091_cast")];
            tensor<string, []> var_60022_equation_0 = const()[name = tensor<string, []>("op_60022_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60022_cast = einsum(equation = var_60022_equation_0, values = (var_59944_cast, var_59901_cast))[name = tensor<string, []>("op_60022_cast")];
            tensor<fp16, []> var_60023_to_fp16 = const()[name = tensor<string, []>("op_60023_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5093_cast = mul(x = var_60022_cast, y = var_60023_to_fp16)[name = tensor<string, []>("aw_5093_cast")];
            tensor<string, []> var_60026_equation_0 = const()[name = tensor<string, []>("op_60026_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60026_cast = einsum(equation = var_60026_equation_0, values = (var_59948_cast, var_59905_cast))[name = tensor<string, []>("op_60026_cast")];
            tensor<fp16, []> var_60027_to_fp16 = const()[name = tensor<string, []>("op_60027_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5095_cast = mul(x = var_60026_cast, y = var_60027_to_fp16)[name = tensor<string, []>("aw_5095_cast")];
            tensor<string, []> var_60030_equation_0 = const()[name = tensor<string, []>("op_60030_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60030_cast = einsum(equation = var_60030_equation_0, values = (var_59952_cast, var_59909_cast))[name = tensor<string, []>("op_60030_cast")];
            tensor<fp16, []> var_60031_to_fp16 = const()[name = tensor<string, []>("op_60031_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5097_cast = mul(x = var_60030_cast, y = var_60031_to_fp16)[name = tensor<string, []>("aw_5097_cast")];
            tensor<string, []> var_60034_equation_0 = const()[name = tensor<string, []>("op_60034_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60034_cast = einsum(equation = var_60034_equation_0, values = (var_59956_cast, var_59913_cast))[name = tensor<string, []>("op_60034_cast")];
            tensor<fp16, []> var_60035_to_fp16 = const()[name = tensor<string, []>("op_60035_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5099_cast = mul(x = var_60034_cast, y = var_60035_to_fp16)[name = tensor<string, []>("aw_5099_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60037_cast = softmax(axis = var_58064, x = aw_5081_cast)[name = tensor<string, []>("op_60037_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60038_cast = softmax(axis = var_58064, x = aw_5083_cast)[name = tensor<string, []>("op_60038_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60039_cast = softmax(axis = var_58064, x = aw_5085_cast)[name = tensor<string, []>("op_60039_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60040_cast = softmax(axis = var_58064, x = aw_5087_cast)[name = tensor<string, []>("op_60040_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60041_cast = softmax(axis = var_58064, x = aw_5089_cast)[name = tensor<string, []>("op_60041_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60042_cast = softmax(axis = var_58064, x = aw_5091_cast)[name = tensor<string, []>("op_60042_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60043_cast = softmax(axis = var_58064, x = aw_5093_cast)[name = tensor<string, []>("op_60043_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60044_cast = softmax(axis = var_58064, x = aw_5095_cast)[name = tensor<string, []>("op_60044_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60045_cast = softmax(axis = var_58064, x = aw_5097_cast)[name = tensor<string, []>("op_60045_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60046_cast = softmax(axis = var_58064, x = aw_5099_cast)[name = tensor<string, []>("op_60046_cast")];
            tensor<string, []> var_60048_equation_0 = const()[name = tensor<string, []>("op_60048_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60048_cast = einsum(equation = var_60048_equation_0, values = (var_59958_cast, var_60037_cast))[name = tensor<string, []>("op_60048_cast")];
            tensor<string, []> var_60050_equation_0 = const()[name = tensor<string, []>("op_60050_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60050_cast = einsum(equation = var_60050_equation_0, values = (var_59962_cast, var_60038_cast))[name = tensor<string, []>("op_60050_cast")];
            tensor<string, []> var_60052_equation_0 = const()[name = tensor<string, []>("op_60052_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60052_cast = einsum(equation = var_60052_equation_0, values = (var_59966_cast, var_60039_cast))[name = tensor<string, []>("op_60052_cast")];
            tensor<string, []> var_60054_equation_0 = const()[name = tensor<string, []>("op_60054_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60054_cast = einsum(equation = var_60054_equation_0, values = (var_59970_cast, var_60040_cast))[name = tensor<string, []>("op_60054_cast")];
            tensor<string, []> var_60056_equation_0 = const()[name = tensor<string, []>("op_60056_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60056_cast = einsum(equation = var_60056_equation_0, values = (var_59974_cast, var_60041_cast))[name = tensor<string, []>("op_60056_cast")];
            tensor<string, []> var_60058_equation_0 = const()[name = tensor<string, []>("op_60058_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60058_cast = einsum(equation = var_60058_equation_0, values = (var_59978_cast, var_60042_cast))[name = tensor<string, []>("op_60058_cast")];
            tensor<string, []> var_60060_equation_0 = const()[name = tensor<string, []>("op_60060_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60060_cast = einsum(equation = var_60060_equation_0, values = (var_59982_cast, var_60043_cast))[name = tensor<string, []>("op_60060_cast")];
            tensor<string, []> var_60062_equation_0 = const()[name = tensor<string, []>("op_60062_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60062_cast = einsum(equation = var_60062_equation_0, values = (var_59986_cast, var_60044_cast))[name = tensor<string, []>("op_60062_cast")];
            tensor<string, []> var_60064_equation_0 = const()[name = tensor<string, []>("op_60064_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60064_cast = einsum(equation = var_60064_equation_0, values = (var_59990_cast, var_60045_cast))[name = tensor<string, []>("op_60064_cast")];
            tensor<string, []> var_60066_equation_0 = const()[name = tensor<string, []>("op_60066_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60066_cast = einsum(equation = var_60066_equation_0, values = (var_59994_cast, var_60046_cast))[name = tensor<string, []>("op_60066_cast")];
            tensor<bool, []> input_781_interleave_0 = const()[name = tensor<string, []>("input_781_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_781_cast = concat(axis = var_58064, interleave = input_781_interleave_0, values = (var_60048_cast, var_60050_cast, var_60052_cast, var_60054_cast, var_60056_cast, var_60058_cast, var_60060_cast, var_60062_cast, var_60064_cast, var_60066_cast))[name = tensor<string, []>("input_781_cast")];
            tensor<int32, [2]> var_60072 = const()[name = tensor<string, []>("op_60072"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60074 = const()[name = tensor<string, []>("op_60074"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_60076_pad_type_0 = const()[name = tensor<string, []>("op_60076_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_60076_pad_0 = const()[name = tensor<string, []>("op_60076_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5023968320)))];
            tensor<fp16, [640]> up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5024787584)))];
            tensor<fp16, [2, 640, 1, 4096]> var_60076_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_60074, groups = var_58064, pad = var_60076_pad_0, pad_type = var_60076_pad_type_0, strides = var_60072, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_781_cast)[name = tensor<string, []>("op_60076_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_405_cast = add(x = var_60076_cast, y = inputs_403_cast)[name = tensor<string, []>("inputs_405_cast")];
            tensor<int32, [1]> var_60080 = const()[name = tensor<string, []>("op_60080"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_405_cast = reduce_mean(axes = var_60080, keep_dims = var_58059, x = inputs_405_cast)[name = tensor<string, []>("channels_mean_405_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_405_cast = sub(x = inputs_405_cast, y = channels_mean_405_cast)[name = tensor<string, []>("zero_mean_405_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_405_cast = mul(x = zero_mean_405_cast, y = zero_mean_405_cast)[name = tensor<string, []>("zero_mean_sq_405_cast")];
            tensor<int32, [1]> var_60084 = const()[name = tensor<string, []>("op_60084"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_60085_cast = reduce_mean(axes = var_60084, keep_dims = var_58059, x = zero_mean_sq_405_cast)[name = tensor<string, []>("op_60085_cast")];
            tensor<fp16, []> var_60086_to_fp16 = const()[name = tensor<string, []>("op_60086_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_60087_cast = add(x = var_60085_cast, y = var_60086_to_fp16)[name = tensor<string, []>("op_60087_cast")];
            tensor<fp16, []> denom_405_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_405_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_405_cast = rsqrt(epsilon = denom_405_epsilon_0_to_fp16, x = var_60087_cast)[name = tensor<string, []>("denom_405_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_405_cast = mul(x = zero_mean_405_cast, y = denom_405_cast)[name = tensor<string, []>("out_405_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_60091_to_fp16 = const()[name = tensor<string, []>("op_60091_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5024788928)))];
            tensor<fp16, [2, 640, 1, 4096]> var_60092_cast = add(x = out_405_cast, y = var_60091_to_fp16)[name = tensor<string, []>("op_60092_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_60094_to_fp16 = const()[name = tensor<string, []>("op_60094_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5024790272)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_543_cast = mul(x = var_60092_cast, y = var_60094_to_fp16)[name = tensor<string, []>("hidden_states_543_cast")];
            tensor<int32, [2]> var_60101 = const()[name = tensor<string, []>("op_60101"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60103 = const()[name = tensor<string, []>("op_60103"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_271_pad_type_0 = const()[name = tensor<string, []>("q_271_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_271_pad_0 = const()[name = tensor<string, []>("q_271_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5024791616)))];
            tensor<fp16, [2, 640, 1, 4096]> q_271_cast = conv(dilations = var_60103, groups = var_58064, pad = q_271_pad_0, pad_type = q_271_pad_type_0, strides = var_60101, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_543_cast)[name = tensor<string, []>("q_271_cast")];
            tensor<int32, [2]> var_60107 = const()[name = tensor<string, []>("op_60107"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60109 = const()[name = tensor<string, []>("op_60109"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_541_pad_type_0 = const()[name = tensor<string, []>("k_541_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_541_pad_0 = const()[name = tensor<string, []>("k_541_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5025610880)))];
            tensor<fp16, [2, 640, 1, 77]> k_541_cast = conv(dilations = var_60109, groups = var_58064, pad = k_541_pad_0, pad_type = k_541_pad_type_0, strides = var_60107, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_541_cast")];
            tensor<int32, [2]> var_60113 = const()[name = tensor<string, []>("op_60113"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60115 = const()[name = tensor<string, []>("op_60115"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_271_pad_type_0 = const()[name = tensor<string, []>("v_271_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_271_pad_0 = const()[name = tensor<string, []>("v_271_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5028232384)))];
            tensor<fp16, [2, 640, 1, 77]> v_271_cast = conv(dilations = var_60115, groups = var_58064, pad = v_271_pad_0, pad_type = v_271_pad_type_0, strides = var_60113, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_271_cast")];
            tensor<int32, [4]> var_60119_begin_0 = const()[name = tensor<string, []>("op_60119_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_60119_end_0 = const()[name = tensor<string, []>("op_60119_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_60119_end_mask_0 = const()[name = tensor<string, []>("op_60119_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60119_cast = slice_by_index(begin = var_60119_begin_0, end = var_60119_end_0, end_mask = var_60119_end_mask_0, x = q_271_cast)[name = tensor<string, []>("op_60119_cast")];
            tensor<int32, [4]> var_60123_begin_0 = const()[name = tensor<string, []>("op_60123_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_60123_end_0 = const()[name = tensor<string, []>("op_60123_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_60123_end_mask_0 = const()[name = tensor<string, []>("op_60123_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60123_cast = slice_by_index(begin = var_60123_begin_0, end = var_60123_end_0, end_mask = var_60123_end_mask_0, x = q_271_cast)[name = tensor<string, []>("op_60123_cast")];
            tensor<int32, [4]> var_60127_begin_0 = const()[name = tensor<string, []>("op_60127_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_60127_end_0 = const()[name = tensor<string, []>("op_60127_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_60127_end_mask_0 = const()[name = tensor<string, []>("op_60127_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60127_cast = slice_by_index(begin = var_60127_begin_0, end = var_60127_end_0, end_mask = var_60127_end_mask_0, x = q_271_cast)[name = tensor<string, []>("op_60127_cast")];
            tensor<int32, [4]> var_60131_begin_0 = const()[name = tensor<string, []>("op_60131_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_60131_end_0 = const()[name = tensor<string, []>("op_60131_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_60131_end_mask_0 = const()[name = tensor<string, []>("op_60131_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60131_cast = slice_by_index(begin = var_60131_begin_0, end = var_60131_end_0, end_mask = var_60131_end_mask_0, x = q_271_cast)[name = tensor<string, []>("op_60131_cast")];
            tensor<int32, [4]> var_60135_begin_0 = const()[name = tensor<string, []>("op_60135_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_60135_end_0 = const()[name = tensor<string, []>("op_60135_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_60135_end_mask_0 = const()[name = tensor<string, []>("op_60135_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60135_cast = slice_by_index(begin = var_60135_begin_0, end = var_60135_end_0, end_mask = var_60135_end_mask_0, x = q_271_cast)[name = tensor<string, []>("op_60135_cast")];
            tensor<int32, [4]> var_60139_begin_0 = const()[name = tensor<string, []>("op_60139_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_60139_end_0 = const()[name = tensor<string, []>("op_60139_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_60139_end_mask_0 = const()[name = tensor<string, []>("op_60139_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60139_cast = slice_by_index(begin = var_60139_begin_0, end = var_60139_end_0, end_mask = var_60139_end_mask_0, x = q_271_cast)[name = tensor<string, []>("op_60139_cast")];
            tensor<int32, [4]> var_60143_begin_0 = const()[name = tensor<string, []>("op_60143_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_60143_end_0 = const()[name = tensor<string, []>("op_60143_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_60143_end_mask_0 = const()[name = tensor<string, []>("op_60143_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60143_cast = slice_by_index(begin = var_60143_begin_0, end = var_60143_end_0, end_mask = var_60143_end_mask_0, x = q_271_cast)[name = tensor<string, []>("op_60143_cast")];
            tensor<int32, [4]> var_60147_begin_0 = const()[name = tensor<string, []>("op_60147_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_60147_end_0 = const()[name = tensor<string, []>("op_60147_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_60147_end_mask_0 = const()[name = tensor<string, []>("op_60147_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60147_cast = slice_by_index(begin = var_60147_begin_0, end = var_60147_end_0, end_mask = var_60147_end_mask_0, x = q_271_cast)[name = tensor<string, []>("op_60147_cast")];
            tensor<int32, [4]> var_60151_begin_0 = const()[name = tensor<string, []>("op_60151_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_60151_end_0 = const()[name = tensor<string, []>("op_60151_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_60151_end_mask_0 = const()[name = tensor<string, []>("op_60151_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60151_cast = slice_by_index(begin = var_60151_begin_0, end = var_60151_end_0, end_mask = var_60151_end_mask_0, x = q_271_cast)[name = tensor<string, []>("op_60151_cast")];
            tensor<int32, [4]> var_60155_begin_0 = const()[name = tensor<string, []>("op_60155_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_60155_end_0 = const()[name = tensor<string, []>("op_60155_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_60155_end_mask_0 = const()[name = tensor<string, []>("op_60155_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60155_cast = slice_by_index(begin = var_60155_begin_0, end = var_60155_end_0, end_mask = var_60155_end_mask_0, x = q_271_cast)[name = tensor<string, []>("op_60155_cast")];
            tensor<int32, [4]> k_543_perm_0 = const()[name = tensor<string, []>("k_543_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_60162_begin_0 = const()[name = tensor<string, []>("op_60162_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_60162_end_0 = const()[name = tensor<string, []>("op_60162_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_60162_end_mask_0 = const()[name = tensor<string, []>("op_60162_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> transpose_4 = transpose(perm = k_543_perm_0, x = k_541_cast)[name = tensor<string, []>("transpose_4")];
            tensor<fp16, [2, 77, 1, 64]> var_60162_cast = slice_by_index(begin = var_60162_begin_0, end = var_60162_end_0, end_mask = var_60162_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_60162_cast")];
            tensor<int32, [4]> var_60166_begin_0 = const()[name = tensor<string, []>("op_60166_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_60166_end_0 = const()[name = tensor<string, []>("op_60166_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_60166_end_mask_0 = const()[name = tensor<string, []>("op_60166_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60166_cast = slice_by_index(begin = var_60166_begin_0, end = var_60166_end_0, end_mask = var_60166_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_60166_cast")];
            tensor<int32, [4]> var_60170_begin_0 = const()[name = tensor<string, []>("op_60170_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_60170_end_0 = const()[name = tensor<string, []>("op_60170_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_60170_end_mask_0 = const()[name = tensor<string, []>("op_60170_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60170_cast = slice_by_index(begin = var_60170_begin_0, end = var_60170_end_0, end_mask = var_60170_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_60170_cast")];
            tensor<int32, [4]> var_60174_begin_0 = const()[name = tensor<string, []>("op_60174_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_60174_end_0 = const()[name = tensor<string, []>("op_60174_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_60174_end_mask_0 = const()[name = tensor<string, []>("op_60174_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60174_cast = slice_by_index(begin = var_60174_begin_0, end = var_60174_end_0, end_mask = var_60174_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_60174_cast")];
            tensor<int32, [4]> var_60178_begin_0 = const()[name = tensor<string, []>("op_60178_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_60178_end_0 = const()[name = tensor<string, []>("op_60178_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_60178_end_mask_0 = const()[name = tensor<string, []>("op_60178_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60178_cast = slice_by_index(begin = var_60178_begin_0, end = var_60178_end_0, end_mask = var_60178_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_60178_cast")];
            tensor<int32, [4]> var_60182_begin_0 = const()[name = tensor<string, []>("op_60182_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_60182_end_0 = const()[name = tensor<string, []>("op_60182_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_60182_end_mask_0 = const()[name = tensor<string, []>("op_60182_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60182_cast = slice_by_index(begin = var_60182_begin_0, end = var_60182_end_0, end_mask = var_60182_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_60182_cast")];
            tensor<int32, [4]> var_60186_begin_0 = const()[name = tensor<string, []>("op_60186_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_60186_end_0 = const()[name = tensor<string, []>("op_60186_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_60186_end_mask_0 = const()[name = tensor<string, []>("op_60186_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60186_cast = slice_by_index(begin = var_60186_begin_0, end = var_60186_end_0, end_mask = var_60186_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_60186_cast")];
            tensor<int32, [4]> var_60190_begin_0 = const()[name = tensor<string, []>("op_60190_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_60190_end_0 = const()[name = tensor<string, []>("op_60190_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_60190_end_mask_0 = const()[name = tensor<string, []>("op_60190_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60190_cast = slice_by_index(begin = var_60190_begin_0, end = var_60190_end_0, end_mask = var_60190_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_60190_cast")];
            tensor<int32, [4]> var_60194_begin_0 = const()[name = tensor<string, []>("op_60194_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_60194_end_0 = const()[name = tensor<string, []>("op_60194_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_60194_end_mask_0 = const()[name = tensor<string, []>("op_60194_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60194_cast = slice_by_index(begin = var_60194_begin_0, end = var_60194_end_0, end_mask = var_60194_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_60194_cast")];
            tensor<int32, [4]> var_60198_begin_0 = const()[name = tensor<string, []>("op_60198_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_60198_end_0 = const()[name = tensor<string, []>("op_60198_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_60198_end_mask_0 = const()[name = tensor<string, []>("op_60198_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60198_cast = slice_by_index(begin = var_60198_begin_0, end = var_60198_end_0, end_mask = var_60198_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_60198_cast")];
            tensor<int32, [4]> var_60200_begin_0 = const()[name = tensor<string, []>("op_60200_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_60200_end_0 = const()[name = tensor<string, []>("op_60200_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_60200_end_mask_0 = const()[name = tensor<string, []>("op_60200_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60200_cast = slice_by_index(begin = var_60200_begin_0, end = var_60200_end_0, end_mask = var_60200_end_mask_0, x = v_271_cast)[name = tensor<string, []>("op_60200_cast")];
            tensor<int32, [4]> var_60204_begin_0 = const()[name = tensor<string, []>("op_60204_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_60204_end_0 = const()[name = tensor<string, []>("op_60204_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_60204_end_mask_0 = const()[name = tensor<string, []>("op_60204_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60204_cast = slice_by_index(begin = var_60204_begin_0, end = var_60204_end_0, end_mask = var_60204_end_mask_0, x = v_271_cast)[name = tensor<string, []>("op_60204_cast")];
            tensor<int32, [4]> var_60208_begin_0 = const()[name = tensor<string, []>("op_60208_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_60208_end_0 = const()[name = tensor<string, []>("op_60208_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_60208_end_mask_0 = const()[name = tensor<string, []>("op_60208_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60208_cast = slice_by_index(begin = var_60208_begin_0, end = var_60208_end_0, end_mask = var_60208_end_mask_0, x = v_271_cast)[name = tensor<string, []>("op_60208_cast")];
            tensor<int32, [4]> var_60212_begin_0 = const()[name = tensor<string, []>("op_60212_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_60212_end_0 = const()[name = tensor<string, []>("op_60212_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_60212_end_mask_0 = const()[name = tensor<string, []>("op_60212_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60212_cast = slice_by_index(begin = var_60212_begin_0, end = var_60212_end_0, end_mask = var_60212_end_mask_0, x = v_271_cast)[name = tensor<string, []>("op_60212_cast")];
            tensor<int32, [4]> var_60216_begin_0 = const()[name = tensor<string, []>("op_60216_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_60216_end_0 = const()[name = tensor<string, []>("op_60216_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_60216_end_mask_0 = const()[name = tensor<string, []>("op_60216_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60216_cast = slice_by_index(begin = var_60216_begin_0, end = var_60216_end_0, end_mask = var_60216_end_mask_0, x = v_271_cast)[name = tensor<string, []>("op_60216_cast")];
            tensor<int32, [4]> var_60220_begin_0 = const()[name = tensor<string, []>("op_60220_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_60220_end_0 = const()[name = tensor<string, []>("op_60220_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_60220_end_mask_0 = const()[name = tensor<string, []>("op_60220_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60220_cast = slice_by_index(begin = var_60220_begin_0, end = var_60220_end_0, end_mask = var_60220_end_mask_0, x = v_271_cast)[name = tensor<string, []>("op_60220_cast")];
            tensor<int32, [4]> var_60224_begin_0 = const()[name = tensor<string, []>("op_60224_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_60224_end_0 = const()[name = tensor<string, []>("op_60224_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_60224_end_mask_0 = const()[name = tensor<string, []>("op_60224_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60224_cast = slice_by_index(begin = var_60224_begin_0, end = var_60224_end_0, end_mask = var_60224_end_mask_0, x = v_271_cast)[name = tensor<string, []>("op_60224_cast")];
            tensor<int32, [4]> var_60228_begin_0 = const()[name = tensor<string, []>("op_60228_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_60228_end_0 = const()[name = tensor<string, []>("op_60228_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_60228_end_mask_0 = const()[name = tensor<string, []>("op_60228_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60228_cast = slice_by_index(begin = var_60228_begin_0, end = var_60228_end_0, end_mask = var_60228_end_mask_0, x = v_271_cast)[name = tensor<string, []>("op_60228_cast")];
            tensor<int32, [4]> var_60232_begin_0 = const()[name = tensor<string, []>("op_60232_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_60232_end_0 = const()[name = tensor<string, []>("op_60232_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_60232_end_mask_0 = const()[name = tensor<string, []>("op_60232_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60232_cast = slice_by_index(begin = var_60232_begin_0, end = var_60232_end_0, end_mask = var_60232_end_mask_0, x = v_271_cast)[name = tensor<string, []>("op_60232_cast")];
            tensor<int32, [4]> var_60236_begin_0 = const()[name = tensor<string, []>("op_60236_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_60236_end_0 = const()[name = tensor<string, []>("op_60236_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_60236_end_mask_0 = const()[name = tensor<string, []>("op_60236_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60236_cast = slice_by_index(begin = var_60236_begin_0, end = var_60236_end_0, end_mask = var_60236_end_mask_0, x = v_271_cast)[name = tensor<string, []>("op_60236_cast")];
            tensor<string, []> var_60240_equation_0 = const()[name = tensor<string, []>("op_60240_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60240_cast = einsum(equation = var_60240_equation_0, values = (var_60162_cast, var_60119_cast))[name = tensor<string, []>("op_60240_cast")];
            tensor<fp16, []> var_60241_to_fp16 = const()[name = tensor<string, []>("op_60241_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5101_cast = mul(x = var_60240_cast, y = var_60241_to_fp16)[name = tensor<string, []>("aw_5101_cast")];
            tensor<string, []> var_60244_equation_0 = const()[name = tensor<string, []>("op_60244_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60244_cast = einsum(equation = var_60244_equation_0, values = (var_60166_cast, var_60123_cast))[name = tensor<string, []>("op_60244_cast")];
            tensor<fp16, []> var_60245_to_fp16 = const()[name = tensor<string, []>("op_60245_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5103_cast = mul(x = var_60244_cast, y = var_60245_to_fp16)[name = tensor<string, []>("aw_5103_cast")];
            tensor<string, []> var_60248_equation_0 = const()[name = tensor<string, []>("op_60248_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60248_cast = einsum(equation = var_60248_equation_0, values = (var_60170_cast, var_60127_cast))[name = tensor<string, []>("op_60248_cast")];
            tensor<fp16, []> var_60249_to_fp16 = const()[name = tensor<string, []>("op_60249_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5105_cast = mul(x = var_60248_cast, y = var_60249_to_fp16)[name = tensor<string, []>("aw_5105_cast")];
            tensor<string, []> var_60252_equation_0 = const()[name = tensor<string, []>("op_60252_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60252_cast = einsum(equation = var_60252_equation_0, values = (var_60174_cast, var_60131_cast))[name = tensor<string, []>("op_60252_cast")];
            tensor<fp16, []> var_60253_to_fp16 = const()[name = tensor<string, []>("op_60253_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5107_cast = mul(x = var_60252_cast, y = var_60253_to_fp16)[name = tensor<string, []>("aw_5107_cast")];
            tensor<string, []> var_60256_equation_0 = const()[name = tensor<string, []>("op_60256_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60256_cast = einsum(equation = var_60256_equation_0, values = (var_60178_cast, var_60135_cast))[name = tensor<string, []>("op_60256_cast")];
            tensor<fp16, []> var_60257_to_fp16 = const()[name = tensor<string, []>("op_60257_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5109_cast = mul(x = var_60256_cast, y = var_60257_to_fp16)[name = tensor<string, []>("aw_5109_cast")];
            tensor<string, []> var_60260_equation_0 = const()[name = tensor<string, []>("op_60260_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60260_cast = einsum(equation = var_60260_equation_0, values = (var_60182_cast, var_60139_cast))[name = tensor<string, []>("op_60260_cast")];
            tensor<fp16, []> var_60261_to_fp16 = const()[name = tensor<string, []>("op_60261_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5111_cast = mul(x = var_60260_cast, y = var_60261_to_fp16)[name = tensor<string, []>("aw_5111_cast")];
            tensor<string, []> var_60264_equation_0 = const()[name = tensor<string, []>("op_60264_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60264_cast = einsum(equation = var_60264_equation_0, values = (var_60186_cast, var_60143_cast))[name = tensor<string, []>("op_60264_cast")];
            tensor<fp16, []> var_60265_to_fp16 = const()[name = tensor<string, []>("op_60265_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5113_cast = mul(x = var_60264_cast, y = var_60265_to_fp16)[name = tensor<string, []>("aw_5113_cast")];
            tensor<string, []> var_60268_equation_0 = const()[name = tensor<string, []>("op_60268_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60268_cast = einsum(equation = var_60268_equation_0, values = (var_60190_cast, var_60147_cast))[name = tensor<string, []>("op_60268_cast")];
            tensor<fp16, []> var_60269_to_fp16 = const()[name = tensor<string, []>("op_60269_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5115_cast = mul(x = var_60268_cast, y = var_60269_to_fp16)[name = tensor<string, []>("aw_5115_cast")];
            tensor<string, []> var_60272_equation_0 = const()[name = tensor<string, []>("op_60272_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60272_cast = einsum(equation = var_60272_equation_0, values = (var_60194_cast, var_60151_cast))[name = tensor<string, []>("op_60272_cast")];
            tensor<fp16, []> var_60273_to_fp16 = const()[name = tensor<string, []>("op_60273_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5117_cast = mul(x = var_60272_cast, y = var_60273_to_fp16)[name = tensor<string, []>("aw_5117_cast")];
            tensor<string, []> var_60276_equation_0 = const()[name = tensor<string, []>("op_60276_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60276_cast = einsum(equation = var_60276_equation_0, values = (var_60198_cast, var_60155_cast))[name = tensor<string, []>("op_60276_cast")];
            tensor<fp16, []> var_60277_to_fp16 = const()[name = tensor<string, []>("op_60277_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5119_cast = mul(x = var_60276_cast, y = var_60277_to_fp16)[name = tensor<string, []>("aw_5119_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60279_cast = softmax(axis = var_58064, x = aw_5101_cast)[name = tensor<string, []>("op_60279_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60280_cast = softmax(axis = var_58064, x = aw_5103_cast)[name = tensor<string, []>("op_60280_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60281_cast = softmax(axis = var_58064, x = aw_5105_cast)[name = tensor<string, []>("op_60281_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60282_cast = softmax(axis = var_58064, x = aw_5107_cast)[name = tensor<string, []>("op_60282_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60283_cast = softmax(axis = var_58064, x = aw_5109_cast)[name = tensor<string, []>("op_60283_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60284_cast = softmax(axis = var_58064, x = aw_5111_cast)[name = tensor<string, []>("op_60284_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60285_cast = softmax(axis = var_58064, x = aw_5113_cast)[name = tensor<string, []>("op_60285_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60286_cast = softmax(axis = var_58064, x = aw_5115_cast)[name = tensor<string, []>("op_60286_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60287_cast = softmax(axis = var_58064, x = aw_5117_cast)[name = tensor<string, []>("op_60287_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60288_cast = softmax(axis = var_58064, x = aw_5119_cast)[name = tensor<string, []>("op_60288_cast")];
            tensor<string, []> var_60290_equation_0 = const()[name = tensor<string, []>("op_60290_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60290_cast = einsum(equation = var_60290_equation_0, values = (var_60200_cast, var_60279_cast))[name = tensor<string, []>("op_60290_cast")];
            tensor<string, []> var_60292_equation_0 = const()[name = tensor<string, []>("op_60292_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60292_cast = einsum(equation = var_60292_equation_0, values = (var_60204_cast, var_60280_cast))[name = tensor<string, []>("op_60292_cast")];
            tensor<string, []> var_60294_equation_0 = const()[name = tensor<string, []>("op_60294_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60294_cast = einsum(equation = var_60294_equation_0, values = (var_60208_cast, var_60281_cast))[name = tensor<string, []>("op_60294_cast")];
            tensor<string, []> var_60296_equation_0 = const()[name = tensor<string, []>("op_60296_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60296_cast = einsum(equation = var_60296_equation_0, values = (var_60212_cast, var_60282_cast))[name = tensor<string, []>("op_60296_cast")];
            tensor<string, []> var_60298_equation_0 = const()[name = tensor<string, []>("op_60298_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60298_cast = einsum(equation = var_60298_equation_0, values = (var_60216_cast, var_60283_cast))[name = tensor<string, []>("op_60298_cast")];
            tensor<string, []> var_60300_equation_0 = const()[name = tensor<string, []>("op_60300_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60300_cast = einsum(equation = var_60300_equation_0, values = (var_60220_cast, var_60284_cast))[name = tensor<string, []>("op_60300_cast")];
            tensor<string, []> var_60302_equation_0 = const()[name = tensor<string, []>("op_60302_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60302_cast = einsum(equation = var_60302_equation_0, values = (var_60224_cast, var_60285_cast))[name = tensor<string, []>("op_60302_cast")];
            tensor<string, []> var_60304_equation_0 = const()[name = tensor<string, []>("op_60304_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60304_cast = einsum(equation = var_60304_equation_0, values = (var_60228_cast, var_60286_cast))[name = tensor<string, []>("op_60304_cast")];
            tensor<string, []> var_60306_equation_0 = const()[name = tensor<string, []>("op_60306_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60306_cast = einsum(equation = var_60306_equation_0, values = (var_60232_cast, var_60287_cast))[name = tensor<string, []>("op_60306_cast")];
            tensor<string, []> var_60308_equation_0 = const()[name = tensor<string, []>("op_60308_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60308_cast = einsum(equation = var_60308_equation_0, values = (var_60236_cast, var_60288_cast))[name = tensor<string, []>("op_60308_cast")];
            tensor<bool, []> input_783_interleave_0 = const()[name = tensor<string, []>("input_783_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_783_cast = concat(axis = var_58064, interleave = input_783_interleave_0, values = (var_60290_cast, var_60292_cast, var_60294_cast, var_60296_cast, var_60298_cast, var_60300_cast, var_60302_cast, var_60304_cast, var_60306_cast, var_60308_cast))[name = tensor<string, []>("input_783_cast")];
            tensor<int32, [2]> var_60314 = const()[name = tensor<string, []>("op_60314"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60316 = const()[name = tensor<string, []>("op_60316"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_60318_pad_type_0 = const()[name = tensor<string, []>("op_60318_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_60318_pad_0 = const()[name = tensor<string, []>("op_60318_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5030853888)))];
            tensor<fp16, [640]> up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5031673152)))];
            tensor<fp16, [2, 640, 1, 4096]> var_60318_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_60316, groups = var_58064, pad = var_60318_pad_0, pad_type = var_60318_pad_type_0, strides = var_60314, weight = up_blocks_1_attentions_1_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_783_cast)[name = tensor<string, []>("op_60318_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_407_cast = add(x = var_60318_cast, y = inputs_405_cast)[name = tensor<string, []>("inputs_407_cast")];
            tensor<int32, [1]> var_60322 = const()[name = tensor<string, []>("op_60322"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_407_cast = reduce_mean(axes = var_60322, keep_dims = var_58059, x = inputs_407_cast)[name = tensor<string, []>("channels_mean_407_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_407_cast = sub(x = inputs_407_cast, y = channels_mean_407_cast)[name = tensor<string, []>("zero_mean_407_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_407_cast = mul(x = zero_mean_407_cast, y = zero_mean_407_cast)[name = tensor<string, []>("zero_mean_sq_407_cast")];
            tensor<int32, [1]> var_60326 = const()[name = tensor<string, []>("op_60326"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_60327_cast = reduce_mean(axes = var_60326, keep_dims = var_58059, x = zero_mean_sq_407_cast)[name = tensor<string, []>("op_60327_cast")];
            tensor<fp16, []> var_60328_to_fp16 = const()[name = tensor<string, []>("op_60328_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_60329_cast = add(x = var_60327_cast, y = var_60328_to_fp16)[name = tensor<string, []>("op_60329_cast")];
            tensor<fp16, []> denom_407_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_407_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_407_cast = rsqrt(epsilon = denom_407_epsilon_0_to_fp16, x = var_60329_cast)[name = tensor<string, []>("denom_407_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_407_cast = mul(x = zero_mean_407_cast, y = denom_407_cast)[name = tensor<string, []>("out_407_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_60333_to_fp16 = const()[name = tensor<string, []>("op_60333_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5031674496)))];
            tensor<fp16, [2, 640, 1, 4096]> var_60334_cast = add(x = out_407_cast, y = var_60333_to_fp16)[name = tensor<string, []>("op_60334_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_60336_to_fp16 = const()[name = tensor<string, []>("op_60336_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5031675840)))];
            tensor<fp16, [2, 640, 1, 4096]> input_785_cast = mul(x = var_60334_cast, y = var_60336_to_fp16)[name = tensor<string, []>("input_785_cast")];
            tensor<int32, [2]> var_60344 = const()[name = tensor<string, []>("op_60344"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60346 = const()[name = tensor<string, []>("op_60346"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_60348_pad_type_0 = const()[name = tensor<string, []>("op_60348_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_60348_pad_0 = const()[name = tensor<string, []>("op_60348_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [5120, 640, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [5120, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5031677184)))];
            tensor<fp16, [5120]> up_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5038230848)))];
            tensor<fp16, [2, 5120, 1, 4096]> var_60348_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_60346, groups = var_58064, pad = var_60348_pad_0, pad_type = var_60348_pad_type_0, strides = var_60344, weight = up_blocks_1_attentions_1_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_785_cast)[name = tensor<string, []>("op_60348_cast")];
            tensor<int32, [2]> var_60349_split_sizes_0 = const()[name = tensor<string, []>("op_60349_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_60349_axis_0 = const()[name = tensor<string, []>("op_60349_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 4096]> var_60349_cast_0, tensor<fp16, [2, 2560, 1, 4096]> var_60349_cast_1 = split(axis = var_60349_axis_0, split_sizes = var_60349_split_sizes_0, x = var_60348_cast)[name = tensor<string, []>("op_60349_cast")];
            tensor<string, []> var_60351_mode_0 = const()[name = tensor<string, []>("op_60351_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 4096]> var_60351_cast = gelu(mode = var_60351_mode_0, x = var_60349_cast_1)[name = tensor<string, []>("op_60351_cast")];
            tensor<fp16, [2, 2560, 1, 4096]> input_787_cast = mul(x = var_60349_cast_0, y = var_60351_cast)[name = tensor<string, []>("input_787_cast")];
            tensor<int32, [2]> var_60355 = const()[name = tensor<string, []>("op_60355"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60357 = const()[name = tensor<string, []>("op_60357"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_60359_pad_type_0 = const()[name = tensor<string, []>("op_60359_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_60359_pad_0 = const()[name = tensor<string, []>("op_60359_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2560, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor<fp16, [640, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5038241152)))];
            tensor<fp16, [640]> up_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5041518016)))];
            tensor<fp16, [2, 640, 1, 4096]> var_60359_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_60357, groups = var_58064, pad = var_60359_pad_0, pad_type = var_60359_pad_type_0, strides = var_60355, weight = up_blocks_1_attentions_1_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_787_cast)[name = tensor<string, []>("op_60359_cast")];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_547_cast = add(x = var_60359_cast, y = inputs_407_cast)[name = tensor<string, []>("hidden_states_547_cast")];
            tensor<int32, [4]> var_60361 = const()[name = tensor<string, []>("op_60361"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> input_789_cast = reshape(shape = var_60361, x = hidden_states_547_cast)[name = tensor<string, []>("input_789_cast")];
            tensor<int32, [2]> var_60365 = const()[name = tensor<string, []>("op_60365"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60367 = const()[name = tensor<string, []>("op_60367"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_549_pad_type_0 = const()[name = tensor<string, []>("hidden_states_549_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_549_pad_0 = const()[name = tensor<string, []>("hidden_states_549_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_proj_out_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5041519360)))];
            tensor<fp16, [640]> up_blocks_1_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_proj_out_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5042338624)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_549_cast = conv(bias = up_blocks_1_attentions_1_proj_out_bias_to_fp16, dilations = var_60367, groups = var_58064, pad = hidden_states_549_pad_0, pad_type = hidden_states_549_pad_type_0, strides = var_60365, weight = up_blocks_1_attentions_1_proj_out_weight_to_fp16, x = input_789_cast)[name = tensor<string, []>("hidden_states_549_cast")];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_551_cast = add(x = hidden_states_549_cast, y = hidden_states_531_cast)[name = tensor<string, []>("hidden_states_551_cast")];
            tensor<bool, []> input_791_interleave_0 = const()[name = tensor<string, []>("input_791_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 960, 64, 64]> input_791_cast = concat(axis = var_58064, interleave = input_791_interleave_0, values = (hidden_states_551_cast, res_hidden_states_11_cast))[name = tensor<string, []>("input_791_cast")];
            tensor<int32, [5]> reshape_144_shape_0 = const()[name = tensor<string, []>("reshape_144_shape_0"), val = tensor<int32, [5]>([2, 32, 30, 64, 64])];
            tensor<fp16, [2, 32, 30, 64, 64]> reshape_144_cast = reshape(shape = reshape_144_shape_0, x = input_791_cast)[name = tensor<string, []>("reshape_144_cast")];
            tensor<int32, [3]> reduce_mean_108_axes_0 = const()[name = tensor<string, []>("reduce_mean_108_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_108_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_108_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_108_cast = reduce_mean(axes = reduce_mean_108_axes_0, keep_dims = reduce_mean_108_keep_dims_0, x = reshape_144_cast)[name = tensor<string, []>("reduce_mean_108_cast")];
            tensor<fp16, [2, 32, 30, 64, 64]> sub_72_cast = sub(x = reshape_144_cast, y = reduce_mean_108_cast)[name = tensor<string, []>("sub_72_cast")];
            tensor<fp16, [2, 32, 30, 64, 64]> square_36_cast = square(x = sub_72_cast)[name = tensor<string, []>("square_36_cast")];
            tensor<int32, [3]> reduce_mean_110_axes_0 = const()[name = tensor<string, []>("reduce_mean_110_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_110_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_110_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_110_cast = reduce_mean(axes = reduce_mean_110_axes_0, keep_dims = reduce_mean_110_keep_dims_0, x = square_36_cast)[name = tensor<string, []>("reduce_mean_110_cast")];
            tensor<fp16, []> add_72_y_0_to_fp16 = const()[name = tensor<string, []>("add_72_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_72_cast = add(x = reduce_mean_110_cast, y = add_72_y_0_to_fp16)[name = tensor<string, []>("add_72_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_36_cast = sqrt(x = add_72_cast)[name = tensor<string, []>("sqrt_36_cast")];
            tensor<fp16, [2, 32, 30, 64, 64]> real_div_36_cast = real_div(x = sub_72_cast, y = sqrt_36_cast)[name = tensor<string, []>("real_div_36_cast")];
            tensor<int32, [4]> reshape_145_shape_0 = const()[name = tensor<string, []>("reshape_145_shape_0"), val = tensor<int32, [4]>([2, 960, 64, 64])];
            tensor<fp16, [2, 960, 64, 64]> reshape_145_cast = reshape(shape = reshape_145_shape_0, x = real_div_36_cast)[name = tensor<string, []>("reshape_145_cast")];
            tensor<fp16, [960]> add_73_mean_0_to_fp16 = const()[name = tensor<string, []>("add_73_mean_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5042339968)))];
            tensor<fp16, [960]> add_73_variance_0_to_fp16 = const()[name = tensor<string, []>("add_73_variance_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5042341952)))];
            tensor<fp16, [960]> add_73_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_73_gamma_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5042343936)))];
            tensor<fp16, [960]> add_73_beta_0_to_fp16 = const()[name = tensor<string, []>("add_73_beta_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5042345920)))];
            tensor<fp16, []> add_73_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_73_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 960, 64, 64]> add_73_cast = batch_norm(beta = add_73_beta_0_to_fp16, epsilon = add_73_epsilon_0_to_fp16, gamma = add_73_gamma_0_to_fp16, mean = add_73_mean_0_to_fp16, variance = add_73_variance_0_to_fp16, x = reshape_145_cast)[name = tensor<string, []>("add_73_cast")];
            tensor<fp16, [2, 960, 64, 64]> input_795_cast = silu(x = add_73_cast)[name = tensor<string, []>("input_795_cast")];
            tensor<int32, [2]> var_60385 = const()[name = tensor<string, []>("op_60385"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60387 = const()[name = tensor<string, []>("op_60387"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_553_pad_type_0 = const()[name = tensor<string, []>("hidden_states_553_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_553_pad_0 = const()[name = tensor<string, []>("hidden_states_553_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 960, 3, 3]> up_blocks_1_resnets_2_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_2_conv1_weight_to_fp16"), val = tensor<fp16, [640, 960, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5042347904)))];
            tensor<fp16, [640]> up_blocks_1_resnets_2_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_2_conv1_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5053407168)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_553_cast = conv(bias = up_blocks_1_resnets_2_conv1_bias_to_fp16, dilations = var_60387, groups = var_58064, pad = hidden_states_553_pad_0, pad_type = hidden_states_553_pad_type_0, strides = var_60385, weight = up_blocks_1_resnets_2_conv1_weight_to_fp16, x = input_795_cast)[name = tensor<string, []>("hidden_states_553_cast")];
            tensor<int32, [2]> var_60393 = const()[name = tensor<string, []>("op_60393"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60395 = const()[name = tensor<string, []>("op_60395"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_27_pad_type_0 = const()[name = tensor<string, []>("temb_27_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_27_pad_0 = const()[name = tensor<string, []>("temb_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1280, 1, 1]> up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [640, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5053408512)))];
            tensor<fp16, [640]> up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5055046976)))];
            tensor<fp16, [2, 640, 1, 1]> temb_27_cast = conv(bias = up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_60395, groups = var_58064, pad = temb_27_pad_0, pad_type = temb_27_pad_type_0, strides = var_60393, weight = up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16, x = input_23_cast)[name = tensor<string, []>("temb_27_cast")];
            tensor<fp16, [2, 640, 64, 64]> input_799_cast = add(x = hidden_states_553_cast, y = temb_27_cast)[name = tensor<string, []>("input_799_cast")];
            tensor<int32, [5]> reshape_148_shape_0 = const()[name = tensor<string, []>("reshape_148_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 64, 64])];
            tensor<fp16, [2, 32, 20, 64, 64]> reshape_148_cast = reshape(shape = reshape_148_shape_0, x = input_799_cast)[name = tensor<string, []>("reshape_148_cast")];
            tensor<int32, [3]> reduce_mean_111_axes_0 = const()[name = tensor<string, []>("reduce_mean_111_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_111_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_111_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_111_cast = reduce_mean(axes = reduce_mean_111_axes_0, keep_dims = reduce_mean_111_keep_dims_0, x = reshape_148_cast)[name = tensor<string, []>("reduce_mean_111_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> sub_74_cast = sub(x = reshape_148_cast, y = reduce_mean_111_cast)[name = tensor<string, []>("sub_74_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> square_37_cast = square(x = sub_74_cast)[name = tensor<string, []>("square_37_cast")];
            tensor<int32, [3]> reduce_mean_113_axes_0 = const()[name = tensor<string, []>("reduce_mean_113_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_113_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_113_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_113_cast = reduce_mean(axes = reduce_mean_113_axes_0, keep_dims = reduce_mean_113_keep_dims_0, x = square_37_cast)[name = tensor<string, []>("reduce_mean_113_cast")];
            tensor<fp16, []> add_74_y_0_to_fp16 = const()[name = tensor<string, []>("add_74_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_74_cast = add(x = reduce_mean_113_cast, y = add_74_y_0_to_fp16)[name = tensor<string, []>("add_74_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_37_cast = sqrt(x = add_74_cast)[name = tensor<string, []>("sqrt_37_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_37_cast = real_div(x = sub_74_cast, y = sqrt_37_cast)[name = tensor<string, []>("real_div_37_cast")];
            tensor<int32, [4]> reshape_149_shape_0 = const()[name = tensor<string, []>("reshape_149_shape_0"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> reshape_149_cast = reshape(shape = reshape_149_shape_0, x = real_div_37_cast)[name = tensor<string, []>("reshape_149_cast")];
            tensor<fp16, [640]> add_75_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_75_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5055048320)))];
            tensor<fp16, [640]> add_75_beta_0_to_fp16 = const()[name = tensor<string, []>("add_75_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5055049664)))];
            tensor<fp16, []> add_75_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_75_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 64, 64]> add_75_cast = batch_norm(beta = add_75_beta_0_to_fp16, epsilon = add_75_epsilon_0_to_fp16, gamma = add_75_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_149_cast)[name = tensor<string, []>("add_75_cast")];
            tensor<fp16, [2, 640, 64, 64]> input_803_cast = silu(x = add_75_cast)[name = tensor<string, []>("input_803_cast")];
            tensor<int32, [2]> var_60405 = const()[name = tensor<string, []>("op_60405"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60407 = const()[name = tensor<string, []>("op_60407"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_555_pad_type_0 = const()[name = tensor<string, []>("hidden_states_555_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_555_pad_0 = const()[name = tensor<string, []>("hidden_states_555_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 640, 3, 3]> up_blocks_1_resnets_2_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_2_conv2_weight_to_fp16"), val = tensor<fp16, [640, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5055051008)))];
            tensor<fp16, [640]> up_blocks_1_resnets_2_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_2_conv2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5062423872)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_555_cast = conv(bias = up_blocks_1_resnets_2_conv2_bias_to_fp16, dilations = var_60407, groups = var_58064, pad = hidden_states_555_pad_0, pad_type = hidden_states_555_pad_type_0, strides = var_60405, weight = up_blocks_1_resnets_2_conv2_weight_to_fp16, x = input_803_cast)[name = tensor<string, []>("hidden_states_555_cast")];
            tensor<int32, [2]> var_60412 = const()[name = tensor<string, []>("op_60412"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60414 = const()[name = tensor<string, []>("op_60414"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_15_pad_type_0 = const()[name = tensor<string, []>("x_15_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_15_pad_0 = const()[name = tensor<string, []>("x_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 960, 1, 1]> up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [640, 960, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5062425216)))];
            tensor<fp16, [640]> up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5063654080)))];
            tensor<fp16, [2, 640, 64, 64]> x_15_cast = conv(bias = up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_60414, groups = var_58064, pad = x_15_pad_0, pad_type = x_15_pad_type_0, strides = var_60412, weight = up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16, x = input_791_cast)[name = tensor<string, []>("x_15_cast")];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_557_cast = add(x = x_15_cast, y = hidden_states_555_cast)[name = tensor<string, []>("hidden_states_557_cast")];
            tensor<int32, [5]> reshape_152_shape_0 = const()[name = tensor<string, []>("reshape_152_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 64, 64])];
            tensor<fp16, [2, 32, 20, 64, 64]> reshape_152_cast = reshape(shape = reshape_152_shape_0, x = hidden_states_557_cast)[name = tensor<string, []>("reshape_152_cast")];
            tensor<int32, [3]> reduce_mean_114_axes_0 = const()[name = tensor<string, []>("reduce_mean_114_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_114_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_114_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_114_cast = reduce_mean(axes = reduce_mean_114_axes_0, keep_dims = reduce_mean_114_keep_dims_0, x = reshape_152_cast)[name = tensor<string, []>("reduce_mean_114_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> sub_76_cast = sub(x = reshape_152_cast, y = reduce_mean_114_cast)[name = tensor<string, []>("sub_76_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> square_38_cast = square(x = sub_76_cast)[name = tensor<string, []>("square_38_cast")];
            tensor<int32, [3]> reduce_mean_116_axes_0 = const()[name = tensor<string, []>("reduce_mean_116_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_116_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_116_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_116_cast = reduce_mean(axes = reduce_mean_116_axes_0, keep_dims = reduce_mean_116_keep_dims_0, x = square_38_cast)[name = tensor<string, []>("reduce_mean_116_cast")];
            tensor<fp16, []> add_76_y_0_to_fp16 = const()[name = tensor<string, []>("add_76_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_76_cast = add(x = reduce_mean_116_cast, y = add_76_y_0_to_fp16)[name = tensor<string, []>("add_76_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_38_cast = sqrt(x = add_76_cast)[name = tensor<string, []>("sqrt_38_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_38_cast = real_div(x = sub_76_cast, y = sqrt_38_cast)[name = tensor<string, []>("real_div_38_cast")];
            tensor<int32, [4]> reshape_153_shape_0 = const()[name = tensor<string, []>("reshape_153_shape_0"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> reshape_153_cast = reshape(shape = reshape_153_shape_0, x = real_div_38_cast)[name = tensor<string, []>("reshape_153_cast")];
            tensor<fp16, [640]> add_77_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_77_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5063655424)))];
            tensor<fp16, [640]> add_77_beta_0_to_fp16 = const()[name = tensor<string, []>("add_77_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5063656768)))];
            tensor<fp16, []> add_77_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_77_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 64, 64]> add_77_cast = batch_norm(beta = add_77_beta_0_to_fp16, epsilon = add_77_epsilon_0_to_fp16, gamma = add_77_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_153_cast)[name = tensor<string, []>("add_77_cast")];
            tensor<int32, [2]> var_60436 = const()[name = tensor<string, []>("op_60436"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60438 = const()[name = tensor<string, []>("op_60438"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_559_pad_type_0 = const()[name = tensor<string, []>("hidden_states_559_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_559_pad_0 = const()[name = tensor<string, []>("hidden_states_559_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_proj_in_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5063658112)))];
            tensor<fp16, [640]> up_blocks_1_attentions_2_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_proj_in_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5064477376)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_559_cast = conv(bias = up_blocks_1_attentions_2_proj_in_bias_to_fp16, dilations = var_60438, groups = var_58064, pad = hidden_states_559_pad_0, pad_type = hidden_states_559_pad_type_0, strides = var_60436, weight = up_blocks_1_attentions_2_proj_in_weight_to_fp16, x = add_77_cast)[name = tensor<string, []>("hidden_states_559_cast")];
            tensor<int32, [4]> var_60443 = const()[name = tensor<string, []>("op_60443"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<fp16, [2, 640, 1, 4096]> inputs_409_cast = reshape(shape = var_60443, x = hidden_states_559_cast)[name = tensor<string, []>("inputs_409_cast")];
            tensor<int32, [1]> var_60453 = const()[name = tensor<string, []>("op_60453"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_409_cast = reduce_mean(axes = var_60453, keep_dims = var_58059, x = inputs_409_cast)[name = tensor<string, []>("channels_mean_409_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_409_cast = sub(x = inputs_409_cast, y = channels_mean_409_cast)[name = tensor<string, []>("zero_mean_409_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_409_cast = mul(x = zero_mean_409_cast, y = zero_mean_409_cast)[name = tensor<string, []>("zero_mean_sq_409_cast")];
            tensor<int32, [1]> var_60457 = const()[name = tensor<string, []>("op_60457"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_60458_cast = reduce_mean(axes = var_60457, keep_dims = var_58059, x = zero_mean_sq_409_cast)[name = tensor<string, []>("op_60458_cast")];
            tensor<fp16, []> var_60459_to_fp16 = const()[name = tensor<string, []>("op_60459_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_60460_cast = add(x = var_60458_cast, y = var_60459_to_fp16)[name = tensor<string, []>("op_60460_cast")];
            tensor<fp16, []> denom_409_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_409_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_409_cast = rsqrt(epsilon = denom_409_epsilon_0_to_fp16, x = var_60460_cast)[name = tensor<string, []>("denom_409_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_409_cast = mul(x = zero_mean_409_cast, y = denom_409_cast)[name = tensor<string, []>("out_409_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_60464_to_fp16 = const()[name = tensor<string, []>("op_60464_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5064478720)))];
            tensor<fp16, [2, 640, 1, 4096]> var_60465_cast = add(x = out_409_cast, y = var_60464_to_fp16)[name = tensor<string, []>("op_60465_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_60467_to_fp16 = const()[name = tensor<string, []>("op_60467_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5064480064)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_561_cast = mul(x = var_60465_cast, y = var_60467_to_fp16)[name = tensor<string, []>("hidden_states_561_cast")];
            tensor<int32, [2]> var_60474 = const()[name = tensor<string, []>("op_60474"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60476 = const()[name = tensor<string, []>("op_60476"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_273_pad_type_0 = const()[name = tensor<string, []>("q_273_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_273_pad_0 = const()[name = tensor<string, []>("q_273_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5064481408)))];
            tensor<fp16, [2, 640, 1, 4096]> q_273_cast = conv(dilations = var_60476, groups = var_58064, pad = q_273_pad_0, pad_type = q_273_pad_type_0, strides = var_60474, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_561_cast)[name = tensor<string, []>("q_273_cast")];
            tensor<int32, [2]> var_60480 = const()[name = tensor<string, []>("op_60480"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60482 = const()[name = tensor<string, []>("op_60482"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_545_pad_type_0 = const()[name = tensor<string, []>("k_545_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_545_pad_0 = const()[name = tensor<string, []>("k_545_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5065300672)))];
            tensor<fp16, [2, 640, 1, 4096]> k_545_cast = conv(dilations = var_60482, groups = var_58064, pad = k_545_pad_0, pad_type = k_545_pad_type_0, strides = var_60480, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_561_cast)[name = tensor<string, []>("k_545_cast")];
            tensor<int32, [2]> var_60486 = const()[name = tensor<string, []>("op_60486"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60488 = const()[name = tensor<string, []>("op_60488"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_273_pad_type_0 = const()[name = tensor<string, []>("v_273_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_273_pad_0 = const()[name = tensor<string, []>("v_273_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5066119936)))];
            tensor<fp16, [2, 640, 1, 4096]> v_273_cast = conv(dilations = var_60488, groups = var_58064, pad = v_273_pad_0, pad_type = v_273_pad_type_0, strides = var_60486, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_561_cast)[name = tensor<string, []>("v_273_cast")];
            tensor<int32, [4]> var_60492_begin_0 = const()[name = tensor<string, []>("op_60492_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_60492_end_0 = const()[name = tensor<string, []>("op_60492_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_60492_end_mask_0 = const()[name = tensor<string, []>("op_60492_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60492_cast = slice_by_index(begin = var_60492_begin_0, end = var_60492_end_0, end_mask = var_60492_end_mask_0, x = q_273_cast)[name = tensor<string, []>("op_60492_cast")];
            tensor<int32, [4]> var_60496_begin_0 = const()[name = tensor<string, []>("op_60496_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_60496_end_0 = const()[name = tensor<string, []>("op_60496_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_60496_end_mask_0 = const()[name = tensor<string, []>("op_60496_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60496_cast = slice_by_index(begin = var_60496_begin_0, end = var_60496_end_0, end_mask = var_60496_end_mask_0, x = q_273_cast)[name = tensor<string, []>("op_60496_cast")];
            tensor<int32, [4]> var_60500_begin_0 = const()[name = tensor<string, []>("op_60500_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_60500_end_0 = const()[name = tensor<string, []>("op_60500_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_60500_end_mask_0 = const()[name = tensor<string, []>("op_60500_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60500_cast = slice_by_index(begin = var_60500_begin_0, end = var_60500_end_0, end_mask = var_60500_end_mask_0, x = q_273_cast)[name = tensor<string, []>("op_60500_cast")];
            tensor<int32, [4]> var_60504_begin_0 = const()[name = tensor<string, []>("op_60504_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_60504_end_0 = const()[name = tensor<string, []>("op_60504_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_60504_end_mask_0 = const()[name = tensor<string, []>("op_60504_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60504_cast = slice_by_index(begin = var_60504_begin_0, end = var_60504_end_0, end_mask = var_60504_end_mask_0, x = q_273_cast)[name = tensor<string, []>("op_60504_cast")];
            tensor<int32, [4]> var_60508_begin_0 = const()[name = tensor<string, []>("op_60508_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_60508_end_0 = const()[name = tensor<string, []>("op_60508_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_60508_end_mask_0 = const()[name = tensor<string, []>("op_60508_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60508_cast = slice_by_index(begin = var_60508_begin_0, end = var_60508_end_0, end_mask = var_60508_end_mask_0, x = q_273_cast)[name = tensor<string, []>("op_60508_cast")];
            tensor<int32, [4]> var_60512_begin_0 = const()[name = tensor<string, []>("op_60512_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_60512_end_0 = const()[name = tensor<string, []>("op_60512_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_60512_end_mask_0 = const()[name = tensor<string, []>("op_60512_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60512_cast = slice_by_index(begin = var_60512_begin_0, end = var_60512_end_0, end_mask = var_60512_end_mask_0, x = q_273_cast)[name = tensor<string, []>("op_60512_cast")];
            tensor<int32, [4]> var_60516_begin_0 = const()[name = tensor<string, []>("op_60516_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_60516_end_0 = const()[name = tensor<string, []>("op_60516_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_60516_end_mask_0 = const()[name = tensor<string, []>("op_60516_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60516_cast = slice_by_index(begin = var_60516_begin_0, end = var_60516_end_0, end_mask = var_60516_end_mask_0, x = q_273_cast)[name = tensor<string, []>("op_60516_cast")];
            tensor<int32, [4]> var_60520_begin_0 = const()[name = tensor<string, []>("op_60520_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_60520_end_0 = const()[name = tensor<string, []>("op_60520_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_60520_end_mask_0 = const()[name = tensor<string, []>("op_60520_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60520_cast = slice_by_index(begin = var_60520_begin_0, end = var_60520_end_0, end_mask = var_60520_end_mask_0, x = q_273_cast)[name = tensor<string, []>("op_60520_cast")];
            tensor<int32, [4]> var_60524_begin_0 = const()[name = tensor<string, []>("op_60524_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_60524_end_0 = const()[name = tensor<string, []>("op_60524_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_60524_end_mask_0 = const()[name = tensor<string, []>("op_60524_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60524_cast = slice_by_index(begin = var_60524_begin_0, end = var_60524_end_0, end_mask = var_60524_end_mask_0, x = q_273_cast)[name = tensor<string, []>("op_60524_cast")];
            tensor<int32, [4]> var_60528_begin_0 = const()[name = tensor<string, []>("op_60528_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_60528_end_0 = const()[name = tensor<string, []>("op_60528_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_60528_end_mask_0 = const()[name = tensor<string, []>("op_60528_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60528_cast = slice_by_index(begin = var_60528_begin_0, end = var_60528_end_0, end_mask = var_60528_end_mask_0, x = q_273_cast)[name = tensor<string, []>("op_60528_cast")];
            tensor<int32, [4]> k_547_perm_0 = const()[name = tensor<string, []>("k_547_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_60535_begin_0 = const()[name = tensor<string, []>("op_60535_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_60535_end_0 = const()[name = tensor<string, []>("op_60535_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_60535_end_mask_0 = const()[name = tensor<string, []>("op_60535_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 640]> transpose_3 = transpose(perm = k_547_perm_0, x = k_545_cast)[name = tensor<string, []>("transpose_3")];
            tensor<fp16, [2, 4096, 1, 64]> var_60535_cast = slice_by_index(begin = var_60535_begin_0, end = var_60535_end_0, end_mask = var_60535_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_60535_cast")];
            tensor<int32, [4]> var_60539_begin_0 = const()[name = tensor<string, []>("op_60539_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_60539_end_0 = const()[name = tensor<string, []>("op_60539_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_60539_end_mask_0 = const()[name = tensor<string, []>("op_60539_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_60539_cast = slice_by_index(begin = var_60539_begin_0, end = var_60539_end_0, end_mask = var_60539_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_60539_cast")];
            tensor<int32, [4]> var_60543_begin_0 = const()[name = tensor<string, []>("op_60543_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_60543_end_0 = const()[name = tensor<string, []>("op_60543_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_60543_end_mask_0 = const()[name = tensor<string, []>("op_60543_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_60543_cast = slice_by_index(begin = var_60543_begin_0, end = var_60543_end_0, end_mask = var_60543_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_60543_cast")];
            tensor<int32, [4]> var_60547_begin_0 = const()[name = tensor<string, []>("op_60547_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_60547_end_0 = const()[name = tensor<string, []>("op_60547_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_60547_end_mask_0 = const()[name = tensor<string, []>("op_60547_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_60547_cast = slice_by_index(begin = var_60547_begin_0, end = var_60547_end_0, end_mask = var_60547_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_60547_cast")];
            tensor<int32, [4]> var_60551_begin_0 = const()[name = tensor<string, []>("op_60551_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_60551_end_0 = const()[name = tensor<string, []>("op_60551_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_60551_end_mask_0 = const()[name = tensor<string, []>("op_60551_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_60551_cast = slice_by_index(begin = var_60551_begin_0, end = var_60551_end_0, end_mask = var_60551_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_60551_cast")];
            tensor<int32, [4]> var_60555_begin_0 = const()[name = tensor<string, []>("op_60555_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_60555_end_0 = const()[name = tensor<string, []>("op_60555_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 384])];
            tensor<bool, [4]> var_60555_end_mask_0 = const()[name = tensor<string, []>("op_60555_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_60555_cast = slice_by_index(begin = var_60555_begin_0, end = var_60555_end_0, end_mask = var_60555_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_60555_cast")];
            tensor<int32, [4]> var_60559_begin_0 = const()[name = tensor<string, []>("op_60559_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_60559_end_0 = const()[name = tensor<string, []>("op_60559_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 448])];
            tensor<bool, [4]> var_60559_end_mask_0 = const()[name = tensor<string, []>("op_60559_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_60559_cast = slice_by_index(begin = var_60559_begin_0, end = var_60559_end_0, end_mask = var_60559_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_60559_cast")];
            tensor<int32, [4]> var_60563_begin_0 = const()[name = tensor<string, []>("op_60563_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_60563_end_0 = const()[name = tensor<string, []>("op_60563_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 512])];
            tensor<bool, [4]> var_60563_end_mask_0 = const()[name = tensor<string, []>("op_60563_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_60563_cast = slice_by_index(begin = var_60563_begin_0, end = var_60563_end_0, end_mask = var_60563_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_60563_cast")];
            tensor<int32, [4]> var_60567_begin_0 = const()[name = tensor<string, []>("op_60567_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_60567_end_0 = const()[name = tensor<string, []>("op_60567_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 576])];
            tensor<bool, [4]> var_60567_end_mask_0 = const()[name = tensor<string, []>("op_60567_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_60567_cast = slice_by_index(begin = var_60567_begin_0, end = var_60567_end_0, end_mask = var_60567_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_60567_cast")];
            tensor<int32, [4]> var_60571_begin_0 = const()[name = tensor<string, []>("op_60571_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_60571_end_0 = const()[name = tensor<string, []>("op_60571_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 640])];
            tensor<bool, [4]> var_60571_end_mask_0 = const()[name = tensor<string, []>("op_60571_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_60571_cast = slice_by_index(begin = var_60571_begin_0, end = var_60571_end_0, end_mask = var_60571_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_60571_cast")];
            tensor<int32, [4]> var_60573_begin_0 = const()[name = tensor<string, []>("op_60573_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_60573_end_0 = const()[name = tensor<string, []>("op_60573_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_60573_end_mask_0 = const()[name = tensor<string, []>("op_60573_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60573_cast = slice_by_index(begin = var_60573_begin_0, end = var_60573_end_0, end_mask = var_60573_end_mask_0, x = v_273_cast)[name = tensor<string, []>("op_60573_cast")];
            tensor<int32, [4]> var_60577_begin_0 = const()[name = tensor<string, []>("op_60577_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_60577_end_0 = const()[name = tensor<string, []>("op_60577_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_60577_end_mask_0 = const()[name = tensor<string, []>("op_60577_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60577_cast = slice_by_index(begin = var_60577_begin_0, end = var_60577_end_0, end_mask = var_60577_end_mask_0, x = v_273_cast)[name = tensor<string, []>("op_60577_cast")];
            tensor<int32, [4]> var_60581_begin_0 = const()[name = tensor<string, []>("op_60581_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_60581_end_0 = const()[name = tensor<string, []>("op_60581_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_60581_end_mask_0 = const()[name = tensor<string, []>("op_60581_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60581_cast = slice_by_index(begin = var_60581_begin_0, end = var_60581_end_0, end_mask = var_60581_end_mask_0, x = v_273_cast)[name = tensor<string, []>("op_60581_cast")];
            tensor<int32, [4]> var_60585_begin_0 = const()[name = tensor<string, []>("op_60585_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_60585_end_0 = const()[name = tensor<string, []>("op_60585_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_60585_end_mask_0 = const()[name = tensor<string, []>("op_60585_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60585_cast = slice_by_index(begin = var_60585_begin_0, end = var_60585_end_0, end_mask = var_60585_end_mask_0, x = v_273_cast)[name = tensor<string, []>("op_60585_cast")];
            tensor<int32, [4]> var_60589_begin_0 = const()[name = tensor<string, []>("op_60589_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_60589_end_0 = const()[name = tensor<string, []>("op_60589_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_60589_end_mask_0 = const()[name = tensor<string, []>("op_60589_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60589_cast = slice_by_index(begin = var_60589_begin_0, end = var_60589_end_0, end_mask = var_60589_end_mask_0, x = v_273_cast)[name = tensor<string, []>("op_60589_cast")];
            tensor<int32, [4]> var_60593_begin_0 = const()[name = tensor<string, []>("op_60593_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_60593_end_0 = const()[name = tensor<string, []>("op_60593_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_60593_end_mask_0 = const()[name = tensor<string, []>("op_60593_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60593_cast = slice_by_index(begin = var_60593_begin_0, end = var_60593_end_0, end_mask = var_60593_end_mask_0, x = v_273_cast)[name = tensor<string, []>("op_60593_cast")];
            tensor<int32, [4]> var_60597_begin_0 = const()[name = tensor<string, []>("op_60597_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_60597_end_0 = const()[name = tensor<string, []>("op_60597_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_60597_end_mask_0 = const()[name = tensor<string, []>("op_60597_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60597_cast = slice_by_index(begin = var_60597_begin_0, end = var_60597_end_0, end_mask = var_60597_end_mask_0, x = v_273_cast)[name = tensor<string, []>("op_60597_cast")];
            tensor<int32, [4]> var_60601_begin_0 = const()[name = tensor<string, []>("op_60601_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_60601_end_0 = const()[name = tensor<string, []>("op_60601_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_60601_end_mask_0 = const()[name = tensor<string, []>("op_60601_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60601_cast = slice_by_index(begin = var_60601_begin_0, end = var_60601_end_0, end_mask = var_60601_end_mask_0, x = v_273_cast)[name = tensor<string, []>("op_60601_cast")];
            tensor<int32, [4]> var_60605_begin_0 = const()[name = tensor<string, []>("op_60605_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_60605_end_0 = const()[name = tensor<string, []>("op_60605_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_60605_end_mask_0 = const()[name = tensor<string, []>("op_60605_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60605_cast = slice_by_index(begin = var_60605_begin_0, end = var_60605_end_0, end_mask = var_60605_end_mask_0, x = v_273_cast)[name = tensor<string, []>("op_60605_cast")];
            tensor<int32, [4]> var_60609_begin_0 = const()[name = tensor<string, []>("op_60609_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_60609_end_0 = const()[name = tensor<string, []>("op_60609_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_60609_end_mask_0 = const()[name = tensor<string, []>("op_60609_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60609_cast = slice_by_index(begin = var_60609_begin_0, end = var_60609_end_0, end_mask = var_60609_end_mask_0, x = v_273_cast)[name = tensor<string, []>("op_60609_cast")];
            tensor<string, []> var_60613_equation_0 = const()[name = tensor<string, []>("op_60613_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60613_cast = einsum(equation = var_60613_equation_0, values = (var_60535_cast, var_60492_cast))[name = tensor<string, []>("op_60613_cast")];
            tensor<fp16, []> var_60614_to_fp16 = const()[name = tensor<string, []>("op_60614_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5121_cast = mul(x = var_60613_cast, y = var_60614_to_fp16)[name = tensor<string, []>("aw_5121_cast")];
            tensor<string, []> var_60617_equation_0 = const()[name = tensor<string, []>("op_60617_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60617_cast = einsum(equation = var_60617_equation_0, values = (var_60539_cast, var_60496_cast))[name = tensor<string, []>("op_60617_cast")];
            tensor<fp16, []> var_60618_to_fp16 = const()[name = tensor<string, []>("op_60618_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5123_cast = mul(x = var_60617_cast, y = var_60618_to_fp16)[name = tensor<string, []>("aw_5123_cast")];
            tensor<string, []> var_60621_equation_0 = const()[name = tensor<string, []>("op_60621_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60621_cast = einsum(equation = var_60621_equation_0, values = (var_60543_cast, var_60500_cast))[name = tensor<string, []>("op_60621_cast")];
            tensor<fp16, []> var_60622_to_fp16 = const()[name = tensor<string, []>("op_60622_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5125_cast = mul(x = var_60621_cast, y = var_60622_to_fp16)[name = tensor<string, []>("aw_5125_cast")];
            tensor<string, []> var_60625_equation_0 = const()[name = tensor<string, []>("op_60625_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60625_cast = einsum(equation = var_60625_equation_0, values = (var_60547_cast, var_60504_cast))[name = tensor<string, []>("op_60625_cast")];
            tensor<fp16, []> var_60626_to_fp16 = const()[name = tensor<string, []>("op_60626_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5127_cast = mul(x = var_60625_cast, y = var_60626_to_fp16)[name = tensor<string, []>("aw_5127_cast")];
            tensor<string, []> var_60629_equation_0 = const()[name = tensor<string, []>("op_60629_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60629_cast = einsum(equation = var_60629_equation_0, values = (var_60551_cast, var_60508_cast))[name = tensor<string, []>("op_60629_cast")];
            tensor<fp16, []> var_60630_to_fp16 = const()[name = tensor<string, []>("op_60630_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5129_cast = mul(x = var_60629_cast, y = var_60630_to_fp16)[name = tensor<string, []>("aw_5129_cast")];
            tensor<string, []> var_60633_equation_0 = const()[name = tensor<string, []>("op_60633_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60633_cast = einsum(equation = var_60633_equation_0, values = (var_60555_cast, var_60512_cast))[name = tensor<string, []>("op_60633_cast")];
            tensor<fp16, []> var_60634_to_fp16 = const()[name = tensor<string, []>("op_60634_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5131_cast = mul(x = var_60633_cast, y = var_60634_to_fp16)[name = tensor<string, []>("aw_5131_cast")];
            tensor<string, []> var_60637_equation_0 = const()[name = tensor<string, []>("op_60637_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60637_cast = einsum(equation = var_60637_equation_0, values = (var_60559_cast, var_60516_cast))[name = tensor<string, []>("op_60637_cast")];
            tensor<fp16, []> var_60638_to_fp16 = const()[name = tensor<string, []>("op_60638_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5133_cast = mul(x = var_60637_cast, y = var_60638_to_fp16)[name = tensor<string, []>("aw_5133_cast")];
            tensor<string, []> var_60641_equation_0 = const()[name = tensor<string, []>("op_60641_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60641_cast = einsum(equation = var_60641_equation_0, values = (var_60563_cast, var_60520_cast))[name = tensor<string, []>("op_60641_cast")];
            tensor<fp16, []> var_60642_to_fp16 = const()[name = tensor<string, []>("op_60642_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5135_cast = mul(x = var_60641_cast, y = var_60642_to_fp16)[name = tensor<string, []>("aw_5135_cast")];
            tensor<string, []> var_60645_equation_0 = const()[name = tensor<string, []>("op_60645_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60645_cast = einsum(equation = var_60645_equation_0, values = (var_60567_cast, var_60524_cast))[name = tensor<string, []>("op_60645_cast")];
            tensor<fp16, []> var_60646_to_fp16 = const()[name = tensor<string, []>("op_60646_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5137_cast = mul(x = var_60645_cast, y = var_60646_to_fp16)[name = tensor<string, []>("aw_5137_cast")];
            tensor<string, []> var_60649_equation_0 = const()[name = tensor<string, []>("op_60649_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60649_cast = einsum(equation = var_60649_equation_0, values = (var_60571_cast, var_60528_cast))[name = tensor<string, []>("op_60649_cast")];
            tensor<fp16, []> var_60650_to_fp16 = const()[name = tensor<string, []>("op_60650_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5139_cast = mul(x = var_60649_cast, y = var_60650_to_fp16)[name = tensor<string, []>("aw_5139_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60652_cast = softmax(axis = var_58064, x = aw_5121_cast)[name = tensor<string, []>("op_60652_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60653_cast = softmax(axis = var_58064, x = aw_5123_cast)[name = tensor<string, []>("op_60653_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60654_cast = softmax(axis = var_58064, x = aw_5125_cast)[name = tensor<string, []>("op_60654_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60655_cast = softmax(axis = var_58064, x = aw_5127_cast)[name = tensor<string, []>("op_60655_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60656_cast = softmax(axis = var_58064, x = aw_5129_cast)[name = tensor<string, []>("op_60656_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60657_cast = softmax(axis = var_58064, x = aw_5131_cast)[name = tensor<string, []>("op_60657_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60658_cast = softmax(axis = var_58064, x = aw_5133_cast)[name = tensor<string, []>("op_60658_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60659_cast = softmax(axis = var_58064, x = aw_5135_cast)[name = tensor<string, []>("op_60659_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60660_cast = softmax(axis = var_58064, x = aw_5137_cast)[name = tensor<string, []>("op_60660_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_60661_cast = softmax(axis = var_58064, x = aw_5139_cast)[name = tensor<string, []>("op_60661_cast")];
            tensor<string, []> var_60663_equation_0 = const()[name = tensor<string, []>("op_60663_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60663_cast = einsum(equation = var_60663_equation_0, values = (var_60573_cast, var_60652_cast))[name = tensor<string, []>("op_60663_cast")];
            tensor<string, []> var_60665_equation_0 = const()[name = tensor<string, []>("op_60665_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60665_cast = einsum(equation = var_60665_equation_0, values = (var_60577_cast, var_60653_cast))[name = tensor<string, []>("op_60665_cast")];
            tensor<string, []> var_60667_equation_0 = const()[name = tensor<string, []>("op_60667_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60667_cast = einsum(equation = var_60667_equation_0, values = (var_60581_cast, var_60654_cast))[name = tensor<string, []>("op_60667_cast")];
            tensor<string, []> var_60669_equation_0 = const()[name = tensor<string, []>("op_60669_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60669_cast = einsum(equation = var_60669_equation_0, values = (var_60585_cast, var_60655_cast))[name = tensor<string, []>("op_60669_cast")];
            tensor<string, []> var_60671_equation_0 = const()[name = tensor<string, []>("op_60671_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60671_cast = einsum(equation = var_60671_equation_0, values = (var_60589_cast, var_60656_cast))[name = tensor<string, []>("op_60671_cast")];
            tensor<string, []> var_60673_equation_0 = const()[name = tensor<string, []>("op_60673_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60673_cast = einsum(equation = var_60673_equation_0, values = (var_60593_cast, var_60657_cast))[name = tensor<string, []>("op_60673_cast")];
            tensor<string, []> var_60675_equation_0 = const()[name = tensor<string, []>("op_60675_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60675_cast = einsum(equation = var_60675_equation_0, values = (var_60597_cast, var_60658_cast))[name = tensor<string, []>("op_60675_cast")];
            tensor<string, []> var_60677_equation_0 = const()[name = tensor<string, []>("op_60677_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60677_cast = einsum(equation = var_60677_equation_0, values = (var_60601_cast, var_60659_cast))[name = tensor<string, []>("op_60677_cast")];
            tensor<string, []> var_60679_equation_0 = const()[name = tensor<string, []>("op_60679_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60679_cast = einsum(equation = var_60679_equation_0, values = (var_60605_cast, var_60660_cast))[name = tensor<string, []>("op_60679_cast")];
            tensor<string, []> var_60681_equation_0 = const()[name = tensor<string, []>("op_60681_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60681_cast = einsum(equation = var_60681_equation_0, values = (var_60609_cast, var_60661_cast))[name = tensor<string, []>("op_60681_cast")];
            tensor<bool, []> input_807_interleave_0 = const()[name = tensor<string, []>("input_807_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_807_cast = concat(axis = var_58064, interleave = input_807_interleave_0, values = (var_60663_cast, var_60665_cast, var_60667_cast, var_60669_cast, var_60671_cast, var_60673_cast, var_60675_cast, var_60677_cast, var_60679_cast, var_60681_cast))[name = tensor<string, []>("input_807_cast")];
            tensor<int32, [2]> var_60687 = const()[name = tensor<string, []>("op_60687"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60689 = const()[name = tensor<string, []>("op_60689"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_60691_pad_type_0 = const()[name = tensor<string, []>("op_60691_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_60691_pad_0 = const()[name = tensor<string, []>("op_60691_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5066939200)))];
            tensor<fp16, [640]> up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5067758464)))];
            tensor<fp16, [2, 640, 1, 4096]> var_60691_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_60689, groups = var_58064, pad = var_60691_pad_0, pad_type = var_60691_pad_type_0, strides = var_60687, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_807_cast)[name = tensor<string, []>("op_60691_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_411_cast = add(x = var_60691_cast, y = inputs_409_cast)[name = tensor<string, []>("inputs_411_cast")];
            tensor<int32, [1]> var_60695 = const()[name = tensor<string, []>("op_60695"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_411_cast = reduce_mean(axes = var_60695, keep_dims = var_58059, x = inputs_411_cast)[name = tensor<string, []>("channels_mean_411_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_411_cast = sub(x = inputs_411_cast, y = channels_mean_411_cast)[name = tensor<string, []>("zero_mean_411_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_411_cast = mul(x = zero_mean_411_cast, y = zero_mean_411_cast)[name = tensor<string, []>("zero_mean_sq_411_cast")];
            tensor<int32, [1]> var_60699 = const()[name = tensor<string, []>("op_60699"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_60700_cast = reduce_mean(axes = var_60699, keep_dims = var_58059, x = zero_mean_sq_411_cast)[name = tensor<string, []>("op_60700_cast")];
            tensor<fp16, []> var_60701_to_fp16 = const()[name = tensor<string, []>("op_60701_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_60702_cast = add(x = var_60700_cast, y = var_60701_to_fp16)[name = tensor<string, []>("op_60702_cast")];
            tensor<fp16, []> denom_411_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_411_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_411_cast = rsqrt(epsilon = denom_411_epsilon_0_to_fp16, x = var_60702_cast)[name = tensor<string, []>("denom_411_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_411_cast = mul(x = zero_mean_411_cast, y = denom_411_cast)[name = tensor<string, []>("out_411_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_60706_to_fp16 = const()[name = tensor<string, []>("op_60706_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5067759808)))];
            tensor<fp16, [2, 640, 1, 4096]> var_60707_cast = add(x = out_411_cast, y = var_60706_to_fp16)[name = tensor<string, []>("op_60707_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_60709_to_fp16 = const()[name = tensor<string, []>("op_60709_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5067761152)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_563_cast = mul(x = var_60707_cast, y = var_60709_to_fp16)[name = tensor<string, []>("hidden_states_563_cast")];
            tensor<int32, [2]> var_60716 = const()[name = tensor<string, []>("op_60716"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60718 = const()[name = tensor<string, []>("op_60718"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_275_pad_type_0 = const()[name = tensor<string, []>("q_275_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_275_pad_0 = const()[name = tensor<string, []>("q_275_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5067762496)))];
            tensor<fp16, [2, 640, 1, 4096]> q_275_cast = conv(dilations = var_60718, groups = var_58064, pad = q_275_pad_0, pad_type = q_275_pad_type_0, strides = var_60716, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_563_cast)[name = tensor<string, []>("q_275_cast")];
            tensor<int32, [2]> var_60722 = const()[name = tensor<string, []>("op_60722"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60724 = const()[name = tensor<string, []>("op_60724"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_549_pad_type_0 = const()[name = tensor<string, []>("k_549_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_549_pad_0 = const()[name = tensor<string, []>("k_549_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5068581760)))];
            tensor<fp16, [2, 640, 1, 77]> k_549_cast = conv(dilations = var_60724, groups = var_58064, pad = k_549_pad_0, pad_type = k_549_pad_type_0, strides = var_60722, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_549_cast")];
            tensor<int32, [2]> var_60728 = const()[name = tensor<string, []>("op_60728"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60730 = const()[name = tensor<string, []>("op_60730"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_275_pad_type_0 = const()[name = tensor<string, []>("v_275_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_275_pad_0 = const()[name = tensor<string, []>("v_275_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5071203264)))];
            tensor<fp16, [2, 640, 1, 77]> v_275_cast = conv(dilations = var_60730, groups = var_58064, pad = v_275_pad_0, pad_type = v_275_pad_type_0, strides = var_60728, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_275_cast")];
            tensor<int32, [4]> var_60734_begin_0 = const()[name = tensor<string, []>("op_60734_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_60734_end_0 = const()[name = tensor<string, []>("op_60734_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_60734_end_mask_0 = const()[name = tensor<string, []>("op_60734_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60734_cast = slice_by_index(begin = var_60734_begin_0, end = var_60734_end_0, end_mask = var_60734_end_mask_0, x = q_275_cast)[name = tensor<string, []>("op_60734_cast")];
            tensor<int32, [4]> var_60738_begin_0 = const()[name = tensor<string, []>("op_60738_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_60738_end_0 = const()[name = tensor<string, []>("op_60738_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_60738_end_mask_0 = const()[name = tensor<string, []>("op_60738_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60738_cast = slice_by_index(begin = var_60738_begin_0, end = var_60738_end_0, end_mask = var_60738_end_mask_0, x = q_275_cast)[name = tensor<string, []>("op_60738_cast")];
            tensor<int32, [4]> var_60742_begin_0 = const()[name = tensor<string, []>("op_60742_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_60742_end_0 = const()[name = tensor<string, []>("op_60742_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_60742_end_mask_0 = const()[name = tensor<string, []>("op_60742_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60742_cast = slice_by_index(begin = var_60742_begin_0, end = var_60742_end_0, end_mask = var_60742_end_mask_0, x = q_275_cast)[name = tensor<string, []>("op_60742_cast")];
            tensor<int32, [4]> var_60746_begin_0 = const()[name = tensor<string, []>("op_60746_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_60746_end_0 = const()[name = tensor<string, []>("op_60746_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_60746_end_mask_0 = const()[name = tensor<string, []>("op_60746_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60746_cast = slice_by_index(begin = var_60746_begin_0, end = var_60746_end_0, end_mask = var_60746_end_mask_0, x = q_275_cast)[name = tensor<string, []>("op_60746_cast")];
            tensor<int32, [4]> var_60750_begin_0 = const()[name = tensor<string, []>("op_60750_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_60750_end_0 = const()[name = tensor<string, []>("op_60750_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_60750_end_mask_0 = const()[name = tensor<string, []>("op_60750_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60750_cast = slice_by_index(begin = var_60750_begin_0, end = var_60750_end_0, end_mask = var_60750_end_mask_0, x = q_275_cast)[name = tensor<string, []>("op_60750_cast")];
            tensor<int32, [4]> var_60754_begin_0 = const()[name = tensor<string, []>("op_60754_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_60754_end_0 = const()[name = tensor<string, []>("op_60754_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_60754_end_mask_0 = const()[name = tensor<string, []>("op_60754_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60754_cast = slice_by_index(begin = var_60754_begin_0, end = var_60754_end_0, end_mask = var_60754_end_mask_0, x = q_275_cast)[name = tensor<string, []>("op_60754_cast")];
            tensor<int32, [4]> var_60758_begin_0 = const()[name = tensor<string, []>("op_60758_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_60758_end_0 = const()[name = tensor<string, []>("op_60758_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_60758_end_mask_0 = const()[name = tensor<string, []>("op_60758_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60758_cast = slice_by_index(begin = var_60758_begin_0, end = var_60758_end_0, end_mask = var_60758_end_mask_0, x = q_275_cast)[name = tensor<string, []>("op_60758_cast")];
            tensor<int32, [4]> var_60762_begin_0 = const()[name = tensor<string, []>("op_60762_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_60762_end_0 = const()[name = tensor<string, []>("op_60762_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_60762_end_mask_0 = const()[name = tensor<string, []>("op_60762_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60762_cast = slice_by_index(begin = var_60762_begin_0, end = var_60762_end_0, end_mask = var_60762_end_mask_0, x = q_275_cast)[name = tensor<string, []>("op_60762_cast")];
            tensor<int32, [4]> var_60766_begin_0 = const()[name = tensor<string, []>("op_60766_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_60766_end_0 = const()[name = tensor<string, []>("op_60766_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_60766_end_mask_0 = const()[name = tensor<string, []>("op_60766_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60766_cast = slice_by_index(begin = var_60766_begin_0, end = var_60766_end_0, end_mask = var_60766_end_mask_0, x = q_275_cast)[name = tensor<string, []>("op_60766_cast")];
            tensor<int32, [4]> var_60770_begin_0 = const()[name = tensor<string, []>("op_60770_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_60770_end_0 = const()[name = tensor<string, []>("op_60770_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_60770_end_mask_0 = const()[name = tensor<string, []>("op_60770_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_60770_cast = slice_by_index(begin = var_60770_begin_0, end = var_60770_end_0, end_mask = var_60770_end_mask_0, x = q_275_cast)[name = tensor<string, []>("op_60770_cast")];
            tensor<int32, [4]> k_551_perm_0 = const()[name = tensor<string, []>("k_551_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_60777_begin_0 = const()[name = tensor<string, []>("op_60777_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_60777_end_0 = const()[name = tensor<string, []>("op_60777_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_60777_end_mask_0 = const()[name = tensor<string, []>("op_60777_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> transpose_2 = transpose(perm = k_551_perm_0, x = k_549_cast)[name = tensor<string, []>("transpose_2")];
            tensor<fp16, [2, 77, 1, 64]> var_60777_cast = slice_by_index(begin = var_60777_begin_0, end = var_60777_end_0, end_mask = var_60777_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_60777_cast")];
            tensor<int32, [4]> var_60781_begin_0 = const()[name = tensor<string, []>("op_60781_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_60781_end_0 = const()[name = tensor<string, []>("op_60781_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_60781_end_mask_0 = const()[name = tensor<string, []>("op_60781_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60781_cast = slice_by_index(begin = var_60781_begin_0, end = var_60781_end_0, end_mask = var_60781_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_60781_cast")];
            tensor<int32, [4]> var_60785_begin_0 = const()[name = tensor<string, []>("op_60785_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_60785_end_0 = const()[name = tensor<string, []>("op_60785_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_60785_end_mask_0 = const()[name = tensor<string, []>("op_60785_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60785_cast = slice_by_index(begin = var_60785_begin_0, end = var_60785_end_0, end_mask = var_60785_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_60785_cast")];
            tensor<int32, [4]> var_60789_begin_0 = const()[name = tensor<string, []>("op_60789_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_60789_end_0 = const()[name = tensor<string, []>("op_60789_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_60789_end_mask_0 = const()[name = tensor<string, []>("op_60789_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60789_cast = slice_by_index(begin = var_60789_begin_0, end = var_60789_end_0, end_mask = var_60789_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_60789_cast")];
            tensor<int32, [4]> var_60793_begin_0 = const()[name = tensor<string, []>("op_60793_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_60793_end_0 = const()[name = tensor<string, []>("op_60793_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_60793_end_mask_0 = const()[name = tensor<string, []>("op_60793_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60793_cast = slice_by_index(begin = var_60793_begin_0, end = var_60793_end_0, end_mask = var_60793_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_60793_cast")];
            tensor<int32, [4]> var_60797_begin_0 = const()[name = tensor<string, []>("op_60797_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_60797_end_0 = const()[name = tensor<string, []>("op_60797_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_60797_end_mask_0 = const()[name = tensor<string, []>("op_60797_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60797_cast = slice_by_index(begin = var_60797_begin_0, end = var_60797_end_0, end_mask = var_60797_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_60797_cast")];
            tensor<int32, [4]> var_60801_begin_0 = const()[name = tensor<string, []>("op_60801_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_60801_end_0 = const()[name = tensor<string, []>("op_60801_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_60801_end_mask_0 = const()[name = tensor<string, []>("op_60801_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60801_cast = slice_by_index(begin = var_60801_begin_0, end = var_60801_end_0, end_mask = var_60801_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_60801_cast")];
            tensor<int32, [4]> var_60805_begin_0 = const()[name = tensor<string, []>("op_60805_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_60805_end_0 = const()[name = tensor<string, []>("op_60805_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_60805_end_mask_0 = const()[name = tensor<string, []>("op_60805_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60805_cast = slice_by_index(begin = var_60805_begin_0, end = var_60805_end_0, end_mask = var_60805_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_60805_cast")];
            tensor<int32, [4]> var_60809_begin_0 = const()[name = tensor<string, []>("op_60809_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_60809_end_0 = const()[name = tensor<string, []>("op_60809_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_60809_end_mask_0 = const()[name = tensor<string, []>("op_60809_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60809_cast = slice_by_index(begin = var_60809_begin_0, end = var_60809_end_0, end_mask = var_60809_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_60809_cast")];
            tensor<int32, [4]> var_60813_begin_0 = const()[name = tensor<string, []>("op_60813_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_60813_end_0 = const()[name = tensor<string, []>("op_60813_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_60813_end_mask_0 = const()[name = tensor<string, []>("op_60813_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_60813_cast = slice_by_index(begin = var_60813_begin_0, end = var_60813_end_0, end_mask = var_60813_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_60813_cast")];
            tensor<int32, [4]> var_60815_begin_0 = const()[name = tensor<string, []>("op_60815_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_60815_end_0 = const()[name = tensor<string, []>("op_60815_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_60815_end_mask_0 = const()[name = tensor<string, []>("op_60815_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60815_cast = slice_by_index(begin = var_60815_begin_0, end = var_60815_end_0, end_mask = var_60815_end_mask_0, x = v_275_cast)[name = tensor<string, []>("op_60815_cast")];
            tensor<int32, [4]> var_60819_begin_0 = const()[name = tensor<string, []>("op_60819_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_60819_end_0 = const()[name = tensor<string, []>("op_60819_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_60819_end_mask_0 = const()[name = tensor<string, []>("op_60819_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60819_cast = slice_by_index(begin = var_60819_begin_0, end = var_60819_end_0, end_mask = var_60819_end_mask_0, x = v_275_cast)[name = tensor<string, []>("op_60819_cast")];
            tensor<int32, [4]> var_60823_begin_0 = const()[name = tensor<string, []>("op_60823_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_60823_end_0 = const()[name = tensor<string, []>("op_60823_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_60823_end_mask_0 = const()[name = tensor<string, []>("op_60823_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60823_cast = slice_by_index(begin = var_60823_begin_0, end = var_60823_end_0, end_mask = var_60823_end_mask_0, x = v_275_cast)[name = tensor<string, []>("op_60823_cast")];
            tensor<int32, [4]> var_60827_begin_0 = const()[name = tensor<string, []>("op_60827_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_60827_end_0 = const()[name = tensor<string, []>("op_60827_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_60827_end_mask_0 = const()[name = tensor<string, []>("op_60827_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60827_cast = slice_by_index(begin = var_60827_begin_0, end = var_60827_end_0, end_mask = var_60827_end_mask_0, x = v_275_cast)[name = tensor<string, []>("op_60827_cast")];
            tensor<int32, [4]> var_60831_begin_0 = const()[name = tensor<string, []>("op_60831_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_60831_end_0 = const()[name = tensor<string, []>("op_60831_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_60831_end_mask_0 = const()[name = tensor<string, []>("op_60831_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60831_cast = slice_by_index(begin = var_60831_begin_0, end = var_60831_end_0, end_mask = var_60831_end_mask_0, x = v_275_cast)[name = tensor<string, []>("op_60831_cast")];
            tensor<int32, [4]> var_60835_begin_0 = const()[name = tensor<string, []>("op_60835_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_60835_end_0 = const()[name = tensor<string, []>("op_60835_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_60835_end_mask_0 = const()[name = tensor<string, []>("op_60835_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60835_cast = slice_by_index(begin = var_60835_begin_0, end = var_60835_end_0, end_mask = var_60835_end_mask_0, x = v_275_cast)[name = tensor<string, []>("op_60835_cast")];
            tensor<int32, [4]> var_60839_begin_0 = const()[name = tensor<string, []>("op_60839_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_60839_end_0 = const()[name = tensor<string, []>("op_60839_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_60839_end_mask_0 = const()[name = tensor<string, []>("op_60839_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60839_cast = slice_by_index(begin = var_60839_begin_0, end = var_60839_end_0, end_mask = var_60839_end_mask_0, x = v_275_cast)[name = tensor<string, []>("op_60839_cast")];
            tensor<int32, [4]> var_60843_begin_0 = const()[name = tensor<string, []>("op_60843_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_60843_end_0 = const()[name = tensor<string, []>("op_60843_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_60843_end_mask_0 = const()[name = tensor<string, []>("op_60843_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60843_cast = slice_by_index(begin = var_60843_begin_0, end = var_60843_end_0, end_mask = var_60843_end_mask_0, x = v_275_cast)[name = tensor<string, []>("op_60843_cast")];
            tensor<int32, [4]> var_60847_begin_0 = const()[name = tensor<string, []>("op_60847_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_60847_end_0 = const()[name = tensor<string, []>("op_60847_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_60847_end_mask_0 = const()[name = tensor<string, []>("op_60847_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60847_cast = slice_by_index(begin = var_60847_begin_0, end = var_60847_end_0, end_mask = var_60847_end_mask_0, x = v_275_cast)[name = tensor<string, []>("op_60847_cast")];
            tensor<int32, [4]> var_60851_begin_0 = const()[name = tensor<string, []>("op_60851_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_60851_end_0 = const()[name = tensor<string, []>("op_60851_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_60851_end_mask_0 = const()[name = tensor<string, []>("op_60851_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_60851_cast = slice_by_index(begin = var_60851_begin_0, end = var_60851_end_0, end_mask = var_60851_end_mask_0, x = v_275_cast)[name = tensor<string, []>("op_60851_cast")];
            tensor<string, []> var_60855_equation_0 = const()[name = tensor<string, []>("op_60855_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60855_cast = einsum(equation = var_60855_equation_0, values = (var_60777_cast, var_60734_cast))[name = tensor<string, []>("op_60855_cast")];
            tensor<fp16, []> var_60856_to_fp16 = const()[name = tensor<string, []>("op_60856_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5141_cast = mul(x = var_60855_cast, y = var_60856_to_fp16)[name = tensor<string, []>("aw_5141_cast")];
            tensor<string, []> var_60859_equation_0 = const()[name = tensor<string, []>("op_60859_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60859_cast = einsum(equation = var_60859_equation_0, values = (var_60781_cast, var_60738_cast))[name = tensor<string, []>("op_60859_cast")];
            tensor<fp16, []> var_60860_to_fp16 = const()[name = tensor<string, []>("op_60860_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5143_cast = mul(x = var_60859_cast, y = var_60860_to_fp16)[name = tensor<string, []>("aw_5143_cast")];
            tensor<string, []> var_60863_equation_0 = const()[name = tensor<string, []>("op_60863_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60863_cast = einsum(equation = var_60863_equation_0, values = (var_60785_cast, var_60742_cast))[name = tensor<string, []>("op_60863_cast")];
            tensor<fp16, []> var_60864_to_fp16 = const()[name = tensor<string, []>("op_60864_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5145_cast = mul(x = var_60863_cast, y = var_60864_to_fp16)[name = tensor<string, []>("aw_5145_cast")];
            tensor<string, []> var_60867_equation_0 = const()[name = tensor<string, []>("op_60867_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60867_cast = einsum(equation = var_60867_equation_0, values = (var_60789_cast, var_60746_cast))[name = tensor<string, []>("op_60867_cast")];
            tensor<fp16, []> var_60868_to_fp16 = const()[name = tensor<string, []>("op_60868_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5147_cast = mul(x = var_60867_cast, y = var_60868_to_fp16)[name = tensor<string, []>("aw_5147_cast")];
            tensor<string, []> var_60871_equation_0 = const()[name = tensor<string, []>("op_60871_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60871_cast = einsum(equation = var_60871_equation_0, values = (var_60793_cast, var_60750_cast))[name = tensor<string, []>("op_60871_cast")];
            tensor<fp16, []> var_60872_to_fp16 = const()[name = tensor<string, []>("op_60872_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5149_cast = mul(x = var_60871_cast, y = var_60872_to_fp16)[name = tensor<string, []>("aw_5149_cast")];
            tensor<string, []> var_60875_equation_0 = const()[name = tensor<string, []>("op_60875_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60875_cast = einsum(equation = var_60875_equation_0, values = (var_60797_cast, var_60754_cast))[name = tensor<string, []>("op_60875_cast")];
            tensor<fp16, []> var_60876_to_fp16 = const()[name = tensor<string, []>("op_60876_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5151_cast = mul(x = var_60875_cast, y = var_60876_to_fp16)[name = tensor<string, []>("aw_5151_cast")];
            tensor<string, []> var_60879_equation_0 = const()[name = tensor<string, []>("op_60879_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60879_cast = einsum(equation = var_60879_equation_0, values = (var_60801_cast, var_60758_cast))[name = tensor<string, []>("op_60879_cast")];
            tensor<fp16, []> var_60880_to_fp16 = const()[name = tensor<string, []>("op_60880_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5153_cast = mul(x = var_60879_cast, y = var_60880_to_fp16)[name = tensor<string, []>("aw_5153_cast")];
            tensor<string, []> var_60883_equation_0 = const()[name = tensor<string, []>("op_60883_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60883_cast = einsum(equation = var_60883_equation_0, values = (var_60805_cast, var_60762_cast))[name = tensor<string, []>("op_60883_cast")];
            tensor<fp16, []> var_60884_to_fp16 = const()[name = tensor<string, []>("op_60884_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5155_cast = mul(x = var_60883_cast, y = var_60884_to_fp16)[name = tensor<string, []>("aw_5155_cast")];
            tensor<string, []> var_60887_equation_0 = const()[name = tensor<string, []>("op_60887_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60887_cast = einsum(equation = var_60887_equation_0, values = (var_60809_cast, var_60766_cast))[name = tensor<string, []>("op_60887_cast")];
            tensor<fp16, []> var_60888_to_fp16 = const()[name = tensor<string, []>("op_60888_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5157_cast = mul(x = var_60887_cast, y = var_60888_to_fp16)[name = tensor<string, []>("aw_5157_cast")];
            tensor<string, []> var_60891_equation_0 = const()[name = tensor<string, []>("op_60891_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_60891_cast = einsum(equation = var_60891_equation_0, values = (var_60813_cast, var_60770_cast))[name = tensor<string, []>("op_60891_cast")];
            tensor<fp16, []> var_60892_to_fp16 = const()[name = tensor<string, []>("op_60892_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5159_cast = mul(x = var_60891_cast, y = var_60892_to_fp16)[name = tensor<string, []>("aw_5159_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60894_cast = softmax(axis = var_58064, x = aw_5141_cast)[name = tensor<string, []>("op_60894_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60895_cast = softmax(axis = var_58064, x = aw_5143_cast)[name = tensor<string, []>("op_60895_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60896_cast = softmax(axis = var_58064, x = aw_5145_cast)[name = tensor<string, []>("op_60896_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60897_cast = softmax(axis = var_58064, x = aw_5147_cast)[name = tensor<string, []>("op_60897_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60898_cast = softmax(axis = var_58064, x = aw_5149_cast)[name = tensor<string, []>("op_60898_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60899_cast = softmax(axis = var_58064, x = aw_5151_cast)[name = tensor<string, []>("op_60899_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60900_cast = softmax(axis = var_58064, x = aw_5153_cast)[name = tensor<string, []>("op_60900_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60901_cast = softmax(axis = var_58064, x = aw_5155_cast)[name = tensor<string, []>("op_60901_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60902_cast = softmax(axis = var_58064, x = aw_5157_cast)[name = tensor<string, []>("op_60902_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_60903_cast = softmax(axis = var_58064, x = aw_5159_cast)[name = tensor<string, []>("op_60903_cast")];
            tensor<string, []> var_60905_equation_0 = const()[name = tensor<string, []>("op_60905_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60905_cast = einsum(equation = var_60905_equation_0, values = (var_60815_cast, var_60894_cast))[name = tensor<string, []>("op_60905_cast")];
            tensor<string, []> var_60907_equation_0 = const()[name = tensor<string, []>("op_60907_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60907_cast = einsum(equation = var_60907_equation_0, values = (var_60819_cast, var_60895_cast))[name = tensor<string, []>("op_60907_cast")];
            tensor<string, []> var_60909_equation_0 = const()[name = tensor<string, []>("op_60909_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60909_cast = einsum(equation = var_60909_equation_0, values = (var_60823_cast, var_60896_cast))[name = tensor<string, []>("op_60909_cast")];
            tensor<string, []> var_60911_equation_0 = const()[name = tensor<string, []>("op_60911_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60911_cast = einsum(equation = var_60911_equation_0, values = (var_60827_cast, var_60897_cast))[name = tensor<string, []>("op_60911_cast")];
            tensor<string, []> var_60913_equation_0 = const()[name = tensor<string, []>("op_60913_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60913_cast = einsum(equation = var_60913_equation_0, values = (var_60831_cast, var_60898_cast))[name = tensor<string, []>("op_60913_cast")];
            tensor<string, []> var_60915_equation_0 = const()[name = tensor<string, []>("op_60915_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60915_cast = einsum(equation = var_60915_equation_0, values = (var_60835_cast, var_60899_cast))[name = tensor<string, []>("op_60915_cast")];
            tensor<string, []> var_60917_equation_0 = const()[name = tensor<string, []>("op_60917_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60917_cast = einsum(equation = var_60917_equation_0, values = (var_60839_cast, var_60900_cast))[name = tensor<string, []>("op_60917_cast")];
            tensor<string, []> var_60919_equation_0 = const()[name = tensor<string, []>("op_60919_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60919_cast = einsum(equation = var_60919_equation_0, values = (var_60843_cast, var_60901_cast))[name = tensor<string, []>("op_60919_cast")];
            tensor<string, []> var_60921_equation_0 = const()[name = tensor<string, []>("op_60921_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60921_cast = einsum(equation = var_60921_equation_0, values = (var_60847_cast, var_60902_cast))[name = tensor<string, []>("op_60921_cast")];
            tensor<string, []> var_60923_equation_0 = const()[name = tensor<string, []>("op_60923_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_60923_cast = einsum(equation = var_60923_equation_0, values = (var_60851_cast, var_60903_cast))[name = tensor<string, []>("op_60923_cast")];
            tensor<bool, []> input_809_interleave_0 = const()[name = tensor<string, []>("input_809_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_809_cast = concat(axis = var_58064, interleave = input_809_interleave_0, values = (var_60905_cast, var_60907_cast, var_60909_cast, var_60911_cast, var_60913_cast, var_60915_cast, var_60917_cast, var_60919_cast, var_60921_cast, var_60923_cast))[name = tensor<string, []>("input_809_cast")];
            tensor<int32, [2]> var_60929 = const()[name = tensor<string, []>("op_60929"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60931 = const()[name = tensor<string, []>("op_60931"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_60933_pad_type_0 = const()[name = tensor<string, []>("op_60933_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_60933_pad_0 = const()[name = tensor<string, []>("op_60933_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5073824768)))];
            tensor<fp16, [640]> up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5074644032)))];
            tensor<fp16, [2, 640, 1, 4096]> var_60933_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_60931, groups = var_58064, pad = var_60933_pad_0, pad_type = var_60933_pad_type_0, strides = var_60929, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_809_cast)[name = tensor<string, []>("op_60933_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_413_cast = add(x = var_60933_cast, y = inputs_411_cast)[name = tensor<string, []>("inputs_413_cast")];
            tensor<int32, [1]> var_60937 = const()[name = tensor<string, []>("op_60937"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_413_cast = reduce_mean(axes = var_60937, keep_dims = var_58059, x = inputs_413_cast)[name = tensor<string, []>("channels_mean_413_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_413_cast = sub(x = inputs_413_cast, y = channels_mean_413_cast)[name = tensor<string, []>("zero_mean_413_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_413_cast = mul(x = zero_mean_413_cast, y = zero_mean_413_cast)[name = tensor<string, []>("zero_mean_sq_413_cast")];
            tensor<int32, [1]> var_60941 = const()[name = tensor<string, []>("op_60941"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_60942_cast = reduce_mean(axes = var_60941, keep_dims = var_58059, x = zero_mean_sq_413_cast)[name = tensor<string, []>("op_60942_cast")];
            tensor<fp16, []> var_60943_to_fp16 = const()[name = tensor<string, []>("op_60943_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_60944_cast = add(x = var_60942_cast, y = var_60943_to_fp16)[name = tensor<string, []>("op_60944_cast")];
            tensor<fp16, []> denom_413_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_413_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_413_cast = rsqrt(epsilon = denom_413_epsilon_0_to_fp16, x = var_60944_cast)[name = tensor<string, []>("denom_413_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_413_cast = mul(x = zero_mean_413_cast, y = denom_413_cast)[name = tensor<string, []>("out_413_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_60948_to_fp16 = const()[name = tensor<string, []>("op_60948_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5074645376)))];
            tensor<fp16, [2, 640, 1, 4096]> var_60949_cast = add(x = out_413_cast, y = var_60948_to_fp16)[name = tensor<string, []>("op_60949_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_60951_to_fp16 = const()[name = tensor<string, []>("op_60951_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5074646720)))];
            tensor<fp16, [2, 640, 1, 4096]> input_811_cast = mul(x = var_60949_cast, y = var_60951_to_fp16)[name = tensor<string, []>("input_811_cast")];
            tensor<int32, [2]> var_60959 = const()[name = tensor<string, []>("op_60959"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60961 = const()[name = tensor<string, []>("op_60961"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_60963_pad_type_0 = const()[name = tensor<string, []>("op_60963_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_60963_pad_0 = const()[name = tensor<string, []>("op_60963_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [5120, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [5120, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5074648064)))];
            tensor<fp16, [5120]> up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5081201728)))];
            tensor<fp16, [2, 5120, 1, 4096]> var_60963_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_60961, groups = var_58064, pad = var_60963_pad_0, pad_type = var_60963_pad_type_0, strides = var_60959, weight = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_811_cast)[name = tensor<string, []>("op_60963_cast")];
            tensor<int32, [2]> var_60964_split_sizes_0 = const()[name = tensor<string, []>("op_60964_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_60964_axis_0 = const()[name = tensor<string, []>("op_60964_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 4096]> var_60964_cast_0, tensor<fp16, [2, 2560, 1, 4096]> var_60964_cast_1 = split(axis = var_60964_axis_0, split_sizes = var_60964_split_sizes_0, x = var_60963_cast)[name = tensor<string, []>("op_60964_cast")];
            tensor<string, []> var_60966_mode_0 = const()[name = tensor<string, []>("op_60966_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 4096]> var_60966_cast = gelu(mode = var_60966_mode_0, x = var_60964_cast_1)[name = tensor<string, []>("op_60966_cast")];
            tensor<fp16, [2, 2560, 1, 4096]> input_813_cast = mul(x = var_60964_cast_0, y = var_60966_cast)[name = tensor<string, []>("input_813_cast")];
            tensor<int32, [2]> var_60970 = const()[name = tensor<string, []>("op_60970"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_60972 = const()[name = tensor<string, []>("op_60972"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_60974_pad_type_0 = const()[name = tensor<string, []>("op_60974_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_60974_pad_0 = const()[name = tensor<string, []>("op_60974_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2560, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [640, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5081212032)))];
            tensor<fp16, [640]> up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5084488896)))];
            tensor<fp16, [2, 640, 1, 4096]> var_60974_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_60972, groups = var_58064, pad = var_60974_pad_0, pad_type = var_60974_pad_type_0, strides = var_60970, weight = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_813_cast)[name = tensor<string, []>("op_60974_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_415_cast = add(x = var_60974_cast, y = inputs_413_cast)[name = tensor<string, []>("inputs_415_cast")];
            tensor<int32, [1]> var_60984 = const()[name = tensor<string, []>("op_60984"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_415_cast = reduce_mean(axes = var_60984, keep_dims = var_58059, x = inputs_415_cast)[name = tensor<string, []>("channels_mean_415_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_415_cast = sub(x = inputs_415_cast, y = channels_mean_415_cast)[name = tensor<string, []>("zero_mean_415_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_415_cast = mul(x = zero_mean_415_cast, y = zero_mean_415_cast)[name = tensor<string, []>("zero_mean_sq_415_cast")];
            tensor<int32, [1]> var_60988 = const()[name = tensor<string, []>("op_60988"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_60989_cast = reduce_mean(axes = var_60988, keep_dims = var_58059, x = zero_mean_sq_415_cast)[name = tensor<string, []>("op_60989_cast")];
            tensor<fp16, []> var_60990_to_fp16 = const()[name = tensor<string, []>("op_60990_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_60991_cast = add(x = var_60989_cast, y = var_60990_to_fp16)[name = tensor<string, []>("op_60991_cast")];
            tensor<fp16, []> denom_415_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_415_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_415_cast = rsqrt(epsilon = denom_415_epsilon_0_to_fp16, x = var_60991_cast)[name = tensor<string, []>("denom_415_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_415_cast = mul(x = zero_mean_415_cast, y = denom_415_cast)[name = tensor<string, []>("out_415_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_60995_to_fp16 = const()[name = tensor<string, []>("op_60995_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5084490240)))];
            tensor<fp16, [2, 640, 1, 4096]> var_60996_cast = add(x = out_415_cast, y = var_60995_to_fp16)[name = tensor<string, []>("op_60996_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_60998_to_fp16 = const()[name = tensor<string, []>("op_60998_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5084491584)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_567_cast = mul(x = var_60996_cast, y = var_60998_to_fp16)[name = tensor<string, []>("hidden_states_567_cast")];
            tensor<int32, [2]> var_61005 = const()[name = tensor<string, []>("op_61005"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61007 = const()[name = tensor<string, []>("op_61007"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_277_pad_type_0 = const()[name = tensor<string, []>("q_277_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_277_pad_0 = const()[name = tensor<string, []>("q_277_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5084492928)))];
            tensor<fp16, [2, 640, 1, 4096]> q_277_cast = conv(dilations = var_61007, groups = var_58064, pad = q_277_pad_0, pad_type = q_277_pad_type_0, strides = var_61005, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_q_weight_to_fp16, x = hidden_states_567_cast)[name = tensor<string, []>("q_277_cast")];
            tensor<int32, [2]> var_61011 = const()[name = tensor<string, []>("op_61011"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61013 = const()[name = tensor<string, []>("op_61013"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_553_pad_type_0 = const()[name = tensor<string, []>("k_553_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_553_pad_0 = const()[name = tensor<string, []>("k_553_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5085312192)))];
            tensor<fp16, [2, 640, 1, 4096]> k_553_cast = conv(dilations = var_61013, groups = var_58064, pad = k_553_pad_0, pad_type = k_553_pad_type_0, strides = var_61011, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_k_weight_to_fp16, x = hidden_states_567_cast)[name = tensor<string, []>("k_553_cast")];
            tensor<int32, [2]> var_61017 = const()[name = tensor<string, []>("op_61017"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61019 = const()[name = tensor<string, []>("op_61019"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_277_pad_type_0 = const()[name = tensor<string, []>("v_277_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_277_pad_0 = const()[name = tensor<string, []>("v_277_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5086131456)))];
            tensor<fp16, [2, 640, 1, 4096]> v_277_cast = conv(dilations = var_61019, groups = var_58064, pad = v_277_pad_0, pad_type = v_277_pad_type_0, strides = var_61017, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_v_weight_to_fp16, x = hidden_states_567_cast)[name = tensor<string, []>("v_277_cast")];
            tensor<int32, [4]> var_61023_begin_0 = const()[name = tensor<string, []>("op_61023_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_61023_end_0 = const()[name = tensor<string, []>("op_61023_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_61023_end_mask_0 = const()[name = tensor<string, []>("op_61023_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61023_cast = slice_by_index(begin = var_61023_begin_0, end = var_61023_end_0, end_mask = var_61023_end_mask_0, x = q_277_cast)[name = tensor<string, []>("op_61023_cast")];
            tensor<int32, [4]> var_61027_begin_0 = const()[name = tensor<string, []>("op_61027_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_61027_end_0 = const()[name = tensor<string, []>("op_61027_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_61027_end_mask_0 = const()[name = tensor<string, []>("op_61027_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61027_cast = slice_by_index(begin = var_61027_begin_0, end = var_61027_end_0, end_mask = var_61027_end_mask_0, x = q_277_cast)[name = tensor<string, []>("op_61027_cast")];
            tensor<int32, [4]> var_61031_begin_0 = const()[name = tensor<string, []>("op_61031_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_61031_end_0 = const()[name = tensor<string, []>("op_61031_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_61031_end_mask_0 = const()[name = tensor<string, []>("op_61031_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61031_cast = slice_by_index(begin = var_61031_begin_0, end = var_61031_end_0, end_mask = var_61031_end_mask_0, x = q_277_cast)[name = tensor<string, []>("op_61031_cast")];
            tensor<int32, [4]> var_61035_begin_0 = const()[name = tensor<string, []>("op_61035_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_61035_end_0 = const()[name = tensor<string, []>("op_61035_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_61035_end_mask_0 = const()[name = tensor<string, []>("op_61035_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61035_cast = slice_by_index(begin = var_61035_begin_0, end = var_61035_end_0, end_mask = var_61035_end_mask_0, x = q_277_cast)[name = tensor<string, []>("op_61035_cast")];
            tensor<int32, [4]> var_61039_begin_0 = const()[name = tensor<string, []>("op_61039_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_61039_end_0 = const()[name = tensor<string, []>("op_61039_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_61039_end_mask_0 = const()[name = tensor<string, []>("op_61039_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61039_cast = slice_by_index(begin = var_61039_begin_0, end = var_61039_end_0, end_mask = var_61039_end_mask_0, x = q_277_cast)[name = tensor<string, []>("op_61039_cast")];
            tensor<int32, [4]> var_61043_begin_0 = const()[name = tensor<string, []>("op_61043_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_61043_end_0 = const()[name = tensor<string, []>("op_61043_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_61043_end_mask_0 = const()[name = tensor<string, []>("op_61043_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61043_cast = slice_by_index(begin = var_61043_begin_0, end = var_61043_end_0, end_mask = var_61043_end_mask_0, x = q_277_cast)[name = tensor<string, []>("op_61043_cast")];
            tensor<int32, [4]> var_61047_begin_0 = const()[name = tensor<string, []>("op_61047_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_61047_end_0 = const()[name = tensor<string, []>("op_61047_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_61047_end_mask_0 = const()[name = tensor<string, []>("op_61047_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61047_cast = slice_by_index(begin = var_61047_begin_0, end = var_61047_end_0, end_mask = var_61047_end_mask_0, x = q_277_cast)[name = tensor<string, []>("op_61047_cast")];
            tensor<int32, [4]> var_61051_begin_0 = const()[name = tensor<string, []>("op_61051_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_61051_end_0 = const()[name = tensor<string, []>("op_61051_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_61051_end_mask_0 = const()[name = tensor<string, []>("op_61051_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61051_cast = slice_by_index(begin = var_61051_begin_0, end = var_61051_end_0, end_mask = var_61051_end_mask_0, x = q_277_cast)[name = tensor<string, []>("op_61051_cast")];
            tensor<int32, [4]> var_61055_begin_0 = const()[name = tensor<string, []>("op_61055_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_61055_end_0 = const()[name = tensor<string, []>("op_61055_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_61055_end_mask_0 = const()[name = tensor<string, []>("op_61055_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61055_cast = slice_by_index(begin = var_61055_begin_0, end = var_61055_end_0, end_mask = var_61055_end_mask_0, x = q_277_cast)[name = tensor<string, []>("op_61055_cast")];
            tensor<int32, [4]> var_61059_begin_0 = const()[name = tensor<string, []>("op_61059_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_61059_end_0 = const()[name = tensor<string, []>("op_61059_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_61059_end_mask_0 = const()[name = tensor<string, []>("op_61059_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61059_cast = slice_by_index(begin = var_61059_begin_0, end = var_61059_end_0, end_mask = var_61059_end_mask_0, x = q_277_cast)[name = tensor<string, []>("op_61059_cast")];
            tensor<int32, [4]> k_555_perm_0 = const()[name = tensor<string, []>("k_555_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_61066_begin_0 = const()[name = tensor<string, []>("op_61066_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_61066_end_0 = const()[name = tensor<string, []>("op_61066_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_61066_end_mask_0 = const()[name = tensor<string, []>("op_61066_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 640]> transpose_1 = transpose(perm = k_555_perm_0, x = k_553_cast)[name = tensor<string, []>("transpose_1")];
            tensor<fp16, [2, 4096, 1, 64]> var_61066_cast = slice_by_index(begin = var_61066_begin_0, end = var_61066_end_0, end_mask = var_61066_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_61066_cast")];
            tensor<int32, [4]> var_61070_begin_0 = const()[name = tensor<string, []>("op_61070_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_61070_end_0 = const()[name = tensor<string, []>("op_61070_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_61070_end_mask_0 = const()[name = tensor<string, []>("op_61070_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_61070_cast = slice_by_index(begin = var_61070_begin_0, end = var_61070_end_0, end_mask = var_61070_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_61070_cast")];
            tensor<int32, [4]> var_61074_begin_0 = const()[name = tensor<string, []>("op_61074_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_61074_end_0 = const()[name = tensor<string, []>("op_61074_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_61074_end_mask_0 = const()[name = tensor<string, []>("op_61074_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_61074_cast = slice_by_index(begin = var_61074_begin_0, end = var_61074_end_0, end_mask = var_61074_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_61074_cast")];
            tensor<int32, [4]> var_61078_begin_0 = const()[name = tensor<string, []>("op_61078_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_61078_end_0 = const()[name = tensor<string, []>("op_61078_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_61078_end_mask_0 = const()[name = tensor<string, []>("op_61078_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_61078_cast = slice_by_index(begin = var_61078_begin_0, end = var_61078_end_0, end_mask = var_61078_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_61078_cast")];
            tensor<int32, [4]> var_61082_begin_0 = const()[name = tensor<string, []>("op_61082_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_61082_end_0 = const()[name = tensor<string, []>("op_61082_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_61082_end_mask_0 = const()[name = tensor<string, []>("op_61082_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_61082_cast = slice_by_index(begin = var_61082_begin_0, end = var_61082_end_0, end_mask = var_61082_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_61082_cast")];
            tensor<int32, [4]> var_61086_begin_0 = const()[name = tensor<string, []>("op_61086_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_61086_end_0 = const()[name = tensor<string, []>("op_61086_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 384])];
            tensor<bool, [4]> var_61086_end_mask_0 = const()[name = tensor<string, []>("op_61086_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_61086_cast = slice_by_index(begin = var_61086_begin_0, end = var_61086_end_0, end_mask = var_61086_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_61086_cast")];
            tensor<int32, [4]> var_61090_begin_0 = const()[name = tensor<string, []>("op_61090_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_61090_end_0 = const()[name = tensor<string, []>("op_61090_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 448])];
            tensor<bool, [4]> var_61090_end_mask_0 = const()[name = tensor<string, []>("op_61090_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_61090_cast = slice_by_index(begin = var_61090_begin_0, end = var_61090_end_0, end_mask = var_61090_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_61090_cast")];
            tensor<int32, [4]> var_61094_begin_0 = const()[name = tensor<string, []>("op_61094_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_61094_end_0 = const()[name = tensor<string, []>("op_61094_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 512])];
            tensor<bool, [4]> var_61094_end_mask_0 = const()[name = tensor<string, []>("op_61094_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_61094_cast = slice_by_index(begin = var_61094_begin_0, end = var_61094_end_0, end_mask = var_61094_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_61094_cast")];
            tensor<int32, [4]> var_61098_begin_0 = const()[name = tensor<string, []>("op_61098_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_61098_end_0 = const()[name = tensor<string, []>("op_61098_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 576])];
            tensor<bool, [4]> var_61098_end_mask_0 = const()[name = tensor<string, []>("op_61098_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_61098_cast = slice_by_index(begin = var_61098_begin_0, end = var_61098_end_0, end_mask = var_61098_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_61098_cast")];
            tensor<int32, [4]> var_61102_begin_0 = const()[name = tensor<string, []>("op_61102_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_61102_end_0 = const()[name = tensor<string, []>("op_61102_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 640])];
            tensor<bool, [4]> var_61102_end_mask_0 = const()[name = tensor<string, []>("op_61102_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_61102_cast = slice_by_index(begin = var_61102_begin_0, end = var_61102_end_0, end_mask = var_61102_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_61102_cast")];
            tensor<int32, [4]> var_61104_begin_0 = const()[name = tensor<string, []>("op_61104_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_61104_end_0 = const()[name = tensor<string, []>("op_61104_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_61104_end_mask_0 = const()[name = tensor<string, []>("op_61104_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61104_cast = slice_by_index(begin = var_61104_begin_0, end = var_61104_end_0, end_mask = var_61104_end_mask_0, x = v_277_cast)[name = tensor<string, []>("op_61104_cast")];
            tensor<int32, [4]> var_61108_begin_0 = const()[name = tensor<string, []>("op_61108_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_61108_end_0 = const()[name = tensor<string, []>("op_61108_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_61108_end_mask_0 = const()[name = tensor<string, []>("op_61108_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61108_cast = slice_by_index(begin = var_61108_begin_0, end = var_61108_end_0, end_mask = var_61108_end_mask_0, x = v_277_cast)[name = tensor<string, []>("op_61108_cast")];
            tensor<int32, [4]> var_61112_begin_0 = const()[name = tensor<string, []>("op_61112_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_61112_end_0 = const()[name = tensor<string, []>("op_61112_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_61112_end_mask_0 = const()[name = tensor<string, []>("op_61112_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61112_cast = slice_by_index(begin = var_61112_begin_0, end = var_61112_end_0, end_mask = var_61112_end_mask_0, x = v_277_cast)[name = tensor<string, []>("op_61112_cast")];
            tensor<int32, [4]> var_61116_begin_0 = const()[name = tensor<string, []>("op_61116_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_61116_end_0 = const()[name = tensor<string, []>("op_61116_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_61116_end_mask_0 = const()[name = tensor<string, []>("op_61116_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61116_cast = slice_by_index(begin = var_61116_begin_0, end = var_61116_end_0, end_mask = var_61116_end_mask_0, x = v_277_cast)[name = tensor<string, []>("op_61116_cast")];
            tensor<int32, [4]> var_61120_begin_0 = const()[name = tensor<string, []>("op_61120_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_61120_end_0 = const()[name = tensor<string, []>("op_61120_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_61120_end_mask_0 = const()[name = tensor<string, []>("op_61120_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61120_cast = slice_by_index(begin = var_61120_begin_0, end = var_61120_end_0, end_mask = var_61120_end_mask_0, x = v_277_cast)[name = tensor<string, []>("op_61120_cast")];
            tensor<int32, [4]> var_61124_begin_0 = const()[name = tensor<string, []>("op_61124_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_61124_end_0 = const()[name = tensor<string, []>("op_61124_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_61124_end_mask_0 = const()[name = tensor<string, []>("op_61124_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61124_cast = slice_by_index(begin = var_61124_begin_0, end = var_61124_end_0, end_mask = var_61124_end_mask_0, x = v_277_cast)[name = tensor<string, []>("op_61124_cast")];
            tensor<int32, [4]> var_61128_begin_0 = const()[name = tensor<string, []>("op_61128_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_61128_end_0 = const()[name = tensor<string, []>("op_61128_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_61128_end_mask_0 = const()[name = tensor<string, []>("op_61128_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61128_cast = slice_by_index(begin = var_61128_begin_0, end = var_61128_end_0, end_mask = var_61128_end_mask_0, x = v_277_cast)[name = tensor<string, []>("op_61128_cast")];
            tensor<int32, [4]> var_61132_begin_0 = const()[name = tensor<string, []>("op_61132_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_61132_end_0 = const()[name = tensor<string, []>("op_61132_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_61132_end_mask_0 = const()[name = tensor<string, []>("op_61132_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61132_cast = slice_by_index(begin = var_61132_begin_0, end = var_61132_end_0, end_mask = var_61132_end_mask_0, x = v_277_cast)[name = tensor<string, []>("op_61132_cast")];
            tensor<int32, [4]> var_61136_begin_0 = const()[name = tensor<string, []>("op_61136_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_61136_end_0 = const()[name = tensor<string, []>("op_61136_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_61136_end_mask_0 = const()[name = tensor<string, []>("op_61136_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61136_cast = slice_by_index(begin = var_61136_begin_0, end = var_61136_end_0, end_mask = var_61136_end_mask_0, x = v_277_cast)[name = tensor<string, []>("op_61136_cast")];
            tensor<int32, [4]> var_61140_begin_0 = const()[name = tensor<string, []>("op_61140_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_61140_end_0 = const()[name = tensor<string, []>("op_61140_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_61140_end_mask_0 = const()[name = tensor<string, []>("op_61140_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61140_cast = slice_by_index(begin = var_61140_begin_0, end = var_61140_end_0, end_mask = var_61140_end_mask_0, x = v_277_cast)[name = tensor<string, []>("op_61140_cast")];
            tensor<string, []> var_61144_equation_0 = const()[name = tensor<string, []>("op_61144_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61144_cast = einsum(equation = var_61144_equation_0, values = (var_61066_cast, var_61023_cast))[name = tensor<string, []>("op_61144_cast")];
            tensor<fp16, []> var_61145_to_fp16 = const()[name = tensor<string, []>("op_61145_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5161_cast = mul(x = var_61144_cast, y = var_61145_to_fp16)[name = tensor<string, []>("aw_5161_cast")];
            tensor<string, []> var_61148_equation_0 = const()[name = tensor<string, []>("op_61148_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61148_cast = einsum(equation = var_61148_equation_0, values = (var_61070_cast, var_61027_cast))[name = tensor<string, []>("op_61148_cast")];
            tensor<fp16, []> var_61149_to_fp16 = const()[name = tensor<string, []>("op_61149_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5163_cast = mul(x = var_61148_cast, y = var_61149_to_fp16)[name = tensor<string, []>("aw_5163_cast")];
            tensor<string, []> var_61152_equation_0 = const()[name = tensor<string, []>("op_61152_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61152_cast = einsum(equation = var_61152_equation_0, values = (var_61074_cast, var_61031_cast))[name = tensor<string, []>("op_61152_cast")];
            tensor<fp16, []> var_61153_to_fp16 = const()[name = tensor<string, []>("op_61153_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5165_cast = mul(x = var_61152_cast, y = var_61153_to_fp16)[name = tensor<string, []>("aw_5165_cast")];
            tensor<string, []> var_61156_equation_0 = const()[name = tensor<string, []>("op_61156_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61156_cast = einsum(equation = var_61156_equation_0, values = (var_61078_cast, var_61035_cast))[name = tensor<string, []>("op_61156_cast")];
            tensor<fp16, []> var_61157_to_fp16 = const()[name = tensor<string, []>("op_61157_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5167_cast = mul(x = var_61156_cast, y = var_61157_to_fp16)[name = tensor<string, []>("aw_5167_cast")];
            tensor<string, []> var_61160_equation_0 = const()[name = tensor<string, []>("op_61160_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61160_cast = einsum(equation = var_61160_equation_0, values = (var_61082_cast, var_61039_cast))[name = tensor<string, []>("op_61160_cast")];
            tensor<fp16, []> var_61161_to_fp16 = const()[name = tensor<string, []>("op_61161_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5169_cast = mul(x = var_61160_cast, y = var_61161_to_fp16)[name = tensor<string, []>("aw_5169_cast")];
            tensor<string, []> var_61164_equation_0 = const()[name = tensor<string, []>("op_61164_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61164_cast = einsum(equation = var_61164_equation_0, values = (var_61086_cast, var_61043_cast))[name = tensor<string, []>("op_61164_cast")];
            tensor<fp16, []> var_61165_to_fp16 = const()[name = tensor<string, []>("op_61165_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5171_cast = mul(x = var_61164_cast, y = var_61165_to_fp16)[name = tensor<string, []>("aw_5171_cast")];
            tensor<string, []> var_61168_equation_0 = const()[name = tensor<string, []>("op_61168_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61168_cast = einsum(equation = var_61168_equation_0, values = (var_61090_cast, var_61047_cast))[name = tensor<string, []>("op_61168_cast")];
            tensor<fp16, []> var_61169_to_fp16 = const()[name = tensor<string, []>("op_61169_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5173_cast = mul(x = var_61168_cast, y = var_61169_to_fp16)[name = tensor<string, []>("aw_5173_cast")];
            tensor<string, []> var_61172_equation_0 = const()[name = tensor<string, []>("op_61172_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61172_cast = einsum(equation = var_61172_equation_0, values = (var_61094_cast, var_61051_cast))[name = tensor<string, []>("op_61172_cast")];
            tensor<fp16, []> var_61173_to_fp16 = const()[name = tensor<string, []>("op_61173_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5175_cast = mul(x = var_61172_cast, y = var_61173_to_fp16)[name = tensor<string, []>("aw_5175_cast")];
            tensor<string, []> var_61176_equation_0 = const()[name = tensor<string, []>("op_61176_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61176_cast = einsum(equation = var_61176_equation_0, values = (var_61098_cast, var_61055_cast))[name = tensor<string, []>("op_61176_cast")];
            tensor<fp16, []> var_61177_to_fp16 = const()[name = tensor<string, []>("op_61177_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5177_cast = mul(x = var_61176_cast, y = var_61177_to_fp16)[name = tensor<string, []>("aw_5177_cast")];
            tensor<string, []> var_61180_equation_0 = const()[name = tensor<string, []>("op_61180_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61180_cast = einsum(equation = var_61180_equation_0, values = (var_61102_cast, var_61059_cast))[name = tensor<string, []>("op_61180_cast")];
            tensor<fp16, []> var_61181_to_fp16 = const()[name = tensor<string, []>("op_61181_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5179_cast = mul(x = var_61180_cast, y = var_61181_to_fp16)[name = tensor<string, []>("aw_5179_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61183_cast = softmax(axis = var_58064, x = aw_5161_cast)[name = tensor<string, []>("op_61183_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61184_cast = softmax(axis = var_58064, x = aw_5163_cast)[name = tensor<string, []>("op_61184_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61185_cast = softmax(axis = var_58064, x = aw_5165_cast)[name = tensor<string, []>("op_61185_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61186_cast = softmax(axis = var_58064, x = aw_5167_cast)[name = tensor<string, []>("op_61186_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61187_cast = softmax(axis = var_58064, x = aw_5169_cast)[name = tensor<string, []>("op_61187_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61188_cast = softmax(axis = var_58064, x = aw_5171_cast)[name = tensor<string, []>("op_61188_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61189_cast = softmax(axis = var_58064, x = aw_5173_cast)[name = tensor<string, []>("op_61189_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61190_cast = softmax(axis = var_58064, x = aw_5175_cast)[name = tensor<string, []>("op_61190_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61191_cast = softmax(axis = var_58064, x = aw_5177_cast)[name = tensor<string, []>("op_61191_cast")];
            tensor<fp16, [2, 4096, 1, 4096]> var_61192_cast = softmax(axis = var_58064, x = aw_5179_cast)[name = tensor<string, []>("op_61192_cast")];
            tensor<string, []> var_61194_equation_0 = const()[name = tensor<string, []>("op_61194_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61194_cast = einsum(equation = var_61194_equation_0, values = (var_61104_cast, var_61183_cast))[name = tensor<string, []>("op_61194_cast")];
            tensor<string, []> var_61196_equation_0 = const()[name = tensor<string, []>("op_61196_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61196_cast = einsum(equation = var_61196_equation_0, values = (var_61108_cast, var_61184_cast))[name = tensor<string, []>("op_61196_cast")];
            tensor<string, []> var_61198_equation_0 = const()[name = tensor<string, []>("op_61198_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61198_cast = einsum(equation = var_61198_equation_0, values = (var_61112_cast, var_61185_cast))[name = tensor<string, []>("op_61198_cast")];
            tensor<string, []> var_61200_equation_0 = const()[name = tensor<string, []>("op_61200_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61200_cast = einsum(equation = var_61200_equation_0, values = (var_61116_cast, var_61186_cast))[name = tensor<string, []>("op_61200_cast")];
            tensor<string, []> var_61202_equation_0 = const()[name = tensor<string, []>("op_61202_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61202_cast = einsum(equation = var_61202_equation_0, values = (var_61120_cast, var_61187_cast))[name = tensor<string, []>("op_61202_cast")];
            tensor<string, []> var_61204_equation_0 = const()[name = tensor<string, []>("op_61204_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61204_cast = einsum(equation = var_61204_equation_0, values = (var_61124_cast, var_61188_cast))[name = tensor<string, []>("op_61204_cast")];
            tensor<string, []> var_61206_equation_0 = const()[name = tensor<string, []>("op_61206_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61206_cast = einsum(equation = var_61206_equation_0, values = (var_61128_cast, var_61189_cast))[name = tensor<string, []>("op_61206_cast")];
            tensor<string, []> var_61208_equation_0 = const()[name = tensor<string, []>("op_61208_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61208_cast = einsum(equation = var_61208_equation_0, values = (var_61132_cast, var_61190_cast))[name = tensor<string, []>("op_61208_cast")];
            tensor<string, []> var_61210_equation_0 = const()[name = tensor<string, []>("op_61210_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61210_cast = einsum(equation = var_61210_equation_0, values = (var_61136_cast, var_61191_cast))[name = tensor<string, []>("op_61210_cast")];
            tensor<string, []> var_61212_equation_0 = const()[name = tensor<string, []>("op_61212_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61212_cast = einsum(equation = var_61212_equation_0, values = (var_61140_cast, var_61192_cast))[name = tensor<string, []>("op_61212_cast")];
            tensor<bool, []> input_815_interleave_0 = const()[name = tensor<string, []>("input_815_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_815_cast = concat(axis = var_58064, interleave = input_815_interleave_0, values = (var_61194_cast, var_61196_cast, var_61198_cast, var_61200_cast, var_61202_cast, var_61204_cast, var_61206_cast, var_61208_cast, var_61210_cast, var_61212_cast))[name = tensor<string, []>("input_815_cast")];
            tensor<int32, [2]> var_61218 = const()[name = tensor<string, []>("op_61218"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61220 = const()[name = tensor<string, []>("op_61220"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_61222_pad_type_0 = const()[name = tensor<string, []>("op_61222_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_61222_pad_0 = const()[name = tensor<string, []>("op_61222_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5086950720)))];
            tensor<fp16, [640]> up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5087769984)))];
            tensor<fp16, [2, 640, 1, 4096]> var_61222_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_out_0_bias_to_fp16, dilations = var_61220, groups = var_58064, pad = var_61222_pad_0, pad_type = var_61222_pad_type_0, strides = var_61218, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn1_to_out_0_weight_to_fp16, x = input_815_cast)[name = tensor<string, []>("op_61222_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_417_cast = add(x = var_61222_cast, y = inputs_415_cast)[name = tensor<string, []>("inputs_417_cast")];
            tensor<int32, [1]> var_61226 = const()[name = tensor<string, []>("op_61226"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_417_cast = reduce_mean(axes = var_61226, keep_dims = var_58059, x = inputs_417_cast)[name = tensor<string, []>("channels_mean_417_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_417_cast = sub(x = inputs_417_cast, y = channels_mean_417_cast)[name = tensor<string, []>("zero_mean_417_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_417_cast = mul(x = zero_mean_417_cast, y = zero_mean_417_cast)[name = tensor<string, []>("zero_mean_sq_417_cast")];
            tensor<int32, [1]> var_61230 = const()[name = tensor<string, []>("op_61230"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_61231_cast = reduce_mean(axes = var_61230, keep_dims = var_58059, x = zero_mean_sq_417_cast)[name = tensor<string, []>("op_61231_cast")];
            tensor<fp16, []> var_61232_to_fp16 = const()[name = tensor<string, []>("op_61232_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_61233_cast = add(x = var_61231_cast, y = var_61232_to_fp16)[name = tensor<string, []>("op_61233_cast")];
            tensor<fp16, []> denom_417_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_417_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_417_cast = rsqrt(epsilon = denom_417_epsilon_0_to_fp16, x = var_61233_cast)[name = tensor<string, []>("denom_417_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_417_cast = mul(x = zero_mean_417_cast, y = denom_417_cast)[name = tensor<string, []>("out_417_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_61237_to_fp16 = const()[name = tensor<string, []>("op_61237_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5087771328)))];
            tensor<fp16, [2, 640, 1, 4096]> var_61238_cast = add(x = out_417_cast, y = var_61237_to_fp16)[name = tensor<string, []>("op_61238_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_61240_to_fp16 = const()[name = tensor<string, []>("op_61240_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5087772672)))];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_569_cast = mul(x = var_61238_cast, y = var_61240_to_fp16)[name = tensor<string, []>("hidden_states_569_cast")];
            tensor<int32, [2]> var_61247 = const()[name = tensor<string, []>("op_61247"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61249 = const()[name = tensor<string, []>("op_61249"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_pad_type_0 = const()[name = tensor<string, []>("q_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_pad_0 = const()[name = tensor<string, []>("q_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5087774016)))];
            tensor<fp16, [2, 640, 1, 4096]> q_cast = conv(dilations = var_61249, groups = var_58064, pad = q_pad_0, pad_type = q_pad_type_0, strides = var_61247, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_q_weight_to_fp16, x = hidden_states_569_cast)[name = tensor<string, []>("q_cast")];
            tensor<int32, [2]> var_61253 = const()[name = tensor<string, []>("op_61253"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61255 = const()[name = tensor<string, []>("op_61255"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_557_pad_type_0 = const()[name = tensor<string, []>("k_557_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_557_pad_0 = const()[name = tensor<string, []>("k_557_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5088593280)))];
            tensor<fp16, [2, 640, 1, 77]> k_557_cast = conv(dilations = var_61255, groups = var_58064, pad = k_557_pad_0, pad_type = k_557_pad_type_0, strides = var_61253, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("k_557_cast")];
            tensor<int32, [2]> var_61259 = const()[name = tensor<string, []>("op_61259"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61261 = const()[name = tensor<string, []>("op_61261"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_pad_type_0 = const()[name = tensor<string, []>("v_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_pad_0 = const()[name = tensor<string, []>("v_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2048, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [640, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5091214784)))];
            tensor<fp16, [2, 640, 1, 77]> v_cast = conv(dilations = var_61261, groups = var_58064, pad = v_pad_0, pad_type = v_pad_type_0, strides = var_61259, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor<string, []>("v_cast")];
            tensor<int32, [4]> var_61265_begin_0 = const()[name = tensor<string, []>("op_61265_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_61265_end_0 = const()[name = tensor<string, []>("op_61265_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_61265_end_mask_0 = const()[name = tensor<string, []>("op_61265_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61265_cast = slice_by_index(begin = var_61265_begin_0, end = var_61265_end_0, end_mask = var_61265_end_mask_0, x = q_cast)[name = tensor<string, []>("op_61265_cast")];
            tensor<int32, [4]> var_61269_begin_0 = const()[name = tensor<string, []>("op_61269_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_61269_end_0 = const()[name = tensor<string, []>("op_61269_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_61269_end_mask_0 = const()[name = tensor<string, []>("op_61269_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61269_cast = slice_by_index(begin = var_61269_begin_0, end = var_61269_end_0, end_mask = var_61269_end_mask_0, x = q_cast)[name = tensor<string, []>("op_61269_cast")];
            tensor<int32, [4]> var_61273_begin_0 = const()[name = tensor<string, []>("op_61273_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_61273_end_0 = const()[name = tensor<string, []>("op_61273_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_61273_end_mask_0 = const()[name = tensor<string, []>("op_61273_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61273_cast = slice_by_index(begin = var_61273_begin_0, end = var_61273_end_0, end_mask = var_61273_end_mask_0, x = q_cast)[name = tensor<string, []>("op_61273_cast")];
            tensor<int32, [4]> var_61277_begin_0 = const()[name = tensor<string, []>("op_61277_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_61277_end_0 = const()[name = tensor<string, []>("op_61277_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_61277_end_mask_0 = const()[name = tensor<string, []>("op_61277_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61277_cast = slice_by_index(begin = var_61277_begin_0, end = var_61277_end_0, end_mask = var_61277_end_mask_0, x = q_cast)[name = tensor<string, []>("op_61277_cast")];
            tensor<int32, [4]> var_61281_begin_0 = const()[name = tensor<string, []>("op_61281_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_61281_end_0 = const()[name = tensor<string, []>("op_61281_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_61281_end_mask_0 = const()[name = tensor<string, []>("op_61281_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61281_cast = slice_by_index(begin = var_61281_begin_0, end = var_61281_end_0, end_mask = var_61281_end_mask_0, x = q_cast)[name = tensor<string, []>("op_61281_cast")];
            tensor<int32, [4]> var_61285_begin_0 = const()[name = tensor<string, []>("op_61285_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_61285_end_0 = const()[name = tensor<string, []>("op_61285_end_0"), val = tensor<int32, [4]>([2, 384, 1, 4096])];
            tensor<bool, [4]> var_61285_end_mask_0 = const()[name = tensor<string, []>("op_61285_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61285_cast = slice_by_index(begin = var_61285_begin_0, end = var_61285_end_0, end_mask = var_61285_end_mask_0, x = q_cast)[name = tensor<string, []>("op_61285_cast")];
            tensor<int32, [4]> var_61289_begin_0 = const()[name = tensor<string, []>("op_61289_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_61289_end_0 = const()[name = tensor<string, []>("op_61289_end_0"), val = tensor<int32, [4]>([2, 448, 1, 4096])];
            tensor<bool, [4]> var_61289_end_mask_0 = const()[name = tensor<string, []>("op_61289_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61289_cast = slice_by_index(begin = var_61289_begin_0, end = var_61289_end_0, end_mask = var_61289_end_mask_0, x = q_cast)[name = tensor<string, []>("op_61289_cast")];
            tensor<int32, [4]> var_61293_begin_0 = const()[name = tensor<string, []>("op_61293_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_61293_end_0 = const()[name = tensor<string, []>("op_61293_end_0"), val = tensor<int32, [4]>([2, 512, 1, 4096])];
            tensor<bool, [4]> var_61293_end_mask_0 = const()[name = tensor<string, []>("op_61293_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61293_cast = slice_by_index(begin = var_61293_begin_0, end = var_61293_end_0, end_mask = var_61293_end_mask_0, x = q_cast)[name = tensor<string, []>("op_61293_cast")];
            tensor<int32, [4]> var_61297_begin_0 = const()[name = tensor<string, []>("op_61297_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_61297_end_0 = const()[name = tensor<string, []>("op_61297_end_0"), val = tensor<int32, [4]>([2, 576, 1, 4096])];
            tensor<bool, [4]> var_61297_end_mask_0 = const()[name = tensor<string, []>("op_61297_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61297_cast = slice_by_index(begin = var_61297_begin_0, end = var_61297_end_0, end_mask = var_61297_end_mask_0, x = q_cast)[name = tensor<string, []>("op_61297_cast")];
            tensor<int32, [4]> var_61301_begin_0 = const()[name = tensor<string, []>("op_61301_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_61301_end_0 = const()[name = tensor<string, []>("op_61301_end_0"), val = tensor<int32, [4]>([2, 640, 1, 4096])];
            tensor<bool, [4]> var_61301_end_mask_0 = const()[name = tensor<string, []>("op_61301_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_61301_cast = slice_by_index(begin = var_61301_begin_0, end = var_61301_end_0, end_mask = var_61301_end_mask_0, x = q_cast)[name = tensor<string, []>("op_61301_cast")];
            tensor<int32, [4]> k_perm_0 = const()[name = tensor<string, []>("k_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_61308_begin_0 = const()[name = tensor<string, []>("op_61308_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_61308_end_0 = const()[name = tensor<string, []>("op_61308_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_61308_end_mask_0 = const()[name = tensor<string, []>("op_61308_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> transpose_0 = transpose(perm = k_perm_0, x = k_557_cast)[name = tensor<string, []>("transpose_0")];
            tensor<fp16, [2, 77, 1, 64]> var_61308_cast = slice_by_index(begin = var_61308_begin_0, end = var_61308_end_0, end_mask = var_61308_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_61308_cast")];
            tensor<int32, [4]> var_61312_begin_0 = const()[name = tensor<string, []>("op_61312_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_61312_end_0 = const()[name = tensor<string, []>("op_61312_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_61312_end_mask_0 = const()[name = tensor<string, []>("op_61312_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_61312_cast = slice_by_index(begin = var_61312_begin_0, end = var_61312_end_0, end_mask = var_61312_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_61312_cast")];
            tensor<int32, [4]> var_61316_begin_0 = const()[name = tensor<string, []>("op_61316_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_61316_end_0 = const()[name = tensor<string, []>("op_61316_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_61316_end_mask_0 = const()[name = tensor<string, []>("op_61316_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_61316_cast = slice_by_index(begin = var_61316_begin_0, end = var_61316_end_0, end_mask = var_61316_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_61316_cast")];
            tensor<int32, [4]> var_61320_begin_0 = const()[name = tensor<string, []>("op_61320_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_61320_end_0 = const()[name = tensor<string, []>("op_61320_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_61320_end_mask_0 = const()[name = tensor<string, []>("op_61320_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_61320_cast = slice_by_index(begin = var_61320_begin_0, end = var_61320_end_0, end_mask = var_61320_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_61320_cast")];
            tensor<int32, [4]> var_61324_begin_0 = const()[name = tensor<string, []>("op_61324_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_61324_end_0 = const()[name = tensor<string, []>("op_61324_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_61324_end_mask_0 = const()[name = tensor<string, []>("op_61324_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_61324_cast = slice_by_index(begin = var_61324_begin_0, end = var_61324_end_0, end_mask = var_61324_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_61324_cast")];
            tensor<int32, [4]> var_61328_begin_0 = const()[name = tensor<string, []>("op_61328_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_61328_end_0 = const()[name = tensor<string, []>("op_61328_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_61328_end_mask_0 = const()[name = tensor<string, []>("op_61328_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_61328_cast = slice_by_index(begin = var_61328_begin_0, end = var_61328_end_0, end_mask = var_61328_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_61328_cast")];
            tensor<int32, [4]> var_61332_begin_0 = const()[name = tensor<string, []>("op_61332_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_61332_end_0 = const()[name = tensor<string, []>("op_61332_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_61332_end_mask_0 = const()[name = tensor<string, []>("op_61332_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_61332_cast = slice_by_index(begin = var_61332_begin_0, end = var_61332_end_0, end_mask = var_61332_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_61332_cast")];
            tensor<int32, [4]> var_61336_begin_0 = const()[name = tensor<string, []>("op_61336_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_61336_end_0 = const()[name = tensor<string, []>("op_61336_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_61336_end_mask_0 = const()[name = tensor<string, []>("op_61336_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_61336_cast = slice_by_index(begin = var_61336_begin_0, end = var_61336_end_0, end_mask = var_61336_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_61336_cast")];
            tensor<int32, [4]> var_61340_begin_0 = const()[name = tensor<string, []>("op_61340_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_61340_end_0 = const()[name = tensor<string, []>("op_61340_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_61340_end_mask_0 = const()[name = tensor<string, []>("op_61340_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_61340_cast = slice_by_index(begin = var_61340_begin_0, end = var_61340_end_0, end_mask = var_61340_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_61340_cast")];
            tensor<int32, [4]> var_61344_begin_0 = const()[name = tensor<string, []>("op_61344_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_61344_end_0 = const()[name = tensor<string, []>("op_61344_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_61344_end_mask_0 = const()[name = tensor<string, []>("op_61344_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_61344_cast = slice_by_index(begin = var_61344_begin_0, end = var_61344_end_0, end_mask = var_61344_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_61344_cast")];
            tensor<int32, [4]> var_61346_begin_0 = const()[name = tensor<string, []>("op_61346_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_61346_end_0 = const()[name = tensor<string, []>("op_61346_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_61346_end_mask_0 = const()[name = tensor<string, []>("op_61346_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61346_cast = slice_by_index(begin = var_61346_begin_0, end = var_61346_end_0, end_mask = var_61346_end_mask_0, x = v_cast)[name = tensor<string, []>("op_61346_cast")];
            tensor<int32, [4]> var_61350_begin_0 = const()[name = tensor<string, []>("op_61350_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_61350_end_0 = const()[name = tensor<string, []>("op_61350_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_61350_end_mask_0 = const()[name = tensor<string, []>("op_61350_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61350_cast = slice_by_index(begin = var_61350_begin_0, end = var_61350_end_0, end_mask = var_61350_end_mask_0, x = v_cast)[name = tensor<string, []>("op_61350_cast")];
            tensor<int32, [4]> var_61354_begin_0 = const()[name = tensor<string, []>("op_61354_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_61354_end_0 = const()[name = tensor<string, []>("op_61354_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_61354_end_mask_0 = const()[name = tensor<string, []>("op_61354_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61354_cast = slice_by_index(begin = var_61354_begin_0, end = var_61354_end_0, end_mask = var_61354_end_mask_0, x = v_cast)[name = tensor<string, []>("op_61354_cast")];
            tensor<int32, [4]> var_61358_begin_0 = const()[name = tensor<string, []>("op_61358_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_61358_end_0 = const()[name = tensor<string, []>("op_61358_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_61358_end_mask_0 = const()[name = tensor<string, []>("op_61358_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61358_cast = slice_by_index(begin = var_61358_begin_0, end = var_61358_end_0, end_mask = var_61358_end_mask_0, x = v_cast)[name = tensor<string, []>("op_61358_cast")];
            tensor<int32, [4]> var_61362_begin_0 = const()[name = tensor<string, []>("op_61362_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_61362_end_0 = const()[name = tensor<string, []>("op_61362_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_61362_end_mask_0 = const()[name = tensor<string, []>("op_61362_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61362_cast = slice_by_index(begin = var_61362_begin_0, end = var_61362_end_0, end_mask = var_61362_end_mask_0, x = v_cast)[name = tensor<string, []>("op_61362_cast")];
            tensor<int32, [4]> var_61366_begin_0 = const()[name = tensor<string, []>("op_61366_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_61366_end_0 = const()[name = tensor<string, []>("op_61366_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_61366_end_mask_0 = const()[name = tensor<string, []>("op_61366_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61366_cast = slice_by_index(begin = var_61366_begin_0, end = var_61366_end_0, end_mask = var_61366_end_mask_0, x = v_cast)[name = tensor<string, []>("op_61366_cast")];
            tensor<int32, [4]> var_61370_begin_0 = const()[name = tensor<string, []>("op_61370_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_61370_end_0 = const()[name = tensor<string, []>("op_61370_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_61370_end_mask_0 = const()[name = tensor<string, []>("op_61370_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61370_cast = slice_by_index(begin = var_61370_begin_0, end = var_61370_end_0, end_mask = var_61370_end_mask_0, x = v_cast)[name = tensor<string, []>("op_61370_cast")];
            tensor<int32, [4]> var_61374_begin_0 = const()[name = tensor<string, []>("op_61374_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_61374_end_0 = const()[name = tensor<string, []>("op_61374_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_61374_end_mask_0 = const()[name = tensor<string, []>("op_61374_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61374_cast = slice_by_index(begin = var_61374_begin_0, end = var_61374_end_0, end_mask = var_61374_end_mask_0, x = v_cast)[name = tensor<string, []>("op_61374_cast")];
            tensor<int32, [4]> var_61378_begin_0 = const()[name = tensor<string, []>("op_61378_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_61378_end_0 = const()[name = tensor<string, []>("op_61378_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_61378_end_mask_0 = const()[name = tensor<string, []>("op_61378_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61378_cast = slice_by_index(begin = var_61378_begin_0, end = var_61378_end_0, end_mask = var_61378_end_mask_0, x = v_cast)[name = tensor<string, []>("op_61378_cast")];
            tensor<int32, [4]> var_61382_begin_0 = const()[name = tensor<string, []>("op_61382_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_61382_end_0 = const()[name = tensor<string, []>("op_61382_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_61382_end_mask_0 = const()[name = tensor<string, []>("op_61382_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_61382_cast = slice_by_index(begin = var_61382_begin_0, end = var_61382_end_0, end_mask = var_61382_end_mask_0, x = v_cast)[name = tensor<string, []>("op_61382_cast")];
            tensor<string, []> var_61386_equation_0 = const()[name = tensor<string, []>("op_61386_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61386_cast = einsum(equation = var_61386_equation_0, values = (var_61308_cast, var_61265_cast))[name = tensor<string, []>("op_61386_cast")];
            tensor<fp16, []> var_61387_to_fp16 = const()[name = tensor<string, []>("op_61387_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5181_cast = mul(x = var_61386_cast, y = var_61387_to_fp16)[name = tensor<string, []>("aw_5181_cast")];
            tensor<string, []> var_61390_equation_0 = const()[name = tensor<string, []>("op_61390_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61390_cast = einsum(equation = var_61390_equation_0, values = (var_61312_cast, var_61269_cast))[name = tensor<string, []>("op_61390_cast")];
            tensor<fp16, []> var_61391_to_fp16 = const()[name = tensor<string, []>("op_61391_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5183_cast = mul(x = var_61390_cast, y = var_61391_to_fp16)[name = tensor<string, []>("aw_5183_cast")];
            tensor<string, []> var_61394_equation_0 = const()[name = tensor<string, []>("op_61394_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61394_cast = einsum(equation = var_61394_equation_0, values = (var_61316_cast, var_61273_cast))[name = tensor<string, []>("op_61394_cast")];
            tensor<fp16, []> var_61395_to_fp16 = const()[name = tensor<string, []>("op_61395_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5185_cast = mul(x = var_61394_cast, y = var_61395_to_fp16)[name = tensor<string, []>("aw_5185_cast")];
            tensor<string, []> var_61398_equation_0 = const()[name = tensor<string, []>("op_61398_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61398_cast = einsum(equation = var_61398_equation_0, values = (var_61320_cast, var_61277_cast))[name = tensor<string, []>("op_61398_cast")];
            tensor<fp16, []> var_61399_to_fp16 = const()[name = tensor<string, []>("op_61399_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5187_cast = mul(x = var_61398_cast, y = var_61399_to_fp16)[name = tensor<string, []>("aw_5187_cast")];
            tensor<string, []> var_61402_equation_0 = const()[name = tensor<string, []>("op_61402_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61402_cast = einsum(equation = var_61402_equation_0, values = (var_61324_cast, var_61281_cast))[name = tensor<string, []>("op_61402_cast")];
            tensor<fp16, []> var_61403_to_fp16 = const()[name = tensor<string, []>("op_61403_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5189_cast = mul(x = var_61402_cast, y = var_61403_to_fp16)[name = tensor<string, []>("aw_5189_cast")];
            tensor<string, []> var_61406_equation_0 = const()[name = tensor<string, []>("op_61406_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61406_cast = einsum(equation = var_61406_equation_0, values = (var_61328_cast, var_61285_cast))[name = tensor<string, []>("op_61406_cast")];
            tensor<fp16, []> var_61407_to_fp16 = const()[name = tensor<string, []>("op_61407_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5191_cast = mul(x = var_61406_cast, y = var_61407_to_fp16)[name = tensor<string, []>("aw_5191_cast")];
            tensor<string, []> var_61410_equation_0 = const()[name = tensor<string, []>("op_61410_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61410_cast = einsum(equation = var_61410_equation_0, values = (var_61332_cast, var_61289_cast))[name = tensor<string, []>("op_61410_cast")];
            tensor<fp16, []> var_61411_to_fp16 = const()[name = tensor<string, []>("op_61411_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5193_cast = mul(x = var_61410_cast, y = var_61411_to_fp16)[name = tensor<string, []>("aw_5193_cast")];
            tensor<string, []> var_61414_equation_0 = const()[name = tensor<string, []>("op_61414_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61414_cast = einsum(equation = var_61414_equation_0, values = (var_61336_cast, var_61293_cast))[name = tensor<string, []>("op_61414_cast")];
            tensor<fp16, []> var_61415_to_fp16 = const()[name = tensor<string, []>("op_61415_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5195_cast = mul(x = var_61414_cast, y = var_61415_to_fp16)[name = tensor<string, []>("aw_5195_cast")];
            tensor<string, []> var_61418_equation_0 = const()[name = tensor<string, []>("op_61418_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61418_cast = einsum(equation = var_61418_equation_0, values = (var_61340_cast, var_61297_cast))[name = tensor<string, []>("op_61418_cast")];
            tensor<fp16, []> var_61419_to_fp16 = const()[name = tensor<string, []>("op_61419_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_5197_cast = mul(x = var_61418_cast, y = var_61419_to_fp16)[name = tensor<string, []>("aw_5197_cast")];
            tensor<string, []> var_61422_equation_0 = const()[name = tensor<string, []>("op_61422_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_61422_cast = einsum(equation = var_61422_equation_0, values = (var_61344_cast, var_61301_cast))[name = tensor<string, []>("op_61422_cast")];
            tensor<fp16, []> var_61423_to_fp16 = const()[name = tensor<string, []>("op_61423_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_cast = mul(x = var_61422_cast, y = var_61423_to_fp16)[name = tensor<string, []>("aw_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_61425_cast = softmax(axis = var_58064, x = aw_5181_cast)[name = tensor<string, []>("op_61425_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_61426_cast = softmax(axis = var_58064, x = aw_5183_cast)[name = tensor<string, []>("op_61426_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_61427_cast = softmax(axis = var_58064, x = aw_5185_cast)[name = tensor<string, []>("op_61427_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_61428_cast = softmax(axis = var_58064, x = aw_5187_cast)[name = tensor<string, []>("op_61428_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_61429_cast = softmax(axis = var_58064, x = aw_5189_cast)[name = tensor<string, []>("op_61429_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_61430_cast = softmax(axis = var_58064, x = aw_5191_cast)[name = tensor<string, []>("op_61430_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_61431_cast = softmax(axis = var_58064, x = aw_5193_cast)[name = tensor<string, []>("op_61431_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_61432_cast = softmax(axis = var_58064, x = aw_5195_cast)[name = tensor<string, []>("op_61432_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_61433_cast = softmax(axis = var_58064, x = aw_5197_cast)[name = tensor<string, []>("op_61433_cast")];
            tensor<fp16, [2, 77, 1, 4096]> var_61434_cast = softmax(axis = var_58064, x = aw_cast)[name = tensor<string, []>("op_61434_cast")];
            tensor<string, []> var_61436_equation_0 = const()[name = tensor<string, []>("op_61436_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61436_cast = einsum(equation = var_61436_equation_0, values = (var_61346_cast, var_61425_cast))[name = tensor<string, []>("op_61436_cast")];
            tensor<string, []> var_61438_equation_0 = const()[name = tensor<string, []>("op_61438_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61438_cast = einsum(equation = var_61438_equation_0, values = (var_61350_cast, var_61426_cast))[name = tensor<string, []>("op_61438_cast")];
            tensor<string, []> var_61440_equation_0 = const()[name = tensor<string, []>("op_61440_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61440_cast = einsum(equation = var_61440_equation_0, values = (var_61354_cast, var_61427_cast))[name = tensor<string, []>("op_61440_cast")];
            tensor<string, []> var_61442_equation_0 = const()[name = tensor<string, []>("op_61442_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61442_cast = einsum(equation = var_61442_equation_0, values = (var_61358_cast, var_61428_cast))[name = tensor<string, []>("op_61442_cast")];
            tensor<string, []> var_61444_equation_0 = const()[name = tensor<string, []>("op_61444_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61444_cast = einsum(equation = var_61444_equation_0, values = (var_61362_cast, var_61429_cast))[name = tensor<string, []>("op_61444_cast")];
            tensor<string, []> var_61446_equation_0 = const()[name = tensor<string, []>("op_61446_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61446_cast = einsum(equation = var_61446_equation_0, values = (var_61366_cast, var_61430_cast))[name = tensor<string, []>("op_61446_cast")];
            tensor<string, []> var_61448_equation_0 = const()[name = tensor<string, []>("op_61448_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61448_cast = einsum(equation = var_61448_equation_0, values = (var_61370_cast, var_61431_cast))[name = tensor<string, []>("op_61448_cast")];
            tensor<string, []> var_61450_equation_0 = const()[name = tensor<string, []>("op_61450_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61450_cast = einsum(equation = var_61450_equation_0, values = (var_61374_cast, var_61432_cast))[name = tensor<string, []>("op_61450_cast")];
            tensor<string, []> var_61452_equation_0 = const()[name = tensor<string, []>("op_61452_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61452_cast = einsum(equation = var_61452_equation_0, values = (var_61378_cast, var_61433_cast))[name = tensor<string, []>("op_61452_cast")];
            tensor<string, []> var_61454_equation_0 = const()[name = tensor<string, []>("op_61454_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_61454_cast = einsum(equation = var_61454_equation_0, values = (var_61382_cast, var_61434_cast))[name = tensor<string, []>("op_61454_cast")];
            tensor<bool, []> input_817_interleave_0 = const()[name = tensor<string, []>("input_817_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 4096]> input_817_cast = concat(axis = var_58064, interleave = input_817_interleave_0, values = (var_61436_cast, var_61438_cast, var_61440_cast, var_61442_cast, var_61444_cast, var_61446_cast, var_61448_cast, var_61450_cast, var_61452_cast, var_61454_cast))[name = tensor<string, []>("input_817_cast")];
            tensor<int32, [2]> var_61460 = const()[name = tensor<string, []>("op_61460"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61462 = const()[name = tensor<string, []>("op_61462"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_61464_pad_type_0 = const()[name = tensor<string, []>("op_61464_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_61464_pad_0 = const()[name = tensor<string, []>("op_61464_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5093836288)))];
            tensor<fp16, [640]> up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5094655552)))];
            tensor<fp16, [2, 640, 1, 4096]> var_61464_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_out_0_bias_to_fp16, dilations = var_61462, groups = var_58064, pad = var_61464_pad_0, pad_type = var_61464_pad_type_0, strides = var_61460, weight = up_blocks_1_attentions_2_transformer_blocks_1_attn2_to_out_0_weight_to_fp16, x = input_817_cast)[name = tensor<string, []>("op_61464_cast")];
            tensor<fp16, [2, 640, 1, 4096]> inputs_cast = add(x = var_61464_cast, y = inputs_417_cast)[name = tensor<string, []>("inputs_cast")];
            tensor<int32, [1]> var_61468 = const()[name = tensor<string, []>("op_61468"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_cast = reduce_mean(axes = var_61468, keep_dims = var_58059, x = inputs_cast)[name = tensor<string, []>("channels_mean_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_cast = sub(x = inputs_cast, y = channels_mean_cast)[name = tensor<string, []>("zero_mean_cast")];
            tensor<fp16, [2, 640, 1, 4096]> zero_mean_sq_cast = mul(x = zero_mean_cast, y = zero_mean_cast)[name = tensor<string, []>("zero_mean_sq_cast")];
            tensor<int32, [1]> var_61472 = const()[name = tensor<string, []>("op_61472"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_61473_cast = reduce_mean(axes = var_61472, keep_dims = var_58059, x = zero_mean_sq_cast)[name = tensor<string, []>("op_61473_cast")];
            tensor<fp16, []> var_61474_to_fp16 = const()[name = tensor<string, []>("op_61474_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_61475_cast = add(x = var_61473_cast, y = var_61474_to_fp16)[name = tensor<string, []>("op_61475_cast")];
            tensor<fp16, []> denom_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_cast = rsqrt(epsilon = denom_epsilon_0_to_fp16, x = var_61475_cast)[name = tensor<string, []>("denom_cast")];
            tensor<fp16, [2, 640, 1, 4096]> out_cast = mul(x = zero_mean_cast, y = denom_cast)[name = tensor<string, []>("out_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_61479_to_fp16 = const()[name = tensor<string, []>("op_61479_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5094656896)))];
            tensor<fp16, [2, 640, 1, 4096]> var_61480_cast = add(x = out_cast, y = var_61479_to_fp16)[name = tensor<string, []>("op_61480_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_61482_to_fp16 = const()[name = tensor<string, []>("op_61482_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5094658240)))];
            tensor<fp16, [2, 640, 1, 4096]> input_819_cast = mul(x = var_61480_cast, y = var_61482_to_fp16)[name = tensor<string, []>("input_819_cast")];
            tensor<int32, [2]> var_61490 = const()[name = tensor<string, []>("op_61490"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61492 = const()[name = tensor<string, []>("op_61492"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_61494_pad_type_0 = const()[name = tensor<string, []>("op_61494_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_61494_pad_0 = const()[name = tensor<string, []>("op_61494_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [5120, 640, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [5120, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5094659584)))];
            tensor<fp16, [5120]> up_blocks_1_attentions_2_transformer_blocks_1_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5101213248)))];
            tensor<fp16, [2, 5120, 1, 4096]> var_61494_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_1_ff_net_0_proj_bias_to_fp16, dilations = var_61492, groups = var_58064, pad = var_61494_pad_0, pad_type = var_61494_pad_type_0, strides = var_61490, weight = up_blocks_1_attentions_2_transformer_blocks_1_ff_net_0_proj_weight_to_fp16, x = input_819_cast)[name = tensor<string, []>("op_61494_cast")];
            tensor<int32, [2]> var_61495_split_sizes_0 = const()[name = tensor<string, []>("op_61495_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_61495_axis_0 = const()[name = tensor<string, []>("op_61495_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 4096]> var_61495_cast_0, tensor<fp16, [2, 2560, 1, 4096]> var_61495_cast_1 = split(axis = var_61495_axis_0, split_sizes = var_61495_split_sizes_0, x = var_61494_cast)[name = tensor<string, []>("op_61495_cast")];
            tensor<string, []> var_61497_mode_0 = const()[name = tensor<string, []>("op_61497_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 4096]> var_61497_cast = gelu(mode = var_61497_mode_0, x = var_61495_cast_1)[name = tensor<string, []>("op_61497_cast")];
            tensor<fp16, [2, 2560, 1, 4096]> input_821_cast = mul(x = var_61495_cast_0, y = var_61497_cast)[name = tensor<string, []>("input_821_cast")];
            tensor<int32, [2]> var_61501 = const()[name = tensor<string, []>("op_61501"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61503 = const()[name = tensor<string, []>("op_61503"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_61505_pad_type_0 = const()[name = tensor<string, []>("op_61505_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_61505_pad_0 = const()[name = tensor<string, []>("op_61505_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2560, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_1_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_ff_net_2_weight_to_fp16"), val = tensor<fp16, [640, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5101223552)))];
            tensor<fp16, [640]> up_blocks_1_attentions_2_transformer_blocks_1_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_1_ff_net_2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5104500416)))];
            tensor<fp16, [2, 640, 1, 4096]> var_61505_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_1_ff_net_2_bias_to_fp16, dilations = var_61503, groups = var_58064, pad = var_61505_pad_0, pad_type = var_61505_pad_type_0, strides = var_61501, weight = up_blocks_1_attentions_2_transformer_blocks_1_ff_net_2_weight_to_fp16, x = input_821_cast)[name = tensor<string, []>("op_61505_cast")];
            tensor<fp16, [2, 640, 1, 4096]> hidden_states_573_cast = add(x = var_61505_cast, y = inputs_cast)[name = tensor<string, []>("hidden_states_573_cast")];
            tensor<int32, [4]> var_61507 = const()[name = tensor<string, []>("op_61507"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> input_823_cast = reshape(shape = var_61507, x = hidden_states_573_cast)[name = tensor<string, []>("input_823_cast")];
            tensor<int32, [2]> var_61511 = const()[name = tensor<string, []>("op_61511"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61513 = const()[name = tensor<string, []>("op_61513"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_575_pad_type_0 = const()[name = tensor<string, []>("hidden_states_575_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_575_pad_0 = const()[name = tensor<string, []>("hidden_states_575_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_1_attentions_2_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_proj_out_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5104501760)))];
            tensor<fp16, [640]> up_blocks_1_attentions_2_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_proj_out_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5105321024)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_575_cast = conv(bias = up_blocks_1_attentions_2_proj_out_bias_to_fp16, dilations = var_61513, groups = var_58064, pad = hidden_states_575_pad_0, pad_type = hidden_states_575_pad_type_0, strides = var_61511, weight = up_blocks_1_attentions_2_proj_out_weight_to_fp16, x = input_823_cast)[name = tensor<string, []>("hidden_states_575_cast")];
            tensor<fp16, [2, 640, 64, 64]> input_825_cast = add(x = hidden_states_575_cast, y = hidden_states_557_cast)[name = tensor<string, []>("input_825_cast")];
            tensor<fp32, []> input_827_scale_factor_height_0 = const()[name = tensor<string, []>("input_827_scale_factor_height_0"), val = tensor<fp32, []>(0x1p+1)];
            tensor<fp32, []> input_827_scale_factor_width_0 = const()[name = tensor<string, []>("input_827_scale_factor_width_0"), val = tensor<fp32, []>(0x1p+1)];
            tensor<fp16, [2, 640, 128, 128]> input_827_cast = upsample_nearest_neighbor(scale_factor_height = input_827_scale_factor_height_0, scale_factor_width = input_827_scale_factor_width_0, x = input_825_cast)[name = tensor<string, []>("input_827_cast")];
            tensor<int32, [2]> var_61522 = const()[name = tensor<string, []>("op_61522"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61524 = const()[name = tensor<string, []>("op_61524"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_577_pad_type_0 = const()[name = tensor<string, []>("hidden_states_577_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_577_pad_0 = const()[name = tensor<string, []>("hidden_states_577_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 640, 3, 3]> up_blocks_1_upsamplers_0_conv_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_upsamplers_0_conv_weight_to_fp16"), val = tensor<fp16, [640, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5105322368)))];
            tensor<fp16, [640]> up_blocks_1_upsamplers_0_conv_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_upsamplers_0_conv_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5112695232)))];
            tensor<fp16, [2, 640, 128, 128]> hidden_states_577_cast = conv(bias = up_blocks_1_upsamplers_0_conv_bias_to_fp16, dilations = var_61524, groups = var_58064, pad = hidden_states_577_pad_0, pad_type = hidden_states_577_pad_type_0, strides = var_61522, weight = up_blocks_1_upsamplers_0_conv_weight_to_fp16, x = input_827_cast)[name = tensor<string, []>("hidden_states_577_cast")];
            tensor<int32, []> var_61532 = const()[name = tensor<string, []>("op_61532"), val = tensor<int32, []>(1)];
            tensor<bool, []> input_829_interleave_0 = const()[name = tensor<string, []>("input_829_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 960, 128, 128]> input_829_cast = concat(axis = var_61532, interleave = input_829_interleave_0, values = (hidden_states_577_cast, res_hidden_states_13_cast))[name = tensor<string, []>("input_829_cast")];
            tensor<int32, [5]> reshape_156_shape_0 = const()[name = tensor<string, []>("reshape_156_shape_0"), val = tensor<int32, [5]>([2, 32, 30, 128, 128])];
            tensor<fp16, [2, 32, 30, 128, 128]> reshape_156_cast = reshape(shape = reshape_156_shape_0, x = input_829_cast)[name = tensor<string, []>("reshape_156_cast")];
            tensor<int32, [3]> reduce_mean_117_axes_0 = const()[name = tensor<string, []>("reduce_mean_117_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_117_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_117_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_117_cast = reduce_mean(axes = reduce_mean_117_axes_0, keep_dims = reduce_mean_117_keep_dims_0, x = reshape_156_cast)[name = tensor<string, []>("reduce_mean_117_cast")];
            tensor<fp16, [2, 32, 30, 128, 128]> sub_78_cast = sub(x = reshape_156_cast, y = reduce_mean_117_cast)[name = tensor<string, []>("sub_78_cast")];
            tensor<fp16, [2, 32, 30, 128, 128]> square_39_cast = square(x = sub_78_cast)[name = tensor<string, []>("square_39_cast")];
            tensor<int32, [3]> reduce_mean_119_axes_0 = const()[name = tensor<string, []>("reduce_mean_119_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_119_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_119_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_119_cast = reduce_mean(axes = reduce_mean_119_axes_0, keep_dims = reduce_mean_119_keep_dims_0, x = square_39_cast)[name = tensor<string, []>("reduce_mean_119_cast")];
            tensor<fp16, []> add_78_y_0_to_fp16 = const()[name = tensor<string, []>("add_78_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_78_cast = add(x = reduce_mean_119_cast, y = add_78_y_0_to_fp16)[name = tensor<string, []>("add_78_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_39_cast = sqrt(x = add_78_cast)[name = tensor<string, []>("sqrt_39_cast")];
            tensor<fp16, [2, 32, 30, 128, 128]> real_div_39_cast = real_div(x = sub_78_cast, y = sqrt_39_cast)[name = tensor<string, []>("real_div_39_cast")];
            tensor<int32, [4]> reshape_157_shape_0 = const()[name = tensor<string, []>("reshape_157_shape_0"), val = tensor<int32, [4]>([2, 960, 128, 128])];
            tensor<fp16, [2, 960, 128, 128]> reshape_157_cast = reshape(shape = reshape_157_shape_0, x = real_div_39_cast)[name = tensor<string, []>("reshape_157_cast")];
            tensor<fp16, [960]> add_79_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_79_gamma_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5112696576)))];
            tensor<fp16, [960]> add_79_beta_0_to_fp16 = const()[name = tensor<string, []>("add_79_beta_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5112698560)))];
            tensor<fp16, []> add_79_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_79_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 960, 128, 128]> add_79_cast = batch_norm(beta = add_79_beta_0_to_fp16, epsilon = add_79_epsilon_0_to_fp16, gamma = add_79_gamma_0_to_fp16, mean = add_73_mean_0_to_fp16, variance = add_73_variance_0_to_fp16, x = reshape_157_cast)[name = tensor<string, []>("add_79_cast")];
            tensor<fp16, [2, 960, 128, 128]> input_833_cast = silu(x = add_79_cast)[name = tensor<string, []>("input_833_cast")];
            tensor<int32, [2]> var_61553 = const()[name = tensor<string, []>("op_61553"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61555 = const()[name = tensor<string, []>("op_61555"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_579_pad_type_0 = const()[name = tensor<string, []>("hidden_states_579_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_579_pad_0 = const()[name = tensor<string, []>("hidden_states_579_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 960, 3, 3]> up_blocks_2_resnets_0_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [320, 960, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5112700544)))];
            tensor<fp16, [320]> up_blocks_2_resnets_0_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5118230208)))];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_579_cast = conv(bias = up_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = var_61555, groups = var_61532, pad = hidden_states_579_pad_0, pad_type = hidden_states_579_pad_type_0, strides = var_61553, weight = up_blocks_2_resnets_0_conv1_weight_to_fp16, x = input_833_cast)[name = tensor<string, []>("hidden_states_579_cast")];
            tensor<int32, [2]> var_61561 = const()[name = tensor<string, []>("op_61561"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61563 = const()[name = tensor<string, []>("op_61563"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_29_pad_type_0 = const()[name = tensor<string, []>("temb_29_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_29_pad_0 = const()[name = tensor<string, []>("temb_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1280, 1, 1]> up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [320, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5118230912)))];
            tensor<fp16, [320]> up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5119050176)))];
            tensor<fp16, [2, 320, 1, 1]> temb_29_cast = conv(bias = up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_61563, groups = var_61532, pad = temb_29_pad_0, pad_type = temb_29_pad_type_0, strides = var_61561, weight = up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16, x = input_23_cast)[name = tensor<string, []>("temb_29_cast")];
            tensor<fp16, [2, 320, 128, 128]> input_837_cast = add(x = hidden_states_579_cast, y = temb_29_cast)[name = tensor<string, []>("input_837_cast")];
            tensor<int32, [5]> reshape_160_shape_0 = const()[name = tensor<string, []>("reshape_160_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 128, 128])];
            tensor<fp16, [2, 32, 10, 128, 128]> reshape_160_cast = reshape(shape = reshape_160_shape_0, x = input_837_cast)[name = tensor<string, []>("reshape_160_cast")];
            tensor<int32, [3]> reduce_mean_120_axes_0 = const()[name = tensor<string, []>("reduce_mean_120_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_120_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_120_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_120_cast = reduce_mean(axes = reduce_mean_120_axes_0, keep_dims = reduce_mean_120_keep_dims_0, x = reshape_160_cast)[name = tensor<string, []>("reduce_mean_120_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> sub_80_cast = sub(x = reshape_160_cast, y = reduce_mean_120_cast)[name = tensor<string, []>("sub_80_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> square_40_cast = square(x = sub_80_cast)[name = tensor<string, []>("square_40_cast")];
            tensor<int32, [3]> reduce_mean_122_axes_0 = const()[name = tensor<string, []>("reduce_mean_122_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_122_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_122_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_122_cast = reduce_mean(axes = reduce_mean_122_axes_0, keep_dims = reduce_mean_122_keep_dims_0, x = square_40_cast)[name = tensor<string, []>("reduce_mean_122_cast")];
            tensor<fp16, []> add_80_y_0_to_fp16 = const()[name = tensor<string, []>("add_80_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_80_cast = add(x = reduce_mean_122_cast, y = add_80_y_0_to_fp16)[name = tensor<string, []>("add_80_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_40_cast = sqrt(x = add_80_cast)[name = tensor<string, []>("sqrt_40_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> real_div_40_cast = real_div(x = sub_80_cast, y = sqrt_40_cast)[name = tensor<string, []>("real_div_40_cast")];
            tensor<int32, [4]> reshape_161_shape_0 = const()[name = tensor<string, []>("reshape_161_shape_0"), val = tensor<int32, [4]>([2, 320, 128, 128])];
            tensor<fp16, [2, 320, 128, 128]> reshape_161_cast = reshape(shape = reshape_161_shape_0, x = real_div_40_cast)[name = tensor<string, []>("reshape_161_cast")];
            tensor<fp16, [320]> add_81_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_81_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5119050880)))];
            tensor<fp16, [320]> add_81_beta_0_to_fp16 = const()[name = tensor<string, []>("add_81_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5119051584)))];
            tensor<fp16, []> add_81_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_81_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 128, 128]> add_81_cast = batch_norm(beta = add_81_beta_0_to_fp16, epsilon = add_81_epsilon_0_to_fp16, gamma = add_81_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_161_cast)[name = tensor<string, []>("add_81_cast")];
            tensor<fp16, [2, 320, 128, 128]> input_841_cast = silu(x = add_81_cast)[name = tensor<string, []>("input_841_cast")];
            tensor<int32, [2]> var_61573 = const()[name = tensor<string, []>("op_61573"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61575 = const()[name = tensor<string, []>("op_61575"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_581_pad_type_0 = const()[name = tensor<string, []>("hidden_states_581_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_581_pad_0 = const()[name = tensor<string, []>("hidden_states_581_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 320, 3, 3]> up_blocks_2_resnets_0_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [320, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5119052288)))];
            tensor<fp16, [320]> up_blocks_2_resnets_0_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5120895552)))];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_581_cast = conv(bias = up_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = var_61575, groups = var_61532, pad = hidden_states_581_pad_0, pad_type = hidden_states_581_pad_type_0, strides = var_61573, weight = up_blocks_2_resnets_0_conv2_weight_to_fp16, x = input_841_cast)[name = tensor<string, []>("hidden_states_581_cast")];
            tensor<int32, [2]> var_61580 = const()[name = tensor<string, []>("op_61580"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61582 = const()[name = tensor<string, []>("op_61582"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_17_pad_type_0 = const()[name = tensor<string, []>("x_17_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_17_pad_0 = const()[name = tensor<string, []>("x_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 960, 1, 1]> up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [320, 960, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5120896256)))];
            tensor<fp16, [320]> up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5121510720)))];
            tensor<fp16, [2, 320, 128, 128]> x_17_cast = conv(bias = up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_61582, groups = var_61532, pad = x_17_pad_0, pad_type = x_17_pad_type_0, strides = var_61580, weight = up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16, x = input_829_cast)[name = tensor<string, []>("x_17_cast")];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_583_cast = add(x = x_17_cast, y = hidden_states_581_cast)[name = tensor<string, []>("hidden_states_583_cast")];
            tensor<bool, []> input_843_interleave_0 = const()[name = tensor<string, []>("input_843_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 128, 128]> input_843_cast = concat(axis = var_61532, interleave = input_843_interleave_0, values = (hidden_states_583_cast, res_hidden_states_15_cast))[name = tensor<string, []>("input_843_cast")];
            tensor<int32, [5]> reshape_164_shape_0 = const()[name = tensor<string, []>("reshape_164_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 128, 128])];
            tensor<fp16, [2, 32, 20, 128, 128]> reshape_164_cast = reshape(shape = reshape_164_shape_0, x = input_843_cast)[name = tensor<string, []>("reshape_164_cast")];
            tensor<int32, [3]> reduce_mean_123_axes_0 = const()[name = tensor<string, []>("reduce_mean_123_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_123_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_123_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_123_cast = reduce_mean(axes = reduce_mean_123_axes_0, keep_dims = reduce_mean_123_keep_dims_0, x = reshape_164_cast)[name = tensor<string, []>("reduce_mean_123_cast")];
            tensor<fp16, [2, 32, 20, 128, 128]> sub_82_cast = sub(x = reshape_164_cast, y = reduce_mean_123_cast)[name = tensor<string, []>("sub_82_cast")];
            tensor<fp16, [2, 32, 20, 128, 128]> square_41_cast = square(x = sub_82_cast)[name = tensor<string, []>("square_41_cast")];
            tensor<int32, [3]> reduce_mean_125_axes_0 = const()[name = tensor<string, []>("reduce_mean_125_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_125_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_125_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_125_cast = reduce_mean(axes = reduce_mean_125_axes_0, keep_dims = reduce_mean_125_keep_dims_0, x = square_41_cast)[name = tensor<string, []>("reduce_mean_125_cast")];
            tensor<fp16, []> add_82_y_0_to_fp16 = const()[name = tensor<string, []>("add_82_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_82_cast = add(x = reduce_mean_125_cast, y = add_82_y_0_to_fp16)[name = tensor<string, []>("add_82_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_41_cast = sqrt(x = add_82_cast)[name = tensor<string, []>("sqrt_41_cast")];
            tensor<fp16, [2, 32, 20, 128, 128]> real_div_41_cast = real_div(x = sub_82_cast, y = sqrt_41_cast)[name = tensor<string, []>("real_div_41_cast")];
            tensor<int32, [4]> reshape_165_shape_0 = const()[name = tensor<string, []>("reshape_165_shape_0"), val = tensor<int32, [4]>([2, 640, 128, 128])];
            tensor<fp16, [2, 640, 128, 128]> reshape_165_cast = reshape(shape = reshape_165_shape_0, x = real_div_41_cast)[name = tensor<string, []>("reshape_165_cast")];
            tensor<fp16, [640]> add_83_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_83_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5121511424)))];
            tensor<fp16, [640]> add_83_beta_0_to_fp16 = const()[name = tensor<string, []>("add_83_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5121512768)))];
            tensor<fp16, []> add_83_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_83_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 128, 128]> add_83_cast = batch_norm(beta = add_83_beta_0_to_fp16, epsilon = add_83_epsilon_0_to_fp16, gamma = add_83_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_165_cast)[name = tensor<string, []>("add_83_cast")];
            tensor<fp16, [2, 640, 128, 128]> input_847_cast = silu(x = add_83_cast)[name = tensor<string, []>("input_847_cast")];
            tensor<int32, [2]> var_61600 = const()[name = tensor<string, []>("op_61600"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61602 = const()[name = tensor<string, []>("op_61602"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_585_pad_type_0 = const()[name = tensor<string, []>("hidden_states_585_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_585_pad_0 = const()[name = tensor<string, []>("hidden_states_585_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 640, 3, 3]> up_blocks_2_resnets_1_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_1_conv1_weight_to_fp16"), val = tensor<fp16, [320, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5121514112)))];
            tensor<fp16, [320]> up_blocks_2_resnets_1_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_1_conv1_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5125200576)))];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_585_cast = conv(bias = up_blocks_2_resnets_1_conv1_bias_to_fp16, dilations = var_61602, groups = var_61532, pad = hidden_states_585_pad_0, pad_type = hidden_states_585_pad_type_0, strides = var_61600, weight = up_blocks_2_resnets_1_conv1_weight_to_fp16, x = input_847_cast)[name = tensor<string, []>("hidden_states_585_cast")];
            tensor<int32, [2]> var_61608 = const()[name = tensor<string, []>("op_61608"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61610 = const()[name = tensor<string, []>("op_61610"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_31_pad_type_0 = const()[name = tensor<string, []>("temb_31_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_31_pad_0 = const()[name = tensor<string, []>("temb_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1280, 1, 1]> up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [320, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5125201280)))];
            tensor<fp16, [320]> up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5126020544)))];
            tensor<fp16, [2, 320, 1, 1]> temb_31_cast = conv(bias = up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_61610, groups = var_61532, pad = temb_31_pad_0, pad_type = temb_31_pad_type_0, strides = var_61608, weight = up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16, x = input_23_cast)[name = tensor<string, []>("temb_31_cast")];
            tensor<fp16, [2, 320, 128, 128]> input_851_cast = add(x = hidden_states_585_cast, y = temb_31_cast)[name = tensor<string, []>("input_851_cast")];
            tensor<int32, [5]> reshape_168_shape_0 = const()[name = tensor<string, []>("reshape_168_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 128, 128])];
            tensor<fp16, [2, 32, 10, 128, 128]> reshape_168_cast = reshape(shape = reshape_168_shape_0, x = input_851_cast)[name = tensor<string, []>("reshape_168_cast")];
            tensor<int32, [3]> reduce_mean_126_axes_0 = const()[name = tensor<string, []>("reduce_mean_126_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_126_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_126_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_126_cast = reduce_mean(axes = reduce_mean_126_axes_0, keep_dims = reduce_mean_126_keep_dims_0, x = reshape_168_cast)[name = tensor<string, []>("reduce_mean_126_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> sub_84_cast = sub(x = reshape_168_cast, y = reduce_mean_126_cast)[name = tensor<string, []>("sub_84_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> square_42_cast = square(x = sub_84_cast)[name = tensor<string, []>("square_42_cast")];
            tensor<int32, [3]> reduce_mean_128_axes_0 = const()[name = tensor<string, []>("reduce_mean_128_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_128_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_128_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_128_cast = reduce_mean(axes = reduce_mean_128_axes_0, keep_dims = reduce_mean_128_keep_dims_0, x = square_42_cast)[name = tensor<string, []>("reduce_mean_128_cast")];
            tensor<fp16, []> add_84_y_0_to_fp16 = const()[name = tensor<string, []>("add_84_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_84_cast = add(x = reduce_mean_128_cast, y = add_84_y_0_to_fp16)[name = tensor<string, []>("add_84_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_42_cast = sqrt(x = add_84_cast)[name = tensor<string, []>("sqrt_42_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> real_div_42_cast = real_div(x = sub_84_cast, y = sqrt_42_cast)[name = tensor<string, []>("real_div_42_cast")];
            tensor<int32, [4]> reshape_169_shape_0 = const()[name = tensor<string, []>("reshape_169_shape_0"), val = tensor<int32, [4]>([2, 320, 128, 128])];
            tensor<fp16, [2, 320, 128, 128]> reshape_169_cast = reshape(shape = reshape_169_shape_0, x = real_div_42_cast)[name = tensor<string, []>("reshape_169_cast")];
            tensor<fp16, [320]> add_85_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_85_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5126021248)))];
            tensor<fp16, [320]> add_85_beta_0_to_fp16 = const()[name = tensor<string, []>("add_85_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5126021952)))];
            tensor<fp16, []> add_85_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_85_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 128, 128]> add_85_cast = batch_norm(beta = add_85_beta_0_to_fp16, epsilon = add_85_epsilon_0_to_fp16, gamma = add_85_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_169_cast)[name = tensor<string, []>("add_85_cast")];
            tensor<fp16, [2, 320, 128, 128]> input_855_cast = silu(x = add_85_cast)[name = tensor<string, []>("input_855_cast")];
            tensor<int32, [2]> var_61620 = const()[name = tensor<string, []>("op_61620"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61622 = const()[name = tensor<string, []>("op_61622"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_587_pad_type_0 = const()[name = tensor<string, []>("hidden_states_587_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_587_pad_0 = const()[name = tensor<string, []>("hidden_states_587_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 320, 3, 3]> up_blocks_2_resnets_1_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_1_conv2_weight_to_fp16"), val = tensor<fp16, [320, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5126022656)))];
            tensor<fp16, [320]> up_blocks_2_resnets_1_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_1_conv2_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5127865920)))];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_587_cast = conv(bias = up_blocks_2_resnets_1_conv2_bias_to_fp16, dilations = var_61622, groups = var_61532, pad = hidden_states_587_pad_0, pad_type = hidden_states_587_pad_type_0, strides = var_61620, weight = up_blocks_2_resnets_1_conv2_weight_to_fp16, x = input_855_cast)[name = tensor<string, []>("hidden_states_587_cast")];
            tensor<int32, [2]> var_61627 = const()[name = tensor<string, []>("op_61627"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61629 = const()[name = tensor<string, []>("op_61629"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_19_pad_type_0 = const()[name = tensor<string, []>("x_19_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_19_pad_0 = const()[name = tensor<string, []>("x_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 640, 1, 1]> up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [320, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5127866624)))];
            tensor<fp16, [320]> up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5128276288)))];
            tensor<fp16, [2, 320, 128, 128]> x_19_cast = conv(bias = up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_61629, groups = var_61532, pad = x_19_pad_0, pad_type = x_19_pad_type_0, strides = var_61627, weight = up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16, x = input_843_cast)[name = tensor<string, []>("x_19_cast")];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_589_cast = add(x = x_19_cast, y = hidden_states_587_cast)[name = tensor<string, []>("hidden_states_589_cast")];
            tensor<bool, []> input_857_interleave_0 = const()[name = tensor<string, []>("input_857_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 128, 128]> input_857_cast = concat(axis = var_61532, interleave = input_857_interleave_0, values = (hidden_states_589_cast, res_hidden_states_cast))[name = tensor<string, []>("input_857_cast")];
            tensor<int32, [5]> reshape_172_shape_0 = const()[name = tensor<string, []>("reshape_172_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 128, 128])];
            tensor<fp16, [2, 32, 20, 128, 128]> reshape_172_cast = reshape(shape = reshape_172_shape_0, x = input_857_cast)[name = tensor<string, []>("reshape_172_cast")];
            tensor<int32, [3]> reduce_mean_129_axes_0 = const()[name = tensor<string, []>("reduce_mean_129_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_129_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_129_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_129_cast = reduce_mean(axes = reduce_mean_129_axes_0, keep_dims = reduce_mean_129_keep_dims_0, x = reshape_172_cast)[name = tensor<string, []>("reduce_mean_129_cast")];
            tensor<fp16, [2, 32, 20, 128, 128]> sub_86_cast = sub(x = reshape_172_cast, y = reduce_mean_129_cast)[name = tensor<string, []>("sub_86_cast")];
            tensor<fp16, [2, 32, 20, 128, 128]> square_43_cast = square(x = sub_86_cast)[name = tensor<string, []>("square_43_cast")];
            tensor<int32, [3]> reduce_mean_131_axes_0 = const()[name = tensor<string, []>("reduce_mean_131_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_131_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_131_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_131_cast = reduce_mean(axes = reduce_mean_131_axes_0, keep_dims = reduce_mean_131_keep_dims_0, x = square_43_cast)[name = tensor<string, []>("reduce_mean_131_cast")];
            tensor<fp16, []> add_86_y_0_to_fp16 = const()[name = tensor<string, []>("add_86_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_86_cast = add(x = reduce_mean_131_cast, y = add_86_y_0_to_fp16)[name = tensor<string, []>("add_86_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_43_cast = sqrt(x = add_86_cast)[name = tensor<string, []>("sqrt_43_cast")];
            tensor<fp16, [2, 32, 20, 128, 128]> real_div_43_cast = real_div(x = sub_86_cast, y = sqrt_43_cast)[name = tensor<string, []>("real_div_43_cast")];
            tensor<int32, [4]> reshape_173_shape_0 = const()[name = tensor<string, []>("reshape_173_shape_0"), val = tensor<int32, [4]>([2, 640, 128, 128])];
            tensor<fp16, [2, 640, 128, 128]> reshape_173_cast = reshape(shape = reshape_173_shape_0, x = real_div_43_cast)[name = tensor<string, []>("reshape_173_cast")];
            tensor<fp16, [640]> add_87_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_87_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5128276992)))];
            tensor<fp16, [640]> add_87_beta_0_to_fp16 = const()[name = tensor<string, []>("add_87_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5128278336)))];
            tensor<fp16, []> add_87_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_87_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 128, 128]> add_87_cast = batch_norm(beta = add_87_beta_0_to_fp16, epsilon = add_87_epsilon_0_to_fp16, gamma = add_87_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_173_cast)[name = tensor<string, []>("add_87_cast")];
            tensor<fp16, [2, 640, 128, 128]> input_861_cast = silu(x = add_87_cast)[name = tensor<string, []>("input_861_cast")];
            tensor<int32, [2]> var_61647 = const()[name = tensor<string, []>("op_61647"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61649 = const()[name = tensor<string, []>("op_61649"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_591_pad_type_0 = const()[name = tensor<string, []>("hidden_states_591_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_591_pad_0 = const()[name = tensor<string, []>("hidden_states_591_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 640, 3, 3]> up_blocks_2_resnets_2_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_2_conv1_weight_to_fp16"), val = tensor<fp16, [320, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5128279680)))];
            tensor<fp16, [320]> up_blocks_2_resnets_2_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_2_conv1_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5131966144)))];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_591_cast = conv(bias = up_blocks_2_resnets_2_conv1_bias_to_fp16, dilations = var_61649, groups = var_61532, pad = hidden_states_591_pad_0, pad_type = hidden_states_591_pad_type_0, strides = var_61647, weight = up_blocks_2_resnets_2_conv1_weight_to_fp16, x = input_861_cast)[name = tensor<string, []>("hidden_states_591_cast")];
            tensor<int32, [2]> var_61655 = const()[name = tensor<string, []>("op_61655"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61657 = const()[name = tensor<string, []>("op_61657"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_pad_type_0 = const()[name = tensor<string, []>("temb_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_pad_0 = const()[name = tensor<string, []>("temb_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1280, 1, 1]> up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [320, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5131966848)))];
            tensor<fp16, [320]> up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5132786112)))];
            tensor<fp16, [2, 320, 1, 1]> temb_cast = conv(bias = up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_61657, groups = var_61532, pad = temb_pad_0, pad_type = temb_pad_type_0, strides = var_61655, weight = up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16, x = input_23_cast)[name = tensor<string, []>("temb_cast")];
            tensor<fp16, [2, 320, 128, 128]> input_865_cast = add(x = hidden_states_591_cast, y = temb_cast)[name = tensor<string, []>("input_865_cast")];
            tensor<int32, [5]> reshape_176_shape_0 = const()[name = tensor<string, []>("reshape_176_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 128, 128])];
            tensor<fp16, [2, 32, 10, 128, 128]> reshape_176_cast = reshape(shape = reshape_176_shape_0, x = input_865_cast)[name = tensor<string, []>("reshape_176_cast")];
            tensor<int32, [3]> reduce_mean_132_axes_0 = const()[name = tensor<string, []>("reduce_mean_132_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_132_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_132_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_132_cast = reduce_mean(axes = reduce_mean_132_axes_0, keep_dims = reduce_mean_132_keep_dims_0, x = reshape_176_cast)[name = tensor<string, []>("reduce_mean_132_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> sub_88_cast = sub(x = reshape_176_cast, y = reduce_mean_132_cast)[name = tensor<string, []>("sub_88_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> square_44_cast = square(x = sub_88_cast)[name = tensor<string, []>("square_44_cast")];
            tensor<int32, [3]> reduce_mean_134_axes_0 = const()[name = tensor<string, []>("reduce_mean_134_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_134_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_134_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_134_cast = reduce_mean(axes = reduce_mean_134_axes_0, keep_dims = reduce_mean_134_keep_dims_0, x = square_44_cast)[name = tensor<string, []>("reduce_mean_134_cast")];
            tensor<fp16, []> add_88_y_0_to_fp16 = const()[name = tensor<string, []>("add_88_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_88_cast = add(x = reduce_mean_134_cast, y = add_88_y_0_to_fp16)[name = tensor<string, []>("add_88_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_44_cast = sqrt(x = add_88_cast)[name = tensor<string, []>("sqrt_44_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> real_div_44_cast = real_div(x = sub_88_cast, y = sqrt_44_cast)[name = tensor<string, []>("real_div_44_cast")];
            tensor<int32, [4]> reshape_177_shape_0 = const()[name = tensor<string, []>("reshape_177_shape_0"), val = tensor<int32, [4]>([2, 320, 128, 128])];
            tensor<fp16, [2, 320, 128, 128]> reshape_177_cast = reshape(shape = reshape_177_shape_0, x = real_div_44_cast)[name = tensor<string, []>("reshape_177_cast")];
            tensor<fp16, [320]> add_89_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_89_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5132786816)))];
            tensor<fp16, [320]> add_89_beta_0_to_fp16 = const()[name = tensor<string, []>("add_89_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5132787520)))];
            tensor<fp16, []> add_89_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_89_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 128, 128]> add_89_cast = batch_norm(beta = add_89_beta_0_to_fp16, epsilon = add_89_epsilon_0_to_fp16, gamma = add_89_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_177_cast)[name = tensor<string, []>("add_89_cast")];
            tensor<fp16, [2, 320, 128, 128]> input_869_cast = silu(x = add_89_cast)[name = tensor<string, []>("input_869_cast")];
            tensor<int32, [2]> var_61667 = const()[name = tensor<string, []>("op_61667"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61669 = const()[name = tensor<string, []>("op_61669"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_pad_type_0 = const()[name = tensor<string, []>("hidden_states_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_pad_0 = const()[name = tensor<string, []>("hidden_states_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 320, 3, 3]> up_blocks_2_resnets_2_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_2_conv2_weight_to_fp16"), val = tensor<fp16, [320, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5132788224)))];
            tensor<fp16, [320]> up_blocks_2_resnets_2_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_2_conv2_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5134631488)))];
            tensor<fp16, [2, 320, 128, 128]> hidden_states_cast = conv(bias = up_blocks_2_resnets_2_conv2_bias_to_fp16, dilations = var_61669, groups = var_61532, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = var_61667, weight = up_blocks_2_resnets_2_conv2_weight_to_fp16, x = input_869_cast)[name = tensor<string, []>("hidden_states_cast")];
            tensor<int32, [2]> var_61674 = const()[name = tensor<string, []>("op_61674"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61676 = const()[name = tensor<string, []>("op_61676"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_pad_type_0 = const()[name = tensor<string, []>("x_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_pad_0 = const()[name = tensor<string, []>("x_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 640, 1, 1]> up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [320, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5134632192)))];
            tensor<fp16, [320]> up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5135041856)))];
            tensor<fp16, [2, 320, 128, 128]> x_cast = conv(bias = up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_61676, groups = var_61532, pad = x_pad_0, pad_type = x_pad_type_0, strides = var_61674, weight = up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16, x = input_857_cast)[name = tensor<string, []>("x_cast")];
            tensor<fp16, [2, 320, 128, 128]> input_871_cast = add(x = x_cast, y = hidden_states_cast)[name = tensor<string, []>("input_871_cast")];
            tensor<int32, [5]> reshape_180_shape_0 = const()[name = tensor<string, []>("reshape_180_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 128, 128])];
            tensor<fp16, [2, 32, 10, 128, 128]> reshape_180_cast = reshape(shape = reshape_180_shape_0, x = input_871_cast)[name = tensor<string, []>("reshape_180_cast")];
            tensor<int32, [3]> reduce_mean_135_axes_0 = const()[name = tensor<string, []>("reduce_mean_135_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_135_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_135_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_135_cast = reduce_mean(axes = reduce_mean_135_axes_0, keep_dims = reduce_mean_135_keep_dims_0, x = reshape_180_cast)[name = tensor<string, []>("reduce_mean_135_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> sub_90_cast = sub(x = reshape_180_cast, y = reduce_mean_135_cast)[name = tensor<string, []>("sub_90_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> square_45_cast = square(x = sub_90_cast)[name = tensor<string, []>("square_45_cast")];
            tensor<int32, [3]> reduce_mean_137_axes_0 = const()[name = tensor<string, []>("reduce_mean_137_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_137_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_137_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_137_cast = reduce_mean(axes = reduce_mean_137_axes_0, keep_dims = reduce_mean_137_keep_dims_0, x = square_45_cast)[name = tensor<string, []>("reduce_mean_137_cast")];
            tensor<fp16, []> add_90_y_0_to_fp16 = const()[name = tensor<string, []>("add_90_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_90_cast = add(x = reduce_mean_137_cast, y = add_90_y_0_to_fp16)[name = tensor<string, []>("add_90_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_45_cast = sqrt(x = add_90_cast)[name = tensor<string, []>("sqrt_45_cast")];
            tensor<fp16, [2, 32, 10, 128, 128]> real_div_45_cast = real_div(x = sub_90_cast, y = sqrt_45_cast)[name = tensor<string, []>("real_div_45_cast")];
            tensor<int32, [4]> reshape_181_shape_0 = const()[name = tensor<string, []>("reshape_181_shape_0"), val = tensor<int32, [4]>([2, 320, 128, 128])];
            tensor<fp16, [2, 320, 128, 128]> reshape_181_cast = reshape(shape = reshape_181_shape_0, x = real_div_45_cast)[name = tensor<string, []>("reshape_181_cast")];
            tensor<fp16, [320]> add_91_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_91_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5135042560)))];
            tensor<fp16, [320]> add_91_beta_0_to_fp16 = const()[name = tensor<string, []>("add_91_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5135043264)))];
            tensor<fp16, []> add_91_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_91_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 128, 128]> add_91_cast = batch_norm(beta = add_91_beta_0_to_fp16, epsilon = add_91_epsilon_0_to_fp16, gamma = add_91_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_181_cast)[name = tensor<string, []>("add_91_cast")];
            tensor<fp16, [2, 320, 128, 128]> input_cast = silu(x = add_91_cast)[name = tensor<string, []>("input_cast")];
            tensor<int32, []> var_61690 = const()[name = tensor<string, []>("op_61690"), val = tensor<int32, []>(1)];
            tensor<int32, [2]> var_61693 = const()[name = tensor<string, []>("op_61693"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_61695 = const()[name = tensor<string, []>("op_61695"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_61697_pad_type_0 = const()[name = tensor<string, []>("op_61697_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_61697_pad_0 = const()[name = tensor<string, []>("op_61697_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [4, 320, 3, 3]> conv_out_weight_to_fp16 = const()[name = tensor<string, []>("conv_out_weight_to_fp16"), val = tensor<fp16, [4, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5135043968)))];
            tensor<fp16, [4]> conv_out_bias_to_fp16 = const()[name = tensor<string, []>("conv_out_bias_to_fp16"), val = tensor<fp16, [4]>([0x1.664p-9, -0x1.72p-10, 0x1.06p-9, -0x1.9b8p-9])];
            tensor<fp16, [2, 4, 128, 128]> var_61697_cast = conv(bias = conv_out_bias_to_fp16, dilations = var_61695, groups = var_61690, pad = var_61697_pad_0, pad_type = var_61697_pad_type_0, strides = var_61693, weight = conv_out_weight_to_fp16, x = input_cast)[name = tensor<string, []>("op_61697_cast")];
            tensor<string, []> var_61697_cast_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_61697_cast_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
            tensor<fp32, [2, 4, 128, 128]> noise_pred = cast(dtype = var_61697_cast_to_fp32_dtype_0, x = var_61697_cast)[name = tensor<string, []>("cast_463")];
        } -> (noise_pred);
}